Hydroacoustic communication and underwater acoustic target recognition constitute critical research areas in underwater acoustic signal processing. The complexity and uncertainty of the underwater acoustic environment pose significant challenges to underwater acoustic signal processing. Specifically, ocean ambient noise, reverberation, as well as other interferences may induce communication instability and recognition performance degradation. Blind source separation (BSS) technique constitutes a fundamental approach to address these challenges. However, conventional statistical characteristic-based BSS methods exhibit limited effectiveness in fast time-varying and spatial-varying hydroacoustic channels. Deep learning-based BSS methods [e.g., time-domain audio separation network (TasNet)] demonstrate enhanced capability in learning feature representations and nonlinear mapping relationships between mixed and source signals, thereby reducing the conventional methods’ reliance on statistical characteristics. Motivated by the exceptional performance of Transformer in speech separation tasks, we propose an early-split Transformer-based blind source separation (BSSepFormer) model for underwater scenarios. To mitigate performance degradation when processing mixed signals with analogous time–frequency characteristics, the proposed method employs the early-split strategy through a dedicated split module integrated between the encoder and decoder for separating the feature representations extracted from different sources. Furthermore, to optimize computational efficiency, the additive attention mechanism is introduced to substitute the dot-product self-attention mechanism in the Transformer. To evaluate the effectiveness, the proposed BSSepFormer is validated on UAMix and ShipsearMix, achieving separation performance improvements of 1.1 and 0.9 dB, respectively, over existing state-of-the-art methods. Experimental results demonstrate that the proposed method not only achieves better separation performance in different scenarios but also enables accurate source signal estimation.

The coprime array, proposed in recent years as a special type of sparse array, combines the advantages of sparse sensing with the unique properties of prime numbers, enabling a larger array aperture and higher degrees of freedom with the same number of physical sensors. In underwater array signal processing, the high-resolution potential of coprime arrays has attracted significant attention. However, in complex ocean environments, leveraging the advantages of coprime arrays to achieve high-resolution and robust target detection still faces challenges posed by sensor failures. Element failures can disrupt the physical structure of the coprime array, leading to significantly increased energy in grating lobes and side lobes of the beam pattern, thereby raising the probability of false target azimuth identification. To address this issue, this paper analyzes the virtual array set mapped from the physical coprime array and proposes a multiplepath matching pursuit method for underwater vector coprime array target azimuth detection based on random virtual array set construction and verification techniques. Cases of continuous and non-continuous virtual arrays are analyzed, and corresponding solutions are proposed. Through simulations and analyses of sea trial data, it is demonstrated that the proposed method can achieve high-resolution target azimuth detection as well as robust target detection in the presence of physical sensor failures.