用修正数据映射方法实现快速稳健子空间跟踪

为改善现有子空间跟踪方法的运算量大或鲁棒性/稳定性差等缺点,基于原数据映射方法(DPM),采用新的正交归一化,提出了一种快速稳健的修正DPM(MDPM)算法。通过改变符号选择跟踪信号子空间或噪声子空间,同时提供了子空间基的正交矢量,运行时无累积性误差。研究和仿真结果表明:与传统子空间跟踪算法相比,MDPM算法的鲁棒性/稳定性好、收敛速度快、运算量小。

Fast and Robust Subspace Tracking Using Modified Data Projection

Aiming at the high computational complexity and nonrobust/instability of the conventional subspace tracking, a new fast and robust modified data projection method （MDPM） algorithm that using orthonormalization step was put forward based on data projection method （DPM） in this paper. The switch form noise subspace to signal subspace estimation could be realized by changing the sign of step size, which would provide orthonormal vector estimates of subspace basis without round-off error accumulation. The research and simulation results showed that the MDPM method had better robust/stability with lower computational cost than the conventional subspace tracking method.