V.I. Marchuk, A.I. Sherstobitov, S.A. Gridin, I.A. Toropov

The paper discusses the reasons increasing uncertainty separate the useful signal in prior uncertainty propagation method estimates. The method of propagation estimates the useful component is a priori assumption that the partition function of the measuring process, some small intervals, it (the function) with sufficient accuracy can be approximated by a polynomial of the second degree. The approximation is made by the method of least squares, but with an increase in the variance of the noise component and reduce the approximation interval there is a «revolution» of the approximating function (of decreasing function, it is increasing and vice versa), which leads to increased uncertainty separate the useful signal. This paper proposes a method of improving the reliability of the estimates and the results of studies that confirm the proposed approach. Studies have shown that the elimination of the «flip» the approximation of the approximation in the method of propagation estimates to reduce the estimation error of the desired signal by 30%, regardless of the variance of the additive noise component.