Streamlined Complex-Valued Neural Network Equalizer Based on Extraction and Fusion Technique in Visible Light Communication

Abstract

Visible Light Communication (VLC) is considered a key technology for the next generation of wireless communication, yet its performance is limited by linear and nonlinear distortions. Neural network has been used to extract impairments in VLC system, and this equalization strategy has been experimentally demonstrated. Here, we propose a novel extraction and fusion neural network (EFNN) that extracts impairments while preserving phase relationships in complex-valued signals, and we conduct experimental verification in the QAM-CAP system. Moreover, we adopt newly designed shared extraction kernel and zero-overlap feature fusion kernel to reduce the number of parameters by up to 49.2% while maintaining excellent signal compensation effect. Experiments indicate that the proposed EFNN can achieve better compensation performance and remain the bit error rate (BER) below the 7% hard-decision forward error correction (HD-FEC) limit of 3.8 × 10 ⁻³ when other equalizers become ineffective under conditions of severe distortion.