Identification of parametric linear models with cyclostationary inputs

Identification of non-parametric linear systems with cyclostationary inputs has received considerable attention in recent years. However, identification of parametric linear models has received very little attention. In this paper, some methods are proposed for identification of moving average (MA) and autoregressive moving average (ARMA) linear models with fractionally spaced data input using only the output sequence. It is shown that q-length MA and MA part of ARMA can be identified using only q points of the cyclic autocorrelation provided it is nonzero at two or more incommensurate cycle frequencies. This can be ensured by using the sum of cycle frequency separated signals or by using signals with a low frequency pilot. Computer simulations are presented to support the methods