Adaptive Prediction of a Random Process Using a Sequential Regression Algorithm

• Main Author: Vladimir A. Golovkov
• Format: Article
• Language: Russian
• Published: Saint Petersburg Electrotechnical University "LETI" 2020-01-01
• Series: Известия высших учебных заведений России: Радиоэлектроника
• Subjects: random process; sample; derivative of random process; non-recursive forecast; adaptation; variance of forecast estimation
• Online Access: https://re.eltech.ru/jour/article/view/385

Introduction. Adaptive statistical prediction of a random process is relevant to a noise compensation in radar and optical location problems. The shape of the signal reflected from the target is often unknown due to the use of short probing pulses passing during their duration in a distance less than the size of the target. Subtracting the noise forecasted in the previous time point from its current value allows one to compensate for the noise.Aim. Investigation of the problem of adaptive linear prediction of random processes by a non - recursive linear filter implementing a sequential regression algorithm for infinitely and finitely differentiable random processes.Materials and methods. Models of random interferences in the form of infinitely and finitely differentiable random processes were considered. The sequential regression algorithm required to estimate the correlation selection matrix, the selection vector of correlation of the forecast and sample units. Due to random process and its derivative incorrelation, the sparse correlation selection matrix was formed. This factor reduced the number of mathematical operations.Results. The results of numerical calculations and the implementation of random process, its optimal and adaptive prediction obtained during the simulation were presented. The adaptive predictive filter with random process derivative sampling provided at least a one third reduction of the number of mathematical operations in comparison with the transversal predictive filter.Conclusion. An algorithm of sequential regression in predicting a random process and its a priori unknown parameters is the closest to the ideal algorithm of direct matrix inversion. It allows to adapt to the changing process parameters. The number of iterations in non-recursive filtering and the value of attenuation of the estimated linear regression coefficients during the adaptation can be used to adapt to the changes in the parameters of the predicted process.