Maximum Correntropy Kalman Filter for Linear Discrete-Time Systems With Intermittent Observations and Non-Gaussian Noise
Xinmin Song, Min Zhang, Wei Xing Zheng, Zheng Liu
Abstract
During data transmission over unreliable communication networks, intermittent observations may appear due to data loss or packet drops. Meanwhile, in practical applications, communication networks are usually disturbed by non-Gaussian noise, e.g., heavy-tailed impulsive noise. To improve the robustness of the Kalman filter with intermittent observations (IOKF) against non-Gaussian noise, this study proposes the maximum correntropy Kalman filter with intermittent observations (MCIOKF), exploiting only the information arrival probability to design and implement the estimator. The robust maximum correntropy, instead of the conventional minimum mean square error, is taken as the optimality criterion to make the estimator perform better than the IOKF. Similar to the traditional IOKF, the MCIOKF performs time update according to the state mean vector and covariance propagation equation. In measurement updates, the developed MCIOKF adopts a widely used fixed-point algorithm and establishes the augmented model of the IOKF by designing a modified error vector whose covariance matrix contains the state covariance function. Finally, the effectiveness and robustness of the proposed algorithm are validated by a vehicle tracking example.