Online Sensor Noise Covariance Identification Using a Modified Adaptive Filter

Modern control systems heavily relay on sensors for closed-loop feedback control. Degradation of sensor performance due to sensor aging affects the closed-loop system performance, reliability, and stability. Sensor aging characterized by the sensor measurement noise covariance. This paper proposes an algorithm used to identify the slow varying sensor noise covariance online based on system sensor measurements. The covariance-matching technique, along with the adaptive Kalman filter is utilized based on the information about the quality of weighted innovation sequence to estimate the slow time-varying sensor noise covariance. The sequential manner of the proposed algorithm leads to significant reduction of the computational load. The covariance-matching of the weighted innovation sequence improves the prediction accuracy and reduces the computational load, which makes it suitable for online applications. Simulation results show that the proposed algorithm is capable of estimating the slow time-varying sensor noise covariance for MIMO systems with white noise whose covariance varies linearly, exponentially, or linearly with added sinusoid perturbation. Furthermore, the proposed estimation algorithm shows a reasonable convergence rate.