Underwater reverberation suppression based on multi-view online subspace tracking in moving target scenarios

Reverberation interference remains a critical bottleneck in active sonar systems. Conventional batch low-rank sparse decomposition methods operate on entire data blocks offline, leading to prohibitive memory consumption, high latency, and poor adaptability to nonstationary underwater acoustic environments. To address these limitations, this paper proposes a Multi-View Online Subspace Tracking (MVOST) framework for streaming reverberation suppression.MVOST preserves spatial locality through Hilbert space-filling curve mapping and multi-view fusion, thereby effectively reducing false alarms,enabling cleaner separation of reverberation from target echoes. To achieve real-time adaptation to nonstationary and abrupt underwater acoustic environments, the framework updates the low-rank subspace frame by frame and incorporates a forgetting mechanism to rapidly track time-varying reverberation statistics. This online update is implemented via recursively maintained cumulative matrices without storing any historical frames, inherently ensuring constant and low per-frame computational complexity and memory footprint. Simulations and field trials in Fuxian Lake demonstrate the effectiveness of the proposed method. It comprehensively outperforms state-of-the-art batch low-rank and sparse decomposition algorithms, achieving an excellent trade-off between real-time processing capability and suppression fidelity in complex underwater environments.