BLIND ESTIMATION OF NOISE VARIANCE FOR 1D SIGNAL DENOISING
A. Kharkov, Viacheslav Oliinyk, Владимир Васильевич Лукин, Sergey Krivenko · Telecommunications and Radio Engineering · 2020
Noise or estimation errors are present in any registered signal (sequence of measurements). Due to this, it is often needed to filter signals. To remove noise efficiently, it is desired to know noise variance a priori or to pre-estimate it for the entire recorded signal (if noise is stationary) or signal fragments (if noise variance can change in time). The presence of a signal component makes such estimation complicated since it is difficult to fully separate the signal and the noise. In this paper, we propose and study a method of blind estimation of noise variance based on discrete cosine transform (DCT) and two robust estimates of data scale. We show that the accuracy of the estimation depends on many factors: signal complexity, signal-to-noise ratio, the block size used, the algorithm of local estimate processing, scale estimator applied. Recommendations concerning the method (algorithm) parameters are given.