The optoelectronic reservoir computing system based on parallel multi-time-delay feedback loops for time-series prediction and optical performance monitoring

Parallel physical reservoir computing (RC) overcomes some obstacles of the single RC, including issues related to large-scale processing and limited performance. In this work, we propose a parallel computing method for optoelectronic RC based on multi-time-delay feedback loops. The reservoir layer consists of multiple nonlinear nodes and loops with different time-delay, which come from different node intervals. Each input signal is mapped to the final state through high-level maps from multiple reservoirs. The results show that in the chaotic time-series prediction task, the proposed multi-time-delay parallel scheme improves the prediction accuracy by nearly one order of magnitude compared to the traditional single-time-delay parallel scheme, even in the case of two time-delay feedback loops. In addition, in the task of optical performance monitoring, the recognition accuracy for all modulation formats (i.e. QPSK, 8QAM, 16QAM, 32QAM and 64QAM) reaches 100 % when the lowest required optical signal-to-noise ratio (OSNR) value are below the corresponding theoretical 20 % FEC limit (BER = 2.4 × 10−2). The mean absolute errors (MAEs) of OSNR monitoring are 0.59 dB, 0.1 dB, 0.3 dB, 0.44 dB, and 0.66 dB, respectively, which are also better than the traditional single-time-delay parallel scheme. Then, we further conduct numerical studies on some influencing factors for the proposed parallel RC, once again proving that the memory capacity of multi-time-delay RC is better than that of the traditional parallel RC. By using the same parameter settings for multiple reservoirs, we simplify the optimization process of reservoir hyper-parameters. The research results demonstrate the enormous potential of parallel reservoirs with multi-time-delay feedback loops in information processing capabilities, and it is expected to reduce size through integrated optical circuits (PICs).