Nonlinearity mitigation in a fiber-wireless integrated system based on low-complexity autoencoder and BiLSTM-ANN equalizer

Abstract

We propose and experimentally demonstrate an intelligent nonlinear compensation method using a stacked autoencoder (SAE) model in conjunction with principal component analysis (PCA) technology and a bidirectional long-short-term memory coupled with ANN (BiLSTM-ANN) nonlinear equalizer for an end-to-end (E2E) fiber-wireless integrated system. The SAE-optimized nonlinear constellation is utilized to mitigate nonlinearity during the optical and electrical conversion process. Our proposed BiLSTM-ANN equalizer is primarily based on time memory and information extraction characteristics, which can compensate for the remaining nonlinear redundancy. A low-complexity 50 Gbps E2E-optimized nonlinear 32 QAM signal is successfully transmitted over a span of 20 km standard single-mode fiber (SSMF) and 6 m wireless link at 92.5 GHz. The extended experimental results indicate that the proposed E2E system can achieve a reduction of up to 78% in BER and a gain in receiver sensitivity of over 0.7 dB at BER of 3.8 × 10 −3 . Moreover, computational complexity is reduced by more than 10 times compared to the classical training model.