Incorporated robustness in narrow-band signal subspace spatial spectral estimators

Signal subspace spectral estimation algorithms are, by design, highly sensitive to differences between assumed source observation models and actual source observations. In spatial spectral estimation, where observations are derived from different array elements and are a function of various source propagation and observation parameters, accurate source modeling is often not possible. As a result, when incorrect values of model parameters are assumed, signal subspace algorithm performance is degraded. In this paper, an approach for incorporating robustness to assumed model parameter values is investigated. The approach is based on: 1) a representation of rank-1 sources in an low-rank subspace; and 2) a signal subspace projection algorithm suggested by Schmidt. In addition to providing robustness to certain types of parameter variations, the approach provides a computationally efficient way of searching through some multidimensional parameter spaces.