Robust Tracking of Multi-Station Underwater Maneuvering Targets Under Skewed Heavy-Tailed Noise

In complex underwater acoustic environments, biases in the effective sound speed and propagation-delay estimation errors may introduce skewness and heavy tails into measurement noise, thereby degrading maneuvering target tracking accuracy. Under these conditions, noise parameter estimates obtained through per-time-step variational updates are susceptible to measurement outliers, while coupling among parameters may hinder effective parameter updates. To exploit recent observations and improve parameter estimation stability, this paper proposes an adaptive interacting multiple model sliding-window variational Bayesian unscented filtering algorithm. Within the interacting multiple-model framework, variational inference and the unscented transform are combined to estimate the target state and noise-related variables at each time step. Sliding-window parameter fusion is applied to the location parameter and mixing variable, while sliding-window maximum a posteriori estimation is employed to update the shape parameter. Furthermore, an adaptive process noise covariance adjustment strategy based on changes in model probabilities is developed to balance rapid response to target maneuvers with robust tracking during steady motion. Simulation and lake-trial results demonstrate that the proposed algorithm outperforms the benchmark methods under skewed heavy-tailed measurement noise and target maneuvers. In the lake trial, the mean position and velocity tracking errors are reduced by 54.5% and 10.7%, respectively.