On Parallel-Incremental Combinations of LMS Filters that Outperform the Affine Projection Algorithm

Data reuse techniques have been successfully applied to improve the performance of adaptive filters, the Affine Projection Algorithm (APA) being one of the most celebrated examplesofsuchachievements.Recently,incrementalcombinations of LMS filters have been shown to either match or outperform theAPAwithlowercomplexity.However,thiscombinationbrings forth a trade-off between convergence speed and steady-state error which depends on the number of component filters. This work proposes a parallel-incremental topology with coefficients feedback to allow transient and steady-state performance to be designed separately. Moreover, to address highly correlated environments, a data reusing incremental combination is employed. An approximate analysis is pursued using the data sharing version of this structure. Numerical experiments show that the novel combination can outperform the convergence rate and misadjustment of the APA and combinations of APAs in different scenarios.