PaperlayerView exact revision on arXiv

Source-linked AI summary

Replacing the Soft FEC Limit Paradigm in the Design of Optical Communication Systems

Alex Alvarado, Erik Agrell, Domanic Lavery, Robert Maher, Polina Bayvel

arXiv:1503.05477v1cs.ITphysics.optics

TL;DR

The paper challenges the channel-independent FEC limit paradigm for soft-decision bit-wise decoding, where pre-FEC BER is assumed to predict post-FEC BER. It evaluates GMI as an alternative and finds it robust across codes, modulation formats, channel conditions, and optical transmission regimes, with spectral-efficiency errors from the soft-FEC paradigm reaching 11% in one reported comparison.

  • Problem

    Pre-FEC BER is not necessarily a reliable predictor of post-FEC BER for soft-decision FEC, although the FEC limit paradigm remains widely used.

  • Method

    The paper evaluates generalized mutual information as a predictor of post-FEC BER for bit-wise soft-decision decoding across simulations and experiments.

  • Results

    The GMI gives very good post-FEC BER predictions across code rates, modulation formats, LDPC and turbo codes, L-value types, and linear and nonlinear optical transmission.

  • Takeaways & Limitations

    The authors suggest replacing the SD-FEC limit with a GMI limit for modern optical communication systems.

Abstract

from arXiv · show

The FEC limit paradigm is the prevalent practice for designing optical communication systems to attain a certain bit-error rate (BER) without forward error correction (FEC). This practice assumes that there is an FEC code that will reduce the BER after decoding to the desired level. In this paper, we challenge this practice and show that the concept of a channel-independent FEC limit is invalid for soft-decision bit-wise decoding. It is shown that for low code rates and high order modulation formats, the use of the soft FEC limit paradigm can underestimate the spectral efficiencies by up to 20%. A better predictor for the BER after decoding is the generalized mutual information, which is shown to give consistent post-FEC BER predictions across different channel conditions and modulation formats. Extensive optical full-field simulations and experiments are carried out in both the linear and nonlinear transmission regimes to confirm the theoretical analysis.

I. INTRODUCTION AND MOTIVATION

Optical systems commonly combine multilevel modulation with FEC, but the SD-FEC limit paradigm assumes pre-FEC BER alone determines post-FEC BER. This paper investigates GMI as a more reliable predictor for bit-wise soft decoding.

  • Coded modulation combines FEC with multilevel modulation to recover sensitivity lost from nonbinary signaling.
  • Bit-wise receivers decouple detection and FEC decoding by computing soft code-bit information before SD-FEC decoding.
  • The FEC limit paradigm designs uncoded systems at a higher pre-FEC BER, assuming verified FEC can reduce it to the desired post-FEC BER.
  • SD-FEC limits assume channels with identical pre-FEC BER produce identical post-FEC BER for a given code, an assumption justified for HD-FEC under suitable interleaving.
  • The paper investigates GMI and reports accurate post-FEC BER prediction across channels, codes, modulation formats, and optical transmission regimes.

L km PMD

The modeled coherent transceiver uses cascaded binary FEC, MQAM, dual polarization, and a bit-wise receiver that computes L-values before sequential SD-FEC and HD-FEC decoding.

  • Each polarization uses an outer encoder serially concatenated with an inner FEC encoder before binary code bits enter an MQAM mapper.
  • The simulations use Gray-mapped 4QAM, 16QAM, 64QAM, and 256QAM, plus a non-Gray 8QAM constellation.
  • The nonlinear channel model contains 11 dual-polarization WDM channels at 32 Gbaud on a 50 GHz grid over a single-mode-fiber span.
  • The bit-wise receiver computes code-bit L-values while ignoring intersymbol and interpolarization interference, then applies SD-FEC followed by HD-FEC.
  • Max-log approximation reduces the computational complexity of L-value calculation.

B. Pre-FEC BER

Pre-FEC BER is a standard uncoded-system metric and predicts post-FEC BER well for ideally interleaved HD-FEC, but not necessarily for SD-FEC.

  • The optimal memoryless HD demapper makes maximum-a-posteriori decisions by hard-deciding the a posteriori L-values.
  • The demapper’s decision rule is only slightly better than standard symbol hard decisions, with noticeable differences only at very high pre-FEC BER.
  • Pre-FEC BER is a good post-FEC BER predictor for HD-FEC with ideal interleaving but is not necessarily reliable for SD-FEC.

C. SD-FEC

Soft-decision FEC sees a binary-input soft-output channel, so the paper evaluates code families and interleaving assumptions to characterize that channel and predict post-FEC BER.

  • The study considers turbo codes and irregular repeat-accumulate LDPC codes with a pseudo-random bit-level interleaver before modulation.
  • The turbo-code family uses code rate 1/3 and puncturing to obtain six additional rates from 2/5 through 5/6.
  • The LDPC family uses standardized codes with rates from 1/3 to 9/10, 64,800-bit frames, and 50 message-passing iterations.
  • The SD-FEC pair sees a binary-input soft-output channel, also called the BICM channel, whose characterization is used to predict post-FEC BER across channels.
  • An interleaver and staircase outer code are assumed so the target post-SD-FEC BER is 4.7 · 10^-3, yielding 10^-15 after HD-FEC under the stated BSC assumptions.

III. ACHIEVABLE RATES

Achievable rates quantify reliable transmission limits for coding schemes over channels with or without memory. The section introduces mutual information as a rate measure and relates receiver interfaces to BISO and BSC models.

  • A coding scheme comprises a codebook, encoder, and decoder that maps noisy observations to an information sequence.
  • An achievable code rate is one for which a coding scheme meets a specified block length and average error-probability target.
  • For channels with memory, achievable rates account for temporal and cross-polarization symbol correlations under information-stability assumptions.
  • Channel capacity is the largest achievable rate supporting vanishing error probability as block length becomes large.
  • The inner SD-FEC interface is modeled as a BISO channel characterized by GMI, while the outer HD-FEC interface is modeled as a BSC characterized by BERpost.
  • For memoryless channels, the ML receiver selects codewords using conditional channel probabilities, and reliable coded modulation is possible when Rcm ≤ I(X; Y ).

C. BW Receivers

Bit-wise receivers separate soft-information calculation from binary SD decoding, making generalized mutual information the relevant achievable-rate measure. GMI accounts for the receiver’s bit-wise metric and can be estimated for exact or approximate L-values.

  • The BW receiver first calculates L-values and then passes them to a binary soft-decision decoder.
  • The BW decoder is generally mismatched to the symbol-wise channel, whereas the ML decoder uses the matched channel metric.
  • An achievable rate for BW decoding is the GMI, which bounds the number of bits per symbol that can be reliably transmitted.
  • The GMI is not proven to be the largest achievable rate for the considered BW receiver, leaving that optimization problem open.
  • For independent input bits, GMI is expressed as a sum of bit-wise mutual informations between code bits and their L-values.
  • With exact L-values, GMI estimation uses s = 1; with approximated L-values, numerical minimization over s is mandatory.

D. AWGN Channel

The AWGN specialization models the optical channel as memoryless complex Gaussian noise with uniformly distributed input symbols. Mutual information and GMI are estimated from transmitted symbols, noise realizations, and demapper L-values.

  • The AWGN model is Y = X + Z, with Z a complex, zero-mean, circularly symmetric Gaussian variable.
  • The AWGN specialization assumes equally likely input bits and therefore equally likely constellation symbols.
  • SNR is defined as ρ ≜ EX[|X|2]/E[|Z|]2 using the signal and noise energies.
  • Exact and max-log L-values provide alternative demapper inputs for estimating GMI from transmitted bits and computed soft values.
  • For max-log L-values, minimizing over s is required because using the exact-L-value estimator produces a rate below the true value.

IV. POST-FEC BER PREDICTION

The study evaluates whether pre-FEC BER, MI, and GMI robustly predict post-FEC BER across channels. AWGN results are presented before nonlinear optical-channel results.

  • The section compares pre-FEC BER, MI, and GMI as predictors of post-FEC BER for a fixed encoder-decoder pair across different channels.
  • Experiments are presented first for the AWGN channel and then for the nonlinear optical channel.

A. AWGN Channel

Across modulation formats and code rates, pre-FEC BER and MI do not consistently predict post-FEC BER for soft-decision bit-wise decoding, whereas GMI provides a consistent prediction. The AWGN results also show that using BER-based or MI-based thresholds can misestimate achievable spectral efficiency.

  • Pre-FEC BER and MI: Pre-FEC BER varies across modulation formats for the same target post-FEC BER, so the SD-FEC limit paradigm is not channel-independent.For 4QAM and code rate Rc = 1/3, BERpre ≈0.2 is required, whereas 256QAM permits BERpre ≈0.23 for the same target BERpost = 4.7·10^-3.
  • Pre-FEC BER and MI: 20%: low-code-rate modulation-format variations can produce errors of up to 20% in estimated spectral efficiency.The errors decrease as code rate increases, but BERpre lacks theoretical justification as a universal SD-FEC performance predictor.
  • L-value distributions: 64QAM and 8QAM can have BERpre ≈0.216 but very different post-FEC BER because their L-value distributions provide different reliability information.The reported post-FEC BER is approximately 5·10^-4 for 64QAM versus 5·10^-2 for 8QAM; high-reliability L-values in 64QAM can be exploited by the iterative SD-FEC decoder.
  • Pre-FEC BER and MI: MI also fails to predict post-FEC BER consistently across modulation formats, especially when 8QAM is included.For 4QAM, 64QAM, and 256QAM, MI appears reliable at high code rates, but this behavior does not extend to 8QAM.
  • GMI prediction: GMI gives consistent post-FEC BER predictions across modulation formats for a given code rate, unlike BERpre and MI.The considered TCs appear universal with respect to GMI, and Fig. 5 shows excellent GMI-based prediction across configurations.
  • GMI prediction: 11%: an MI-threshold paradigm can overestimate spectral efficiency for 8QAM at Rc = 2/3, whereas GMI aligns the markers for the same code.At normalized MI ≈0.71, 4QAM supports Rc = 2/3 for BERpost = 4.7·10^-3, but 8QAM requires Rc = 3/5.

B. Optical Channel—Simulations

Optical-channel simulations show that normalized GMI predicts post-FEC BER consistently across linear and nonlinear channels, modulation formats, code rates, and launch powers. Pre-FEC BER is less reliable, especially at lower code rates.

  • GMI-based prediction of post-FEC BER is excellent for nonlinear-channel simulations, whereas pre-FEC BER is unreliable except as an approximation at high code rates.
  • A random interleaver between the binary encoder and mapper makes the GMI-based prediction more precise.
  • The simulations compare LPDC code rates 1/3, 1/2, 3/4, and 9/10 across 4QAM, 16QAM, 64QAM, and 256QAM.Post-FEC BER is plotted against pre-FEC BER, normalized MI, and normalized GMI.
  • GMI gives the same post-FEC BER predictions for AWGN and NLSE simulations, supporting its robustness across different channels.The agreement also suggests that a Gaussian noise model is reasonable in this setting.
  • GMI accurately predicts post-FEC BER across highly nonlinear conditions when launch power varies from 2.6 dBm to 12.6 dBm for 64QAM with Rc = 3/4.The fixed-distance setup uses L = 210 km.

C. Optical Channel—Experiments

Experiments on a dual-polarization 64QAM Nyquist-spaced WDM system validate normalized GMI as a predictor of post-FEC BER across transmission distances and launch powers. The measured results remain predictive regardless of transmit power.

  • The experiment implements an LDPC code in a dual-polarization 64QAM Nyquist-spaced WDM transmission system to test normalized GMI prediction.The setup uses a seven-line frequency-locked comb and a recirculating loop.
  • Figure 10 plots post-FEC BER against normalized GMI for code rates 2/5, 1/2, 3/5, 3/4, and 9/10, with markers for different span counts and solid lines for AWGN.
  • Transmission distances span 81.8 km to 1308.8 km, while launch powers range from −18 dBm to +2 dBm.These conditions produce normalized GMI values from 0.39 to 0.93.
  • Normalized GMI predicts post-FEC BER regardless of transmit power in the experimental results.Each marker corresponds to a launch power, code rate, and number of spans.

V. CONCLUSIONS

The conclusions identify GMI as a robust way to predict post-FEC BER without encoding and decoding, while finding the FEC limit unreliable for soft-decision bit-wise decoding. The scope is limited to noniterative binary decoding.

  • GMI predicts post-FEC BER without actually encoding and decoding data.
  • The FEC limit is unreliable for optical systems using soft-decision FEC with bit-wise decoders.
  • GMI performs well across code rates, modulation formats, LDPC and turbo codes, L-value variants, and linear and nonlinear transmission.
  • The paper suggests replacing the SD-FEC limit with a GMI limit for modern optical communication systems.
  • The study considers only noniterative binary decoding; different results are expected for nonbinary decoding or iterative detection, which remain future work.
Loading 1503.05477v1…