Real-valued parallel module decomposition of maximally decimated block filter bank
Ikuya Murakami · IEEE Transactions on Signal Processing · 2000
This study discusses real-value parallel decomposition of M-channel maximally decimated filter banks and cosine-modulated filter banks. The input signal and filter lengths are assumed to be an even multiple of M. Using real-value fast transforms, a filter bank is decomposed into a parallel set of module filter banks of size M independent of the filter lengths. This decomposition is applied to a cosine-modulated filter bank. The decomposition is advantageous for achieving fast module hardware implementation because the module filter banks and fast transforms can be computed in parallel, and the size of the module filter bank is fitted as long as the channel number does not change. In addition to the computational advantage, the perfect reconstruction condition is less restrictive than that of conventional cosine-modulated filter banks. This property makes it possible to use standard methods for a prototype filter design and improve the stopband attenuation level for a given filter length.