Wideband Subspace Estimation Through Projection Matrix Smoothing
J. Selva · arXiv (Cornell University) · 2017
This paper presents a method for improving the performance of wideband direction-of-arrival (DOA) subspace estimators. The method exploits the fact that the signal subspace varies smoothly along the spectrum to improve the estimation of this subspace and, in turn, the DOA estimates that may be subsequently computed. In an initial step, it computes the sample covariance matrix at a set of frequency bins and obtains from them the corresponding signal projection matrices. It then smooths this last set by means of a least squares fitting to a low-order polynomial and, finally, yields a small set of signal projection matrices, that can then be employed by wideband DOA estimators such as Incoherent MUltiple SIgnal Classification (IC-MUSIC) and Test of Orthogonality of Projected Subspaces (TOPS). The method provides a significant improvement in the RMS error performance of these two estimators with a small increase in computational burden. Its performance is assessed in several numerical examples.