Power-of-two adaptive filters using tabu search

In a digital filter, when the coefficients assume power-of-two or a sum of power-of-two terms, the filter multiplier can be build using simple shift registers. In fact these filters are used in order to reduce both the VLSI silicon area and the computation time. In this paper power-of-two filters are extended to the case of adaptive filters. Due to the quantization and the non-uniform distribution in the coefficients domain, a classical gradient based approach, like LMS, cannot be successfully applied to perform the adaptation process. To overcome the problem of the filter adaptation a specific tabu search for time varying problems, suitable for real-time signal processing, is presented. Several experimental results demonstrate the effectiveness of the proposed method