Adaptive filters have been widely used in system identification

echo cancellation and channel equalization. The sign subband adaptive filter (SSAF) is robust against impulsive noise. However

when the input is corrupted by noise

using the SSAF to estimate the coefficients of the unknown system will result in estimation bias. To address this problem

this paper proposes a bias-compensated SSAF (BC-SSAF) based on an unbiased-estimation criterion. To overcome the problem of tradeoff between convergence rate and steady-state misalignment existing in fixed step-size adaptive filters

this paper uses the method of stochastic gradient to update regularization parameter and thus presents a variable regularization BC-SSAF (VR-BC-SSAF). Simulation results verify the superiority of the BC-SSAF and VR-BC-SSAF.