Analysis of subspace fitting based methods for sensor array processing

The problem of estimating signal parameters from sensor array measurements is addressed. A general multidimensional signal subspace method, called the weighted subspace fitting (WSF) method, is proposed. The relationship of WSF to other signal subspace methods as well as the relation to the deterministic maximum-likelihood (ML) method is discussed. The asymptotic properties of WSF are presented for a general weighting. This result includes the properties of ML as a special case. The weighting that minimizes the estimation error covariance is given, resulting in a method that always outperforms ML. A numerical example is presented, demonstrating that the optimally weighted WSF method can give notably lower variance for highly correlated signals. Simulations are included to substantiate the analysis