[Paper Review] Statistical Properties of Nonlinear Phase Noise

Keang-Po Ho|ArXiv.org|Mar 21, 2003

Optical Network TechnologiesEngineering8 references16 citations

TL;DR

This paper analytically derives the joint characteristic functions of nonlinear phase noise with electric field, received intensity, and amplifier noise phase, enabling exact error probability calculations for PSK and DPSK signals under non-Gaussian nonlinear phase noise. The key contribution is the derivation of optimal linear and nonlinear minimum mean-square error (MMSE) and maximum a posteriori (MAP) compensators, showing that nonlinear MMSE compensators outperform linear ones by up to 0.23 dB in SNR penalty, and that optimal nonlinear MAP detection offers only marginal gains over linear MAP detection (≤0.14 dB).

ABSTRACT

The statistical properties of nonlinear phase noise, often called the Gordon-Mollenauer effect, is studied analytically when the number of fiber spans is very large. The joint characteristic functions of the nonlinear phase noise with electric field, received intensity, and the phase of amplifier noise are all derived analytically. Based on the joint characteristic function of nonlinear phase noise with the phase of amplifier noise, the error probability of signal having nonlinear phase noise is calculated using the Fourier series expansion of the probability density function. The error probability is increased due to the dependence between nonlinear phase noise and the phase of amplifier noise. When the received intensity is used to compensate the nonlinear phase noise, the optimal linear and nonlinear minimum mean-square error compensators are derived analytically using the joint characteristic function of nonlinear phase noise and received intensity. Using the joint probability density of received amplitude and phase, the optimal maximum a posteriori probability detector is derived analytically. The nonlinear compensator always performs better than linear compensator.

Motivation & Objective

• To analytically characterize the statistical properties of nonlinear phase noise (Gordon-Mollenauer effect) in long-haul optical fiber systems with a large number of spans.
• To address the limitation of Gaussian noise assumptions by deriving the joint probability density function (p.d.f.) of nonlinear phase noise and signal phase, which are non-Gaussian and dependent.
• To develop optimal nonlinear compensation techniques—specifically MMSE and MAP detectors—that minimize error probability beyond what is achievable with linear compensation.
• To quantify the performance gain of nonlinear versus linear compensation in terms of SNR penalty and error probability for PSK and DPSK modulation formats.

Proposed method

• Models the nonlinear phase noise as a distributed process over a large number of fiber spans (≥32), replacing discrete span-by-span summation with an integral for analytical tractability.
• Derives the joint characteristic function of nonlinear phase noise with the phase of amplifier noise, enabling the inverse Fourier transform to obtain the joint p.d.f. of the received phase.
• Uses Fourier series expansion of the joint p.d.f. to compute the error probability of PSK and DPSK signals, accounting for the non-Gaussian and dependent nature of nonlinear phase noise and amplifier noise.
• Derives optimal linear and nonlinear MMSE compensators by minimizing the variance of residual nonlinear phase noise using the joint characteristic function with received intensity.
• Derives the optimal MAP detector by maximizing the posterior probability of the transmitted signal given the noisy received signal, using the exact joint distribution of amplitude and phase.
• Performs numerical optimization to find the optimal linear MAP compensator and compares it with the nonlinear MAP and MMSE detectors using error probability and SNR penalty metrics.

Experimental results

Research questions

• RQ1 How does the dependence between nonlinear phase noise and amplifier noise phase affect the error probability of PSK and DPSK signals?
• RQ2 What is the optimal linear and nonlinear compensator for minimizing residual nonlinear phase noise variance when using received intensity as a compensation reference?
• RQ3 How does the error probability of nonlinearly compensated signals compare to linearly compensated signals, and what is the resulting SNR penalty difference?
• RQ4 Can the optimal MAP detector for phase-modulated signals be derived analytically under non-Gaussian nonlinear phase noise, and how does it compare to MMSE-based detection?
• RQ5 To what extent does nonlinear compensation improve system performance beyond linear MMSE, and what is the quantitative gain in terms of SNR penalty reduction?

Key findings

• The joint p.d.f. of nonlinear phase noise and amplifier noise phase is non-Gaussian and asymmetric, leading to a non-centered optimal decision region for PSK signals, contrary to symmetric assumptions in prior work.
• The dependence between nonlinear phase noise and amplifier noise phase increases the error probability compared to independent noise models.
• The nonlinear MMSE compensator reduces SNR penalty by up to 0.23 dB compared to the linear MMSE compensator, despite similar residual phase variance, demonstrating that variance alone is not a reliable performance metric.
• The optimal nonlinear MAP detector performs only up to 0.14 dB better than the optimal linear MAP detector, indicating that linear MAP compensation via numerical optimization closely approximates the optimal nonlinear solution.
• The optimal operating point for system design—defined as the mean nonlinear phase shift minimizing 1-dB SNR penalty—is lower for MAP detectors (e.g., 2.12 rad for nonlinear MAP) than for MMSE schemes, due to steeper error probability curves.
• The MMSE criterion does not minimize error probability, as shown by the fact that the nonlinear MMSE compensator performs better than the linear one even when both have the same residual variance.