A measurement-differencing robust kalman filter for SINS/DVL integrated navigation in the presence of complex measurement noise

In underwater navigation using a strapdown inertial navigation system (SINS), integration with a Doppler velocity log(DVL) can effectively suppress the time-accumulating error divergence inherent in standalone inertial navigation. However, under complex underwater operating conditions, DVL measurements are susceptible to various noise disturbances that often exhibit both temporal correlation, i.e., colored noise, and non-Gaussian statistical characteristics. The coupled effects of these two types of noise can severely contaminate the filter innovation, resulting in substantial performance degradation or even divergence of conventional Kalman filters and robust filtering methods designed for only a single type of noise. To address the challenge posed by the coexistence of colored and non-Gaussian noise, this study proposes a measurement-difference robust Kalman filter (MeDRKF) based on measurement reconstruction for navigation under compound noise disturbances. First, the colored noise in DVL measurements is modeled as a first-order Markov process, and a whitened equivalent measurement is constructed by differencing measurements at two consecutive time instants. Since the differencing operation introduces correlation between the process noise and the equivalent measurement noise, the state transition matrix and the associated noise covariance matrices are jointly reconstructed to establish an equivalent filtering model suitable for colored-noise conditions. On this basis, a Mahalanobis-distance-based chi-square test is introduced to identify residual non-Gaussian abnormal measurements online. An adaptive inflation factor is subsequently employed to dynamically adjust the weighting of measurement information, thereby mitigating the adverse effects of non-Gaussian disturbances without compromising the whitened measurement structure. Shipborne experimental data collected on the Yangtze River are used for validation. The temporal correlation characteristics of DVL noise are further investigated using the autocorrelation function and partial autocorrelation function supporting the rationality of modeling the colored DVL noise as a first-order Markov process. Experimental results demonstrate that, compared with conventional filtering methods, the proposed MeDRKF exhibits enhanced noise adaptability and robustness in complex noise environments characterized by the coexistence of colored and non-Gaussian noise.