快速单边补偿逼近幂迭代子空间跟踪

Fast Single Compensation Approximated Power Iteration Subspace Tracking Algorithm

针对子空间估计与跟踪问题,在快速逼近幂迭代子空间跟踪算法的基础上,得到了一种计算复杂度更低的快速子空间跟踪算法——快速单边补偿逼近幂迭代子空间跟踪算法.该算法采用与Badeau等人不同的逼近方式,简化了互相关矩阵的迭代计算,从而达到降低计算复杂度的目的.实验结果表明:该算法的最大主角性能远远优于PAST子空间跟踪算法,与FAPI和OPAST子空间跟踪算法相当;在正交误差度方面,其正交误差度优于FAPI子空间跟踪算法,远远优于PAST子空间跟踪算法.

For the subspace estimating and tracking, based on the well-known fast approximated power iteration （FAPI） subspace tracking algorithm, a new fast subspace tracking algorithm with lower computational complexity was proposed, that is fast single compensation approximated power iteration （FSCAPI） subspace tracking algorithm. The algorithm used different approximations with Badeau et al, simplifying the iterative calculation of the correlation matrix, so as to achieve the purpose to reduce the computational complexity. The experimental results show that the maximum principal angle performance of the algorithm is far superior to the PAST subspace tracking algorithm and performances fairly the same with FAPI and OPAST subspace tracking algorithms; the deviation from orthonormality of the algorithm is better than the FAPI subspace tracking algorithm and far better than the PAST subspace tracking algorithm.