Abstract

We propose and numerically demonstrate a novel performance enhancement optical reservoir computing (RC) system based on a three-loop mutual coupling electro-optical feedback (TLCEO) structure. The nonlinear capability of the proposed RC system is greatly improved by introducing the mutual coupling structure. The proposed system can operate in two modes: single-channel task high-quality processing mode and three-channel tasks parallel processing mode. We evaluate the performance of the proposed TLCEO system by using the chaotic time-series prediction, waveform classification, Nonlinear Autoregressive Moving-average (NARMA) and memory capacity (MC). In single-channel task mode, the proposed system exhibits the best performance in all three tasks compared to other single-node RC systems and multi-node RC systems. The NMSE of Santa-Fe time series prediction, waveform classification and NARMA-10 are only 0.0021, $\text{1.068} \times \text{10}^{-9} $ and 0.0658, respectively. Moreover, in three-channel tasks mode, the proposed system has the ability to process three different tasks in parallel, and its performance is the strongest compared to other single-node RC systems. Finally, we analyze in detail the impact of different optical parameters on the performance of the RC system. To sum up, the proposed RC system has both high performance and high efficiency, which has excellent application prospects in communication, sensing and fault prediction in optical networks.