Pilot-Efficient Channel Estimation with FortiTran for Underwater Wireless Optical OFDM Systems

In underwater wireless optical communication (UWOC) systems with orthogonal frequency division multiplexing (OFDM), water absorption, scattering, and transceiver misalignment increase the difficulty of accurate channel estimation. Estimation accuracy generally improves with pilot density, but a higher pilot density reduces the resources available for data transmission. This work adapts Fortified Transformer (FortiTran) to sparse pilot UWOC OFDM channel estimation. FortiTran captures both local correlations and broader time frequency variations in UWOC channels through its local and global modeling capabilities. For each pilot density, a separate model is trained and evaluated in both simulation and experiment to compare the normalized mean square error (NMSE), bit error ratio (BER), and pilot overhead across the tested pilot configurations. FortiTran achieves lower NMSE and BER than ChannelNet across the evaluated pilot densities in both simulation and experiment. At an experimental pilot density of 9.52%, FortiTran achieves a BER of 3.69×10−3, compared with 5.24×10−3 for ChannelNet, corresponding to a reduction of 29.56%. For the matched BER comparison, ChannelNet at a pilot density of 25% provides the reference BER, and the corresponding interpolated pilot overhead reductions of FortiTran are 60.10% in simulation and 49.26% in the experiment.