R.M. Kurbanaliev, S.S. Zhukov

The problem of signals change detection occupies an important place in signal processing. In this work a model of signal is stochastic process, and a change is refers to a discontinuous change in its properties, which occurs at an unknown time. The proposed method can be referred to a posterior method for which the sampling is made in advance and their task is to approximate the moment of change in the signal. Most algorithms are applied to the problem of change detection, using different kinds of a priori information (the probability that a disorder generally occurs, the distribution of various provisions of the change-point, a priori distribution of parameters, etc.) [2]. However, the interesting case when a priori information is missing, and only some estimates of the signal itself are available. In this paper we propose a method for signal change detection in conditions of intensive noise and denoised waveform recovery. Using pseudogradient adaptation we obtain two estimations of the signal. And the first estimation is performed in the direction of the signal from the beginning to the end («left»), a second estimation ? «right to left». Their co-processing can significantly improve the waveform estimation of the original signal. The experimental results show the effectiveness of the proposed method in comparison with averaging and nonrecursive median filtering.