System identification using a pair of adaptive linear phase filters

A novel adaptive filter structure for system identification is presented. The unknown plant is modelled by two linear phase filters, one antisymmetric and the other symmetric, and the two filters are connected in parallel. By adjusting the filter weights simultaneously using Widrow´s LMS (least mean square) algorithm, the transfer function of the plant can be derived. Theoretical analysis has shown that the eigenvalue spread of the correlation matrix of the effective input sequence of the two adaptive filters is usually smaller than that of the simple transversal structure, which thus provides a faster adaptation speed. Indeed, when the step sizes of the two adaptive filters are chosen properly, the convergence characteristics of the system could be improved considerably. Simulation results are included to evaluate the performance of the proposed configuration