A real-time autocovariance least-squares algorithm

Autocovariance least-squares (ALS) is an off-line noise covariance estimation method that uses steady-state gain to establish a linear model of noise covariance; however, it is unable to correct the noise covariance in real time. The gain will reconverges after the correction, and the varying gain destroys the original estimation model. This study presents a real-time autocovariance least-squares (RT-ALS) algorithm that continuously estimates and corrects the noise covariance while filtering. First, a noise covariance estimation model was established separately for each window. A forgetting factor is then combined to suppress the effect of gain reconvergence on innovation. Finally, the historical model was coupled in a sequential manner to improve real-time estimation accuracy and computational efficiency. Numerical simulations and real-world examples demonstrate that the RT-ALS algorithm can estimate accurate and reasonable noise covariance in real-time, as well as improve the filtering accuracy. (c) 2022 Elsevier Inc. All rights reserved.