Beruflich Dokumente
Kultur Dokumente
8, AUGUST 2001
Abstract—Joint iterative decoding of multiple forward error An important step forward is the realization that CDMA
control (FEC) encoded data streams is studied for linear multiple combined with convolutional coding can be viewed as the
access channels, such as code-division multiple access (CDMA). concatenation of a bank of error control codes with a “channel
It is shown that such systems can be viewed as serially concate-
nated coding systems, and that iterative soft-decision decoding can encoder”—the CDMA channel—separated by interleavers (if
be performed successfully. To improve power efficiency, powerful present). Serial turbo decoding principles can now be used on
FEC codes are used. These FEC codes are themselves serially con- such systems and have been proposed [17], [14], [1], [23]. In
catenated. The overall transmission system can be viewed as the these systems, a CDMA a posteriori probability (APP) decoder
concatenation of two error control codes with the linear multiple and a bank of single-user error control APP decoders form a
access channel, and soft-decision decoders are used at each stage.
A variance transfer function approach applied to the analysis closed loop system and updated soft information is exchanged
of this system captures the role of the component decoders in an in an iterative fashion analogously to turbo decoding [3], [7].
overall iterative decoding system. We show that this approach Moher [14] demonstrated by simulation that the combination
forms a methodology to study the effects of the component codes of FEC and random interleaving can achieve optimal asymp-
as well as that of the iteration schedule. Analysis and simulation totic efficiencies even with highly correlated signals of the dif-
examples are presented for transmission systems that operate close
to the Shannon limit and illustrate the accuracy of the analysis. ferent users. Powerful FEC codes combined with large inter-
leaver sizes improve performance over that of using only simple
Index Terms—CDMA, iterative decoding, multiuser detection, FEC codes, and create a turbo cliff region, that is a region where
serial concatenation, turbo decoding.
the error probability drops rapidly, following an almost ver-
tical asymptote ([2] and this paper). The difficulty is that the
I. INTRODUCTION CDMA channel APP decoder has a large state complexity of
order . In [23], Wang and Poor replace the APP CDMA
C ODE-DIVISION multiple access (CDMA) is a popular
multiple access method whereby a number of simulta-
neous users access a channel jointly by spreading their signals
decoder by a simple canceler and a single-user minimum-mean
square error (MMSE) filter to perform interference suppression
with user-specific spreading or signature sequences. CDMA as an approximation to APP decoding. The system by Alexander
finds application in a number of modern wireless commu- et al. [1] uses only a simple interference canceler which con-
nications systems, most notably the IS-95 and cdma2000 sists of a subtraction of interference estimates from the received
standards for cellular telephony [13], [19]. CDMA is a linear matched-filter signals and has the lowest complexity. Simula-
accessing method, i.e., the signals from the different users tion results show that the bit error rate (BER) performance for
superpose linearly. Correlation reception is typically used at both approaches is close to the single-user performance of the
the individual receivers, where matched filters suppress part of code alone even for heavily loaded CDMA systems. In [17] the
the interference generated by the joint users. Due to theoretical use of Turbo codes achieved excellent power efficiency with
limitations of the simple correlation receiver, joint detection a complex APP CDMA decoder approximated by an -al-
gorithm. In this paper, we present a comprehensive approach
has been proposed and studied as an alternative to improving
which achieves superior performance using a simple interfer-
performance.
ence canceler, and present an analysis methodology which al-
Recent work has focused on designing efficient multiuser
lows us to understand and predict the performance limits of such
receivers for CDMA systems which operate in conjunction
systems.
with forward error control coding (FEC). The philosophy
For single user communication over additive white Gaussian
behind using FEC in a CDMA system is to move performance
(AWGN) channels, concatenated error control codes have
closer to the limit, in this case the Shannon capacity of the
proven to be extremely powerful, reaching the Shannon ca-
multiple access channel. Due to the excessive complexity of
pacity to within fractions of a dB [20], [3]. Two classes of
an optimal joint decoder however, suboptimal decoders with
concatenation, namely serial and parallel, are common. Parallel
lower implementation complexity are considered as candidates
concatenated coding (PCC) has recently achieved fame under
for practical applications.
the heading of Turbo codes [7] while serially concatenated
coding (SCC) has existed much longer, but has only recently
Manuscript received December 15, 2000; revised May 1, 2001. This work was been recognized to develop its full coding power under soft
supported in part by National Science Foundation under Grant CCR 9732962 iterative decoding.
The authors are with the Department of Electrical Engineering, University of
Utah, Salt Lake City, UT 84112 USA (e-mail: schlegel@ee.utah.edu). Much effort has been spent on the analysis of both of these
Publisher Item Identifier S 0733-8716(01)07231-6. cases, and the behavior of these codes in the turbo cliff region is
0733–8716/01$10.00 © 2001 IEEE
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
SHI AND SCHLEGEL: JOINT ITERATIVE DECODING OF SERIALLY CONCATENATED ERROR CONTROL CODED CDMA 1647
arguably best understood via the extrinsic information exchange The sources generate independent binary information bits
approach presented in [20]–[22], or a variance analysis as in . These bits are written into blocks
[10]. The error floor or error flare region, the area where the or frames of size and fed into parallel outer convolu-
error probability flattens out and away from its vertical drop, is tional encoders of rate . Random interleavers separate the
best understood as the effect of the finite minimum distance of inner from the outer encoders as is customary in SCC, real-
the code [16], [3], [18]. In [3], SCC and PCC systems are ana- izing the interleaver gain so important in Turbo codes. The inner
lyzed via the code distance spectrum and some important con- encoders operate at rate , and their output bits are mapped
clusions are drawn regarding the component codes. Most im- into BPSK symbols and, interleaved again, modulated by se-
portant for our purposes is the conclusion that the “inner code” quence spreaders. Assuming equal power of all users, the signal
of a SCC should be a recursive code in order to realize the Turbo from the th spreader is given by
gain.
There have been some efforts to apply Turbo codes to the
CDMA channel [2], however without explicit consideration of (1)
the iteration schedules when more than two component APPs
are involved. In this paper, we present a step forward in the
analysis and design of the concatenation of error control codes where
with CDMA, based on basic results for SCC systems. The first number of encoded symbols per frame;
idea is to introduce a recursive encoder between the original en- time delay of user ;
coder and the CDMA channel as shown in Fig. 1. This encoder supported on the interval , is the en-
should increase the small distances of the original system and ergy-normalized spreading waveform for user
produce the turbo effect. The serial Turbo decoding strategy is during symbol
extended and applied to a receiver with three component de- and is generated in the usual way as
coders. Variance transfer (VTR) functions are defined and used
to analyze the convergence behavior of the overall system. It
is shown how the component decoders interact with the inter- (2)
ference canceler, and how the system decoding schedule has
to be designed for low error probability and efficient conver-
gence. Finally, we present simulation examples which exhibit where
near-Shannon performance on the CDMA channel with small spreading gain, i.e., the number
component decoders. of chips per symbol;
chip interval;
th spreading chip for user ;
II. SYSTEM MODEL
unit-energy chip waveform.
Fig. 1 shows the transmitter side of a coded CDMA system We assume random spreading in this paper, i.e., the are
consisting of single user error control encoders, each com- chosen randomly and independently, and identically distributed.
prising an outer and an inner FEC code separated by random Such a system is referred to as random CDMA and finds appli-
interleavers. cation, among others, in the IS-95 and cdma2000 standards.
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
1648 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 8, AUGUST 2001
(7)
III. JOINT ITERATIVE DECODING
The structure of an iterative receiver for the serially concate-
nated CDMA system is shown in Fig. 2, and is made up of two by inverting the LLR function defined previously. It is worth
parts: single-user FEC iterative decoders and a joint inter- noting that constitutes the minimum mean square error
ference soft-output decoder, which is approximated by an in- (MMSE) estimator due to the APP decoder,
terference canceler for complexity reasons. Its structure is the and, consequently, we observe that the variance of is
least complex since it merely calculates a canceled version of reduced from that of . The nonlinearity is a crucial
the matched filter outputs [1]: component and can be viewed as a soft bit function. The
function constitutes the hard bit function and generates a
(5) significantly larger error variance.
The estimated symbols are now used to update the
where are soft estimates of . The iterative single-user canceled matched filter values according to (5), starting a
decoders operate with the log-likelihood ratios (LLRs) new iteration. In this way, the iterative FEC decoders and the
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
SHI AND SCHLEGEL: JOINT ITERATIVE DECODING OF SERIALLY CONCATENATED ERROR CONTROL CODED CDMA 1649
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
1650 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 8, AUGUST 2001
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
SHI AND SCHLEGEL: JOINT ITERATIVE DECODING OF SERIALLY CONCATENATED ERROR CONTROL CODED CDMA 1651
Fig. 9. Successful and unsuccessful iteration schedules for operation at (E =N ) = 2:6 dB.
variances larger than that, the error control loop will not reach The particular example in Fig. 7 uses , and ,
zero-variance convergence. The inner code is the same which give a close to best performance. The trajectory of mea-
recursive convolutional code as in Fig. 5, but the mea- sured variances from an actual simulation is nicely contained
surement is on the bits . between the two transfer curves, and progresses inside the con-
It becomes evident that a some point the input variance vergence channel from point with four iterations of loop 1 to
too large for the FEC decoders in loop 2 to progress very far. the convergence point with , leading to very small BERs
According to Fig. 6, this happens for which translates as shown in the Fig. 8, which presents the BERs of this iteration
into dB. Below this value iterations of only the schedule as a function of the channel . As predicted, for
error control loop will not lead to zero-variance convergence. dB, the open channel leads to convergence with
The number of iterations inside each loop at each call five cycles.
is referred to as the iteration schedule Fig. 9 illustrates that different iteration schedules exhibit dif-
, where is the number of iterations in ferent convergence behaviors. In fact, it is evident from our pre-
loop at its -th call. The situation is not quite as difficult vious discussion that the schedule with four iterations in each
to analyze as may appear. It is evident from Figs. 5 and 6 that loop is superior to schedules with fewer iterations inside one
increasing the number of iterations monotonically improve or both loops. Three such unsuccessful schedules are shown in
the variance, and therefore, more iterations in each block are Fig. 9.
always better from a performance point of view. However, there
is a point of diminishing return, especially in loop 1, where
V. FURTHER RESULTS AND APPLICATIONS
additional iterations deliver very little once the loop is near the
equilibrium. We have used the described variance transfer methodology to
Fig. 7 captures the behavior of the iteration schedule, i.e., the gain some insight into some basic questions. In order to illustrate
switching between the two loops. The two functions plotted are its usefulness in the code design, we have interchanged the inner
the full circle variance transfer functions of each loop with the and outer FEC codes. The outer code is now the lower rate
variance of the other loop as input. That is, the functions code and the inner code operates at the high rate
and (14) .
Fig. 10 shows both the transfer exchange graph and the BERs
These functions are extracted from executing iterations in for this switched system. Note that a loss of about 0.3 dB oc-
the computation for Fig. 5, and iterations in Fig. 6. curs w.r.t. the original ordering of codes., which is accurately
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
1652 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 8, AUGUST 2001
(a) (b)
Fig. 10. BERs of the triple-iterative joint multiuser detector with reversed FEC codes.
(a) (b)
Fig. 11. BERs of an overloaded system with K = 20 and N = 10.
predicted by the analysis. Again, four iterations in each loop VI. CONCLUSION
achieve the point of limiting return.
The second analysis example we study is one for an over- We have presented a coding and decoding methodology
loaded system. We use users with processing gain for joint detection systems for CDMA which achieves
, thus doubly load the random CDMA system using near-Shannon limit performance by using serially concatenated
the same concatenated codes in the original configuration with simple FEC codes. The decoder is an iterative decoder for
, and . Fig. 11 again shows the transfer the triple serially concatenated system of FEC codes and
exchange graph and the associated BERs of the system. It be- CDMA channel which iterates soft information between three
comes evident that a significantly larger is required in component systems, forming two loops, where the inner FEC
order to open a channel, in fact dB for the decoder participates in both loops. The iteration schedule has
transfer exchange graph. Despite the larger SNR required, the been defined as the sequence of successive APP iterations to be
turbo cliff is still clearly present. We can see that the channel is executed in each loop. The multiple access decoder has been
quite unbalanced, pinching off early. This is due to the fact that realized as a simple interference canceler.
we did not undertake any code design, and future designs will A variance transfer analysis has been presented and used to
target to open that channel at that point, at the cost of channel study the iteration schedules of the decoder for a specific system
width for small variances. We also observe that the measure- with user with processing gain , each encoded
ments are somewhat inaccurate in this example. This is due, we by a concatenated FEC system using low-complexity inner and
suspect, to the relatively small interleaver sizes of 10 000 in- outer recursive convolutional codes. It has been shown that for
formation bits only, which causes some residual correlation to this system an unconventional iteration schedule with several
degrade the variances. For larger interleavers, this effect disap- iterations inside each loop is superior to other schedules and
pears. achieves near-Shannon limit performance.
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.
SHI AND SCHLEGEL: JOINT ITERATIVE DECODING OF SERIALLY CONCATENATED ERROR CONTROL CODED CDMA 1653
The presented methodology is general and can be used to de- [18] C. Schlegel, Trellis Coding. Piscataway, NJ: IEEE Press, 1997.
sign component systems, predict performance, as well as study [19] The cdma2000 ITU-R RTT Candidate Submission (1998, June). [On-
line]. Available: http://www.itu.int/imt/
their interaction and design successful iteration schedules. Fu- [20] S. ten Brink, “Convergence behavior of iteratively decoded parallel con-
ture work will target the design of optimal code combinations catenated codes,” IEEE Trans. Commun., Mar. 2000, submitted for pub-
and the study of very large systems with . lication.
[21] , “A rate one-half code for approaching the Shannon limit by 0.1
dB,” Electron.Lett., vol. 36, no. 15, pp. 1293–1294, July 2000.
REFERENCES [22] , “Design of serially concatenated codes based on iterative decoding
convergence,” presented at the 2nd Int.Symp.Turbo Codes & Related
[1] P. Alexander, A. Grant, and M. Reed, “Iterative detection in code-di- Topics, Brest, France, Sept. 2000.
vision multiple-access with error control coding,” Eur. Trans.
[23] X. Wang and H. V. Poor, “Iterative (turbo) soft interference cancellation
Telecommun., vol. 9, no. 5, pp. 419–426, Sept. 1998. and decoding for coded CDMA,” IEEE Trans. Commun., vol. 47, July
[2] P. D. Alexander, M. C. Reed, J. A. Asenstorfer, and S. B. Schlegel, “It-
1999.
erative multiuser interference reduction: Turbo CDMA,” IEEE Trans.
Commun., vol. 47, pp. 1008–1014, July 1999.
[3] S. Benedetto, D. Divsalar, G. Montorsi, and F. Pollara, “Serial concate-
nation of interleaved codes: Performance analysis, design, and itera-
tive decoding,” IEEE Trans. Inform. Theory, vol. 44, pp. 909–926, May
1998. Zhenning Shi was born in Beijing, China, in 1975.
[4] , “Soft-Output Decoding Algorithms in Iterative Decoding of Turbo He received the B.S. degree in electrical engineering
Codes,”, TDA Progress Rep., Feb. 1996. from Tsinghua University, China, in 1998. Since
[5] , “A Soft-Input Soft-Output Maximum a Posteriori (MAP) Module September 1998, he has been working toward the
to Decode Parallel and Serial Concatenated Codes,”, TDA Progress Ph.D. degree with the Department of Electrical
Rep., Nov. 1996. Engineering, University of Utah.
[6] C. Berrou, A. Glavieux, and P. Thitimajshima, “Near Shannon limit error His research interests includes multiaccess com-
correcting coding and decoding: Turbo-codes,” Proc. IEEE Int. Conf. on munications and digital signal processing.
Communications, pp. 1064–1070, 1993.
[7] C. Berrou and A. Glavieux, “Near Shannon limit error correcting
coding and decoding: Turbo-codes,” IEEE Trans. Commun., vol. 44,
pp. 1261–1271, Oct. 1996.
[8] R. M. Buehrer, S. P. Nicoloso, and S. Gollamudi, “Linear versus non-
linear interference cancellation,” J. Commun. Networks, vol. 1, no. 2,
June 1999.
[9] L. Brunel and J. Boutros, “Code division multiple access based on inde- Christian Schlegel (S’86–M’88–SM’97) received
pendent codes and turbo decoding,” Ann. Télecommun., vol. 54, no. 7, the Dipl. El. Ing. ETH degree from the Federal
pp. 401–410, July 1999. Institute of Technology (ETH), Zürich, Switzerland,
[10] D. Divsalar, S. Dolinar, and F. Pollara, “Iterative Turbo Decoder Anal- in 1984, and the M.S.E.E. and Ph.D. degrees in
ysis Based on Density Evolution,”, JPL TMO Progress Report 42–144, electrical engineering from the University of Notre
Feb. 15, 2001. Dame, South Bend, Indiana, in 1987 and 1989,
[11] T. R. Giallorenzi and S. G. Wilson, “Multiuser ML sequence estimator respectively.
for convolutional code asynchronous DS-CDMA systems,” IEEE Trans. From 1988 to 1992, he was with Asea Brown
Commun., vol. 44, pp. 997–1008, Aug. 1996. Boveri, Ltd., Baden, Switzerland. He spend the
[12] , “Suboptimum multiuser receivers for convolutionally coded 1991–1992 academic year as Visiting Assistant Pro-
asynchronous DS-CDMA systems,” IEEE Trans. Commun., vol. 44, fessor with the University of Hawaii, Manoa, before
pp. 1183–1196, Sept. 1996. joining the Digital Communications Group, University of South Australia,
[13] TIA/EIA/IS-95 Interim Standard, mobile station—Base station compat- Adelaide. From 1994–1996, he was with the University of Texas, San Antonio,
ibility standard for dual-mode wideband spread spectrum cellular sys- and since 1996, he has been with the University of Utah, Salt Lake City. His
tems, 1993. research interests are in the area of digital communications, code modulation,
[14] M. Moher, “An iterative multiuser decoder for near-capacity communi- multiple access communications, space–time coding for high-capacity wireless
cations,” IEEE Trans. Commun., vol. 47, pp. 870–880, July 1998. systems, and the design of analog VLSI decoders for digital systems. He is
[15] K. R. Narayanan and G. L. Stuber, “A serial concatenation approach to the author of the research monograph “Trellis Coding” published by IEEE
iterative demodulation and decoding,” IEEE Trans. Commun., vol. 47, Press in 1997, and is currently working on “Coordinated Multiple User
pp. 956–961, July 1999. Communications,” co-authored with Professor Alex Grant from the University
[16] L. C. Perez, J. Seghers, and D. J. Costello Jr., “A distance spectrum of South Australia, and on an expansion of his research monograph as “Trellis
interpretation of turbo codes,” IEEE Trans. Inform. Theory, vol. 42, pp. and Turbo Coding” with Professor Lance Pérez. Dr. Schlegel received an NSF
1698–1709, Nov. 1996. 1997 Career Award in support of his research in multiuser communications.
[17] M. C. Reed, C. B. Schlegel, P. D. Alexander, and J. A. Asenstorfer, “It- Dr. Schlegel is currently an Associate Editor for coding theory and techniques
erative multiuser detection for CDMA with FEC: Near-single-user per- for the IEEE TRANSACTIONS ON COMMUNICATIONS, and the technical co-chair
formance,” IEEE Trans. Commun., vol. 46, pp. 1693–1699, Dec. 1998. of the IEEE Information Theory Workshop 2001, to be held in Cairns, Australia.
Authorized licensed use limited to: Rizvi College of Engineering. Downloaded on August 09,2010 at 06:32:14 UTC from IEEE Xplore. Restrictions apply.