Multi-mission cascaded DNN-based backscattering interference compensation for full-duplex underwater wireless optical communication.

Underwater wireless optical communication (UWOC) attracts increasing attention for offshore exploration and marine engineering remote vehicle applications, characterized by higher data rates, larger bandwidth, and lower latency compared to traditional underwater wireless communication. Full-duplex (FD) UWOC, as an important research direction enabling bidirectional data exchange, faces severe backscattering interference (BI) from co-located transmitters. Existing systems rely on time-division or wavelength-division duplexing to mitigate BI at the cost of bandwidth utilization. A multi-mission cascaded deep neural network (MMC-DNN) is proposed in this research for BI compensation, which more accurately estimates multipath channel impulse responses in realistic volumetric scattering environments and effectively handles the nonlinear saturation effects and signal distortion induced by BI on highly sensitive detectors compared with linear mathematical methods. A single-wavelength simultaneous FD-UWOC system achieving one-way 300 Mbps error-free communication is demonstrated in a 1 m water tank with realistic volumetric backscattering under various water types. Experimental results show that, at the minimum effective signal optical power (ESOP), MMC-DNN reduces BER from 4.4 × 10-1 to 7.0 × 10-3 with 17.8 dB SNR improvement, outperforming LMMSE, which achieves only 4.1 × 10-2 with 13.0 dB improvement. As ESOP increases, MMC-DNN reaches error-free performance at 27.3 dB SNR, whereas LMMSE remains at 6.1 × 10-5 with a maximum SNR of 20.0 dB.