Signal, Image and Video Processing Обработка сигналов
Robust widely-linear recursive least M-estimate algorithm: design, analysis and echo cancellation application
Аннотация
Adaptive filters based on widely-linear estimation can fully exploit the second-order statistical information of noncircular signals to achieve superior filtering performance, and thus have been extensively investigated in the field of complex-valued adaptive signal processing. The widely-linear recursive least squares (WL-RLS) algorithm adopts the forgetting mechanism to accelerate convergence significantly, yet at the expense of high computational complexity. Furthermore, the WL-RLS algorithm fails to converge stably under impulsive noise interference. To tackle these drawbacks, this paper proposes a robust widely-linear recursive least M-estimate (WL-RLM) algorithm based on the modified Huber function, as well as its low-complexity variant based on dichotomous coordinate descent, termed DCD-WL-RLM. Subsequently, the mean square deviation (MSD) performance of the WL-RLM algorithm is analyzed, and the theoretical expression for steady-state MSD is derived. Simulation experiments involving system identification and stereophonic acoustic echo cancellation (SAEC) verify the estimation superiority of the proposed algorithms over comparison algorithms, together with the accuracy of the steady-state MSD theoretical analysis.