Linear methods for TFARNA parameter estimation and system approximation

Time-frequency autoregressive moving-average (TFARMA) models have recently been introduced as parsimonious parametric models for underspread nonstationary random processes. In this paper, we propose linear TFARMA and TFMA parameter estimators based on a high-order TFAR model. These estimators extend the Graupe-Krause-Moore and Durbin methods for time-invariant parameter estimation to underspread nonstationary processes. We also derive linear methods for approximating an underspread time-varying linear system by a TFARMA-type system. The linear equations obtained have Toeplitz/block-Toeplitz structure and thus can be solved efficiently by the Wax-Kailath algorithm. Simulation results demonstrate the performance of the proposed methods