A comparative study on subband identification

This paper reports some new study on the subband identification approach. We first provide a complete characterization for the subbands to be decoupled in the sense that different subband channels are statistically independent and the subband model is diagonal. We then apply this result to studying the performance of subband identification schemes based on recursive least-squares algorithms. We provide expressions for the computational cost, asymptotic residual error, and convergence rate. These expressions can be used to determine when the subband approach is advantageous over the full-band approach. A simulation example is given to demonstrate these advantages