Abstract: Variable step-size segment proportionate affine projection algorithm，namely VSS-SPAPA，incorporates the advantages of fast convergence speed and low mis-adjustment.However，the large number of filter coefficients in echo cancellation applications diminishes the usefulness of this algorithm owing to increased complexity.To deal with this obstacle，a selective partial-update method for VSS-SPAPA is proposed which partitions the filter coefficients into multiple subsets and focuses on updating a selected subset per iteration because the computational complexity is proportional to the number of filter coefficients.Echo cancellation simulation results verify that the proposed algorithm can maintain the convergence and tracking performance close to the full update VSS-SPAPA when its updating counts range from around 50% to 100% of the number of updates in standard algorithm.

Key words: echo cancellation, sparse impulse response, variable step-size, partial update

摘要： 变步长分割式比例仿射投影算法（VSS-SPAPA）具有收敛快、失调小等优点，但运用到有大量抽头系数需要更新的回波抵消器中时，因计算量较大，限制了其应用。为降低其计算复杂度，基于局部迭代思想提出了一种改进算法。该算法将滤波器系数分成多个系数子集，通过仅更新权系数某一子集的方法，减少了每次迭代中需更新的抽头权系数个数，从而减少了VSS-SPAPA的运算量。理论分析和仿真结果表明：新算法每次迭代更新的抽头权系数只需达到全部系数的一半时，收敛速度和稳态误差均与原算法相当，但所提算法较大程度降低了计算复杂度，从而具备更好的实时性。

关键词: 回声消除, 稀疏冲激响应, 变步长, 局部迭代