IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 57, NO. 5, SEPTEMBER 2008 2815
BER Evaluation of BICM-ID via
Bonferroni-Type Bounds
Rolando Bettancourt, Leszek Szczecinski, Senior Member, IEEE, and Rodolfo Feick, Senior Member, IEEE
Abstract—In this paper, we evaluate the bit error rate (BER) of
bit-interleaved coded modulation (BICM) receivers that operate
with a priori information obtained from the decoder during the so-
called turbo iterations. We use Bonferroni-type lower and upper
bounds to tightly estimate the BER performance in the range of
signal-to-noise ratio (SNR) of interest. The results are shown to be
useful for the convergence analysis of the iterative demapping via
the so-called extrinsic information transfer (EXIT) charts and for
the evaluation of the coded performance of BICM with iterative
demapping (BICM-ID).
Index Terms—BICM with iterative demapping (BICM-ID),
bit-interleaved coded modulation (BICM), Bonferroni bounds,
iterative demapping, turbo decoding, turbo demapping.
I. I NTRODUCTION
I
N THIS PAPER, we propose a method for the evaluation
of the bit error rate (BER) of bit-interleaved coded modu-
lation (BICM) [1] receivers that operate with random a priori
information. Such a scenario occurs in the context of BICM
with iterative demapping (BICM-ID) [2], [3], where the soft-
input soft-output (SISO) decoder and the demapper exchange
information about the coded bits.
The performance of BICM-ID strongly depends on the ap-
propriate choice of the bits-to-symbols mapping, which has
been subject to various propositions, e.g., [4]–[6]. These mostly
heuristic approaches have also been complemented, e.g., in
[7] and [8], by a formal algorithmic design of the mapping.
In [7] and [8], the design criteria are based on the assump-
tion of a sufficiently large number of iterations and a high
operating signal-to-noise ratio (SNR) (which are necessary to
ensure convergence), i.e., optimization of the error floor is
targeted.
Manuscript received January 23, 2007; revised July 5, 2007, August 17,
2007, October 11, 2007, and November 14, 2007. This work was supported in
part by the Comisión Nacional de Investigación Científica y Tecnológica, Chile,
under Programa Bicentenario de Ciencia y Tecnología Project ACT-11/2004,
by Fundación Andes, Chile, and by the Natural Sciences and Engineering
Research Council, Canada, under Grant 249704-02. The review of this paper
was coordinated by Prof. J. Choi.
R. Bettancourt was with the Department of Electronics Engineering, Univer-
sidad Tecnica Federico Santa Maria, Valparaíso 110-V,Chile. He is now with
VTR Globalcom, Santiago 3340, Chile (e-mail: rolando.bettancourt@vtr.cl).
L. Szczecinski is with the Institut National de la Recherche Scientifique-
Énergie, Matériaux et Télécommunications (INRS-EMT), University of
Quebec, Montreal, QC H5A 1K6, Canada (e-mail: leszek@emt.inrs.ca).
R. Feick is with the Department of Electronics Engineering, Universidad
Tecnica Federico Santa Maria, Valparaíso 110-V, Chile (e-mail: feick@elo.
utfsm.cl).
Digital Object Identifier 10.1109/TVT.2007.913182
However, often, the so-called waterfall region matters more
than obtaining a low error floor, particularly if retransmissions
are available to take care of erroneously received packets.
In such a case, optimizing the waterfall region (or finding
mappings that guarantee the early convergence) translates into
throughput gains.
The EXIT analysis is probably the most popular method to
evaluate the performance of iterative receivers [9], [10] and
consists of tracking the evolution of the mutual information
(MI) between L-values and coded bits throughout the iterative
process. It yields the so-called extrinsic information transfer
(EXIT) charts that are helpful to analyze convergence after
a theoretically unbounded number of iterations and may also
be used to approximate the performance when the number of
iterations is limited [9], [11], [12].
Despite such appealing features, using EXIT charts for map-
ping optimization is made difficult by the fact that simulations
are necessary to obtain the EXIT functions. To circumvent this
limitation, i.e., to avoid simulations, we propose to accurately
approximate the EXIT functions. The proposed approach is
based on a numerically efficient method that we developed for
the evaluation of the BER at the demapper’s output. Under
some simplifying assumptions, e.g., [9] and [11], the BER may
be accurately related to the MI that is used in the EXIT analysis.
The proposed method may be used to track MI throughout the
iterations, producing results that are more accurate than those
that result from BER tracking, as done in [10] and [13].
The main challenge is the evaluation of the demappers’ BER
with random a priori L-values. For that purpose, the general
BER-evaluation method proposed in [14] was already used in
[15] via integration over the space of the random L-values. The
complexity of such an approach grows as K
log
2
M
, with M
being the constellation size and K—the number of integration
nodes per dimension. To deal with this problem, we propose to
apply the Bonferroni-type algorithmic bounds [16] on the BER.
Their computational complexity is modest, and the advantage
offered is that the effect of the a priori L-values is reduced to
a single random variable, which is marginalized analytically.
Numerical results indicate that the bounds tightly approach the
bit error probabilities obtained via simulations.
This paper is organized as follows. Section II introduces the
notation and provides a quick overview of the principles of
BICM-ID. In Section III, we make the derivations necessary
to apply the algorithmic bounds to estimate the BER of the
MAP receiver with random a priori information. These results
are then applied in Section IV to obtain the EXIT functions of
the demapper that are further used to estimate the coded BER.
Concluding remarks are made in Section V.
0018-9545/$25.00 © 2008 IEEE