Convex Combination of Affine Projection and Error Coded Least Mean Square Algorithms

Abstract

Affine Projection (AP) algorithms offer a relatively good convergence speed which can be increased by augmenting the projection order (L), however, in addition to presenting a high computational complexity, their steady-state misadjustment worsens in direct ratio to the rise of L. Convex combinations of AP algorithms have been devised in an attempt to address the misadjustment issue, albeit at the cost of doubling the aforementioned computational complexity. This work introduces the convex combination of an AP algorithm with an Error Coded Least Mean Square (ECLMS) algorithm, in order to reduce the twofold increase in computational complexity of dual AP combinations while retaining the high convergence speed and improving the steady-state misadjustment level. The proposed algorithm was tested in a system identification application, results demonstrate that the proposal performs as good or better than dual AP solutions, while considerably reducing computational complexity.