Time-Varying Polynomial System Approach to Multichannel Optimal Linear Filtering

A new approach to linear estimation in time-varying discrete multivariable systems is described. The signal model is taken to be a time-varying vector difference equation which can be expressed in ARMA polynomial system form. An optimal linear filter and predictor is derived in terms of time-dependent polynomial operators and this can also be implemented as a recursive algorithm using difference equations. The system model and filter are particuarly relevant in self-tuning filtering applications.