Robust Augmented-Diffusion Filtered-x Maximum Correntropy Criterion for Distributed Active Noise Control Under Impulsive Noise

Abstract

Distributed multichannel active noise control (ANC) is scalable, but filtered-x least-mean-square updates are highly sensitive to impulsive disturbances. This paper proposes a robust augmented-diffusion filtered-x maximum correntropy criterion algorithm (R-ADFxMCC). It keeps the fixed admissible combination of augmented diffusion and replaces the local squared-error update with a normalized correntropy-gated recursion, whose kernel width follows a median-absolute-deviation estimate of the local residual scale. The resulting influence function is bounded and redescending. Under the stated independence and linearization approximations, the analysis gives a combination-weight-independent mean stability condition and a local steady-state excess mean-square error approximation; a symmetry argument proves zero steady-state mean bias without requiring independence between the gate and the regressors. In 30-realization simulations with one validation protocol for all step sizes, the median steady-state gains over augmented-diffusion FxLMS are 25.79, 21.84, 13.80 and 6.46 dB for single-tone, multitone, narrowband and broadband references under localized Bernoulli–Gaussian impulses. The augmented-diffusion M-estimate algorithm reaches the same steady state within 0.45 dB at the validation-selected step sizes, but on single-tone and narrowband references, R-ADFxMCC remains stable at step sizes at which the M-estimate algorithm diverges. Tests with α-stable noise, reverberation, secondary-path errors and measured per-sample cost are also reported.