Dzieciol, Hubert; (2023) Signal Design and Machine Learning Assisted Nonlinearity Compensation for Coherent Optical Fibre Communication Links. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text PhD_Thesis_final.pdf - Accepted Version Download (9MB) | Preview

Abstract

This thesis investigates low-complexity digital signal processing (DSP) for signal design and nonlinearity compensation strategies to improve the performance of single-mode optical fibre links over different distance scales. The performance of a novel ML-assisted inverse regular perturbation technique that mitigates fibre nonlinearities was investigated numerically with a dual-polarization 64 quadrature amplitude modulation (QAM) link over 800 km distance. The model outperformed the heuristically-optimised digital backpropagation approach with <5 steps per span and mitigated the gain expansion issue, which limits the accuracy of an untrained model when the balance between the nonlinear and linear components becomes considerable. For short reach links, the phase noise due to low-cost, high-linewidth lasers is a more significant channel impairment. A novel constellation optimisation algorithm was, therefore, proposed to design modulation formats that are robust against both additive white Gaussian noise (AWGN) and the residual laser phase noise (i.e., after carrier phase estimation). Subsequently, these constellations were numerically validated in the context of a 400ZR standard system, and achieved up to 1.2 dB gains in comparison with the modulation formats which were optimised only for the AWGN channel. The thesis concludes by examining a joint strategy to modulate and demodulate signals in a partially-coherent AWGN (PCAWGN) channel. With a low-complexity PCAWGN demapper, 8- to 64-ary modulation formats were designed and validated through numerical simulations. The bit-wise achievable information rates (AIR) and post forward error correction (FEC) bit error rates (BER) of the designed constellations were numerically validated with: the theoretically optimum, Euclidean (conventional), and low-complexity PCAWGN demappers. The resulting constellations demonstrated post-FEC BER shaping gains of up to 2.59 dB and 2.19 dB versus uniform 64 QAM and 64-ary constellations shaped for the purely AWGN channel model, respectively. The described geometric shaping strategies can be used to either relax linewidth and/or carrier phase estimator requirements, or to increase signal-to-noise ratio (SNR) tolerance of a system in the presence of residual phase noise.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONCSVXML

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item