Parametric modelling of cyclostationary processes

Abstract

This paper solves the problem of fitting a time-varying autoregressive model of order ‘p’, (AR(p)), to a cyclostationary process, {x(n)}, via the solution of the time-varying single-step linear prediction problem. A set of ‘Yule-Walker’ equations are derived and are shown to be structurally equivalent to the normal equations from the linear prediction problem. Both block and adaptive simulations are used to implement the linear predictor, and the resulting AR model coefficient estimates are shown to be consistent with the theory. Finally, a technique is developed to estimate the statistical periodicity of any cyclostationary process for use in this modelling process.