A fast transversal filtering algorithm for pole-zero modeling (Q1329430)

The problem that is discussed in the paper concerns poor performance, yielded by auto regressive models in presence of sharp zeros in the transfer function of an unknown system. The idea is to solve this problem using an auto regressive moving-average (ARMA) model in least squares system identification and spectral estimation in speech signal processing. The paper describes a pole zero auto regressive moving- average (ARMA) modelling of discrete time linear system using a recursive least squares (RLS) fast transversal filter (FTF) algorithm. The algorithm is derived using geometric projections. The approach gives base for various filters' variants from the one of algorithm. As a result the algorithm can exactly estimate unknown filter coefficients. Simulation results are presented, which shows the rapid convergence speed and the numerical accuracy of the algorithm. It is shown that this algorithm converges more accurately than the other known ARMA FTF algorithms. This algorithm also requires less computations than RLS lattice filters and other known algorithms of the same type. The proposed approach is of wide applicability. The achieved results show that ARMA FTF algorithm indicates more exactly the unknown system coefficients in a few iteration of the given pole-zero orders than the algorithms of the same type. The modelling results illustrate that proposed algorithm converges to almost the same error energies in spite of using different white noises.