As communication network technologies rapidly evolve towards the 6th Generation Mobile Communication Technology (6G), the future may see the emergence of an integrated internet of everything that covers space, sky, land, and sea scenarios. In this context, cross-water Visible Light Communication (VLC) has emerged as one of the key enabling technologies to support such a new network.

In this paper, we address the shortcomings of existing Water-to-Air (W2A) VLC systems with a Deep Learning (DL) aided Link Alignment (LA) scheme for a W2A-VLC system communication between Unmanned Aerial Vehicle (UAV) and Autonomous Underwater Vehicles (AUV), based on the proposed extended W2A (eW2A) VLC channel model. Upon the aligned optical link between the transmitter and receiver, we further design a signal detection scheme based on a wind speed estimator and a channel classifier operating in a channel with unknown wind speeds.

The results show that the fitted eW2A channel model based on Lognormal and Gamma distributions matches well with the channel data generated by Monte Carlo simulations, accurately modeling the cross-water VLC channel. Simulation results show that under high Signal-to-Noise Ratio (SNR) conditions and different wind speeds, the angle estimation accuracy of the four angular layers located between 2~30° for the UAV can reach 100%. The results exhibit good convergence and can effectively support the UAV path decision-making in the DL-LA scheme. Additionally, in the wind speed range of 1~10 m/s, the Wind-speed Estimation (WE) and Channel Classification (CC) aided Symbol Detection (SD), namely the WE-CC-SD scheme, can work under untrained unknown wind speeds. This scheme can achieve a high accuracy of WE and a Bit Error Rate (BER) performance close to that achieved under ideal channel state conditions.

The above results validate the superiority of the proposed DL-LA and WE-CC-SD schemes over traditional methods. The findings can provide valuable references for the improvement and development of cross-water VLC technologies.