Pełne materiały konferencyjne z PWT 2012

Transkrypt

Pełne materiały konferencyjne z PWT 2012
XVI Poznańskie Warsztaty Telekomunikacyjne
14 grudnia 2012 – Poznań
Wydział Elektroniki i Telekomunikacji – Politechnika Poznańska
XVI Poznańskie Warsztaty Telekomunikacyjne - PWT 2012
14 grudnia 2012 – Poznań
KOMITET PROGRAMOWY PWT 2012
Marek Amanowicz
Daniel Józef Bem
Anna Cysewska-Sobusiak
Andrzej Dobrogowski
(przewodniczący)
Adam Dąbrowski
Marek Domański
Janusz Filipiak
Wojciech Kabaciński
Józef Lubacz
Zdzisław Papir
Wojciech Sobczak
Paweł Szulakiewicz
Zbigniew Szymański
Krzysztof Wesołowski
Marian Zientalski
Wojskowa Akademia Techniczna
Politechnika Wrocławska
Politechnika Poznańska
Politechnika Poznańska
Politechnika Poznańska
Politechnika Poznańska
Akademia Górniczo-Hutnicza
Politechnika Poznańska
Politechnika Warszawska
Akademia Górniczo-Hutnicza
Politechnika Gdańska
Politechnika Poznańska
Politechnika Poznańska
Politechnika Poznańska
Politechnika Gdańska
KOMITET ORGANIZACYJNY PWT 2012
Robert Kotrys
Piotr Remlein
Paweł Szulakiewicz
Barbara Wagrowska
Przewodniczący
Wydawca:
POLITECHNIKA POZNAŃSKA
Wydział Elektroniki i Telekomunikacji
ul. Polanka 3, 60-965 POZNAŃ
Email: [email protected] Strona Internetowa: http://www.pwt.et.put.poznan.pl
Wszystkie artykuły zostały zrecenzowane i opublikowane bez korekty w formie przekazanej przez autorów.
All papers have been peer-reviewed and published as received without alteration of their technical contents.
ISBN 978‐83‐925227‐0‐6 Wydział Elektroniki i Telekomunikacji – Politechnika Poznańska
I
Spis referatów
Sesja I. Nowe obszary badań w telekomunikacji
1. Automatic lane detection
Mateusz Buczkowski, Ryszard Stasiński
3
2. Porównanie iteracyjnych metod synchronizacji fazy dla sygnałów z turbokodowaniem
Piotr Tyczka, Michał Kuśmirek
7
3. Implementation of the static round-robin dispatching scheme in the MSM
and SMM clos-network switch
Janusz Kleban
13
4. Pomiar właściwości spektralnych optycznej krotnicy przeplotowej
Jan Lamperski
17
5. Metoda precyzyjnego pomiaru FSR rezonatora optycznego
Jan Lamperski, Zofia Planner
20
6. Coherent data transmission in optical DWDM systems
Piotr Rydlichowski, Piotr Turowicz
23
7. An admission control algorithm for multiuser OFDMA-based cognitive
radio networks
Jerzy Martyna
28
8. Bayesian methods in optimization of analog communication systems with
feedback
Anatoliy Platonov
32
9. Path loss modelling in industrial environment
Sławomir J. Ambroziak
36
10. The effect of voice transmission over IP on text-independent speaker
verification performance
Waldemar Maciejko
40
11. Maximization of utility in computer network with application of game
theory
Magdalena Turowska
44
II
Sesja II. Synchronizacja w telekomunikacji
Sesja specjalna z okazji jubileuszu 50-lecia pracy prof. Andrzeja Dobrogowskiego
1. System wspomagania synchronizacji
Andrzej Dobrogowski, Mieczysław Jessa, Michał Kasznia, Krzysztof Lange
51
2. Przegląd metod wyznaczania maksymalnego błędu przedziału czasu
Andrzej Dobrogowski, Michał Kasznia
55
3. Synchronizacja w sieci Telekomunikacji Polskiej
Marek Dymowski, Tadeusz Pawszak
59
4. Sondy pomiarowe w systemie wspomagania synchronizacji
Andrzej Dobrogowski, Mieczysław Jessa, Michał Kasznia, Krzysztof Lange
63
5. FPGA-based time counters with a wander measurement mode
Ryszard Szplet, Paweł Kwiatkowski, Zbigniew Jachna, Krzysztof Rozyc
67
6. Metody wyznaczania parametrów sygnałów synchronizacji w czasie
rzeczywistym dla pomiarów wielokanałowych
Michał Kasznia
71
Sesja III. Dydaktyka w telekomunikacji
1. Signal processor as a master of local bus and as a slave of global bus (VME
standard)
Bogusław J. Wiśniewski, Barbara E. Szecówka – Wiśniewska, Jacek A. Ostrowski
77
2. Design and implementation of GEPON architecture in laboratory testbed
Magdalena Młynarczuk, Lech Smoleński
81
3. Tani sprzęt dla astrofotografii
Tomasz Kossowski, Ryszard Stasiński
85
4. A software environment for multicast routing testing
Paweł Krzyżaniak, Marek Michalski
88
5. Aplet JAVA do demonstracji modulacji cyfrowych
Maciej Krasicki, Bartosz Kaczmarek
92
III
SesjaI
Nowe obszary badań w telekomunikacji
XVI Poznańskie Warsztaty Telekomunikacyjne
1
2
PWT 2012 - Poznań, 14 grudnia 2012
2012
AUTOMATIC LANE DETECTION
Automatic Lane Detection
Mateusz Buczkowski∗ , Ryszard Stasiński∗
∗ Poznan
University of Technology,
Faculty of Electronics and Telecommunications
Abstract—In the paper a new lane detection algorithm is
described. The algorithm is accurate and simple, which allows its
real time implementation on a simple computer. Its speed is due
to problem-oriented postprocessing of Hough transform results,
and masking of next video frame on the basis of previous frame
processing results. Moreover, the algorithm does not require any
camera-dependent calibration.
Index Terms—Digital images, Image analysis, Object recognition
I. I NTRODUCTION
A
DVANCES in integrated circuits technology make it
possible to use computers in increasing number of applications. Decreasing size and growing computing power allows
implementation of more and more complicated functionalities
in commonly used equipment. In the paper application of a
small-size computer within a car for lane detection is considered. Data obtained from detection algorithm can be used
to improve driving comfort as well as safety on the road [3].
Some previous works on this subject have been reported in [2],
[5], [6], [10]. This paper describes an algorithm that is simple,
while having improved accuracy and usability. The paper
consists of five sections. In the following one basic image
processing theory that algorithm is based on is introduced. In
the third section main stages of the algorithm are described,
and intermediate results are shown. Fourth section presents
results of the algorithm and provides input parameters that
were used. The paper finishes with a conclusion in which main
features of the algorithm are recapitulated.
II. T HEORY
A. 2D Correlation
Correlation is function that shows dependencies between
signals. For a given time shift the more alike signals are, the
higher value of the correlation is obtained.
Correlation for one-dimensional signals is defined as
follows[11]:
Rx,y (i) =
∞
x(n) y ∗ (n + i)
B. Filtering
Filtering is based on 2D convolution which is defined as
follows[11]:
∞
y(i, j) =
Convolution in time corresponds to multiplication in the frequency domain, so filtering can be also done in the following
way:
Y (i, j) = X(i, j) · H(i, j)
A filter is defined by the kernel h(i, j) which determines
behavior and type of the filter. Two most common group of
filters are low-pass and high-pass filters. Low-pass filters are
responsible for removing details and noise from an image.
High-pass filters on the other hand are good for finding edges
and removing average value.
C. Hough transform
Hough transform changes (x, y) coordinates of an image
to coordinates (r, θ) where r is the distance of a point from
the lower-left corner of an image, and θ is the rising angle of
a line linking the corner with a point. Hough transform can
be used to find certain shapes in the input image, depending
on the used formula [8]. Two most common applications for
Hough transform is to find a line or ellipsis. General idea of
Hough transform can be described by the following algorithm:
1) Set increments for coordinates Δr, Δθ according to
desired accuracy and resolution
2) For each point (x0 , y0 ) in the input image, increment
values in output image whenever a line is going through
it which is described by parameters (r, θ) that fulfill the
following formula (in the case of line detection):
r(θ) = x0 · cos θ + y0 · sin θ
The higher the value in the output image, the higher the
probability that line with parameters (r, θ) is present in input
image.
III. A PPLICATION
The definition for two-dimensional signals is simply its
generalization:
∞
∞
n=−∞m=−∞
x(m, n) h(i − m, j − n)
n=−∞m=−∞
n=−∞
Rx,y (i, j) =
∞
x(m, n) y ∗ (m + i, n + j)
The new algorithm consists of preprocessing and main
stages. This section describes in details steps of these parts.
In preprocessing video is stabilized while each frame is
preprocessed individually. Main stage consists of frame filtration, detection of lines, selection of proper lines and their
combination.
XVI Poznańskie Warsztaty Telekomunikacyjne
3
AUTOMATIC LANE DETECTION
A. Stabilization and frame preprocessing
1) Stabilization: This step is not mandatory, but it can improve output results. Of course, it increases algorithm running
time, hence, stabilization can be turned on or off, depending
on used hardware capabilities. The stage is simplified with
respect to stand-alone stabilization algorithms[7]. Taking into
account nature of camera shake and movements of the car,
main direction in which stabilization should be performed is
vertical one. Another simplification that can be done is to
compensate only for image shift and neglect its tilt.
Image shift between consecutive frames is found by applying 2D correlation. Algorithm that is used to determine shift
vector is as follows [9]:
1) Discrete Fourier transform is calculated for images for
which shift is calculated (ia and ib ):
Ia = F{ia },
Ib = F{ib }
2) Cross-power spectrum is calculated according to the
following formula:
R=
Ia Ib∗
|Ia Ib∗ |
3) Cross-correlation is obtained by the inverse Fourier
transform:
r = F −1 {R}
4) Shift between images is equal to the position of the
maximum in the r image:
Figure 1.
Preprocessed image
“edges” are analyzed. If “candidate” has connection to “edge”
(even through other “candidates”) it becomes an “edge”.
Output image is binary which simplifies further processing.
Good results were obtained with low threshold set to 50 and
high one to 200.
2) Lines detection: After edge detection image is transformed using Hough transform. This transform, as described
in the previous section, can be used to find lines in an image.
Image obtained using Hough transform contains information
about how probable it is that input image contains line with
specific parameters. For all parameters (r, θ) whenever an
image value exceeds a threshold a line is created. Lines are
stored as set of pairs (rn , θn ). Exemplary image with found
lines is presented in the figure 2.
(x, y) = arg max{r}
(x,y)
Process of stabilization is obtained by shifting latter image in
vertical direction by value of y.
2) Frame preprocessing: Video sequence should have
proper input format for next steps of algorithm. This step
consists of image resizing and transforming color space from
RGB to gray scale. High resolution is not required to obtain
accurate results, so downscaling is performed to decrease
required processing time. Execution order of the first two
steps (stabilization and preprocessing) can be interchanged. If
preprocessing is performed firstly, stabilization will give worse
results, but is faster. An exemplary image that is used as input
data to main processing stage is presented in the figure 1.
B. Main processing
1) Filtration: Two filters are applied - low-pass filter and
Canny filter. Low-pass filter used in this algorithm is Gaussian
one of size 7x7. The operation removes unnecessary details
from the image, which could cause false lane detection. Second filter is the Canny one used in edge detection[4]. Canny
filter has two main parameters - low and high thresholds.
Canny filtering is performed in the following steps. Firstly,
edge detection is performed using Laplacian. Then, points
which values exceed low threshold are marked as “candidates”.
Points with values exceeding high threshold are marked as
“edge”. In the last step relations between “candidates” and
4
Figure 2.
Lines found using Hough transform
3) Lines selection: This step of processing is necessary for
removing lines that with high probability are not part of a lane.
The main criterion used to remove a line from the input set is
its angle. It is assumed, that the angle of the lines which are
borders of the lane can vary in a certain range. On the other
hand, a range is determined by a line location. Figure 3 shows
how does the ranges change with line location. It can be seen
that, the closer the line is to the middle of an image (A2 ), the
more vertical it can be. On the other hand, if a line is near to
the border it should be tilted - either to the left (A3 ) or the
right (A1 ), depending on which side is the line. This operation
removes fair amount of lines which could affect correctness of
results. Using position-depended angle ranges allows to detect
a lane border while the car is changing the lane. In the case of
location-independent ranges wider ranges would be required
to achieve such functionality and this would lead to too weak
filtration of lines - a lot of lines that should be removed would
be present in the output set. Formula describing angle of the
PWT 2012 - Poznań, 14 grudnia 2012
AUTOMATIC LANE DETECTION
Figure 3.
Dependency between angle ranges and line position
the case of two borders the algorithm has to assign each line
to two borders (of neighboring and middle lane). This is done
by comparing slope of each line with mean value of slopes.
If it is greater or equal, then line is assigned to one group.
Otherwise, a line is assigned to the second one. At this stage
of the algorithm, there are several groups of lines, where each
group corresponds to a different lane border. Within each of
these groups lines are being averaged, which gives single line
per each found lane border. Image obtained after this step
of proessing figure 4 is presented in figure 5. At this stage
lanes are found and the output can be used to run different
algorithms e.g. road surface markings detection [3].
line center is as follows:
αc = arctan
2x − W
H
where:
W - image width
H - image height
x - x coordinate of the middle of the line
αc - center angle of the range
Line angle can vary from αc − Δα/2 to αc + Δα/2, where
Δα is range width. Figure 4 presents the set of remaining lines
from image 2after applying angle condition.
Figure 5.
Output lines
5) Improving next image: As lanes are found the algorithm
can generate a mask, which can improve its results in following frames. Next input frame (in main processing stage) should
be multiplied by the mask generated in current step - or from
current input frame point of view, it should be multiplied with
mask generated in previous algorithm run:
yt (i, j) = xt (i, j) · mt−1 (i, j)
Figure 4.
Selected lines
4) Combining lines: Lines set, which is obtained after step
3 consists of many lines that do not differ much from some
others. Some groups of lines might have only slightly different
angles or positions, and their origin is common lane border. In
this step such lines will be combined together and the output
set of single lines per each lane border is obtained.
Firstly, coarse grouping is performed basing on positions.
Lines are assigned to group “left” and “right”. Within each of
these groups mean value and standard deviation of slopes is
computed. It is assumed that algorithm can detect middle lane
(where car is going) as well as neighboring lanes. This means
that maximum number of lines at the output of this step is
equal to four, i.e. two per group. Algorithm not always finds all
four lines - either because of cars occluding neighboring lanes
or simply because there is only a single lane on the road. This
means that before splitting groups into two and combining
lines, algorithm should check how many different lane borders
are found. This is done by verifying standard deviation of
slopes. If its value exceeds a threshold it is assumed that two
lane borders are found on the analyzed side. Otherwise, if
standard deviation is low, which means that all slopes are very
similar, it is assumed that only one lane border is found. In
where:
xt - current input image
yt - new (modified) input image
mt−1 - mask generated in previous frame processing
The mask is created in such a way that it multiplies area
on and near the lane border by the factor k1 which is equal to
1 or greater than 1 if input image is dark. Area between lane
borders is multiplied by the factor k2 which is smaller than 1,
but usually greater than 0. The mask is filtered with low-pass
filter, so the transition between areas multiplied by k1 and k2
is smooth. This mask attenuates areas where lane border is not
expected to be found, so it is assumed that lines that would be
detected in this area would come from noise or obstacles. In
this way algorithm is potentially decreasing number of false
lines in the next run. On the other hand, if k1 > 1 then the
absolute differences between values are enlarged, so edge is
detected more easily. Figure 6 presents mask obtained from
figure 5 that can be used to enhance next input image.
Gray background of the mask is value 1.0, i.e there are no
modifications in this area. Values near the found lane borders
are amplified by the value 1.3. Area in the middle of the lane
is attenuated (multiplied by factor 0.6).
This step is vulnerable to shake of the camera, so video
stabilization can bring significant improvement to this step.
Differences in position between consecutive images should be
as small as possible.
XVI Poznańskie Warsztaty Telekomunikacyjne
5
AUTOMATIC LANE DETECTION
Table I
I NPUT PARAMETERS
Figure 6.
Parameter
Gaussian filter size (n)
Canny filter thresholds (TC1 , TC2 )
Hough transform steps (Δr, Δθ)
Line detection threshold (TH )
Angle range width (Δφ)
Standard deviation threshold (Tσ )
Mask multiplication factors (k1 , k2 )
Enhancing mask
IV. R ESULTS
Figure 7 presents an exemplary result of the algorithm.
Processing time required for single frame processing is below
14 ms for 2.0 GHz processor (Intel® T5750) and below 28 ms
for 1.0 GHz one and that allows algorithm to work in real-time
with video stream respectively up to 70 or 35 FPS.
Default value
7x7
30; 160
1; π/180
30
45o
0.1
1.2; 0.4
lines found and more computations have to be done. This is,
however, not a significant change. If processing time has to
be decreased or on the other hand, there is a great overhead
of processing power or one can afford to increase processing
time to get better results, the image size in the preprocessing
stage can be altered.
V. C ONCLUSIONS
Figure 7.
Exemplary output frame
Another processed frame is presented in the figure 8. In
this case lane location imperfection caused by approximation
of its borders by straight lines is visible. Namely, in the more
distant part of the road, markings are not exactly followed.
This could be overcome by applying either a spline or set of
lines to approximate lane border[10].
Figure 8.
Exemplary output frame
Modification of algorithm parameters presented in TableI
does not have great influence on processing time but do change
its accuracy and correctness. Processing time depends on e.g.
line detection threshold - the lower the threshold, the more
6
The presented in the paper algorithm is able to detect lane
borders in real-time. Part of its efficiency is due to introduced
improvements: position-dependent line angle classification and
enhancing the next frame processing by generating a mask
from the previous frame. This resulted in construction of
low-complexity while accurate algorithm, nevertheless, its
accuracy and processing time can be improved even further.
Data generated by this algorithm can be used in different
applications e.g. driving assistant, or road sufrace markings, or
obstacle detection. Its important advantage is that it does not
transform image geometry while extracting information about
lane, as is done in other techniques [1][2]. This makes the
algorithm camera independent, setting parameters individually
for any camera is no longer needed.
R EFERENCES
[1] M. Aly. Real time detection of lane markers in urban streets. IEEE
Intelligent Vehicles Symposium, pages 7–12, 2008.
[2] A. Broggi. Robust real-time lane and road detection in critical shadow
conditions. International Symposium on Computer Vision. Proceedings.,
pages 353–358, 1995.
[3] M. Buczkowski. (in Polish) Automatyczne wykrywanie poziomych
znakow drogowych. 2012.
[4] J. Canny. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 679–698,
1986.
[5] J. V. Kittler J. Matas, C. Galambos. Robust detection of lines using the
progressive probabilistic hough transform. Computer Vision and Pattern
Recognition, pages 119–137, 1999.
[6] A. Broggi M. Bertozzi. Real-time lane and obstacle detection on the gold
system. IEEE Intelligent Vehicles Symposium, pages 213–218, 1996.
[7] L. Wang R. Jia, H.Zhang. Digital image stabilization based on phase
correlation. International Conference on Artificial Intelligence and
Computational Intelligence, pages 485–489, 2009.
[8] P. E. Hart R. O. Duda. Use of the hough transformation to detect lines
and curves in pictures. Communications of the ACM, pages 11–15, 1971.
[9] B. S. Reddy and B. N. Chatterji. An fft-based technique for translation,
rotation, and scale-invariant image registration. IEEE Transactions on
Image Processing, pages 1266–1271, 1996.
[10] D. Shen Y. Wang, E. K. Teoh. Lane detection and tracking using bsnake. Image and Vision Computing, pages 269–280, 2004.
[11] T. P. Zielinski. (in Polish) Cyfrowe przetwarzanie sygnalow. WKL,
2009.
PWT 2012 - Poznań, 14 grudnia 2012
2012
Porównanie iteracyjnych metod synchronizacji
fazy dla sygnałów z turbo-kodowaniem
Piotr Tyczka, Michał KuĞmirek
Streszczenie—Turbo-kody pozwalają na transmisjĊ w
warunkach małego stosunku sygnału do szumu co bardzo
utrudnia działanie tradycyjnych układów synchronizacyjnych. W
ostatnich latach zaproponowano w literaturze kilka algorytmów
turbo-synchronizacji fazy noĞnej. W artykule przedstawiono
dwie iteracyjne metody synchronizacji fazy dla sygnałów z turbokodowaniem: algorytm Oha-Cheuna oraz Zhang-Burra.
Algorytmy te zostały zaimplementowane i zaprezentowano oraz
porównano wyniki jakoĞci ich działania.
Słowa kluczowe—Synchronizacja fazy
synchroni-zacja, sygnały z turbo-kodowaniem
noĞnej,
turbo-
I. WPROWADZENIE
W
ostatnich latach nastąpił olbrzymi rozwój systemów
telekomunikacyjnych. Ciągle rosnący popyt na nowe
usługi telekomunikacyjne, a w tym na usługi multimedialne
oraz rozpowszechnienie radiokomunikacji ruchomej wymaga
wprowadzania systemów o wiĊkszej wydajnoĞci, pojemnoĞci i
szybkoĞci transmisji. Dla potrzeb tych systemów konieczne
jest stosowanie zaawansowanych technik transmisji
korzystających z nowoczesnych metod kodowania, modulacji,
wielodostĊpu itp. Konsekwencją tego jest prowadzenie
intensywnych badaĔ w tym obszarze i powstawanie coraz
skuteczniejszych i nowszych metod. Jednym z ostatnich
osiągniĊü w dziedzinie kodowania zabezpieczającego przed
błĊdami jest opracowanie turbo-kodów [1] [2].
C. Shannon w swojej fundamentalnej pracy [3] okreĞlił
teoretyczną granicĊ zysku kodowania i podał twierdzenie,
które gwarantuje, Īe istnieją kody osiągające dowolnie małe
prawdopodobieĔstwo błĊdu jeĞli przepływnoĞü transmitowanych danych jest mniejsza od pojemnoĞci kanału. Turbo-kody
swoimi właĞciwoĞciami znacznie zbliĪyły siĊ do granicy
Shannona. Niestety wpływ rzeczywistego kanału transmisyjnego powoduje miĊdzy innymi powstanie błĊdu
synchronizacji, który znacząco obniĪa jakoĞü systemu z turbokodowaniem. Systemy te są stosowane zwykle w Ğrodowisku,
gdzie panują warunki małego stosunku mocy sygnału do mocy
szumu, wyraĪonego poprzez stosunek Eb/N0. W takich
warunkach tradycyjne metody synchronizacji nie zapewniają
wymaganej jakoĞci estymat poszukiwanych parametrów, lub
do ich działania konieczne jest przesyłanie duĪej iloĞci
dodatkowej informacji synchronizacyjnej. Wobec powyĪszych
problemów, w ostatnich latach opracowano nowe techniki
synchronizacji przeznaczone dla systemów z turbokodowaniem. Algorytmy te wykorzystują iteracyjne dekodowanie do wyznaczenia poszukiwanych parametrów i z tego
wzglĊdu techniki te okreĞlane są nazwą turbo-synchronizacji.
W artykule przedstawiono aspekty implementacyjne oraz
porównanie jakoĞci działania dwóch algorytmów turbosynchronizacji fazy noĞnej: algorytmu Oha-Cheuna [4] oraz
algorytmu Zhang-Burra [5] dla wybranych systemów z turbokodowaniem. W rozdziałach 2 i 3 pracy zaprezentowano
szczegółowo powyĪsze układy synchronizacji. Opis modelu
symulacyjnego oraz uwagi dotyczące realizacji rozwaĪanych
układów synchronizacji zawarto w rozdziale 4. W rozdziale 5
przedstawiono wyniki przeprowadzonych badaĔ symulacyjnych i ich dyskusjĊ. Ostatni, 6. rozdział stanowi podsumowanie.
II. ALGORYTM OHA-CHEUNA
Algorytm ten został zaproponowany przez W. Oha i K.
Cheuna w pracy [4]. Schemat blokowy systemu przedstawia
rysunek 1. Zgodnie z rys. 1 bity wejĞciowe po zakodowaniu w
turbo-koderze podawane są na modulator BPSK i przesyłane
do odbiornika przez kanał AWGN. Po przejĞciu przez kanał
symbole są demodulowane. Za układem demodulatora
nastĊpuje korekcja błĊdu fazy i dekodowanie w turbodekoderze MAP. W metodzie tej zakłada siĊ, Īe błąd fazy jest
jednakowy dla całej ramki, a estymata tego błĊdu wyznaczana
jest jednokrotnie na załoĪoną liczbĊ iteracji turbo-dekodera.
Autorzy wykorzystali w swoim algorytmie fakt, Īe „moc”
informacji zewnĊtrznej ‫ܮ‬௘ ሺ‫ݔ‬௞ ሻ, generowanej przez moduły
SISO (ang. Soft-Input Soft-Output), maleje w stosunku do
błĊdu fazy (ang. phase offset). Zakłada on wykorzystanie
estymaty mocy ‫ܮ‬௘ ሺ‫ݔ‬௞ ሻ do obliczenia błĊdu fazy. Dla
zmniejszenia złoĪonoĞci implementacyjnej przyjĊto, Īe moc
informacji dodatkowej wyraĪa siĊ za pomocą metryki
obliczanej ze wzoru:
ேିଵ
‫ܯ‬ଵ௟ ൌ
ͳ
෍ห‫ܮ‬௘ ௟ଶ ሺ‫ݔ‬௞ ሻห ሺͳሻ
ܰ
௞ୀ଴
‫݁ܮ‬ଶ௟ ሺ‫ݔ‬௞ ሻ
Piotr Tyczka, Michał KuĞmirek, Katedra Radiokomunikacji, Politechnika
PoznaĔska, ul. Polanka 3, 60-965 PoznaĔ (tel. +48 61 665 39 18; e-mail:
tyczka@ et.put.poznan.pl).
jest informacją dodatkową dla k-tego bitu w lgdzie
tej ramce obliczonej przez drugi dekoder składowy w ostatniej
iteracji. Dla jeszcze wiĊkszego zmniejszenia złoĪonoĞci
obliczeniowej metrykĊ tĊ moĪna obliczaü według
XVI Poznańskie Warsztaty Telekomunikacyjne
7
݊݇
‫݇ݔ‬
Modulator
BPSK
Turbo-koder
ܰܽ
;
݁ െ݆Ȳ ݅
݈
;
݁ െ݆ߠ ݅
;
Demodulator
WstĊpna
inicjacja fazy
Estymacja
fazy
III. ALGORYTM ZHANG-BURRA
‫ ݈ʹ݁ܮ‬ሺ‫ ݇ݔ‬ሻ
‫ݔ‬ො݇
Turbo-dekoder
MAP
Rys. 1. Model systemu z układem synchronizacji Oha-Cheuna.
uproszczonego wzoru:
ͳ
‫ܯ‬ଶ௟ ൌ ቂ ƒš ൛ห‫ܮ‬௘ ௟ଶ ሺ‫ݔ‬௞ ሻหൟ ൅ ‹ ൛ห‫ܮ‬௘ ௟ଶ ሺ‫ݔ‬௞ ሻหൟቃ ሺʹሻ
଴ஸ௞ஸேିଵ
ʹ ଴ஸ௞ஸேିଵ
Jak pokazano w [4], wartoĞci Ğrednich metryk ‫ܯ‬ଵ௟ i ‫ܯ‬ଶ௟
osiągają maksimum dla zerowego błĊdu fazy i gwałtownie
maleją wraz z jego wzrostem. Okazuje siĊ, Īe kompensacja
błĊdu fazy sprowadza siĊ do maksymalizacji wartoĞci metryk
‫ܯ‬௜௟ w zakresie േ͹Ͳι, co wymaga zastosowania wstĊpnej
korekcji fazy tak, aby błąd znalazł siĊ w akceptowalnych
granicach.
WstĊpna inicjacja fazy odbywa siĊ przez wyznaczenie fazy
Ȳ୧୒ୟ za pomocą wzoru:
ேିଵ
Ȳ୧୒ୟ ൌ ƒ”‰ ƒš
ట೘
12 ramek przy SNR równym 3 dB, Īeby osiągnąü stratĊ
nieprzekraczającą 0,1 dB w stosunku do przypadku idealnej
synchronizacji dla BER na poziomie 10-3. NaleĪy zwróciü
uwagĊ, Īe estymata błĊdu fazy zaleĪy od wartoĞci estymat
wyznaczonych dla dwóch poprzednich ramek. Wobec
powyĪszego, algorytm ten nie bĊdzie skutecznie działał w
przypadku szybkozmiennego błĊdu fazy.
ͳ
೘ ቚ ሺ͵ሻ
෍ ቚ‫ܮ‬௘ ௟ǡே௔ǡట
ଶǡ௞
ܰ
௞ୀ଴
೘ jest
gdzie ߰௠ ൌ ݉ߨΤʹ, dla m = { 0, 1, 2, 3 } a ‫ܮ‬௘ ௟ǡே௔ǡట
ଶǡ௞
informacją dodatkową generowaną przez drugi dekoder
składowy dla k-tego bitu, Na-tej iteracji dla l-tej ramki, dla
sygnału kompensowanego przez ߰௠ . RównieĪ w tym
wypadku, moĪemy dla zmniejszenia złoĪonoĞci obliczeniowej,
tak jak w przypadku obliczania metryk, zastąpiü uĞrednienie
po wszystkich symbolach wzorem (2). Błąd inicjacji fazy
danej wzorem (3) nie wystĊpuje dla ܰ௔ ൒ ͵. Oznacza to, Īe
juĪ dla niewielkiej liczby nadmiarowych iteracji mamy bardzo
małe prawdopodobieĔstwo błĊdnego wyboru wstĊpnej fazy
korygującej.
Po zakoĔczeniu procesu wstĊpnej inicjacji fazy pozostaje
szczątkowy błąd fazy, który ma wartoĞü mniejszą niĪ ߨȀͶ.
WartoĞü tego błĊdu oblicza siĊ ze wzoru:
Innym algorytmem działającym zgodnie z ideą turbosynchronizacji jest algorytm autorstwa Li Zhang i Alistera G.
Burra, opisany w pracy [5]. Przedstawili oni algorytm dla
modulacji BPSK i QPSK, który z kaĪdą iteracją turbodekodera oblicza estymatĊ błĊdu fazy, którą wykorzystuje siĊ
w nastĊpnej iteracji w celu poprawienia wyników
dekodowania.
Przedstawmy sygnał odebrany jako:
‫ݎ‬௞ ൌ ‫ܣ‬௞ ‡š’ሺ݆߮௞ ሻሺͷሻ
gdzie ‫ܣ‬௞ jest amplitudą a ߮௞ reprezentuje argument, oba
zmieniające siĊ pod wpływem szumu i błĊdów fazy. Biorąc
pod uwagĊ kanał AWGN o mocy szumu ߪ ଶ , funkcja
wiarygodnoĞci dla estymacji błĊdu fazy ߮ na podstawie N
odebranych symboli wynosi:
ே
ͳ
Ȧሺ߮ሻ ൌ ‡š’ ൥ ଶ ෍ ܴ݁൫‫ݎ‬௞ ‫ݏ‬௞‫ כ‬ሺ߮ሻ൯൩ሺ͸ሻ
ߪ
௞ୀଵ
gdzie ‫ݏ‬௞ ሺ߮ሻ jest k-tym transmitowanym symbolem, który jest
funkcją błĊdu fazy ߮Ǥ Błąd fazy jest nieznany, ale stały dla
całej ramki danych.
Autorzy metody uĞrednili funkcjĊ wiarygodnoĞci po
wszystkich danych korzystając ze statystyk sygnału. W
tradycyjnych metodach wykorzystuje siĊ w tym celu rozkład
jednostajny. Natomiast w turbo-kodach, po pierwszej iteracji
znane są pewne prawdopodobieĔstwa dla kaĪdego
transmitowanego symbolu w sygnale odebranym. Oznaczają
one prawdopodobieĔstwo, Īe k-ty transmitowany symbol jest
m-tym
punktem
konstelacji.
UĞrednienie
funkcji
wiarygodnoĞci (6), biorąc pod uwagĊ te prawdopodobieĔstwa,
a nastĊpnie zlogarytmowanie prowadzi do logarytmicznej
funkcji wiarygodnoĞci (ang. Log-Likelihood Function −LLF)
postaci:
୒
୩ୀଵ
8
௠ୀ଴
ܴ݁ሺ‫ݎ‬௞ ‫ כ ݏ‬ሺ݉ǡ ߮ሻ
൨ൡ ߪଶ
୒
ൌ ෍ Ȧ୩୐ ሺɔሻ ሺ͹ሻ
ߠ෠ ௟ ൌ ߠ෠ ௟ିଵ ൅ ߤ൛‫݊݃ݏ‬൫ߠ෠ ௟ିଶ െ ߠ෠ ௟ିଵ ൯൫‫ܯ‬௜௟ െ ‫ܯ‬௜௟ିଵ ൯ൟሺͶሻ
gdzie ߤ jest krokiem korekcji fazy a ‫݊݃ݏ‬൫ߠ෠ ௟ିଶ െ ߠ෠ ௟ିଵ ൯
okreĞla jej kierunek. Współczynnik ߤ ma wpływ na szybkoĞü
korekcji błĊdu fazy czyli na iloĞü kroków potrzebnych do
właĞciwego wyznaczenia nieznanej wartoĞci błĊdu fazy.
Współczynnik ten wpływa równieĪ na stabilnoĞü algorytmu
czyli na zafalowania błĊdu fazy. Na podstawie wyników
opublikowanych w [4], algorytm ten potrzebuje jedynie około
ெିଵ
Ȧ୐ ሺ߮ሻ ൌ ෍ Ž ൝ ෍ ܲ௞ ሺ݉ሻ ‡š’ ൤
୩ୀଵ
Ȧ୩୐ ሺ߮ሻ
jest funkcją wiarygodnoĞci dla k-tego symbolu.
gdzie
Na podstawie obserwacji wartoĞci LLF w funkcji błĊdu
fazy ߮ okazuje siĊ, Īe zgodnie z regułą maksimum
wiarygodnoĞci estymata ߠ෠ jest wartoĞcią ߮ maksymalizującą
LLF. Do obliczenia maksimum funkcji LLF słuĪy pochodna
tej funkcji, która spełnia warunek:
PWT 2012 - Poznań, 14 grudnia 2012
݀Ȧ୐ ሺ߮ሻ
ൌ Ͳሺͺሻ
݀߮
PoniewaĪ obliczenie błĊdu fazy z zastosowaniem
bezpoĞrednio wzoru (7) wymagałoby bardzo duĪej liczby
obliczeĔ, co mogłoby wprowadziü opóĨnienia i problemy
implementacyjne, w pracy [5] funkcjĊ tĊ przedstawiono w
postaci szeregu Fouriera. W przypadku sygnałów BPSK, po
zastosowaniu odpowiednich przybliĪeĔ wynikających z
reprezentacji funkcji LLF za pomocą szeregu Fouriera,
uzyskuje siĊ ostatecznie nastĊpującą zaleĪnoĞü pozwalającą
෡
wyznaczyü estymatĊ ߠǣ
ߠ෠ ൌ
‫ݓ‬ଵ ߜଵ ൅ ʹ‫ݓ‬ଶ ߜଶ
ሺͻሻ
‫ݓ‬ଵ ൅ Ͷ‫ݓ‬ଶ
gdzie ‫ݓ‬ଵ i ‫ݓ‬ଶ są amplitudami, odpowiednio, pierwszej i
drugiej harmonicznej, a ߜଵ i ߜଶ są argumentami otrzymanymi
przez sumowanie po N symbolach według wzorów:
ே
‫ݓ‬ଵ …‘•ሺߜଵ ሻ ൌ ෍ ܽଵ௞ …‘•ሺ߮௞ ሻ
௞ୀଵ
ே
‫ݓ‬ଵ •‹ሺߜଵ ሻ ൌ ෍ ܽଵ௞ •‹ሺ߮௞ ሻ
௞ୀଵ
ே
‫ݓ‬ଶ …‘•ሺߜଶሻ ൌ
෍ ܽଶ௞
ሺͳͲሻ
…‘•ሺʹ߮௞ ሻ
‫ݓ‬ଶ •‹ሺߜଶሻ ൌ
ሾ‫ ݇ܣ‬ሿ݈
Obliczanie
fazy i
amplitudy
ሾ‫ ݇ݎ‬ሿ݈
ሾܽ݊݇ ܾ݊݇ ሿ݅െͳ
݈
Bufor danych
ߠ෠݈݅െͳ
Bufor
danych
ሾ‫ ݇݁ܮ‬ሿ݅െͳ
݈
Turbo-dekoder
Rys. 2. Schemat blokowy układu synchronizacji fazy Zhang-Burra z turbodekoderem.
IV. MODEL SYMULACYJNY I ASPEKTY IMPLEMENTACYJNE
Schemat blokowy symulowanego systemu przedstawia rys.
3. Parametrami wejĞciowymi symulowanego systemu są:
• wielkoĞü bloku danych wraz z bitami ogona – N
• wielomiany generujące dla turbo-kodera – g1 i g2
• wybór sprawnoĞci turbo-kodera – R = 1/2 lub R = 1/3
• wartoĞü Eb/N0 dla kanału AWGN wyraĪona w
decybelach
• iloĞü iteracji turbo-dekodera – Nit
Błąd fazy
‫݇ݕ‬
‫݇ݔ‬
Turbo-koder
•‹ሺʹ߮௞ ሻ
Tablica
współczynników
ሾ߮݇ ሿ݈
௞ୀଵ
ே
෍ ܽଶ௞
Modulator
BPSK
݀݇
;
Kanał
AWGN
௞ୀଵ
ܽ௡௞
jest współczynnikiem szeregu Fouriera,
przy czym
zdefiniowanym nastĊpująco:
ܽ௡௞ ൌ
ͳ గ ୩
න Ȧ ሺ߮ሻ …‘•ሺ݊߮ሻ݀߮ ሺͳͳሻ
ߨ ିగ ୐
Dodatkowo, aby zmniejszyü złoĪonoĞü obliczeniową związaną
z obliczeniami współczynników Fouriera, autorzy algorytmu
stablicowali je jako funkcjĊ ‫ܣ‬௞ i ‫ܮ‬௘ ሺ‫ݔ‬௞ ሻ. Tablica współczynników jest prosta i nie wymaga duĪej iloĞci pamiĊci.
DziĊki tym uproszczeniom, estymata błĊdu fazy moĪe byü
szybko obliczona i sam układ synchronizacji, którego schemat
blokowy jest przedstawiony na rys. 2 nie wprowadza Īadnego
opóĨnienia. Blok danych trafia jednoczeĞnie do układu
korekcji fazy oraz układu estymacji fazy. Estymata jest
inicjowana jako 0 i liczona osobno dla kaĪdej ramki. Układ
korekcji fazy dokonuje korekty na podstawie estymaty
wyliczonej w poprzedniej iteracji czyli blok danych
skorygowanych za pomocą estymaty wyliczonej w i-tej
iteracji bĊdzie dekodowany w i+1-szej iteracji.
Znaczącą zaletą algorytmu Zhang-Burra jest wyznaczanie
estymaty błĊdu fazy dla kaĪdej ramki niezaleĪnie. DziĊki temu
algorytm zachowuje swoje własnoĞci równieĪ w przypadku,
gdy błąd fazy zmienia siĊ dla kaĪdej ramki.
‫݇ݎ‬
Demodulator
Turbo-dekoder
‫ݔ‬ො݇ Rys. 3. Schemat blokowy symulowanego systemu.
Zgodnie z opisem zawartym w rozdziale II, algorytm OhaCheuna składa siĊ z dwóch etapów: inicjacji wstĊpnej i
właĞciwej estymacji błĊdu fazy. Działanie algorytmu opiera
siĊ o metryki liczone na podstawie informacji zewnĊtrznej
‫ܮ‬௘ ሺ݇ሻ. W implementacji algorytmu Oha-Cheuna wykorzystano dokładniejszy wzór (1), obliczający wartoĞü metryki dla ltej ramki. We wzorze tym korzysta siĊ z informacji
zewnĊtrznej ‫ܮ‬௘ ሺ݇ሻ z drugiego dekodera składowego,
wyliczonej w ostatniej iteracji turbo-dekodera.
W fazie wstĊpnej wykonuje siĊ po trzy iteracje dla czterech
róĪnych wartoĞci fazy równych 0, ʌ/2, ʌ oraz (3/2)ʌ,
wyraĪonych w radianach. Jak wykazano w [4] trzy iteracje są
wystarczające, aby uniknąü błĊdu inicjacji fazy. Po kaĪdej
trzeciej iteracji liczona jest metryka według wzoru (1) i
wybierana jest wartoĞü fazy, dla której metryka jest
maksymalna.
W fazie właĞciwej, układ synchronizacji korzysta z metryk
obliczonych na podstawie informacji zewnĊtrznej ‫ܮ‬௘ ሺ݇ሻ,
otrzymanych po ostatniej właĞciwej iteracji turbo-dekodera,
gdzie symbole na wejĞciu turbo-dekodera są skorygowane o
wartoĞü estymaty błĊdu fazy. PoniewaĪ estymata błĊdu fazy
XVI Poznańskie Warsztaty Telekomunikacyjne
9
ߠ෠ ௟ , zgodnie ze wzorem (4), liczona jest na podstawie wartoĞci
fazy dla ramki l-2 i l-1 oraz metryki ‫ܯ‬௟ dla ramki
l-1 i l, konieczne było ustalenie warunków początkowych. Na
podstawie analizy algorytmu przyjĊto, Īe wartoĞü ߠ෠ ௟ିଶ ൌ
Ͳǡnatomiast ߠ෠ ௟ିଵ powinna równaü siĊ wartoĞci fazy
wyliczonej w etapie wstĊpnej inicjacji. Ponadto przyjĊto
wartoĞü ‫ܯ‬௟ିଵ ൌ Ͳ, a ‫ ܯ‬௟ równe wartoĞci metryki, dla której
wyznaczono wartoĞü fazy w czasie inicjacji. Z uwagi na
zastosowanie algorytmu Log-MAP konieczne było
przeprowadzenie optymalizacji wartoĞci kroku korekcji fazy
μ, wystĊpującego we wzorze (4).
Układ
synchronizacji
fazy
Zhang-Burra,
który
przedstawiony został na rys. 2, jako parametry wejĞciowe
przyjmuje amplitudĊ ‫ܣ‬௞ i argument ߮௞ odebranego symbolu z
l-tej ramki oraz ‫ܮ‬௘ ሺ݇ሻ. Symbol wejĞciowy jest
reprezentowany jako liczba zespolona. Amplituda ‫ܣ‬௞ jest
liczona jako pierwiastek kwadratowy z sumy kwadratów
czĊĞci rzeczywistej i urojonej. Argument ߮௞ jest wartoĞcią
kąta z zakresu od 0 do 2ʌ i jest liczony za pomocą funkcji
atan2, która jako argumenty przyjmuje czĊĞü urojoną i
rzeczywistą liczby zespolonej. PoniewaĪ funkcja ta zwraca
wartoĞci w zakresie od –ʌ do ʌ, dla symboli ujemnych
wprowadzana jest korekta równa wartoĞci 2ʌ. Obliczenie
współczynników szeregu Fouriera niezbĊdnych w równaniach
(10) zostało zrealizowane na podstawie tablicy zawartej w [5]
za pomocą komend warunkowych. Rozwiązanie układu
równaĔ (10) doprowadziło do uzyskania nastĊpujących
zaleĪnoĞci:
‫ݕ‬
ߩ ൌ ƒ–ƒ ሺͳʹሻ
‫ݔ‬
‫ݔ‬
‫ݓ‬ൌ
ሺͳ͵ሻ
…‘• ߩ
Do porównania zostały wybrane krzywe otrzymane w
wyniku symulacji cyfrowej systemu dla odstrojenia równego
20°. W systemie z prostym czterostanowym koderem i ramką
N = 400 oraz N = 5000, moĪna przyjąü iĪ oba układy uzyskują
podobne wyniki (rys. 4 i 5). W obu przypadkach układy
synchronizacji nie wnoszą poprawy do systemu dla
odstrojenia mniejszego niĪ 10° oraz zapewniają działanie
systemu ze stratą nie przekraczającą 0,2 dB na poziomie BER
= 10-4 w porównaniu z przypadkiem idealnej synchronizacji.
Wyniki dla systemu z koderem oĞmiostanowym i ramką N =
400 oraz N = 5000 przedstawiają rys. 6 i 7. RównieĪ w tym
przypadku moĪna w ogólnoĞci stwierdziü, Īe układy OhaCheuna i Zhang-Burra uzyskują podobne wyniki działania.
Rys. 4. Porównanie jakoĞci działania algorytmów Oha-Cheuna i Zhang-Burra
(N = 400, R = 1/2, Nit = 5, g = (7, 5)8, Log-MAP).
ே
gdzie ‫ ݔ‬ൌ σே
௞ୀଵ ܽ௞ …‘•ሺ߮௞ ሻ, a ‫ ݕ‬ൌ σ௞ୀଵ ܽ௞ •‹ሺ߮௞ ሻ, które
zostały zaimplementowane w symulowanym systemie do
obliczenia estymaty błĊdu fazy na podstawie wzoru (9).
WartoĞü informacji zewnĊtrznej ‫ܮ‬௘ ሺ݇ሻ przed pierwszą iteracją
turbo-dekodera jest zerowa, dlatego dla pierwszej iteracji
przyjmuje siĊ wartoĞü estymaty ߠ෠ ൌ Ͳ.
V. WYNIKI BADAē
Badania przeprowadzone zostały dla trzech róĪnych błĊdów
fazy równych 10°, 20° i 30° oraz dla dwóch długoĞci ramek
N = 400 i N = 5000. Zastosowane zostały równieĪ dwa turbokodery o róĪnych wielomianach generujących koderów
składowych: czterostanowy koder RSC o wielomianach
generującym g = (7, 5)8 oraz bardziej złoĪony, oĞmiostanowy
koder RSC o wielomianach generujących g = (13, 15)8. W
badaniach, jak juĪ wczeĞniej wspomniano, przyjĊto Nit = 5
iteracji turbo-dekodera. Przeprowadzona optymalizacja
parametru ߤ dla algorytmu Oha-Cheuna wskazała na wartoĞü
0,2 i taka wartoĞü tego parametru została przyjĊta w
badaniach. Wyniki porównania jakoĞci działania algorytmów
synchronizacji fazy Oha-Cheuna i Zhang-Burra w róĪnych
warunkach pracy zostały przedstawione na rys. 4-10.
10
Rys. 5. Porównanie jakoĞci działania algorytmów Oha-Cheuna i Zhang-Burra
(N = 5000, R = 1/2, Nit = 5, g = (7, 5)8, Log-MAP).
Z badaĔ wynika, Īe zaprezentowane układy synchronizacji
uzyskują podobne wyniki dla róĪnych parametrów systemu.
Nawet w systemie, dla Ğredniego bloku danych N = 5000 i
oĞmiostanowego kodera RSC, który osiąga wartoĞü BER na
poziomie 10-6 przy Eb/N0 = 1,5 dB (rys. 7), oba układy
synchronizacji uzyskują jakoĞü działania ze stratą
nieprzekraczającą 0,3 dB dla BER = 10-4 w porównaniu do
przypadku z idealną synchronizacją. Oznacza to, Īe układy te
PWT 2012 - Poznań, 14 grudnia 2012
mogą byü stosowane w róĪnych systemach, co jest
niewątpliwie zaletą i elementem zachĊcającym do ich
stosowania.
Rys. 6. Porównanie jakoĞci działania algorytmów Oha-Cheuna i Zhang-Burra
(N = 400, R = 1/2, Nit = 5, g = (13, 15)8, Log-MAP).
Rys. 7. Porównanie jakoĞci działania algorytmów Oha-Cheuna i Zhang-Burra
(N = 5000, R = 1/2, Nit = 5, g = (13, 15)8, Log-MAP).
Dotychczas zaprezentowane wyniki zostały uzyskane dla
stałego zakłócenia fazy, niezmiennego dla wszystkich ramek
danych. W takim przypadku oba algorytmy uzyskują
jednakowe wyniki. Jedną z róĪnic jaka moĪe Ğwiadczyü na
korzyĞü algorytmu Zhang-Burra jest mniejszy dodatkowy
nakład obliczeniowy jaki wprowadza układ synchronizacji. W
algorytmie Oha-Cheuna istnieje bowiem koniecznoĞü
stosowania fazy inicjującej, która dla zapewnienia
zadowalających
wyników
wymaga
zastosowania
dodatkowych dwunastu iteracji turbo-dekodera. Jak pokazano
w [4], zwiĊkszanie liczby iteracji powyĪej dwunastu nie
powoduje poprawy w działaniu systemu. Dlatego nadmiarowe
iteracje fazy inicjacji mogą mieü wpływ na zwiĊkszenie iloĞci
obliczeĔ i opóĨnienie dekodowania.
Rysunki 8 i 9 pokazują wykresy estymacji fazy dla układu
synchronizacji Oha-Cheuna i Zhang-Burra w przypadku
stałego odstrojenia równego 30° oraz dla odstrojenia
zmieniającego siĊ liniowo w zakresie od 0° do 30°. Zmiana
błĊdu fazy w drugim przypadku wynosi jeden stopieĔ na
kaĪdą ramkĊ.
Rys. 8. Wykres wartoĞci estymaty fazy w przypadku stałego błĊdu fazy
równego 30° dla algorytmów Oha-Cheuna i Zhang-Burra.
Rys. 9. Wykres wartoĞci estymaty fazy w przypadku zmieniającego siĊ
liniowo błĊdu fazy dla algorytmów Oha-Cheuna i Zhang-Burra.
Jak widaü na rys. 8, w przypadku układu synchronizacji
Zhang-Burra róĪnica miĊdzy estymatą a wartoĞcią rzeczywistą
fazy wynosi 6°, czego wynikiem jest decybelowa strata w
systemie równa 0,2 dB przy BER = 10-4. Estymata ta
charakteryzuje siĊ duĪą stabilnoĞcią, a róĪnica jej wartoĞci dla
róĪnych ramek wynosi dziesiąte czĊĞci stopnia. Dla liniowo
zmieniającego siĊ błĊdu fazy (rys. 9) widaü wyraĨnie
zaleĪnoĞü wartoĞci estymaty od zmian odstrojenia. Pomijając
błąd wyznaczonej estymaty w stosunku do rzeczywistego
odstrojenia bĊdący wynikiem niskiej wartoĞci stosunku
sygnału do szumu, układ synchronizacji Zhang-Burra wiernie
odtwarza kształt krzywej prezentującej zmiany wartoĞci błĊdu
fazy.
Jak juĪ wczeĞniej nadmieniono, algorytm Oha-Cheuna przy
wyznaczaniu estymaty błĊdu fazy korzysta z wartoĞci
estymaty dla dwóch poprzednich ramek. Estymata dla ramki l
zaleĪy zatem od estymaty dla ramki l-1-szej i l-2-iej.
ZaleĪnoĞü ta negatywnie wpływa na działanie układu
synchronizacji w przypadku zmiennego odstrojenia fazy.
XVI Poznańskie Warsztaty Telekomunikacyjne
11
Estymata fazy oscyluje wokół wartoĞci rzeczywistego błĊdu
fazy, czego wynikiem jest strata jakoĞci w systemie, podobna
jak w przypadku algorytmu Zhang-Burra. W przypadku
stałego odstrojenia fazy oscylacje te są niewielkie, dlatego nie
powodują duĪej straty. Jednak dla zmiennego błĊdu fazy, jak
pokazuje rys. 9, algorytm nie nadąĪa z dostrojeniem siĊ do
wartoĞci rzeczywistego błĊdu fazy. Powoduje to powstanie
strat jakoĞci transmisji w systemie, które pokazane zostały na
rys. 10. Rysunek 10 przedstawia wykres zaleĪnoĞci BER w
funkcji Eb/N0 w systemie ze zmieniającą siĊ liniowo fazą w
zakresie 0° - 30° dla obu rozpatrywanych algorytmów
synchronizacji fazy. Krzywa BER dla algorytmu Zhang-Burra
przebiega w podobny sposób jak dla systemu ze stałym
zakłóceniem fazy. Strata w porównaniu do systemu z idealną
synchronizacją wynosi jedynie 0,2 dB na poziomie BER=10-4.
Strata w systemie z układem synchronizacji Oha-Cheuna jest
juĪ wiĊksza i na tym samym poziomie BER wynosi ponad 0,6
dB.
zmniejszająca straty jakoĞci transmisji w systemie wynikające
z istnienia błĊdu fazy. Straty Eb/N0 w systemie z badanymi
układami turbo-synchronizacji wynoszą mniej niĪ 0,3 dB w
porównaniu do systemu z idealną synchronizacją dla BER na
poziomie 10-4. Mimo niewielkich strat, dziĊki zastosowaniu
układów synchronizacji uzyskuje siĊ poprawĊ rzĊdu 1 dB w
stosunku do systemu bez układu synchronizacji w obecnoĞci
błĊdu fazy noĞnej, co ma duĪe znaczenie w przypadku małych
wartoĞci SNR. Badania pokazały równieĪ, Īe zastosowane
układy synchronizacji fazy wykazują zbliĪone wyniki przy
róĪnych warunkach transmisji.
Zestawienie rezultatów symulacyjnych dla zaimplementowanych algorytmów i ich porównanie prowadzi do wniosku,
Īe algorytm Zhang-Burra pozwala uzyskaü lepsze wyniki od
algorytmu Oha-Cheuna dla systemów ze zmiennym błĊdem
fazy, co czĊsto zachodzi w rzeczywistych warunkach pracy.
Poza tym algorytm Zhang-Burra, dziĊki stablicowaniu
zmiennych niezbĊdnych do wyznaczenia estymaty fazy, nie
zwiĊksza znacząco złoĪonoĞci obliczeniowej odbiornika.
Przeprowadzone badania potwierdziły, Īe rozpatrywane
algorytmy turbo-synchronizacji stanowią dobrą technikĊ
synchronizacji fazy w cyfrowych systemach telekomunikacyjnych stosujących turbo-kodowanie.
LITERATURA
[1]
[2]
[3]
Rys. 10. Porównanie jakoĞci działania algorytmów Oha-Cheuna i ZhangBurra w przypadku zmieniającego siĊ liniowo błĊdu fazy (N = 400, R = 1/2,
Nit = 5, g = (7, 5)8, Log-MAP).
Z analizy przedstawionych wyników badaĔ oraz
porównania działania algorytmów wydaje siĊ, Īe algorytm
Zhang-Burra jest korzystniejszym niĪ algorytm Oha-Cheuna
do zastosowania w cyfrowych systemach telekomunikacyjnych z turbo-kodowaniem.
[4]
[5]
Berrou, C., Glavieux, A., Thitimajshima, P., “Near Shannon limit errorcorrecting coding and decoding: turbo-codes,” Proc. IEEE Int. Conf. on
Communications (ICC’93), Geneva, Switzerland, May 1993, 1064-1070.
Berrou, C. Glavieux, A., “Near optimum error correcting coding and
decoding: turbo-codes” IEEE Trans. on Communications, vol. 44, No.
10, October 1996, 1261-1271.
Shannon C. E.: “A mathematical theory of communication”, The Bell
System Technical Journal, vol. 27, July/October 1948, 379-423, 623656.
Oh W., Cheun K.: “Joint decoding and carrier phase recovery
algorithm for turbo codes,” IEEE Commun. Lett., vol. 5, No. 9, Sep.
2001, 375–377.
Zhang L.,. Burr A. G.: “Iterative carrier phase recovery suited to turbocoded systems,” IEEE Trans. Wireless Commun., vol. 3, no. 6, Nov.
2004, 2267–2276.
VI. PODSUMOWANIE
W artykule zaprezentowano dwie iteracyjne metody
synchronizacji fazy, opracowane dla systemów korzystających
z techniki turbo-kodowania: algorytm Oha-Cheuna i algorytm
Zhang-Burra. Omówiono zasadĊ działania kaĪdego z tych
algorytmów oraz przedstawiono praktyczne podejĞcie do ich
realizacji. Przeprowadzone zostały równieĪ komputerowe
badania symulacyjne tych algorytmów przy wykorzystaniu
programu napisanego w jĊzyku Matlab.
Wyniki badaĔ pokazały, Īe za pomocą obu tych układów
synchronizacji moĪliwa jest w warunkach niskiego stosunku
sygnału do szumu, estymacja i korekcja błĊdu fazy znacząco
12
PWT 2012 - Poznań, 14 grudnia 2012
2012
Implementation of the Static Round-Robin
Dispatching Scheme in the MSM and SMM
Clos-Network Switch
Janusz Kleban, Member, IEEE
Abstract—In this paper, results obtained under the Static
Round-Robin (SRR) technique implemented in Memory-SpaceMemory (MSM) and Space-Memory-Memory (SMM) Clos
switching fabric were analyzed and compared. The SRR
technique was first proposed for the single-stage switching fabric
(crossbar). The intuition behind this packet dispatching scheme is
to desynchronize the pointers of the arbiters in a static way to
considerably improve the delay performance. The implementation
of this technique in the MSM Clos-network switches is known as
the Static Round-Robin Dispatching (SRRD) arbitration
algorithm. We contribute to this research area by implementing
the SRR idea in the SMM Clos switching fabric made of the
Crosspoint Queued (CQ) switches with virtual crosspoint queues
(VCQs). The performance of the MSM and SMM Clos-network
switches under the SRR technique and uniform and nonuniform
traffic distribution patterns has been evaluated using computer
simulation. Three main performance measures were investigated:
cell delay, throughput and input buffers size. Selected results are
shown in this paper.
Index Terms—Clos-network, Dispatching Algorithm, Packet
Switching, Packet Scheduling.
I. INTRODUCTION
T
HE architecture of switching fabric for high-performance
switches/routers is still a big challenge, for both researches
and equipment vendors. A switching fabric is responsible for
transmitting packets from the ingress side to the egress side of
a switch, and may be organized as a single-stage (crossbar) or
multiple-stage, such as the Clos switching fabric [1]. The
crossbar fabric is not scalable, so the Clos-network
architecture is a very attractive solution for constructing a
multiple-stage switching fabric made of smaller-size switching
elements, where each such element is usually a crossbar.
Currently, network equipment vendors such as Cisco, Juniper,
and Brocade use this switching fabric to build core routers:
CRS series, T series, and BigIron RX Series, respectively.
Taking into consideration buffer allocation schemes Closnetwork packet switches may be classified as: Space-SpaceJanusz Kleban is with the Chair of Telecommunication and Computer
Networks,
Faculty
of
Electronics
and
Telecommunications,
Poznan
University
of
Technology,
Poznan,
Poland
(e-mail: [email protected]).
Space (SSS or S3), Memory-Memory-Memory (MMM),
Memory-Space-Memory (MSM), Space-Memory-Memory
(SMM), and Space-Space-Memory (SSM) switches [2]. Two
of these architectures, namely the MSM and SMM are
investigated in this paper.
The MSM Clos switching fabric is made from nonblocking
crossbars as switching modules. This architecture uses
bufferless modules in the second stage, but it has buffers in the
first and third stage. Since the architecture has no buffer in the
second-stage modules, the out-of-sequence problem is
eliminated but how to dispatch cells from the first stage to the
second stage becomes an important issue. The cells are fixsized packets obtained by segmentation of variable-length
packets arriving to the ingress line cards of a large-capacity
switch. Packets are assembled from cells at the egress line
cards, before they depart. Cells are transmitted through the
switching fabric during a time slot. Since buffers are allocated
in the first stage the Virtual Output Queuing (VOQ)
mechanism may be implemented very easy, to avoid the HeadOf-Line (HOL) blocking phenomenon [2]. Internal blocking
and output port contention problems in the MSM Closnetwork switches must be solved by fast arbitration schemes.
The well known dispatching schemes for buffered Closnetwork switches were proposed in [2-5]. The basic idea of
these algorithms is to use the effect of desynchronization of
arbitration pointers and common request-grant-accept
handshaking routine. Most of these schemes can achieve 100%
throughput under the uniform traffic, but under the nonuniform
traffic the throughput is usually reduced. A switch can achieve
100% throughput under the uniform or nonuniform traffic if
the switch is stable, as it was defined in [6].
The SMM architecture was proposed in [7]. This
architecture employs bufferless modules in the first stage, and
output-queued modules in the central and output stage. Since
the input stage is bufferless the SRR mechanism may be used
to dispatch cells to the second stage. We propose to replace
the output-queued modules in the central and output stage with
the CQ switches with VCQs.
In this paper, implementation of the SRR packet dispatching
scheme in the MSM and SMM Clos-network switch is
presented. Performance parameters of these two architectures
are also evaluated. The remainder of this paper is organized as
XVI Poznańskie Warsztaty Telekomunikacyjne
13
follows. Section II introduces some background knowledge
concerning the MSM and SMM Clos switching fabric. The CQ
switches with VCQs are also presented in this section. Section
III presents packet dispatching schemes. Section IV is devoted
to performance evaluation of the investigated switching fabric
architectures. The paper is concluded in Section V.
II. MSM AND SMM CLOS SWITCHING FABRIC
Clos-networks were proposed by C. Clos in [1]. The threestage Clos-network architecture is denoted by C(m, n, k),
where parameters m, n, and k entirely determine the structure
of the network. The MSM Clos switching fabric is shown in
Fig. 1. To define the architecture the terminology proposed in
[3] is used (see Table I).
IM (0)
IP (0,0)
IP (0,n-1)
CM (0)
OP (0,0)
OP (0,n-1)
VOQ(0,k-1,n-1)
CM (r)
IM (i)
IP (i,0)
IP (i,n-1)
OM (0)
VOQ(0,0,0)
OM (j)
OP (j,0)
VOQ(i,0,0)
The SMM Clos-network architecture can be defined by
analogy with the MSM architecture. To transform the MSM
architecture into SMM, the central stage must be replaced with
the input stage. The organization of buffers in the switching
modules must be also changed, because the central stage in the
SMM switching fabric uses only output buffers. The VOQs are
not used this architecture.
Currently, the implementation of crosspoint queued
switching fabric with large crosspoint buffers has become
feasible, because the technology for VLSI chip manufacture
has matured. Recently, the CQ switches with VCQs (Fig. 2)
have been proposed to eliminate the large RTT (Round Trip
Time) delay between the line card and switching fabric the
CICQ (Combined Input and Crosspoint Queued) switch suffers
from [8]. This kind of switch consists of NxN crosspoint
queues and N VCQs associated with individual input ports. All
schedulers, crosspoint buffers (CPBs), and VCQs are located
inside a single chip. In the CQ switch with VCQs the capacity
of buffers to store cells was considerably increased because
cells may be buffered in both CPBs and VCQs.
OP (j,n-1)
VOQ(i,k-1,n-1)
Input 1
VCQ11
VCQ-S1
CM (m-1)
IM (k-1)
OM (k-1)
CPB
CPB
CPB
VCQ1N
OP (k-1,0)
IP (k-1,0)
CPB
VOQ(k-1,0,0)
OP (k-1,n-1)
IP (k-1,n-1)
VOQ(k-1,k-1,n-1)
LI (i, r)
LC (r, j)
Fig. 1. The MSM Clos switching fabric architecture.
Input N
VCQ-SN
TABLE I
A NOTATION FOR THE MSM CLOS SWITCHING FABRIC
Notation
Description
IM
CM
OM
m
n
k
i
j
h
r
IM(i)
CM(r)
OM(j)
IP(i, h)
OP(j, h)
LI(i, r)
LC(r, j)
VOQ(i, j, h)
Input module at the first stage
Central module at the second stage
Output module at the third stage
Number of CMs
Number of input/output ports in each IM/OM
Number of IMs/OMs
IM number, where 0 d i d k-1
OM number, where 0 d j d k-1
Input/output port number in IM/OM, where 0 d h d n-1
CM number, where 0 d r d m-1
The (i+1)th input module
The (r+1)th central module
The (j+1)th output module
The (h+1)th input port at IM(i)
The (h+1)th output port at OM(j)
Output link at IM(i) that is connected to CM(r)
Output link at CM(r) that is connected to OM(j)
Virtual output queue at IM(i) that stores cells from IM(i) to
OP(j, h)
In the MSM Clos switching fabric architecture the first stage
consists of k IMs, and each of them has an
n u m dimension and nk VOQ(i, j, h) to eliminate Head-OfLine blocking. The second stage consists of m bufferless CMs,
and each of them has a k u k dimension. The third stage
consists of k OMs of capacity m u n, where each OP(j, h) has
an output buffer. Each output buffer can receive at most m
cells from m CMs, so a memory speedup is required here.
14
VCQN1
VCQNN
CP-S1
Output 1
CP-SN
Output N
Fig. 2. CQ switch with VCQ: VCQ-S – VCQ scheduler, CP-S – CPB
scheduler.
We propose employing the CQ switches with VCQs in the
second and third stages of the SMM Clos-switching fabric to
investigate the architecture made of real switches.
III. IMPLEMENTATION OF THE SRR IN THE MSM AND SMM
CLOS SWITCHING FABRIC
The SRR scheme was first proposed for crossbar switches.
The basic idea of this mechanism is to desynchronize the
arbitration pointers at the beginning and to update the pointers
in a static way according to the round-robin routine (Fig. 3).
The pointers are kept desynchronized all the time. Since the
SRR algorithm is very simple it may be implemented easily in
hardware.
Implementation of the SRR in the MSM Clos-network
switches has been proposed in [4], and is called SRRD
scheme. The SRRD scheme is the same as the CMSD
(Concurrent Master-Slave Round-Robin Dispatching) [3]
algorithm except the routine concerning updating the values of
round robin pointers. The CMSD algorithm uses hierarchical
PWT 2012 - Poznań, 14 grudnia 2012
round-robin arbitration during iterative matching within the
IMs. Two types of round-robin arbiters (master and slave) are
necessary to perform arbitration process in the first stage of
Clos-network switch.
Fig. 3. Matching sequence in the SRR.
Within the SRRD scheme the pointers are set up as follows:
PV(i, j, h) = h, PSL(i, j, r) = r, PML(i, r) = (i+r) % k, PC(r, j) = i
if PML(i, r) = j. In each time slot pointer PML(i, r) and PC(r, j)
are always incremented by 1 (mod k), and pointer PSL(i, j, r)
and PV(i, j, h) remain unchanged, no matter there is a match or
not. PV(i, j, h) – represents the pointer of the VOQ(i, j, h)
arbiter. Pointer PML(i, r) is assigned to the master arbiter in
LI(i, r), whereas PSL(i, j, r) works as the pointer of the jth slave
arbiter in LI(i, r). Pointer PC(r, j) works for the arbiter in
LC(r, j). The initial configuration of the round-robin pointers is
set to preserve the paths as shown in Fig. 3. To improve the
performance of the SRRD scheme under nonuniform traffic
rotation of the search directions of round-robin arbiters was
proposed. Now, some round-robin arbiters are allowed to
search the requests in clockwise direction and anti-clockwise
direction alternatively, each for one time slot. The track of
time is kept by a 0/1 counter, which will increment by one
(mod 2) in each time slot. Therefore, if the counter=0, the
master arbiter in LI(i, r) searches one request in clockwise
round-robin fashion, and the other way round if the counter=1.
The implementation of the SRR mechanism in the SMM
Clos-network switches is much simpler. Since the first stage of
the SMM switching fabric is bufferless no schedulers are
needed. In this case the input stage is used to distribute cells to
buffers located in the central stage. We propose to shift the
static connection patterns in IMs under the influence of the
backpressure signal (BP), sent by any CM module. The BP
signal is transmitted to IMs controller when the cell, sent to
any central module exceeds the value of the assumed threshold
representing the number of cells waiting in VCQs. The
algorithm works as follows:
Phase 1: Cells forwarding between IM and CM
o Step 1: Each cell arriving to IP(i, h) is transmitted,
according to the static connection pattern of IM(i), to
LI(i, r), and next to CM(r).
o Step 2: Each cell arriving to CM(r) is stored in proper
VCQ(r, i, j) buffer; e.g. the cell destined to OM(3) is
buffered in the VCQ(r, i, 3) buffer. This buffer is
associated with CPB(r, i, 3) buffer. All CPB buffers with
j=3 are associated with LC(r, 3) output link. Cells buffered
in these buffers will be sent by this link to OM(3).
o Step 3: If arriving cell exceeds the value of assumed
threshold send the BP signal to IMs controller to change
the connection pattern used in IMs.
o Step 4: If there is a room in CPB(r, i, j) buffers, the VCQ-S
arbiters transfer cells from VCQ(r, i, j) to CPB(r, i, j).
Phase 2: Cells forwarding between CM and OM
o Step 1: Each CP-S arbiter searches CPB(r, i, j) buffers with
the same value of j, and selects cell to be sent to the OM(j).
The round-robin arbitration is employed for this selection.
o Step 2: All selected cells in step 1 are sent to OMs modules
and stored in proper VCQ(j, i, h) buffers related to
OP(j, h).
o Step 3: If there is a room in CPB(j, i, h) buffers, the VCQS arbiters transfer cells from VCQ(j, i, h) to CPB(j, i, h).
o Step 4: Each CP-S round-robin arbiter searches CPB(j, i, h)
buffers with the same value of h, and selects cell to be sent
to the OP(j, h).
IV. SIMULATION EXPERIMENTS
Two packet arrival models are considered in simulation
experiments: the Bernoulli arrival model and the bursty traffic
model. The probability that a cell may arrive in a time slot is
denoted by p and is referred to as the load of the input. In the
bursty traffic model, each input alternates between active and
idle periods. During active periods, cells destined for the same
output arrive continuously in consecutive time slots. The
average burst (active period) length is set to 16 cells.
Several traffic distribution models (the most popular in this
research area) have been considered, which determine the
probability pij that a cell, which arrives at an input, will be
directed to a certain output. The considered traffic models are:
uniform, Chang’s, bidiagonal and transdiagonal [9].
The experiments have been carried out for the MSM and
SMM Clos-switching fabric of size 64 u 64. There are no
capacity limited buffers in the MSM as well as in the SMM
Clos switching fabric. In each simulation experiment traffic
load per input port, from p = 0,05 to p = 1, with the step 0.05,
was considered. The 95% confidence intervals have been
calculated after t-student distribution for five series with
200,000 cycles, and are at least one order lower than the mean
value of the simulation results, so they are not shown in the
figures. The starting phase comprised 50,000 time slots,
enabled to reach the stable state of the switching fabric. The
threshold for BP signals was set to 15 cells. One and four
iterations were considered for the MSM switch. Average cell
delay in time slots, maximum size of queues in CMs and OMs
in the SMM switching fabric, and throughput have been
evaluated. Owing to the space limits, only selected results for
the average cell delay will be presented in this paper (Fig. 4–
Fig. 8).
Fig. 4 shows the average cell delay for uniform traffic and
Bernoulli arrival model. The SMM architecture produces
better results for high input load p>0.95. For lower input load
the differences between both architectures are very small.
Analogous trends, in the average cell delay can be observed
for Chang’s traffic (Fig. 5). Fig. 6 shows average cell delay
under the bidiagonal traffic. In this case, the throughput of
both architectures is limited to 95% and 90% for the MSM
XVI Poznańskie Warsztaty Telekomunikacyjne
15
1000
MSM, SRRD, 1 itr
Average cell delay (time slots)
switch and SMM switch respectively. Simulation results have
shown that the MSM architecture under the SRR mechanism is
not able to support the transdiagonal traffic. The achievable
throughput is equal only to 70% for four iterations. The SMM
architecture behaves in a different way. The throughput under
the transdiagonal traffic is equal to 100%, and average cell
delay is about 140. The results for bursty model and uniform
traffic are shown in Fig. 8. The MSM switch produces slightly
better results than the SMM switch. The maximum queue size
(total VCQ and CPB) in CMs modules in the SMM switch was
about 50 for uniform and Chang’s traffic, 180 for
transdiagonal traffic, 4000 for bidiagonal traffic and 500 for
bursty model. A little bit longer queues were observed in OMs.
SMM, BP
SMM, no BP
100
10
1
0,00
0,10
0,20
0,30
0,40
0,50
0,60
0,70
0,80
0,90
1,00
Input load
Fig. 7. Average cell delay, transdiagonal traffic.
1000
MSM, SRRD, 1 itr
MSM, SRRD, 4 itr
SMM, BP
SMM, no BP
Average cell delay (time slots)
1000
Average cell delay (time slots)
MSM, SRRD, 4 itr
100
10
MSM, SRRD, 1 itr
MSM, SRRD, 4 itr
SMM, BP
SMM, no BP
100
10
1
0,00
1
0,00
0,10
0,20
0,30
0,40
0,50
0,60
0,70
0,80
0,90
0,10
1,00
Input load
Average cell delay (time slots)
0,40
0,50
0,60
0,70
0,80
0,90
1,00
Fig. 8. Average cell delay, uniform traffic, bursty model.
V. CONCLUSION
1000
MSM, SRRD, 1itr
MSM, SRRD, 4 itr
SMM, BP
SMM, no BP
Simulation results for the MSM and SMM Clos-network
switches under the SRR scheme are presented in this paper. To
our knowledge, there are first results obtained for the SMM
architecture made of CQ switches with VCQ. Proposed
solution is implementable in hardware due to its simplicity.
100
10
REFERENCES
[1]
[2]
1
0,00
0,10
0,20
0,30
0,40
0,50
Input load
0,60
0,70
0,80
0,90
1,00
[3]
Fig. 5. Average cell delay, Chang’s traffic
[4]
1000
MSM, SRRD, 1 itr
Average cell delay (time slots)
0,30
Input load
Fig. 4. Average cell delay, uniform traffic.
[5]
MSM, SRRD, 4 itr
SMM, BP
SMM no BP
100
[6]
[7]
10
[8]
1
0,00
0,10
0,20
0,30
0,40
0,50
0,60
0,70
0,80
Input load
Fig. 6. Average cell delay, bidiagonal traffic.
16
0,20
0,90
1,00
[9]
C. Clos, “A study of non-blocking switching networks”, Bell Sys. Tech.
Jour., 1953, pp. 406-424.
H. J. Chao, B. Liu, High performance switches and routers, Wiley
Interscience, New Jersey, 2007.
E. Oki, Z. Jing, R. Rojas-Cessa, and H. J. Chao: “Concurrent roundrobin-based dispatching schemes for Clos-network switches”,
IEEE/ACM Trans. on Networking, vol. 10, no.6, 2002, pp. 830-844.
K. Pun and M. Hamdi, “Dispatching schemes for Clos-network
switches”, Computer Networks, no. 44, pp. 667–679, 2004.
J. Kleban and A. Wieczorek, “CRRD-OG – a packet dispatching
algorithm with open grants for three-stage buffered Clos-network
switches”, in Proc. High Performance Switching and Routing 2006 –
HPSR 2006, pp. 315–320.
N. McKeown, A. Mekkittikul, V. Anantharam, J. Walrand, “Achieving
100% throughput in an input-queued switch”, IEEE Trans. Commun.,
pp. 1260-1267, Aug. 1999.
X. Li, Z. Zhou, and M. Hamdi, “Space-Memory-Memory architecture
for Clos-network packet switches”, Proc. IEEE International Conference
on Communications – ICC 2005, May 2005, vol. 2, pp. 1031-1035.
K. Yoshigoe, The Crosspoint-Queued Switches with Virtual Crosspoint
Queueing, Proc. 5th International Conference on Signal Processing and
Communication Systems, ICSPCS 2011, Dec. 2011. pp. 277-281.
J. Kleban: Packet Dispatching Scheme Employing Distributed Arbiters
for Modified MSM Clos Switching Fabric, Poznańskie Warsztaty
Telekomunikacyjne '2011, 09 grudnia 2011, Poznań, pp.43-47.
PWT 2012 - Poznań, 14 grudnia 2012
2012
Pomiar właściwości spektralnych optycznej
krotnicy przeplotowej
Jan Lamperski
Streszczenie— W prezentowanej pracy omówiono problemy
pomiarowe oraz przedstawiono wyniki pomiaru właściwości
optycznej krotnicy przeplotowej zrealizowanej w konfiguracji
interferometru Macha-Zehndera przeznaczonej dla selekcji
kanałów o gigahercowych odstępach.
Słowa kluczowe — Optyczne
interferometr Macha-Zehndera.
krotnice
przeplotowe,
I. WSTĘP
W
pracy skoncentrowano się na pomiarze właściwości
spektralnych nietypowych krotnic przeplotowych
przeznaczonych dla systemów UDWDM, których odstępy
międzykanałowe wynoszą od kilku do kilkunastu GHz.
Planuje się, że krotnice będą współpracowały z opracowanymi
w laboratorium wielofalowymi źródłami optycznymi. Źródła
te umożliwiają uzyskanie odstępów międzykanałowych od 1
do ok. 3 GHz. Zwiększenie odstępu pomiędzy optycznymi
częstotliwościami nośnymi do kilkunastu GHz możliwe jest
dzięki separacji kanałów za pomocą krotnic przeplotowych.
Dla systemów UDWDM zaproponowano zastosowanie
krotnic, których konstrukcja bazuje na konfiguracji
światłowodowego interferometru Macha-Zehndera (Rys. 1).
ͳ
ሾܶଶ ሺߥሻሿ ൌ ൤
Ͳ
gdzie: W – różnica czasów propagacji w ramionach
interferometru, Q – częstotliwość optyczna.
Charakterystykę transmitancyjną krotnicy otrzymujemy
jako iloczyn macierzy oddzielnych elementów:
ൣܶ௜௝̴ெ௓ ሺߥሻ൧ ൌ
ൌ
ͳ
ͳ
൤
െ݆
ξʹ
ͳ
ͳ
െ݆
ξʹ
൤
െ݆
൨
ͳ
oraz ramion (linii opóźniających):
Jan
Lamperski,
Katedra
Systemów
Telekomunikacyjnych
i
Optoelektroniki, Politechnika Poznańska, ul. Polanka 3, 60-965 Poznań, (Tel.:
+48-61-6653809; e-mail: [email protected]).
ͳ ͳ
Ͳ
൨
൤
݁‫݌ݔ‬ሺെʹߨ݆ߥ߬ሻ ξʹ െ݆
െ݆
൨
ͳ
Transmitancję mocy optycznej opisuje zależność:
ଶ
ቂหܶ௜௝̴ெ௓ ሺߥሻห ቃ ൌ ൤
ܿ‫ ݏ݋‬ଶ ሺߨߥ߬ሻ
‫݊݅ݏ‬ଶ ሺߨߥ߬ሻ
‫݊݅ݏ‬ଶ ሺߨߥ߬ሻ
൨
ܿ‫ ݏ݋‬ଶ ሺߨߥ߬ሻ
Zwykle sygnał wejściowy doprowadzony jest do jednego
wejścia i wtedy:
ܲଵ̴ௐ௒
ܿ‫ ݏ݋‬ଶ ሺߨߥ߬ሻ
൨ ൈ ܲଵ̴ௐா
൨ൌ൤ ଶ
ܲଶ̴ௐ௒
‫ ݊݅ݏ‬ሺߨߥ߬ሻ
Powyższe równanie pokazuje, że transmitancja mocy jest
funkcję okresową z okresem równym 1/W.
Krotnica przeplotowa zbudowana w oparciu o opisany
układ będzie rozdzielała nieparzyste i parzyste częstotliwości
nośne WDM gdy odstępy między nimi będą wynosiły:
Krotnica składa się z dwóch sprzęgaczy kierunkowych, o
transmitancji wyrażonej równaniem:
ሾܶଵ ሿ ൌ
െ݆ ͳ
൨൤
ͳ Ͳ
ͳ ͳ െ ݁‫݌ݔ‬ሺെʹߨ݆ߥ߬ሻ
െ݆ሾͳ ൅ ݁‫݌ݔ‬ሺെʹߨ݆ߥ߬ሻሿ
൤
൨
ʹ െ݆ሾͳ ൅ ݁‫݌ݔ‬ሺെʹߨ݆ߥ߬ሻሿ െሾͳ െ ݁‫݌ݔ‬ሺെʹߨ݆ߥ߬ሻሿ
൤
Rys. 1. Krotnica przeplotowa w konfiguracji interferometru Mach-Zehndera.
Ͳ
൨
݁‫݌ݔ‬ሺെʹߨ݆ߥ߬ሻ
ߜߥ ൌ
ܿ
ͳ
ൌ
ʹ߬ ʹ݊ߜ݈
gdzie:
c – prędkość światła w próżni,
n – współczynnik załamania włókna,
Gl – różnica długości ramion interferometru MachaZehndera.
Pomiary spektralne filtrów optycznych, krotnic falowych
oraz optycznych krotnic
transferowych
najczęściej
realizowane są z wykorzystaniem analizatorów widma
stosując jako źródło światła przestrajalne lasery lub
szerokopasmowe źródła ASE.
XVI Poznańskie Warsztaty Telekomunikacyjne
17
Podjęte badania miały na celu opracowanie metodyki
pomiaru oryginalnych konstrukcji optycznych krotnic
przeplotowych przeznaczonych do systemów z ultra gęstym
zwielokrotnieniem falowym.
II. POMIAR KROTNIC PRZEPLOTOWYCH PRZEZNACZONYCH
DLA SYSTEMÓW Z GIGAHERCOWYMI ODSTĘPAMI POMIĘDZY
OPTYCZNYMI CZĘSTOTLIWOŚCIAMI NOŚNYMI
Rozdzielczość klasycznych analizatorów widma nie jest
wystarczająca do realizacji pomiarów krotnic przeznaczonych
dla systemów, w których odstępy pomiędzy optycznymi
częstotliwościami nośnymi wynoszą kilka GHz (Rys. 2) [2].
precyzyjnym określeniu bezwzględnych i względnych
wartości częstotliwości. Problem określenia względnej skali
częstotliwości spowodowany jest, korzystną z innych
względów, możliwością regulacji w szerokim zakresie
parametru FSR.
W związku z tym w celu wyznaczenia parametrów krotnicy,
a w szczególności odległości pomiędzy zerami transmitancji
zastosowano metodę tradycyjną wykorzystującą precyzyjny
przestrajalny laser.
Parametry najlepszych dostępnych laserów nie są
wystarczające do zrealizowania założonych pomiarów.
W szczególności:
- względna dokładność długości fali wynosi ±0,035 nm,
- rozdzielczość wynosi 0,001nm (125 MHz),
- powtarzalność ustawienia długości fali wynosi: ±0,035
nm.
Niektóre z problemów, związanych z ograniczoną
rozdzielczością, powtarzalnością ustawień długości fal oraz
stabilnością ilustrują poniższe rysunki.
Rys. 4 ilustruje problem nierównomierności skali długości
fal.
Rys. 2.
Rezultaty próby pomiaru krotnicy UDWDM klasycznym
analizatorem widma.
W związku z tym w proponowanych wcześniej pomiarach
[2] wykorzystano własną konstrukcję analizatora widma
wysokiej rozdzielczości wykorzystującego przestrajalny
elaton Fabry-Perot.
Przykładowe wyniki pomiarów charakterystyk spektralnych
światłowodowej krotnicy M-Z pokazano na Rys. 3 [2].
Rys. 4. Charakterystyka spektralna krotnicy, wyjścia A i B, problem
nierównomierności przestrajania długości fali.
Na Rys. 5 pokazano problem
powtarzalności dostrajania długości fali.
Rys. 3. Charakterystyka krotnicy w szerokim zakresie częstotliwości.
Jak widać zbudowany analizator posiada wystarczającą
rozdzielczość jednakże jego wadą jest trudność w
18
dokładności
oraz
Rys. 5. Charakterystyka spektralna krotnicy, wyjścia A i B, problem
dokładności oraz powtarzalności dostrajania długości fali.
PWT 2012 - Poznań, 14 grudnia 2012
Na Rys. 6 pokazano charakterystyki
uśrednieniu 4 cykli pomiarowych.
krotnicy
po
Rys. 5. Charakterystyka spektralna krotnicy, wyjścia A i B, wyniki
uśrednione po czterech cyklach pomiarów.
Problem precyzji oraz powtarzalności dostrajania długości
fali lasera przestrajalnego może być rozwiązany stosując
dodatkowy miernik długości fal.
Stosując układ pomiarowy zawierający przestrajalny laser,
miernik mocy optycznej oraz miernik długości fal, a także
stosując procedurę uśredniania w szerokim zakresie długości
fal oraz dopasowania wyników pomiarowych do znanej
funkcji transmitancji uzyskano dokładność określenia FSR
krotnicy na poziomie ok. 10 MHz. Dokładność potwierdzono
porównując wyniki z danymi katalogowymi wzorcowego
interferometru.
III. WNIOSKI
Uzyskane wyniki potwierdzają możliwość pomiaru i
precyzyjnego zestrojenia budowanych krotnic przeplotowych
przeznaczonych do zastosowania w systemach UDWDM.
LITERATURA
[1]
[2]
K. Perlicki, Pomiary w optycznych systemach telekomunikacyjnych, WKŁ,
2002.
Jan Lamperski, Problemy pomiaru charakterystyki spektralnej optycznej
krotnicy przeplotowej UDWDM, XV Poznań Telecommunications
Workshop, 9 December 2011, s 86-87
XVI Poznańskie Warsztaty Telekomunikacyjne
19
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Metoda precyzyjnego pomiaru FSR rezonatora
optycznego
Jan Lamperski, Zofia Planner
Streszczenie— W pracy przedstawiono ideę działania
rezonatora Fabry-Perot wraz z najważniejszymi jego
parametrami. Zaproponowano oryginalną metodę pomiaru FSR.
FSR
Pojęcia —rezonator Fabry-Perot, analizator widma .
I. WSTĘP
R
ezonator optyczny typu Fabry-Perota(etalon) składa się z
dwóch płaskich zwierciadeł , umieszczonych równolegle
względem siebie, między którymi znajduje się płasko
równoległa warstwa powietrza.
Odległość pomiędzy
zwierciadłami wynosi L. Fale przebiegając przez rezonator
tworzą fale stojące , dla których natężenie pola elektrycznego
na powierzchniach zwierciadeł jest w przybliżeniu równe
zeru. Zatem możliwe częstotliwości rezonansowe są
uwarunkowane odległością między zwierciadłami, która
powinna być równa całkowitej wielokrotności połowy
długości fali. Częstotliwości własne rezonatora różnią się o ten
sam przedział częstotliwości.
Wzór na charakterystykę transmitancyjną filtru Fabry-Perot
jest przedstawiony poniżej:
TFP
It
Ii
ͳ
ͳ
ͶR /(ͳ
(1)
R ) ʹ sin ʹ (δ / ʹ)
Wynik ten jest znany jako suma Airy. Przyjmuje ona
wartość
maksymalną
i
równą
1,
gdy
ʹ
.
sin (Ɂ / ʹ) Ͳ Ɂ / ʹ mɎ
gdzie:
It,Ii – natężenie fali przetransmitowanej i padającej;
ͶɎnL cos Ʌ jest różnicą fazy między dwoma kolejnymi
Ɂ
ɉ
przesłanymi falami, kąt θ jest kątem propagacji we wnęce;
n jest współczynnikiem załamania ośrodka wnęki;
R- jest współczynnikiem odbicia zwierciadeł.
Iloraz [4R/(1-R)2] nazywa się współczynnikiem finezji lub
współczynnikiem kontrastu.
Ważnym parametrem rezonatora jest odstęp między
sąsiednimi częstotliwościami rezonansowymi. Jest on okresem
charakterystyki transmitancji i oznacza się symbolem FSR
Jan Lamperski, Politechnika Poznańska, Poznań, Polska
[email protected] );
Zofia Planner, Politechnika Poznańska, Poznań, Polska
[email protected]).
20
(free spectral range). Jest to zakres, w którym można
przestrajać filtr nie pokrywając sąsiednich pasm
przepustowych :
(email:
(email:
ɋm
ͳ
ɋm
c .
ʹnL
(2)
Kolejnym parametrem jest szerokość pasma 3dB rezonatora
Fabry-Perot (full width at half maximum FWHM). Jest to
odstęp między dwoma najbliższymi częstotliwościami dla
których transmisja spada do połowy wartości szczytowej.
Wyraża się ona wzorem:
ȟɋ͵dB
c ͳ R.
ʹnL R
(3)
Dla filtru Fabry-Perot definiuje się także parametr zwany
finezją F. Jest to iloraz FSR do szerokości pasma 3dB i
przybliża liczbę kanałów jaką może wyselekcjonować filtr.
Finezja jest miarą rozdzielczości etalonu. Fizycznym
ograniczeniem finezji są absorpcja i rozpraszanie zwierciadeł
oraz fakt wielokrotnych przejść między zwierciadłami.
Zwierciadła rzadko osiągają odbicie równe 99,7%. Typowo,
interferometr Fabry-Perot ze zwierciadłami płaskimi ma
98,4%).
wartość finezji 100 200 ( R = 96,9%
F
FSR
ȟɋ FWHM
Ɏ R .
ͳ R
(4)
Rezonator zbudowany ze zwierciadeł sferycznych
rozwiązuje problem występowania przy każdym przejściu fali
strat w wyniku ugięcia na krawędziach zwierciadeł płasko
równoległych, a także jest prostszy w realizacji. Na Rys. 1
pokazano układ rezonatora z wnęką konfokalną i
niekonfokalną. Niestety przy zastosowaniu zwierciadeł
sferycznych geometria rezonatora nie może być dowolnie
wybrana. Kształt rezonatora ogranicza tzw. warunek
stabilności:
L
L
(5)
) ͳ,
)(ͳ
Rʹ
Rͳ
gdzie R1 i R2 oznaczają promienie krzywizny zwierciadeł.
Spełnienie tego warunku gwarantuje małe straty dyfrakcyjne
modów rezonatora. Rezonator konfokalny (współogniskowy)
jak i niekonfokalny spełnia warunek stabilności.
Ͳ
(ͳ
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
Rys. 1. Rezonator konfokalny i niekonfokalny.
II. OPTYCZNY ANALIZATOR WIDMA WYSOKIEJ
ROZDZIELCZOŚCI
A. Opis
Rezonator Fabry-Perot wykorzystywany jest do analizy
widmowej ze względu na możliwości przestrajania długości
wnęki rezonansowej (o pół długości fali) oraz odpowiedniej
finezji. Kluczem do zrozumienia istoty jego działania jest
znajomość rozkładu modów rezonatora, który zależy od
geometrii rezonatora, szczególnie zwierciadeł. Struktura
modowa wnęki składa się z modów poprzecznych (drgania
własne poprzeczne rezonatora różnią się swoim rozkładem
prostopadle do kierunku propagacji) i z modów podłużnych
(drgania własne różnią się wzdłuż kierunku propagacji).
Rozkład modowy dla rezonatora sferycznego jest określony
wzorem:
f qmn
c
q
ʹL
(m
n
ͳ)
cos
ͳ
(
gͳg ʹ )
Ɏ
,
(6)
gdzie:
c/2L jest odległością między modami wzdłużnymi;
m, n – rząd modów poprzecznych, liczby całkowite;
q – numer modu podłużnego;
gi = 1-L/Ri , i = 1,2;
Ri – promień krzywizny i-tego zwierciadła.
Stosowany w laboratorium rezonator posiada konfigurację
niekonfokalną. Ma on dwa identyczne zwierciadła, których
promień krzywizny spełnia nierówność Ri > L , a zatem gi
(0;1). Promień ten wynosi 300 mm. Rezonator jest
zaprojektowany dla środkowej długości fali λ = 1550nm i
długość jego wnęki wynosi 25,4 mm. Odległość między
modami poprzecznymi określa wzór:
ȟf
c cos ͳ (g )
ʹL
Ɏ
.
(7)
Analizator charakteryzuje się następującymi parametrami:
- FSR =6±0,3 GHz,
- odległość między modami poprzecznymi = 780±40 MHz,
- ∆ν3dB < 1,2 MHz,
- finezja F > 5000 ,
- efektywność > 50%,
- straty wtrącenia < 3 dB.
Dla rezonatora podaje się iloraz odległości między modami
poprzecznymi do odległości między modami podłużnymi i
2
nazywa się go ułamkowym podziałem modów (z ang.
fractional mode splitting FMS). Analizator F-P wysokiej
rozdzielczości ma współczynnik FMS równy 0,131.
W analizatorze zastosowane są zwierciadła typu
Supermirror, które dzięki zaawansowanej technologii
nanoszenia warstw wiązką jonów charakteryzują się bardzo
małymi stratami, wysokim współczynnikiem odbicia 99,99% i
poziomem finezji rzędu 30000. Zwierciadła te składają się z
wielu warstw dielektrycznych i zaprojektowane są na
konkretną długość fali. Stąd też dany rezonator
zaprojektowany jest na pewien zakres długości fal przy
którym finezja nie spadnie poniżej pewnej gwarantowanej
wartości. Finezja związana jest ze stratami A i
współczynnikiem transmisji T w sposób następujący:
F
Ɏ
T
A
;
oraz
R
T
A
ͳ
.
(8)
Ze względu na swoje wyjątkowe właściwości analizator
znajduje zastosowanie jako wąskopasmowy filtr optyczny, do
stabilizacji pracy lasera, w diagnostyce linii widmowych
lasera, w pomiarach przesunięcia dopplerowskiego a także
spektroskopii modulacji.
Zasada działania rezonatora jako analizatora widma polega
na ciągłej, periodycznej zmianie długości wnęki. Zmiana jest
dokonywana przy pomocy piezoprzesuwu. Sygnałem
sterującym piezoprzesuw jest napięcie piłokształtne
100V±10% o częstotliwości 30 Hz. Zakres przestrajania
rezonatora można zmieniać poprzez regulację amplitudy
napięcia piłokształtnego.
Prawidłowość analizy spektralnej opiera się na regułach:
1. Częstotliwość maksymalnej transmisji przestraja się w
zakresie FSR poprzez zmianę długości wnęki o
półdługości fali, czyli 775nm w próżni. Całkowita
długość skanowania jest ograniczona do dl<(λ/2n).
2. Jeżeli szerokość piku transmisji jest mała w
porównaniu ze szczegółami badanej wiązki padającej,
na wyjściu interferometru na ekranie oscyloskopu
powstanie replika profilu widma analizowanego
sygnału.
3. Szerokość spektralna badanego sygnału musi być
mniejsza od odległości międzymodowych rezonatora.
Rozdzielczość rezonatora ogranicza szerokość piku
transmisyjnego .
B. Przykład
Analizie widmowej poddano laser helowo-neonowy LHIP0101-152 firmy Research Electro-Optics Inc. o następujących
parametrach: długość fali 1,52 μm, moc 1mW, struktura
modowa TEM00 > 99%, odległość między modami
podłużnymi 316 MHz, rozbieżność wiązki 1,43mrad oraz
średnica wiązki 1,36 mm. Wiązkę lasera podczerwonego
sprzężono z włóknem światłowodowym przy pomocy
obiektywu mikroskopowego, ponieważ analizator posiada
wejście z wyprowadzonym światłowodem. Analizator jest
zintegrowany z fotodetektorem , posiada układ stabilizacji
temperatury wnęki, a także układ elektroniczny utrzymujący
XVI Poznańskie Warsztaty Telekomunikacyjne
21
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
3
wybrany prążek w polu widzenia. Obraz ekranu oscyloskopu
pokazano na Rys. 2. Na jego podstawie i znajomości FSR
analizatora oszacowano odległość między modami
podłużnymi lasera. Otrzymano FSR wnęki lasera równy
(0,00045*6*109)/0,009 = 300 MHz.
Rys. 3. Obraz z oscyloskopu, gdy częstotliwość różnicowa zbliża się do
wartości FSR.
Moment, w którym różnica pomiędzy częstotliwościami
zrównuje się z wartością FSR odpowiada sytuacji, gdy na
ekranie oscyloskopu następuje pokrycie prążków obu laserów
(Rys. 4).
Rys. 2. Obraz z oscyloskopu.
C. Metoda precyzyjnego pomiaru FSR
Dokładność analizy widmowej realizowanej za pomocą
analizatora Fabry-Perot wysokiej rozdzielczości jest tym
większa im precyzyjniej określona jest wartość FSR.
Stosowany w laboratorium analizator posiada FSR określony
przez producenta z dokładnością do 10% (FSR=6±0,3 GHz).
Wartość ta spowodowana jest średnim rozrzutem parametrów
uzyskiwanych na etapie montażu rezonatorów. Precyzja
wartości FSR bezpośrednio przekłada się na precyzję analizy
widmowej.
W pracy zastosowano oryginalną metodę pomiaru FSR.
Najprostszy sposób określenia FSR rezonatora F-P bazuje na
pomiarze transmitancji w funkcji długości fali. Dokładność
takiego pomiaru wynika ze względnej dokładności
przestrajania źródła optycznego lub błędu pomiarowego
miernika długości fal i zwykle jest na poziomie kilkuset
megaherców. Zaproponowana metoda precyzyjnego pomiaru
FSR polega na zastosowaniu dwóch laserów tj. lasera
przestrajalnego oraz lasera pracującego na określonej długości
fali; oraz heterodynowej detekcji koherentnej w celu
dokładnego wyznaczenia częstotliwości różnicowej pomiędzy
laserami.
Początkowo lasery dostrojone były na podobne długości fal
i częstotliwość różnicowa mierzona na mikrofalowym
analizatorze widma była bliska zeru. Następnie zwiększano
odstęp częstotliwości pomiędzy laserami do nominalnej
wartości FSR obserwując prążki widmowe na ekranie
oscyloskopu (Rys. 3).
Rys. 3. Obraz z oscyloskopu, gdy częstotliwości laserów różnią się o wartość
FSR.
Potencjalnie moment ten można ustalić z dokładnością
równą rozdzielczości analizatora optycznego (1,2 MHz), w
którym rzeczywistą wartość FSR odczytujemy z analizatora
mikrofalowego.
III. WNIOSKI
Zaproponowana w pracy metoda pozwala na bardzo
precyzyjne określenie wartości FSR optycznego analizatora
widma.
LITERATURA
[1]
[2]
[3]
22
W. Brunner, W.Radloff, K.Junge , „Elektronika kwantowa”,
Wydawnictwo Naukowo Techniczne , Warszawa 1980.
SuperCavity Optical Spectrum Analyzer Instruction Manual, Newport
Corporation, Irvine, California 1991.
A. Yariv, „Optical Electronics in Modern Communications”, wyd.5,
Oxford University Press 1997.
PWT 2012 - Poznań, 14 grudnia 2012
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Coherent Data Transmission in Optical DWDM
Systems
Piotr Rydlichowski, Piotr Turowicz
Abstract—Paper presents theoretical and practical aspects of
coherent data transmission in optical DWDM transmission
systems. Recent advancements in commercial systems are
presented and future guideline outlined. Authors also present
real-life challenges and problems that can be encountered during
implementing coherent transmission technology in current
generation systems. Measurements aspects and problems are
discussed. Didactical aspects is proposed.
Index Terms—DWDM systems, coherent transmission, optical
transport systems
I. INTRODUCTION
R
ECENT advancements in electronic and optical equipment
enabled to develop new fully coherent optical data
transmission schemes. It can be utilized in DWDM
transmission systems. New transmission schemes and signal
coding allowed to achieve and standarize 40 Gbit/s, 100 Gbit/s
data rates on single optical wavelength. 400 Gbit/s and 1Tbit/s
data rates are currently under development. TELCO operators
are currently heavily investing in these technologies as it
allows to reduce operational costs and meet increasing
demands for packet transmission devices, particularly from
increasing number of mobile devices.
Authors were actively testing coherent transmission
equipment and some general conclusion were made. As this
technology matures it will allow to develop efficient design
and maintenance techniques, it will require new general view at
optical network and especially optical fibers. Networks with
old and of questionable quality fibers cannot accommodate this
new fast data transmission rates.
Along with the new DWDM transmission equipment new
measurement equipment needs to follow. New types of devices
need to be introduced to on-field testing methods. Many of
them were previously only laboratory type equipment.
Technicians also need to properly trained and aware of various
potential transmission problems.
Manuscript received November 12, 2012.
Piotr Rydlichowski is with the Institute of Bioorganic Chemistry of Polish
Academy of Sciences, Poznań Supercomputing and Networking Center, ul.
Noskowskiego
12/14,
61-704
Poznań,
Poland
(e-mail:
[email protected]).
Piotr Turowicz is with the Institute of Bioorganic Chemisty of Polish
Academy of Sciences, Poznań Supercomputing and Networking Center, ul.
Noskowskiego
12/14,
61-704
Poznań,
Poland
(e-mail:
[email protected]).
II. THEORETICAL ASPECTS OF COHERENT OPTICAL
TRANSMISSION
Optical Coherent transmission schemes share much in
common with techniques used in RF and microwave systems.
In previous ones data is transmitted using two orthogonal
carriers. Each carrier is specifically coded, transmitted and
received. Receiver uses optimal receiver concept.
In optical systems data is transmitted on two independently
polarized optical carriers. Each of them is also uniquely coded.
Advancements in electronics and photonics allowed to
precisely modulate, monitor and receive dual polarized optical
signals.
Similarly to RF and microwave signals various coding
schemes has been presented for systems with optical
transmission.
The main module in optical transponder is the transmitreceive module. Each modulation scheme is designed in such a
way that specific transmission capabilities are met. In
commercial modules economic aspects also need to be
examined and compromises has to be found. Emerging 40G
and 100+G solutions’ propositions are compromise between
assumptions mentioned above. New signal coding is built using
several basic modulation methods. In this paper focus will be
made on 40G and 100G signal coding solutions as they are
commercially available and will probably stay for long time,
especially 100G technology.
Looking at the basic signal characteristics four values can be
used to produce a modulation:
• Amplitude on/off keying (OOK)
- Non-return-to-zero (NRZ)
- Return-to-zero (RZ)
- Carrier-suppressed-RZ (CS-RZ)
- Single-sideband RZ (SSB-RZ)
• Phase-shift keying (PSK)
• Frequency-shift keying (FSK)
• Polarization-shift keying (PoSK)
Modulation format has to match specific optical system
transmission characteristics. It needs also to ensure limited
linear and non-linear impairments. Given modulation format
with a narrow optical spectrum may suite DWDM transmission
with narrow channel spacing and consequently tolerate more
chromatic distortion. Additionally, specific signals with
XVI Poznańskie Warsztaty Telekomunikacyjne
23
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
constant optical power may be less susceptible to non-linear
effects such as SPM and XPM.
A. 10G Technology
10G is today the most common used technology. It met
various requirements for WAN and LAN transmission systems.
10G transponders are available in wide range of types and
functionalities. Many of these support multiplexing and
demultiplexing of low-speed local ports (e.g. ranging from
100Mbps to 2,5 Gbps). Protocols like STM-64, OC-192,
10GbE WAN and LAN, 10G-FC, on both sides – local and
network are supported. On line side OTU2 framing is used.
Looking at the transponder functionalities, two different types
of transponders can be found: fully transparent without FEC
correction and second supporting OTH and OTU-compatible
framing (on the line side) with various FEC modes. DWDM
systems vendors use both the bit error correction that is defined
in FEC standard and proprietary solution (EFEC, GFEC).
Looking at the signal coding the most popular and efficient
way is NRZ (Non-Return to Zero). It is simple light intensity
modulation.
Below, the list of the most important parameters’ values of
10G transponders that are commonly supported by vendors is
presented:
-
laser power is less then 4dBm and it can’t be higher
because of nonlinear effects. Typical is value between
0 and 3,5 dBm
typical receiver input range is between -22 to -10 dBm and
generally higher receiver sensitivity means lower
acceptable OSNR
OSNR lever typical about 12dB but some vendors
producing cards match better with 20dB OSNR lever
CD tolerance: typical is r700 ps/nm more advanced
receivers -1000…+3400 ps/nm.
Used format coding NRZ determine the PMD tolerance at
the level of 10 ps PMD (30 ps DGD) within 1 dB penalty
maximal haul is dependent on count of regeneration points
(approx. 7) hence haul of over 2000 km
state and tunable cards are produced for 100, 50 and
25 GHz ITU grid
no onboard dispersion compensation –
only line
compensation is available (DCMs, Bragg gratings and
circulators)
Cost of 10G technology is relatively low and is still
decreasing. It will position this technology in end-client
markets where previously it was too expensive. Increase in this
client-side traffic will allow TELCO operator to move into
40G/100G technology in the backbone networks. Technology
upgrade will be cheaper than hire of more dark fibers.
However, it needs to be noted that not all of currently used
optical fibers (especially older ones) will sustain 40G/100G
transmission. CD/PMD transmission parameters are the
deciding factor.
24
2
B. 40G Technology
40G transmission is regarded as technology transition from
10G to 100G. Standardization of 40G and 100G technologies
was conducted simultaneously. 40G equipment is nowdays cost
attractive and is ideal solution for operators that do not have
technical and financial capabilities to move to 100G
technology.
At the local side, 40G transponders may support multiplexing
of various range of protocols e.g. STM-64, OC-192, 10GbE
WAN and LAN. In similar to 10G technology 40G
transponders from a functional point of view can be divided
into two different types: fully transparent without FEC
correction and supporting OTH and OTU-compatible framing
(on the line side) with a EFEC-3 modes.
Different modulation scheme for 40G guarantees
transmission reach and PMD tolerance in comparison to 10G.
RZ-DQPSK (Return-to-Zero Differential Quadrature Phase
Shift Keying) modulation format was chosen as the most
convenient and suitable for 40G transmission. Transmitter is
shown in Fig. 1. This type of modulation proved to be robust
and currently is also used for Ultra Long Haul submarine
DWDM systems. This kind of modulation format is
characterized by superior filtering tolerance. Experiments have
shown that after crossing through elements with filters i.e.
OADMs and ROADMs the signal still has good shape and is
resistant to many impairments like optical noise, CD and PMD,
optical nonlinearity. Practical implementations have shown that
this format can support 40G WDM transmission over existing
networks. In case of PMD tolerance and OSNR performance it
is possible to establish longer transmission spans with fewer
regeneration sites and increased number of ROADM nodes per
network. RZ-DQPSK modulation allowed the spectral
efficiency to increase and also to improve PMD and CD
tolerance.
However, more sophisticated transmitter design is required.
Transmitter is equipped with two phase modulator and one
intensity modulator. Moreover pulse train has similar RZ
format shape and its optical spectrum is relatively narrow. It is
applicable in 100GHz ITU grid. Using RZ-DQPSK compared
to NRZ gives 3dB RX sensitivity gain.
Fig. 1 Transmitter scheme and RZ-DQPSK modulation
diagrams.
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
The following parameters characterize examplary RZ-DQPSK
modulation transponders:
-
Tunable cards with 50GHz channel spacing
OTU-3 frame compliant
On board TDC (Tunable Dispersion Compensator)
with dispersion tuning range from -360 to 700 ps/nm
Transmit output power from -1dBm to +4dBm
Receiver input power range from -18dBm to +5dBm
PMD tolerance: 8ps DGD for 1 dB OSNR penalty
with PMDC
-
interfaces. Nowadays the vendors present the following
solutions:
-
As it was previously stated fiber PMD is a determining factor
for 40G design. PMD mainly originates from the irregularity of
the fiber due to imperfections in the manufacturing process,
and transient responses to the environment such as temperature
changes, tensions, pressure, vibrations, etc. these changes
occur both in time and different points along the fiber.
Different sections of the fiber cable could and usually have
different PMD. For currently used fibers, typical PMD value is
0,1ps/km0,5 . Older fibers have ca. 1ps/km0,5 In the newest
fibers PMD coefficient is 0.05 ps/km0,5. Influence of PMD on
transmission range is given in tab. 1
Tab. 1. Comparison system range with different type of PMD
coefficient.
PMD
tolerance
3
8
12
Fiber
with
0.05
ps/km0,5
3600 km
>4000
km
>4000
km
Fiber
with
0.1
ps/km0,5
900 km
>4000
km
>4000
km
Fiber
with
0.2
ps/km0,5
225 km
Fiber
with
1
ps/km0,5
9 km
1600 km
64 km
3600 km
144 km
The 40G transmission is four-times more sensitivity to noise.
In case CD this tolerance increases 16 times.
C. 100G Technology
Currently 100G interfaces are coming to the market and
promise to be next 10G technology in the backbone systems.
The commercial systems emerged in year 2011. IEEE
standardized the following types of interfaces:
-
10x10,2Gb/s SDM with the maximal haul of 10m
over copper cable
10x10,2Gb/s 850nm SDM with the maximal haul of
100m and based on MMF Ribbon
4x25,8Gb/s 1,3Pm LWDM with the maximal haul of
10km and based on SMF
4x25,8Gb/s 1,3Pm LWDM with the maximal haul of
40km and based on SMF
3
4*28Gb/s (OFDM) transmission for the short haul of
<200km for SMF
1*112Gb/s transmission using M-ary (ASK-)PSK
modulation for Metro Ethernet solutions <600km for
SMF
1*112Gb/s serial transmission taking advantage of
Coherent PolMux-QPSK modulation for the
Backbone solutions <2000km fot SMF
1*112Gb/s serial transmission taking advantage of
DP-QPSK modulation for the Backbone solutions
<2000km fot SMF
The first 4x28Gb/s OFDM solution is designed for short-haul
transmission. It is operating on the four independent
wavelengths with the speed of 28Gbps. For DWDM systems
this approach is less effective as it has four-times wider
spectrum comparing to serial transmission. This disadvantage
is eliminated in serial transmission of 1x112Gbps using M-ary
(ASK-) PSK modulation, it could be used in Metro Ethernet.
DQPSK modulation has been chosen for the 100G serial
transmission. The advantage of this solution is 50Gbaud/s
symbol rate and hence lower requirements for electronic.
Transponder that utilizes such technique can operate with
100GHz ITU grid.
Figure 2. DQPSK modulator.
Another solution for 100+G serial transmission is DP-DQPSK
modulation (POL MUX DQPSK). This scheme is combined
Polarization Division Multiplexing and QPSK Quadri-Phase
Shift Keying. It decreases the baud-rate by factor of 4. Each
symbol codes 4 bits. At 100G this scheme suffers less impact
than 40G from nonlinear effects. CD tolerance increases
because of doubled symbol duration. Electronic components
are not critical, it operates at lower speed of 25 Gbps. Common
100GHz ITU grid can be used due to narrower optical
spectrum. Thus 100G transmission could be implemented in
existing networks. In many cases it is impossible to exchange
optical cables and build infrastructure with new fibers both
from technical and economical point of view.
Two kinds of approaches in the 100+ Gbps Serial
Transmission are shown on figure 3.
The main problem and challenge for the vendors of the optical
transmission systems was the construction of the line side
XVI Poznańskie Warsztaty Telekomunikacyjne
25
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
4
with a coherent receiver, which has the potential to increase
spectral efficiency to 4 (bit/s)/Hz.
Tab. 2 Comparison of modulation schemes[1].
Figure 3. Block diagram of TX – RX modules of 100+ Gbps
serial transmission.
Another long haul 100G transmission is based on coherent,
single-carrier, dual-polarization QPSK with a digital intradyne
receiver and NRZ pulse shaping (coherent NRZ-DPQPSK).
This modulation scheme yields high CD and PMD tolerance
using digital-receive filter. It achieves high spectral efficiency
in the range of 2(bit/s)/Hz, which allows it to fit within a
50GHz WDM grid. Drawbacks include complexity and the
associated cost of the digital filter that must be able to process
four A/D-converted bit streams of 28Gbit/s in real time (the
capacity to process ~1.1Tbit/s). Figure 4 shows a block
diagram of a DP-QPSK system.
Modulati
on
10G
NRZOOK
NRZDPQPSK
NRZDPDQPSK
NRZDCDPQPSK
RZDPSK3ASK
4x28G
OOFDM
binary
NRZ
4x28G
Inv Mx
bin
NRZ
NRZDP16QAM
Detection
Direct,
no
preamp
Coh
Idyne,
DSP,
no pre
Incoh,
DLI,
DPC,
pre-amp
Cog
Idyne,
DSP,
no pre
Direct,
self-coh
(DLI),
pre-amp
Diect,
no preamp
Diect,
no preamp
Coh
Idyne,
DSP, no
pre
50
(25)
50
50
50
100
100
4 x 50
(4 x
25)
25
0.2
(0.4)
2
2
2
1
1
0.5
(1)
4
21 FEC
29 FEC
off
19
-300
+500
>1000
WDM
grid
(GHz)
Spect Eff
(b / s) /
Hz
OSNR
(dB)
15
13
15
12
21
21
FEC,
32 FEC
off
CD +/(ps/nm)
600
>1000
600
(TDC)
>1000
600
(TDC)
-300
+500
PMD
(ps)
NL
effects
Max
reach
(km)
10
>10
~5
>10
~4
~5
~5
>10
Mod
Mod
Mod
Strong
Low
Strong
Mod
Strong
2000+
2000
1500
2000
600
200 km
FEC
600 km
FEC,
1000
Migration to 100G technology is a significant step in capacity
and in system complexity. It required significant R&D
investments.
Figure 4. Coherent intradyne DP-QPSK transmission system.
LPF denotes a low-pass filter, PC denotes a passive optical
polarization controller, and PBS denotes a polarization beam
splitter [1].
This diagram shows the location of additional optional pulse
carvers, should RZ-DP-QPSK be the goal. Incoherent DPDQPSK requires ultra-fast, reliable polarization controllers.
Figure 5 shows incoherent DP-DQPSK for comparison.
High requirements on PMD and CD for the new transponders
will require the best parameters of the fibers or shorter
distances between amplifier or regeneration elements.
Moreover the quantity of possible regeneration point should be
shorten probably to three point only. Presented factors
influence on high-requirements of transmission path
maintenance
and necessity of rebuilding old DWDM
installations. Significant investments has to be made in special
100G testing equipment.
Figure 6 shows example of measurement taken for the 100G
transceiver line module. Optical Modulation Analyzer was
specially developed and is used to test and verify 40G/100G
and future solutions. It allows to investigate signal
constellation and eye-diagram.
III. CONCLUSION
Figure 5. Incoherent NRZ-DP-DQPSK system. DPC denotes a
dynamic polarization controller and R denotes a directdetection receiver [1].
100GbE modulation techniques offer a wide range of
performance and cost points. If a cost differential is sufficiently
high at 25% or above, this justifies the use of more than a
single, one-size-fits-all modulation scheme. An analysis is
required which would take transmission performance, spectral
efficiency, cost and latency into account.
In Table 2, modulation schemes and transmission systems
discussed are compared. It includes basic 10G NRZ-OOK for
reference. The comparison also includes NRZ-DP-16QAM
26
Paper presented recent advancements in optical data
transmission systems. Significant R&D investment was
required. Technologies will require new approach to network
maintenance. New set of measurement equipment and
techniques was developed. Time will tell how these new
technologies work in practice in different conditions.
40G/100G technology paved the way for future systems –
400G, 800G and 1T.
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
5
Figure 6. Example of measurements for Optical Modulation
Analyzer type of equipment.
REFERENCES
[1]
[2]
[3]
ADVA Optical Networking white paper “100GBE – The Future of
Ethernet”, August 2009.
EXFO application notes.
IEEE P802.3ba standard.
Piotr Rydlichowski (M’03) finished faculty of Electronics and
Telecommunication at Poznan University of Technology. In 2003-2008 PhD
student at Poznań University of Technology, chair of Multimedia
Telecommunications and Microelectronics. In 2008 joined Institute of
Bioorganic Chemistry PAS Poznan Supercomputing and Networking Centre.
Works in the team responsible for the maintenance of Polish NREN PIONIER network. Area of interest are: electromagnetic wave propagation
theory and microwave theory and techniques.
Piotr Turowicz graduated in Electronics and Telecommunication Science at
Poznan University of Technology with MSc degree in 1999. Since that year he
has been working in Network Department in Poznan Supercomputing and
Networking Center. His research interests are in the areas of broadband and
optical networks and new generation networks. From the beginning of his
work in PSNC he has been responsible for maintenance and development of
Poznan Metropolitan Area Network - POZMAN and Polish NREN -PIONIER
network.
XVI Poznańskie Warsztaty Telekomunikacyjne
27
2012
An Admission Control Algorithm for Multiuser
OFDMA-based Cognitive Radio Networks
Jerzy Martyna
Faculty of Mathematics and Computer Science
Institute of Computer Science, Jagiellonian University
ul. Prof. S. Łojasiewicza 6, 30-348 Cracow, Poland
Email; [email protected]
Abstract—Cognitive radio (CR) has been proposed as a way of
exploiting unused spectrum bands (also called spectrum holes).
However, several experiments have indicated that the currently
allocated spectrum is underutilized due the static nature of the
spectrum assignment. This paper presents an algorithm that
allows us to obtain admission control of secondary users (SUs)
in a multiuser OFDMA-based CR network implemented in the
cellular system. We also analyse the effect of primary unit
(PU) and secondary base station (SBS) activities on available
resource allocation. Thus, the suboptimal sub-carrier and power
allocation scheme with the minimization of the total transmission
power keeping the allowed interference to the PUs and SBS
are obtained. The simulation results demonstrate that the joint
admission control and resource allocation problems are treated
in a practical way.
Index Terms—cognitive radio networks, admission control
algorithm, spectrum management
I. I NTRODUCTION
The so-called cognitive communications, originally used
to control radios [1], [2], promise to offer fast and more
effective resource utilization by offering intelligent resource
assignment. In order to achieve this goal, cognitive communications will rely on information that is to be used by cognitive
radio devices and network elements. This is a complex task to
accomplish, thus making various learning methods for building
these models truly useful.
As shown in Fig. 1, the CR system coexists with the
primary system in the same geographic location. A primary
system operated in the licensed band has the highest priority
to use that frequency band (e.g. 2G/3G cellular, digital TV
broadcasts, etc.). Other unlicensed users and/or systems can
neither interfere with the primary system in an obtrusive way
nor occupy the licensed band. By using the pricing scheme,
each of the primary service providers (operators) maximises
its profit under the QoS constraint for primary users (PUs).
All of the unlicensed secondary users (SUs) are equipped
with cognitive radio technologies, usually static or mobile.
The primary users (PUs) are responsible for throwing unused
frequencies to the secondary users for a fee. While the existing
literature has focused on the communications needed for CR
system control, this paper assumes a network of secondary
base stations (SBSs). Every SBS can only have information
on a small number of PUs or channels. It causes interference
to PUs and SBSs.
28
Fig. 1. Downlink/uplink CR network.
The Orthogonal Frequency-Division Multiplexing (OFDM)
has been introduced by Saltzberg [3] and Chang [4]. The
idea behind it is the division of the broadband band into
parallel sub-bands, called sub-carriers, where the high-rate
data stream is split into low-rate streams. As the number
of sub-carriers increases, the bandwidth of each sub-channel
becomes narrower. The OFDM is currently used in digital
audio and video broadcasting standards, and in broadband
wireless access systems such as IEEE 802.16 (WiMAX), IEEE
802.20 (Mobile Broadband Wireless Access, MBWA). The
OFDM system used in the multiuser version is called the
Orthogonal Frequency-Division Multiple Access (OFDMA).
In all of these systems, the multiple accesses are achieved by
allocating a group of sub-carriers to a given user.
The OFDM system has been recommended as a candidate
for the CR system due its ability to perform underlying
sensing. The ability of the OFDM system to meet CR requirements can be studied according to approaches suggested
by Mahmond et al. [5], as follows: spectrum sensing, efficient spectrum utilization, interoperability, multiple access
and spectral allocation. As explained in [6], the resource
allocation technique has the potential to greatly improve power
efficiency. To increase spectral efficiency, resource allocation
schemes for OFDMA-based CR systems, such as a two-step
PWT 2012 - Poznań, 14 grudnia 2012
resource allocation algorithm in multicarrier based CR system
[7] or power/bit loading in OFDM-based CR networks [8],
have been proposed. Nevertheless, the traditional resource
allocation algorithms for OFDMA systems cannot be directly
used for CR systems due to the fact that the number of
available sub-carriers varies according to time.
Admission control for the CR network has been used in a
number of papers. For example, the optimal admission control
of secondary users at cognitive radio hotspots using a semiMarkov decision process to solve maximization problems has
been presented by Kim and Shin [9]. A fractional guard
channel reservation method has been used by Pacheco-Paramo
et al. [10] to control the balance between the blocking rate of
arriving users and the dropping rate of ones already admitted.
Nevertheless, none of the papers mentioned above analyze the
dependencies between interference constraints and transmission power in secondary systems.
The main contribution of this paper to the field is to provide
an admission control algorithm devoted to implementation
of SBSs in the downlink of multiuser OFDMA-based CR
networks. It will allow us to allocate the available system
resources optimally. It shows that by using the admission
control algorithm, the number of SUs is maximized and
the power allocation is optimized. Moreover, the proposed
admission control algorithm can be efficiently used to reduce
the drop rate of admitted SUs.
The rest of this paper is organized as follows. Section
2 provides the system model and describes the problem
formulation. The solution is proposed in Section 3. Section
4 presents the numerical results. Conclusions are drawn in
Section 5.
convex increasing in rk,n as it is necessary in optimization
methods.
In this paper, we assume that all system events such as SU
arrival, SU departure, and PU activity changes are synchronized. It is achieved by use of a CR system clock.
III. A DMISSION C ONTROL A LGORITHM
In this section, we present an admission control algorithm
in the CR system. We suppose that the admission control
algorithm should optimize the number of SUs in the CR
system and minimize the number of SUs being dropped due to
a lack of resources. Assuming the cellular structure of the CR
network, some channels are reserved in this system to serve
call handing over from neighbouring cells. Analogously, new
admission requests are dropping from the lack of frequency
carriers.
In order to obtain the admission control algorithm, the
problem can be mathematically formulated as follows:
min
K
N (3)
ρk,n pk,n
n=1 k=1
subject to
N
ρk,n bk,n ≥ Rktar ∀k ∈ KRR
n=1
E
N K
(4)
ρk pk,n Sk,n
≤ In
(5)
n=1 k=1
K
ρk,n ≤ 1
∀n ∈ Nf ree
(6)
k=1
II. S YSTEM M ODEL
This paper considers a multiuser downlink OFDMA-based
cellular system with an implemented cognitive radio system
(see Fig. 1). The accessible bandwidth B by the SU can be
utilized as long as it is not actively used by the PU. This
bandwidth is divided in the SBS belonging to the CR system
into N sub-bands. The received signal at the k-th SU on the
n-th carrier is given by
yk,n = hk,n xk,n + vk,n
(1)
where hk,n , xk,n , vk,n are the channel gain, transmitted signal and the zero means unit variance, i.e. circularly symmetric
complex Gaussian noise associated with the kth SU and the
nth carrier respectively.
To make the problem tractable, we define pk,n as the
transmission power associated with the n-th carrier and the
k-th SU, namely
pk,n ≈ βk,n (2rk,n − 1)
ρk,n ∈ {0, 1} ∀k, n
pk,n ≥ 0
∀k, n
(8)
where N free is the set of free carriers, ρk,n is a subcarrier
assignment indicator, indicating whether the n-th subcarrier
is assigned to the k-th SU or not, Rktar is the target bit rate
required by the k-th SU and the given average interference
threshold of the PU, KRR . is the set of the already admitted
rate required by SUs plus the newly arriving SUs.
The constraint (5) allows us to allocate the transmission
power among SUs over fading channels. The resulting interference Ik,n to the PUs band caused by the nth subcarrier is
given [11] by:
Ik,n = pk,n · Sk,n
(9)
where Sk,n is the power spectrum density (PSD) of the fading
process and is defined as
(2)
where βk,n > 0 is a function of hk,n and the Bit Error Rate
(BER) requirement rk,n is the rate associated with the k-th
SU and the n-th carrier. Additionally, we assume that pk,n is
(7)
Sk,n =|
2
hsp
k,n |
XVI Poznańskie Warsztaty Telekomunikacyjne
dn +
dn −
Wp
2
Wp
2
Ts
sin(πf Ts )
πf Ts
2
df
(10)
29
Fig. 3. Total transmission power versus the number of SUs for
different value of Rktar .
Fig. 4. The average SU drop rate versus the SU arrival rate.
dn representing the frequency distance between the n-th
subcarrier and the band of PU.
An algorithm for admission control in the CR system is
outlined in Fig. 2. The first procedure, called suboptimal
carrier assignment, finds the suboptimal power allocation for
the SUs in the CR system. Firstly, in the inner loop one SU
at each iteration to determine its suboptimal ρ∗k,n is selected.
If the target requirements are satisfied for the k-th SU, the number
of sub-bands is increased until the number of SUs or sub-bands are
exhausted.
The second procedure tests the all SU combinations that include
a single SU for eventual removal. If a single SU removal provides no
feasible combinations, all SUs combinations with two users removed
are tested until the set of feasible combinations is obtained. In both
cases, the number SUs of dropped in this way is established.
IV. S IMULATION R ESULTS
procedure suboptimal_carrier_assign;
begin
Cmax := 0; n := 1
set ρk,n ← Ωk ∀n;
t∗
obtain pk,n using pk,n = ρk,n
k,n
1
where t∗k,n = K·N
· Sk,n
;
k,n
{compute rate requirements of all SUs and Ccurr };
for k = 1 to K do
begin
k
Rreq
=
k∈KRR
n∈Nf ree pk,n · ρk,n · Sk,n ;
k
Rreq
< Rktar then n := n + Δn;
compute Ccurr ;
else
Cmax := Ccurr ;
end;
end;
procedure obtain_the_no._of_SUs_with_highest_sum_rate;
begin
let d := 1; { thenumber ofSUs }
K
compute | L |=
;
K −d
{ L is the set of all SUs combinations }
find | A |;
{ A is the set of admissible combination of SUs }
if | A |>| L | then suboptimal_carrier_assign;
else { | L ≥ A |}
d := d + 1;
end;
I
Fig. 2. The algorithm for admission control in CR network.
30
In this section the simulation results of the presented control
admission algorithm for the CR system are presented.
The system parameters in our simulation are as follows. All SUs
and PUs are distributed uniformly over 1 kms × 1 km. We set some
parameters, such as: the PU transmit power equal to 100 mW, the
SU transmit power equal to 10 mW for all SUs, and the noise level
σ 2 = −90 dBm. The target bit rate requirement increases from 0.5
bps/Hz to 2.0 bps/Hz.
Fig. 3 depicts the total transmission in the function of the number
of SUs for different values of Rreq. It can be seen that the transmission power increases with the value of Rktar . It means that the
decrease of Rreq is associated with a smaller total transmission power.
Fig. 4 shows the average number of dropped SUs in dependence
of the SU arrival rate for two admitted density of SUs (No. of SUs /
m2 ). We see that the admission control algorithm causes a large SU
drop rate for the high density of SUs in the simulated scenario.
V. C ONCLUSIONS
In this paper, an algorithm for admission control of SU traffic
rate in the multiuser OFDM-based cellular cognitive system was
presented. An iterative algorithm is powerful solution for such
problems because its iterative and distributed nature allows for many
applications such as target rate requirements and eliminating central
schedulers, taking them from possible commercial deployment. Due
to its high complexity, the suboptimal algorithm was proposed. Numerical results showed that the proposed admission control algorithm
may be an effective candidate to employ in the CR networks.
R EFERENCES
[1]
[2]
J. III Mitola, G. Q. Maquire Jr., Cognitive Radio: Making Software
Radios More Personal, IEEE Pers. Commun., vol. 6 no. 4, 1999, pp. 13
- 18.
J. Mitola, Cognitive Radio: An Integrated Agent Architecture for Software Defined Radio, Ph.D. Thesis, Teleinformatics, Royal Institute of
Technology (KTH), 2000.
PWT 2012 - Poznań, 14 grudnia 2012
[3]
B. Saltzberg, Performance of an Efficient Parallel Data Transmission
System, IEEE Trans. on Communication Technology, vol. 15, no. 6,
1967, 805 - 811.
[4] R. Chang, R. Gibby, A Theoretical Study of Performance of an
Orthogonal Multiplexing Data Transmission Scheme, IEEE Trans. on
Communication Technology, vol. 16, no. 4, 1968, 529 - 540.
[5] H. Mahmond, T. Yucek, H. Arslan, OFDM for Cognitive Radio: Merits
and Challenges, IEEE Wireless Communication, vol. 16, no. 2, 2009, 6
- 15.
[6] G. Han, T. Harrold, S. Armour, I.Krikidis, S. Vidrev, P.M. Grant„ Green
Radio: Radio Techniques to Enable Energy Efficient Wireless Networks,
IEEE Communication Magazine, 2011, vol. 49, 46 - 54.
[7] Shaat Musbah, Bader Faouzi, A Two-step Resource Allocation Algorithm in Multicarrier Based Cognitive Radio Systems, in: Proc. IEEE
Wireless Communication Networking Conference (WCNC), 2010, 1 6.
[8] C. Zhao, K. Kyungsup, Power/bit Loading in OFDM-based Cognitive
Networks with Comprehensive Interference Considerations: the SingleSU Case, IEEE Trans. on Vehicular Technology, vol. 59, 2010, 1910
-22.
[9] H. Kim, K. Shin, Optimal Admission and Eviction Control of Secondary Users of Cognitive Radio Hotspots, in: Proc. 6th Annual IEEE
Communication Society Conference on Sensors, Mesh, and Ad Hoc
Communication and Networks, June 2009, 1 - 9.
[10] D. Pacheco-Paramo, V. Pla, J. Martinez-Bauset, Optimal Admission
Control in Cognitive Radio Networks, in: Proc 4th Int. Conf. on
Cognitive Radio Oriented Wireless Networks and Communications, June
2009.
[11] T. Weiss, J. Hillebrand, A. Krohn, F. K. Jondrol, Mutual Interference
in OFDM-based Spectrum Pooling System, in: Proc. IEEE Vehicular
Technology Conference, 2004, 1873 - 7.
XVI Poznańskie Warsztaty Telekomunikacyjne
31
2012
Bayesian Methods in Optimization of Analog
Communication Systems with Feedback
Anatoliy Platonov, Senior Member, IEEE
Abstract –The paper shows a possibility and advantages of consideration of the feedback communication systems (FCS) as the
estimation systems. There are considered general effects appearing in optimal FCS and caused by them changes in limit information characteristics of the systems. The obtained results enable a
transition to practical design of optimal FCS. They remain valid
also for the adaptive estimation systems with feedback.
Index Terms—Adaptive modulation, analog transmission,
feedback systems, power-bandwidth efficiency. threshold effect
x
Qk
x
xˆk 1
N
R2
Calculation
of Mk
Pk
Pk
Pk
Calculation
of Pk
xˆn
z-1
Fig. 1. Block diagram of estimation system with adjusted nonlinear observing
device [11].
[t
ek
x
S&H
OWADAYS,
32
xˆn
Estimation
Unit
xˆk 1
xt
analogue signal transmission is considered as
an initial stage of communications development which
has no future in comparison with evident superiority of digital
communication systems (CS). However, naither C.E. Shannon,
nor founders of information and communication theory, considered digital transmission as more perspective than the analog
one. Moreover, in the 1950-1960s, analog communication
systems with feedback channels were a subject of very intensive and promising researches (e.g. works [1]-[9]). However,
results of these researches have no practical application even
now. In turn, impressive successes of digital CS theory and
applications created, in 1960s, a great field of new general and
particular theoretical tasks with granted, not less great interest
of the industry to new solutions. This switched the attention of
researchers to digital communications, and analog feedback
systems (FCS) became the subject of relatively rare theoretical
investigations dominated by the great amount of works devoted to digital CS.
We have studied basic series of works in analog FCS to understand what had hampered implementation of the obtained
excellent analytical results and their further development? The
answer was somethat unexpected – the main reason was commonly used linear model of the transmitting part of FCS. The
second reason was a solution of optimization task under constraints on the mean or peak-power of emitted signals. Both
these reasons did not allow a consideration of the cases of saturation of the transmitters (or overmodulation) which are the
sources of abnormal errors.
The particular feature of researches in analog FCS is a possibility to determine analytical form of the mean square error
(MSE) of transmission and to derive minimizing MSE optimal
transmission-receiving algorithm using methods of applied
Bayesian estimation theory [10]. (Let us notice that this is impossible for digital CS due to impossibility to search minimum
of MSE on the set of possible codes). However, MSE does not
yk
Mˆ k
Qk
I. INTRODUCTION
[t
yk
ek
x
Σ
Bˆk
M1
Ch1
st,k
DM1
yk
DSPU
Mˆ k
R2
Transmission unit (TU)
st,k
xˆn
Bk , M k
Ch2
T2
Kt
z-1
Base station (BS)
Fig 2. Block-diagram of the analog FCS [13]-[16].
allow adequate consideration of abnormal errors which cause
irreversible loss of information about the samples and increase
mean percentage of erroneous bits (bit-error rate - BER) in
information streams at the channel and system outputs. Rare
but crucially important for practice, abnormal errors add to the
value of MSE negligibly small values. They also cannot be
included into the varied form of estimating rule. This was the
main reason of unsuccessful the probes implementation of the
results obtained in the frame of only MSE approach.
In the paper, we give deeper explanation of key ideas of
new approach to optimization of FCS which removes the described above difficulties and enables designing the topefficient systems working at the upper (Shannon’s) boundaries.
There are discussed not studied before effects appearing in optimal FSC. The paper develops the results of works [13]-[16].
II.
CONSIDERATION OF ABNORMAL ERRORS IN FCS THEORY
Optimization of FCS is based on the original approach ([11]
,[12] and other works) to optimization of the adaptive feedback estimation systems (FES) with adaptively adjusted nonlinear observing part (the simplified block diagram of FES is
presented in Fig. 1). Particularity of the approach is explicit
consideration of always existing possibility of saturation of the
analog part of the system.
A. The key ideas of the approach
The main idea is introduction, apart from MSE, of additional criterion of FES performance – permissible probability μ
PWT 2012 - Poznań, 14 grudnia 2012
of its saturation for each moment of the system work (statistical fitting condition, [11]-[16]). This condition reduces the sets
of possible values of the parameters of analogue part to the
“permissible” sets. These sets consist of the parameters guaranteeing that, for each time instant, saturation may appear
with a probability not greater than prescribed value μ ,
(
10 4 ). The statistically fitted FES (with parameters from
the permissible sets) will be working as the linear systems with
a probability not smaller than 1 - μ .
In practice linear statistically fitted FES permit to use Bayesian methods and to find the parameters of the analog and digital parts of FES minimizing MSE. Solution of this task – optimal observation-estimation algorithms determine the parameters of the analog part and estimation algorithm, as well as
optimal rules of the analog part adjusting. This information
enables systematic design of optimal FES, whose performance
attains theoretically achievable upper boundary (minimal MSE
of estimates). Non-linearity of these systems and saturation
errors will be excluded at the confidence level 1
.
One should say that Bayesian methods can’t be applied to
optimization of digital CS due to the lack of continuous analytical models of digitizing units and encoders. For this reason,
digital theory employs several criterions of the CS performance. One of them is BER.
As it was shown in [14]-[16], probability of saturation μ
plays the same role, as BER in digital CS but was never used
in the analysis of analog CS. Deeper analysis of the question
showed that just its introduction into the theory (by means of
the statistical fitting condition) removes the problems which
blocked application of earlier theoretical results [1]-[9], and
enables their further development and generalizations.
Similarity of the block diagrams of FES (Fig. 1) and FCS
(Fig. 2), as well as coinciding mathematical models of their
analogue parts and transmission lines allows direct application
of the approach presented in [11],[12] to FCS optimization.
Differences between the technical units, data transfer channels
and application of the systems are not essential and require
only careful consideration in the parameters of the models.
B. Main models used in FCS and FES analysis.
The block diagram of the point-to-point analog FCS is presented in Fig. 2. The system includes adaptively adjusted peripheral transmitting unit (TU) and base station (BS) connected
by the forward M1-Ch1-DM1 and feedback T2-Ch2-R2 channels. Both channels are assumed to be stationary, linear and
memoryless. The channel noises ξt , ηn are additive white
Gaussian noises (AWGN) with known spectral power densities
N ξ , N η . The input signals xt are limit-band stationary Gaussian processes with known mean value x0 and variance σ 02 .
The input sample-and-hold unit (S&H) forms sequences of
the samples x ( m ) = x(mT ) , m 1,2,.. of the input signal xt ,
T 1/ 2F is the sampling period. Each sample is transmitted
in n T / Δt0 F0 / F cycles ( Δt0 1/ 2 F0 is duration of
the single cycle of transmission, 2F0 determines the channels
bandwidth). As in [14]-[16], we assume that each sample is
A0
Bˆk A0 / Mˆ k
x
sk
x
x
Bˆ k Bˆk A0 / Mˆ k
x
-A0
Fig. 3. Transition characteristic of adaptive transmitter Σ +M1.
transmitted independently and in the same way. This permits
to reduce the analysis of the systems transmitting single sample
and omit index m in notations of the input samples x ( m ) . Simultaneously, this reduces the analysis to the particular case of
the random values estimation considered in [11],[12].
Adaptive transmitter includes the subtracting unit Σ and
amplitude modulator M1 with the controlled modulation index
Mˆ k . There is assumed that amplitude of the emitted signals
cannot be greater than A0 (saturation level). In this case,
transmitter can be described by the model (see also Fig. 3):
­
° Mˆ ( x Bˆ k ) if Mˆ k | x Bˆ k | d 1½
°
st , k A0 ® k
¾ cos(2S f 0t Mk )
ˆ
ˆ
ˆ
°sgn ( x Bk ) if M k | x Bk | >1¿
°
¯
(1)
where A0 , f0 , φk are the parameters of the carrier, and
(k -1)'t0 d t d k 't0 , k 1,..., n are the current numbers of the
transmission cycles. In each cycle, adjusted parameters of the
transmitter: modulation index Mˆ k and signal Bˆ k at the second
input of the subtractor Σ are set to proper values (see below).
To simplify the form of analytical results the DSB-SC AM is
considered, but the obtained results can be easily extended to
the other types of AM. This model corresponds to non-linear
analog unit of FES block diagram in Fig. 1.
x Bˆ k
Formed by the subtractor Σ difference signal ek
modulates the carrier transmitted to BS through the forward
channel M1-Ch1-DM1 with AWGN ξt . Signal st , k at the
input of demodulator DM1 is described by the relationship:
γ0
(2)
sk ,t
sk ,t ξt
r
where γ0 is the channel gain and r is the distance between
TU and BS. Signal y k (“observation”) formed by the demodulator DM1 is routed to the input of digital signal processing
unit (DSPU) of BS. This ppart of FCS corresponds to communication line yk
yk ξk of FES.
Unit DSPU computes the current estimate xˆk of the sample
according to the Kalman-type equation:
xˆk
xˆk 1 Lk [ yk E ( yk | y1k 1 )] ; ( x̂0
x0 )
(3)
k
1
1
1
( y1 ,..., yk 1 ) denotes the sequence of observawhere y
tions received at the previous cycles and E ( yk | y1k 11 ) is the
one-cycle prediction of the signal at the demodulator DM1
output. Parameter Lk determines the rate of algorithm (3)
convergence. Equation (3) represents the mathematical model
of digital part of FCS and has the same form as in [11].
In [11],[12] it was shown that, in the Gaussian case, optimal
values of the parameter Mˆ k do not depend on observation y k
XVI Poznańskie Warsztaty Telekomunikacyjne
33
p
and can be computed and set independently. Therefore, apart
of the estimate, DSPU computes only control Bk B( y1k )
routed to the TU through the feedback channel R2-Ch2-T2.
Receiver R2 routes the received signals to corresponding input
of the subtracting unit Σ which forms the next difference signal ek 1 x Bˆ k 1 . Simultaneously, synchronizing unit of
DSPU resets the modulation index to the value Mˆ k 1 and the
gain Lk to new value Lk 1 , and the next cycle of the sample
transmission begins. After n cycles, final estimate of the sample xˆn is routed to the addressee, and AFCS begins a transmission of the next sample.
It is important to notice that feedback channel with AWGN
does not require detailed modeling. It is sufficient to describe
the error of the parameter Bˆ k setting as AWGN νk (shown by
upper arrows in Figs. 1, 2) with known variance σ v2 and to
model the feedback chain by the relationship Bˆ k 1 Bk ν k 1 .
C. Optimal transmission-reception algorithm
The presented models allow derivation of analytical form of
E[( x xˆk )2 ] for each
the MSE of transmission errors Pk
k 1,..., n and statistical fitting condition excluding appearance of the transmitter saturation at the confidence level 1
.
In turn, identity of the mathematical models of FES and FCS
permits to perform full optimization of FSC using approach
proposed in [11]. The result of optimization – optimal joint
transmission - reception algorithm is as follows [13]-[16]:
a). Receiving algorithm for DSPU of BS, ( k 1,..., n ):
xˆk
xˆk 1 Lk yk ; ( x̂0
x0 )
(4)
where gains Lk are set, in each cycle, to the values
Lk
Pk 1
Q2
1
.
AM k 1 1 Q 2 VQ2 Pk 1
1
(1 Pk Pk11 )
AM k 1
b). Transmitter TU adjusting algorithm ( k
st , k
Bˆ k
Bk 1
1,..., n ):
A0 Mˆ k ( x Bˆk ) cos(2S f0t Mk )
Bk 1 ν k , Mˆ k
xˆk 1 ; M k 1
1
D σ Pk 1
2
ν
(5)
M k 1
; ( M0
(6)
(7)
(D σ 0 ) 1 ). (8)
c). Parameters Pk in (5), (8) equal to minimal MSE (MMSE)
of estimates determined by the relationship:
ª (1 Q 2 )V v2 Pk 1 º
2
(1 Q 2 )1 «
» Pk 1 ; ( P0 V 0 ) . (9)
2
VQ Pk 1
¬
¼
2
Q
in (5), (9) describes the signal-to-noise ratio
Parameter
(SNR) at the channel M1-Ch1-DM1 output:
2
W A2 M k21 E[( x Bˆk )2 ] § A ·
1
Ch1
(10)
Q2
SNRout
¨
¸
2
2
V[
V[
© α ¹ Nξ F0
Pk
where W ( A / α ) 2 ,is the power of emitted signal and
A A0 γ0 / r is the amplitude of received signal, respectively; σξ2 N ξ F0 . Parameter α (saturation factor) is connected
with permissible probability of TU saturation by the equation:
P 1- 2)(D ) , where )(D ) is Gaussian error function. Observations yk in (4) are formed according to the relationship:
34
yk
A Mˆ k ( x Bˆ k ) ξ k , where A
A0
J0
.
(11)
r
Algorithm (4)-(10) contains the basic information permitting to design FCS with MMSE of transmission (9).
III. NEW EFFECTS APPEARING IN OPTIMAL FCS
Depending on the task, optimal FCS analysis may have two
scenarios: under fixed channel bandwidth F0 or the system
baseband F . Below, we consider the case of the given F0 . In
this case, duration of a single cycle Δt0 1/ 2 F0 , and transmission of the sample in n cycles means corresponding narrowing of the bandpass by n times.
A. Changes of MMSE versus number of cycles
The results presented below are obtained under condition:
Ch1
SNRinp
V 02
V ν2
!! 1 Q 2
Ch1
1 SNRout
(12)
which describes a typical situation: SNR at the output of the
forward channel is small but SNR at the modulator input is
large (resources of BS provide the sufficiently small feedback
errors). In this case, MMSE (9) can be replaced by the approximate relationship [13]-[16]:
2
2 n
*
­
for 1 d n d n
° V 0 (1 Q )
® 2
1
*
*
°̄ σ ν (n n 1) for n ! n
where threshold point n* is solution of the equation Pn
Ch1
log 2 ( SNR inp )
§ σ 02 ·
1
*
.
n
log
2 ¨
2 ¸
Ch1
log 2 (1 Q2 )
© σ v ¹ log 2 (1 SNR out )
Pn
*
(13)
VQ2 :
(14)
B. Capacity of forward channel and of FCS as a whole
According
g to (8),(11), E ( yk | y1n 1 ) 0 and E ( yk2 | y1n 1 )
2
E ( yk ) ( A / D ) 2 σ ξ2 . Computation of the prior and posterior
y1n , and of the mutual amount
entropies of received sequences
q
n
n
n
of information I (Y1 ; e1 ) in y1 and e1n gives the relationships:
n
§
A2 ·
n
H (Y1n ) ¦ H (Yk )
log 2 2πeσ ξ2 ¨1 2 2 ¸ ;
(15)
¨ ασ ¸
2
k 1
ξ ¹
©
n
n
n
H (Y1 | e1 )
log 2 (2S eV [2 ) ;
(16)
2
n
n
n
I (Y1 ; e1 ) n[ H (Y ) H (Y | e)]
log 2 (1 Q 2 ) .
(17)
2
If the only information about the input signal is its mean
value and variance, then (15) describes maximal, on the set of
possible distributions of input sequences e1n , entropy of the
received sequences y1n . This entropy is additionally maximized over the mean power of the signals emitted by TU under given probability of saturation. The latter means that (17)
determines the upper boundary of the amount of information
delivered to BS in n cycles of transmission. In this case relationship
n
n
§ W sign ·
I (Y1 ; e1 )
ch1
RCh1
F0 log 2 (1 Q 2 ) F0 log 2 ¨1 ¸ C (18)
N
F
nΔt0
0
[
©
¹
determines the capacity of the forward channel. The single
difference from classical expression is its dependence on saturation factor α in SNR Q 2 . In turn, the mean bit-rate at the
optimal FCS output attains maximal
PWT 2012 - Poznań, 14 grudnia 2012
*
*
n
*
25
n
1
*
+
*
Fig. 4. Dependence of FCS capacity R AS on number of cycles
y
under different
power of received signals (from the top: for W 100 2 ; 1100 3 ; 10 4 ; 10 5 ; 10 6 ,
V ξ2 105 , dash lines refer to the capacities RCh1 , stars - to threshold points).
Ch1
Fig. 5. Changes of limit values of P-efficiency Eˆ nbit AS / N ξ f n ( SNRout
) in
different cycles of transmission (lower trajectory coincides with that for forward channel transmission, see also [18]). Equation Chapter (Next) Section 1
value and determines the capacity of FCS as a whole:
RnFCS
CnFCS
I ( X ; Xˆ n )
max
Tn
Fn log 2
σ
2
0
F0
n
Pn
log 2
σ
REFERENCES
2
0
Pn
= (19)
*
­C Ch1
for 1 d n d n
°
[bit/s]
ª σ 02
º
® F0
*
*
° n log 2 « σ 2 (n n 1) » for n ! n
¬ ν
¼
¯
Unlike the constant channel capacity (18), capacity of the
system depends on n and diminishes beginning with n n * .
This effect is confirmed by the results of simulations (Fig. 4).
C. Spectral-power efficiency of optimal FCS
WTn
nW
N ξ I ( X , Xˆ n )
2 N ξ F0 I ( X , Xˆ n )
nQ
2
σ2
log 2 0
Pn
Substitution of (9) into (20) gives the expression:
Nξ
[2]
[3]
[4]
[5]
[6]
The spectral (S)-efficiency of FCS RnAS / F0 [bit/s/Hz] is determined by (19) and depends on n in a similar way. The power (P)-efficiency of FCS can be assessed according to the basic
definition [17] as normalized to the spectral density of channel
noise, energy necessary for delivering of one bit to addressee:
Enbit FCS
[1]
[7]
[8]
[9]
[J/bit] . (20)
2
­E
Q
*
for 1 d n d n ;
° N
2
Q
log
(1
)
2
bit FCS
° [
En
°
2
®
nQ
(21)
Nξ
for n t n*
°
2
σ
ª
º
*
0
° log
( n n 1) »
2 « 2
°
¬ σν
¼
¯
for theoretically achievable efficiency of transmission. Formula (21) and results of simulations (Fig. 5.) show that the “enerCh1
) grows for n n * .
gy of bit” Eˆ nbit AS / N ξ f n ( SNRout
Claim: Transmission of the input signals with the sampling
frequency n* times smaller than F0 (with the spectrum expansion greater than n* ) decreases efficiency of transmission.
The obtained results are valid also for estimation systems
discussed in the Introduction. For n n * , expressions for the
limit characteristics of optimal FCS coincide with corresponding results in [1]-[9].
[10]
[11]
bit Ch 1
[12]
[13]
[14]
[15]
[16]
[17]
[18]
P. Elias, “Channel capacity without coding,” MIT Research Laboratory
of Electronics, Quarterly Progress Rep., October 1956, pp. 90–93, also
in Lectures on Comm. System Theory, Ed. E. Baghdady, McGraw-Hill,
NY, 1961 pp.90-93.
C. E. Shannon, "The zero-error capacity of a noisy channel." IRE Trans.
on Inf. Theory, vol.2, no.3, 1956, pp. 8-19.
P. E. Green, “Feedback communication systems,” in Lectures on Communication System Theory, McGraw-Hill, NY, 1961 pp. 345-366.
T. J. Goblick, “Theoretical limitations on the transmission of data from
analog sources”, IEEE Trans. on Inf. Theory, vol.11, no.4, 1965, pp.
558 – 567.
J. Schalkwijk and T. Kailath, “A coding scheme for additive noise
channels with feedback - Part I” IEEE Trans. on In. Theory, vol. 12,
April 1966 pp. 172–182,
T. Kailath, “An application of Shannon’s rate-distortion theory to
analog communication over feedback channels”, Proc. IEEE, vol. 55,
no. 6, 1967, pp. 1102-1103.
J. P. M. Schalkwijk, L. I. Bluestein, “Transmission of analog waveforms through channels with feedback”, IEEE Trans. on Inf. Theory,
vol. 13, no.4, 1967, pp.617-619.
J. K. Omura, “Optimal transmission of analog data for channels with
feedback”. IEEE Trans. on Inf. Theory, vol. 14, no. 1, 1968, pp. 38-43.
S. Butman, “A general formulation of linear feedback communicationsystems with solutions,” IEEE Trans. on Inf. Theory, vol. 15, no. 5,
1969, pp. 392–400.
H. L. Van Trees, Detection, estimation and modulation theory, J. Wiley,
New York, 1972.
A. Platonov, “Optimal identification of regression-type processes under
adaptively controlled observations”, IEEE Trans. on Sign. Proc., vol.
42, no. 9, Sept. 1994, pp. 2280-2291.
A. Platonov, Analytical methods of analog-digital adaptive estimation
systems design, D.Sc. monograph, Publishing House of Warsaw Univ.
of Technology, s. “Electronics”, vol. 154, Warsaw, 2006 (in Polish).
Platonov A., “Wysokoefektywne iteracyjne systemy transmisji sygnałów
z nadajnikami AM małej mocy (High efficient iterative transmission
systems with low-power AM transmiters), Materiały Konferencyne X
Poznańskich Warsztatów Telekomunikacyjnych, PWT - 2005, Poznań,
2005, pp. 347-352 (in Polish).
Platonov A., “Optimization of adaptive communication systems with
feedback channels”, IEEE Wireless Comm. and Networking Conf.
WCNC’2009, Budapest, 2009, (IEEE Xplore).
Platonov A., “Capacity and power-bandwidth efficiency of wireless
adaptive feedback communication systems”, IEEE Comm. Letters, vol.
16. no. 5, pp. 573-576, 2012.
Platonov A., Analysis of power-bandwidth efficiency and MSE of
transmission as the measures of communication systems performance,
Przegląd Telekomunikacyjny, no. 4, pp. 116-119, 2012.
E. Bedrosian, “Spectrum conservation by efficient channel utilization
IEEE Communication Society Magazine, vol. 15, no. 7, 1997, pp.20-27.
J.S. Lee, L.E. Miller, CDMA Systems Engineering Handbook, Artech
House, Boston-London, 1988.
XVI Poznańskie Warsztaty Telekomunikacyjne
35
2012
XVI Poznań Telecommunications Workshop - PWT 2012
1
Path Loss Modelling in Industrial Environment
Sławomir J. Ambroziak
Abstract — In the paper adjusted selected propagation models
as well as the new path loss model have been evaluated in terms of
designing mobile radio networks in investigated environment.
Index Terms — path loss modelling, radio waves propagation,
measuring research, container terminal environment
I. INTRODUCTION
The container port area is a particular example of the industrial environment and should be treated as a very difficult radio
waves propagation environment. There are a number of propagation models, mainly for urban, suburban or rural environments. There is also propagation model destined for container
port environment, but this model has been developed for designing only fixed radio links. So there was a need to adjust
existing models and to develop new propagation model based
on results of measurement research. Such tests have been carried out by author in Gdansk Deepwater Container Terminal
(hereinafter called DCT Gdansk) in accordance with normative
requirements [1, 2], that have to be met during research. The
analysis contained in [3] has been also taken into account.
During research in DCT Gdansk nearly 290 thousand propagation cases have been collected. These cases concern propagation routes with various lengths (up to 620 m), various frequencies of test signal (from a range of 500 MHz up to
4 GHz),various heights of transmitting antenna installation
(from a range of 12 m up to 36 m) and various average heights
of container stacks (from a range of 2.6 m up to 8.7 m). It
should be noted that width of the streets (transport routs between containers stacks) as well as theirs length are not variables and equal 10 m and 620 m, respectively.
In the paper adjusted selected propagation models have
been evaluated in terms of designing mobile radio networks in
investigated environment. These models are: ITU-R P.1411
models for the case of propagation over roof-tops for urban
and suburban areas [4], COST231 – Walfisch-Ikegami model
[5] and the multi-variant empirical model for designing fixed
radio systems in container terminal [6]. The new propagation
model for designing mobile radio links in container terminal
has also been evaluated [7]. At the end of the paper results
have been summarized and discussed.
II. ADJUSTED PROPAGATION MODELS
Analysis contained in [8] has pointed the necessity of statistical adjustment of the selected models. In order to increase
accuracy of path loss estimation in investigated environment, a
36
method of statistical adjustment of selected models has been
proposed. This method relies on adding functional correction
factors to original path loss formulas. Coefficients of these
functions have been calculated on the basis of empirical data
and using multivariate linear regression. The statistical significance of particular coefficients has been proved (with 95%
confidence) using the Student's t-test.
A. Adjusted COST 231 Walfisch-Ikegami
The COST231 Walfisch-Ikegami model has been developed
as a result of COST 231 Project [5]. The model allows to improved path loss estimation by consideration of more data to
characterized of the urban environment, namely: heights of
buildings, widths of roads, buildings separation and road orientation with relation to the direct radio path. For NLOS situation the basic transmission loss is depended on a free space
loss L0 [dB], a multiple screen diffraction loss Lmsd [dB], and a
roof-top-to-street diffraction and scatter loss Lrts [dB] [5].
The Walfisch-Ikegami model has been modified by adding
two functional correction factors to original equation:
ΔLc1(f, d, Δht, Δhb, ϕ) and ΔLc2(f, d), as follows:
­ L Lrst Lmsd ' Lc1
L'WI > dB @ ® 0
L0 ' Lc2
¯
for Lrst Lmsd ! 0,
for Lrst Lmsd d 0.
These factors are expressed by following polynomials:
'Lc1 > dB @
­ for suburban areas original scenario :
°
°17.9 lg f 15lg d 12.9 lg 'ht °
5.2 lg 'hb 0.08I 48.2,
°
®
° for urban centres original scenario :
° 31.9 lg f 16.7 lg d 8.8lg 'h t
°
°
'
5.5lg
h
0.09
I 86.7,
b
¯
'Lc 2 > dB @
­ for suburban areas original scenario :
°
° 2.3lg f 9.1lg d 8.7,
®
° for urban centres original scenario :
° 5.9lg f 7.5lg d 16.8,
¯
where f [MHz] – signal frequency, d [km] – distance between
base station antenna and mobile terminal, Δht [m] - difference
between average height of containers stacks and height of mobile terminal, Δhb [m] - difference between height of base station antenna and average height of containers stacks and ϕ [°] street orientation angle.
PWT 2012 - Poznań, 14 grudnia 2012
XVI Poznań Telecommunications Workshop - PWT 2012
B. Adjusted ITU-R P.1411 (§4.2.1)
Recommendation [4] includes propagation models destined
for designing short-range outdoor radio communication systems for different types of environments. Two models for typical cases (called NLOS1), where the base station antenna is
mounted above roof-top level, have been selected. For these
situations radio wave propagation is mainly over the roof-tops.
The first one is the model described in §4.2.1 of [4]. It is a
modified version of the Walfisch-Ikegami model, extending
the frequency range of its applicability up to 5 GHz. In addition, this model describes situations in which the length of the
path covered by buildings is less than so called "settled field
distance". This situation hasn't been taken under consideration
in Walfisch-Ikegami model.
In case of propagation model, described in §4.2.1 of ITU-R
P.1411, adjustment has been done by adding to original model
equation
one
functional
correction
factor
ΔLc3(f, d, Δht, Δhb, ϕ), as it is outlined below:
2
'Lc 5 > dB @ 22.5lg d 15.6lg hb ht 16.6lg hr ht 0.2I 46.2,
'Lc 6 > dB @ 3.8lg d 0.7 lg O 1.7 lg hb ht 15.1lg hr ht 0.02I 6.2,
where λ [m] – wavelength, d [m] –distance between base station antenna and mobile terminal, hb [m] - height of base station antenna, hr [m] – average height of containers stacks,
ht [m] - height of mobile terminal. Other parameters have the
same meaning as in the previous cases.
D. Adjusted Empirical Model
There is also known the empirical model for designing fixed
radio links in container terminal. This model makes the basic
transmission loss dependent on: frequency f [MHz], propagation path length d [km], path type qualification (LOS or NLOS
condition) and difference between transmitter antenna height
hb [m] above terrain level and average height hav [m] of con­ L0 Lrst Lmsd for Lrst Lmsd ! 0,
tainers stacks [6, 9]. From among four variants, two describes
L'1411,4.2.1 > dB @ ' Lc3 ®
L
for
L
L
d
0.
propagation situations that occurred during tests (both for
0
rst
msd
¯
hb t hav), namely: LOS1 and NLOS1.
This factor is expressed by following equation:
This model has been modified by adding functional correction factors: ΔLc7(f, d, hb–hav) and ΔLc8(f, d, hb–hav) to two in­ for suburban areas original scenario :
vestigated variants, namely:
°
°6 lg f 19.7 lg d 6.9 lg 'ht °
L 'LOS1 > dB@ LLOS1 > dB@ 'Lc7 >dB@ , 0.7 lg 'hb 0.2I 10,
°
'Lc 3 > dB @ ®
° for urban centres original scenario :
° 6 lg f 20 lg d 9.4 lg 'h t
L 'NLOS1 > dB@ LNLOS1 > dB@ 'Lc8 >dB@. °
°¯
0.1lg 'hb 0.2I 12,
Function parameters have the same meaning as in (2) and (3).
C. Adjusted ITU-R P.1411 (§4.2.2)
The third model has been characterized in §4.2.2 of [4]. It
may be used to calculate the basic transmission loss in suburban environment. Depending on the distance between base
station and mobile station this model distinguishes three regions in terms of dominant arrival waves at the mobile station:
a direct wave dominant region (with L0 [dB]), a reflected wave
dominant region (with Lrw [dB]) and a diffracted wave dominant region (with Ldw [dB]).
Adjustment of this model was done by adding to original
equation three functional correction factors: ΔLc4(d, λ), ΔLc5(d,
hb-ht, hr-ht, ϕ) and ΔLc6(d, λ, hb-ht, hr-ht, ϕ), as follows:
These factors are expressed by following polynomials:
0.2lg f 14lg d 11.3lg hb hav , 'Lc 7
'Lc8
0.2lg f 11.9lg d 11.6lg hb hav 22.5. III. NEW EMPIRICAL MODEL FOR MOBILE RADIO LINKS
The environment under investigation has a relatively regular
structure. However, the diversity of conditions occurring in
different places of the container terminal was taken into account in the new model (named the MCT model, as an abbreviation for: mobile, container, terminal). For this reason, the
terminal was divided into three subareas, where different prop­ L0 ' Lc4
direct wave dom. reg. ,
agation mechanisms have a crucial influence on basic trans°
L'1411,4.2.2 > dB @ ® Lrw ' Lc5 reflected wave dom. reg. , mission loss, namely: LOS Area, Containers Area and Off° L ' L diffracted wave dom. reg. .
Terminal Area [7].
c6
¯ dw
Based on preliminary analysis of propagation conditions,
the relevant factors affecting the basic transmission loss value
These factors are expressed by following polynomials:
in a container terminal environment were defined, namely:
'Lc 4 > dB@ 6.5lg d 0.8lg O 21.5, frequency (f [MHz]) of the radio signal, propagation path
length (d [m]), base station antenna height (hb [MHz]), angle
XVI Poznańskie Warsztaty Telekomunikacyjne
37
XVI Poznań Telecommunications Workshop - PWT 2012
3
(ϕ [°]) of radio wave arrival The characteristic parameters for
the investigated industrial environment are also very important. These are:
x terminal surface occupancy ratio (St), defined as a ratio of surface occupied by containers to all terminal
surface destined for container storage,
x i-th row surface occupancy ratio (Si), defined as a ratio
of surface occupied by containers in i-th row to all surface destined for container storage in this row,
x average height of container stacks (hc,t) throughout the
terminal,
x average height of container stacks (hc,i) in i-th row,
x average height of container stacks (hc,d) over a propagation path length,
x average height of container stacks (hc,r+1) in the row
behind the mobile station and causing the reflection of
radio waves.
On the basis of the analysis of propagation conditions in
each subarea of the terminal under investigation, the basic
transmission loss (LMCT) in such an environment may be expressed by the following equation:
LMCT > dB @
­ LLOS
°
® LCont
°L
¯ OffT
for
d d d1 (LOSArea),
for d1 d d d R (Containers Area), for d ! d R (Off-Terminal Area).
The particular components are:
x for the LOS Area:
LLOS > dB@
L0 4.2lg hb hc,1 11.6, L0 > dB@ 20log f 20lg d 27.6, and the (hb–hc,1) factor is related to the wave reflected from
containers in the first row of storage fields;
x for the Containers Area:
LCont > dB @
20lg f 25lg d 18lg hb hc , d 6.2lg hb hc , r 1 4lg I 21.8,
where the (hb–hc,d) factor is related to the path loss due to diffraction at the edges of containers on the propagation path
over the containers, where:
r
¦h
c ,i
hc , d
˜ Si
i 1
r
¦S
13.5lg St 4lg I 21.8,
R
and the (hb–hc,r+1) factor is related to the wave reflected from
the containers in the next row behind the mobile station;
x for the Off-Terminal Area:
¦h
c ,i
˜ Si
i 1
hc ,t
R
¦ Si
i 1
and the St factor reflects the influence of the number of containers in the whole container terminal area.
In the above equations r=1,2,…,R-1 means the number of
the last row of storage fields before the mobile station and R is
the number of all rows of storage fields.
Coefficients of the above equations were calculated on the
basis of empirical data and using multivariate linear regression
with the least-squares method, which minimizes the sum of
squared differences between measured path loss value and the
regression function. Statistical significance of particular coefficients was proved with 95% confidence interval (5% level of
significance) using the t-test with Student’s distribution, and
statistical significance of regression functions was proved with
the same confidence interval using the F-test with FisherSnedecor distribution.
IV. THE EVALUATION
Above described models were evaluated in order to use
them for designing mobile radio links in container terminal
environment. This evaluation was based on the measure of
matching measured data to mathematical models, namely
standard error of estimate (SEE), which is used to verify accuracy of the path loss models [10] and it is defined as follows:
N
1
N 1 i
¦ (L
SEE > dB @
m ,i
> dB @ Lc,i >dB @)2 , 1
where Lm,i is measured value of the basic transmission loss in
i-th position of the receiver (i=1,...,N), Lc,i means the basic
transmission loss value computed using propagation model for
i-th position, and N is a sample size.
Apart from this parameter, very important is a coefficient of
determination R2, which is a statistical measure of how well
the adjusted model approximates the real path loss values and
it is expressed by following equation [11]:
N
¦L
c ,i
i
where the (hb–hc,t) factor is related to the path loss due to diffraction at the edges of containers on the propagation path
over the containers, where:
i 1
38
LOffT > dB @ 20lg f 30lg d 18lg hb hc ,t where the L0 factor is related to the direct wave, expressed by
the well-known equation:
R2
i 1
N
Lm , av 2
¦ Lm,i Lm,av 2
,
i 1
where Lm,av – averaged value of measured basic transmission
loss.
PWT 2012 - Poznań, 14 grudnia 2012
XVI Poznań Telecommunications Workshop - PWT 2012
4
TABLE 1 VALUES OF STANDARD ERROR OF ESTIMATE AND COEFFICIENT OF DETERMINATION FOR SELECTED PROPAGATION MODELS
Model
Original scenario
SEE [dB]
R2
Suburban areas
5.2
0.758
Urban centers
5.7
0.718
Suburban centers
7.6
0.5
Urban centers
7.8
0.46
Suburban areas
5.0
0.785
LOS1
5.3
0.752
NLOS1
5.3
0.752
LOS Area
4.40
0.81
Containers Area
4.53
0.80
Off-Terminal Area
4.30
0.77
All Areas
4.45
0.82
Adjusted COST 231
Walfisch-Ikegami
Adjusted ITU-R P.1411 NLOS1
(§4.2.1 of [4])
Adjusted ITU-R P.1411 NLOS1
(§4.2.2 of [4])
Adjusted Empirical Model for
Fixed Radio Systems
The MCT Model
Comparison of SEE and R2values for described selected
propagation models is presented in Table 1. It is seen, that the
best results were obtained for the MCT model for each area of
the container terminal and for all areas in general.
V. CONCLUSION
Since there was no propagation model for designing mobile
radio networks in a container terminal environment, in practice
other models were used. Therefore, there was a need to adjust
existing models and to develop new empirical propagation
model for mobile radio links working in such a difficult environment as is the container terminal. The new model takes into
account all essential factors that occur in this environment and
that affect basic transmission loss of radio wave.
The MCT model is the first model for the accurate estimation of path loss in the investigated environment. It was developed on the basis of almost 290 thousand propagation path
measurements in a real container terminal environment, collected in accordance with the appropriate requirements. The
obtained standard error of estimate is 4.45 dB. What is more,
the obtained value of the coefficient of determination is 0.82,
what additionally proves the accuracy and usefulness of the
MCT model.
The MCT model may be used for frequencies from a range
of 500 MHz up to 4 GHz, propagation path length between
50 m and 620 m and the base station antenna height between
12 m and 36 m.
Described model is a prelude to elaborate more universal
model for the outdoor industrial environments in general. For
this reason a comprehensive measurement research in different
environments should be carried out. For obvious reasons this
should be done in international cooperation.
REFERENCES
[1]
ITU-R, SM.378-7, “Field-strength measurements at monitoring stations”, 2007.
[2] ITU-R, SM.1708, “Field-strength measurements along a rout with
geographical coordinate registration”, 2005
[3] W.C.Y.Lee, Mobile communications design fundamentals, Wiley, 1993
[4] ITU-R, P.1411-4, “Propagation data and prediction methods for the
planning of short-range outdoor radiocommunication systems and radio
local area networks in the frequency range 300 MHz to 100 GHz”, 2007
[5] D.J.Cichon, T.Kürner, “Propagation Prediction Models”, Digital Mobile
Radio Towards Future Generation Systems, COST 231 Final Report
[6] R.J.Katulski, J.Sadowski, and J.Stefanski, „Propagation path modelling
in container terminal environment”, Proceedings of IEEE 68th Vehicular Technology Conference, Calgary, Canada, pp. 1-4, 2008
[7] Ambroziak S.J., Katulski R.: “New empirical propagation model for
designing mobile radio networks in industrial environment” (in Polish),
Telecommunication review and telecommunication news, No. 4/2012,
pp. 165-168
[8] S.J.Ambroziak, R.J.Katulski, „On the usefulness of selected radio waves
propagation models for designing mobile wireless systems in container
terminal environment“, Proceedings of the XXX General Assembly and
Scientific Symposium of URSI, Istanbul, Turkey, 2011
[9] S.J.Ambroziak, R.J.Katulski, J.Sadowski, and J.Stefanski, „Propagation
Path Loss Modelling in Container Terminal Environment“, Vehicular
Technologies: Increasing Connectivity, InTech, 2011
[10] R.J.Katulski, The radio wave propagation in wireless telecommunication (in Polish), WKŁ, 2009
[11] H.Cramer, Mathematical Methods of Statistics, Princeton University
Press, 1999
XVI Poznańskie Warsztaty Telekomunikacyjne
39
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
The Effect of Voice Transmission over IP on
Text-Independent Speaker Verification
Performance
Waldemar Maciejko
Abstract—Despite the growing importance of Voice
Transmission over IP (VoIP) systems, there is still a shortage of
thorough analyses of VoIP transmission effect on speech signal,
speech and speaker recognition performance. In this study the
effect on speaker verification performance was investigated.
Automatic system based on EM-MAP-GMM modeling methods
was used. Four various coders of H.323 standard were
investigated with special emphasis placed on packet loss
phenomenon.
Index Terms— Automatic Speaker Verification, packet loss,
speech compression, voice over IP.
I. INTRODUCTION
T
he popularity of automatic speaker recognition as one of
the methods of biometric human identification, is
constantly growing. The natural consumer of this type of
technology is a bank sector. The voice biometrics is becoming
supplemental to traditional security methods like passwords.
Another potential beneficiary of voice biometrics’ discipline
are forensic sciences. Forensic speaker identification experts
may use not only aural-perceptual methods but also unbiased
parametric analysis based on automatic verification [14][19].
VoIP technology is an effect of a widespread access to the
internet and continuing development of IP technologies. The
most important advantages of VoIP are: lower costs of
telephone conversations and possibility of a parallel non-audio
data transmission. According to telecommunication market
predictions, there will be 88 millions of VoIP users in Western
Europe by the end of 2012, which represents a 240 % growth
since 2007 [4]. These market data show that the role of VoIP
is still growing. The effect of GSM and PSTN transmission
degradations on speaker verification performance was a
subject of many previous studies. The majority of
investigations focused at evaluating the influence of additive
noise like white and colored noise [6] as well as the influence
of non-linear spectral distortions [6][9]. Results showed that
this type of distortions caused the degradation of speaker
verification performance. Other researchers evaluated the
influence of audio GSM codecs, such as GSM 06.10, GSM
06.20 and GSM 06.60 [7][8]. All GSM and PSTN speech
transmission phenomenon lead to speaker verification
performance decline. Degree of degradation depends on
transmission technology.
40
II. VOICE TRANSMISSION OVER IP
Each telecommunication system makes up a collection of
transmission and switching devices. The older telephone
systems like Public Switch Telephone Network used the
circuit switching in which, two network nodes established a
communication channel. When the connection was established
the full bandwidth was guaranteed and the communication
path remained busy for the entire duration of session [3].
Alternatively, the VoIP systems use the packet switching. A
packet is a fundamental binary information in
telecommunication network, circulating between the terminals
as a series of bits. Packets consist two kinds of data: the
header and the body. The header contains, among others the
control information such as terminal IP, the number of packets
into which the message has been divided and synchronization
information. The body is a part comprising the coded and
divided speech signal. Each packet is individually transmitted
through the network. Unlike in the circuit switching, the
communication path remains busy only during packet
transmission. When packet transmission is finished, the path
becomes immediately accessible [3].
A. Packet loss and packet loss concealment
The IP transmission may cause many different packet
errors. Among others, packets can be lost, damaged or
delayed. Most of the IP traffic is under control of TCP
protocol, which provides solution of packet retransmission.
However, the VoIP technology uses UDP protocol that doesn’t
provide any recovery method.
The investigation of packet loss phenomenon reveals that
voice packets are lost in bursts. The IP traffic is described by a
two-state simple Gilbert model [15][16], which allows using
only two parameters to describe the loss process and provides
good approximation of this phenomenon [17]. Example of
Gilbert model is presented on Figure 1 [5].
PWT 2012 - Poznań, 14 grudnia 2012
Fig 1 Gilbert model
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
The probability p11 stands for conditional loss probability
(clp). Mean Burst Loss Length (MBLL) based on [15] is
computed as:
1
1 - clp
MBLL
(1)
Finally, based on equation (1), clp can be computed as
[15][16]:
clp 1 1
MBLL
(2)
The probability of being in state B, representing the mean loss,
is denoted as unconditional loss probability (Pz):
PZ
p G p 01 p B p11
(3)
where pG and pB are the stationary probabilities pG+ pB=1 and
Pz= pB. Consequently, p01 can be expressed as
p01
PZ (1 - clp)
(1 - PZ )
(4)
Gilbert model based on equations (4) and (2) is described by
[17]:
p01
PZ
MBLL˜ (1 - PZ )
(5)
2
the test utterance Y and the voice of target speaker M is
quantified by similarity score, according to Bayes’ defined by:
accept Ÿ match score !
p(Y M)
p(Y ™M UBM )
(6)
The likelihood ratio score is compared to a threshold δ and
in case of a match the speaker is accepted. The target speaker
model M can be calculated with maximum accuracy by using
training speech. The model ¬MUBM, called Universal
Background Model (UBM) represents the entire spectrum of
possible alternatives to the hypothesized speaker and can not
be estimated with maximum accuracy [12]. The UBM
represents a set of selected speakers according to Gaussian
distribution. The criterions for speaker selection to the
alternative population are for example: quality of speech,
gender and language. The exemplary application of the
automatic speaker verification based on UBM is presented on
the figure 2.
To determine the likelihood functions of the nominator and
denominator shown on equation 6, is the main objective of an
automatic verification model. This functions depend among
others, on the low level of used spectral features. The current
system extracts 16 MFCC coefficients, 16 Δ MFCC first order
derivatives and the signal log energy using the HTK binary
HCopy [20]. The core of the Mel Frequency Cepstral
Coefficients is the short-term Fourier spectrum. The melfrequency analysis uses filters spaced linearly at low
frequencies and logarithmically at high frequencies. Mel-scale
is related to human pitch sensation. Such signal representation
captures
individually
and
phonetically
important
characteristics of speech [10]. The MFCC signal
representation underlies the subsequent speech analysis.
B. Speech signal compression
The final effect of speech signal compression (quality and
compression ratio) depends on applied codec. Usually, higher
compression causes quality loss and allows to use more
narrow band to voice transmission.
The first standard of audio-video transmission by IP
network was defined by H.323. This standard uses many audio
codecs [2], of which: G.711 (codec with the highest bit rate),
G.729 (codec with medium bit rate) and G.723.1 (codec with
the lowest bit rate) were investigated in this study and are
summarized in Table 1.
TABLE I
SELECTED H.323 STANDARD AUDIO CODERS [2].
Fig 2 Basic building blocks of speaker verification system.
Standard
Compression
algorithm
Frame
(ms)
Compression
ratio
Bit rate kbps
G.711
G.729
G.729
G.723.1
PCM
+CS-ACELP
CS-ACELP
MP-MLQ
0,125
10
10
30
1:1
9:1
8:1
10:1
64,0
6,4
11,8
6,3
III. AUTOMATIC SPEAKER VERIFICATION
The task of automatic speaker verification is to determine if
a hypothesized test utterance Y was spoken by a hypothesized
speaker M. The general approach to this task is to test the
likelihood ratio. In automatic systems, the similarity between
The next step is the cepstral mean variance-normalization
(CMS). This technique permits to use recognizer even in case
of a mismatch between training and testing environment [11].
The applied algorithm uses the voice automatic detection
(VAD) and the CMS normalization is conducted only on a
speech, non zero frames. The VAD method is based on a 2 nd
order GMM of energy distribution. The assumption of this
method is a difference between the speech energy and the
energy of non speech frames [23].
In text-independent speaker recognition there is no previous
knowledge of what the speaker will say. Over the past dozen
XVI Poznańskie Warsztaty Telekomunikacyjne
41
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
years, Gaussian mixture models (GMM) has been remaining
the most successful method of estimation of text-independent
speaker model. The GMM is based on assumption that the
probability density function of some features, being defined in
a multidimensional space, can be estimated with function of
mixture models. Each model of the mixture represents some
high level phonetic sound [12]. Gaussian mixture model is a
weighted linear combination of R Gaussian densities pi(y).
p( y | M )
3
presented on Figures 3-6 in the form of Detection Error
Tradeoff curves [18].
R
¦ Z p ( y)
i
(7)
i
i 1
where R is the model order, ωiis the vector of weights.
Gaussian PDF pi(y) for D-dimensional spectral features is
given by:
pi ( y)
1
(2 ˜ S )
D/2
6i
1/ 2
­ 1
½
˜ exp ® ˜ ( y Pi )c(6i ) 1 ( y Pi )¾
¯ 2
¿
(8)
where μiis the mean vector and Σi is a covariance matrix where
i=1…R. Because of an assumption that MFCC feature vectors
are independent, the likelihood of appearance of a sequence of
MFCC vectors Y = {y1, y2, ... yT} in model M is defined as:
log p(Y | M )
T
¦ log p( y
t
| M)
Fig 3 Effect of voice transmission over IP on text-independent speaker
verification performance for varying degrees of packets loss, from 0 up to
25%. The G.711 coder was used.
(9)
t 1
where model M is described by parameters {μi, Σi, ωi}, where i
= 1, …, R
To estimate the Universal Background Model (¬MUBM), the
Expectation Maximization (EM) algorithm described in [12],
[13], with model order 128 is used. The EM method is an
iterative algorithm for approximating Maximum Likelihood in
the case of incomplete data. To estimate target speaker model
(M), Maximum a Posteriori algorithm is used where relevance
factor r equals 14 [12][13]. MAP procedure allows to adapt
the parameters of the Universal Background Model using the
speaker’s training speech. To build probabilistic speaker
models and background models, the BECARS C++ library
was used [22].
IV. SPEAKER VERIFICATION OVER IP NETWORKS
Fig 4 Effect of voice transmission over IP on text-independent speaker
verification performance for varying degrees of packet loss, from 0 up to 25%.
The G.723.1 6,3 kbps coder was used.
EXPERIMENTS
A. Database
The verification experiments were conducted on a database
containing recordings of 38 polish language speakers
performed with a high quality condenser microphone in PCM
format with 44,1 kHz sampling frequency and 16 bit
resolution. Both the speakers’ training and the test utterance
were performed on approximately 30-second long,
phonetically rich sentences.
B. Packet loss and H.323 audio coders speakers
verification performance degradation
The speaker database was transcoded with coders described
in section II.B. The simulation of packet loss was performed
according to Gilbert model (described in section II.A). Packet
loss phenomenon was analyzed at varying degrees of packet
loss from 0 up to 25%. The results of the performed tests are
42
Fig 5 Effect of voice transmission over IP on text-independent speaker
verification performance for varying degrees of packet loss, from 0 up to 25%.
The G.729 11,8 kbps coder was used.
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
4
References
[1]
[2]
[3]
[4]
[5]
[6]
Fig 6 Effect of voice transmission over IP on text-independent speaker
verification performance for varying degrees of packet loss, from 0 up to 25%,
G.729 6,4 kbps coder was used.
In all plots, Pz represents the probability of a packet loss,
while EER represents Equal Error Rate and was marked as
black dots.
[7]
[8]
[9]
[10]
V. SUMMARY
Speaker verification performance is summarized on Fig 7.
Notably, the results do not depend on applied speech
compression method. There are no obvious differences
between coders with high bit rate and low bit rate.
Additionally, the obtained results are opposite to scores
presented in [1], where bit rate was an important parameter. A
source of this dissonance lies probably in the methods used for
coding of target and test utterance. Bit rate becomes important
if both test and target are coded with different methods [1].
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Fig 7 Summary of the experiment. Equal Error Rate as a function of the
probability packet loss for various methods of speech compression.
The main factor of speaker verification performance
degradation is a packet loss. The dependence between packet
loss probability and Equal Error Rate is almost linear,
regardless of applied codec, which is shown on Fig 7. On the
other hand, the highest degree of packet loss ensuring
acceptable voice quality must not exceed 1%. At this rate, the
EER factor has increased by no more than 1%, compared to no
packet loss .
[21]
[22]
[23]
L. Besacier, A. M. Ariyaeeinia, J. S. Mason, J. F.Bonastre, P. Mayorga,
C. Fredouille, S. Meignier, J. Siau, N. W. D. Evans, R. Auckenthaler and
R. Stapert,“Voice biometrics over the internet in the framework of
COST action 275” , EURASIP Journal on Applied Signal Processing
2004:4, 466-479, Hindawi Publishing Corporation.
ITU-T H.323 Series H: Audiovisual and multimedia systems.
Infrastructure of audiovisual services – Systems and terminal equipment
for audiovisual services. Packet-based multimedia communications
systems. Recommendation ITU-T H.323, International
Telecommunication Union12/2009.
A. Jajszczyk, “Wstęp do telekomunikacji”, Podręczniki akademickie
WNT 2009.
Fierce Enterprise Communications (2008, March) Europe VoIP to reach
88 million,Available:http://www.fierceenterprisecommunications.com.
E. N.Gilbert,“Capacity of a burst-noise channel”,The Bell System
Technical Journal, September 1960.
D. A. Reynolds, M. A. Zissman, T.F. Quatieri,G. C. O’Leary,B.
A.Carlson,“The effect of telephone transmission degradation on speaker
recognition performance”, Acoustics, Speech, and Signal Processing,
1995, ICASSP-95.
L. Besacier, S. Grassi, A. Dufaux, M. Ansorge, F. Pellandini, GSM
speech coding and speaker recognition, ICASSP 2000.
C. Byrne, P. Foulkes, “The mobile phone effect on vowel formants”
International Journal of Speech Language and the Law, Vol 11. No 1
2004.
D.A. Reynolds, “The effects of handset variability on speaker
recognition performance: experiments on the switchboard corpus”,
Acoustics, Speech and Signal Processing, ICASSP-96, Conference
Proceedings.
S.B. Davis, P. Mermelstein, “Comparison of parametric representations
for monosyllabic word recognition in continuously spoken sentences”,
IEEE Transactions on Acoustics, Speech and Signal Processing28(4).
1980. s. 357–366.
S. Furui, “Cepstral analysis technique for automatic speaker
verification”, IEEE Transactions Acoustics, Speech, Signal Processing,
ASSP-29. 1981. s. 254-272.
D. A. Reynolds, T. F. Quatieri, R. B. Dunn, “Speaker verification using
adapted gaussian mixture models”, Digital Signal Processing, nr. 10,
2000, s. 19– 41.
D. A. Reynolds, R. C. Rose, “Robust text-independent speaker
identification using gaussian mixture speaker models”, IEEE
Transactions on Speech and Audio Processing, nr. 3(1), 1995, s.72–83.
P. Rose “Forensic speaker identification”, Taylor & Francis. 2002. s. 72.
H. Sanneck, “Packet loss recovery and control for voice transmission
over the interne”, Ph.D. thesis Technischen Universität Berlin 2000,
unpublished.
S. Jelassi, G. A. Rubino, “A study of artificial speech quality assessor of
VoIP calls subject to limited bursty packet losses” EURASIP Journal on
Image and Video Processing 2011, 2011:9.
S. Mohamed, G. Rubino, M. Varela Perfomance evaluation of real-time
speech through a packet network: a random networks-based approach.
Performance evaluation. An international Journal. 57 (2004) 141-161.
A. Martin, G. Doddington, T. Kamm, M. Ordowski, M. Przybocki, The
DET curve in assessment of detection task performance In Proc.
Eurospeech ’97, pages 1895–1898, Rhodes, Greece 1997.
W. Maciejko, “Biometryczne rozpoznawanie mówców w
kryminalistyce”, Problemy Kryminalistyki 275, Warszawa 2012.
S. Young, G. Evermann, M. Gales, T. Hain, D. Kershaw, X. Liu, G.
Moore, J. Odell, D. Ollason, D. Povey, V. Valtchev, P. Woodland,
(2009),“The HTK book v3.4”,Cambridge 2009.
O. Viikki, K. Laurila, “Cepstral domain segmental feature vector
normalization for noise robust speech recognition”,Speech
Communication 25 (1998) 133-147.
C. Mokbel, H. Mokbel, R. Blouet, G. Aversano, BECARS library and
tools for speaker verification 1.1.0, April 2005
I. Margin-Chagnolleau, G. Gravier, R. Blouet Overview of the 20002001 ELISA consortium research activities ISCA A speaker Odyssey
The Speaker Recognition Workshop Crete 2001.
XVI Poznańskie Warsztaty Telekomunikacyjne
43
2012
Maximization of Utility in Computer Network
with Application of Game Theory
Magdalena Turowska
Abstract—In this paper we propose joint multipath routing
and rate allocation under assumption of strategic behavior of
users using mechanism design. We use decomposition techniques
and pricing mechanism that generate efficient allocations,
maximize net utilities and can be implemented as the distributed
algorithm. The algorithm is fair and robust against manipulation
from users who usually are selfish and behave strategically.
allocation. We formulate the problem of joint rate allocation
and routing and propose the decomposition in Section III. In
Section IV we present mechanism design for the subproblem
dependent on the behavior of users. The conclusions are
presented in Section V.
II. FLOW MODEL
network,
network
utility
Index
Terms—computer
maximization, rate allocation, routing, mechanism design, game
theory
I. INTRODUCTION
A
important approach to design a computer network
system is by formulating the design as the aggregate
maximization of the utilities of the nodes subject to physical
and economic constraints in the network. This is referred to as
Network Utility Maximization (NUM). Much research effort
has been put in the design of distributed algorithms for NUM
[10]-[13]. The main ingredient to obtain distributed algorithms
is the decomposition techniques, widely used in optimization
theory [6].
One approach to joint routing and rate allocation is to allow
multi-path routing, i.e. a source can transmit its data on
multiple paths to its destination. In this formulation, a decision
is decomposed into two – how much traffic to send (rate
allocation) and how to distributed it over the available paths
(multi-path routing) – in order to maximize aggregate network
utility.
The main disadvantage of existing solutions on resource
allocation based on Kelly’s approach [8] is that they strongly
depend upon the users declaring their resource demand in
a truthful way. They lack robustness against strategic
manipulation from users.
This paper presents a decomposition method for joint rate
allocation and routing complemented by explicitly considering
the strategic behaviors of users using mechanism design [1],
[2], [3], [4], [5].
The paper is organized as follows: In Section II we present
the flow model in computer network. In Section II we
introduce the utility maximization framework for resource
N
Magdalena Turowska is with the Institute of Informatics, Wroclaw
University of Technology, Wyb. Wyspianskiego, 30-570 Wroclaw (e-mail:
[email protected]).
44
We consider a data network with a topology which can be
represented by a directed graph. A collection of nodes, labeled
= 1, … , can send, receive, and relay data across
communication links. A communication link is represented by
an ordered pair , of distinct nodes. The presence of a link
, means that the network is able to send data from the start
node to the end node . The links are labeled by integers
= 1, … . The network topology can be represented by a
matrix =
∈
, whose entry
= 1if ∈
,
and
= 0 otherwise,where
is the
= −1 if ∈
set of links that are incoming to node , and
is the set of
links that are outgoing from node .
We identify the flows in the network by their destinations,
i.e., flows with the same destination are considered as one
single commodity, regardless of their sources. The destination
nodes are labeled by = 1, … , = 1, … ,
denotes index
of transmission demand from source to destination , ! "
is data rate of
-th demand/user with source
and
destination .
On each link , $ ≥ 0 is the amount of flow destined for
node . At each node , flows satisfy the constraint
'()
&!
"*+
"
+ & $
∈-
= & $ , = 1, … , .
∈.
We assume finite capacity of links, thus
&$
*+
≤0,
where 0 is the capacity of link .
This model describes the average behavior of data
transmissions, i.e., the average data rates on the
communication links, and ignores packet-level details of
transmission protocols and forwarding mechanisms. The link
capacity in practical communication systems should be
defined appropriately, taking into account packet loss and
PWT 2012 - Poznań, 14 grudnia 2012
retransmission, so the flow conservation law holds for the
effective throughput or goodput [9].
III. NETWORK UTILITY MAXIMIZATION (NUM)
Many network resource allocation problems can be
formulated as constrained maximization of some utility
function [7, 9]. In the Network Utility Maximization (NUM)
framework, each transmission demand (or user) has its utility
function and link bandwidths are allocated so that network
utility (i.e., the sum of all users’ utilities) is maximized. A
utility function can be interpreted as the level of satisfaction
attained by a user as a function of resource allocation.
Efficiency of resource allocation algorithms can thus be
measured by the achieved network utility. Utility functions can
also be interpreted as the ‘knobs’ to control the tradeoff
between efficiency and fairness.
To formulate the joint rate allocation and routing problem in
communication network we need to consider the following
network utility maximization problem
'()
max & & & 7
()4 ,5)6
"
*+ *+, "*+
8
!
"
where λ , β
are the Lagrange’s multipliers (prices)
associated with link capacity and network structure
constraints, respectively.
The convex optimization (optimization a convex or concave
function over a convex constraint set) has useful Lagrange
duality properties, which lead to decomposability structures.
Lagrange duality theory links the original maximization
problem, termed primal problem, with a dual minimization
problem, which sometimes readily presents decomposition
possibilities [9]. The basic idea in Lagrange duality is to relax
the original problem by transferring the constraints to the
objective in the form of a weighted sum.
IV. DECOMPOSITION
The main idea of decomposition is to decompose the
original large problem into distributively solvable subproblems
which are then coordinated by a high-level master problem by
means of some kind of signaling [6].
Dual decomposition methods is used correspond to
a resource allocation. The master problem sets the price for the
resources to each subproblem, which has to decide the amount
of resources that can be used depending on the price.
The Lagrange dual problem associated with the primal
problem under consideration is given by
'()
&!
"*+
" + & $
!
∈-
"
≥ 0,
A,B
= & $
$
∈.
≥0
λ > 0,
where
*+
over ! " , $ , for = 1, … ; , = 1, … ; = 1, …
,
where 7 " is the utility of k-th network transmission from
source to destination (k-th user).
The objective function is differentiable and strictly concave
and the feasible region is compact; hence a maximizing value
of ! " , $ exists and can be found by Lagrangian methods.
The Lagrangian of formulated maximization problem is
defined as
= & & &7
*+ *+, "*+
8
"
"
'()
*+ *+, "*+
8
−& & λ $ + & & & >
*+ *+
*+
8
"
DEF
x, β = max & & & 7
()4
5
*+
y, λ, β + & λ 0
*+
for
!
"
= 1, … ;
*+,…,
*+,…,
"*+,…,DEF
H8I G*+
"
!
−> !
"
" y, λ, β = max{− & & λ $ − & >
5)6
−
$ + &λ 0
5
and
−& & &> !
x, β +
λ, β = : x, y, λ, β =
!
(2)
subject to constraints:
&$ ≤ 0
'()
λ, β min
subject to constraints:
H8I
, = 1, … ;
= 1, …
and y = $
*+,… .
*+,…
$
(3)
},
(4)
; where x =
It is well known that for a convex optimization, a local
optimum is also a global optimum and solving the problem
XVI Poznańskie Warsztaty Telekomunikacyjne
45
after decomposition is equivalent to solving the original
(primal) problem.
For solving dual problem we propose to use three-level
decomposition presented in Fig. 1.
level. R and > are usually interpreted as users (demands) and
links prices.
V. MECHANISM DESIGN
Fig. 1. A hierarchical dual decomposition used to solving
NUM problem (L- number of links, P- number of paths).
The first (upper) level solves problem (2), the second (middle)
level solves problems (3) and (4), which can be decomposed.
On the third (the lowest) level each source and each link can
independently compute ! ∗ " > and$ ∗ ∑H8I > ; λ
as
min 7
"
()4
!
−> !
"
"
(5)
and
min λ $ − & >
5)6
!
min & & & 7
A6 ,O()
*+ *+, "*+
8
"
!∗
"
− & & & > !∗
*+ *+, "*+
8
− & & λ $ ∗ P&
H8I
+&& &>
*+ *+
*+
8
$∗ &
H8I
0 − ∑ S *+, S 8 ∑ ∈-
$
∑ S*+, S8
>
>̅
"
>
W
−
> ;λ Q
> ; λ + & λ 0
*+
, where
for
= 1, … ;
, = 1, … ;
= 1, …
! ∗ " > and$ ∗ ∑H8I > ; λ are the optimal ! ∗ " and
$ ∗ for the given R and > – the results obtained on the lower
46
∑ ∈.
(6)
+
=
"
'()
.
"
=
'
U
()
∑ S*+, S8 ∑"*V
>
∑ S*+,
8
S"
−>
"
S
which represents the average price per unit of resource from
the other users, and
>
'()
=
for each user (demand) and initial price
respectively. On the upper level the problem we can formulate
as
'()
"
Step 1:
Each user receives information from the network:
$
*+
8
As results of the decomposition we have received to solve
subproblems A which can be resolved by users (sources)
independently and subproblems B which links can solve
independly.
In order to prevent selfish behaviors of users, because such
behaviors make impossible maximization of the network
utility, we introduce game-theoretic mechanism for
subproblems A. We refer to the class of noncooperative games
derived from such mechanism as auction, in which Nash
equilibrium is usually assumed as a solution concept. A Nash
equilibrium captures the notation of a stable solution and is the
solution from which no user (demand/player) can individually
improve his utility by deviating.
In this paper we assume that each user is restricted to
communicate his demand to the network with the price that
user is willing to pay for the resource. We present distributed
and dynamic process that converges to Nash equilibrium.
Initialization: We assume an initial resource allocation
"
=
& 0 −
∈.
−
&
S *+, 8
&
& $ −
∈'()
U
&!
S *+, S 8 "*V
S"
+!
"
(7)
which is the excess resource demand excluding the demand
from -th user with source and destination .
We define the cost of using resource:
X
"
!
",
>
"
̅
≜!
+{>
" >
"
− >̅
"
"
+
1−Z W
"
−max{0, Z W
PWT 2012 - Poznań, 14 grudnia 2012
−!
"
"
−!
−
"
}}[
which represents the price that user pays for ! " amount of
resource corrected by penalty that user pays due to the
mismatch of its price to the average price of the other users
and by the term that is introduced to prevent the solution from
reaching an inefficient Nash equilibrium.
Step 2:
Each user calculate !
max
()4 ,O()4
"
7
and >
!
"
"
"
"
!
", >
"
,
that means
max
()4 ,O()4
+{−>
"
7
+ >̅
"
"
!
1−Z W
+max{0, Z W
"
̅
−!
"
−!
" >
"
"
−!
"
}}[ .
+
"
REFERENCES
+
The maximization of the utility can only occur when
>
̅
" =>
"
1−Z W
"
−!
"
VII. CONCLUSION
In this paper we study the problem of joint rate allocation
and routing in the network with competing and selfish users.
The important feature of proposed mechanism is indicated: it
gives the optimal solution of allocation problem in spite of
autonomous of users. The simulation and implementation for
the proposed framework are part of future research.
as
−X
means that the Nash equilibrium and the optimal solution are
the same.
+
+ max{0, Z W
"
−!
"
},
where Z is a parameter of algorithm.
The draft of algorithm for joint rate allocation and routing
with application of game-theoretic mechanism design is
following:
1) Each source nodes locally computes >̅ " and W " , and
transmit they to the users.
2) Each user based on received information calculates
> " and ! " >̅ " .
3) Each
network
node
calculate
locally
'() ̅
∑
∑
$
H8I
"*V > " , λ , using current value of λ
and >̅ " .
4) Each link (node where the link begin) updates R .
5) Return to Step 1.
VI. PRICE OF ANARCHY
The price of anarchy of the game is the most popular
measure of efficiency and is defined as the ratio between the
worse objective function value of an equilibrium of the game
and that of an optimal outcome [1].
The mechanism presented in Section V implements in Nash
equilibrium problem described in Section III. It can be shown
that Nash equilibrium of mechanism presented in Section V
satisfy the optimality conditions of problem from Section III.
The necessary and sufficient conditions for the efficient
allocations can be determined from Karsh-Kuhn-Tucker
(KKT) conditions. In order to show that the mechanism in
Section V implements in Nash equilibrium allocation problem
presented in Section III, it can be shown that the Nash
allocation satisfies the KKT conditions for primal problem.
This result indicates that the price of anarchy equals 1, which
[1] N. Nissan, T. Roughgarden, E. Tardos, V. Vazirani (eds),
“Algoritmic Game Theory”, Cambridge University Press, 2007.
[2] S. Yang, B. Hajek, “VCG-Kelly Mechanisms for Allocation of
Divisible Goods: Adapting VCG Mechanisms to OneDimensional Signals”, IEEE Journal on Selected Areas in
Communications, vol. 25, no. 6, August 2007.
[3] M. Dramitinos, G. D. Stamoulis, C. Courcoubetis, “An auction
mechanism for allocating the bandwidth of networks to their
users”. Computer Networks, 2007, nr 51, pp. 4979–4996.
[4] R. Jain, J. Walrand. “An efficient mechanism for network
bandwidth auction”. Network Operations and Management
Symposium Workshops 2008, NOMS Workshops 2008, pp.
227–234.
[5] Y. Su. M. van der Schaar, "Linearly Coupled Communication
Games", IEEE Trans. Commun., vol. 59, no. 9, pp. 2543-2553,
Sep. 2011.
[6] D.P. Bertsekas, Network Optimization: Continuous and Discrete
Models. Belmont, MA: Athena Scientific, 1998.
[7] A.
Ephremides,
B.Hajek,
Information
theory
and
communication networks: An unconsummated union, IEEE
Transaction on Information Theory, vol. 44, pp. 2416–2434,
Sept. 1998.
[8] F. P. Kelly, A. Maulloo, A. Tan, Rate control for communication
networks: Shadow prices, proportional fairness and stability,
Journal of Operational Research Society, vol. 49, pp. 237–252,
Mar. 1998.
[9] S. Low, D. Lapsley, Optimization flow control—I: Basic
algorithm and convergence, IEEE/ACM Transaction on
Networking, vol. 7, pp. 861–874, Dec. 1999.
[10] A. Ouorou, P. Mahey, J.-PH. Vilal, A survey of algorithms for
convex multi-commodity flow problems, Management Science,
vol. 46, pp. 126–147, Jan. 2000.
[11] D. Palomar, M. Chiang, Alternative Distributed Algorithms for
Network Utility Maximization: Framework and Applications.
IEEE Transaction on Automatic Control, vol. 52, no. 12, pp.
2254-2269, Dec. 2007
[12] L. Tassiulas, A. Ephremides, Jointly optimal routing and
scheduling in packet radio networks, IEEE Transaction on
Information Theory, vol. 38, pp. 165–168, Jan. 1992.
[13] Y. Yi, M. Chiang, Stochastic Network Utility Maximization.
European Transaction on Telecommunications. vol. 22, pp. 122, 2008.
XVI Poznańskie Warsztaty Telekomunikacyjne
47
48
PWT 2012 - Poznań, 14 grudnia 2012
SesjaII
Sesja specjalna z okazji jubileuszu 50-lecia
pracy prof. Andrzeja Dobrogowskiego
Synchronizacja w telekomunikacji
XVI Poznańskie Warsztaty Telekomunikacyjne
49
50
PWT 2012 - Poznań, 14 grudnia 2012
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
System Wspomagania Synchronizacji
A. Dobrogowski, M. Jessa, M. Kasznia, K. Lange
Streszczenie—W artykule przedstawiono genezĊ, podstawowe
elementy skáadowe, rozwój oraz wáaĞciwoĞci systemu
wspomagania synchronizacji (SWS) opracowanego przez zespóá
zajmujący siĊ w Politechnice PoznaĔskiej synchronizacją sieci
telekomunikacyjnych.
Sáowa
kluczowe
—
Synchronizacja
sieci,
pomiary
synchronizacyjne, aparatura pomiarowa, zalecenia ITU-T
W
I. WPROWADZENIE
SPÓàCZESNE sieci telekomunikacyjne cechuje
róĪnorodnoĞü urządzeĔ i systemów transportu
informacji pochodzących z wielu Ĩródeá. NiezaleĪnie od
charakteru sygnaáu Ĩródáowego jest on zawsze zamieniany na
sygnaá cyfrowy. PrzepáywnoĞci strumieni bitów przesyáane
w wyĪszych warstwach sieci telekomunikacyjnej dochodzą do
40 Gbit/s dla pojedynczej dáugoĞci fali. Po zastosowaniu
zwielokrotnienia falowego mogą przekraczają nawet 1Tbit/s.
Podstawową barierą w dalszym zwiĊkszaniu tej
przepáywnoĞci w odniesieniu do jednego wáókna optycznego
nie są juĪ wzglĊdy techniczne czy technologiczne lecz
ekonomia. Ewentualne straty operatora spowodowane
uszkodzeniem jednego wáókna ze strumieniem o
przepáywnoĞci 40Gbit/s lub wiĊkszej są na tyle duĪe, Īe musi
on zapewniü szybką protekcjĊ uszkodzonego áącza. Czas
wymiany áącza na áącze zapasowe powinien byü jak
najkrótszy, na przykáad krótszy od 50 ms w przypadku
Synchronicznej Hierarchii Cyfrowej (SDH) uĪywanej obecnie
w warstwie transportowej sieci. NiezaleĪnie od tego, czy w
dalszej perspektywie transport informacji na duĪe odlegáoĞci
pomiĊdzy róĪnymi punktami warstwy dostĊpowej sieci bĊdzie
realizowaá system SDH czy jakiĞ inny, oczekuje siĊ, Īe
dáugoterminowy báąd czĊstotliwoĞci sygnaáów taktujących
strumienie bitów nie bĊdzie wiĊkszy od 10-11. Spodziewamy
siĊ takĪe, Īe wolnozmienne fluktuacje fazy sygnaáów
taktujących nie przekroczą 21 Ps, gdzie 18 Ps przypada na
áącze, a 3 Ps przypadają na dowiązanie fazy do
miĊdzynarodowego czasu koordynowanego UTC [1-6]. Cel
ten moĪna osiągnąü albo poprzez wyposaĪenie kaĪdego
urządzenia cyfrowego w super dokáadny zegar, co jest zbyt
drogie, albo poprzez zbudowanie systemu synchronizacji
taktów strumieni cyfrowych do taktu tzw. zegara odniesienia
Autorzy są pracownikami Wydziaáu Elektroniki i Telekomunikacji
Politechniki PoznaĔskiej (e-mail [email protected];
[email protected];
[email protected];
[email protected]).
(ang. Primary Reference Clock), nazywanego takĪe zegarem
gáównym odniesienia lub zegarem gáównym sieci.
ĩądanie wysokiej jakoĞci taktowania strumieni cyfrowych
w początkowym okresie rozwoju sieci cyfrowych wynikaáo
przede wszystkim z koniecznoĞci unikania báĊdów transmisji
spowodowanych brakiem synchronicznoĞci strumieni bitów
wymienianych
pomiĊdzy
róĪnymi
wĊzáami
sieci
telekomunikacyjnej. Obecnie to nie wystarcza. Dokáadne
taktowanie, w szczególnoĞci minimalne wartoĞci fluktuacji
fazy, są niezbĊdne samym usáugom. Trudno sobie wyobraziü,
przesyáanie na duĪe odlegáoĞci obrazu ruchomego HDTV,
a w niedalekiej przyszáoĞci takĪe 3D, bez zapewnienia ĞciĞle
okreĞlonych relacji fazowych w przestrzenie oddalonych
obiektach sieci telekomunikacyjnej. Istotnym elementem,
który moĪe zapewniü odpowiednią jakoĞü taktowania, áącznie
z procesem kontroli tej jakoĞci, jest system wspomagania
synchronizacji opracowany w Instytucie Elektroniki
i Telekomunikacji (od 2006 roku na Wydziale Elektroniki
i Telekomunikacji) Politechniki PoznaĔskiej.
II. GENEZA I ROZWÓJ SYSTEMU WSPOMAGANIA
SYNCHRONIZACJI (SWS)
W zespole zajmującym siĊ zagadnieniami synchronizacji
sieci
telekomunikacyjnej
opracowano
koncepcjĊ,
zaprojektowano i wdroĪono do produkcji siáami wáasnymi
i wspóápracujących firm, nastĊpujące urządzenia tworzące
SWS:
1. Unikatowy w skali Ğwiatowej system pomiarowy do
kompleksowego badania wolnozmiennych parametrów
sygnaáów taktowania i synchronizacji na zgodnoĞü z normami
miĊdzynarodowymi. Najpierw byá to system SP-2000, od roku
1999 system SP-3000, a od 2009 roku – system SP-4000.
2. Zestaw sond pomiarowych do pobierania sygnaáów
badanych z pracujących linii telekomunikacyjnych bez
pogarszania jakoĞci informacji transmitowanej w sieci.
3. ħródáo sygnaáu czasu i czĊstotliwoĞci z termostatowanym
generatorem kwarcowym lub rubidowym, synchronizowane
z satelitarnego globalnego systemu pozycjonowania GPS.
4. Unikatowe
Ĩródáo
czĊstotliwoĞci
wzorcowej
z generatorem rubidowym i automatyczną kompensacją
starzeniowych zmian czĊstotliwoĞci generatora za pomocą
zewnĊtrznego sygnaáu taktowania lub sygnaáu z odbiornika
GPS. ħródáo pozbawione jest báĊdu retrygowalnoĞci, dziĊki
wbudowaniu podtrzymania zasilania na czas transportu.
5. Tanie i proste w obsáudze dystrybutory sygnaáów
taktowania DST pozwalające na zwielokrotnienie sygnaáu
wejĞciowego podstawowego lub zapasowego na wiele
sygnaáów wyjĞciowych. W dystrybutory wbudowano
XVI Poznańskie Warsztaty Telekomunikacyjne
51
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
sygnalizacjĊ alarmową zwarciową, co pozwala operatorowi na
bezinwestycyjnie (za pomocą istniejących zasobów sieci
transmisyjnej
lub
komutacyjnej)
zorganizowanie
scentralizowanego nadzoru nad obecnoĞcią wejĞciowych
i wyjĞciowych sygnaáów synchronizacji w przestrzennie
oddalonych obiektach.
Prace nad SWS rozpoczĊto z inicjatywy prof. Andrzeja
Dobrogowskiego w roku 1992 od budowy Ĩródáa sygnaáu
czasu i czĊstotliwoĞci synchronizowanego sygnaáami
powstającego wtedy globalnego systemu pozycjonowania
GPS (system oficjalnie oddano do uĪytku w roku 1994). Rok
póĨniej ruszyáy prace nad wielokanaáowym systemem
pomiarowym do kompleksowego badania wolnozmiennych
parametrów sygnaáów taktowania i synchronizacji na
zgodnoĞü z normami miĊdzynarodowymi. System do roku
1998 nie miaá swojego odpowiednika w postaci urządzenia
oferowanego przez firmĊ zagraniczną. PrzyjĊto wówczas
nowatorską w skali Ğwiatowej koncepcje umieszczenia
w jednej obudowie wielu urządzeĔ sterowanych za
poĞrednictwem standardowego, zewnĊtrznego wobec systemu,
komputera personalnego. DziĊki tej koncepcji nie tylko
obniĪono cenĊ systemu pomiarowego ale znacząco wydáuĪono
jego eksploatacjĊ. Dostosowanie do aktualnych wymagaĔ
odbywa siĊ poprzez wymianĊ oprogramowania i ewentualnie
samego komputera na nowszy.
Opracowanie
systemu
wspomagania synchronizacji
poprzedziáa budowa w 1990 roku modelu piĊciowĊzáowej
sieci synchronizacyjnej, w którym generatory kwarcowe
termostatowane OCXO trzech wĊzáów byáy sterowane za
pomocą komputera a czwarty generator miaá sterowanie
wáasne, zrealizowane za pomocą mikroprocesora Z80. Piąty
generator nie podlegaá regulacjom. W roku 1992 opracowano
dwukanaáowy miernik báĊdu przedziaáu czasu (symbol TIEM1) zbudowany w postaci karty montowanej w komputerze PC.
Oprogramowanie pracowaáo pod kontrolą systemu DOS.
W tym samym roku powstaáo Ĩródáo STS-1 sygnaáu czasu
synchronizowane
sygnaáami
Globalnego
Systemu
Pozycjonowania (GPS). Oprócz dowiązania do czasu UTC
z báĊdem mniejszym od +/-300 ns Ĩródáo dostarczaáo kilku
czĊstotliwoĞci wzorcowych np. 10 MHz i 1 Hz. W roku 1993
opracowano i uruchomiono „Wzorzec czasu i czĊstotliwoĞci
STFS/GPS synchronizowany sygnaáami GPS”, który oprócz
dowiązania do czasu UTC z báĊdem mniejszym od +/-150 ns,
STFS/GPS dostarczaá szeĞciu czĊstotliwoĞci wzorcowych
2048
kHz
na
potrzeby
synchronizacji
sieci
telekomunikacyjnej. W tym samym roku opracowano takĪe
koncepcjĊ systemu pomiarowego SP-2000. NastĊpnie, w 1994
roku, Instytut àącznoĞci na zlecenie Telekomunikacji Polskiej
(TP) opracowuje „Wymagania Techniczno-Eksploatacyjne na
Wielokanaáowy Miernik BáĊdu Przedziaáu Czasu (MBPC)”.
W wymaganiach wykorzystano pozytywne doĞwiadczenia
z pracy TP z miernikiem TIEM-1 oraz zaáoĪenia systemu SP2000 przedstawione przez nasz zespóá. W roku 1995 powstają
pierwsze 4 egzemplarze miernika speániającego wymagania
Instytutu àącznoĞci. PoniewaĪ w tej samej obudowie
umieszczone są cztery urządzenia do kompleksowego badania
52
2
sygnaáów synchronizacji w sieciach cyfrowych sterowane za
pomocą zewnĊtrznego komputera ustalono, Īe caáoĞü zostanie
nazwana systemem pomiarowym SP-2000. System SP-2000
pracuje pod kontrolą systemu Win 3.0/3.11, a póĨniej Win 98.
W roku 1995 opracowano i uruchomiono sondy pomiarowe
E1 i E1/E1 do pozyskiwania sygnaáu badanego przez system
SP-2000 z pracujących linii telekomunikacyjnych bez
zakáócania transmisji. Sonda E1/E1 dodatkowo umoĪliwia
wyprowadzenie danych dla jednoczesnego badania jakoĞci
taktowania i stopy báĊdów. Rok 1996 to opracowanie
i uruchomienie sond E3 oraz E4 do pozyskiwania sygnaáu
taktowania bezpoĞrednio ze strumieni, odpowiednio E3 oraz
E4. Pozyskany sygnaá moĪe byü dalej badany za pomocą
systemu SP-2000.
Kolejnym skáadnikiem SWS jest opracowany w 1998 roku
na zlecenie firmy ALCATEL dystrybutor sygnaáów
taktowania.
Dystrybutor
DST-8
dostarcza
oĞmiu,
separowanych galwanicznie i synfazowych sygnaáów 2048
kHz o parametrach zgodnych z zaleceniem ITU-T G.703-13
[7]. Báąd fazy sygnaáów na wyjĞciu nie przekracza 4 ns.
W roku 1999 uruchomiono nastĊpcĊ systemu SP-2000
o zwiĊkszonej funkcjonalnoĞci i mniejszych gabarytach, tj.
system SP-3000. Zastosowanie ukáadów CPLD, wbudowanie
ukáadu automatycznej korekcji czĊstotliwoĞci wewnĊtrznego
generatora kwarcowego albo rubidowego do sygnaáu GPS lub
zewnĊtrznego sygnaáu 2048kHz, praca pod kontrolą systemu
Win 2000, a póĨniej takĪe Win XP oraz Win 7, to
podstawowe cechy nowego systemu, odróĪniające go od
poprzednika. W tym samy roku powstaje takĪe sonda STM-1
do pozyskiwania sygnaáu taktowania bezpoĞrednio ze
strumienia elektrycznego STM-1 sieci SDH. Parametry
wolnozmienne pobranego sygnaáu mogą byü dalej badane za
pomocą systemu SP-3000. W roku 2000 dystrybutor sygnaáu
taktowania i synchronizacji DST-16 jako urządzenie stacyjne
otrzymuje wymaganą wówczas homologacjĊ Instytutu
àącznoĞci i rozpoczyna siĊ proces wytwarzania dystrybutorów
na potrzeby wĊzáów sieci telekomunikacyjnej. Kolejny istotny
krok w rozwoju SWS przypada na rok 2001. W tym roku
powstaje przenoĞne Ĩródáo czĊstotliwoĞci wzorcowej Syn-Rb,
zastĊpujące w pomiarach synchronizacyjnych wielokrotnie od
niego droĪsze wzorce cezowe. ħródáo ma akumulatorowe
podtrzymanie zasilania na czas transportu i jest
synchronizowane sygnaáem 1pps z mobilnego odbiornika GPS
lub sygnaáem 2048 kHz z synchronizowanej sieci
telekomunikacyjnej. W roku 2005 dochodzi do rozszerzenia
funkcjonalnoĞci dystrybutorów DST-16 o dostarczanie
sygnaáów kodowanych HDB-3 z tzw. „pustą” ramką PCM
30/32
na
potrzeby
warstwy
dostĊpowej
sieci
telekomunikacyjnej, a w roku 2009 powstaje pierwszy
egzemplarz nastĊpcy systemu SP-3000, tj. system SP-4000.
Wszystkie urządzenia skáadowe systemu SP-4000 są
sterowanego za poĞrednictwem záącza USB 2.0. Ukáady
CPLD zostają zastąpione ukáadami FPGA, co czterokrotne
zwiĊksza rozdzielczoĞü wszystkich czterech niezaleĪnych
kanaáów pomiarowych. Nowy system pracuje pod kontrolą
Win XP/7. Szybki rozwój sieci optycznej wymusza powstanie
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
w 2011 roku sond optycznych OSTM-1 i OSTM-16 do
wydzielania taktu z pracujących áączy optycznych z sygnaáami
odpowiednio STM-1 (155,52 Mbit/s) i STM-16 (2,4 Gbit/s).
Wydzielone sygnaáy mogą byü dalej badane za pomocą
systemu pomiarowego SP-3000 lub SP-4000.
W pracach nad SWS, na róĪnych etapach jego powstawania
i rozwoju, uczestniczyli:
o Prof. dr hab. inĪ. Andrzej Dobrogowski1
o Dr hab. inĪ. Mieczysáaw Jessa1
o Dr inĪ. Henryk Batycki2
o Dr inĪ. Tomasz Bilski3
o Dr inĪ. Michal Kasznia1
o Dr inĪ. Krzysztof Lange1
o Mgr inĪ. Michaá Jaworski1
o Mgr inĪ. àukasz Matuszewski1
o InĪ. Jerzy Gumny1
o InĪ. Wáodzimierz Mankiewicz4
o InĪ. Andrzej Nowak3
1
Wydziaá Elektroniki i Telekomunikacji, Katedra Systemów
Telekomunikacyjnych i Optoelektroniki
2
Wydziaá Elektroniki i Telekomunikacji, Katedra
Telekomunikacji Multimedialnej i Mikroelektroniki
3
Wydziaá Elektryczny, Instytut Automatyki i InĪynierii
Informatycznej
4
Wydziaá Elektroniki i Telekomunikacji, Katedra
Radiokomunikacji
Urządzenia byáy montowane w dwóch firmach:
x Zakáad Techniki Mikroprocesorowej „EXE”,
Ul. AnyĪowa 15, 61-680 PoznaĔ,
x Zakáad Elektroniczny „Securus”
Ul. PopliĔskich 11, 61-573 PoznaĔ.
Podzespoáy sprowadzano z nastĊpujących krajów: Polska,
USA, Niemcy, Szwajcaria, Francja, Wielka Brytania, Austria.
Na wszystkie urządzenia zaprojektowane w zespole
zajmującym siĊ synchronizacją i firmowane logo Politechniki
PoznaĔskiej jest udzielana trzyletnia gwarancja.
sygnaáów synchronizacji. Najpierw powstaá model DST-16
o zadanej (na etapie produkcji) postaci sygnaáu wyjĞciowego
(G.703.9 albo G.703.13) (rys. 6). Od roku 2006 jest dostĊpny
dystrybutor z programowalnymi bankami wyjĞü 2x8 wyjĞü.
Dla kaĪdego banku uĪytkownik moĪe wybraü przeáącznikiem
z páyty czoáowej sygnaá G.703.9 albo G.703.13 (rys. 7).
ħródáo STFS/GPS sygnaáu czasu i czĊstotliwoĞci
synchronizowane sygnaáem systemu satelitarnego
GPS
Wybrane parametry Ĩródáa STFS/GPS
•
•
•
•
•
•
•
•
Synchronizacja do sygnaáu pochodzącego
z odbiornika GPS zintegrowanego z anteną.
OdlegáoĞü od odbiornika GPS - max 150
metrów.
Wbudowany lokalny generator kwarcowy
OCXO lub rubidowy Rb.
Sygnaáy wyjĞciowe TTL: 1Hz, 50Hz, 1kHz,
5MHz, 10MHz,
Sygnaáy wyjĞciowe sinusoidalne: 6x2048kHz.
ZgodnoĞü skali czasu generowanej lokalnie
ze skalą UTC z báĊdem nie wiĊkszym od
r50ns.
Dáugoterminowy báąd czĊstotliwoĞci nie
wiĊkszy od 8˜10-13.
MoĪliwoĞü synchronizacji czasu dowolnego
komputera poprzez záącze RS-232
Wersja z wejĞciami i wyjĞciami z tyáu obudowy
Wersja z wejĞciami i wyjĞciami z przodu obudowy
Rys 1. ħródáo STFS/GPS sygnaáu czasu i czĊstotliwoĞci synchronizowane
sygnaáem systemu satelitarnego GPS
System Pomiarowy SP-2000
Podstawowe funkcje systemu SP-2000
•
•
ವ
•
•
•
•
ವ
III. ELEMENTY SKàADOWE SYSTEMU WSPOMAGANIA
SYNCHRONIZACJI
Pierwszym
urządzeniem,
które
znalazáo
szersze
zastosowanie w sieci telekomunikacyjnej jest Ĩródáo
STFS/GPS sygnaáu czasu i czĊstotliwoĞci synchronizowane
sygnaáem systemu satelitarnego GPS (rys.1) [8]. Najbardziej
zaawansowane pod wzglĊdem sprzĊtowym i programowym są
systemy pomiarowe SP-2000 (rys. 2), SP-3000 (rys 3) i SP4000 (rys. 4) [8]. Kolejnym istotnym skáadnikiem SWS jest
przenoĞne Ĩródáo sygnaáu wzorcowego z atomowym
generatorem czĊstotliwoĞci wzorcowej – model Syn-Rb (rys.
5) [8]. ħródáo to, po uprzednim zsynchronizowaniu do
sygnaáu 1pps pochodzącego z zewnĊtrznego odbiornika GPS
lub sieci telekomunikacyjnej, moĪe przez co najmniej dwie
doby zastąpiü w pomiarach synchronizacyjnych wielokrotnie
od niego droĪsze wzorce cezowe. MoĪe byü takĪe Ĩródáem
sygnaáu synchronizacji o dáugoterminowym báĊdzie
czĊstotliwoĞci nie wiĊkszym od 10-12. Kolejnym istotnym
elementem SWS są dystrybutory sygnaáu synchronizacji DST16 [8]. Są to relatywnie proste urządzenia, bez sterowania
mikroprocesorowego, dostarczające sieci jednoczeĞnie wielu
3
Pomiar báĊdu czasu z filtracją w filtrze
dolnopasmowym o regulowanej
czĊstotliwoĞci odciĊcia od 1nHz do 10Hz.
Cztery niezaleĪne kanaáy pomiarowe.
)XQNFMDV]\ENLHMNRPSDUDFMLF]ÛVWRWOLZRĝFL
Obliczanie parametrów sygnaáów taktowania
i synchronizacji, np. MTIE, odchylenia Allana,
odchylenia czasu, niedokáadnoĞci
czĊstotliwoĞci, niestaáoĞci czĊstotliwoĞci itp.
Odtwarzanie off-line procesu pomiarowego
wyzwalanego automatycznie i trwającego
zaprogramowany czas (np. 3 doby).
ವ
Zliczanie i rejestracja tzw. poĞlizgów o
125Ps, bĊdących „zgrubnym” wyznacznikiem
jakoĞci synchronizacji.
Analiza wyników pomiarów z wykorzystaniem ವ
funkcji kursora, kompresja do jednego
ekranu itp.
3RUµZQDQLHZ\QLNµZSRPLDUµZ]QRUPDPL
ವ
PLÛG]\QDURGRZ\PL
6SRU]ÇG]DQLHVWDQGDU\]RZDQ\FK
JUDILF]Q\FKLWHNVWRZ\FKUDSRUWµZ
]SRPLDUµZ
6WHURZDQLH]DSRPRFÇ]HZQÛWU]QHJR
NRPSXWHUD3&SRSU]H]]ĄÇF]H
&HQWURQLFV.
2SURJUDPRZDQLHdla systemu
:LQdows 3.0/RUD]:LQdows Rys 2. System pomiarowy SP-2000
System Pomiarowy SP-3000
Podstawowe funkcje systemu SP-3000
ವ
ವ
ವ
ವ
ವ
ವ
ವ
3RPLDUEĄÛGXF]DVX]ILOWUDFMÇZILOWU]H
GROQRSDVPRZ\PRUHJXORZDQHMF]ÛVWRWOLZRĝFL
RGFLÛFLDRGQ+]GR+]
&]WHU\QLH]DOHľQHNDQDĄ\SRPLDURZH
.DUWDG\VWU\EXFMLNRQZHUVMLF]ÛVWRWOLZRĝFL
0+]QDF]WHU\V\JQDĄ\N+]*
:EXGRZDQ\JHQHUDWRUUXELGRZ\]IXQNFMÇ
DXWRPDW\F]QHMNRUHNFMLF]ÛVWRWOLZRĝFLGR
LPSXOVµZSSV]RGELRUQLND*36OXE
]HZQÛWU]nHJRV\JQDĄXZ]RUFRZHJRN+]
2EOLF]DQLHSDUDPHWUµZV\JQDĄµZWDNWRZDQLD
L V\QFKURQL]DFMLQS07,(RGFK\OHQLD$OODQD
RGFK\OHQLDF]DVXQLHGRNĄDGQRĝFL
F]ÛVWRWOLZRĝFLQLHVWDĄRĝFLF]ÛVWRWOLZRĝFLLWS
2GWZDU]DQLHRIIOLQH SURFHVXSRPLDURZHJR
Z\]ZDODQHJRDXWRPDW\F]QLHLWUZDMÇFHJR
]DSURJUDPRZDQ\F]DVQSGRE\
=OLF]DQLHLUHMHVWUDFMDW]ZSRĝOL]JµZRPV
EÛGÇF\FKಱ]JUXEQ\PರZ\]QDF]QLNLHPMDNRĝFL
V\QFKURQL]DFML
XVI Poznańskie Warsztaty Telekomunikacyjne
ವ
ವ
ವ
ವ
ವ
$QDOL]DZ\QLNµZSRPLDUµZ
]Z\NRU]\VWDQLHPIXQNFMLNXUVRUD
NRPSUHVMDGRMHGQHJRHNUDQXLWS
3RUµZQDQLHZ\QLNµZSRPLDUµZ
]QRUPDPLPLÛG]\QDURGRZ\PL
6SRU]ÇG]DQLHVWDQGDU\]RZDQ\FK
JUDILF]Q\FKLWHNVWRZ\FKUDSRUWµZ
]SRPLDUµZ
6WHURZDQLH]DSRPRFÇ]HZQÛWU]QHJR
NRPSXWHUD3&SRSU]H]]ĄÇF]H
&HQWURQLFV.
2SURJUDPRZDQLHGODV\VWHPX
:LQdows 2000 RUD]:LQdows XP
Rys 3. System pomiarowy SP-3000
53
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
Dystrybutor sygnaáu taktującego DST-16
z programowanymi bankami wyjĞü
System Pomiarowy SP-4000
Podstawowe funkcje systemu SP-4000
ವ
ವ
ವ
ವ
ವ
ವ
ವ
3RPLDUEĄÛGXF]DVX]ILOWUDFMÇZILOWU]H
GROQRSDVPRZ\PRUHJXORZDQHM
F]ÛVWRWOLZRĝFL RGFLÛFLDRGQ+]GR+]
&]WHU\QLH]DOHľQHNDQDĄ\SRPLDURZH
.DUWDG\VWU\EXFMLNRQZHUVMLF]ÛVWRWOLZRĝFL
0+]QDF]WHU\V\JQDĄ\N+]*
:EXGRZDQ\JHQHUDWRUUXELGRZ\]IXQNFMÇ
DXWRPDW\F]QHMNRUHNFMLF]ÛVWRWOLZRĝFLGR
ವ
LPSXOVµZSSV]RGELRUQLND*36OXE
]HZQÛWU]nHJRV\JQDĄXZ]RUFRZHJR
N+]
ವ
2EOLF]DQLHSDUDPHWUµZV\JQDĄµZWDNWRZDQLD
L V\QFKURQL]DFMLQS07,(RGFK\OHQLD$OODQD
ವ
RGFK\OHQLDF]DVXQLHGRNĄDGQRĝFL
F]ÛVWRWOLZRĝFLQLHVWDĄRĝFLF]ÛVWRWOLZRĝFLLWS
2GWZDU]DQLHRIIOLQH SURFHVXSRPLDURZHJR
ವ
Z\]ZDODQHJRDXWRPDW\F]QLHLWUZDMÇFHJR
]DSURJUDPRZDQ\F]DVQSGRE\
=OLF]DQLHLUHMHVWUDFMDW]ZSRĝOL]JµZR
PVEÛGÇF\FKಱ]JUXEQ\PರZ\]QDF]QLNLHP ವ
MDNRĝFLV\QFKURQL]DFML
Wybrane parametry dystrybutora
DST-16
•
•
$QDOL]DZ\QLNµZSRPLDUµZ
]Z\NRU]\VWDQLHPIXQNFMLNXUVRUD
NRPSUHVMDGRMHGQHJRHNUDQXLWS
3RUµZQDQLHZ\QLNµZSRPLDUµZ
]QRUPDPLPLÛG]\QDURGRZ\PL
6SRU]ÇG]DQLHVWDQGDU\]RZDQ\FK
JUDILF]Q\FKLWHNVWRZ\FKUDSRUWµZ
]SRPLDUµZ
6WHURZDQLH]DSRPRFÇ]HZQÛWU]QHJR
NRPSXWHUD3&SRSU]H]]ĄÇF]HUSB
2.0.
2SURJUDPRZDQLHGODV\VWHPX
:LQdows Vista RUD]:LQdows 7
•
•
•
Rys 4. System pomiarowy SP-4000
WejĞcie dla sygnaáu okresowego
sinusoidalnego, prostokątnego,
trójkątnego o czĊstotliwoĞci 2048kHz.
WejĞcie dla sygnaáu sinusoidalnego
2048kHz lub sygnaáu liniowego
2048kbit/s/HDB-3.
Liczba wyjĞü: 2x8 w dwóch bankach
Sygnaáy wyjĞciowe G.703.13 (sinus) lub
G.703.9 (HDB-3) wybierane niezaleĪnie
dla kaĪdego banku.
Galwaniczna separacja wejĞü od wyjĞü
oraz wyjĞü pomiĊdzy sobą.
RóĪnice faz pomiĊdzy sygnaáami
wyjĞciowymi tego samego banku
mniejsze od 2ns.
Funkcje Syn-Rb:
• ħródáo sygnaáu wzorcowego
w badaniach jakoĞci synchronizacji
sieci telekomunikacyjnej.
• ħródáo sygnaáu synchronizacji dla
fragmentów sieci telekomunikacyjnych
(tzw. przenoĞne PRS).
Rys 5. PrzenoĞne Ĩródáo sygnaáu wzorcowego z atomowym generatorem
czĊstotliwoĞci wzorcowej – model Syn-Rb
Dystrybutor sygnaáu taktującego DST-16
Wybrane parametry dystrybutora
DST-16
•
•
•
•
W pracy przestawiono genezĊ, rozwój i podstawowe
elementy skáadowe systemu wspomagania synchronizacji
opracowanego przez pracowników Politechniki PoznaĔskiej.
Poszczególne urządzenia byáy montowane w firmach
zewnĊtrznych. Na terenie Politechniki PoznaĔskiej powstaáa
koncepcja SWS i caáoĞü oprogramowania. Czáonkowie
zespoáu zajmują siĊ takĪe badaniem poprawnoĞci dziaáania
elementów SWS przed dostarczeniem do odbiorcy koĔcowego
oraz serwisowaniem po upáywie trzyletniego okresu
gwarancyjnego. Obecnie w sieciach róĪnych operatorów,
gáównie w sieci Telekomunikacji Polskiej, pracuje áącznie
ponad tysiąc elementów skáadowych SWS. W roku 2005
zespóá w skáadzie Tomasz Bilski, Andrzej Dobrogowski,
Mieczysáaw Jessa, Michaá Kasznia i Krzysztof Lange
otrzymaá nagrodĊ zespoáowa Prezesa Rady Ministrów
w kategorii „Wybitne krajowe osiągniĊcia naukowotechniczne”, za „System Wspomagania Synchronizacji w
sieciach telekomunikacyjnych”.
Funkcje DST-16
•
•
•
•
Wzmocnienie/wytáumienie sygnaáu
wejĞciowego.
Standaryzacja postaci sygnaáów
wyjĞciowych zgodnie z zaleceniami
ITU-T.
Detekcja zaników sygnaáów
wejĞciowych oraz wyjĞciowych.
Sygnalizacja alarmów zwarciowa, nie
wymagająca dla przesáania alarmu
tworzenia oddzielnego systemu
informatycznego.
Rys 6. Dystrybutor sygnaáu taktującego DST-16 o zadanej postaci sygnaáu
wyjĞciowego G.703.9 albo G.703.13
54
•
Wzmocnienie/wytáumienie sygnaáu
wejĞciowego.
Standaryzacja postaci sygnaáów
wyjĞciowych na zgodne z zaleceniami
miĊdzynarodowymi.
Detekcja zaników sygnaáów
wejĞciowych oraz wyjĞciowych.
Sygnalizacja alarmów zwarciowa, nie
wymagająca dla przesáania alarmu
tworzenia oddzielnego systemu
informatycznego.
IV. PODSUMOWANIE
Sygnaáy synchronizujące:
- impulsy 1pps pochodzące z odbiornika
satelitarnego systemu nawigacyjnego GPS
- sygnaá 2048kHz z sieci telekomunikacyjnej.
Sygnaáy wyjĞciowe:
1Hz (TTL), 1MHz (TTL), 5MHz (TTL), 10MHz
(TTL), 4x2048kHz (sinus, o poziomach
zgodnych z zaleceniami miĊdzynarodowymi)
Wbudowany atomowy generator rubidowy
z akumulatorowym podtrzymaniem czĊstotliwoĞci
pracy na czas transportu.
•
•
•
System wspomagania synchronizacji uzupeánia zestaw sond
pomiarowych opisany w innym artykule.
Wybrane parametry dystrybutora
Ĩródáa Syn-Rb
WejĞcie dla sygnaáu okresowego
sinusoidalnego, prostokątnego,
trójkątnego o czĊstotliwoĞci 2048kHz.
WejĞcie dla sygnaáu sinusoidalnego
2048kHz lub sygnaáu liniowego
2048kbit/s/HDB-3.
Liczba wyjĞü: 2x8 w dwóch bankach
(postaü sygnaáu wyjĞciowego w kaĪdym
z banków jest wybierana na etapie
zamawiania dystrybutora).
Galwaniczna separacja wejĞü od wyjĞü
oraz wyjĞü pomiĊdzy sobą.
RóĪnice faz pomiĊdzy sygnaáami
wyjĞciowymi tego samego banku
mniejsze od 2ns.
Funkcje DST-16
Rys 7. Dystrybutor sygnaáu taktującego DST-16 z programowalnymi
bankami wyjĞü
PrzenoĞne Ĩródáo Syn-Rb sygnaáu wzorcowego
z atomowym generatorem czĊstotliwoĞci
wzorcowej
•
4
LITERATURA
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Zalecenie ITU-T G.803.
Zalecenie ITU-T G.811.
Zalecenie ITU-T G.812.
Zalecenie ITU-T G.815.
Zalecenie ITU-T G.823.
Zalecenie ITU-T G.825.
Zalecenie ITU-T G.703.
Instrukcje eksploatacyjne elementów skáadowych systemu wspomagania
synchronizacji SWS.
PWT 2012 - Poznań, 14 grudnia 2012
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
1
Przegląd metod wyznaczania maksymalnego
báĊdu przedziaáu czasu
Andrzej Dobrogowski, Michaá Kasznia
Streszczenie—W pracy dokonano przeglądu metod
wyznaczania jednego z parametrów sygnaáów synchronizacji –
maksymalnego báĊdu przedziaáu czasu MTIE. Przedstawiono
metody umoĪliwiające analizĊ po wykonaniu pomiaru báĊdu
czasu (off-line) oraz metody wyznaczania parametru w czasie
rzeczywistym (real-time).
Sáowa kluczowe—sygnaá synchronizacji,
maksymalny báąd przedziaáu czasu
M
báąd
czasu
I. WPROWADZENIE
AKSYMALNY báad przedziaáu czasu (ang.: Maximum
Time Interval Error, MTIE) jest jednym z istotnych
parametrów opisujących jakoĞü sygnaáów synchronizacji
sieci telekomunikacyjnej. Maksymalny báąd przedziaáu
czasu jest parametrem o charakterze granicznym,
opisującym maksymalne wartoĞci wahaĔ fazy zegara w
okreĞlonym przedziale czasu (przedziale obserwacji).
Parametr ten wykorzystywany jest do wymiarowania
pojemnoĞci buforów na granicy skal czasu. EstymatĊ
maksymalnego báĊdu przedziaáu czasu wyznacza siĊ na
podstawie ciągu próbek báĊdu czasu zmierzonych w
interfejsie sieci telekomunikacyjnej [1].
Wyznaczanie maksymalnego báĊdu przedziaáu czasu jest
jedną z najbardziej czasocháonnych (obok pomiaru wartoĞci
báĊdu czasu) procedur w procesie analizy jakoĞci sygnaáu
taktowania. W szczególnych przypadkach czas trwania
obliczeĔ tego parametru moĪe przekroczyü czas
pozyskiwania danych (pomiaru próbek báĊdu czasu)
wykorzystywanych do obliczeĔ, co stanowiáo istotną
niedogodnoĞü w przypadku starszej generacji komputerów
uĪywanych do obliczeĔ. W celu uáatwienia analizy
konieczne staáo siĊ znalezienie metod umoĪliwiających
uzyskanie wartoĞci MTIE w czasie znacznie krótszym od
czasu pozyskiwania danych. Dlatego teĪ w literaturze
moĪna znaleĨü szereg efektywnych czasowo metod
wyznaczania parametru [2, 3, 4], obok których obecne są
takĪe oryginalne metody opracowane przez autorów
niniejszej pracy [5–13]. Przyczyną zainteresowania
problematyką wyznaczania MTIE byáo opracowanie i
wdraĪanie do eksploatacji systemów pomiarowych
sáuĪących do analizy jakoĞci sygnaáów synchronizacji sieci
telekomunikacyjnej od poáowy lat dziewiĊüdziesiątych XX
Autorzy są pracownikami Katedry Systemów Telekomunikacyjnych
i Optoelektroniki Politechniki PoznaĔskiej, ul. Polanka 3, 60-965 PoznaĔ
([email protected], [email protected]).
PracĊ wykonano w ramach projektu nr N N517 470540 finansowanego
przez Narodowe Centrum Nauki w latach 2011-2013.
w. Efektem prac byáo opracowanie szeregu efektywnych
czasowo metod wykorzystujących losowy charakter ciągu
danych bądĨ teĪ jego sekwencyjną redukcjĊ [5–8], co
pozwoliáo uzyskaü czas obliczeĔ zdecydowanie krótszy od
czasu pomiaru. Z tego wzglĊdu moĪliwe staáo siĊ
zaproponowanie metod pozwalających na wykonywanie
obliczeĔ parametru w trakcie pomiaru báĊdu czasu [9–12].
Opracowane metody moĪna podzieliü na dokáadne i
przybliĪone, zaleĪne i niezaleĪne od danych, metody z
redukcją i bez redukcji danych oraz metody analizy off-line i
analizy w czasie rzeczywistym. W pracy omówiono cechy
charakterystyczne i zastosowanie poszczególnych metod.
Wskazane zostaáy kierunki dalszych prac nad metodami
wyznaczania MTIE.
II. ESTYMATOR MTIE
Maksymalny báąd przedziaáu
estymowany wedáug wzoru:
czasu
MTIE(W)
jest
max §¨ max xi min xi ·¸
(1)
k di d k n
¹
gdzie W=nW0 jest przedziaáem obserwacji, ciąg {xi} jest
ciągiem N próbek báĊdu czasu miĊdzy sygnaáem badanym a
sygnaáem odniesienia wziĊtych z odstĊpem W0, natomiast
wartoĞü n moĪe zmieniaü siĊ od 1 do Ní1.
W celu znalezienia wartoĞci MTIE w przedziale
obserwacji W zgodnie ze wzorem (1), naleĪy dokonaü
przejrzenia wszystkich przedziaáów (okien) o szerokoĞci
W=nW0 wystĊpujących w ciągu N próbek báĊdu czasu. W tym
celu „okno” obejmujące n+1 kolejnych próbek przesuwane
jest o odstĊp W0, a wiĊc o jedną próbkĊ od początku do
koĔca ciągu {xi}. Dla kaĪdego usytuowania okna
znajdowana jest wartoĞü miĊdzyszczytowa báĊdu czasu xpp.
WartoĞü
MTIE(W)
jest
maksymalną
wartoĞcią
miĊdzyszczytową báĊdu czasu xpp znalezioną dla wszystkich
moĪliwych usytuowaĔ okna o szerokoĞci W. ZasadĊ
wyznaczania MTIE ilustruje Rys. 1.
W celu oceny jakoĞci badanego sygnaáu ciąg wartoĞci
MTIE wyznaczonych dla przedziaáów obserwacji W z
okreĞlonego zakresu (np. od 0.1 s do 1000 s) przedstawia
siĊ w postaci wykresu o logarytmicznej skali wartoĞci i
logarytmicznej skali czasu, a nastĊpnie porównuje siĊ z
naáoĪonymi wartoĞciami granicznymi (normami) MTIE
zdefiniowanymi dla okreĞlonych punktów sieci. NiezbĊdne
jest wiĊc wyznaczenie ciągu wartoĞci MTIE dla przedziaáów
obserwacji z zakresu od Wmin do Wmax. Zastosowanie do
obliczeĔ opisanej wyĪej metody opartej bezpoĞrednio na
definicji estymatora parametru, polegającej na przesuwaniu
i przeglądaniu okien moĪe byü czasocháonne [4, 5].
MTˆIE nW0 1d k d N n© k d i d k n
XVI Poznańskie Warsztaty Telekomunikacyjne
55
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
przesuniĊcie okna
xi
nastĊpne poáoĪenie okna
T=(N-1)W0
W0
0 1
xpp
. . .
k k+1
k+n k+n+1 N i
W=nW0
okno
Rys. 1. Zasada wyznaczania maksymalnego báĊdu przedziaáu czasu
III. METODY WYZNACZANIA MTIE
A. Metody wyznaczania w trybie „off-line”
Wyznaczanie wartoĞci MTIE w tak zwanym trybie
off-line jest procesem dwuetapowym. Pierwszym etapem
jest pomiar báĊdu czasu, a obliczenia wartoĞci parametru
(drugi etap) wykonywane są na podstawie zmierzonych
wczeĞniej próbek báĊdu czasu. Efektywne czasowo metody
wyznaczania MTIE opierają mechanizm swojego dziaáania
na redukcji liczby usytuowaĔ „okna”, w których
wykonywany jest przegląd caáej zawartoĞci okna lub na
sekwencyjnej redukcji danych wraz ze wzrostem szerokoĞci
analizowanego przedziaáu obserwacji.
Metodami opierającymi zasadĊ swojego dziaáania na
pierwszym mechanizmie są: metoda decyzji brzegowych,
metoda skoku do ekstremum [5] i metoda przedziaáów
rozáącznych [2], przy czym ta ostatnia metoda jest metodą
przybliĪoną, dającą bardzo zgrubną estymatĊ wartoĞci
parametru, natomiast dwie pierwsze metody są metodami
dokáadnymi. W przypadku metody decyzji brzegowych
wykorzystano fakt, Īe w procesie przesuwania okna o jedną
pozycjĊ w ciągu próbek, zmiana zawartoĞci okna dotyczy
tylko dwóch próbek: pierwsza wartoĞü opuszcza okno, a
pojawia siĊ nowa wartoĞü na koĔcu okna. O koniecznoĞci
przeszukania okna w nowym poáoĪeniu decyduje rezultat
porównania wartoĞci brzegowych (nowa i stara wartoĞü) z
wartoĞcią maksymalną i minimalną dla poprzedniego
poáoĪenia okna. Dziaáanie metody skoku do ekstremum,
zwanej teĪ metodą poáoĪenia wartoĞci ekstremalnych (ang.:
extreme fix method, EF) opiera siĊ na ustaleniu poáoĪenia
próbek o wartoĞciach ekstremalnych (maksimum i
minimum), tworzących wartoĞü miĊdzyszczytową dla
danego usytuowania okna. Na odcinku miĊdzy
najwczeĞniejszą (pierwszą) próbką w oknie, a najbliĪszą jej
próbką o wartoĞci ekstremalnej, nie wystąpi juĪ tego typu
wartoĞü dla kaĪdego z nowych usytuowaĔ okna
zaczynających siĊ na tym odcinku. Nie jest zatem konieczne
przeglądanie tego odcinka i moĪna wykonaü przesuniĊcie
okna w ten sposób, Īe pierwszą próbką w nowym poáoĪeniu
bĊdzie wáaĞnie najwczeĞniejsza próbka o wartoĞci
ekstremalnej z poprzedniego poáoĪenia okna. DziĊki temu
pomija siĊ w analizie usytuowania okna wyznaczone
opuszczonymi próbkami. Po wykonaniu przesuniĊcia okna
do najbliĪszej próbki ekstremalnej przeszukiwana jest
jedynie czĊĞü usytuowania okna zawierająca nowe wartoĞci,
które pojawiáy siĊ w oknie. W przypadku, gdy
56
2
najwczeĞniejszą próbka ekstremalną jest pierwsza próbka
dla danego usytuowania okna, przesuniĊcie wykonywane
jest o jedną próbkĊ. W tej sytuacji zachodzi koniecznoĞü
znalezienia takiego ekstremum, jakie wáaĞnie opuĞciáo okno.
O przeglądaniu zawartoĞci nowego usytuowania okna
decyduje analiza sytuacji na kraĔcach okna, jak w
przypadku metody decyzji brzegowych. PrzesuniĊcia okna
w procesie wyznaczania wartoĞci MTIE metodą EF
przedstawione zostaáy na Rys. 2 [5].
Metodami opierającymi zasadĊ swojego dziaáania na
drugim mechanizmie są: metoda z zastosowaniem
dekompozycji binarnej (ang.: binary decomposition) [4],
metoda skoku do ekstremum z sekwencyjną redukcją
danych (ang.: extreme fix with sequential data reduction,
EFSDR) [7] oraz metoda bezpoĞrednia z sekwencyjną
redukcją danych (ang.: direct serach with sequential data
reduction, DSDR) [8]. Metody te przeznaczone są do
wyznaczania wartoĞci parametru dla przedziaáów obserwacji
uszeregowanych od Wmin do Wmax. Efektem poszukiwania
wartoĞci MTIE dla najmniejszego przedziaáu obserwacji,
oprócz znalezionej wartoĞci parametru, są zredukowane
ciągi
danych
(ciągi
wartoĞci
ekstremalnych),
wykorzystywane w kolejnym kroku obliczeĔ (dla kolejnego
przedziaáu obserwacji). W efekcie, dla kaĪdego kolejnego
przedziaáu obserwacji dziaáania wykonuje siĊ na krótszych
ciągach danych, co w znaczny sposób przyspiesza
wykonywanie obliczeĔ. W przypadku metody z
dekompozycją binarną, rozmiary przedziaáów obserwacji
muszą byü potĊgą liczby dwa – porównywane są pary
próbek, a proces redukcji danych jest regularny. W
przypadku metod DSDR i EFSDR, wielkoĞci przedziaáów
obserwacji są dowolne, a na proces redukcji danych wpáyw
ma ich losowy charakter. Poáączenie mechanizmów skoku
do ekstremum i sekwencyjnej redukcji danych w metodzie
EFSDR przyniosáo dobry efekt z punktu widzenia czasu
wykonywania obliczeĔ [7]. Niestety, takie poáączenie
spowodowaáo teĪ, Īe dla specyficznego uáoĪenia próbek
moĪe dojĞü do pominiĊcia usytuowaĔ okna mających wpáyw
na ostateczny wynik obliczeĔ. W rezultacie uzyskany wynik
obliczeĔ moĪe byü mniejszy od rzeczywistej wartoĞci MTIE
charakteryzującej badany ciąg próbek [8]. Dziaáanie metody
z dekompozycją binarną przedstawione jest na Rys. 3,
natomiast metody DSDR na Rys. 4.
xi
usytuowanie okna k
0 1 ... k
p1
p2
k+n
p1+n
przesuniĊcie okna
usytuowanie okna p1
przesuniĊcie okna o 1 próbkĊ
usytuowanie okna p1+1
Rys.2. PrzesuniĊcia okna w metodzie skoku do ekstremum EF
PWT 2012 - Poznań, 14 grudnia 2012
i
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
ciąg próbek báĊdu czasu
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 pozycja
4 2 5 8 6 9 3 7 5 9 7 10 8 9 6 7 wartoĞü
k=1
W=W0
max 4 5 8 8 9 9 7 7 9 9 10 10 9 9 7MTIE(W )=6
0
min 2 2 5 6 6 3 3 5 5 7 7 8 8 6 6
k=2
W=3W0
max 8 8 9 9 9 8 9 9 10 10 10 10 9
min 2 2 5 3 3 3 3 5 5 7 7 6 8
k=3
W=7W0
max 9 8 9 9 10 10 10 10 10
min 2 2 3 3 3 3 3 5 5
MTIE(3W0)=6
MTIE(7W0)=7
k=4
max 10
W=15W0 min 2
MTIE(15W0)=8
Rys. 3. Wyznaczanie MTIE z zastosowaniem dekompozycji binarnej
ciąg próbek báĊdu czasu
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 pozycja
4 7 6 2 3 11 4 7 5 8 4 1 5 0 2 5 wartoĞü
(max=7, min=2)
(11, 2)
(11, 2)
(11, 2)
(11, 3)
(11, 4)
(8, 4)
...
7
11
2 3
8
4
(5, 0)
okno 5-próbkowe
nowe – zredukowane – ciągi danych
max
min
MTIE=9
5
4 1
5
0
(11, 2)
(11, 2)
(11, 2)
(11, 2)
(11, 3)
MTIE=10
...
(8, 0)
okno 7-próbkowe
Rys. 4. Wyznaczanie MTIE metodą bezpoĞrednią z sekwencyjną redukcją
danych
B. Metody wyznaczania w czasie rzeczywistym
Wyznaczając wartoĞü MTIE wyznacza siĊ wartoĞci
ekstremalne dla okreĞlonego zbioru wartoĞci (usytuowania
okna), czyli dokonuje porównania wartoĞci z wartoĞcią
maksymalną i minimalną. Porównywanie to moĪe byü
wykonywane na bieĪąco w trakcie pomiaru próbek báĊdu
czasu. Nowo pomierzona próbka porównywana jest z
bieĪącymi wartoĞciami ekstremalnymi dla danego
usytuowania okna o zadanym rozmiarze (szerokoĞci
przedziaáu obserwacji), nastĊpnie ustalana jest wartoĞü
miĊdzyszczytowa xpp báĊdu czasu dla tego usytuowania, a
uzyskana wartoĞü xpp porównywana jest z maksymalną
wartoĞcią miĊdzyszczytową znalezioną do tej chwili.
Procedura
ta
przypomina
w
swej
zasadzie
funkcjonowanie metody decyzji brzegowych. Do opisanych
wyĪej operacji naleĪy wiĊc dodaü takĪe ewentualne
przeglądanie caáej zawartoĞci usytuowania okna w
przypadku utraty wiedzy o aktualnym ekstremum.
3
CzynnoĞci te muszą zostaü wykonane (w przypadku
zaáoĪenia jednowątkowoĞci procesu) w odstĊpie czasu
pomiĊdzy chwilami pomiaru kolejnych próbek báĊdu czasu.
W zaleĪnoĞci od odstĊpu próbkowania i moĪliwoĞci
obliczeniowych systemu pomiarowego w odstĊpie tym
moĪna wykonaü obsáugĊ kilku okien (przedziaáów
obserwacji) o wiĊkszych rozmiarach. Zastosowanie takiej
procedury pozwoli uzyskaü informacje o wartoĞciach MTIE
dla ustalonych przedziaáów obserwacji znalezionych do
bieĪącej chwili pomiaru, a wiĊc moĪliwa bĊdzie obserwacja
bieĪących wartoĞci MTIE podczas pomiaru báĊdu czasu.
Krytycznym problemem w przypadku realizacji
wyznaczania MTIE w czasie rzeczywistym jednoczeĞnie dla
szeregu wartoĞci przedziaáu obserwacji jest kumulacja
operacji porównywania wykonywanych w trakcie jednego
odstĊpu próbkowania. Liczba tych operacji roĞnie w
przypadku koniecznoĞci przeglądania caáego usytuowania
okna. Zakáadając, Īe przeglądanie caáego usytuowania okna
odpowiadającego
najmniejszemu
rozpatrywanemu
przedziaáowi obserwacji nie powinno sprawiaü problemu
(zwykle Wmin=3W0, a wiĊc przedziaá ten obejmuje 4 próbki),
w przypadku wiĊkszych przedziaáów obserwacji korzystne
jest zastosowanie metod efektywnych czasowo.
Autorzy pracy zaadaptowali szereg efektywnych czasowo
metod wyznaczania MTIE do obliczeĔ prowadzonych w
czasie rzeczywistym [9–12]. Obliczanie MTIE w czasie
rzeczywistym z wykorzystaniem metody bezpoĞredniej z
sekwencyjną redukcją danych DSDR rozpoczyna siĊ wraz z
pierwszą zmierzoną próbką báĊdu czasu. KaĪda kolejna
próbka jest porównywana z aktualnymi wartoĞciami
ekstremalnymi dopóki nie zostanie wypeánione próbkami
pierwsze
usytuowanie
okna
odpowiadającego
najmniejszemu przedziaáowi obserwacji. Po wypeánieniu
znane są wartoĞci ekstremalne dla tego usytuowania. KaĪda
kolejna zmierzona próbka tworzy nowe usytuowanie okna.
WartoĞci ekstremalne znalezione dla kaĪdego usytuowania
okna tworzą nowe ciągi danych o zredukowanej liczbie
elementów, które są uĪywane do wyznaczenia wartoĞci
ekstremalnych dla kolejnego, wiĊkszego przedziaáu
obserwacji. Usytuowanie wiĊkszego okna nie jest
analizowane dopóki wszystkie próbki znajdujące siĊ w tym
usytuowaniu nie zostaną przejrzane przez poprzednie,
mniejsze okno. Ewentualne przeglądanie caáej zawartoĞci
usytuowania okna angaĪuje mniejszą liczbĊ próbek niĪ
wynikająca z szerokoĞci okna. Przykáad wyznaczania MTIE
w czasie rzeczywistym przedstawiony jest na Rys. 5 [9]. Po
zakoĔczeniu pomiaru znana jest wartoĞü MTIE dla
najmniejszego przedziaáu obserwacji. WartoĞci parametru
dla wiĊkszych przedziaáów bĊdą znane po „dosuniĊciu”
odpowiadających im okien do koĔca ciągu danych.
Podobnie przebiegają obliczenia w czasie rzeczywistym z
zastosowaniem metody z dekompozycją binarną. KaĪda
kolejna mierzona próbka porównywana jest z próbką
zmierzoną w chwili poprzedniej. Efekt porównania – w
postaci pary ekstremów (maksimum i minimum) – jest
zapisywany do dalszych obliczeĔ. RóĪnica uzyskanych
wartoĞci (wartoĞü miĊdzyszczytowa pary próbek) jest
porównywana
z
bieĪącą
maksymalną
wartoĞcią
miĊdzyszczytową. NastĊpnie, jeĞli wystarczająca liczba
XVI Poznańskie Warsztaty Telekomunikacyjne
57
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
próbek zostanie zmierzona, dokonywana jest analiza dla
okna 4-próbkowego – bieĪący efekt analizy okna
2-próbkowego porównywany jest z rezultatem uzyskanym 2
odstĊpy próbkowania wczeĞniej. BieĪący rezultat analizy
okna 4-próbkowego jest zapisywany w celu wykorzystania
do analizy okna 8-próbkowego. Kolejne okna zawierające
liczbĊ próbek bĊdącą potĊgą liczby 2 są analizowane (po
zmierzeniu wystarczającej liczby próbek) z wykorzystaniem
zapisanych wczeĞniej rezultatów analizy odpowiednio
mniejszych okien. Przykáad wyznaczania MTIE w czasie
rzeczywistym z zastosowaniem dekompozycji binarnej
przedstawiony jest na Rys. 6 [10].
W przypadku realizacji wyznaczania MTIE w czasie
rzeczywistym z zastosowaniem metody skoku do
ekstremum EF, obliczenia dla kaĪdego okna (przedziaáu
obserwacji) są realizowane niezaleĪnie. Usytuowania okien
odpowiadających przedziaáom obserwacji dáuĪszym niĪ Wmin
są analizowane po wypeánieniu próbkami bez czekania na
analizĊ tych pozycji dla odpowiednio mniejszych okien.
WartoĞci ekstremalne znalezione dla danego przedziaáu
obserwacji nie wpáywają na proces obliczeĔ realizowany dla
innych przedziaáów. Przykáad wyznaczania MTIE w czasie
rzeczywistym z zastosowaniem metody EF przedstawiony
jest na Rys. 7 [9]. WartoĞci ekstremalne (biaáe i czarne
gwiazdki) znalezione dla aktualnie analizowanego
usytuowania okien pozwalają wyznaczyü nastĊpne
usytuowania okien (oznaczone liniami przerywanymi). Po
zakoĔczeniu pomiaru báĊdu czasu, wartoĞci parametru dla
wszystkich przedziaáów obserwacji są znane natychmiast,
bez opóĨnienia spowodowanego „dosuwaniem” okien.
próbki báĊdu czasu
xpp maxpp
4
5
5
ciągi zredukowane
max
min
7
6
2
8
4
5
2
7 8 9
6
1
4
1
7
2
1
LITERATURA
[1]
[2]
[3]
[4]
7
[7]
1
Rys. 5. Przykáad wyznaczania MTIE w czasie rzeczywistym metodą DSDR
[8]
próbki báĊdu czasu
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 …
4 7 6 2 4 7 4 7 5 8 2 5 5 1 4 3 …
okno
2-próbkowe
max 7 7 6 4 7 7 7 7 8 8 5 5 5 4 4
min 4 6 2 2 4 4 4 5 5 2 2 5 1 1 3
okno
max 7
4-próbkowe min 2
[9]
xpp maxpp
1
6
4
6
[10]
[11]
7 7 7 7 7 8 8 8 8 5 5 5
2 2 2 4 4 4 2 2 2 1 1 1
[12]
okno
8-próbkowe
max 7 7 8 8 8 8 8 8 8
min 2 2 2 2 2 2 1 1 1
7
7
[13]
Rys. 6. Przykáad wyznaczania MTIE w czasie rzeczywistym metodą z
dekompozycją binarną
58
(8, 2)
Zaproponowane przez autorów metody wyznaczania
MTIE znalazáy zastosowanie w opracowywanych systemach
pomiarowych. Kolejnym krokiem w rozwoju metod
wyznaczania MTIE moĪe byü zastosowanie obliczeĔ
wielowątkowych (np. dla procesorów wielordzeniowych)
oraz realizacja obliczeĔ w czasie rzeczywistym dla
wielokanaáowych pomiarów báĊdu czasu [13].
11
6
8
6
IV. PODSUMOWANIE
[6]
2
(7, 2)
Autorzy zaimplementowali takĪe dla metod DSDR i EF
skuteczny mechanizm kontrolujący proces obliczeĔ i
zabezpieczający przed przekroczeniem przez obliczenia
odstĊpu próbkowania [11, 12].
ciągi zredukowane
max
min
(8, 4)
Rys. 7. Przykáad wyznaczania MTIE w czasie rzeczywistym metodą skoku
do ekstremum
[5]
7
1 2 3 4 5 6 7 8 9 10 11 12 13 14 … pozycja
4 7 6 2 4 7 4 7 5 8
… wartoĞü
5
próbki báĊdu czasu
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
4 7 6 2 4 7 4 7 5 8 2 5 5 1 4 3 1 6 4 7 8 9 7 11 9
4
Zalecenia ETSI EN 300 462, ITU-T Rec. G.810, ANSI T1.101-1999.
S. Bregni, “Measurement of Maximum Time Interval Error for
telecommunications clock stability characterization,” IEEE Trans.
Instrum. Meas., vol. 45, No. 5, pp. 900-906, 1996.
S. Bregni, P. Tavella, “Estimation of the percentile Maximum Time
Interval Error of gaussian white phase noise,” Proc. of IEEE ICC’97,
vol. 3, pp. 1597-1601, Montreal, Canada, June 1997.
S. Bregni, S. Maccabruni, “Fast computation of Maximum Time
Interval Error by binary decomposition,” IEEE Trans. Instrum. Meas.,
vol. 49, No. 6, pp. 1240-1244, Dec. 2000.
A. Dobrogowski, M. Kasznia, “Time effective methods of calculation
of Maximum Time Interval Error,” IEEE Trans. Instrum. Meas., vol.
50, No. 3, pp. 732-741, June 2001.
A. Dobrogowski, M. Kasznia, “Quasi-parallel computation of
Maximum Time Interval Error estimates,” Proc. 2002 IEEE Freq.
Contr. Symp., pp. 733-738, New Orleans, 29-31 May 2002.
A. Dobrogowski, M. Kasznia, “Maximum Time Interval Error
assessment based on the sequential reducing data volume,” IEEE
Trans. Ultras. Ferr. Freq. Contr., vol. 49, no. 7, pp. 987-994, July
2002.
A. Dobrogowski, M. Kasznia, “Some problems and their solutions in
quasi-parallel MTIE assessment,” Proc. 2004 IEEE Int. Ultras., Ferr.,
Freq. Contr. Joint 50th Anniv. Conf., pp. 494-499, Montreal, 23-27
August 2004.
A. Dobrogowski, M. Kasznia, “Testing of the methods of real-time
MTIE calculation,” Proc. 2005 Joint IEEE Freq. Contr. Symp. and
PTTI Syst. Appl. Meeting, pp. 397-403, Vancouver, 29-31 August
2005.
A. Dobrogowski, M. Kasznia, “On-line computation of MTIE using
binary decomposition and direct search with sequential data
reducing,” Proc. 2007 IEEE Int. Freq. Contr. Symp. and 21st EFTF,
pp. 877-882, Geneva, 29 May – 01 June 2007.
A. Dobrogowski, M. Kasznia, “Real-time MTIE assessment with
flexible control of computation process,” Proc. EFTF’09 IEEEFCS’09 Joint Conf., pp. 1102-1107, Besancon, France, 20-24 April
2009.
A. Dobrogowski, M. Kasznia, “Implementation of real-time MTIE
assessment method”, Proc. 2011 Joint Conf. IEEE FCS and EFTF,
pp. 304-309, San Francisco, California, USA, May 2-5, 2011.
A. Dobrogowski, M. Kasznia, “Some Concepts of the Real-Time
MTIE Assessment for Multi-Channel Time Error Measurement,”
Proc. 2012 IEEE Freq. Contr. Symp., pp. 493-498, Baltimore, USA,
May 22-24, 2012.
PWT 2012 - Poznań, 14 grudnia 2012
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Synchronizacja sieci Telekomunikacji Polskiej
Marek Dynowski, Tadeusz Pawszak
Streszczenie—W pracy przedstawiono ogólne zasady
synchronizacji sieci Telekomunikacji Polskiej. Przedstawiono
strukturę sieci synchronizacyjnej oraz urządzenia używane w tej
sieci.
Słowa Kluczowe — Synchronizacja, sieć telekomunikacyjna,
urządzenia synchronizacji sieci.
W
S
I. POTRZEBA SYNCHRONIZACJI
TELEKOMUNIKACYJNYCH SIECIACH CYFROWYCH
ZYBKI rozwój telekomunikacji powoduje, że sieci
telekomunikacyjne ewoluują od sieci TDM (ang. Time
Division Multiplexing) wykorzystujących zwielokrotnienie
z podziałem czasu do tzw. Sieci Nowej Generacji (ang. Next
Generatation Networks) bazowanych na przełączaniu pakietów
PSN (ang. Packet Switched Networks) – Ethernet. Kołem
napędowym tej ewolucji jest znaczna redukcja kosztów.
Technicznym celem jest transport wszystkich usług
telekomunikacyjnych
przez
zunifikowaną
platformę
przełączania pakietów.
Ethernet został użyty w stacjach bazowych sieci
komórkowych, w sieciach przełączeniowych, w sieciach
metropolitalnych i w różnych technologiach sieci
dostępowych.
Aby
zapewnić
odpowiednią
jakość
świadczonych usług telekomunikacyjnych, bazujących na sieci
TDM i sieciach pakietowych, konieczne jest zapewnienie
właściwej synchronizacji tych sieci.
Należy podkreślić, że zarówno dla sieci TDM jak i sieci
pakietowych dystrybucja sygnałów synchronizacyjnych jest
realizowana na bazie sieci transportowych SDH (ang.
Synchronous Digital Hierarchy) z wykorzystaniem dotychczas
stosowanych urządzeń w strukturze sieci synchronizacyjnej.
Na przykład dla transportu synchronizacji w sieci Ethernet dla
zapewnienia odpowiedniej jakości różnego rodzaju usług
konieczne było wprowadzenie Synchronicznego Ethernetu
(SyncE). W sieci SyncE przyjęto metodę (master – slave),
analogiczną jak dla sieci SDH. Elementami sieci SyncE są
zegary EEC (ang. Ethernet Equipment Clock) analogiczne jak
zegary SEC (ang. Synchronous Equipment Clock) w sieci
SDH. Dystrybucja sygnałów synchronizacyjnych od zegara
PRC (ang. Primary Referency Clock) może być realizowana
w strukturze mieszanej zawierającej zarówno elementy sieci
SDH (zegary SEC) i elementy sieci SyncE (zegary EEC).
Sieć telekomunikacyjna składa się z kilku płaszczyzn. Jedną
Autorzy są pracownikami Telekomunikacji Polskiej w Warszawie.
(e-mail: [email protected], [email protected]).
z nich jest płaszczyzna synchronizacyjna, w której dokonuje
się zsynchronizowanie rozdzielonych geograficznie zegarów
do taktu pochodzącego od najlepszego źródła. Taki proces
nazywa się stanem synchronizmu i oznacza taką zgodność
taktów zegarów, przy której utrata informacji spowodowana
przesunięciami czasowymi taktów nie przekracza wcześniej
przyjętych granic.
Im bardziej złożone są to usługi oraz im większe szybkości
transmisji są stosowane do ich obsługi, tym ważniejszym
zagadnieniem jest zapewnienie prawidłowej synchronizacji
sieci. Brak właściwej synchronizacji powoduje obniżenie
jakości lub degradację oferowanych usług. Na przykład:
transmisja telefoniczna objawia się trzaskami w słuchawce,
transmisja telefaksowa powoduje zamazanie do 8 linii,
transmisja danych (przekazywanie obrazu) zatrzymanie
obrazu na pewien czas,
pakietowa transmisja danych powoduje utratę przypadkowej
liczby bloków
II. OGÓLNE ZASADY SYNCHRONIZACJI SIECI
Ogólne zasady synchronizacji sieci są następujące:
1. Dla synchronizacji sieci cyfrowej wykorzystuje się
wyspecjalizowane źródła częstotliwości, wytwarzające
bardzo stabilny sygnał taktowania.
2. Podstawowym źródłem sygnału synchronizacyjnego jest
pierwotny zegar odniesienia PRC, który dostarcza sygnał
wzorcowy o długoterminowej niedokładności częstotliwości
o wartości nie większej niż 1x10-11.
3. W ważniejszych węzłach sieci są stosowane wtórne źródła
sygnałów synchronizacyjnych SSU (ang. Synchronization
Supply Unit) służące do regeneracji sygnałów
synchronizacyjnych
oraz
dystrybucji
sygnałów
synchronizacyjnych na potrzeby różnych elementów
sieciowych w węźle.
4. W węzłach mniejszych sieci, gdzie nie są stosowane
urządzenia SSU, a ilość wyjść synchronizacyjnych
z zegarów SEC krotnic SDH jest niewystarczająca,
stosujemy
dystrybutory
sygnałów
taktowania
i synchronizacji SDU (ang. Synchronization Distribution
Unit).
5. Sieć synchronizacyjna powinna tworzyć hierarchiczną
strukturę przekazywania sygnałów taktowania pomiędzy
poziomami sieci telekomunikacyjnej zgodnie z zasadą
master-slave.
6. Przesyłanie sygnałów taktowania pomiędzy węzłami: każdy
węzeł sieci synchronizacyjnej powinien być zasilany
z wyższego
lub
tego
samego
poziomu
sieci
XVI Poznańskie Warsztaty Telekomunikacyjne
59
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
synchronizacyjnej, co najmniej dwiema fizycznie
rozdzielnymi drogami, tzn. do każdego węzła dostarczony
jest sygnał drogą główną i rezerwową.
7. Dystrybucja sygnałów synchronizacyjnych odbywała się
początkowo za pomocą teletransmisyjnych systemów PDH
(ang. Plesiochronous Digital Hierarchy) przez sygnał
użytkowy o przepływności 2048 kbit/s, a obecnie za pomocą
systemów SDH (ang. Synchronous Equipment Digital
Hierarchy) lub ASON (ang. Automatically Switched Optical
Network) z wykorzystaniem płaszczyzny sterowania
GMPLS (ang. Generalized Multiprotocol Label Switching) nowa generacja SDH przez sygnał zbiorczy STM-N (ang.
Synchronous Transport Module Level N) reprezentowanych
w strukturze łańcucha przez zegary SEC.
8. Unikanie pętli dla sygnałów taktowania. Należy unikać dróg
zamkniętych tzw. pętli dla sygnałów taktowania, łącznie ze
stanami awaryjnego przełączania na rezerwę w przypadku
uszkodzenia jakiegokolwiek elementu sieci transportowej.
Długookresowe odstrojenie częstotliwości, powodowane
przez pętle czasowe, może być większe od 10 -7 i prowadzi
do degradacji jakości usług oferowanych w danej sieci.
9. Sieć synchronizacyjna powinna umożliwiać bezpieczne,
automatyczne przełączanie na odpowiednie struktury
rezerwowe w przypadku awarii łączy lub zegarów.
10. W celu automatycznego przełączenia się krotnicy SDH na
rezerwowe sygnały synchronizacyjne, w nagłówku SOH
(ang. Section Overhead) ramki sygnału zbiorczego STM-N
istnieje bajt S1 zawierający cztery bity przenoszące
informację SSM (ang. Synchronization Status Message),
w którym są zapisane informacje o jakości sygnału
taktowania niesionego przez sygnał STM-N. Na tej
podstawie urządzenie odbiorcze dokonuje wyboru sygnału
taktowania. Istnieje zasada, że do synchronizacji zegara SEC
wybierany jest sygnał taktowania z tego wejścia STM-N,
w którym informacja SSM wskazuje na najwyższą jakość.
W przypadku, gdy na kilku wejściach krotnicy SDH
informacja SSM wskazuje tą samą jakość taktowania, dla
prawidłowego funkcjonowania procesu odzyskiwania
synchronizacji stosuje się zasadę priorytetów określających
kolejność, z jaką te sygnały mogą być pobierane do
synchronizacji. Należy zauważyć, że aby nie dopuścić do
tworzenia się pętli dla sygnałów synchronizacyjnych w sieci,
w informacji SSM wprowadzono kod DNU (ang. Do Not
Use) oznaczający, że sygnał w określonej sytuacji jest
zabroniony do użytku.
11. Sieć synchronizacyjna powinna być wyposażona
w system nadzoru nad pracą zegarów i jakością sygnałów
taktowania w przypadku awarii łączy lub zegarów oraz
środki zarządzania tą siecią.
12. Struktura sieci synchronizacyjnej powinna umożliwiać
stopniową, harmonijną rozbudowę tej sieci, bez
konieczności
dokonywania
zasadniczych
zmian
w wybudowanych wcześniej fragmentach sieci.
60
2
III. STRUKTURA SIECI SYNCHRONIZACYJNEJ
TELEKOMUNIKACJI POLSKIEJ
A. Wprowadzenie
Pierwsza sieć synchronizacyjna w Telekomunikacji Polskiej
(TP) była zbudowana w 1996 r. Doprowadzono
synchronizację z centrali międzymiastowej tranzytowej EWSD
do wszystkich central tranzytowych przy wykorzystaniu
użytkowych strumieni 2 Mbit/s przesyłanych w sieci PDH.
W kolejnym etapie, tj. w 1999 r. zbudowano sieć
synchronizacyjną w płaszczyźnie tranzytowej w oparciu
o urządzenia synchronizacyjne firmy Oscilloquartz , tj. PRC
oraz 13 urządzeń SSU oraz systemy SDH firmy
MITSUI/NEC. W latach 2002 -2004 r. nastąpiła rozbudowa
sieci TP w oparciu o systemy SDH firmy NEC, Alcatel,
Lucent, Siemens oraz trakty optyczne DWDM (ang. Dense
Wavelenght Division Multiplexing). Spowodowało to
rozbudowę sieci synchronizacyjnej o kolejne 47 urządzeń
SSU.
Stosowany w sieci sprzęt firmy Oscilloquartz (PRC, SSU,
źródła GPS) są to urządzenia o dużej niezawodności
eksploatacyjnej.
W strukturze sieci synchronizacyjnej od 1999 r.
wprowadzamy na różnych płaszczyznach sieci dystrybutory
taktowania i synchronizacji, które wspierają węzły z zegarami
SEC i dają możliwość dostarczenia większej ilości sygnałów
synchronizacyjnych. Pierwsze wprowadzone do sieci
dystrybutory zostały wykorzystane na potrzeby synchronizacji
105 central S-12, które potrzebowały 4 sygnałów synfazowych
2,048 MHz. Należy podkreślić, że firma Alcatel dla realizacji
ww. zadania wybrała dystrybutory DST-16 opracowane
w Politechnice Poznańskiej.
Aktualnie dystrybutory DST-16 stanowią 95%, tj. około 650
sztuk tego typu urządzeń stosowanych w sieci TP. Są to
urządzenia niezawodne w eksploatacji, wykorzystywane do
synchronizacji różnych elementów sieciowych, pełnią ważną
rolę w sieci synchronizacyjnej TP.
Ważną rolę w strukturze sieci synchronizacyjnej pełnią
opracowane w Politechnice Poznańskiej źródła GPS
oznaczone STFS/GPS. Urządzenia te również charakteryzują
się dużą niezawodnością pracy. Stanowią około 70% tego typu
sprzętu stosowanego w TP, 30% są to urządzenia firmy
Oscilloquartz.
Dla oceny jakości parametrów sieci synchronizacyjnej
i sieci telekomunikacyjnej wprowadzono w 1995 r. w ramach
współpracy z Politechniką Poznańską systemy pomiarowe SP2000 wraz z osprzętem (sondy pomiarowe elektryczne) oraz
w późniejszym okresie systemy pomiarowe SP-3000, a od
2009 r. systemy SP-4000.
W 2001 r. wprowadzono do stosowania opracowane
w Politechnice Poznańskiej źródła rubidowe częstotliwości
wzorcowej oznaczone Syn-Rb. W 2011 r. wprowadzono do
użytku sondy optyczne OSTM-1 i OSTM-16, które zostały
zastosowane w układach pomiarowych dla potrzeb realizacji
projektu EURO2012.
Sprzęt pomiarowy Politechniki Poznańskiej stanowi
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
podstawowe narzędzie utrzymania sieci synchronizacyjnej
i telekomunikacyjnej TP. Sprzęt ten charakteryzuje się dużą
niezawodnością pracy oraz bardzo dobrym rozwiązaniem
technicznym, tj. między innymi stosowanie w w/w. systemach
pomiarowych czterech kanałów pomiarowych, co umożliwia
jednoczesne pomiary w 4 punktach pomiarowych
przyczyniając się do szybszej realizacji wyznaczonych zadań,
jak również dużego usprawnienia przy lokalizacji awarii sieci
synchronizacyjnej lub telekomunikacyjnej.
B. Aktualny stan sieci synchronizacyjnej
W sieci Telekomunikacji Polskiej przyjęto system
hierarchiczny rozprowadzania sygnałów zegarowych według
zasad opisanych w punkcie II.
W modelu synchronizacyjnym przyjętym w TP rozróżnia się
następujące rodzaje zegarów (rys. 1):
- Pierwotne zegary odniesienia PRC,
- Wtórne źródła sygnałów synchronizacyjnych SSU dla
węzłów tranzytowych, regionalnych
- Zegary SEC (zegary wewnętrzne krotnic SDH),
- Zegary central tranzytowych (CMN, CT, CMT),
- Zegary central końcowych (CK, CKL).
W modelu tym przyjęto, że rezerwowym sygnałem
synchronizacyjnym dla urządzeń SSU, wykorzystanym
w przypadku niemożności pozyskania tego sygnału z sieci
SDH, jest źródło częstotliwości wzorcowej sterowane
sygnałem GPS zainstalowane przy urządzeniu SSU.
W nowej strukturze sieci synchronizacyjnej przyjęto
następujące zasady:
-łańcuch synchronizacji tworzą węzły z zegarami SSU
w liczbie do 8 oraz urządzenia z zegarami SEC umieszczone
między kolejnymi węzłami SSU w liczbie do 16.
Obecnie podstawowa sieć synchronizacyjna TP bazuje na
sprzęcie firmy Oscilloquartz. Zbudowana jest w oparciu o dwa
węzły z zegarami klasy PRC. Podstawowy węzeł z zegarami
klasy PRC wyposażony jest w trzy wzorce cezowe. Natomiast
rezerwowy węzeł klasy PRC wyposażony jest w dwa masery
wodorowe. W sieci synchronizacyjnej znajduje się
kilkadziesiąt urządzeń SSU zlokalizowanych w płaszczyznach:
tranzytowej i regionalnej.
Aktualnie sieć synchronizacyjna w płaszczyźnie tranzytowej
zbudowana jest na bazie nowobudowanej sieci transmisyjnej
ASON wykorzystującej w tej strukturze urządzenia
A1678MCC firmy Alcatel-Lucent oraz systemy DWDM. firmy
Alcatel-Lucent.
Rozprowadzanie
sygnałów
synchronizacyjnych w sieci ASON do około 40 węzłów
z urządzeniami A1678MCC jest realizowane z dwóch
ośrodków z zegarami PRC. Sieć synchronizacyjna
w płaszczyźnie regionalnej i strefowej jest zbudowana na
bazie sieci transmisyjnej SDH wykorzystującej urządzenia
1660SM oraz systemy DWDM firmy Alcatel-Lucent.
Doprowadzenie sygnałów synchronizacyjnych do węzłów
regionalnych i strefowych jest realizowane poprzez istniejące
urządzenia SSU płaszczyzny tranzytowej. Przyjęto zasadę, że
do urządzeń SSU podstawowe i rezerwowe sygnały
synchronizacyjne doprowadzone są z dwóch różnych
3
pierścieni pracujących w jednej płaszczyźnie sieci.
W przypadku niemożności pozyskania rezerwowego sygnału
synchronizacyjnego zgodnie z wyżej podaną zasadą, sygnał
powinien być doprowadzony ze źródła częstotliwości
wzorcowej sterowanego sygnałami systemu pozycyjnego GPS.
Transport sygnałów synchronizacyjnych w płaszczyźnie
tranzytowej sieci ASON jest realizowany w oparciu o łącza
STM-64 w systemach DWDM LH OLS400 i w systemach
DWDM A1696MS. Plan dystrybucji tych sygnałów
przedstawiono na planie synchronizacji ASON/GMPLS
pokazanym na rysunku 2. W planie tym zaznaczono dwa
węzły z zegarami klasy PRC, z których jest rozprowadzony
sygnał synchronizacyjny o jakości PRC, czyli o wartości Q=2
do wszystkich węzłów występujących w strukturze kratowej
sieci ASON. Zgodnie z zasadami dystrybucji do każdego
węzła sieci ASON doprowadzony jest podstawowy
i rezerwowy sygnał taktowania, który jest przekazywany
dwiema różnymi drogami. Sposób automatycznego
przełączania sygnałów taktowania z wykorzystaniem
informacji o jakości (Q) sygnału synchronizacyjnego
przesyłanej w komunikacie SSM opisano w punkcie II.
Sygnały synchronizacyjne są wyprowadzane z krotnic
1678MCC sieci ASON na zewnątrz przy wykorzystaniu
następujących interfejsów:
- wyjść traktów liniowych z sygnałem zbiorczym STM-N
(T1),
- wyjść synchronizacyjnych T4-sygnał 2 MHz,
- wyjść synchronizacyjnych T5-sygnał 2 Mbit/s.
Z wyjść synchronizacyjnych T4 dostarczane są sygnały do
urządzeń
SSU
płaszczyzny
tranzytowej
sieci
synchronizacyjnej. Z tych urządzeń SSU są synchronizowane
różne elementy sieciowe występujące w węźle oraz
dostarczane są sygnały zegarowe do niższych płaszczyzn sieci.
IV. PODSUMOWANIE
W następnych latach przewiduje się rozbudowę sieci ASON
w oparciu o krotnice 1678MCC oraz łącza STM-64
realizowane w systemach DWDM LH OLS400 i DWDM
1696MS. W związku z powyższym rozbudowa i modernizacja
sieci synchronizacyjnej będzie podporządkowana zmianom
jakie zostaną wprowadzone w optycznych sieciach
teletransmisyjnych.
Należy również wziąć pod uwagę konieczność stopniowej
wymiany sprzętu synchronizacyjnego, biorąc pod uwagę, że
pierwsze urządzenia SSU zostały wprowadzone do sieci TP 14
lat temu oraz to, że firma Oscilloquartz w 2013 roku
zaprzestaje produkcję urządzeń SSU typu OSA 5548B.
LITERATURA
[1]
[2]
Zarządzenie nr 77 z dnia 7 lipca 2010 r. Dyrektora PS i PU S.Soulie
„Wymagania techniczno-eksploatacyjne dotyczące synchronizacji sieci
telekomunikacyjnej”.
Reguły inżynieryjne dotyczące synchronizacji sieci teletransmisyjnej
SDH i sieci ASON/GMPLS, Warszawa czerwiec 2012.
XVI Poznańskie Warsztaty Telekomunikacyjne
61
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
PRC-1
GPS
PRC-2
GPS
4
GPS
Interfaces 2,048MHz
SSU
SSU
SSU
SSU
Interfaces
2,048MHz or
2,048Mbit/s
CMN & CT
CMN & CT
CT
SEC
SEC
GPS
SEC
GPS
SSU
GPS
SSU
Transit Layer
SEC
SEC
GPS
SSU
Interfaces
2,048MHz or
2,048Mbit/s
SSU
Interfaces
2,048MHz or
2,048Mbit/s
CT
SEC
SEC
SEC
SEC
SEC
SEC
SEC
GPS
GPS
SSU
Interfaces
2,048MHz or
2,048Mbit/s
SDU
SSU
CMT i CT
Interfaces
2,048MHz or
2,048Mbit/s
CT i Cx
SEC
CT
SEC
SEC
SEC
SEC
SEC
SEC
Interfaces
2,048MHz or
2,048Mbit/s
CMT&Cx
Cx
SEC
Regional Layer
SEC
SDU
Main Local Layer
SSU
SDU
ONU
NT
ONU
Cx
NT
primary route of synchronization signal within STM-N
reserve route of synchronization signal within STM-N
external signal 2MHz
external signal 2MBit/s
clock with E1/T1/J3
SEC – Synchronus Equipment Clock
ONU – Optical Network Unit
NT – Network Termination
NT
Lower
Local
Layer
Interfaces
2,048MHz or
2,048Mbit/s
NT
Cx
PRC – Primary Referency Clock
GPS – surce of frequncy standard controlled by (GPS) signal
SSU – Synchronization Supply Unit
SDU – Synchronization Distribution Unit
CMN – internatinal exchange
CT – tranzit exchange
CMT – metropolitan tranzit exchange
Cx – local exchange
Synchronization model within TP network
Rys. 1. Model synchronizacji sieci Telekomunikacji Polskiej
SSU
SSU
GPS
P=2
Q=2
1678MCC
P=1
Q=2
SSU
GPS
GPS
SSU
1678MCC
1678MCC
P=2
Q=F
P=2
Q=2
SSU
P=2
Q=2
SSU
GPS
SSU
P=1
Q=2
1678MCC
P=1
Q=2
P=2
Q=F
SSU
P=1
Q=2
1678MCC
SSU
P=1
Q=2
SSU
P=1
Q=2
1678MCC
T3 T3
P=1
Q=2
SSU
GPS
P=1
Q=2
1678MCC
1678MCC
SSU
SSU
GPS
1678MCC
SSU
GPS
1678MCC
Q=2
P=2
Q=F
GPS
SSU
P=2
Q=2
1678MCC
GPS
1678MCC
P=1
Q=2
1678MCC
P=1
Q=2
P=2
Q=2
1678MCC
P=1
Q=2
SSU
P=2
Q=2
S S U (r)
P=3
Q=2
P=1
Q=2
PRC-1
SSU
P=1
Q=2
1678MCC
SSU
droga podstawowa
droga rezerwowa
STM-N optyczny
STM-1 elektryczny
2,048 Mbit/s po retimingu
GPS
P=2
Q=2
1678MCC
1678MCC
SSU
P=1
Q=2
1678MCC
P=1
Q=2
P=1...6
Q=2, 4, 8, B, F
P=2
Q=F
GPS
1678MCC
P=2
Q=F
P=1
Q=2
1678MCC
SSU
P=2
Q=2
GPS
P=1
Q=2
P=1
Q=2
P=2
Q=F
1678MCC
SSU
P=1
Q=2
GPS
priorytet sygnału synchronizacji
jakość sygnału synchronizacji
P=2
Q=F
1678MCC
SSU
GPS
P=2
Q=2
PWT 2012 - Poznań, 14 grudnia 2012
1678MCC
P=2
Q=F
1678MCC
P=1
Q=2
P=1
Q=2
1678MCC
SSU
GPS
sygnał podstawowy 2MHz/2Mbit/s
sygnał rezerwowy 2MHz/2Mbit/s
Rys. 2. Synchronizacja w płaszczyźnie tranzytowej sieci ASON/GMPLS Telekomunikacji Polskiej
62
SSU
P=1
Q=2
1678MCC
P=1
Q=2
GPS
P=2
Q=2
P=3
Q=F
GPS
GPS
P=2
Q=F
P=1
Q=2
GPS
P=1
Q=2
PRC-2
1678MCC
P=2
Q=F
1678MCC
P=2
Q=2
SSU
GPS
GPS
T4 T4
1678MCC
SSU
GPS
1678MCC
P=3 Q=F
P=2
Q=2
P=2
Q=2
SSU
1678MCC
P=1
Q=2
T3 T3
1678MCC
1678MCC
P=2
Q=2
P=2
Q=2
2M
P=1
Q=2
GPS
P=1
Q=2
P=2
Q=F
SSU
STM - 1
S S U (p)
Q=2
P=1
Q=2
P=1
Q=2
P=1
Q=2
P=2
Q=F
1678MCC
P=2
Q=2
STM - N
GPS
SSU
P=1
Q=2
P=2
Q=F
P=1
Q=2
1678MCC
1678MCC
1678MCC
P=2
Q=2
1678MCC
SSU
SSU
P=1
Q=2
P=1
Q=2
WARSZAWA
1678MCC
GPS
GPS
1678MCC
1678MCC
P=2
Q=2
P=2
Q=F
P=1
Q=2
P=2
Q=2
1678MCC
1678MCC
P=2
Q=F
SSU
P=2
Q=2
P=2
Q=2
P=2
Q=2
SSU
P=1
Q=2
P=2
T4 T4
GPS
P=1
Q=2
P=1
Q=2
P=1
P=1
Q=2
P=1
Q=2
GPS
GPS
1678MCC
P=1
Q=2
1678MCC
P=2
Q=F
P=1
Q=2
P=2
Q=2
P=1
Q=2
GPS
GPS
P=2
Q=F
SSU
GPS
1678MCC
P=2
Q=2
1678MCC
GPS
SSU
P=2
Q=2
P=1
Q=2
P=1
Q=2
1678MCC
SSU
GPS
1678MCC
P=2
Q=F
P=2
Q=F
P=1
Q=2
GPS
GPS
1678MCC
P=2
Q=2
Plan synchronizacji
ASON/GMPLS
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Sondy Pomiarowe w Systemie Wspomagania
Synchronizacji
A. Dobrogowski, M. Jessa, M. Kasznia, K. Lange
Streszczenie—W referacie przestawiono sondy pomiarowe
uĪywane w systemach pomiarowych opracowanych przez zespóá
z Politechniki PoznaĔskiej. Zaprezentowano rodzaje sond
opisując ich zastosowanie, miejsce w torze pomiarowym
i podstawowe parametry.
Sáowa kluczowe — Synchronizacja sieci,
synchronizacyjne, sondy pomiarowe, zalecenia ITU-T
W
pomiary
I. WPROWADZENIE
ramach
prac
nad
systemem
wspomagania
synchronizacji zaprojektowano i zbudowano trzy
systemy pomiarowe: SP-2000, SP-3000 i SP-4000. Ich
zadaniem jest precyzyjne wyznaczenie podstawowych
parametrów sygnaáów taktowania i sygnaáów synchronizacji.
Filozofia dziaáania pomiarów tymi systemami pomiarowymi
zakáada pewne ograniczenia na ksztaát i czĊstotliwoĞü
wejĞciowych sygnaáów badanych. Przyrządy są przede
wszystkim przewidziane dla sygnaáów sinusoidalnych,
dopasowanych impedancyjnie dla 75 : w paĞmie:100 kHz –
10 MHz oraz dla standardu TTL w paĞmie: 1 Hz – 30 MHz.
Stąd, gdy zachodzi potrzeba pomiaru parametrów sygnaáu o
innym ksztaácie, lub o czĊstotliwoĞci, spoza zakresów
wejĞciowych systemów pomiarowych konieczne staje siĊ
zastosowanie specjalizowanych urządzeĔ poĞredniczących
zwanych tutaj sondami. Urządzenia te umoĪliwiają
wprowadzenie sygnaáów z innych Ĩródeá, a szczególnie z
kodów liniowych o róĪnych postaciach i przepáywnoĞciach.
Wszystkie prezentowany sondy zostaáy opracowane przez
zespóá pracowników Wydziaáu Elektroniki i Telekomunikacji.
II. PREZENTACJA OGÓLNA SYSTEMU SOND
Podstawowym zadaniem sond jest galwaniczne oddzielenie
Ĩródáa sygnaáu od urządzenia pomiarowego. Potrzeba
oddzielenia galwanicznego zachodzi wtedy, gdy Ĩródáo
sygnaáu i odbiornik sygnaáu są zasilane z róĪnych Ĩródeá.
Przykáadowo, strumieĔ E-1 pochodzi z krotnicy SDH
zasilanej napiĊciem staáym 48V z uziemionym biegunem
dodatnim, a miernik stopy báĊdów na wejĞcie którego
wprowadzamy strumieĔ badany jest zasilany z sieci prądu
przemiennego 230V/50Hz. Przy róĪnych zasilaczach Ĩródáa i
odbiornika istnieje niebezpieczeĔstwo páyniĊcia prądów
wyrównawczych, które istotnie zmniejszają odpornoĞü na
zakáócenia od sieci energetycznej, np. spowodowane
wáączeniem oĞwietlenia jarzeniowego pomieszczenia. Efekt
ten moĪna zmniejszyü przez odpowiednie zaprojektowanie
sieci energetycznej na danym obiekcie (czĊste stosowanie
uziomów, prowadzenie oddzielnego przewodu ochronnego,
itp.). Niestety, w wielu obiektach jest to niemoĪliwe (przerwa
w pracy obiektu) lub nieopáacalne.
Drugim zadaniem sond pomiarowych jest usuniĊcie
z badanego sygnaáu kodu linowego i wprowadzenie w to
miejsce sygnaáu taktu uzyskanego z tego kodu, który bĊdzie
kompatybilny
z
ukáadami
wejĞciowymi
systemu
pomiarowego. Oznacza to zapewnienie dopasowania
impedancyjnego, umieszczenie sygnaáu w odpowiednim
zakresie amplitudowym oraz nie wprowadzanie dodatkowych
fluktuacji fazowych. Sonda musi wiĊc zapewniaü równieĪ
ochronĊ przez oddziaáywaniem sygnaáów zakáócających.
Dla zapewnienia wszechstronnej wspóápracy systemów
pomiarowych z róĪnymi Ĩródáami sygnaáów opracowano
róĪne sondy, odpowiednie dla stosowanych w sieciach
telekomunikacyjnych przepáywnoĞci i hierarchii. Wykonane
sondy zapewniają wspóápracĊ z sieciami przewodowymi
i Ğwiatáowodowymi. Dla sieci przewodowych hierarchii PDH
rodzina sond obejmujĊ nastĊpujące sondy: sonda E1, sonda
E3, sonda E4, sonda E1/E1. Dla sieci SDH i sygnaáu
elektrycznego jest to sonda STM-1. Dla sygnaáu optycznego
są to sondy OSTM-1, OSTM-4, OSTM-16 i OSTM-64.
III. SONDY DLA HIERARCHII PDH
A. Sonda wydzielająca sygnaá taktowania ze strumienia
2048kbit/s, model E1
Sonda E-1 jest autonomicznym urządzeniem przeznaczonym
do wydzielenia ze strumienia liniowego 2048kbit/s sygnaáu
taktowania. Konstrukcja sondy zapewnia wydzielenie taktu
dla sygnaáu wejĞciowego kodowanego kodem HDB-3 oraz
sygnaáu niekodowanego. Postaü kodu HDB - 3 przedstawia
rysunek 1.
Autorzy są pracownikami Wydziaáu Elektroniki i Telekomunikacji
Politechniki PoznaĔskiej (e-mail [email protected];
[email protected];
[email protected];
[email protected]).
XVI Poznańskie Warsztaty Telekomunikacyjne
63
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
2
Rys.1. Przykáadowy fragment sygnaáu kodowanego HDB-3
Podstawowym zadaniem sondy E1 jest wydzielenie sygnaáu
taktowania ze strumienia E1 kodowanego kodem HDB-3 bez
pogarszania jakoĞci transmisji pracujących traktów.
Dodatkowo, sonda oddziela galwaniczne Ĩródáo sygnaáu E1
od urządzenia odbiorczego oraz stwarza moĪliwoĞü przesyáu
wydzielonego sygnaáu taktującego na odlegáoĞci znacznie
przekraczające 2-3 metry, np. 100 metrów, aĪ do stanowiska
z aparaturą kontrolno-pomiarową, np. z systemem SP-2000,
SP-3000 lub SP-4000. Jest to podstawowa sonda dedykowana
pomiarom synchronizacyjnym. Na wejĞcie sondy moĪna
podaü sygnaá HDB-3 lub sygnaá sinusoidalny. W pierwszym
przypadku sonda wydziela sygnaá taktujący, zapewnia
separacjĊ galwaniczną zarówno od Ĩródáa sygnaáu, jak i od
urządzenia odbiorczego oraz dostarcza standaryzowaną postaü
sygnaáu wyjĞciowego (sygnaá harmoniczny G.703-13) [1]. W
drugim przypadku, sonda E1 separuje galwaniczną oraz
dostarcza sygnaáu o standaryzowanej postaci, (sygnaá
harmoniczny G.703-13). Wygląd sondy E1 przedstawia
rysunek 2.
Rys.2. Páyta czoáowa sondy E1
B. Sonda wydzielająca sygnaá taktowania ze strumienia
34368kbit/s, model E3
Rys.3. Wygląd sondy E3
C. Sonda wydzielająca sygnaá taktowania ze strumienia
139264 kbit/s, model E4
Sonda E4 jest autonomicznym urządzeniem przeznaczonym
do wydzielenia ze strumienia liniowego 139264kbit/s sygnaáu
taktowania. Konstrukcja sondy zapewnia wydzielenie taktu
dla sygnaáu wejĞciowego kodowanego kodem CMI. Postaü
kodu CMI przedstawia rysunek 4.
Rys.4. Postaü kodu CMI
Podstawowym zadaniem sondy E4 jest wydzielenie sygnaáu
taktowania ze strumienia E4 kodowanego kodem CMI bez
pogarszania jakoĞci transmisji pracujących traktów.
Dodatkowo, sonda oddziela galwaniczne Ĩródáo sygnaáu E4
od urządzenia odbiorczego oraz stwarza moĪliwoĞü przesyáu
wydzielonego sygnaáu taktującego na odlegáoĞci znacznie
przekraczające 2-3 metry, np. 50 metrów, aĪ do stanowiska z
aparaturą kontrolno-pomiarową, np. z systemem SP-2000 lub
SP-3000. Jest to zatem sonda dedykowana pomiarom
synchronizacyjnym. Sonda E4 separuje galwaniczną oraz
dostarcza sygnaáu o standaryzowanej postaci, (sygnaá
harmoniczny G.703-13). Ze wzglĊdu na duĪą przepáywnoĞü
kabel sygnaáu wejĞciowego sondy musi byü krótki. Wygląd
sondy E4 przedstawia rysunek 5.
Sonda E3 jest autonomicznym urządzeniem przeznaczonym
do wydzielenia ze strumienia liniowego 34368kbit/s sygnaáu
taktowania. Konstrukcja sondy zapewnia wydzielenie taktu
dla sygnaáu wejĞciowego kodowanego kodem HDB-3. Sonda
wydziela sygnaá taktujący, zapewnia separacjĊ galwaniczną
zarówno od Ĩródáa sygnaáu, jak i od urządzenia odbiorczego
oraz dostarcza
wyjĞciowy sygnaá harmoniczny
o czĊstotliwoĞci 2148kHz i poziomie zgodnym z G.703-13.
Sygnaá ten jest akceptowany przez systemy pomiarowe SP3000 i SP-4000. Ze wzglĊdu na wysoką czĊstotliwoĞü
wejĞciową sonda E3 musi pracowaü z krótkim kablem
pobierającym z traktu sygnaá wejĞciowy (kilkadziesiąt cm).
Wygląd sondy E3 przedstawia rysunek 3.
64
PWT 2012 - Poznań, 14 grudnia 2012
Rys.5. Páyta czoáowa sondy E4
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
3
D. Sonda E1/E1
Sondy E1/E1 są w istocie rozwiniĊciem konstrukcyjnym
sond aktywnych E1. Sonda posiada dwa wejĞcia i dwa
wyjĞcia. Na wejĞcie E1 podajemy sygnaá HDB-3 E1
(zalecenie ITU-T G.703-9)[2]. W tym przypadku sonda
przeniesie sygnaá HDB-3 z wejĞcia na wyjĞcie HDB-3 oraz
wydzieli takt podając go na drugie wyjĞcie SIN sondy (záącze
BNC, 75:, sygnaá wyjĞciowy G.703-13). Oba wyjĞcia są
separowane galwanicznie od Ĩródáa sygnaáu. Przy podaniu
sygnaáu okresowego na wejĞcie SIN sonda E1/E1 separuje
galwanicznie Ĩródáo, np. wyjĞcie T4 krotnicy, od odbiornika
np. systemu pomiarowego SP-3000. Oprócz separacji
galwanicznej sonda dokonuje standaryzacji poziomu sygnaáu
wyjĞciowego dla szerokiego zakresu zmian napiĊcia sygnaáu
wejĞciowego (od 1Vpp do 10Vpp) z wejĞcia BNC.
Zastosowanie w systemie sondy E1/E1 umoĪliwia
jednoczesne wydzielenie z sygnaáu taktu dla pomiaru
wolnozmiennych fluktuacji fazy oraz bitowej stopy báĊdów.
Wygląd sondy przedstawia rysunek 6.
Podobnie jak dla sondy E3 czĊstotliwoĞü wyjĞciowa
sygnaáu z sondy STM-1 jest inna niĪ 2048KHz. Wynosi ona w
tym przypadku 2430KHz, co przy wspóápracy z systemami
SP-3000 i SP-4000 nie ma znaczenia dla jakoĞci pomiarów.
Znaczną czĊĞü sondy opracowaá dr inĪ. Henryk Batycki.
B. Sonda OSTM-1 do wydzielania taktu ze strumienia
optycznego 155.52 Mb/s
Sonda OSTM-1 jest autonomicznym urządzeniem
przeznaczonym do wydzielenia ze strumienia liniowego
155.52Mb/s sygnaáu taktowania. Konstrukcja sondy zapewnia
wydzielenie taktu dla sygnaáu wejĞciowego kodowanego
kodem NRZ. Postaü kodu NRZ przedstawia rys. 8.
Rys. 8. Sygnaá transmisyjny kodowany NRZ
Rys.6. Páyta czoáowa sondy E1/E1
Podstawowym zadaniem sondy OSTM-1 jest wydzielenie
sygnaáu taktowania ze strumienia optycznego kodowanego
kodem NRZ bez pogarszania jakoĞci transmisji pracujących
traktów. Jest przeznaczona do wspóápracy z systemami
pomiarowymi SP-3000 i SP-4000. Sonda
zapewnia
moĪliwoĞü przesyáu wydzielonego sygnaáu taktującego na
odlegáoĞci znacznie przekraczające 2-3 metry, np. 50 metrów,
aĪ do stanowiska z aparaturą kontrolno-pomiarową, np.
z systemem SP-3000 lub SP-4000. Jest to zatem sonda
dedykowana pomiarom synchronizacyjnym. Na wejĞcie
Optyczne sondy naleĪy podaü sygnaá optyczny na wejĞciowe
záącze SC/APC (gniazdo zielone). Wygląd sondy OSTM-1
przedstawia rysunek 9.
IV. SONDY DLA HIERARCHII SDH
A. Sonda STM-1 do wydzielania taktu ze strumienia
elektrycznego 155.52 Mb/s
Sonda STM-1 jest urządzeniem, którego zadaniem jest
wydzielanie taktu ze strumienia liniowego STM-1
o przepáywnoĞci 155,52 Mbit/s sieci SDH. Parametry sygnaáu
wejĞciowego dla sondy są zgodne z zaleceniem ITU-T G.703
- 12 (kod CMI). Wygląd sondy przedstawia rysunek 7.
INPUT 1
Rys.9. Páyta czoáowa sondy OSTM-1
Podobnie jak sonda STM-1, sonda optyczna dostarcza na
wyjĞcie sygnaá o czĊstotliwoĞci 2430 kHz i parametrach
impedancyjnych i amplitudowych zgodnych z zaleceniem
G.703-13.
SIGNAL
POWER
INPUT 2
POWER
Sonda STM-1
Rys.7. Páyta czoáowa sondy STM-1
XVI Poznańskie Warsztaty Telekomunikacyjne
65
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
4
C. Sonda OSTM-16 do wydzielania taktu ze strumienia
optycznego 2488,52 Mb/s
SP-3000/4000
Sonda OSTM-16 jest autonomicznym urządzeniem
przeznaczonym do wydzielenia ze optycznego strumienia
liniowego 2488,32Mb/s sygnaáu taktowania. Konstrukcja
sondy zapewnia wydzielenie taktu dla sygnaáu wejĞciowego
kodowanego kodem NRZ. PáytĊ czoáową sondy przedstawia
rysunku 10.
WejĞcie A
2048 kHz
Sygnaá referencyjny
WejĞcie B
2430 kHz
IN
OUT
2%
INPUT PROBE
Sonda OSTM-16
Rys.10. Páyta czoáowa sondy OSTM-16
Rys.11. Wspóápraca sondy OSTM-16 z SP-4000
Podobnie jak sonda STM-16, sonda optyczna posiada na
wyjĞciu sygnaá o czĊstotliwoĞci 2430KHz i parametrach
impedancyjnych i amplitudowych zgodnych z zaleceniem
G.703-13.
Sonda OSTM-4 nie jest w Polsce stosowana powszechnie.
Jest podobna do sond OSTM-1 i OSTM-16. RóĪnica polega
na innym kolorze páyty czoáowej. Natomiast sonda OSTM-64
obsáugująca przepáywnoĞü 9,95328Gb/s, ze wzglĊdu na
wymogi pracy w wysokiej czĊstotliwoĞci ma inna strukturĊ
elektroniczną. Natomiast jej páyta czoáowa ma równieĪ swój
róĪnicujący ją od inny sond indywidualny kolor.
V. EKSPLOATACJA SOND POMIAROWYCH
VI. PODSUMOWANIE
System sond jest niezwykle istotnym elementem caáego
systemu wspomagania synchronizacji. Znacząca wiĊkszoĞü
sygnaáów (wszystkie kodowane) są mierzone dziĊki
zastosowaniu sond. Pozwoliáo to odciąĪyü ukáady wejĞciowe
systemów pomiarowych od rozbudowanych interfejsów
zapewniających wspóápracĊ z róĪnymi kodami i
przepáywnoĞciami.
LITERATURA
[1]
[2]
[3]
Zalecenie ITU-T G.703-9.
Zalecenie ITU-T G.703-13.
Zalecenie ITU-T G.957.
Eksploatacja sond dotyczy zasad ich odpowiedniego
wáączenia w tor pomiaru fluktuacji fazy przez system
wspomagania synchronizacji. WspóápracĊ sond optycznych z
systemami pomiarowymi przedstawia rysunek 11.
66
PWT 2012 - Poznań, 14 grudnia 2012
FPGA-based time counters
with a wander measurement mode
2012
R. Szplet, P. Kwiatkowski, Z. Jachna, K. Rozyc
Department of Electronics
Military University of Technology
Warsaw, Poland
[email protected]
Abstract – This paper presents the design and basic technical
features of the series of virtual time interval counters with a wander
measurement mode. The counters are developed as computer card
with PCI interface and as portable modules with USB or Wi-Fi
interfaces. The measurement functions are executed by the time
counter unit integrated in an FPGA device. To achieve high
precision and wide measurement mode a two-stage interpolation
method is used. The precise and fast time-to-digital (T/D)
conversion in the second stage of interpolation is performed with
the aid of carry chains built-in the FPGA chip. The use of up to 16
such chains in a single interpolator leads to the high measurement
resolution of 1.2 ps. The short death time of T/D converters allowed
to introduce several specialized measurement modes, and among
them the wander measurement.
Keywords—time interval counter, two-stage interpolation,
wander measurement
I.
INTRODUCTION
The modern microelectronic technology allows to develop
of high-precision, low-cost time interval counter as a single
FPGA (Field Programmable Gate Array) device [1–5].
Typically in such a time counter a tapped delay line is used for
T/D conversion. Thus the resolution of conversion is limited
by the propagation time of a single delay element. To increase
the resolution two shifted delay lines [6] or independent delay
lines [7] may be used. The highest resolution and precision are
provided in time counters combining the two-stage
interpolation and multiple independent delay lines [8]. To make
such integrated counters easy applicable and more useful for
average user we considered their possible applications and
designed three models of virtual time counters in the forms of
a computer card and small portable modules. They differ each
other mainly with respect to the computer interface. The
computer card is equipped with the PCI interface, while
portable modules have USB or Wi-Fi interface.
Precise time interval measurements are crucial in many
areas of research and industry. In telecommunication, slow
variations in signal timing through a system (wander) are
investigated [9–12]. Due to these needs we introduced the
specialized operation modes of the counter, including
measurement of Time-Interval Error (TIE) and Time
Deviation (TDEV). Each developed counter is controlled by a
user-friendly software working in the Windows environment
and creating a virtual front-panel of the counter on the monitor
screen.
In this paper we describe the design of the integrated
counter chip, its operation in the wander measurement mode
and the respective technical details. We also present the results
of some performed tests.
II.
INTERPOLATION METHOD
The wide measurement range and high resolution in the
advanced integrated time counters are obtained by the use of
two-stage interpolation method (fig. 1). The wide range of the
measured time intervals T is achieved by counting an integer
number N of the reference clock periods T0, whose leading
pulse edges appear at the counter inputs between the leading
edges of the START and STOP pulses. These pulses represent,
respectively, the begin and end of the measured time interval
T. Then the product NT0 is calculated giving a rough estimate
of the T. The higher resolution is provided by two two-stage
interpolators, one on each input channel, which precisely
measure the fractional parts of T within single clock periods.
In the First Interpolation Stage (FIS) of each interpolator a
multiphase clock is commonly used. In our case we applied
Four Phase Clock (FPC). The FIS detects in which segment of
FPC the input pulse START (STOP) appears that allows to
evaluate of TST1 (TSP1). Simultaneously, the time interval TST2
(TSP2) between the START (STOP) pulse and the nearest edge
of the FPC is measured in the Second Interpolation Stage
(SIS). The final measurement result is calculated as:
T = NT0 + (TST 1 + TST 2 ) − (TSP1 + TSP 2 )
(1)
Figure 1. Two-stage interpolation method
This work was supported by the Polish National Science Centre under
contract no. DEC-2011/01/B/ST7/03278
XVI Poznańskie Warsztaty Telekomunikacyjne
67
III.
TIME COUNTER
The block diagram of time counter based on the two-stage
interpolation method is shown in figure 2. The FIS detects the
FPC segment in which an input pulse (START or STOP)
appears and synchronizes the latter with the nearest FPC edge.
This edge is then used both to enable the main counter and for
precise conversion in the SIS. A synchronizer is needed to
ensure proper work of the main counter. The results from FIS
and SIS are converted to NB code and then they are
transferred to the code processor. This processor executes the
statistical code density test [13], performed for identification
of the nonlinearity of interpolator, calculates and stores the
resulting transfer characteristic. The measurement results from
interpolator are calculated and corrected on-the-fly in that
block as well.
To obtain high measurement rate and minimize the dead
time between successive measurements a FIFO memory in the
FPGA device is used. It allows increasing the maximum
measurement rate up to 5 million measurements per second
(when measuring the “zero time interval”).
Figure 3. External view of the counter board with PCI interface and
the counter module with USB nterface
IV.
MEASUREMENT MODES
The counter can operate either in the calibration mode or
in the measurement modes. The calibration is performed
automatically in two consecutive procedures. In the first one,
the transfer characteristics of interpolators are identified. In
the second one, the time offset between two input channels
(START and STOP) is calculated.
Each of five measurement modes may be selected by
clicking the predefined virtual keys on the virtual front panel
of the counter (fig. 4).
Figure 2. Block diagram of the time counter
A simple FPC generator can be built as a tapped delay line.
Such line consists of few logic gates, which have appropriate
delays to obtain required clock phase segments. As the paths
delays are crucial to achieve high uniformity of the clock
segments, the manual routing is needed.
For precise T/D conversion within a single phase segment
of the FPC, tapped delay line is used. It consists of
multiplexers forming the fast carry chains in FPGA devices.
Such multiplexers offer the shortest propagation time among
all logical elements available in programmable devices. Due to
recent improvement in CMOS technology the delay of a single
multiplexer improved from tens of picoseconds (45 ps –
Spartan-3) to several picoseconds (19 ps – Spartan-6). To
further improve the delay line resolution beyond its cell delay
we proposed to use multiple independent delay lines for
creating an equivalent coding line [7, 8]. In this way the ndelay lines increase the resolution of about n-times.
Figure 3 shows the external view of the computer card
(left) and the portable module (right). The FPGA counter is
located in the center of the card (yellow label) and below is
the PCI interface chip.
68
Figure 4. Example of the virtual front panel of the counter operating
in the Time Interval mode
The fundamental measurement mode is the Time Interval,
which allows to measure of intervals with the resolution up to
1.2 ps and precision below 10 ps. The thresholds of the input
comparators can be set manually in the range –4V to 4V, or
can be preset on the fixed (TTL or CMOS) level. The
threshold level can also be adjusted automatically.
The Pulse Width mode allows for measurement of the
width of pulses applied to the input START or STOP. The
polarity of pulses (active edges) and threshold level are
defined in the same way as in the Time Interval mode.
PWT 2012 - Poznań, 14 grudnia 2012
In the Frequency mode, the signal frequency up to 3.5
GHz is measured with the use of the reciprocal method by
measuring the time interval consisting of a known integer
number of signal periods, calculating the duration of a single
period, and calculating its reciprocal.
In the Period mode the measurements are performed in a
similar way as in the Frequency mode.
In the Totalize mode the input pulses are counted within a
preset time gate. The duration of the gate may be selected
from 1 ȝs to 10 s, or not set (Open Gate mode). For each gate
the number of counted pulses and the respective mean
frequency in pulses per second are displayed.
V.
The TIE sample is defined over the sampling interval IJ0 as the
difference TE2 – TE1 measured in both ends of the
synchronized sampling interval IJ0s = t2 – t1. The worst-case
difference between the interval IJ0 and IJ0s is ±Tt .
We can observe a train of ‘TE samples’ (TE1, TE2,...) and a
train of ‘TIE samples’ of the form
TIE1(IJ0) = TE(t2) – TE(t1) = TE2 – TE1
TIE2(IJ0) = TE(t3) – TE(t2) = TE3 – TE2
.............
The result of TIE measurement over the observation
interval IJ is the root-mean-square value TIErms(IJ) calculated
from a finite number N of TE samples collected within that
interval. Utilizing the number n = N – 1 of the TIE(IJ0) samples
WANDER MEASUREMENT
A commonly used parameter for characterizing the wander
is the Time Interval Error (TIE) [11]. The maximum value of
TIE (MTIE), computed from an array of TIE data, can
characterize the frequency offsets and phase transients of a
tested signal to obtain a clear view of quality of relevant
electronic apparatus or systems. The time interval counters
offer possibility of precise measurement of TIE, MTIE and
Time Deviation (TDEV) related to wander.
The time error function TE(t) of a clock (in standards also
x(t)), with respect to a frequency standard, is the difference
between the time of that clock and the reference time
TIErms(IJ0,IJ) =
1 n
¦ (TIEi )2
n i =1
This is shown on the display panel (fig. 6) after collecting
data for a specified observation interval. The maximum time
interval error (MTIE) is defined as the maximum peak-to-peak
variation of TE in the all possible observation intervals IJ = nIJ0
within a measurement period T. The array of N samples of TE
(collected within an observation interval of IJ = nIJ0) may be
searched by software to find the minimum value TEmin and the
maximum value TEmax . Then MTIE(IJ0,IJ) = TEmax – TEmin.
x(t) Ł TE(t) = T(t) – Tref(t).
Referring to figure 5, the sampling interval IJ0 is specified
as follows. It begins with the START ENABLE signal
generated to allow measurement of the first TE (marked by
TE1) between the nearest zero-crossing of the tested signal (of
frequency ft ) and that of the reference signal (of frequency fr ).
Since the interval IJ0 is generated by the counter
asynchronously with reference to the tested signal, the time
lag from beginning of the sampling interval to the START
moment (t1) is randomly variable within the range (0 – 1)Tt .
That measurement is repeated in the end of the interval IJ0 and
the second TE equal to the delay TE2 is measured.
Figure 6. Virtual front panel of the counter operating in the TIE mode
Figure 5. Measurement of Time Interval Error
In many telecommunication applications, following the ITU
Recommendations there is a need to check if the measurement
data conform to the requirements specified in a relevant ITU
document. In such a case a relevant limit mask can also be
displayed. An example is shown in fig. 7, where the mask
defined in the ITU Recommendation G812 (clock 1) is
displayed. It allows for easy checking the quality of the clock
under test.In accordance with ITU-T Recs G.813 and G.812
the maximum sampling interval is limited to IJ0max = 1/30 s §
33 ms. The counters provide the IJ0 magnitude selectable in 12-5-10-20-33ms sequence.
XVI Poznańskie Warsztaty Telekomunikacyjne
69
the uncertainty was diminished below 35 ps. The newest
counter with 16 delay lines in SIS (T4100U), implemented in
45 nm CMOS technology (Spartan-6), allows to achieve the
precision better than 10 ps within the range of 100 ms. The
latter results are much lower than derived from popular and
commonly used counters such as HP53132 (Hewlett Packard)
and SR620 (Stanford Research Systems).
VII. CONCLUSIONS
Figure 7. Graph MTIE(IJ) with limit mask (G.812 clock 1)
VI.
The described time counters allow precise measurements
of long time intervals with picosecond accuracy at a lower
price than standalone counters of comparable parameters. The
natural applications of the counters are advanced systems for
time keeping, laser ranging and navigation. They can also be
used in industrial and research laboratories, and in ATE
systems. Thanks to the use of programmable devices (FPGA
and CPLD) the counter boards can be customized to match the
user applications.
EXAMPLE TEST RESULTS
REFERENCES
The measurement uncertainty of the successive generations
of our time counter is shown in figure 8. As a reference signal
of 10 MHz we used rubidium frequency standard FS725
(Stanford Research Systems). As sources of time intervals we
used in turn three delay generators: GFT1004 (Greenfield
Technology) for the range 10 ns – 200 ns, 81130A (Hewlett
Packard) for the range 500 ns – 5 Ns and T5300U (Vigo
System) for the time intervals longer than 10 Ns. The use of
three generators allowed to select their best operation ranges
(the lowest jitter) for delay generation.
The time counter with a single tapped delay line in SIS
(T2700U), implemented in 90 nm CMOS technology
(Spartan-3), provides measurement uncertainty below 50 ps.
By the use of the equivalent coding line composed of two
delay lines implemented in the same technology (T3200U)
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
Figure 8. Comparison of the precision (standard measurement uncertainty) of
succesisive versions of our time counter (T2700U with a single coding line,
T3200U with a double coding line and T4100U with multiple coding lines)
and selected commercial time counters (HP53132A, Hewlett-Packard and
SR620, Stanford Research Systems)
70
[13]
R. Szplet, J. Kalisz, R. Szymanowski, Interpolating Time Counter with
100-ps Resolution on a Single FPGA Device, IEEE Transactions on
Instrumentation and Measurement, vol. 49, no. 4, 2000, pp. 879-883.
J. Wu, Z. Shi, I. Y. Wang, Firmware-only Implementation of Time-toDigital Converter (TDC) in Field-Programmable Gate Array (FPGA),
IEEE 2003 Nuclear Science Symposium Conference Record, Portland,
Oregon, USA, pp. 177-181 Vol. 1
J. Song, Q. An, S. Liu, A High-resolution Time-to-Digital converter
Implemented in Field-Programmable-Gate-Arrays, IEEE Transactions
on Nuclear Science, vol. 53, no. 1, 2006, pp. 236-241
R. Szplet, J. Kalisz, Z. Jachna, A 45 ps time digitizer with two-phase
clock and dual-edge two-stage interpolation in Field Programmable
Gate Array device, Measurement Science and Technology, vol. 20
(2009) 025108, pp. 11
M. Daigneault, J. David, A Novel 10 ps Resolution TDC Architecture
Implemented in a 130nm Process FPGA, 8th IEEE Int. NEWCAS
Conference, 2010, pp. 281-284
M. ZieliĔski, D. Chaberski, M. Kowalski, R. Frankowski, S. Grzelak,
High-resolution time-interval measuring system implemented in single
FPGA device, Measurement vol. 35, 2004, pp. 311-317
R. Szplet, Z. Jachna, J. Kalisz, A Flash Time-to-Digital Converter with
Two Independent Time Coding Lines, IEEE 2011 ADC Forum, Orvieto,
Italy, 2011, 6 pp.
R. Szplet, Z. Jachna, P Kwiatkowski, K Rozyc, A 1.2 ps resolution
interpolating time counter based on multiple independent coding lines,
Sent for publication in Measurement Science and Technology
A. Dobrogowski, M. Kasznia, Time Effective Methods of Calculation of
Maximum Time Interval Error, IEEE Trans. Instr. Meas., vol. 50, no. 3,
2001, pp. 732-741.
S. Bregni, Synchronization of Digital Telecommunications Networks, J.
Wiley, London, 2002.
A. Dobrogowski, Sygnał czasu, Wyd. Politechniki PoznaĔskiej, 2003
A. Dobrogowski, M. Kasznia, Real-Time MTIE Assessment with
Flexible Control of Computation Process, Proc. IEEE Int. Frequency
Control Symposium, 2009, pp. 1102–1107
S. Cova, M. Bertolaccini, “Differential linearity testing and precision
calibration of multichannel time sorters”, Nuclear Instruments and
Methods, vol. 77, no. 2, 1970, pp. 269–276
PWT 2012 - Poznań, 14 grudnia 2012
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
1
Metody wyznaczania parametrów sygnaáów
synchronizacji w czasie rzeczywistym
dla pomiarów wielokanaáowych
Michaá Kasznia
Streszczenie—W pracy przedstawiono metody wyznaczania
w czasie rzeczywistym podstawowych parametrów sygnaáów
synchronizacji
–
dewiacji
Allana,
dewiacji
czasu
i maksymalnego báĊdu przedziaáu czasu – dla wielokanaáowych
pomiarów báĊdu czasu. Opisano koncepcjĊ obliczeĔ w czasie
rzeczywistym oraz struktury danych istotne w obliczeniach dla
pomiarów wielokanaáowych.
Sáowa kluczowe—sygnaá synchronizacji, dewiacja Allana,
dewiacja czasu, maksymalny báąd przedziaáu czasu
D
I. WPROWADZENIE
Allana (Allan deviation, ADEV), dewiacja
czasu (time deviation, TDEV) oraz maksymalny báąd
przedziaáu czasu (Maximum Time Interval Error, MTIE) są
podstawowymi parametrami opisującymi jakoĞü sygnaáów
synchronizacji sieci telekomunikacyjnej. Dewiacja Allana i
dewiacja czasu są miarami niestaáoĞci czĊstotliwoĞci
generowanej przez sygnaá taktowania [1]. Parametry te
opisują nierównomiernoĞü „chodu” badanego zegara.
Obydwa parametry pozwalają na zidentyfikowanie
dominującego typu szumu fazy sygnaáu taktowania oraz
wykrycie wahaĔ fazy sygnaáu o charakterze okresowym.
Maksymalny báąd przedziaáu czasu jest parametrem o
charakterze granicznym, opisującym maksymalne wartoĞci
wahaĔ fazy zegara w okreĞlonym przedziale obserwacji.
Parametr ten wykorzystywany jest do wymiarowania
pojemnoĞci buforów na granicy skal czasu. KaĪdy z tych
parametrów dostarcza innych informacji o zachowaniu
zegara, dlatego w celu uzyskania dokáadnej wiedzy o
jakoĞci badanego sygnaáu naleĪy wyznaczyü dewiacjĊ
Allana lub dewiacjĊ czasu i maksymalny báąd przedziaáu
czasu.
W praktyce pomiarowej sygnaáów synchronizacji czĊsto
spotyka siĊ koniecznoĞü wykonania oceny wiĊcej niĪ
jednego sygnaáu taktowania. Taka sytuacja ma miejsce, gdy
trzeba porównaü dwa zegary nie posiadając wiedzy, który z
nich jest lepszy (charakteryzuje siĊ wiĊkszą stabilnoĞcią).
Ze wzglĊdu na charakter estymatorów parametrów
uzyskanie wartoĞci estymaty parametru na podstawie
porównania dwóch zegarów nie wskazuje, który z nich jest
EWIACJA
Autor jest pracownikiem Katedry Systemów Telekomunikacyjnych
i Optoelektroniki Politechniki PoznaĔskiej, ul. Polanka 3, 60-965 PoznaĔ
([email protected]).
PracĊ wykonano w ramach projektu nr N N517 470540 finansowanego
przez Narodowe Centrum Nauki w latach 2011-2013.
lepszy (wartoĞci estymat są zawsze dodatnie), a uzyskany
parametr ma charakter wariancji „áącznej” dwóch zegarów.
W celu wskazania lepszego zegara trzeba wykorzystaü
dodatkowy zegar i wykonaü trzy porównania (trzy serie
pomiarów báĊdu czasu i obliczeĔ parametrów) metodą
„kaĪdy z kaĪdym”, a nastĊpnie odpowiednio dodając i
odejmując otrzymane wariancje uzyskaü parametry
charakteryzujące poszczególne zegary [3]. Metoda taka,
zwana takĪe metodą „3-corner hat”, wymaga znacznego
zaangaĪowania czasowego (trzy nastĊpujące po sobie
dáugotrwaáe serie pomiarowe oraz seanse obliczeniowe
parametrów) lub sprzĊtowego (zastosowanie trzech
mierników báĊdu czasu) w przypadku stosowania
jednokanaáowych urządzeĔ pomiarowych. Zastosowanie
wielokanaáowego miernika báĊdu czasu oraz algorytmów
obliczeniowych umoĪliwiających wyznaczanie parametrów
w czasie rzeczywistym jednoczeĞnie dla wielu kanaáów
(strumieni próbek báĊdu czasu) skróci czas i uproĞci
procedurĊ analizy. Jednoczesne pomiary wielokanaáowe
przydatne są takĪe w pomiarach sygnaáów synchronizacji w
sieci
telekomunikacyjnej.
Procedura
pomiaru
wielokanaáowego wraz obliczeniami w czasie rzeczywistym
umoĪliwia
bieĪące
Ğledzenie
jakoĞci
sygnaáów
synchronizacji wydobywanych ze strumieni danych
dopáywających do wĊzáa sieci telekomunikacyjnej z róĪnych
kierunków i natychmiastową reakcjĊ sáuĪb utrzymania sieci
w przypadku pogorszenia jakoĞci sygnaáu.
W pracy przedstawiono metody wyznaczania w czasie
rzeczywistym parametrów sygnaáów synchronizacji dla
pomiarów wielokanaáowych báĊdu czasu. Opisano metody
wyznaczania dewiacji Allana i dewiacji czasu, a nastĊpnie
przedstawiono koncepcje wyznaczania maksymalnego
báĊdu przedziaáu czasu w tej konfiguracji pomiarowej.
Rozpatrzono róĪne konfiguracje struktur danych istotne w
pomiarach wielokanaáowych i obliczeniach w czasie
rzeczywistym.
II. WYZNACZANIE ADEV I TDEV
WartoĞü dewiacji Allana oraz dewiacji czasu estymuje siĊ
wykorzystując ciąg N próbek báĊdu czasu ze wzorów:
ADˆ EV W TDˆ EV nW0 1
2n 2 W02
N 2n
xi 2n 2 xi n xi N 2n ¦
i 1
N 3n 1 ª
1
1
1
˜
«
¦
6 N 3n 1 j 1 «¬ n
(1)
2
º
¦ xi 2n 2 xi n xi »
»¼
i j
j n 1
2
(2)
gdzie: xi – wartoĞci báĊdu czasu wziĊte z odstĊpem W0; W=nW0
XVI Poznańskie Warsztaty Telekomunikacyjne
71
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
jest przedziaáem obserwacji, a N jest liczbą równomiernie
odlegáych wartoĞci báĊdu czasu xi [1].
Realizując obliczanie parametrów w czasie rzeczywistym,
podczas pomiaru próbek báĊdu czasu, naleĪy wziąü pod
uwagĊ, Īe dla bieĪącej chwili próbkowania i, nie są
dostĊpne próbki báĊdu czasu oznaczone indeksami i+n,
i+2n, oraz i+3n, gdyĪ te próbki nie zostaáy jeszcze
zmierzone. DostĊpna jest aktualnie zmierzona próbka
(oznaczona indeksem i) oraz próbki zmierzone wczeĞniej o
mniejszych indeksach. Dlatego teĪ konieczna jest zmiana
indeksowania oraz przeksztaácenie postaci wzorów (1-2).
Dodatkowo, w celu uproszczenia obliczania podwójnych
sum przy wyznaczania TDEV moĪna zastosowaü procedurĊ
opisaną w [2, 3] i zastąpiü podwójne sumowanie
pojedynczym sumowaniem trzeciej róĪnicy procesu báĊdu
czasu.. Przeksztaácenia estymatorów do postaci dogodnej do
obliczeĔ w czasie rzeczywistym przedstawiono w pracach
[4, 5]. WartoĞü dewiacji Allana dla i-tej chwili próbkowania
moĪna estymowaü ze wzoru:
ADˆ EVi nW0 2n2W02
1
A n xi 2xi n xi 2n 2
i 2n i 1
(3)
gdzie Ai 1 n jest sumą kwadratów drugich róĪnic
obliczoną w chwili ií1:
Ai 1 n i 1
¦ x j 2 x jn x j 2n 2
(4)
j 2 n 1
Operacje obliczania dewiacji czasu dla bieĪącej chwili
próbkowania i przebiegają wedáug wzoru:
TDˆ EVi nW0 >
1
1
S ov, i 1 n Si 1 n ' i n 2
˜
i 3n 1 6n 2
@
(5)
gdzie:
S ov, i n jest sumą caákowitą (zewnĊtrzną) daną wzorem:
S ov , i n S ov , i 1 n Si2 n (6)
Si n (7)
S i n jest sumą wewnĊtrzną obliczaną wedáug wzorów:
S3n n Si 1 n xi 3n 3 xi 2 n 3 xi n xi dla i ! 3n
¦ x j 2 x j n x j 2n 3n
(8)
j 2 n 1
a ' i n jest trzecią róĪnicą báĊdu czasu
' i n xi 3n 3 xi 2 n 3 xi n xi
(9)
WartoĞü parametrów dla bieĪącej chwili próbkowania i
zaleĪą od sum Ai 1 n , Sov,i-1(n) i Si-1(n) wyznaczonych dla
chwili i–1, próbki báĊdu czasu zmierzonej w bieĪącej chwili
próbkowania i oraz próbek zmierzonych n, 2n oraz 3n
odstĊpów próbkowania wczeĞniej.
Format wzorów (3-9) pozwala na realizowanie
wspólnego,
jednoczesnego
wyznaczanie
obydwu
parametrów w czasie rzeczywistym [5]. W obliczeniach
kaĪdego z parametrów do aktualizacji wartoĞci
odpowiednich sum, wykorzystywana jest próbka bieĪąca
oraz próbki zmierzone n, 2n lub 3n odstĊpów wczeĞniej..
Odczyt danych (odczyt próbek zmierzonych n, 2n lub 3n
odstĊpów wczeĞniej) odbywa siĊ w takiej sytuacji raz dla
liczonych wspólnie parametrów, co redukuje czas odczytu
danych wykonywanego osobno. Dodanie wyáącznie
dodatkowych operacji obliczeniowych (bez procedury
72
2
odczytu danych) nie wpáywa znacząco na czas obliczeĔ
prowadzonych w ramach jednego odstĊpu próbkowania [5].
III. WYZNACZANIE ADEV I TDEV DLA POMIARÓW
WIELOKANAàOWYCH
Wyniki testów przeprowadzonych takĪe przez autora
pracy pokazaáy duĪy „zapas” czasowy na przeprowadzanie
obliczeĔ w ramach pojedynczego odstĊpu próbkowania
[4, 5]. Pozwala to na rozwaĪenie dodatkowego „obciąĪenia”
– umieszczenie dodatkowych obliczeĔ wykonywanych w
odstĊpie
pomiĊdzy
kolejnymi
próbkami.
Takim
dodatkowym obciąĪeniem zaproponowanym przez autora w
pracy [6] byáo zwiĊkszenie liczby jednoczeĞnie
analizowanych sygnaáów – rozwaĪono realizacjĊ
wielokanaáowych obliczeĔ dewiacji Allana oraz dewiacji
czasu w czasie rzeczywistym podczas pomiaru próbek báĊdu
czasu. Do wyznaczania wartoĞci parametrów zastosowano
postaci estymatorów dane wzorami (3-9). PrzyjĊto, Īe
próbki báĊdu czasu pobierane są z równym odstĊpem
próbkowania w kaĪdym kanale pomiarowym. Rozpatrzone
zostaáy dwa sposoby organizacji danych bĊdących efektem
pomiaru wielokanaáowego. W pierwszym przypadku, dane
uzyskane z pomiaru báĊdu czasu w kaĪdym z mierników
(detektorów fazy) są przesyáane do komputera i zapisywane
jako osobne ciągi w osobnych strukturach danych (Rys. 1).
W drugim przypadku, próbki báĊdu czasu zmierzone w tej
samej chwili próbkowania (lub w nastĊpujących po sobie
chwilach wyznaczonych przez zegar wielofazowy, a
związanych z jednym odstĊpem próbkowania) áączone są w
interfejsie w wektory i zapisywane w postaci ciągu
wektorów w jednej strukturze danych (Rys. 2). KaĪdy ze
sposobów organizacji danych wymaga innej procedury
obliczeniowej przeprowadzanej w ramach jednego odstĊpu
próbkowania. W przypadku osobnych (rozdzielonych)
struktur danych, procedura odczytu danych dla danego
przedziaáu obserwacji (dla danego n) musi byü wykonana
osobno dla kaĪdego kanaáu pomiarowego (dla kaĪdej
struktury). W przypadku wspólnej struktury danych
obiektem dziaáaĔ obliczeniowych są M-wymiarowe
wektory. Odczyt danych wykonywany jest áącznie dla
próbek związanych z daną chwilą próbkowania we
wszystkich kanaáach. Procedura obliczeĔ dla rozdzielnych
struktur danych jest nastĊpująca:
wielokanaáowy system pomiarowy
sM(t)
miernik M
sM ref(t)
interfejs
...
sm(t)
...
...
strumienie danych
miernik m
sm ref(t)
interfejs
...
s1(t)
...
...
miernik 1
interfejs
s1 ref(t)
rozdzielone struktury danych
...
Mxi-3n
...
Mxi-2n
...
Mxi-n
...
Mxi
...
...
mxi-3n
...
mxi-2n
...
mxi-n
...
mxi
...
...
1xi-3n
...
1xi-2n
...
1xi-n
...
1xi
operatory TDEV i ADEV
Rys. 1. Wielokanaáowy pomiar báĊdu czasu z rozdzielonymi strukturami
danych
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
wielokanaáowy system pomiarowy
sM(t)
miernik M
sM ref(t)
...
sm(t)
...
strumieĔ danych
miernik m
sm ref(t)
...
s1(t)
interfejs
...
miernik 1
s1 ref(t)
wspólna struktura danych
...
...
...
Mxi-3n
...
mxi-3n
...
1xi-3n
...
...
...
Mxi-2n
...
mxi-2n
...
1xi-2n
...
...
...
Mxi-n
...
mxi-n
...
1xi-n
...
Mxi
...
mxi
...
1x i
...
...
operator TDEV i ADEV
Rys. 2. Wielokanaáowy pomiar báĊdu czasu ze wspólną strukturą danych
1. Pomiar próbek báĊdu czasu i zapis do osobnych
plików.
2. Odczyt próbek báĊdu czasu zmierzonych n, 2n i/lub 3n
odstĊpów próbkowania wczeĞniej w kanale nr 1.
3. Aktualizacja bieĪących sum dla kanaáu nr 1.
4. Wyznaczenie odpowiednich wartoĞci Ğrednich i ich
pierwiastków (bieĪących wartoĞci parametrów) dla
kanaáu nr 1.
5. Wykonanie kroków 2-4 dla kolejnych wiĊkszych
przedziaáów obserwacji (dla wiĊkszych n).
6. Wykonanie kroków 2-5 dla kolejnych kanaáów
pomiarowych.
Zastosowanie wspólnej struktury danych wymaga
pojedynczego odczytu danych dla danej wartoĞci n.
Procedura obliczeĔ dla tej struktury danych jest nastĊpująca:
1. Pomiar próbek báĊdu czasu i zapis w postaci wektora
do wspólnego pliku.
2. Odczyt próbek báĊdu czasu zmierzonych n, 2n i/lub 3n
odstĊpów próbkowania wczeĞniej dla wszystkich
kanaáów.
3. Aktualizacja bieĪących sum kolejno dla kaĪdego
kanaáu.
4. Wyznaczenie odpowiednich wartoĞci Ğrednich i ich
pierwiastków (bieĪących wartoĞci parametrów)
kolejno dla kaĪdego kanaáu.
5. Wykonanie kroków 2-4 dla kolejnych wiĊkszych
przedziaáów obserwacji (dla wiĊkszych n).
Zastosowanie pojedynczej wspólnej struktury danych dla
pomiaru wielokanaáowego skutkuje prostszą procedurą
obliczania wartoĞci parametrów. Testy obliczeniowe
obydwóch sposobów organizacji danych wykazaáy lepsze
„zachowanie”
takiej
procedury
w
obliczeniach
prowadzonych w czasie rzeczywistym [6].
IV. WYZNACZANIE MTIE
Maksymalny báąd przedziaáu
estymowany wedáug wzoru:
czasu
MTIE(W)
jest
3
W celu znalezienia wartoĞci MTIE w przedziale
obserwacji W zgodnie ze wzorem (10), naleĪy dokonaü
przejrzenia wszystkich przedziaáów (okien) o szerokoĞci
W=nW0 wystĊpujących w ciągu N próbek báĊdu czasu. W tym
celu „okno” obejmujące n+1 kolejnych próbek przesuwane
jest o odstĊp W0, a wiĊc o jedną próbkĊ od początku do
koĔca ciągu {xi}. Dla kaĪdego usytuowania okna
znajdowana jest wartoĞü miĊdzyszczytowa báĊdu czasu xpp.
WartoĞü
MTIE(W)
jest
maksymalną
wartoĞcią
miĊdzyszczytową báĊdu czasu xpp znalezioną dla wszystkich
moĪliwych usytuowaĔ okna o szerokoĞci W.
EstymatĊ MTIE z reguáy wyznacza siĊ po zakoĔczeniu
pomiaru báĊdu czasu. Szereg efektywnych czasowo metod
wyznaczania MTIE zaadoptowany zostaá jednak do obliczeĔ
parametru prowadzonych w czasie rzeczywistym, podczas
pomiaru próbek báĊdu czasu [7-9]. Najlepsze efekty
uzyskano wykorzystując metodĊ z zastosowaniem
dekompozycji binarnej [2, 7] oraz metodĊ skoku do
ekstremum EF [8, 9]. W pierwszej metodzie wykorzystuje
siĊ mechanizm dekompozycji binarnej ciągu próbek do
redukcji liczby danych wykorzystywanych do szukania
wartoĞci MTIE dla kolejnych przedziaáów obserwacji. W
drugiej metodzie opuszczane są takie usytuowania okna,
których przeszukanie nie przyniesie znalezienia wartoĞci
„bardziej ekstremalnych” od znalezionych do tej pory. Okno
przesuwane jest nie o jedną próbkĊ, lecz do najbliĪszego
ekstremum.
Realizując wyznaczanie MTIE w czasie pomiaru báĊdu
czasu, nowo zmierzoną próbkĊ porównuje siĊ z aktualnymi
wartoĞciami ekstremalnymi dla danego przedziaáu
obserwacji (w przypadku metody EF) lub z poprzednią
próbką (w przypadku dekompozycji binarnej). Wyznaczone
wartoĞci ekstremów znalezione dla aktualnie analizowanego
usytuowania okien pozwalają wyznaczyü nastĊpne
usytuowania okien. W przypadku metody z dekompozycją
binarną efekt porównania – w postaci pary ekstremów
(maksimum i minimum) – jest zapisywany do dalszych
obliczeĔ. RóĪnica uzyskanych wartoĞci jest porównywana z
bieĪącą
maksymalną
wartoĞcią
miĊdzyszczytową.
NastĊpnie, jeĞli wystarczająca liczba próbek zostanie
zmierzona, dokonywana jest analiza dla okna 4próbkowego, 8-próbkowego itd. Wszystkie niezbĊdne
operacje wykonywane są w odstĊpie pomiĊdzy kolejnymi
próbkami báĊdu czasu. Przykáady wyznaczania MTIE w
czasie rzeczywistym za pomocą tych metod przedstawione
są na Rys. 3 i Rys. 4 [7, 8].
próbki báĊdu czasu
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 …
4 7 6 2 4 7 4 7 5 8 2 5 5 1 4 3 …
okno
2-próbkowe
max 7 7 6 4 7 7 7 7 8 8 5 5 5 4 4
min 4 6 2 2 4 4 4 5 5 2 2 5 1 1 3
xpp maxpp
1
6
max §¨ max xi min xi ·¸
(10)
okno
max 7 7 7 7 7 7 8 8 8 8 5 5 5
k di d k n
¹
4
6
4-próbkowe min 2 2 2 2 4 4 4 2 2 2 1 1 1
gdzie W=nW0 jest przedziaáem obserwacji, ciąg {xi} jest
ciągiem N próbek báĊdu czasu miĊdzy sygnaáem badanym a Rys. 3. Wyznaczanie MTIE w czasie rzeczywistym metodą z dekompozycją
sygnaáem odniesienia wziĊtych z odstĊpem W0, natomiast binarną
wartoĞü n moĪe zmieniaü siĊ od 1 do Ní1.
MTˆIE nW0 1d k d N n© k d i d k n
XVI Poznańskie Warsztaty Telekomunikacyjne
73
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
próbki báĊdu czasu
xpp maxpp
4
5
próbki báĊdu czasu, M-ty kanaá
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 . . .
2 4 6 3 4 8 5 6 4 7 2 5 5 2 4 3 ...
1 2 3 4 5 6 7 8 9 10 11 12 13 14 … pozycja
4 7 6 2 4 7 4 7 5 8
… wartoĞü
.
.
(8, 4)
. . . current sampling interval
5
5
6
próbki báĊdu czasu, m-ty kanaá
(7, 2)
6
(8, 2)
.
.
...
Rys. 4. Wyznaczanie MTIE w czasie rzeczywistym metodą EF
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 . . .
6 7 3 3 5 4 5 7 2 7 3 8 2 5 5 1 ...
.
.
V. WYZNACZANIE MTIE DLA POMIARÓW
WIELOKANAàOWYCH
Opisane w poprzednim rozdziale metody wyznaczania
MTIE w czasie rzeczywistym, ze wzglĊdu na swoje
wáaĞciwoĞci, mogą byü wykorzystane do obliczeĔ parametru
dla pomiarów wielokanaáowych [13]. Zasada wyznaczania
parametru jest taka sama, jak w przypadku pomiarów
jednokanaáowych. Obydwie konfiguracje struktur danych,
opisane w rozdziale III, mogą byü wykorzystane do
obliczeĔ MTIE w zaleĪnoĞci od zastosowanej metody.
W przypadku zastosowania metody EF, obliczenia dla
kaĪdego przedziaáu obserwacji w kaĪdym kanale
realizowane są niezaleĪnie od siebie. Dla bieĪącej chwili
próbkowania przedziaáy obserwacji analizowane są w
kolejnoĞci od najkrótszego do najdáuĪszego, przy czym
najpierw wykonywane są operacje dla wszystkich
przedziaáów o tej samej szerokoĞci we wszystkich kanaáach.
Ze wzglĊdu na niezaleĪnoĞü wykonywanych operacji,
mierzone próbki mogą byü zapisywane w rozáącznych
strukturach danych. W celu przyspieszenia obliczeĔ,
ostatnie zmierzone nmax+1 próbek (Wmax=nmaxW0 jest
najdáuĪszym rozwaĪanym przedziaáem obserwacji) w
kaĪdym kanale moĪe byü buforowane w postaci tablicy w
pamiĊci operacyjnej komputera. Przyspieszy to dostĊp do
danych podczas operacji przeszukiwania usytuowaĔ
analizowanych okien. Idea wyznaczania MTIE w tej
konfiguracji przedstawiona jest na Rys. 5.
W przypadku zastosowania metody z dekompozycją
binarną, obliczenia realizowane są niezaleĪnie w kaĪdym
kanale. Ze wzglĊdu jednak na regularnoĞü procesu redukcji
danych, do porównaĔ wykorzystuje siĊ próbki o tych
samych indeksach. W celu uáatwienia dostĊpu do danych,
dla tej metody efektywne bĊdzie zastosowanie wspólnej
struktury danych ulokowanej na dysku twardym komputera
oraz, w miarĊ moĪliwoĞci, w postaci bufora w pamiĊci
operacyjnej. Idea wyznaczania MTIE z zastosowaniem
dekompozycji binarnej przedstawiona jest na Rys. 6.
bieĪąca chwila próbkowania
próbki báĊdu czasu, M-ty kanaá
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 . . .
2 4 6 3 4 8 5 6 4 7 2 5 5 2 4 3 1 7 4 7
(7, 1)
...
(7, 1)
próbki báĊdu czasu, m-ty kanaá
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 . . .
4 7 6 3 4 7 4 6 5 3 2 5 5 1 4 6 1 3 4 6
(6, 1)
...
(6, 1)
próbki báĊdu czasu, 1. kanaá
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 . . .
6 7 3 3 5 4 5 7 2 7 3 8 2 5 5 1 4 3 1 6
(5, 1)
.
.
.
.
7 7 5 5 5 4 4
4 2 2 5 2 2 3
.
.
.
.
.
.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 . . .
4 7 6 3 4 7 4 6 5 3 2 5 5 1 4 6 ...
próbki báĊdu czasu, 1. kanaá
kolejnoĞü wykonywania dziaáaĔ
dla przedziaáu obserwacji Wi=niW0
(8, 1)
Rys. 5. Wyznaczanie MTIE w czasie rzeczywistym metodą EF dla
pomiarów wielokanaáowych
74
4
.
.
.
.
7 7 8 8 5 5 5
2 3 3 2 2 5 1
.
.
.
.
.
.
.
.
.
.
5 3 5 5 5 4 6
3 2 2 5 1 1 4
.
.
.
.
.
.
7 7 7 5 5 5
2 2 2 2 2 2
.
.
.
.
.
.
5 5
2 2
6 5 5 5 5 6
2 2 2 1 1 1
.
.
.
.
.
.
6 6
1 1
wspólna struktura danych dla wszystkich kanaáów
7 8 8 8 8 5
2 2 2 2 2 1
.
.
.
.
.
.
8 8
2 1
Rys. 6. Wyznaczanie MTIE w czasie rzeczywistym metodą z dekompozycją
binarną dla pomiarów wielokanaáowych
VI. PODSUMOWANIE
Opisane w pracy metody wyznaczania podstawowych
parametrów
sygnaáów
synchronizacji
w
czasie
rzeczywistym dla pomiarów wielokanaáowych wpáywają w
istotny sposób na uáatwienie przyspieszenie procesu oceny
jakoĞci badanych sygnaáów. Zastosowanie przetestowanej w
[10] metody buforowania wektorów danych przyspiesza
znacząco obliczenia przeprowadzane w czasie pomiĊdzy
kolejnymi chwilami próbkowania. Metoda ta pozwoli
uniknąü przekroczenia przez wykonywane operacje czasu
równego dáugoĞci odstĊpu próbkowania. W celu
wykorzystania moĪliwoĞci komputerów stosowanych do
obliczeĔ,
wskazane
jest
zastosowanie
obliczeĔ
wielowątkowych, szczególnie w przypadku zastosowania
metody EF dla rozáącznych struktur danych.
LITERATURA
[1]
[2]
Zalecenia ETSI EN 300 462, ITU-T Rec. G.810, ANSI T1.101-1999.
M. Kasznia, “Some Approach to Computation of ADEV, TDEV and
MTIE”, Proc. 11th European Frequency and Time Forum, pp. 544548, Neuchatel 4-6 March 1997.
[3] S. Bregni, “Synchronization of Digital Telecommunications
Networks”, J. Wiley & Sons, 2002.
[4] A. Dobrogowski, M. Kasznia, “Real-time assessment of Allan
deviation and time deviation,” Proc. 2007 IEEE Int. Freq. Contr.
Symp. and 21st EFTF, pp. 887-882, Geneva, 29 May – 01 June 2007.
[5] A. Dobrogowski, M. Kasznia, “Joint Real-time Assessment of Allan
Deviation and Time Deviation”, Proc. of 22nd European Frequency
and Time Forum, 22-25 April 2008, Toulouse, France.
[6] M. Kasznia, “Multi-channel Real-time Computation of ADEV and
TDEV”, Proc. 24th European Frequency and Time Forum, 13-16
April 2010, Noordwijk, Netherlands.
[7] A. Dobrogowski, M. Kasznia, “On-line computation of MTIE using
binary decomposition and direct search with sequential data
reducing,” Proc. 2007 IEEE Int. Freq. Contr. Symp. and 21st EFTF,
pp. 877-882, Geneva, 29 May – 01 June 2007.
[8] A. Dobrogowski, M. Kasznia, “Real-time MTIE assessment with
flexible control of computation process,” Proc. EFTF’09 IEEEFCS’09 Joint Conf., pp. 1102-1107, Besancon, France, 20-24 April
2009.
[9] A. Dobrogowski, M. Kasznia, “Implementation of real-time MTIE
assessment method”, Proc. 2011 Joint Conf. IEEE FCS and EFTF,
pp. 304-309, San Francisco, California, USA, May 2-5, 2011.
[10] A. Dobrogowski, M. Kasznia, “Real-time assessment of dynamic
Allan deviation and dynamic time deviation,” Proc. 26th European
Frequency and Time Forum, Goeteborg, Sweden, 24-26 April 2012.
[11] A. Dobrogowski, M. Kasznia, “Some Concepts of the Real-Time
MTIE Assessment for Multi-Channel Time Error Measurement,”
Proc. 2012 IEEE Freq. Contr. Symp., pp. 493-498, Baltimore, USA,
May 22-24, 2012.
PWT 2012 - Poznań, 14 grudnia 2012
SesjaIII
Dydaktyka w telekomunikacji
XVI Poznańskie Warsztaty Telekomunikacyjne
75
76
PWT 2012 - Poznań, 14 grudnia 2012
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
1
Signal processor as a master of local bus and as
a slave of global bus (VME standard)
Bogusáaw J. WiĞniewski, Barbara E. Szecówka – WiĞniewska, Jacek A. Ostrowski
Abstract—The supervisor system based on MC68040/60
microprocessor is used to service the laboratory stand with
TMS32C5402 signal processor. Memory address space sharing
technique is utilized in order to achieve mutual
communication. The laboratory stand uses VME standard and
its own defined local bus for specific input/output card.
Index Terms—VME standard, signal processor card,
memory sharing technique
T
I. INTRODUCTION
he modern market of signal processors (sμp) has been
dominated by solutions delivered by companies such as:
Texas Instruments (TI), Analog Devices (AD), and
Freescale Semiconductor (FS) – previously Motorola. These
companies, except from Motorola produce a wide range of
fixed as well as floating point processors. Motorola’s
successful M9600X floating point family disappeared from
the mass production excluding substantive and economic
reasons. There are development trends like: expansion of
cache
memory
capacity
(code-data,
two-levels),
constructions with double MAC (TMSC55X5), separated
multiprocessor (ADSP2106X), multicore (TMSC647X and
’C66X; ADSP-BF5XX; MSC8XXX) and also tandem: the
microcontroller with sμp components (ADUC 8XXX;
‘C7XXX; MC/DSP 56F8XX). In this last case, producers
rely on their own microcontrollers and also on constructions
based on the ARM7 architecture. Students, who study
Electronics and Telecommunication at the Faculty of
Computer Science, Electronics and Telecommunication
have an optional subject of signal processors. Before the
students begin get to know advanced systems in the form of
so-called kits, connected with the computers, they are
introduced into specific construction of sμp produced by TI,
AD and FS. These basic processors are: TMS32C5402,
ADSP2161 and MC56311 [1],[2],[3]. The laboratory stands,
Manuscript received November 12, 2012.
B. J. WiĞniewski, Department of Electronics, AGH University of
Science and Technology , al. Mickiewicza 30, 30-059 Krakow, Poland
(corresponding author phone: 0048 609 192 139; fax: 0048 12 633 23 98 ;
e-mail: [email protected]).
B. E. Szecówka-WiĞniewska, Department of Electronics,
AGH
University of Science and Technology, al. Mickiewicza 30, 30-059
Krakow, Poland (e-mail: [email protected]).
J. A. Ostrowski, Department of Electronics, AGH University of Science
and Technology, al. Mickiewicza 30, 30-059 Krakow, Poland (e-mail:
[email protected]).
with regard to illustrative effects are not “closedconstruction”, but “open-construction” in module form, with
the dedicated bus.
II. LABORATORY STAND
All the laboratory stands with signal processors were
mutually standardized. As the base, the VME bus was used.
This assumption comes from the fact, that all 32-bits
microprocessor systems in this laboratory use this standard.
In this case, a central bus arbiter is required and at least one
card has to behave as MASTER. With regard to usage this
stand for development and didactic purposes TMS32C5402
will operate in „microprocessor” mode – using external
code and data memories. Activated application’s program
could be transfer from master through HOST port.
Transfered program comes to internal memory. In this case,
it is impossible to observe of the bus cycle by cycle
application’s operation (using !READY line logic - !means
negation). This mode is irreplaceable during hardware
activation and it allows to better understanding of system
operation. It is necessary to provide direct access to external
code and data space with omitting TMS32C5402 circuit.
Taking over control by sȝp screen/keyboard service and for
example self-assembling is however unprofitable. So it was
decided that one of the microprocessor laboratory system
take over application with TMS32C5402 circuit service.
These systems cooperate with keyboard/screen interfaces
and they are equipped with resident software (editorassembler, disassembler, monitor-debugger).
This system would prepare binary form of application
program for TMS32C5402.
Supervisor system should mutually share out its own
memory space with TMS32C5402 circuit – program and
data sȝp memories will be a part of address space of VME
bus [4]. This standard is not able to predict lines for all
individual interfaces. It has however numerous group of
lines in order to define them by user. Every sμp has other
internal lines. That is why only two serial ports and simple
synchronic bus (addresses LA5-0, !LCS, !LRD, !LWR,
!INTA, !INTB, LD15 - 0) has been defined. Cards
cooperating with this bus will be mapped in data space of
sμp. Above-cited idea is shown in the Fig. 1
XVI Poznańskie Warsztaty Telekomunikacyjne
77
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
In order to operation of sȝp interfaces also in other
configurations (without sμp), the card which realize
2
-
edition of source file and check of syntax,
assembling with syntax detection,
loading binary code form into code memory space
of sμp,
- software step by step work of sμp application, with
viewing registers and data memory of sμp.
First two functions are internal problem of the system with
μp M68040/60. The other functions base on mutually
sharing out memory spaces. Detailed using of both
superposed spaces is shown in Fig. 3.
Some spaces are related to special activities. So there are:
- reset of sμp,
- interrupt request to sμp (emergency stop of
Fig. 1. Organization of address space
mapping local bus space into the VME bus memory space
has been constructed. Supervisor system is based on
M68040/60 processor [6]. It has their own interfaces and
resident software. It can also effectively ensure edition and
assembling for TMS32C5402 circuit. Supervisor system
should also have possibility instruction step by step work
for application with sμp, with simultaneous viewing its
contents and resources. Complete laboratory stand is shown
in the Fig. 2.
TMS32C5402 circuit has also testing port JTAG with basic
functions set. There is a possibility of viewing sμp signals,
after adding standard console of JTAG port. The solution,
which is applied in laboratory has unified hardware and
Fig. 3. Organization of shared address space
Fig. 2. Laboratory stand with signal processor
successively added software for the next μp, equipped with
JTAG port [5].
III. COOPERATION BETWEEN SYSTEMS
Supervisor system has to take over a number of functions in
relation to the application with sμp:
78
application),
- returnable interrupt to supervisor system.
Supervisor system, after code loading to common space,
places there also procedure of INT0 interrupt service (the
highest priority). The interrupt is used to software step by
step work. INT0 interrupt service transfers registers states
to supervisor system through the isolated space in data
memory. Next it reads status word situated in the data
memory. When supervisor system changes this status, sμp
receives registers state and executes the next instruction of
tested program. Because supervisor system sets earlier INT0
interrupt, all this procedure repeats again.
In case of necessity of setting the software trap(s),
supervisor system places the software interrupt instruction
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
3
(SINT) at the given address instead of original instruction.
The supervisor system has already loaded procedure of
software interrupt (SINT) during the initialization handler.
When the sȝp goes to SINT instruction service, it writes any
values into its own data space (FF80yFFBF). As an effect,
supervisor system receives the IRQ5 interrupt. Signal
processor waits for supervisor system reaction (INT0
interrupt). Further actions will be determined by the bits of
status location in the data memory. For viewing purposes,
the possibility of emulation HOST port signals, through
locations in address space has been predicted. The detailed
description of such solutions is not in the frame of this
article.
IV. CHOSEN HARDWARE IMPLEMENTATIONS
Normally, the access to the memory from the both system
side should be assured by two-port memory. Difficult to
solve problems, which can lead to software cycle repetition
may occur during the superposition of addresses. In order to
avoid such situations, the safe method using
!HOLD/!HOLDA lines of signal processor is applied. When
the supervisor system addresses the common area, then sȝp
receives bus request !HOLD signal. Until there is no answer
on the !HOLDA line, supervisor system adds wait states
into the cycle. Next, the finite state machine on the sȝps
card realizes the transfer. Only after finishing the transfer,
the supervisor
system receives acknowledge on the
!DTACK line and transfer cycle can be finished. Block
diagram is shown in the fig. 4.
After reading from supervisor system, it is necessary to save
the word, which is read from memory. Buffering is
necessary
only from the supervisor system. We take
Fig. 6. View of the complete TMS32C5402 card.
advantage of the fact that after setting !HOLDA signal, sȝp
goes into high impedance state on its lines. Detailed timing,
which illustrates the operation of finite state machine is
shown in the fig. 5. The “safe” timings without finite state
machine compression is assured. This ensures dependable
transfer. Speed during the step by step work/trap is not an
essential parameter.
Fig. 4. Block diagram of shared memory.
systems for signal processors such as: Analog Devices and
Motorola/Freescale Semiconductor was implemented. In the
Fig. 5. Timing for finite state machine
V. CONCLUSIONS
The constructed stand has been checked during the creation
of application and laboratory classes. The view of the
complete TMS32C5402 card is shown in Fig. 6. Processor
M68040/60 has turned out to be very effective during the
process of creating editor/assembler and cooperate
procedures. Therefore in the similar way, the development
case of sȝp, currently introduced into production, we can
already use effective tools, connected with tests ports. It will
be possible to build independent development system with
embedded editor/assemble/debugger software.
XVI Poznańskie Warsztaty Telekomunikacyjne
79
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT)
<
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
80
Company website: http://www.ti.com
Company website: http://www.freescale.com
Company website: http://www.analog.com
http://www.vita.com, VME Standard, 10.04.1995
B. J. WiĞniewski, B. E. Szecówka-WiĞniewska, J. A. Ostrowski,
“Using the JTAG test port in the microprocessor system Lab”, PWT
2008, XIII PoznaĔskie Warsztaty Telekomunikacyjne, PoznaĔ 11
grudnia 2008.
B. J. WiĞniewski, „ĝrodki uruchomieniowe w laboratorium
dydaktycznym Techniki Mikroprocesorowej”, Elektronika :
konstrukcje, technologie, zastosowania, vol. 50, no.10, pp. 42-45,
Oct. 2009.
PWT 2012 - Poznań, 14 grudnia 2012
4
2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Design and Implementation of GEPON
Architecture in Laboratory Testbed
Magdalena Młynarczuk, Lech Smoleński
Abstract—This paper presents a proposition of GEPON
architecture for the didactic purpose. The GEPON architecture
is implemented in access networks laboratory testbed. The
paper includes a brief description of GEPON standardization, a
description of laboratory GEPON equipment architecture and a
short presentation of proposed laboratory exercises. The
exemplary performance results are included.
The proposition of GEPON architecture testbed enables
students to investigate Passive Optical Network (PON)
architecture, services and maintenance of the system. The aim
of proposed architecture is to familiarize the students with PON
solution and cooperation PON solution with xDSL systems in
access network or with METRO network. The proposed testbed
architecture as a part of Next Generation Network is well
inscribed in the needs of information society. For this reason its
design and implementations is crucial in the didactic process of
the technical university.
Index Terms—PON, GEPON, access network, ONU, OLT
C
Switch replacement by low-cost passive optical element
additionally minimizes the amount of optical transceivers,
terminations, with no need to power supply. This solution
named as Passive Optical Network (PON) [1] is generally
considered to be the best in aspect of operating expenditures
(OPEX) and capital expenditures (CAPEX).
In the article basics of design and implementation of
Gigabit Ethernet Passive Optical Network (GEPON)
architecture are presented. The paper is organized as follows.
General
information
about
GEPON
architecture
standardization and laboratory testbed are depicted in section
II. Section III is devoted to proposition GEPON laboratory
testbed for the didactic purpose. The exercises in laboratory
GEPON network are proposed in section IV. Section V
presents performance tests results. Section VI concludes the
paper.
II.GEPON ARCHITECTURE STANDARDIZATION
I. INTRODUCTION
society demands concerned with highdefinition HDTV, e-learning, online games,
sophisticated multimedia services, require high capacity in
the access network. Typical Digital Subscriber Line (DSL)
solutions with one copper pair per user, with few kilometers
distance and a speed limitation are not sufficient to satisfy
information society requirements.
The better proposition in such case is point-to-point (P2P)
optical access network solution with dedicated fibers from
Central Office (CO) to each end user subscriber. That
approach from economic aspects is expensive due to the fact
that it requires dedicated fiber deployment and optical
transponders.
To reduce fiber installation cost and resolve traffic
aggregation problem remote switch should be placed in the
neighborhood of end users. Related to previous approach in
this solution seriously limitation is requirement of electrical
power supply.
ONTEMPORARY

Manuscript received November 12, 2012. PON access network equipment,
described in this paper was supported by the Regional Operational Program for
Pomorskie Voivodeship for 2007-2012 years, Poland, under the Project
"Technical infrastructure modernization of student labs at the Faculty of
Electronics, Telecommunications, and Informatics at Gdansk University of
Technology."
The authors are with the Department of Teleinformation Networks, Faculty
of Electronics, Telecommunications and Informatics, Gdansk University of
Technology, Gabriela Narutowicza 11/12 Street, 80-233 Gdansk, (e-mail:
[email protected], [email protected])
There are two major organizations involved in PON
standardization: Institute of Electrical and Electronics
Engineers (IEEE) and Telecommunication Standardization
Sector of the International Telecommunication Union (ITUT). Both organizations publish standards dedicated to PON.
This section is a brief description of evolutionary of PON
toward GEPON.
First recommendations concerned with Access Network
(AN) are presented in [1], [2], [3]. The recommendations
define basic elements of PON. The reference configuration
shown in Fig. 1 includes AF (block of Adaptation Function),
ONU (Optical Network Unit), ODN (Optical Distribution
Network), OLT (Optical Line Termination), references
points, management interface (Q3).
The AF block is responsible for providing DSL
transmission over copper to the end user. ODN consists of
only passive elements splitters, connectors and fiber optics.
The ONU acts as an optical interface on the ODN side and
implements the UNI interfaces at the end user side. The OLT
is defined by core shell, common shell and service shell
functions. The core shell functions include digital cross
connects functions, transmission multiplex function, ODN
interface function. The common shell is responsible for power
supply and Operation and Maintenance (OAM) function. Due
to service shell OLT is able to support two or more different
services. The conception of PON architecture refers to all
XVI Poznańskie Warsztaty Telekomunikacyjne
81
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
2
has a downstream capacity of 2.488 Gb/s and an upstream
capacity of 1.244 Gbp/s that is shared among users.
Encryption is used to keep each user's data secured and
private.
Apart from GEPON the management of GPON is solved by
using of new separate channel, where the GEPON
management uses SNMP protocol.
Although there are other technologies that could supply
fiber to the home, passive optical networks (PONs) like
GPON and GEPON are generally considered the strongest
candidate for widespread deployments. The rest of the paper
is devoted to GEPON implementation, which is used in the
laboratory.
Fig. 1. Reference configuration for the PON [1].
PON solution presented in Fig 2.
The first PON solution is All Passive Optical Network
(APON). The APON has appeared in the conception of Full
Service Access Network (FSAN) as a PON-based network
that uses asynchronous transfer mode (ATM) [1].
The extended functionally of APON offers Broadband
Passive Optical Network (BPON) architecture. Apart from
APON, BPON [3] offers higher symmetric transmission
speed, up to 622Mb/s downlink and uplink.
The EPON (Ethernet Passive Optical Network) is an
optical network, able to carry on data encapsulated in
Ethernet frames. The EPON standard has been introduced in
the 802.3ah document. The basic configuration of EPON is
defined in the [4]. In downstream direction Ethernet frames
converted to optical form are transmitted by the OLT, pass
through a 1:N splitters until reach ONU. ODN splitting ratio
is typically from 4 to 64. In upstream direction data frames
from all ONUs are transmitted through ODN towards OLT.
Communication with another ONU is possible only via OLT
switch.
GEPON supports 1 Gbit/s symmetric data rate. Apart from
GEPON, 10G-EPON [5] offers EPON both symmetric
10Gbit/s downstream and upstream, and asymmetric 10Gbit/s
downstream and 1Gbit/s upstream data rates [6].
The Gigabit Passive Optical Network (GPON)
recommended in [7] also support Gigabit speeds but GPON
Fig. 2. The PON standardization evolutionary.
82
III. GEPON LABORATORY TESTBED AND ITS FUNCTIONALITY
The network topology was implemented in accordance with
the principles of GEPON recommendations for ODN. Its
architecture is presented in Fig. 3. GEPON architecture was
implemented in purpose to familiarize students with GEPON
solutions. The architecture consists of: OLT, five ONU, ODN,
two PC computers, WWW/FTP + Video server (see Fig. 3).
The OLT is based on Raisecom GEPON (ISCOM5504) [8].
The OLT provides a single PON optical interface (1 of 4) for
communicated with ONU_1- ONU_5.
For management purpose the OLT module is additionally
equipped with out-of band management Ethernet port and
console port serial interface (RS-232). All of the commands,
including remote ONU management’s configuration
commands are carried out at the OLT configuration side in
encryption mode.
The ODN is implemented according to optical access
network requirements. The ODN is built with two levels of
splitters and attenuators for modeling fiber optics power loss.
The ONU is represented by Raisecom ISCOM5101 [9],
which is stand-alone ONU device with one EPON interface,
one Ethernet interface and one console interface. All devices
are placed in GEPON architecture in laboratory testbed,
implemented as point-to-multipoint (P2MP) access network
(see Fig. 3). During labs ODN can be easily reconfigured into
point-to-point (P2P) configuration. Due to limited number of
PON interfaces in P2P configuration only four ONU can be
installed (see Fig. 4).
Apart from basic functionality GEPON architecture could
be implemented in cooperation with xDSL (one of FITL
topology) and METRO access network, as it presents Fig. 5.
The GEPON laboratory testbed presented in Fig. 5 enables
students to examine broadband services realization for
various types of FITL access network solutions as Asymetric
Digital Subscriber Line (ADSL) as well as Very high speed
Digital Subsriber Line (VDSL) for the “first mile” distance.
Adittionaly the testbed presented in Fig. 5 can integrate
GEPON network with METRO DWDM network. Such
cooperation requires additional equipment between servers
and GEPON OLT, such as devices for Carrier Ethernet
PWT 2012 - Poznań, 14 grudnia 2012
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
3
T
_1
T
_
T
8
T
_2
8
T
8
T
7
T
6
5
T
T
T
7
T
6
T
5
T
_1
_2
_
_3
_4
_3
4
T
_4
T
/
+
T
M
M
_5
8
/
_ (
+
)
_ (
_1
-C
)
-R
_
8
_2
2
7
-R
6
_
5
_3
_2
/
/
M
+
/
_ (
)
/
+
service demarcation, extension and aggregation (in Fig. 5
depicted as GbE/Opt) with optical links. to optical add/drop
multiplexers and GEPON OLT.
IV. EXERCISES IN LABORATORY GEPON NETWORK
In general student can act in this laboratory as user of
broadband services or as network OAM operator. This section
focuses on presentation exemplary exercises performed by
students during labs.
First of the GEPON laboratory functions is presentation of
optical access for broadband services as video distribution.
There are fixed parameters configured in SLA (for example
FIR, CIR, PIR). For this item main observations are
subjective video quality evaluation and in addition passive
monitoring of downstream transmission parameters.
More advanced version of GEPON functionality testing is
interactive choice of GEPON transmission parameters on the
basis of video service quality with optimization of system
resources usage. Because of need for OLT management
access this possibility is limited to only one student team and
can disturb another active links in GEPON.
The most advanced GEPON configuration case is usage of
VPN and QoS managing with data streams for different CoS.
The application of QoS can deal with certain kind of data
priority or adopts specifically management schedule strategy
to make the network performance predictable and the
management of bandwidth more efficient. QoS of ONU has
three parts to implement: classification, tag and queue
XVI Poznańskie Warsztaty Telekomunikacyjne
83
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
scheduling. The OLT model supports remark functions based
on port ID, MAC address, VLAN, IEEE 802.1 priority,
DiffServ and IP ToS. The students as users have an
opportunity to compare service quality in different network
conditions and topologies limitations.
Another topic for practice is GEPON network Operation,
Administration and Maintenance (OAM). One of main
function is remote fault indication. With maintenance port is
possible to get actual parameters and statistics for OLT and
ONU interfaces (optical and Ethernet) and also data
transmission volume and quality information. Another OAM
function is link loopback test which evaluates the quality and
performance of link.
Apart from functional and performance testing students
have an opportunity for plug and play testing. Manual
configuration from P2MP to P2P is very crucial in aspect of
practical knowledge.
V. PERFORMANCE TESTS RESULTS EXAMPLES
This section presents some examples of laboratory tests
results from OAM terminal in command line interface mode.
These examples present typical answers for EPON global
configuration mode (fttx).
Fig. 6 presents informations for ODN, such as distances
and signal delay and ONU activity.
Raisecom(fttx)#show interface onu 0/5/1-4 online-information
Distance
ONU ID
State
LLID RTT(TQ) (m)
Login Date
-----------------------------------------------------------0/5/1
online
5
54
12
2012-10-01,08:00
0/5/2
online
1
50
6
2012-10-01,08:00
0/5/3
online
3
48
3
2012-10-01,08:00
0/5/4
online
4
54
12
2012-10-01,08:00
Fig. 6 Round trip time and distance information for online ONUs
Fig.7 is an example of service level agreement for uplink,
without priority differentiation.
Raisecom(fttx)#show interface onu 0/5/1 sla
FIR: Fixed information rate
CIR: Commited information rate
PIR: Peak information rate
ONU ID
FIR (Kbps)
CIR(Kbps)
PIR(Kbps) Priority
-----------------------------------------------------------0/5/1
0
64
30720
0
Raisecom(fttx)#show
ONU ID: 0/5/1
InUnicast
:
InMulticasts :
InBroadcasts :
InGoodOctets :
InGoods
:
InErrors
:
InDiscards
:
OutUnicast
:
OutMuliticasts:
OutBroadcasts :
OutGoodOctets :
OutGoods
:
OutErrors
:
OutDiscards
:
Raisecom(fttx)#show interface onu 0/5/1 uni ethernet policing
Port ID: 0/5/1/1
Ingress policing
: disable
Ingress policing CIR: 4836 Kbps
Ingress policing CBS: 4194304 Bytes
Ingress policing EBS: 1514 Bytes
Egress policing
: disable
Egress policing CIR : 4836 Kbps
Egress policing PIR : 64022 Kbps
84
33,994
11,217,440
16
2,221,358,768
11,251,450
0
0
18,676
11,103,726
356
824,298,252
11,122,758
0
0
Fig. 9 Performance statistic information of ONU_1.
Raisecom(fttx)#show interface onu 0/5/1 statistics linkquality
US : upstream
FER : frame error rate
US Errored
US BER
US Errored US FER
ONU ID
Bytes
in 1e-9 Frames
in 1e-9
-------------------------------------------------0/5/1
0
0
0
0
Fig. 10 Uplink quality statistic for ONU_1
VI. CONCLUSION
In the paper GEPON architecture in laboratory testbed is
presented. The possibility of GEPON network testbed
development toward cooperation with xDSL systems and
METRO network is presented. The propositions for students’
labs are presented. Exemplary results are also included.
The presented didactics laboratory GEPON architecture is
a valuable illustration of practical GEPON realization. It is
complementary to the theoretical view on optical access
networks. It enables students to get familiarized with GEPON
conception and its typical applications, it gives an opportunity
to configuration, broadband services testing, and practice
OAM functions.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
Fig. 8 Policy configuration of ONU_1.
PON statistic information query for ONU_1 performance is
presented on Fig. 9. This command presents performance
statistics for receive/send (in/out) single cast, multicast and
broadcast package as well as receive/send (in/out) bytes of
good packets, errors, discards.
interface onu 0/5/1 pon statistic
Fig. 10 is an example result of byte and frame error
monitoring for uplink data transmission.
Fig. 7 Service level agreement configuration parameters for ONU_1
Fig. 8 shows policing configuration in both directions for
ONU_1 Ethernet port from Fig. 3 network topology.
4
[7]
[8]
[9]
ITU-T Recommendation G.982, Optical access networks to support
services up to the ISDN primary rate or equivalent bit rates, November
1996.
ITU-T Recommendation G.983.1, Broadband optical access systems
based on Passive Optical Networks (PON), January 2005
ITU-T Recommendation G.983.3, A broadband optical access system
with increased service capability by wavelength allocation, March 2001
http://standards.ieee.org/findstds/standard/802.3ah-2004.html
K. Tanaka, A. Agata, Y. Horiuchi, IEEE 802.3av 10G-EPON
Standardization and Its Research and Development Status, Journal of
Lightwave Technology, vol. 28, February 2010, pp.651-661.
Z..A. Manaf, K. Khairi,, R. Mohamad, Z . Lambak, D. Adriyanto, D.
Tarsono, .The challenge for active and passive components design in
CWDM PON system co-exist in GEPON and 10 GEPON architecture,
Photonics ICP, Conference Publication IEEE, 17-19 October 201.
ITU-T Recommendation G.984.1, Gigabit-capable passive optical
networks (GPON), General statistics, March 2008.
EPON Configuration Guide, www.raisecom.com
EPON Configuration Guide, Remote ONU Device Management
www.raisecom.com
PWT 2012 - Poznań, 14 grudnia 2012
2012
TANI SPRZET
˛ DLA ASTROFOTOGRAFII
1
Tani sprz˛et dla astrofotografii
Tomasz Kossowski, Ryszard Stasiński
Politechnika Poznańska, Wydział Elektroniki i Telekomunikacji
Streszczenie—Główna˛ idea˛ badań była konstrukcja taniego
sprz˛etu do fotografowania obiektów nieba gł˛ebokiego. W tym
celu przebudowano cyfrowy aparat kompaktowy, dopasowujac
˛
go do teleskopu i pracy z długimi czasami naświetlania. Drugim
działaniem było dodanie do niedrogiego i popularnego teleskopu
w układzie Newtona nap˛edu paralaktycznego. Uzyskane rezultaty
dowodza,˛ że tanie elementy o wystarczajacej
˛
jakości na zrealizowanie idei badań istnieja,˛ ich integracja w urzadzenie
˛
jest
kwestia˛ wiedzy i pomysłowości.
I. W ST EP
˛
O
BIEKTY nieba gł˛ebokiego, takie jak mgławice czy
galaktyki, daja˛ pi˛ekne obrazy. Niestety w wi˛ekszości
przypadków te malownicze pyłowe obłoki sa˛ źródłem światła
odbitego. Jeśli dodamy do tego ogromna˛ odległość, jaka nas
od nich dzieli, to okazuje si˛e, że sa˛ one źródłem bardzo
słabego światła. Do obserwowania i fotografowania mgławic
wykorzystuje si˛e duże teleskopy i silnie chłodzone matryce
CCD, co umożliwia bardzo długie czasy ekspozycji. Taki profesjonalny sprz˛et jest bardzo drogi i niedost˛epny dla każdego.
Stad
˛ narodził si˛e pomysł, by zbudować sprz˛et pozwalajacy
˛
fotografować obiekty nieba gł˛ebokiego każdemu, kto ma
1 tys. zł. na zakup odpowiednich podzespołów. Oczywiście
nie chodzi o uzyskanie rezulatów porównywalnych do osia˛
galnych z zestawów droższych o conajmniej rzad
˛ wielkości,
ale o umożliwienie fotografowania takich obiektów w ogóle.
Na powyższym rysunku widać, że teleskop ten może zbierać
duża˛ ilość światła, które ogniskuje w obiektywie okularu. Dla
zastosowań astrofotografii w miejscu okularu umieszcza si˛e
matryc˛e światłoczuła,˛ która rejestruje padajacy
˛ na nia˛ obraz.
B. Matryca CCD
Drugim elementem użytym w urzadzeniu
˛
jest aparat kompaktowy wyposażony w przetwornik CCD (ang. Charge Coupled Device [4]). Jego zadaniem jest gromadzenie ładunku
proporcjonalnego do ilości padajacego
˛
światła, a nast˛epnie
zamiana tego ładunku na wartości liczbowe, odpowiadajace
˛
jasnościom pikseli obrazu. Każdy kolorowy piksel obrazu
składa si˛e z czterech subpikseli z filtrami o odpowiednim
kolorze. Sa˛ to filtry barw podstawowych (zielony powtórzony jest dwukrotnie), układ ten nosi nazw˛e macierzy Bayera. Każdy punkt na matrycy rejestrujacy
˛ obraz jest studnia˛
potencjału, która gromadzi ładunek na podstawie zjawiska
fotoelektrycznego. Padajace
˛ fotony wybijaja˛ elektrony, które
gromadzone sa˛ pomi˛edzy elektrodami i przechowywane podczas całego czasu ekspozycji. Po zakończeniu naświetlania
nast˛epuje transfer każdej paczki ładunku do komparatorów
i przetworzenie ładunku na napi˛ecie, które dalej podlega konwersji na postać cyfrowa˛ [2]. Schemat działania przedstawiono
na rysunku 2 [6].
II. T EORIA
A. Teleskop
Teleskop w układzie Newtona, jest najbardziej popularna˛
konstrukcja˛ dost˛epna na rynku. Wynika to z prostej budowy,
co przekłada si˛e na niski koszt produkcji i dobry stosunek ceny
do parametrów optycznych. Jest to teleskop zwierciadlany
w odróżnieniu do teleskopów soczewkowych (refraktorów).
Konstrukcja ta cechuje si˛e duża˛ ilościa˛ zbieranego światła oraz
małymi zniekształceniami obrazu. Ide˛e budowy omawianej
konstrukcji oraz bieg promieni przedstawiono na rysunku 1
[4].
Rysunek 2. Schemat konwersji w matrycy CCD [6]
III. M ODYFIKACJE SPRZ ETU
˛
A. Nap˛ed w osi rektascencji
Rysunek 1. Bieg promieni w teleskopie Newtona [4]
Do przeprowadznia badań wybrany został teleskop SkyWatcher Synta 900/130, ponieważ charakteryzuje si˛e najlepszym stosunkiem jakości do ceny w przedziale do 1 tys. zł.
Teleskop ten wyposażony jest w montaż paralaktyczny EQ2, który umożliwia obracanie teleskopem zgodnie z ruchem
XVI Poznańskie Warsztaty Telekomunikacyjne
85
TANI SPRZET
˛ DLA ASTROFOTOGRAFII
2
obrotowym Ziemi [5]. Dzi˛eki temu możliwe jest kompensowanie tego ruchu, jednakże r˛eczne korygowanie położenia
w osi rektascencji jest niemożliwe jeśli chcemy zapewnić
idealna˛ stabilność obrazu. Konieczne zatem okazało si˛e zbudowanie nap˛edu, który obracałby teleskopem przeciwnie do
ruchu nieboskłonu, tak by automatycznie go kompensować
i umożliwić naświetlanie w danym miejscu matrycy dokładnie
tego samego punktu nieba przez cały czas ekspozycji. Brak
takiego układu powoduje rozmywanie si˛e obiektów w kadrze
podczas naświetlania dłuższego niż 1 sekunda i dużym powi˛ekszeniu. Pojedyncze fotony sa˛ zbierane przez te same studnie
potencjału, dzi˛eki czemu zostaje zebrana dostatecznie duża
porcja ładunku umożliwiajaca
˛ zarejestrowanie obrazu [2].
Zbudowany został mechanizm w oparciu o silnik krokowy
Mitsumi M35SP-8 [3] oraz układ kół z˛ebatych zapewniajacych
˛
dopasowanie do montażu teleskopu. Cała konstrukcja wykonana została z odpowiednio wyprofilowanego aluminiowego
płaskownika. Silnik obracajac
˛ si˛e z pr˛edkościa˛ 1 obr/s pozwala
na kompensacj˛e ruchu obrotowego Ziemi z dokładnościa˛ zapewniajac
˛ a˛ niezmienność kadru podczas minutowego naświetlania. Sterowanie zostało zrealizowane przy pomocy programowalnego mikroprocesora ATmega 8 i układu tranzystorów sterujacych.
˛
Dzi˛eki temu możliwa jest dalsza rozbudowa
układu o automatyczne ustawianie teleskopu w osi rektascencji
[2]. Fotografia 3 pokazuje działajace
˛ urzadzenie.
˛
ramy został też przymocowany radiator odprowadzajacy
˛ ciepło
z ogniwa Peltiera, które chłodzi matryc˛e.
Przetwornik został wymontowany z wn˛etrza aparatu
i umieszczony w tulei, która˛ montuje si˛e do teleskopu. Zabieg
ten był konieczny z dwóch powodów: po pierwsze należało
ominać
˛ zintegrowana˛ optyk˛e aparatu, ponieważ wprowadzała
zniekształcenia i powodowała utrat˛e ilości wpadajacego
˛
światła oraz zapewnić chłodzenie przetwornika CCD. Na
fotografii 4 przedstawiono wszystkie omawiane modyfikacje
aparatu.
Fotografia 4. Rama wraz z radiatorem i wentylatorem
Chłodznie zostało zrealizowane przy pomocy ogniwa
Peltiera, które jest przymocowane pomi˛edzy radiatorem
a matryca.˛ Moduł pozwala na obniżenie temperatury do ok.
0°C. Redukcja temperatury jest kluczowa dla długich czasów ekspozycji i małych ilości padajacego
˛
światła. W takiej
sytuacji szum termiczny może być wi˛ekszy niż informacja,
która jest zbierana. Żeby nie dopuścić do takiej sytuacji,
gdzie w obrazie wyst˛epuje jedynie szum, należy ochłodzić
pracujacy
˛ przetwornik do jak najniższej temperatury [2]. Do
odprowadzania ciepła z goracej
˛ strony ogniwa Peltiera służy
stalowy radiator z wentylatorem, a dobre przewodnictwo
cieplne zapewnia silikonowa pasta termoprzewodzaca
˛ [1].
Wywiew ciepłego powietrza skierowany jest dodatkowo na
elektronik˛e aparatu, dzi˛eki czemu możliwa jest praca przy
niskich temperaturach otoczenia.
Na fotografii 5 zaprezentowano chłodzenie przetwornika
CCD.
Fotografia 3. Nap˛ed teleskopu w osi rektascencji
B. Modyfikacja aparatu fotograficznego
Aparatem cyfrowym wybranym do doświadczeń został
Canon PowerShot A530. Jest to mały i tani aparat kompaktowy, który nie ma możliwości montażu zewn˛etrznych elementów optyki takich jak obiektywy czy telekop. W zwiazku
˛
z tym konieczne było wprowadzenie kilku modyfiakcji:
1) Umożliwienie montażu w teleskopie
2) Wyprowadzenie matrycy na zwenatrz
˛
3) Zapewnienie chłodzenia przetwornika CCD
4) Zmiana oprogramowania
Pierwsza˛ zmiana˛ było zbudowanie aluminiowo-stalowej
ramy, która umożliwiłaby zainstalowanie urzadzenia
˛
w miejscu
standardowego okularu teleskopu. Konstrukcja ta musiała
być solidna oraz zapewniać szybki montaż i demnotaż. Do
86
Fotografia 5. Ogniwo Peltiera i matryca CCD
Ostatnia˛ zmiana˛ w aparacie było wgranie nowego opro-
PWT 2012 - Poznań, 14 grudnia 2012
TANI SPRZET
˛ DLA ASTROFOTOGRAFII
3
gramowania. Standardowe umożliwiało ustwienie czasu ekspozycji na maksymalnie 15 sekund, nie pozwalało na
wyłaczenie
˛
redukcji szumu i zapisywanie zdj˛eć w innym formacie niż JPEG. Nowe oprogramowanie odblokowało dost˛ep
do wszytskich możliwości wbudowanego procesora obrazu,
dzi˛eki czemu możliwe stało si˛e ustawianie ekspozycji nawet
na pół godziny, wyłaczenie
˛
redukcji szumów oraz zapis zdj˛eć
w formacie RAW, dzi˛eki czemu można uzyskać zdj˛ecia bez
jakiejkolwiek kompresji [7]. Jest do bardzo ważne, gdyż
kompresja JPEG może usuwać ważne informacje.
Ciekawa jest tu ilustracja działania chłodzenia przetwornika
CCD. Na fotografiach 8.1 i 8.2. zaprezentowano szum przy
wyłaczonym
˛
i właczonym
˛
chłodzeniu.
IV. W YNIKI
Zaprezentowane modyfikacje pozwoliły na zwi˛ekszenie
możliwości elementów omawianego sprz˛etu optycznego. Dzi˛eki temu udało si˛e przeprowadzić obserwacje astronomiczne
i zebrać materiały do innych badań, których celem jest
poprawa jakości obrazów. Przedstawione poniżej fotografie
zostały wykonane przy nienajlepszych warunkach atmosferycznych pod dużym miastem. Miało to duży i negatywny
wpływ na jakość zdj˛eć, co dodatkowo świadczy o możliwościach samego sprz˛etu. Na fotografiach 6 i 7 zaprezentowano
przykładowe zdj˛ecia przy użyciu całego urzadzenia.
˛
Fotografia 8.1. i 8.2. szum w temperaturze pokojowej (po lewej) oraz
działajacym
˛
chłodzeniem (po prawej)
Z powyższego przykładu wyraźnie widać jak duże znaczenie ma redukcja szumu termicznego. Jest to decydujacy
˛ czynnik szumowy w fotografiach z długim czasem ekspozycji.
Jego ilość jest zależna zarówno od temperatury jak i czasu
gromadzenia ładunku. Przy konieczności długiego naświetlania jedyna˛ możliwościa˛ ograniczenia jego ilości jest obniżenie temeratury przetwornika. Brak chłodzenia matrycy może
powodować, że szum przewyższa informacj˛e na zdj˛eciach i nie
jest możliwe wyodr˛ebnienie danych, czyli nie można uzyskać
obrazu obiektu, który emituje bardzo mało światła.
V. W NIOSKI
Fotografia 6. M31 - Wielka Mgławica Andromedy (jadro)
˛
Przedstawione w artykule wyniki pokazały, że można
skutecznie, przy niewielkim nakładzie kosztów zmodyfikować
tani sprz˛et optyczny by zwi˛ekszyć jego możliwości. Nap˛ed
w osi rektascencji umożliwia długie naświetlanie jednego
obszaru nieba bez efektu rozmywania si˛e obiektów w kadrze.
Modyfikacja aparatu sprawiła, że można go szybko i łatwo
zamontować w wyciagu
˛ teleskopu. Chłodzenie matrycy CCD
pozwala w znacznym stopniu zredukować szum termiczny
powstajacy
˛ przy długim czasie ekspozycji, a nowe oprogramowanie umożliwiło ustawianie takich czasów naświetlania. Wszystkie te modyfikacje przyczyniły si˛e do finalnego efektu jakim jest możliwość fotografowania galaktyk
i mgławic. Łaczny
˛
koszt omawianego sprz˛etu nie przekroczył
1 tys. zł, co obala mit, że tanim sprz˛etem można fotografować
jedynie tarcz˛e Ksi˛eżyca.
B IBLIOGRAFIA
Fotografia 7. M42 - Wielka Mgławica Oriona (jadro)
˛
[1] Piotr Górecki, artykuł: “Radiatory w sprz˛ecie elektronicznym”, Elektronika dla Wszystkich, Grudzień 1999 r.
[2] Tomasz Kossowski, “Poprawianie jakości obrazu przy
słabych warunkach naświetlenia”, Politechnika Poznańska,
Poznań 2012 r.
[3] Mitsumi, M35SP-8, dokumentacja techniczna producenta
[4] http://pl.wikipedia.org (czerwiec 2012)
[5] http://deltaoptical.pl/blizej_nieba/teleskop-sky-watchersynta-sk-1309- eq2,d697.html (marzec 2012)
[6] http://www.teledynedalsa.com/corp/markets/ccd_vs_cmos.aspx
(kwiecień 2012)
[7] http://chdk.wikia.com/wiki/CHDK (czerwiec 2012)
XVI Poznańskie Warsztaty Telekomunikacyjne
87
2012
A software environment for multicast routing testing
Paweł Krzyżaniak, Marek Michalski, Member, IEEE
Poznan University of Technology, Chair of Communication and Computer Networks,
ul. Polanka 3, 60-965 Poznań, Poland, e-mail: (pawel.krzyzaniak, marek.michalski)@et.put.poznan.pl
Abstract—Presented paper introduces a software environment
facilitating routing protocols testing (with emphasis on multicast
routing protocols). The main purpose of the program is to
increase the efficiency of various network topologies analysis
by means of configuration process automation. The program
was designed to be used in conjunction with Juniper MX-series
routers, although its universal nature allows it to cooperate with
other routing platforms running Junos OS as well.
I. I NTRODUCTION
Nowadays world becomes virtual. The virtualization technology is very popular in many areas, the word virtualization
can have different meanings. Basically, it is logical division of
physical resources and realization of independent access for
different users. We will focus on virtualization in computer
networks. Even here we can have different approaches to
virtualization. The host machines can be virtualized [1], [2],
the processors can be divided into virtual processors [3],
networks segments can be virtualized as a Virtual Local Area
Networks (VLANs) on layer 2 [4] or in layer 3 as a Virtual
Private Networks (VPNs) [5] which can be realized in many
different ways. Also the network equipment can be virtualized,
huge hardware machines can be treated as a set of several
smaller ones [6]. In this paper we will focus on mechanism
of logical systems, which is developed by Juniper Networks
Company [7]. It allows to divide huge router into smaller ones,
but still fully functional. With them we will analyze multicast
protocols and another network protocols and mechanisms.
II. G ENERAL D ESCRIPTION
The program was initially devised as a part of master
thesis and was subsequently custom-tailored to serve as an
effective tool for teaching the basics of multicast routing
mechanisms in computer networks laboratory. All tests were
performed on Juniper MX240 Ethernet Services Router [8]
shown in Fig. 1. The software makes use of logical systems
virtualization technology provided by Juniper Networks Junos
OS [7]. This technology allows the user to create up to
15 logical systems (restriction imposed by Junos OS) on a
single hardware platform. It has to be noted however that
logical system concept is not synonymous with the widely
recognized notion of virtual router. Although similar in its core
assumptions, logical system is a much more versatile tool. The
most significant difference between these two concepts is that
for every logical system created, the routing engine starts a
new routing protocol process [9]. It means that every logical
system is a fully separate entity that works independently.
Should one of the logical systems fail, the routing processes
88
Fig. 1.
Juniper MX240 Ethernet Services Router
of other logical systems will not be affected, thus ensuring a
more stable functioning of the whole network. Complete list
of features supported by logical systems technology can be
found in [9].
Presented software environment has a form of a functional
and easy to use graphical interface (Fig. 2) [10]. Main window
is divided into four sections. The largest portion of the window
is occupied by a panel containing current topology. The user
can create a logical system by double clicking left mouse
button anywhere in the panel. By double clicking right mouse
button the user can create a host (representing an arbitrary
physical device). Every object in the panel is fully interactive
- it can be moved around the panel, has a related context
menu as well as a list of modifiable parameters displayed in
a property grid located under the treeview. The treeview itself
enables a quick access to every logical system and its interfaces. The bottom section contains a window displaying all
commands that are sent to the router. The user can customize
the appearance of the main window as well as numerous
functional capabilities of the program to suit his own needs,
e.g. resize all sections composing the main window, switch
between different modes of creating and sending commands
(instantaneous and explicit), modify the colors of elements in
network topology, show/hide selected labels (like IP addresses
and subnet masks on interfaces) to improve the clarity of
examined topology, etc. One of the most important aspects of
the program is that the knowledge of Junos OS syntax is not
required. The configuration process is done entirely through
the graphical interface. The program translates the choices
made by user into Junos OS commands and sends them to
the router. The following section contains a more detailed
description of program’s features.
PWT 2012 - Poznań, 14 grudnia 2012
III. S ELECTED F EATURES
As stated at the beginning of this paper, the main purpose of
the program is to relieve the user from the necessity of manually entering configuration commands on every single router.
The program delivers many conveniences providing users
with maximum comfort. One of the most important features
realizing this approach is automatic IP address generation. The
software uses a predefined pool to assign IP addresses from the
same subnet to interfaces that are being connected. This feature
provides the ability to create a complex topology consisting of
many interconnected logical systems within minutes (or even
seconds). The logical systems (as well as other objects) can
be easily removed from current topology (and from physical
router’s configuration) by choosing appropriate option from
their context menu. Similar rule applies to links and interfaces
- right-clicking on a connection between two objects (routers,
hosts or external network) or on a particular interface in
treeview brings up a context menu allowing to disconnect
it. The interfaces themselves are automatically detected by
software. When a new network is created, user is prompted to
enter all information required to establish a connection with
physical routers. Confirmation of all entered information (by
clicking appropriate button) starts a new thread (a separate
thread is run for every router the application is attempting to
connect to). This thread first establishes a connection with
router. Then it performs all necessary operations to create
a list of available interfaces. All interfaces that are either
logical in nature or not configurable are filtered in this process.
If the router is equipped with hardware (a specialized PIC
(Physical Interface Card) [9]) enabling the use of logical
tunnel interfaces, the user can utilize them to interconnect
logical systems. If logical tunnel interfaces are not available
on router, the user can still use physical interfaces to connect
objects in network topology. Moreover, the user can define
the way these interfaces are connected in a separate window.
This information is then used to automatically select proper
interfaces when two logical systems are being connected. Of
course, the user can switch between logical tunnel interfaces
(if they are available) and physical interfaces at any time.
The software uses a set of Visual Basic scripts to start
a new process of PuTTy [11] program and establish an
individual connection session with a particular logical system
or host (if user defined IP address of that host). These scripts
automatically fill in all necessary information to properly login
to any logical system. An example use of this feature is
shown in Fig. 3. Right-clicking on a logical system brings up
its context menu. "Console" option executes a parameterized
script starting a new PuTTy process. After successful login
the script issues a command switching the console view to
the selected logical system. As seen in Fig. 3, logical systems
emulate a real router to a great extent. User can work in both
operational and configuration mode and issue commands that
are characteristic to these modes. Although most important
and frequently used commands are accessible via the context
menu, PuTTy terminal is very handy for issuing commands
Fig. 2.
The main window with example topology
that require constant refreshing (e.g. ping command).
Another useful feature of the software is the ability to switch
between two modes of creating and sending router commands:
instantaneous and explicit. In first mode, the commands are
created and sent right after any change in configuration-related
parameter is detected, which applies to almost every parameter
that does not refer to object’s appearance (like its color or
position). In second mode, it is up to user to decide when
the commands should be created and sent. The appropriate
buttons are available in configuration windows and context
menus of logical systems and interfaces. The commands can
also be sent in a cumulative way by choosing a proper option
from the context menu of a physical router. It results in
creating commands for every modified logical system (and
its interfaces) that exist within that physical router. Modes
can be changed at any time - user can choose the one that
better suits his needs in a given situation. Presented software
environment focuses on multicast routing protocols. Users can
configure one of many available protocols - PIM-DM [12],
PIM-SM [13] (along with various Rendezvous Point election
mechanisms), PIM-SSM [14], MSDP [15] and DVMRP [16]
(which is considered to be an obsolete protocol, but was
included for comparison purposes). Since PIM does not build
its own routing table - it relies on other sources of routing
information (like unicast routing protocols, but also static
routes) - the program enables the user to configure several
dynamic unicast routing protocols (OSPF, RIP and BGP [17])
as well as static routes. Both OSPF and BGP use the notion
of areas. All the routers (or interfaces) within an area share
common configuration characteristics. It would be then highly
impractical and time consuming to manually configure all the
XVI Poznańskie Warsztaty Telekomunikacyjne
89
Fig. 3. The context menu and terminal window with logical router LR1
connected
routers within a given area (even through graphic interface).
For this reason, the objects representing these areas (seen on
Fig. 2) were developed. By choosing appropriate option from
toolbar, the user can draw an OSPF or AS area of an arbitrary
size. As every object in topology, areas are described by a
set of modifiable parameters. By placing a router (or link)
within range of a particular area, the software automatically
performs proper operations (like setting AS area number,
configuring default metric of OSPF stub area, etc.). This
feature is especially useful when configuring multi-area OSPF.
Every router forwarding multicast packets is a part of multicast
tree (whether it is a shortest path tree or a shared tree). These
trees are composed of links and routers that are a part of
multicast data transmission path. Learning the path traversed
by multicast packets requires tracing the route from last-hop
router back to the source network by issuing proper command.
This method however does not scale well when multiple end
stations are connected to different routers. Described software
environment performs all necessary actions to learn multicast
routes automatically and pictures the results in topology section of the main window (Fig. 4 ). The data transmission path
is marked by a thick green line. Host icons (Fig. 4), in this
case, are representing Linux computers acting as multicast
transmitters and receivers. As with logical systems, hosts can
be accessed via their context menu. Both transmitters and
receivers use an open source software called MGEN (MultiGenerator) [18] to generate and request multicast data stream
for a particular group address, respectively. MGEN can, among
its numerous applications, generate packets (both TCP and
UDP) according to specific traffic pattern (periodic, burst,
etc.) destined for a particular IP address and define the output
interface used to transmit this data stream. Similar parameters
can be used to set the end station in listening mode and store
information about received packets in log files for further
analysis. The software allows the user to save and then load a
network topology from a file. This file stores all information
needed to recreate the exact topology that was previously
saved. The ability to save and load can be used to create
90
Fig. 4.
The multicast tree in the example topology
various lab scenarios. The same topology can be loaded and
viewed on multiple workstations thus allowing many users to
access logical systems simultaneously via graphical interface.
When user is finished working with application and wants to
erase all configuration changes, he can do this by clicking
"Clear network" icon on toolbar. The software issues a set
of commands deleting all logical systems that were created
during current session. Network is also automatically cleared
when user decides to load topology from file, ensuring that
existing configuration won’t be merged with the loaded one.
IV. F URTHER DEVELOPMENT
The efforts are currently focused on providing the application with mechanisms supporting interoperability between
multiple lab groups working on the same physical router. The
external networks are currently depicted by a cloud symbol.
The intention is to provide all lab groups with the full picture
of network, that is - to provide all groups with complete
information about networks managed by other groups in a
graphical way. Because there is no possibility to exchange data
directly between computers in lab environment, all required
information regarding logical system topology has to be gathered from physical router itself. Another interesting and unique
feature supported by Junos is a virtual switch [19]. This technology is available only on MX-series routers. Incorporation
of this technology into software environment would expand its
capabilities to support data link layer specific functions, which
could considerably improve software’s overall usefulness.
PWT 2012 - Poznań, 14 grudnia 2012
V. S UMMARY
In this paper, a software environment for multicast routing
mechanisms testing was presented. Its purpose is to provide a
tool that will enable the user to effectively test various network
topologies and learn the basics of multicast routing without the
knowledge of Junos OS syntax. The paper described numerous
features of presented software and its proposed applications.
R EFERENCES
[1] Virtualization
Overview,
VMware
Inc.
available
at:
http://www.vmware.com/pdf/virtualization.pdf.
[2] Quick Migration with Hyper-V, White Paper, Microsoft Company, January 2008.
[3] White Paper: Intel Virtualization Technology FlexMigration Enablement,
available at: http://communities.intel.com/docs/DOC-5062.
[4] IEEE Std. 802.1Q-2005, Virtual Bridged Local Area Networks. ISBN
0-7381-3662-X.
[5] OpenVPN - The Open Source VPN. available at: openvpn.net.
[6] Cisco Nexus 1000V Series Switches Data Sheet, available at:
http://www.cisco.com/en/US/prod/collateral/switches/ps9441/ps9902/data
_sheet_c78-492971.html.
[7] Junos
Network
Operating
System,
available
at:
http://www.juniper.net/us/en/products-services/nos/junos/.
[8] D. R. Hanks Jr., H. Reynolds, Juniper MX Series, A Comprehensive Guide
to Trio Technologies on the MX, O’Reilly Media, 2012.
[9] OS Logical Systems Configuration Guide, Juniper Networks, Inc., 2012.
[10] Michalski M.: A software and hardware system for a fully functional
remote access to laboratory networks. The Fifth International Conference
on Networking and Services, Valencia, Spain April 20-25, 2009 str. 561565.
[11] PuTTy official website, available at: http://www.putty.org/.
[12] Protocol Independent Multicast - Dense Mode (PIM-DM):
Protocol Specification (Revised), RFC 3973, available at:
http://tools.ietf.org/html/rfc3973.
[13] Protocol Independent Multicast - Sparse Mode (PIM-SM):
Protocol Specification (Revised), RFC 4601 available at:
http://tools.ietf.org/html/rfc4601.
[14] An Overview of Source-Specific Multicast (SSM), RFC 3569, July 2003,
available at: http://tools.ietf.org/html/rfc3569.
[15] Multicast Source Discovery Protocol (MSDP), RFC 3618, October 2003,
available at: http://tools.ietf.org/html/rfc3618.
[16] Distance Vector Multicast Routing Protocol, RFC 1075, November 1988,
available at: http://tools.ietf.org/html/rfc1075.
[17] Andrew. S. Tannenbaum, Computer Networks, Prentice-Hall Inc., third
edition, 1996.
[18] Multi-Generator, avail. at: http://cs.itd.nrl.navy.mil/work/mgen/index.php.
[19] Junos OS MX Series Ethernet Services Routers Solutions Guide, Juniper
Networks, Inc., Rel. 11.1, 2011.
XVI Poznańskie Warsztaty Telekomunikacyjne
91
2012
1
Aplet JAVA do demonstracji
modulacji cyfrowych
Maciej Krasicki, Bartosz Kaczmarek
Wydział Elektroniki i Telekomunikacji
Politechnika Poznańska
ul. Polanka 3, 61-131 Poznań
e-mail: [email protected]
Streszczenie—W referacie przypomniano podstawowe pojęcia i
przekształcenia, prowadzące do opisu modulacji nośnych
sinusoidalnych w równoważnym pasmie podstawowym.
astępnie przedstawiono aplet do demonstracji modulacji
cyfrowych, wykorzystywany w trakcie zajęć dydaktycznych w
zakresie systemów transmisji cyfrowej.
Słowa kluczowe—modulacja cyfrowa, konstelacja sygnału,
obwiednia zespolona
W
II. MODULACJA CYFROWA NOŚNEJ SINUSOIDALNEJ
A. Modulacje fazy i amplitudy
W ogólności sygnał pasmowy z modulacją fazy i/lub
amplitudy można zapisać w postaci
(1)
gdzie ௖ jest częstotliwością nośną sygnału, – jego
amplitudą, a – fazą. Dwie ostatnie z wymienionych
wartości mogą zależeć od aktualnie nadawanego elementu
ciągu danych.
Korzystając z własności trygonometrycznych, (1) można
przepisać w następującej formie:
ൌ …‘•
…‘•ʹ௖ െ
െ •‹
•‹ʹ௖ ,
I. WSTĘP
YKŁAD i ćwiczenia audytoryjne w zakresie systemów
transmisji cyfrowej odbywają się na Wydziale
Elektroniki i Telekomunikacji Politechniki Poznańskiej na
piątym semestrze studiów I stopnia.
Ponieważ temat modulacji cyfrowych w paśmie nośnym
nastręczał wielu trudności, prowadzący ćwiczenia – pierwszy
współautor niniejszego referatu – stworzył w środowisku
MATLAB niewielki program przedstawiający kolejne etapy
powstawania pasmowego sygnału cyfrowego. Program ów
sterował graficznym interfejsem użytkownika (ang. Graphical
User Interface, GUI), który pokazano na rys. 1. Przebiegi
prezentowane były w trakcie zajęć z odpowiednim
komentarzem prowadzącego. Studenci wielokrotnie prosili o
udostępnienie podobnego programu, z którego mogliby
korzystać w domu, przygotowując się do zajęć. Niestety,
MATLAB nie jest środowiskiem darmowym, a cena jego
zakupu przekracza możliwości finansowe polskiego studenta.
Z drugiej strony, nie ma możliwości łatwego utworzenia w
pakiecie MATLAB aplikacji typu stand-alone, czyli
niewymagającej środowiska uruchomieniowego. Dlatego
powstała koncepcja napisania podobnego symulatora w
darmowym środowisku programowania, a jej stworzenie
powierzono drugiemu z współautorów niniejszego referatu w
ramach pracy inżynierskiej [1].
92
ൌ …‘•ʹ௖ ൅ ,
(2)
przy czym …‘•
ൌ ூ nazywa się składową
synfazową sygnału, a •‹
ൌ ொ – składową
kwadraturową. Korzystając z powyższych oznaczeń,
wygodnie jest posługiwać się pojęciem równoważnego
sygnału w paśmie podstawowym lub, inaczej, obwiednią
zespoloną sygnału:
ൌ ூ ൅ ொ .
(3)
Związek pomiędzy obwiednią zespoloną, a rzeczywistym
sygnałem pasmowym jest następujący:
ൌ ࣬ሼା ሽ,
(4)
ା ൌ ௝ଶ஠௙೎ ௧
(5)
przy czym
jest tzw. sygnałem analitycznym. Dzięki zastosowanym
zależnościom, kolejne operacje powstawania sygnału
pasmowego mogą być reprezentowane w równoważnym
pasmie podstawowym.
Składowe ூ i ொ są kształtowane, odpowiednio, w
torze synfazowym i kwadraturowym nadajnika, jak pokazano
na rys. 2. Wektor binarny a jest odwzorowywany w wektor
zespolonych symboli danych d o elementach
௜ ൌ ௜ூ ൅ ௜ொ .
(6)
Składowe (rzeczywista i urojona) symboli danych
reprezentują łącznie tzw. punkt konstelacji, który – przy
założeniu konkretnej postaci filtrów kształtujących impuls w
pasmie
podstawowym
–
jednoznacznie
wyraża
rozpoczynający się w i-tym odstępie modulacji składnik
sygnału pasmowego. Należy zaznaczyć, że w ogólności w
PWT 2012 - Poznań, 14 grudnia 2012
2
Rys. 1
Graficzny interfejs użytkownika programu do demonstracji modulacji
cyfrowych w środowisku MATLAB
Rys. 2. Model obrazujący tworzenie sygnału pasmowego w nadajniku
każdym torze nadajnika mogą być stosowane filtry o różnych
odpowiedziach impulsowych. W omawianym przypadku
założono, że obie odpowiedzi są jednakowe i wynoszą .
Źródłem problemów interpretacyjnych dla studentów jest
stałe położenie punktów konstelacji na płaszczyźnie
zespolonej, jak na rys. 3, stanowiących ilustrację procesu
modulacji pasmowej w podręcznikach. Czytelnik może
przeoczyć w ten sposób fakt, że nośne kosinusoidalna i
sinusoidalna sygnału pasmowego są w istocie modulowane
przez składowe synfazową i kwadraturową obwiedni
zespolonej, która jest funkcją ciągłą. Między innymi ten
problem miał być wyjaśniony w tworzonym programie.
B. Modulacja częstotliwości z ciągłą fazą
W trakcie ćwiczeń ze studentami, poza liniowymi
modulacjami amplitudy i fazy, dużo uwagi poświęca się
modulacji częstotliwości o stałej obwiedni i z ciągłą fazą
(Continuous Phase Modulation, CPM), a to ze względu na jej
znaczenie praktyczne we współczesnych systemach
telekomunikacji bezprzewodowej.
Sygnał CPM może być reprezentowany za pomocą
obwiedni zespolonej (3), przy czym tym razem
ூ cos ,
ொ sin .
(7)
(8)
Rys. 3. Punkty konstelacji 16-QAM; pokazano
przykładowe odwzorowanie bitów ciągu kodowego w
elementy sygnału
XVI Poznańskie Warsztaty Telekomunikacyjne
93
3
Skoro jest stałą, oczywistym następstwem jest stała moc
sygnału pasmowego, bowiem niezmienny jest również
kwadrat modułu obwiedni zespolonej:
||ଶ ଶ cos ଶ sinଶ ଶ .
(9)
Tymczasem chwilowa wartość fazy sygnału nośnego dana
jest następującym wzorem:
௧
2 ∑௡௜ୀିஶ ௜ ିஶ d,
(10)
w którym 2∆ jest tzw. indeksem modulacji cyfrowej,
∆ to maksymalne odstrojenie częstotliwości sygnału od
częstotliwości nośnej, ௜ reprezentuje kolejne (rzeczywiste,
najczęściej bipolarne) symbole danych, a jest tzw.
impulsem częstotliwościowym. Jak można stwierdzić
analizując (10), to impuls częstotliwościowy decyduje o
szybkości i sposobie zmiany fazy rzeczywistego sygnału
pasmowego. Modulację częstotliwości na płaszczyźnie
zespolonej można interpretować jako ruch punktu,
stanowiącego aktualną wartość obwiedni zespolonej, po
okręgu wokół początku układu współrzędnych. Jeżeli ruch
tego punktu następuje w kierunku zgodnym z kierunkiem
wirowania zespolonej sinusoidy, składającej się na sygnał
analityczny (5), można powiedzieć, że w ten sposób
zwiększana jest efektywna (wypadkowa) częstotliwość
sygnału pasmowego. Jeżeli punkt reprezentujący wartość
obwiedni zespolonej wiruje przeciwnie niż zespolona
sinusoida, efektywnie zmniejsza on częstotliwość sygnału
pasmowego. Nadajnik CPM, w szczególnych przypadkach,
może być zrealizowany w postaci liniowej, podobnie jak dla
modulacji fazy i częstotliwości (por. rys. 4.38 w [2]).
III. WŁASNOŚCI APLETU DEMONSTRACYJNEGO
Założeniem przy realizacji programu była jego bezpłatna
dostępność dla możliwie szerokiego kręgu odbiorców,
niezależnie od używanego systemu operacyjnego. Możliwości
takie daje programowanie w języku Java. (Aplet Java może
być uruchomiony z poziomu dowolnej przeglądarki
internetowej, dzięki użyciu wirtualnej maszyny Java.) Aby
zadośćuczynić wymienionym założeniom, program został
udostępniony w Internecie [3].
W oknie apletu, pokazanym na rys. 4 widoczne są trzy
wykresy, przedstawiająco kolejno: przebieg obwiedni
zespolonej, sygnału analitycznego oraz rzeczywistego sygnału
pasmowego. Wyświetlane są także fragmenty ciągu danych a,
odwzorowywane w kolejne elementy sygnału. Użytkownik
otrzymuje możliwość wyboru rodzaju modulacji spośród
BPSK, QPSK, 16-QAM oraz CPM (CPFSK) oraz kształtu
impulsu (w przypadku modulacji CPM rzecz dotyczy
wyboru impulsu częstotliwościowego).
Aplet pokazuje, że punkt reprezentujący obwiednię
zespoloną może poruszać się na płaszczyźnie zespolonej, a
także, że wypadkowy sygnał pasmowy może być wypadkową
dwu kolejnych sygnałów w przypadku, gdy odpowiedź
impulsowa filtru kształtującego impuls w pasmie
podstawowym przekracza jeden odstęp modulacji.
Symulacja powstawania sygnału pasmowego przebiega
automatycznie po kliknięciu przycisku „Start”. Godna uwagi
jest możliwość powrotu do dowolnego momentu symulacji po
zakończeniu jej automatycznego przebiegu (wystarczy w tym
celu jedynie przesunąć suwak umieszczony nad wykresem
sygnału pasmowego).
Rys. 4. Okno apletu w trakcie demonstracji przebiegu sygnału 16-QAM
z impulsem podstawowym o kształcie podniesionego kosinusa
94
PWT 2012 - Poznań, 14 grudnia 2012
4
Rys. 5. Demonstracja kształtowania sygnału CPM
Wyświetlanie na bieżąco przebiegu obwiedni zespolonej dla
sygnału CPM (rys. 5) jest bodaj najbardziej efektownym
wynikiem uruchomienia apletu. Użytkownik może bowiem
przekonać się o stałości obwiedni rzeczywistego sygnału
pasmowego, a także stwierdzić, że obrót punktu
reprezentującego chwilową wartość obwiedni zespolonej
zmniejsza lub zwiększa częstotliwość sygnału nośnego. Na
podstawie otrzymanego przebiegu można łatwo wytłumaczyć
sens kratownic fazowych dla sygnałów CPM.
Nietypowym, a posiadającym duże walory dydaktyczne,
rozwiązaniem jest wyświetlenie, jak na rys. 6, pełnej
trajektorii obwiedni zespolonej, uwzględniającej wszystkie
możliwe przejścia na konstelacji. Uzyskanie podobnego
obrazu (por. rys. 4.23 w [2]) w rzeczywistych warunkach
mogłoby wymagać bardzo długiego czasu obserwacji.
Rys. 6 Kompletna trajektoria obwiedni zespolonej dla
sygnału 16-QAM przy zastosowaniu impulsu w kształcie
podniesionego kosinusa
IV. PODSUMOWANIE
W połączeniu z innymi materiałami informacyjnymi
dostarczanymi studentom, opisany aplet stanowi nowoczesną i
skuteczną metodę nauczania w dziedzinie podstaw
telekomunikacji w trakcie zajęć na uczelni. Studenci mogą
korzystać z niego także w domu, celem powtórzenia materiału
i przygotowania się do kolokwium.
W przyszłości strona internetowa [3], na której
umieszczony jest aplet, mogłaby zostać wzbogacona o
stosowny opis modulacji cyfrowych i wówczas wykorzystana
do nauczania na odległość.
LITERATURA
[1]
[2]
[3]
B. Kaczmarek, Program do demonstracji właściwości modulacji
cyfrowych, Praca inżynierska pod kier. dr. inż. M. Krasickiego, Wydział
Elektroniki i Telekomunikacji Politechniki Poznańskiej, styczeń 2012.
K. Wesołowski, Podstawy cyfrowych systemów telekomunikacyjnych,
WKiŁ, Warszawa, 2003.
cygnus.et.put.poznan.pl/~mkrasic/Modulations/launch.html
XVI Poznańskie Warsztaty Telekomunikacyjne
95
96
PWT 2012 - Poznań, 14 grudnia 2012