Transient performance of an incremental combination of LMS filters

Incremental combinations were first introduced as a solution to the convergence stagnation issue of parallel-independent combinations, both convex and affine. Since then, this topology has been shown to enhance performance to the point of allowing a combination of LMS filters to outperform the APA with lower complexity. In order to better understand and improve this structure, the present work develops mean and mean square transient models for the incremental combination of two LMS filters, that is shown to be a generalization of the data reuse LMS (DR-LMS). By formulating the optimal supervisor and deriving its constraints, the previously proposed adaptive combiner is redesigned to improve the combination's overall performance, as shown by simulations and analysis.