基于可变数据重用因子的变步长仿射投影算法

A variable step-size affine projection algorithm based on a variable data-reuse factor

提出的仿射投影算法可根据输入数据矩阵条件数的变化情况,实时地决定下一步输入矩阵的维数,即数据重用因子和步长,并利用这两个动态量去折中收敛速度和稳态误差之间的矛盾.仿真结果表明,在较大程度节省了运算量的情况下,该算法的收敛速度接近于使用高维输入矩阵的情况,而稳态误差与使用低维输入矩阵时相当,同时收敛稳定性更优.

The dimension of a next input signal matrix can be determined by the variation of the condition number between current and previous steps, which are the data-reuse factor and step-size. Then the conflict between convergence rate and stable-state error was settled by these two dynamic factors. Simulation results show that under the presupposition of considerable computation reduction, the convergence rate of this algorithm is close to the high dimension situation, and the stable-state error is nearly the same as the low dimension condition. The convergence stability is also improved.