Rigorous Performance Characterisation of AI-Enhanced 6G Underwater Optical Wireless Networks: Link Budget Correction and Empirical Gain Validation

Abstract

The incorporation of sixth-generation (6G) technologies into underwater optical wireless communication (UWOC) holds the potential for Gbps-class maritime connectivity; however, the field is presently affected by significant overestimations in the systematic link budget found in existing literature. Certain modelling discrepancies found in the literature—such as using absorption instead of total beam attenuation, assumptions of coherent detection in IM/DD systems, neglecting the Hermitian symmetry factor in DCO–OFDM throughput and the unexamined application of laboratory-measured AI gains to analytical budgets—result in range forecasts that surpass physically consistent predictions by two to three times. This study introduces a revised intensity-modulated direct-detection (IM/DD) framework for DC-biased optical OFDM (DCO–OFDM) that corrects for each of these issues. The net throughput is rigorously calculated to be 776 Mbps for QPSK, explicitly considering the 511 usable subcarriers dictated by Hermitian symmetry in a 1024-point transform. Symbol-level Monte Carlo simulations across 40.9 million bits per distance point confirm the analytical BER predictions within 0.3 orders of magnitude at error rates relevant to operational use, whereas a controlled experiment with an MLP equaliser shows that, in a known, stationary and linear LTI channel with perfect channel state information (CSI) provided to the minimum mean squared error (MMSE), the gains of +4 to −6 dB that are often associated with deep learning equalisers diminish to zero. This highlights the necessity of channel uncertainty, transmitter nonlinearity and rapid fading as essential conditions for achieving the benefits of AI-assisted equalisation. The revised framework offers a self-validated range of 44.8 m (only with MRC diversity) and a projected range of 46 m (assuming literature-based CSI prediction) at BER in clear ocean conditions. System-level evaluations of proton-exchange membrane fuel cells indicate that AI-driven load smoothing yields minimal efficiency improvements (< 0.2%) at average AUV power levels, confirming that the 1.8× endurance benefit over lithium-ion batteries is solely due to energy density rather than intelligent power management. By providing fully disclosed parameters and empirical validation for every asserted advantage, this research establishes a scientifically credible benchmark for future calibration of 6G subsea architectures.