Nonlinear Post Equalizer for Visible Light Communication System Based on Deep Residual Convolutional Neural Network with Feature Concatenation

Visible light communication (VLC) is a promising wireless communication technology but its performance is limited by linear and nonlinear distortions. In this paper, we proposed a post-equalization method based on deep residual convolutional neural network (DRCNN) with feature concatenation and demonstrated the effectiveness of the residual structure and the excellent compensation performance of DRCNN through experiments on a PAM-8 VLC system. The residual structure reduces the bit error rate (BER) by 46.7% on severely distorted data collected with direct current (DC) bias of 140mA, voltage peak-to-peak (VPP) of 0.8V, and bit rate of 450Mbps. Compared to the conventional linear FIR equalizer, the DRCNN improves the Q factor by 1.4 dB at maximum and the operating current range and voltage range of the system below the hard decision forward error correction (HD-FEC) threshold of 3.8 × 10 ⁻³ by 43.5% and 19.5%, respectively.