Optimal Adaptive Waveform Design Utilizing an End-to-End Learning-Based Pre-Equalization Neural Network in an UVLC System

Visible light communication (VLC) has emerged as a promising communication method in the special field of 6G, such as intersatellite communication (ISC) and underwater wireless optical communication (UWOC). Green light-emitting diode (LED)-based underwater visible light communication (VLC) is considered a potential candidate for UWOC of mobile autonomous underwater vehicle (AUV) and remotely operated underwater vehicle, particularly in coastal and harbor environments. However, due to bandwidth limitations and variable underwater channels, implementing high-speed underwater VLC remains a challenge. In this paper, a neural network-based auto equalization model (NNAEM) using end-to-end learning is proposed to achieve a high-speed underwater VLC link with a bandwidth-limited green LED. The entire NNAEM includes a neural network-based channel model, a pre-equalization neural network, and a post-equalization neural network. Based on the data-driven channel model, our auto equalization (AE) method can dynamically pre-equalize the discrete multitone (DMT) modulated signal in different bandwidth-limited cases and various nonlinearity cases. An underwater VLC link with a data rate of 3.451 Gbps and a transmission distance of 1.2 m is experimentally demonstrated by employing the proposed NNAEM under the hard decision-forward error correction (HD-FEC) threshold of 3.8 x 10(-3). The data rate improvement is up to 30.4% compared to other advanced digital signal processing (DSP) methods. The scenario adaptation and robustness of the NNAEM are also verified in the experiments, which demonstrate that the NNAEM can serve an important role in underwater high-speed VLC transmission.