Principal and minor subspace computation with applications

Abstract

Algorithms for computing signal subspace frequency or bearing estimates without eigendecomposition were described. Fast algorithms based on the power method were developed to estimate the principal and minor subspaces of the sample correlation matrices. These subspaces were then utilized to develop high-resolution methods such as MUSIC and ESPRIT for sinusoidal frequency and direction of arrival (DOA) problems. A simple squaring procedure was suggested which provides significant computational saving in comparison with exact eigendecomposition methods.