Adaptive Variance Estimation of Sensor Noise within a Sensor Data Fusion Framework

In the field of signal filtering sensor noise variance estimation is of high interest. Various approaches exist for different applications. Within this work, we propose a novel online noise variance estimation scheme based on an online algorithm with exponential forgetting. The approach serves as an extension of a sensor data fusion algorithm that was presented earlier for the application within multi-cell battery systems equipped with cell-individual sensors. Utilizing measurements of electrical-linked sensors the signals and their noises are separated, and the noise variance is adaptively determined. Experiments show that sensor data fusion is equivalent to common methods like low-pass filtering to gain the target signal. Consequently, the variance is estimated with high accuracy especially with regard to signals featuring high dynamic range. Moreover, the results are on a par with difference-based noise estimation. Furthermore, the influence of relevant parameters on the method is investigated namely the adaptivity of the algorithm and the necessary number of involved sensors. As a result, with just eight sensors decent results are achieved within an exemplary application.