Simulation of processes with nonseparable covariances

The problem of simulating a class of nonstationary processes with nonseparable covariances is considered. When a process covariance function vanishes outside a finite interval, standard techniques of shaping filter realization fail because the covariance function is no longer separable. In this correspondence a finite-memory moving-average (FMMA) structure is proposed as a method of handling this class of processes. Necessary and sufficient conditions are established for the existence of a factorization, and two illustrative examples are presented.