（复旦大学，中国 上海 200433）
摘要：可见光通信（VLC）系统被认为是6G 通信的重要组成部分。它具有高信道容量、低电磁辐射、高保密性等优势。然而高速发光二极管（LED）VLC 系统受限于调制带宽与高功率下的非线性。提出了一种基于贪婪算法的几何整形编码方式，并使用了一个具有3 层隐藏层的深度神经网络（DNN）作为接收端的解码器。相比于使用传统的无载波幅度相位调制（CAP）解调策略，该技术使系统传输速率得到提升，且能够承受更大的信号峰峰电压和偏置电流，适合用于高速、高功率通信系统。实验证明了DNN 能够替代传统的解调策略，并且误码率表现更加优越，是未来可见光通信中一项很有潜力的技术。
Constellation Shaping and AI-Driven Demodulation Techniques in Visible Light Communication
CAI Jifan, XU Zengyi, CHI Nan
(Fudan University, Shanghai 200433, China)
Abstract: The visible light communication (VLC) system is predicted to be a vital part of 6G communication. It features high channel capacity,low electromagnetic radiation, high security, and other advantages. However, high-speed light-emitting diode (LED) based VLC system usually suffers from limited modulation bandwidth and nonlinearity that resides in high power signals. In this paper, a greedy algorithm-based encoding of geometric shaping is proposed, and a deep neural network (DNN) with three hidden layers as the decoder at the receiver is introduced. Compared with the traditional carrierless amplitude and phase modulation (CAP) demodulation strategy, the system is improved in data rate and shows a greater ability to withstand high bias current and peak-to-peak voltage, which is desired in high-speed high-power communication system. In this experiment DNN proves that it can successfully replace the traditional communication CAP decoding process
with a superior BER performance. It is a promising technology in the future VLC system.
Keywords: geometric shaping; deep neural network; carrierless amplitude and phase modulation