Joint equalization–demodulation using deep neural networks with embedded signal-dependent noise variance for SPAD-based UWOC systems

作者：Sheng Xie, Dan Li, Yan Dong, Liying Chen, Renhai Feng, Z.X. Han, Yuanxiang Chen, Shuang Wang, Jun Li · 发表于：Applied Optics · 年份：2025 · DOI：10.1364/ao.586110 · 被引用次数：1 · 研究领域：Optical Wireless Communication Technologies、Underwater Vehicles and Communication Systems、Advanced Photonic Communication Systems

In underwater wireless optical communication (UWOC) systems, using single-photon avalanche diodes (SPAD) as a detector can improve the detection sensitivity and thus improve the transmission distance. However, the signal detection for SPAD-based systems is greatly challenged by the complex optical channel characteristics and SPAD nonlinear distortion. In order to mitigate the nonlinear distortion caused by SPAD detectors, signal-dependent noise (SDN), and the effect of underwater channel attenuation on the transmitted signal in UWOC systems, this paper proposes a signal detection scheme using deep neural networks with embedded SDN variance (DNN-SDN). This scheme performs joint equalization and demodulation of the received signal, avoiding information loss and error propagation between these two processes, thereby achieving more accurate signal recovery. In this work, we first characterize the statistical noise model for the SPAD array and establish a nonlinear relationship between the number of detected photons and the transmitted signal power. The SDN variance is integrated as an input feature during network training. Performance is quantified via a joint loss function with parameters optimized through a systematic ablation study. Finally, the performance of the proposed signal detection scheme is also compared with other schemes in different waters. Simulation results demonstrate that the scheme can effectively mitigate the effects of SPAD-induced nonlinear distortion and S...