Professional Documents
Culture Documents
b = --1 ( p l
2
+ l2) a 2 = - $ (pl +9 3 034)
1.0-
-
n
~ = -1
( p 1 3 + $ p 1 2 + $ p ~ 15
z
0 . 24
U
n becomes degradation the Therefore,
2
e -
0.5.
D = - l o log [ I - + (pl + l ) e-z + L ( p l +
2 8
+)3F2
-
- A (pl + p12 + $pl + ~ 1 ~ 4 1 . 05)
Using the series expansion for the logarithm,we get
-
D = 10 (loge)[+(pl + +I-.’
- $ ( p l + +>2(pl -;)F2
5 IO
-
Fig. 6 .
SYMBOL ASYMMETRY
Degradationcausedbysymbolasymmetryfor
(%)
PE =
4 ( p 1 3 + $p12
+ 21 +$pl + ?lE41 (B6)
and p = 1/2.
to fourth order in E. This has to be expressed in terms of po
rather than p l . From (3) and(B5),
APPENDIX A
-
Expand p1 = po[ 1 + l2( p o + +)e2 + higherorderterms].(B7)
f(e)= 4 erfc1-4 =-
J;; /-J z c z e-x2 dx Substitution of (B7)into
small quantities.
(B6) yields(12)tofourthorderin
aylor in a E = 0: REFERENCES
f(E)=f(0)+fdE
df / L O
+p?l de2 E =
+...
~
(A21
[ 11 E. Y. Ho and Y. S. Yeh, “A new approach for evaluating the error
probability in the presence of intersymbol interference and additive
Gaussian noise”. Bell Syst. Tech. J.. vol. 49. p. 2249, 1970.
and use Liebnitz’s rule for differentiating an integral: [2] 0. ShimboandM.Celebiler.“Theprobability of error due to
intersymbol interference and Gaussian noise in digital communica-
b(x) db systems.”
tion lEEECommun.
Trans.
Technol.. COM-19.
vol. p.
y ( x ,z ) dz =
Ia(x)
- dz
dx
+ .v(x, b ( x ) ) -
dx
113,1971.
[3] J . Spilker, Diyiral Communicarion b.v Satellire. Englewood
Cliffs, NJ: Prentice-Hall, 1977.
141 M . Abramowitzand I . A . S t e p ” , Hondbook of.Mathemutical
Functions with Formulas. Graphs. and Tables. Washington.DC:
Nat. Bur. Stand., 1964, p. 932, no. 76.2.17.
[51 M . K. Simon, K. Tu. and B.H.Batson.“Effects of dataasym-
toobtain ( 5 ) . Thisresultisequivalent tothemoregeneral metry on shuttleKu-bandcommunications link performance.”
expansion of f ( e ) in terms of the Hermite polynomials in p1 IEEE Trans. Commun.. vol. COM-26, pp. 1639-1651. Nov. 1978.
) a well-behaved function for E < 1, its Taylor
’[31 . Since f ( ~ is
series expansion converges absolutely.
APPENDIXB
To derive (1 2), average (5) term by term, retaining only the
even powers of E :
.-
High-Rate Punctured Convolutional Codes for Soft Decision
Viterbi Decoding
PE = 4 erfc 4 p 1 ( 1 - a -
e2
-
- be22
-
- ce4) (B2)
Absrruct-The high-rate punctured codes of rates 2/3 through 13/14
are derived from rate 1/2 specific convolutional codes with maximal
where a , b , and c are constants to be determined. The quantity free distance. Coding gains of derived codes are compared based on
inparenthesesisthemostgeneralfourth-orderexpression, their bit error rate
performances under
soft decision Viterbi
involving only the even-order moments of E, that can be writ- decoding.
ten for the factor multiplying p l . Using ( 5 ) again to expand
(B2), we obtain I. INTRODUCTION
It is known that Viterbi decoding for high-rate codes can
besignificantlysimplifiedbyemploying“punctured”con-
)r--
Cain et al. [ 2 ] clarifiedthecodestructureanddecoding for Origlnol code data
procedure of the punctured code, and derived the best punc-
tured‘codesofrates2/3and3/4fromamong all possible (orlgmal
combinations of code generators. Coded Data b1.z bz.2 ---- b1.z b,+I,z - - -
Punctured coding technique is very, attractive not only for
simplifying the Viterbi decoder for high-rate codes but also
forimplementing rateselectable
a convolutionalencoder/
Viterbidecoder.Forthelatterpurpose,it is essential t o ( 2 x L )I * [ transmlrtlng
0I :: deletlng
employ various rate punctured codes derived from the same u ( m blts 1
)I
originalcode.Inthiscorrespondence,high-ratepunctured Punctured i3.1 X - - - - b1,1 b1tl.1 - - -
codesofrate ( n - l ) / n ( n = 3 , 4 , *-, 14)arederivedfrom coded Data bt,z bz,z - - - - X b1tt.z - - -
rate
1/2
specific
convolutional codes with
maximal free
’ X ---deleted bits I
distance[4J,andthecodinggainsarecomparedbasedon
theirbiterrorrateperformanceundersoftdecisionViterbi Fig. 1. Basic procedure of punctured codingfrom rate 1 / 2 convolu-
decoding. tional code.
TABLE I
M A P O F DELETING BITS FOR PUNCTURED CODES DERIVED
FROM 1/2 CODES WITH v = 2, ..., 8 (1: TRANSMITTING,
0: DELETING)
TABLE I1
d , cd. AND c d FOR PUNCTURED CODES LISTEDIN TABLE I
received
(threshold value1 bo-l--- bf,, b Q b+-q---
APPENDIX
CALCULATION O F P k
dg: Free distance of best punctured codes [ 21.
d : Free distance of punctured codes listed in Table I1
In general, the Q-level soft decision channel can be regarded
as thebinaryinputQ-aryoutputsymmetricmemoryless
channel where input symbol x = 0 or 1 and output symbol
y = y l , y z , -., Y Q . Then,assume P(i) to be the channel
transition probability that symbol yi is received when symbol
x = 1 is transmitted. In the transmission channel where white
Gaussiannoise is added, P(i) can be uniquely calculated by
using the error function if EJNo (symbol energy versus one-
sided noise power density ratio) and the soft decision thresh-
olds {bi} (i = 1, 2 , --,Q - 1) are determined. (See'Fig.4.)
In such a channel model, Pk in ( l a ) can be calculated by the
followingequationswhere it is assumedthatthesymbol
metric for Viterbi decoding takes integers0, 1, .*.,Q - 1 [ 71 .
-1
- Eb/~otdBl
Speech A c t l v l t y F r a c t i o n (SAF)
= 1.366 = ,432
1.366+1. do2
P = H/M x SAF
t
/ / / P = .755 (N-14)
Absrmcr-In thiscorrespondence we complement, by means Of an b
analyticalmodel,anearliersimulationstudyontradeoffbetween
delay and TASI advantage in a packetized speech multiplexer.
z
; c = ,701 (N-13)
L.
I. INTRODUCTION
In two previous papers [ 2 ] , [3] the authors addressed the P = .647 (N=12),
problem of multiplexingvoiceand/ordataterminals. As an
extension of this, we provide here the delay characteristics of
the packet voice multiplexer model and a n analytical verifica-
tion of the simulation study reported by Weinstein and Hof-
stetter [41 on tradeoff between delay and TASI advantage. I I I 1
60 80 1m 120 140
BUFFER S I Z E CK1
11. T H E PACKET VOICE MULTIPLEXER Fig. 1. Mean VU queueing delays versus buffer size in VU’s in a packet
For assumptions, definitions, and model development the voicemultiplexer.Notethattheprob. of VU loss is a variable.
reader is referred to the earlier publications [ 2 1 , [ 3 ] . It will,
however, be useful to restate here that due to the discrete time
nature of the queueing model used in this study, we assume ferential delays between talkspurts that affect speech quality.
that the talkspurt and silence durations are distributed accord- However, the talkspurt delays are more difficult to determine
ing toageometricdistributionratherthananexponential andtheresults given herearemeanttoprovidesomefirst-
one as observed by Brady [ 1 ] and used in the simulation study order guidelines on delay performance.
referred to above. Having said this, we proceed directly to dis-
cuss the packet voice multiplexer performance. B. Delay Versus TASI Advantage
A . Delay Characteristics Here we complement the simulation study due to Weinstein
and Hofstetter [ 4 ] by means of our analytical tool primarily
Fig. 1 portrays the variation of mean queueing delays with t o verify their main conclusion, namely, thatif one can tolerate
respect to buffer size, for a system with eight servers, each one moderate delays for the VU’s, then improvements in TASI ad-
operating at 1/8 of the total transmission capacity. The param- vantageforthepacketizedsystem over conventionalcircuit
eters of the system are as shown in the figure. For the multi- switchedTASIaresignificant,particularlyforcaseswherea
plexer shown in Fig. 1, the talkspurt and idle time statistics are small number of users are multiplexed.
assumed t o be nonvarying with time and identical for all the Inourcomparison,wehave used exactlythesamemean
talkers. The utilization factor p , therefore, is varied by chang- talkspurtandsilencedurations,delaycriteriaand VU loss,
ing the number of off-hook callers. etc., as Weinstein. However, we have used an equivalent multi-
In dealing with voice unit (VU) delay, one should be aware server queueing model rather than the single server one used in
of its limitations. At the receiving end, talkspurts are the en- the simulation. This difference in the two models may be seen
tities that are delivered to the subscriber and it is the peak dif- reflected in Figs. 2 and 3, which show the curves relating TASI
advantage and delay for both an average delay criterion and a
Paper approved by the Editor for Computer Communications of the maximumdelaycriterion.Althoughourresultsdo have the
IEEE Communications Society for publication without oral presenta- samebehavior as thoseduetothesimulation,theyshowa
tion. Manuscript received March 31, 1981; revised December 22, 1982. somewhat higher TASI advantage for the same number of “cir-
N. Janakiraman is withSatelliteBusinessSystems,McLean, VA
22102. cuits” M and delay. This difference may be given an intuitive
B. Pagurek is with the Department of Systems and Computer Engi- explanation as follows. For the same utilization, a multiserver
neering, Carleton University, Ottawa, Ont.. Canada K1S 5B6. systemhassomewhatsmallerqueueingtimethanits single
J. E. Neilson is with the School of Computer Science, Carleton Uni- server counterpart. In other words, a multiserver system should
versity, Ottawa, Ont., Canada K1S 5B6. operateatahigherutilization to experience the same mean
.I . . ’