Improving the Generalization Ability of NN Equalizers of VLCs Using Channel Representation

This research focuses on enhancing the generalization capabilities of neural network (NN)-based equalizers within visible light communication (VLC) systems. NN equalizers often suffer from performance degradation when exposed to varying system operational conditions and external environments, sometimes necessitating retraining. By integrating the principles of representation learning, we developed a novel NN equalizer structure termed the Channel Representation-Generalizable Neural Equalizer (CRGNE). This structure comprises two components: a channel representation network that encodes prior information into channel features, and a signal compensation network that uses these features to adaptively perform signal equalization. Our experimental results demonstrate that CRGNE maintains robust performance under novel conditions, showcasing superior generalization abilities compared to traditional NN equalizers. This study offers valuable insights into improving the generalization of NNs across diverse communication fields, such as optical fiber and wireless radio frequency communications.