A novel structure for implementing DFT-filter banks

The paper develops a concept of sharing multipliers for the realization of FIR filters. Considering the realization in the transposed direct form, the complete DFT-filter bank can be realized using the same number of multipliers required to implement a single channel. If a multiple of 4 is chosen for the filter length N, then the implementation of the filter bank requires only (N/4-1) real multiplications per sample. For small N, e.g. N=16, the realization in the direct form leads to DFT structures which require a smaller number of multiplications than the corresponding FFT structures. When N is a power of 2, the novel structures are shown to be preferable, if the required number of frequency components is less than((16/3)\log_{2}N -10).