A Comparative Survey of Convex Combination of Adaptive Filters

Over the last twenty years, combination of adaptive filters has emerged to provide a potential solution in the development of advanced applications, such as channel equalization, active noise control, acoustic echo cancelation, adaptive beamforming, among others. Several authors have demonstrated that the combination of two filters with complementary capabilities improves the overall filter performance in comparison when a single filter is used. Commonly, these combinations employ a fast filter and slow filter to guarantee fast convergence speed and low steady-state mean square error (MSE), respectively, at the cost of increasing their computational complexity. However, those combinations have not been well analysed and compared. Currently, this aspect is crucial since most of the current convex combination of adaptive filters could be implemented in advanced embedded digital devices to be used in real-time signal processing applications. In this work, we evaluate the convex combination of adaptive filters, mainly including those based on least mean square (LMS) algorithm, the affine projection (AP) algorithm and the recursive least mean square (RLS) algorithm. We carry out an extensive performance evaluation to demonstrate the advantages and disadvantages of each approach. Specifically, we simulate them in fixed point, which makes feasible the rapid development of advanced prototypes. This aspect could be potentially valuable to the signal processing engineering communities.