CORRELATION FUNCTIONS AND QUASI-DETERMINISTIC SIGNALS

Abstract

When processing data on random functions, they are most often limited to constructing an empirical correlation function. In this regard, the problem arises of constructing a random function (a quasi-deterministic signal) determined by a finite set of random variables and having a given correlation function. Moreover, a random function can often be considered Gaussian, since in many cases a random signal is obtained at the output of the system, which is fairly well approximated by a Gaussian. For stationary random processes and for random fields, this problem has been considered. For random sequences and discrete random fields, as well as for non-stationary random signals, the problem remained open. The article considers the problem of restoring a random sequence from known mathematical expectation and correlation function. Such a model random sequence is constructed, in which the mathematical expectation and correlation function coincide with the given ones. The mathematical expectation and the correlation function are the simplest probabilistic numerical characteristics, but they do not uniquely determine the corresponding set of probability distribution densities that satisfy the conditions of normalization and consistency, provided that for each fixed integer value of the parameter, the random sequence is a continuous random variable. The article considers the restoration of a quasi-deterministic signal in stationary and non-stationary cases. For the stationary case, three examples are given for constructing a quasi-deterministic discrete signal EMBED Equation.DSMT4 , provided that the spectral density has three different forms. For the non-stationary case, the corresponding quasi-deterministic signal was obtained for various cases of the spectrum. The use of a random function model determined by a finite number of parameters makes it possible to significantly simplify the analysis of applied problems, the solution of which is associated with differential equations with random coefficients, which are such quasi-deterministic signals. In this case, there is no need to use a complex apparatus of stochastic differential equations, since the solution of such an equation simply depends on random variables as on parameters.