Model-Based Machine Learning for Joint Digital Backpropagation and PMD Compensation

In this paper, we propose a model-based machine-learning approach for dual-polarization systems by parameterizing the split-step Fourier method Manakov-PMD equation. The resulting combines hardware-friendly time-domain nonlinearity mitigation via recently proposed learned digital backpropagation (LDBP) with distributed compensation of polarization-mode dispersion (PMD). We refer to as LDBP-PMD. train LDBP-PMD on multiple PMD realizations and show that it converges within 1% its peak dB performance after 428 training iterations average, yielding effective signal-to-noise ratio only 0.30 below PMD-free case. Similar state-of-the-art lumped algorithms in practical systems, our does not assume any knowledge about particular realization along link, nor total accumulated PMD. This is significant improvement compared prior work compensation, where typically assumed. also compare different parameterization choices terms performance, complexity, convergence behavior. Lastly, demonstrate models can be successfully retrained an abrupt change fiber.