PHOTONIC MACHINE LEARNING BASED SIGNAL RECOVERY IN SELF-COHERENT DETECTORS

A computer-implemented method of training a self-coherent detector module (200) to recover distorted optical transmission signals, and processing devices adapted to perform the method are disclosed. The detector module comprises an optical filter unit (210), a detection unit (221) for detecting an amplitude component of an optical input, and digital signal processing circuitry (230) configured to retrieve a phase component of the optical input from the amplitude component. The method comprises iteratively computing updates (104) of adjustable filter coefficients of the filter unit, based on a received first dataset (100), and replacing training signals (105) of the first dataset, taking the updated filter coefficients into account. An initial set of filter coefficients is pre-trained (102) on a received second dataset (103). Training signals of the first and second dataset comprise distorted optical transmission signals that are subjected to filtering by the filter unit.