Fiber nonlinearity compensation using nonlinear equalizer based on complex-valued reservoir computing

We investigated the performance of a nonlinear equalizer based on complex-valued reservoir computing (CVRC) for compensation of nonlinear waveform distortion caused by fiber nonlinearity in optical fiber communication systems. Reservoir computing (RC) is a kind of neuromorphic signal-processing algorithm. One important advantage of RC is the high-speed training capability compared to feedforward artificial neural networks (ANNs) with multiple (three or more) layers. CVRC can lower the computational complexity and directly deal with the complex amplitude of the IQ-modulated optical signals. We analyzed the impact of various parameters in CVRC on the equalization performance. The parameters include the number of reservoir units, the spectral radius, the timing of the training signal, and sparse connectivity. By optimizing these parameters, we achieved efficient operation of the CVRC-based nonlinear equalizers, making them applicable to various optical transmission distances. A comparative performance evaluation between CVRC and conventional real-valued RC (RVRC) was conducted through numerical simulations and experiments. We clarified that the CVRC-based nonlinear equalizer requires only about half the number of reservoir units compared to the RVRC-based nonlinear equalizer and reduces the computational complexity to approximately one-quarter.