This article presents a modification of Kalman filters for the problem
of predicting random processes. It differs from modifiable-value filt
ers in the use of constant coefficients in the corresponding expressio
ns and the absence of a need for knowledge of the correlation function
of the random signals subject to optimal filtration.