Chalmers Publication Library

LDPC codes for optical channels: Is the "FEC limit" a good predictor of post-FEC
BER?

This document has been downloaded from Chalmers Publication Library (CPL). It is the author´s
version of a work that was accepted for publication in:
2015 Optical Fiber Communications Conference and Exhibition, (OFC) 2015; Los Angeles;
United States; 22-26 March 2015

Citation for the published paper:

Alvarado, A. ; Agrell, E. ; Lavery, D. et al. (2015) "LDPC codes for optical channels: Is the
"FEC limit" a good predictor of post-FEC BER?". 2015 Optical Fiber Communications
Conference and Exhibition, (OFC) 2015; Los Angeles; United States; 22-26 March 2015 pp.
7121706.
http://dx.doi.org/10.1364/OFC.2015.Th3E.5

Downloaded from: http://publications.lib.chalmers.se/publication/226348

Notice: Changes introduced as a result of publishing processes such as copy-editing and

formatting may not be reflected in this document. For a definitive version of this work, please refer
to the published source. Please note that access to the published version might require a
subscription.

Chalmers Publication Library (CPL) offers the possibility of retrieving research publications produced at Chalmers
University of Technology. It covers all types of publications: articles, dissertations, licentiate theses, masters theses,
conference papers, reports etc. Since 2006 it is the official tool for Chalmers official publication statistics. To ensure that
Chalmers research results are disseminated as widely as possible, an Open Access Policy has been adopted.
The CPL service is administrated and maintained by Chalmers Library.

(article starts on next page)

 LDPC Codes for Optical Channels: Is the “FEC
Limit” a Good Predictor of Post-FEC BER?
Alex Alvarado, Erik Agrell§ , Domaniç Lavery, and Polina Bayvel
Optical Networks Group, University College London (UCL), Torrington Place, London, WC1E 7JE, UK
§ Departmentof Signals and Systems, Chalmers University of Technology, SE-41296 Gothenburg, Sweden
alex.alvarado@ieee.org

Abstract: We answer the question in the title negatively. More precisely, the FEC limit is
invalid for soft decision decoding and low to medium code rates. A better predictor is the
generalized mutual information.
OCIS codes: (060.4080) Modulation, (060.4510) Optical communications.

1. Introduction and Motivation

Soft-decision forward error correction (SD-FEC) and multilevel modulation formats are key technologies for realizing
high spectral efficiencies in optical communications. The combination of FEC and multilevel modulation is known as
coded modulation (CM), where FEC is used to recover the sensitivity loss caused by the nonbinary modulation.
Current digital coherent receivers are based on powerful digital signal processing (DSP) algorithms, which are used
to detect the transmitted bits and to compensate for channel impairments and transceiver imperfections. The optimal
DSP should find the most likely coded sequence; however, this is hard to realize in practice, and thus, most receivers
are implemented suboptimally. In particular, detection and FEC decoding are typically decoupled at the receiver: soft
information on the code bits is calculated first, and then, an SD-FEC decoder is used.
When optical communications research began to incorporate digital coherent receivers, it became impractical to
implement realtime receivers. Instead, offline DSP was used, typically operating on a sample of data less than 1 µ s
long. In this scenario, the BER after FEC decoding, which is here called post-FEC BER or BERpos , should be as low
as 10−12 or 10−15 to meet higher-layer quality requirements. Since such low BER values cannot be reliably estimated
by Monte Carlo simulations, the conventional design paradigm has been to simulate the system without FEC encoding
and decoding, and optimize it for a much higher BER requirement, the so-called FEC limit or FEC threshold. The
rationale is that a certain BER without coding, which is here called pre-FEC BER or BERpre , supposedly can be
brought down to the desired post-FEC BER by previously verified FEC implementations.
The use of FEC limits assumes that the decoder’s performance is fully characterized by BERpre , and that different
channels with the same BERpre will result in the same BERpos . Under some assumptions on interleaving the code bits,
and that the FEC operates only on hard decisions, this assumption is justifiable. The use of FEC limits, however, did
not change with the adoption of SD-FEC in optical communications.
For any given channel, BERpre can be used to predict BERpos of an SD-FEC decoder. This has been done for
example for some of the SD-FEC decoders in the G.975.1 standard [1], where BERpos values are tabulated versus
BERpre . However, a problem arises when uncoded experiments or simulations rely on tabulated values and claim
(without encoding and decoding information) the existence of an SD-FEC decoder that can deal with the measured
BERpre . The caveat with this approach is that it relies on the strong assumption that the same SD-FEC encoder and
decoder pair will perform identically for two different channels with the same BERpre . As we will see later, there is
no information-theoretic basis for this assumption, and it is indeed often incorrect. This paper investigates the range
of validity of this assumption, and discusses alternative metrics for estimating BERpos of SD-FEC decoders.

2. System Model and Achievable Rates

We consider 11 WDM channels of 32 GBaud in a 50 GHz grid over a single-span of standard single mode fiber (SSMF)
shown in Fig. 1. Optical fiber transmission is simulated using the nonlinear Schrödinger equation (NLSE) via the split
step Fourier method with a step size of 100 m, oversampling factor of 4 samples/symbol, and root-raised-cosine (RRC)
filters with 1% rolloff. The detection process was modeled as an ideal phase- and polarization-diverse coherent receiver
Binary Encoder Binary-input soft-output channel SD-FEC Decoder
SSMF 50 iterations 6.25% OH
C1 NF 3 dB Λ1

Encoder
Staircase

Staircase
Decoder

Decoder
Encoder

.. MQAM PM MQAM ..

LDPC
LDPC
X PM Y
. EDFA .
Cm Mapper Tx α = 0.2 dB/km Rx Demapper Λm
γ = 1.2 (W km)−1
D = 17 ps/nm/km
Mutual Information
I(X;Y )
Generalized Mutual Information
∑m
k=1 I(Ck ; Λk )

Fig. 1. Polarization-multiplexed single-span optical with 11 WDM channel under consideration.

with electronic chromatic dispersion compensation. Data for the central channel was recorded. Gray-mapped MQAM
constellations with M = 2m = 4, 16, 64, 256 equally likely symbols are considered.
Irregular repeat-accumulate low-density parity-check (LDPC) codes with rates Rc ∈ {1/3, 1/2, 3/4, 9/10} were
used. This leads to FEC overheads (OHs) of {200, 100, 33.3, 11.1}%. An outer staircase code with 6.25% OH that
gives a final BERpos of 10−15 for a post-LDPC BER of 4.7 · 10−3 [2, Table I] is assumed (see Fig. 1). Each transmitted
frame consists of 64, 800 code bits C1 , . . . ,Cm , which are assigned cyclically to the modulating bits, with no interleaver.
At the receiver, soft information on the code bits is calcu- 1
8
lated in the form of logarithmic likelihood ratios (LLRs), QPSK
which we denote by Λk with k = 1, . . . , m. These LLRs Λk 7

Achievable Rate [bit/symbol]

16QAM

are then passed to the LDPC decoder. 0.8 64QAM

6 256QAM
An achievable rate for the receiver in Fig. 1 is the gener-
5
alized mutual information (GMI) [3–6] 0.6
4
m m
GMI ! ∑ I(Ck ;Y ) = ∑ I(Ck ; Λk ), (1)
0.4
3
k=1 k=1
2
where I(·; ·) is the mutual information (MI) between two
0.2 1
arbitrary random variables. The GMI is then the sum of
0
bit-wise MIs between code bits and LLRs. Fig. 2 shows the 150 200 250 300 350
GMI as a function of the span length, for MQAM constel- 0 Distance [km]
lations. For each distance and M, we used the launch power Fig. 2. Achievable rates versus span length.
that gave the highest GMI. In this figure, we also show the
distance required for the SD-FEC decoder in Fig. 1 to give BERpos = 10−15 for each combination of 4 constella-
tions and 4 LDPC codes. The vertical position of these 16 markers represent the resulting achievable rates and clearly
show that the results follow the GMI curves. This agrees perfectly with the results in [6]. The penalties due to the
suboptimality of the LDPC and staircase code are between 5 and 15 km and are highest for high code rates.

3. Post-FEC BER Prediction

To study the robustness of the pre-FEC BER as a metric to predict post-FEC BER, we show in Fig. 3 the post-LDPC
BER as a function of BERpre for the same 16 cases (markers) as in Fig. 2. Ideally, all the lines for the same rate should
fall on top of each other, indicating that measuring BERpre is enough to predict BERpos when the channel (in this case,
the modulation format) changes. The results in this figure show that this is indeed the case for high code rates. For
low and moderate code rates, however, BERpre fails to predict the performance of the SD-FEC decoder. An intuitive
explanation for this is that the SD-FEC in Fig. 1 does not operate on bits, and thus, it is not surprising that a metric
that is based on bits (i.e., the pre-FEC BER) cannot be used to predict the performance of the decoder.
Using BERpre to predict the performance of SD-FEC decoders has no information-theoretic justification. To remedy
this, it was proposed in [7] to use the symbol-wise MI I(X;Y ) (see Fig. 1) as a metric to better predict BERpos . The
values of BERpos as a function of the normalized MI I(X;Y )/m are shown in Fig. 4. Again, the prediction works well
only for high code rates. An explanation for this is that the MI is an achievable rate for the optimum receiver, but not
for the (suboptimal) receiver in Fig. 1.
The main contribution of this paper is to put forward the idea of using the GMI to predict the post-FEC BER.
The rationale behind this is that an SD-FEC decoder is fed with LLRs, and thus, the GMI (see (1)) is a better
metric than BERpre and MI. The values of BERpos as a function of the normalized GMI are shown in Fig. 5.
 1 100 1 100
QPSK QPSK
16QAM 16QAM
64QAM 64QAM
0.8 256QAM 0.8 256QAM
-1 -1
10 10

Rc
Rc
0.6 0.6

=3
BER

BER
Rc

Rc
=1

/4
=1

=9
-2 /3 -2

/2
10 10

/10
0.4 0.4
6.25% OH [2] 6.25% OH [2]

-3 -3
0.2 10 0.2 10
0.75 0.8 0.85 0.9 0.95 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 1−Pre-FEC BER 0 Normalized Mutual Information

Fig. 3. Post-FEC BER vs. pre-FEC BER. Fig. 4. Post-FEC BER vs. (normalized) MI
1 10 0
These results show that for a given code rate, changing the QPSK
constellation does not greatly affect the post-FEC BER pre- 16QAM
64QAM
diction based on the GMI. More importantly, and unlike for 0.8 256QAM
-1
BERpos and MI, the prediction based on the GMI appears 10
to work across all code rates. For high rates, however, the0.6 BER
obtained results based on pre-FEC BER, MI, and GMI are
-2
very similar. We believe this could be explained using to the 10
recently discovered asymptotics relationships between these0.4
6.25% OH [2]
quantities [8].
-3
In Fig. 5, we also show the GMI values needed by an 0.2 10
“ideal” SD-FEC code (vertical lines). The horizontal differ-
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ences between these lines and the results obtained by the
LDPC codes represent the rate penalty caused by the use 0 Normalized Generalized Mutual Information
of a suboptimal code. For this particular family of LDPC
Fig. 5. Post-FEC BER vs. (normalized) GMI.
codes, this penalty is around 4% (0.04m bit/symbol) at the
staircase code threshold. Combined with the 6.25% OH of the staircase code, an overall penalty of about 10% with
respect of the GMI prediction was observed.

4. Conclusions
The pre-FEC BER and mutual information were shown to be poor predictors of the performance of soft-decision
FEC decoders, except in the special case of very high code rates and square QAM. The so-called FEC limit is hence
an unreliable design criterion for optical communication systems with soft-decision FEC. On the other hand, the
generalized mutual information was found to give very good results for all code rates. We believe that these results are
valid for other capacity-approaching soft-decision FEC codes, not only the LDPC codes studied in this paper.

References
1. ITU, “Forward error correction for high bit-rate DWDM submarine systems,” Tech. rep., ITU-T Rec. G.975.1 (2004).
2. L. M. Zhang and F. R. Kschischang, “Staircase codes with 6% to 33% overhead,” J. Lightw. Technol. 32, 1999–2001 (2014).
3. G. Caire, et. al., “Bit-interleaved coded modulation,” IEEE Trans. Inf. Theory 44, 927–946 (1998).
4. A. Guillén i Fàbregas, et. al., “Bit-interleaved coded modulation,” F. and Trends in Comm. and Inf. Theory 5, 1–153 (2008).
5. H. Bülow, et. al., “Coded modulation of polarization- and space-multiplexed signals,” in “Asia Communications and Pho-
tonics Conference and Exhibition (ACP),” (Shanghai, China, 2011). (Invited Paper).
6. A. Alvarado and E. Agrell, “Achievable rates for four-dimensional coded modulation with a bit-wise receiver,” in “Proc.
Optical Fiber Communication Conference (OFC),” (San Francisco, CA, 2014).
7. A. Leven, et. al., “Estimation of soft FEC performance in optical transmission experiments,” IEEE Photon. Technol. Lett.
23, 1547–1549 (2011).
8. A. Alvarado, et. al., “High-SNR asymptotics of mutual information for discrete constellations with applications to BICM,”
IEEE Trans. Inf. Theory 60, 1061–1076 (2014).