MMFFNet: Multi-Node Fusion Network for Passive Fusion Detection of Underwater Acoustic Communication Signals

In underwater acoustic communication (UWAC) signal detection, integrating data from multi-node can resolve issues caused by the spatial variability of acoustic wave propagation. This research presents a multi-node multi-scale feature fusion network (MMFFNet), which enhances the probability of detection (Pd) in UWAC signals by leveraging joint training of the Preprocessor and Classifier modules. Numerical simulations and field tests conducted in a lake environment verified the efficiency of the model. It is found that when the symbol signal-to-noise ratio (SNR) across three nodes is uniformly −2 dB, MMFFNet outperforms a single-node ResNet, providing an 8 dB gain at a Pd of 95%. Furthermore, it improves the Pd by 10% to 55% compared to existing fusion algorithms. The performance of this method is verified through practical lake trials.