Uniform DFT Filter Bank with Finite Precision Prototype Filters

Multirate signal processing is important in a wide range of signal processing applications. Drawbacks when using multirate processing are mainly related to aliasing, reconstruction effects and the delays. In this paper, the authors investigate the design of uniform DFT filter bank with finite precision prototype filters and a minimum total number of power-of-two for the coefficients. The design problem is formulated as a mixed integer optimization problem. This problem can then be solved by using e.g. the mean field annealing algorithm. Design examples show that the filter bank can be designed with less total number of power-of-two than the quantized filter bank while achieving approximately the same performance