Modification Algorithms for a Class of Subspace Tracking Methods

Subspace analysis is a frequently encountered method in many signal processing fields. Computational complexity is a very important point for real time implementation. Starting from analysis of DPM and OJA classes of subspace tracking methods, we proposed modified methods for them, named as MFDPM and MFOOJA, which are numerically stable and less computational complexity. Simulation results verified that the proposed algorithms have similar convergence speed and steadystate error compared to FDPM or FOOJA. And the proposed algorithms are less sensitive to roundoff error accumulation under the finite word length condition and guarantee orthonormality convergence, so which are robust.