Locally stationary stochastic processes and Weyl symbols of positive operators (Q539275)

Locally stationary stochastic processes \(z(t)\) extend the class of stationary processes, namely the corresponding covariance function has the form \(E(z(t)\overline{z(s)})=f_1(\frac{t+s}{2})f_2(t-s)\) for two functions or distributions \(f_1,f_2\). The paper uses results from [\textit{A. J. E. M. Janssen}, Philips J. Res. 44, No.~1, 7--42 (1989; Zbl 0684.47021)] about Weyl symbols of nonnegative Hilbert-Schmidt operators to make connections between locally stationary stochastic processes and Weyl symbols of positive operators. The main part of the paper consists of an analysis of necessary conditions for \(f_1\) and \(f_2\) to describe a locally stationary stochastic process. Some discussions about sufficient conditions are also given. A characterization of locally stationary stochastic processes in terms of the existence of a stochastic process on \(\mathbb{C}\) whose restriction to the diagonal has the same covariance function defined on \(\mathbb{R}\) is given. Various properties of locally stationary stochastic processes are analyzed and some consequences of the time-frequency analysis are obtained. A subclass of Cohen's class of time-frequency representations having separable kernels, related to locally stationary stochastic processes is characterized by the result of the last section of the paper.