Optimum low complex tree structured filter banks using overlap-add method

Abstract

Multirate filter banks are bank of filters used for extracting spectral components of input signal. Tree structured filter bank is obtained by cascading two channel filter banks in a uniform manner. This paper proposes an optimum tree structured filter banks using Parks McClellan algorithm which minimizes the error between desired and actual response. Perfect reconstruction of proposed filter bank can be achieved by adjusting low pass filter coefficients using Remez exchange algorithm and finite impulse response filter is used for getting stable and linear output. The computational complexity of filter bank can be reduced by adopting overlap-add convolution at each channel. Proposed method was also compared with Discrete Fourier Transform based filter bank and Cosine modulated filter bank using signal to noise ratio.