Modulation Doping For Iterative Demapping of Bit-Interleaved Coded Modulation

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

IEEE COMMUNICATIONS LETTERS, VOL. 9, NO.

12, DECEMBER 2005

1031

Modulation Doping for Iterative Demapping of


Bit-Interleaved Coded Modulation
Leszek Szczecinski, Houda Chafnaji, and Cesar Hermosilla

Abstract In this letter, the problem of the transmitter design


for bit-interleaved coded modulation (BICM) with the receiver
employing iterative demapping, is addressed. Conventionally, the
design is focused on the appropriate choice of the constellation
mapping so that the iterative process converges to small values
of bit- or block- error rates. Here, instead of a difficult design of
a new mapping, we propose to use symbols modulated using two
different mappings. Through numerical simulation, our simple
design is shown to outperform the conventional approach without
increase in the receivers complexity.
Index Terms Bit Interleaved Coded Modulation (BICM),
iterative demapping, iterative decoding.

I. I NTRODUCTION

IT INTERLEAVED coded modulation (BICM) is a


coded modulation transmission scheme suited for fading
and non-fading channels [1]. To improve the performance
of the BICM, iterative exchange of information between the
de-mapper and the decoder was proposed in [2] [3]. Such
BICM with iterative demapping (BICM-ID), provides gains
throughout the iterations if the bits-to-symbol mapping is
appropriately chosen. To design the mapping, heuristics were
used in [4] and computer search was employed in [5]. The
main challenge of the design lies in the lack of a well defined
criterion reflecting the performance [expressed e.g., in terms
of bit error rate (BER) or block error rate (BLER)] for a
limited number of iterations and for a given signal to noise
ratio (SNR). To avoid this issue, the design proposed in [5]
optimizes the performance when a perfect a priori information
about the bits is provided by the decoder (the so-called perfect
feedback [4]). The mappings designed in such way will be
called herein anti-Gray (name used, e.g., in [6],[7]).
The assumption of the perfect feedback indeed materializes
after sufficiently large number of iterations and for the SNR
high enough to trigger the convergence [4] [5]. On the other
hand, for the low SNR and/or for a small number of iterations
the feedback is not perfect and the performance of the BICMID is adversely affected by the anti-Gray mapping. To take
the imperfect feedback into account, a method to predict
the performance of BICM-ID for arbitrary mapping would be
needed. Unfortunately such method does not exist, so the only
way to design the mapping would be via extensive computer

Manuscript received June 10, 2005. The associate editor coordinating the
review of this letter and approving it for publication was Dr. James Ritcey.
The research supported by the government of Canada under research grant
NSERC 249704-02.
The authors are with the Institut National de la Recherche Scientifique
(INRS-EMT), University of Quebec, Montreal, Canada (e-mail: {leszek,
chafnaji, hermosil}@emt.inrs.ca).
Digital Object Identifier 10.1109/LCOMM.2005.12008.

simulations and the results would not generalize to different


codes, block lengths or channel characteristics.
In this letter we present a new approach allowing us to
take advantage of the BICM-ID in all ranges of SNR, for an
arbitrary number of iterations, and for arbitrary block-length.
To circumvent the difficulty of the aforementioned extensive
simulations search we propose to employ, within the same data
block, two different mappings: the Gray and the anti-Gray.
Mixture of symbols obtained with different mappings ensures
desired behavior of the iterative process. Transmission with
such modulation doping1 implies that only one parameter (a
doping ratio - a scalar defining the proportion of Gray and
anti-Gray mapped symbols) has to be chosen, which greatly
reduces the search complexity. The very idea of modulation
doping was presented first in [9][10] and was called therein
irregular modulation. The doping ratio was selected in [9]
using the so-called EXIT charts. Our approach is slightly
different being based on the actual outcome of the simulations;
in this way the finite block-length and/or limited number of
iteration are taken care of.
This letter is organized as follows. In Section II the system
model and principle of iterative demapping are introduced. In
Section III we show how to select the systems parameter
and we show that the proposed approach yields gain over
the conventional BICM-ID (with anti-Gray mapping only);
the so-called EXIT chart is also used to give an explanation
of the obtained improvement. The conclusions are drawn in
Section IV.
II. S YSTEM M ODEL
We consider the BICM system shown in Fig. 1 [4][5], where
a block of information bits b( ), = 1, . . . , Nb is encoded by
a convolutional encoder of rate R. The output of the encoder
is randomly interleaved through {} and transformed into
binary codewords of length m, c(t) = [c1 (t), . . . , cm (t)]
{0, 1}m , where t = 1, . . . , Ns denotes discrete time and
Nb
. The codewords c(t) are mapped into symbols
Ns = Rm
s(t) taken from the complex constellation X using the oneto-one mapping t {}, i.e.,
 sm (t) = t {cm (t)} X . The
constellation
is
unbiased
aX a = 0 and power-normalized

1
2
|a|
=
1.
The
mapping
may vary in time t; here
aX
2m
we consider only the case where t {} 1 , t = 1, . . . , N1
and t {} 2 , t = N1 + 1, . . . , Ns , i.e., only two different
mappings 1 {} and 2 {}, being respectively anti-Gray and
Gray, are used. As we learned, the full credit for this idea
goes to [9][10]; this modulation doping being an important
1 Name coined by analogy to the method employed in [8] in the context of
repeat-accumulate codes.

c 2005 IEEE
1089-7798/05$20.00 

1032

c(t)
Encoder

ex

L k (t)

ap

L b ()

t { . }

s(t)

Decoder

.
1
t { }

r(t)

g(t)

(t)

L k (t)

Fig. 1.

BLER

b()

IEEE COMMUNICATIONS LETTERS, VOL. 9, NO. 12, DECEMBER 2005

10

10

System model.
10

distinctive feature comparing to the conventional schemes


proposed for BICM-ID, e.g. in [4] [5], practically does not
increase the complexity of the de-mapper, which only needs
to change the operation mode depending on the time t.
The modulated signals are transmitted over a memoryless
channel yielding the signal r(t) = g(t)s(t) + (t), where g(t)
is the complex channels gain and (t) is a complex, zeromean white Gaussian noise whose real and imaginary parts
are independent, each with variance N0 /2. Since the channel
identification is out of scope of this letter, the channel state
[i.e., g(t) and N0 ] is assumed known, cf. [4] [5].
The extrinsic logarithmic likelihood ratios (LLRs) Lex
k (t)
for the coded bits ck (t) are produced by the de-mapper 1
t {}
using max-log simplification

|r(t) g(t)a|2 

Laj (t)j {a}


Lex
k (t) = mink

N0
aX0
j=k

|r(t) g(t)a|2 
min

Laj (t)j {a} , (1)

N0
aX1k
j=k

where Lak (t) is the a priori LLR for the coded bit ck (t)
obtained from the decoder in the previous iteration, j {a} is
an operator extracting the j-th bit labelling the symbol a X
and Xk is the set of symbols having the k-th bit set to ,
i.e. Xk = {a : k {a} = }. Since the sets Xk depend on
the mappings t {}, the de-mapper must adjust its operation
according to the time t. The decisions about the information
bits b( ) are made from the sign of a-posteriori LLRs Lap
b ( )
produced by the decoder.
The LLRs calculated by the demapper are deinterleaved
(inverting the operation of the bit-level interleaver {} applied prior to the modulation) and fed to the soft-input softoutput (SISO) a-posteriori decoder [11], implemented here
using the max-log simplification. Thanks to the deinterleaver
the LLRs obtained from symbols with different mappings are
spread throughout the data block. The extrinsic LLRs related
to the coded bits and produced by the decoder are used in the
subsequent iterations as a-priori LLRs in (1).
For the numerical simulations presented in this letter, X
is a 16-ary quadrature amplitude modulation (16-QAM) and
Gray and anti-Gray mappings are shown in Fig. 3; the latter
taken from [5]. A systematic recursive convolutional code of
rate R = 12 with generating polynomials {1 23/35}8 is used
[5], bits are not flushed at the blocks end, i.e., the trellis is

Gray
= 0.5
= 0.6
= 0.7
anti-Gray
7

SNR

10

11

Fig. 2. Performance of the BICM-ID after 8 turbo-iterations in AWGN


channel for different values of the doping ratio .

not terminated. The interleaver is generated randomly and the


number of information bits in the block is equal to Nb = 2000,
thus Ns = 1000.
We consider here additive white Gaussian noise (AWGN)
channel, i.e. g(t) 1 so the SNR is equal to SN R = N10 ,
however, the same approach can be used in fading channels.
III. D ESIGN P ROCEDURE
Using the model defined in Section II, the systems design
1
consists in choosing the doping ratio = N
Ns {0, 1} to
obtain a desirable performance.
The design itself is very simple indeed: the value of
producing the best results (e.g., in terms of BER or BLER)
is chosen on the basis of the results of simulations. Since
only the scalar parameter needs to be adjusted, the search
over whole range of its values does not introduce a significant
computational overhead and may be easily repeated if the
systems parameters (code or block length) change. We show
in Fig. 2 the BLER obtained in AWGN channel after eight
turbo-iterations at the receiver.
We observe that, to optimize the performance (BLER),
should vary with SNR, which may be accomplished through
the SNR measurement at the receiver sent over a feedback
channel to the transmitter. If we target the BLER BLERt =
101 we should use = 0.5, but for BLERt = 102 ,
= 0.6 is the most appropriate. Only for sufficiently high
SNR, using pure anti-Gray mapping ( = 1) provides the
best results (not shown in the figure), but the improvement
is rather marginal since the BLER value is already very low
(less than 102 ), so the effective throughput is very close to
the nominal spectral efficiency (here, 2 bits per channel use).
A reasonable tradeoff is obtained with constant = 0.6; then
the proposed BICM-ID scheme outperforms the conventional
pure anti-Gray mapping design by 1.4dB (at BLERt = 101 )
or 1dB at BLERt = 102 . Such improvement is obtained
without increasing the complexity of the receiver.
The value of may be adjusted depending on the number of
iterations, the latter being dictated by available computational
resources at the receiver (results not shown for lack of space).

SZCZECINSKI et al.: MODULATION DOPING FOR ITERATIVE DEMAPPING OF BIT-INTERLEAVED CODED MODULATION

1
Decoders input/demappers output MIs

0.9
0.8
0.7

is not triggered and proposed modulation doping improves the


performance.

Traj
DEC
anti-Gray
= 0.4
Gray

IV. C ONCLUSIONS

0.6
0.5
0.4
0.3
0.2
0.1
0
0

1033

antiGray

Gray

0.2
0.4
0.6
0.8
Demappers input/decoders output MIs

Fig. 3.
Decoders EXIT function (DEC) and the demappers functions
obtained for SN R=6.5dB when using mappings Gray, anti-Gray and for the
proposed modulation doping with = 0.4. For the latter, the EXIT trajectory
(TRAJ) is shown to illustrate the evolution of the MI throughout the iterations.
The labellings Gray and anti-Gray are shown in hexadecimal format.

In this letter we have presented a new design paradigm


for the bit-interleaved coded modulation when the iterative
demapping is used at the receiver (BICM-ID). Instead of a
difficult approach of designing the mapping for a limited
SNR and/or for a small number of iterations, we propose
to use two different but known mappings to modulate the
symbols within the same transmitted data block. The symbols
from each mapping must be inserted with adequate doping
ratio, which is easy to find through simulations. The proposed
approach does not increase the processing complexity of the
receiver, leads to a simple and flexible design and is shown
to outperform the conventional approach of using only one
prescribed mapping.
ACKNOWLEDGMENT
We thank the anonymous reviewer for bringing our attention
to the publications [9][10].

Finally, to give the reader an intuitive understanding of the


working principle of the proposed scheme we present in Fig. 3
the so-called extrinsic information transfer (EXIT) chart [12],
obtained in AWGN channel. The axes show mutual information (MI) between extrinsic LLRs and the corresponding coded
bits measured at the input/output of the de-mapper/decoder.
For more details about EXIT charts we refer the reader to [6]
[12].
As known, the Gray mappings EXIT function ( = 0)
is quasi-flat [6], thus only slight improvement over iterations
may be obtained. On the other hand, the anti-Gray mappings
EXIT function ( = 1) increases with MI, thus leads to
improvement over iterations, provided it does not intersect
the decoder EXIT function too early (as it is the case for
SN R=6.5dB in Fig. 3). To make the EXIT functions cross
for high value of MI, the demappers function should be an
average of the Gray and the anti-Grays functions. Employing
both mappings within the same data block corresponds indeed
to such MI averaging. In the shown example, using = 0.4,
the de-mappers function is flattened enough (with respect to
the anti-Gray mapping) to open the so-called tunnel [12],
which allows for iterative improvement, as indicated by the
so-called trajectory, shown as well in the figure.
Using the EXIT chart interpretation, the dependance of optimal (i.e, minimizing the BLER) on the SNR, is intuitively
clear: the anti-Gray mapping produces the best results for high
SNR; on the other hand, for moderate SNR, the convergence

R EFERENCES
[1] G. Caire, G. Tarrico, and E. Biglieri, Bit-interleaved coded modulation,
IEEE Trans. Inform. Theory, vol. 44, pp. 927946, May 1998.
[2] X. Li and J. A. Ritcey, Bit-interleaved coded modulation with iterative
decoding, IEEE Commun. Lett., vol. 1, pp. 169171, Nov. 2002.
[3] S. ten Brink, Iterative demapping for QPSK modulation, IEE Electronics Lett., vol. 34, pp. 14591460, July 1998.
[4] A. Chindapol and J. A. Ritcey, Design, analysis, and performance
evaluation for BICM-ID with square QAM constellations in rayleigh
fading channels, IEEE J. Select. Areas Commun., vol. 19, pp. 944
957, May 2001.
[5] F. Schreckenbach, N. Gortz, J. Hagenauer, and G. Bauch, Optimization
of symbol mappings for bit-interleaved coded modulation with iterative
decoding, IEEE Commun. Lett., vol. 07, pp. 593595, Dec. 2003.
[6] S. ten Brink, Convergence of iterative decoding, IEE Electronics Lett.,
vol. 35, pp. 806808, May 1999.
[7] M. Tuchler, Design of serially concatenated systems depending on the
block length, IEEE Trans. Commun., vol. 52, pp. 209218, Feb. 2004.
[8] S. ten Brink and G. Kramer, Design of repeat-accumulate codes for
iterative detection and decoding, IEEE Trans. Signal Processing, vol.
51, pp. 27642772, Nov. 2003.
[9] F. Schreckenbach and G. Bauch, Irregular signal constellations, mappings and precoder, in Proc. International Symposium on Information
Theory and its Applications (ISITA), Oct. 2004.
[10] F. Schreckenbach and G. Bauch, Adaptive bit-interleaved coded irregular modulation, in Proc. 14th IST Mobile & Wireless Communications
Summit, pp. 13321336, June 2005.
[11] S. Benedetto, D. Divsalar, G. Montorsi, and F. Pollara, A soft-input
soft-output APP module for iterative decoding of concatenated codes,
IEEE Commun. Lett., vol. 1, pp. 2224, Jan. 1997.
[12] S. ten Brink, Convergence behaviour of iteratively decoded parallel
concatenated codes, IEEE Trans. Commun., vol. 49, pp. 1727 1737,
Oct. 2001.

You might also like