Analysis on adaptability and training methods of Photonic Reservoir Computing in compensating nonlinear effects

Photonic reservoir computing (RC) is proved to be an effective method to recover optical communication signal. In this paper, the adaptability and effectiveness of RC to recover signals at different transmission distances are verified by employing photonic RC model; in addition, performance of the system is investigated using three different linear regression training methods. The analysis results show that when the number of virtual nodes is more than 600, the bit-error-rate (BER) is much lower using Levenberg-Marquardt than the other two methods.