An Optimal Adaptive Constellation Design Utilizing an Autoencoder-Based Geometric Shaping Model Framework

Since visible-light communication (VLC) has become an increasingly promising candidate for 6G, the field of underwater visible-light communication (UVLC) has also garnered significant attention. However, the impairments introduced by practical systems and the time-varying underwater channels always limit the performance of underwater visible-light communication. In this paper, we propose and experimentally demonstrate an autoencoder-based geometric shaping model (AEGSM) framework to jointly optimize quadrature amplitude modulation (QAM) signals at the symbol-wise and bit-wise levels for underwater visible-light communication. Unlike traditional geometric shaping (GS) methods, which only give theoretically optimal shaping solutions, our framework can always obtain the globally optimal shaping scheme for a specific channel condition or different application scenarios. In our AEGSM framework, an autoencoder is used to find the optimal shaping scheme at the symbol-wise level and a revised pairwise optimization (RPO) algorithm is applied to achieve bit-wise optimization. In a real UVLC system, 2.05 Gbps transmission is achieved under the hard decision–forward error correction (HD-FEC) threshold of 3.8 × 10−3 by employing the autoencoder-based 8QAM (AE-8QAM) optimized by the AEGSM, which is 103 Mbps faster than the Norm-8QAM. The AE-8QAM also shows its resistance to nonlinearity and enables the UVLC system to operate within a larger dynamic range of driving voltages. The results substantiate the potential and practicality of the proposed AEGSM framework in the realm of underwater visible-light communication.