[Paper Review] Mutual Information as a Figure of Merit for Optical Fiber Systems

Tobias Fehenberger, Norbert Hanik|arXiv (Cornell University)|Apr 10, 2014

Optical Network Technologies38 references7 citations

TL;DR

This paper proposes using mutual information (MI) as a precise figure of merit for optical fiber systems, estimating it via histogram-based methods with blind bin selection. For a 6000 km standard fiber link, 16-QAM achieves 3.1 bits/symbol MI, requiring 29% FEC overhead; digital back-propagation increases MI by 0.25 bits/symbol, enabling 5.7 bits/s/Hz spectral efficiency at Nyquist spacing.

ABSTRACT

Advanced channel decoders rely on soft-decision decoder inputs for which mutual information (MI) is the natural figure of merit. In this paper, we analyze an optical fiber system by evaluating MI as the maximum achievable rate of transmission of such a system. MI is estimated by means of histograms for which the correct bin number is determined in a blind way. The MI estimate obtained this way shows excellent accuracy in comparison with the true MI of 16-state quadrature amplitude modulation (QAM) over an additive white Gaussian noise channel with additional phase noise, which is a simplified model of a nonlinear optical fiber channel. We thereby justify to use the MI estimation method to accurately estimate the MI of an optical fiber system. In the second part of this work, a transoceanic fiber system with 6000 km of standard single-mode fiber is simulated and its MI determined. Among rectangular QAMs, 16-QAM is found to be the optimal modulation scheme for this link as to performance in terms of MI and requirements on components and digital signal processing. For the reported MI of 3.1 bits/symbol, a minimum coding overhead of 29% is required when the channel memory is not taken into account. By employing ideal single-channel digital back-propagation, an increase in MI by 0.25 bits/symbol and 0.28 bits/symbol is reported for 16-QAM and 64-QAM, respectively, lowering the required overhead to 19% and 16%. When the channel spacing is decreased to be close to the Nyquist rate, the dual-polarization spectral efficiency is 5.7 bits/s/Hz, an increase of more than 2 bits/symbol compared to a 50 GHz spacing.

Motivation & Objective

• To establish mutual information (MI) as a reliable, information-theoretic figure of merit for optical fiber systems, especially for soft-decision decoding.
• To develop and validate a histogram-based MI estimation method without prior assumptions on noise distribution, using a blind bin selection technique.
• To evaluate the performance of optical fiber systems in terms of achievable rate, spectral efficiency, and coding overhead using MI as a metric.
• To quantify the impact of digital back-propagation and reduced channel spacing on system capacity and spectral efficiency.

Proposed method

• MI is estimated using empirical histograms of received signal distributions, with the optimal number of bins determined via a blind, accurate method to minimize estimation error.
• The method is validated on a simplified QAM-AWGN channel with phase noise, showing high accuracy compared to analytical MI values for 16-QAM.
• A long-haul 6000 km standard single-mode fiber link is simulated using dual-polarization QAM formats, with MI computed across varying launch powers and channel spacings.
• Digital back-propagation (DBP) is applied in a single-channel mode to assess nonlinear impairment mitigation and quantify MI gains.
• Spectral efficiency (SE) is computed as net SE, accounting for ideal FEC overhead and assuming arbitrarily low BER.
• MI and SE are used as relative metrics to compare modulation formats, channel spacing, and receiver techniques.

Experimental results

Research questions

• RQ1 Can mutual information be accurately estimated in optical fiber systems without assuming Gaussian noise, using only empirical histograms?
• RQ2 What is the optimal modulation format (e.g., 16-QAM vs. 64-QAM) for a 6000 km standard fiber link in terms of MI and system complexity?
• RQ3 How does digital back-propagation affect the achievable mutual information and spectral efficiency in long-haul optical transmission?
• RQ4 What is the impact of reducing WDM channel spacing toward the Nyquist rate on mutual information and spectral efficiency?
• RQ5 To what extent can MI serve as a reliable predictor of post-FEC BER and coding overhead in optical systems?

Key findings

• The histogram-based MI estimation method achieves high accuracy, closely matching analytical MI values for 16-QAM over an AWGN channel with phase noise.
• For a 6000 km standard fiber link, 16-QAM achieves the highest MI of 3.1 bits/symbol at 50 GHz channel spacing, requiring a minimum FEC overhead of 29% when channel memory is ignored.
• Digital back-propagation increases MI by 0.25 bits/symbol for 16-QAM and 0.28 bits/symbol for 64-QAM, reducing required FEC overhead to 19% and 16%, respectively.
• At a 30 GHz channel spacing, the net dual-polarization spectral efficiency reaches 5.7 bits/s/Hz, a significant increase over 50 GHz spacing.
• Reducing channel spacing to 27.5 GHz causes a drop in MI due to spectral overlap and increased interference, especially in the nonlinear regime.
• The MI metric effectively quantifies the performance gain of receiver techniques, such as DBP, and enables comparison of system components in terms of achievable rate.