Low complexity and fast system identification using adaptive IIR filter employing subband decomposition

Adaptive infinite impulse response (IIR) filters are particularly beneficial in modeling real systems because they require lower computational complexity and can achieve performance more efficiently as compared to the finite impulse response (FIR) filters. In this paper, an adaptive IIR-RLS filter employing subband decomposition is investigated. The proposed subband IIR-RLS filter is based on oversampled uniform DFT banks. To investigate the performance and the stability of the proposed subband IIR-RLS filter in a practical implementation, real data for far-end speech signal and its echo were used in car environment. The computer simulations are included in order to verify the performance of the algorithm. The proposed subband IIR-RLS filter produces results that are significantly favorable than fullband versions of FIR-RLS and IIR-RLS filters and traditional subband FIR-NLMS filter. Moreover, this approach greatly reduces the computational complexity while the stability of the low-order IIR filters in the proposed subband IIR-RLS filter could be monitored easily. Therefore, the proposed subband system identification technique is very attractive for the practical use with real time implementation.