Principal and Minor Subspace Tracking: Algorithms & Stability Analysis

We consider the problem of tracking the minor or principal subspace of a positive Hermitian covariance matrix. We first propose a fast and numerically robust implementation of Oja algorithm (FOOja: fast orthogonal Oja). The latter is said fast in the sense that its computational cost is of order O(np) flops per iteration where n is the size of the observation vector and p < n is the number of minor or principal eigenvectors we need to estimate. FOOja guarantees the orthogonality of the weight matrix at each iteration. Moreover, this algorithm is analyzed and compared with two other fast algorithms (OOjaH and FDPM) with respect to their numerical stability. Simulation results highlights the relatively good stability behavior of FOOja

[1]  Y. Hua,et al.  Orthogonal Oja algorithm , 2000, IEEE Signal Processing Letters.

[2]  Gene H. Golub,et al.  Matrix computations , 1983 .

[3]  Joos Vandewalle,et al.  Updating Singular Value Decompositions. A Parallel Implementation. , 1989, Optics & Photonics.

[4]  K. Abed-Meraim,et al.  Subspace estimation and tracking using enhanced versions of Oja's algorithm , 2001, 2001 IEEE Third Workshop on Signal Processing Advances in Wireless Communications (SPAWC'01). Workshop Proceedings (Cat. No.01EX471).

[5]  Qin Lin,et al.  A unified algorithm for principal and minor components extraction , 1998, Neural Networks.

[6]  Erkki Oja,et al.  Principal components, minor components, and linear neural networks , 1992, Neural Networks.

[7]  George V. Moustakides,et al.  The fast Data Projection Method for stable subspace tracking , 2005, 2005 13th European Signal Processing Conference.