Abstract: This paper proposes two improved affine projection sign subband adaptive filters (APSSAF) based on the multiband structure, which called variable regularization parameter APSSAF (VRPAPSSAF) and proportionate APSSAF (PAPSSAF). For sparse system identification, two APSSAFs are designed. On the one hand, VRPAPSSAF is presented to get a tradeoff between fast convergence rate and low steadystate misalignment by minimizing the meansquare deviation (MSD) of the system weight vector. In addition, the update of regularization parameter is achieved by the normalized stochastic gradient of MSD. On the other hand, PAPSSAF incorporates a gain distribution matrix into the APSSAF to proportionately adapt the tapweight vector of adaptive filter, which can both utilize the sparsity of system and raise convergence performance for sparse system identification. Simulation results show that the proposed VRPAPSSAF and PAPSSAF used in general system identification and echo cancellation, not only maintain the robustness against impulsive noise and have a good result for tracking capacity, but also have an improved performance in convergence rate and steadystate misalignment in their proper context.