Low Rate Coded Interleave 2

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

1

Low Rate Coded Interleave Division Multiple


Access on Rician Fading Channels
Mustafa Eroz and Lin-Nan Lee, Fellow, IEEE

performance gap between the two schemes grows bigger for


Abstract—The performance of Interleave Division Multiple heavily loaded systems [6].
Access (IDMA) with a new set of low rate turbo codes is Another key advantage of IDMA over CDMA is that it
evaluated on Additive White Gaussian Noise (AWGN) and allows the use of low rate codes since user separation is
Rician fading channels. Despite the simplicity of the scheme and
excessive multiuser interference, very good performance close to
achieved by user specific interleavers instead of signature
theoretical limit is demonstrated. The effect of varying number spreading sequences [1],[2]. Simple [8] or high performing [7]
of users, code rate and channel characteristics on the low rate codes can be used in coded IDMA to provide
performance is investigated. additional coding gain with respect to coded CDMA systems.
In the next section, we first overview IDMA and describe a
Index Terms—Interleave division multiple access, turbo codes, new set of turbo codes having good performance with more
Rician channel.
modest complexity. In Section III, we demonstrate the
performance of the new coded IDMA in Additive White
Gaussian Noise (AWGN) and Rician fading channels. It is
I. INTRODUCTION
concluded that as the number of users increases beyond

A S the bandwidth and power become scarce to support the


ever increasing throughput requirements, more complex
but more efficient techniques play more important roles in
certain number in AWGN channel, the gap to the
corresponding channel capacity increases. On the other hand,
for certain Rician channel cases, it is demonstrated by
future communication systems. One such recent technique is simulation that having multiple users can actually improve the
called Interleave Division Multiple Access (IDMA) [1],[2]. performance.
With IDMA, different users that share the same bandwidth
and time slots are separated by user specific and possibly
random interleavers. Therefore IDMA is a non-orthogonal II. LOW RATE CODED IDMA
multiple access technique similar to random waveform Code
Division Multiple Access (CDMA) [3]. A. Overview of IDMA
While orthogonal multiple access schemes such as Time Figure 1 depicts a simple block diagram of a low-rate turbo
Division Multiple Access (TDMA) or Frequency Division coded IDMA. Each of the U users are turbo encoded with the
Multiple Access (FDMA) are implicitly too restrictive to same turbo encoder. Then each codeword is interleaved with a
achieve theoretical limits in fading channels, non-orthogonal user specific interleaver before transmission. At the receiver,
CDMA or IDMA have the potential of achieving these limits an iterative multiuser detection or soft interference
[4]. Coding is typically used to improve the performance. At cancellation algorithm and turbo decoding is performed.
the receiver, iterative multiuser detection or interference
cancellation followed by decoding is performed to approach
maximum likelihood (ML) receiver performance without
excessive complexity. But for coded CDMA systems, even
this iterative receiver may lead to complicated algorithms
especially when the number of users is large. Typically with
CDMA, the complexity of multiuser detection or soft
interference cancellation algorithms grows in polynomial form
with the number of users [5]. IDMA, on the other hand, lends
itself to a simple chip by chip detection algorithm whose total Fig. 1. Block diagram of turbo coded IDMA
complexity grows only linearly with the number of users
[1],[2]. Furthermore uncoded IDMA systems perform at least
as well as and usually better than uncoded CDMA. The
B. Low Rate Turbo Codes
Turbo codes [9] and low-density parity check (LDPC) [10]
The authors are with Hughes Network Systems, Germantown, Maryland,
USA codes are powerful forward error correcting schemes that
2

approach the ultimate Shannon limit within 1 dB [11]. While


properly designed, irregular LDPC codes with simple 1.E+00
1/4
encoding have superior performance for higher code rates,
turbo codes give comparable or better performance for lower 1.E-01
code rates. But for very low codes such as 1/6 or lower, 1/7

Frame Error Rate


classical turbo code design can be improved. The room of 1.E-02
improvement exists because turbo codes have traditionally
1/10
been designed to maximize the minimum Hamming weight of 1.E-03
Classical Design

systematic codewords where the information part of the 1/3


codeword has a Hamming weight of two [12]. At relatively 1.E-04 0.8 dB from
1/6
high signal to ratios (SNR), this approach yields good codes Shannon 1/8
1/5
since two codewords are most easily confused when their 1.E-05
information part are differed by two bits, because of the -0.7 -0.5 -0.3 -0.1 0.1 0.3
recursive nature of the constituent codes. But for very low Ebi/No (dB)

SNR where low code rates operate, an investigation of the Fig. 3. AWGN performance of low rate turbo codes (10000 bit interleaver)
erroneous turbo code frames reveals that the number of errors
in the information part is in general more than two. This
observation suggests that by targeting the weight spectrum 1.E+00

corresponding to information sequences with Hamming


weight more than two, we can in principle improve the 1.E-01

performance of low code rates. Following this strategy, we

Frame Error Rate


searched for new turbo codes using the uniform interleaving 1.E-02
1/6
assumption in [12] and constraining ourselves to input
sequences of Hamming weight up to seven. The new 1.E-03

constituent code is shown in Figure 2. 1/8 1/4


1.E-04 1.4 dB from 1/10
Shannon

1.E-05
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8
Ebi/No (dB)
Fig. 4. AWGN performance of low rate turbo codes (1000 bit interleaver)

1.E+00

1.E-01
Fig. 2. Constituent code of the new turbo code
Frame Error Rate

1.E-02
For rate 1/10 turbo code all five parity bits of the constituent
codes are transmitted. For rate 1/8 turbo code, Y4 is punctured
1.E-03
from both constituent code, for rate 1/6 turbo code Y3 and Y4 50000
10000 1000
are punctured and for rate 1/4 turbo code, Y2, Y3 and Y4 are
1.E-04
punctured from both constituent codes. Performance of the 5000

new codes are shown in Figures 3 and 4 with an interleaver


1.E-05
size of 10000 and 1000 bits, respectively. The advantage of -0.7 -0.5 -0.3 -0.1 0.1 0.3
the new codes with respect to classical turbo code design is Ebi/No (dB)

apparent in Figure 3, where the new rate 1/6 turbo code Fig. 5. AWGN performance of rate 1/10 turbo codes with different
outperforms a rate 1/7 classical design turbo code by 0.4 dB. interleaver lengths
Classical turbo codes formed the basis of coding for 3GPP
standards as described in [13]. At a frame error rate (FER) of III. PERFORMANCE OF CODED IDMA
10-3, the new rate 1/10 codes with interleaver sizes of 10000
A. AWGN Channel
and 1000 bits are about 0.8 dB and 1.4 dB, respectively away
from AWGN channel capacity using QPSK modulation with We investigated the performance of coded IDMA using the
0.2 bits/symbol throughput. Figure 5 shows the performance rate 1/10 turbo codes described in Section II. B with a turbo
of rate 1/10 turbo codes with different interleaver lengths. code interleaver size of 1000 bits. At the receiver, we
employed the chip by chip detection algorithm [1],[2]. The
results with different number of users, from two to eight users,
are shown in Figure 6. The x-axis is the total received SNR
3

for all users. For each case, we measured the gap between
simulation results at FER=10-3 versus AWGN channel 1.E+00
capacity using QPSK modulation with the appropriate
throughput. For example with three users, the system has a 1.E-01
total throughput of 3 x 1/10 x 2 = 0.6 bits/channel use and 3 user, R=1/6

Frame Error Rate


therefore the Shannon limit of -2.8 dB. Note that the Shannon 1.E-02
limit depends only on the total throughput and is the same
whether the channel is for a single user or multiple users. The 1.E-03
gap between simulation results and Shannon limit is shown in 4 user, R=1/8
Table 1. 1.E-04
7 user, R=1/14

1.E+00 1.E-05 5 user, R=1/10


1.3 1.5 1.7 1.9 2.1 2.3 2.5
1 user 6 users Total Es/No (dB)
1.E-01 4 users
2 users Fig. 7. AWGN performance of coded IDMA (throughput: 1 bit/sym)
5 users
Frame Error Rate

3 users
1.E-02 B. Rician Channel
For Rician channel simulations we set the Rician factor at
1.E-03 K=15 dB and investigate three different normalized fading
bandwidth BTs=0.00111, 0.000111 or 0.0000111 where B is
1.E-04 the fading bandwidth and Ts is the symbol period. With a
turbo code block size of 5000 symbols (1000 bit input block,
1.E-05 rate 1/10 and QPSK modulation), these normalized fading
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4
Total Es/No (dB)
bandwidths correspond to roughly 5, 0.5 or 0.05 fade cycles
Fig. 6. AWGN performance of coded IDMA (each user uses 1/10 coding)
per block, respectively. Therefore the case with BTs=0.00111
can be characterized as a fast fading channel, whereas the case
with BTs=0.0000111 as a fairly static channel, resembling
TABLE I AWGN channel due to relatively high K factor. The
DISTANCE TO SHANNON LIMIT
performance of coded IDMA over these channels are given in
Number Total Es/No at Shannon
of Users FER=10-3 (dB) Limit (dB) Distance (dB) Figures 8 through 10. For comparison AWGN results of
1 -6.9 -8.3 1.4
Figure 6 are also included in dotted lines.
2 -3.7 -4.9 1.2
3 -1.7 -2.8 1.1
4 0.1 -1.2 1.3
5 1.8 0.2 1.6 1.E+00
6 3.6 1.5 2.1
1 user 6 user
1.E-01 4 user
2 user
5 user
Frame Error Rate

From the results of Figures 4 and 6, a natural question then 3 user


1.E-02
would be how changing the number of users and code rate
simultaneously affects the performance while keeping a
1.E-03
constant total throughput. To investigate this, we simulated 3,
4, 5 or 7 users with user code rates of 1/6, 1/8, 1/10 and 1/14
1.E-04
respectively. Total throughput is 1 bit/channel use in each
0.4 dB
case. QPSK modulation is employed. The turbo interleaver
1.E-05 0.4 dB 0.2 dB
block size is 1000 bits. The results are given in Figure 7. From -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4
the results, it is clear that the power of low rate codes and Total Es/No (dB)
larger coded block sizes are enough to overcome the Fig. 8. Rician channel performance of coded IDMA (BTs=0.00111)
interference of additional users.
4

users.
1.E+00

1 user
1.E-01 4 user
6 user IV. CONCLUSION
2 user
5 user
Combined with carefully designed low rate codes, IDMA is
Frame Error Rate

3 user
1.E-02 a simple but powerful multiple access technique that approach
the channel capacity in AWGN and Rician channels.
1.E-03
1.8 dB REFERENCES
1.E-04 1.8 dB 1.1 dB [1] L. Ping, L. Liu and W. K. Leung, “A simple approach to near optimal
1.6 dB
multiuser detection: interleave division multiple access,” in Proc.
0.5 dB WCNC’03, New Orleans, LA, Mar. 2003, pp. 391-396.
1.E-05
[2] L. Ping, L. Liu, K. Y. Wu and W. K. Leung, “Interleave division
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4
multiple access (IDMA) communications”, in Proc. Int. Symp. Turbo
Total Es/No (dB)
Codes Related Topics, Brest, France, Sept. 2003, pp.173-180.
Fig. 9. Rician channel performance of coded IDMA (BTs=0.000111) [3] K. S. Gilhousen, I. M. Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver,
Jr. and C. E. Wheatley, “On the capacity of a cellular CDMA system,”
From Figures 8 and 10, we observe that the fast fading IEEE Trans. Veh. Technol.., May 1991, pp. 303-312.
[4] D. N. C. Tse and P. Viswanath, Fundamentals of Wireless
and almost static Rician channel results are quite close the Communications, Cambridge University Press, Cambridge, UK, 2005.
AWGN channel results with the two set of curves getting [5] X. Wang and H. V. Poor, “Iterative (turbo) soft interference cancellation
closer with increasing number of users. The similar and decoding for coded CDMA,” IEEE Trans. Commun., vol. 48, pp.
1772-1793, July 2002.
performance of AWGN and Rician channel results in Figure 8 [6] K. Kusume and G. Bauch, “CDMA and IDMA: Iterative multiuser
can be explained by the effectiveness of forward error detections for near-far asynchronous communications,” in Proc. IEEE
correction codes in fast fading environments, whereas that in Int. Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC 2005), September 2005.
Figure 10 can be explained by the similarity of the two [7] L. Ping, W. K. Leung and K. Y. Wu, “Low-rate turbo-Hadamard codes,”
channel characteristics due to high K factor and small IEEE Trans. Info. Theory, vol. 49, no. 12, pp. 3213-3224, Dec. 2003.
fading [8] A. J. Viterbi, “Very low rate convolutional codes for maximum
theoretical performance of spread spectrum multiple-access channels,”
IEEE J. Select. Areas Commun., vol. 8, pp. 641-649, August 1990.
1.E+00 [9] C. Berrou, A. Galvieux and P. Thitimajshima, “Near Shannon limit error
correcting coding and cecoding: turbo codes,” in Proc. ICC (Geneva,
1 user 6 user Switzerland), May 1993.
1.E-01 4 user [10] R. G. Gallager, “Low density parity check codes,” IRE Trans. Info.
2 user
5 user
3 user Theory, IT-8, pp. 21-28., 1962
Frame Error Rate

1.E-02 [11] M. Eroz, F. Sun and L. Lee, “An innovative low density parity check
code design with near Shannon limit performance and simple
implementation”, IEEE Trans. on Commun.., pp. 13-17, Jan. 2006.
1.E-03 [12] S. Benedetto and G. Montorsi, “Design of parallel concatenated
convolutional codes,” IEEE Trans. on Commun., vol. COM-44, pp.
591-600., May 1996.
1.E-04
[13] L. Lee, R. Hammons, F. Sun and M.Eroz, “Application and
standardization of turbo codes in third-generation high speed wireless
1.E-05 0.35 dB 0.25 dB 0.2 dB
data services”, IEEE Trans. on Vehicular Tech., pp. 2198-2207, Nov.
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 2000
Total Es/No (dB)
Fig. 10. Rician channel performance of coded IDMA (BTs=0.0000111)

bandwidth. In Figure 9, on the other hand, we observe that


the Rician channel, being neither very fast fading nor almost
static, poses the biggest challenge for the coded system.
However, we notice that the difference between AWGN and
Rician channel results diminish with increasing number of
users. This is due to the improved performance of the
multiuser detection algorithm when the signal powers of
different users are unequal in a fading environment. With
more users, this power difference becomes more apparent,
therefore helping Rician, multiuser performance. As a matter
of fact this multiuser gain in the fading channel of Figure 9
offsets the additional interference stemming from other users.
As a result, required Eb/No per user to reach FER=10-5 is
approximately constant at about 2.5 dB from one up to five

You might also like