Knowledge-Aided Parameter Estimation Based on Conjugate Gradient Algorithms

The performance of many parameter estimation algorithms used for direction finding and localization techniques depends on the accuracy of the signal covariance matrix estimate. For a small number of sensors, the commonly used sample covariance matrix estimation procedure may only provide a poor estimate of the unknown true covariance matrix. In scenarios with low signal-to-noise ratio, stationary and non-stationary signal sources, a more accurate estimate of the signal covariance matrix can be achieved by incorporating a priori knowledge about the direction of arrival (DOA) of dominant signals. In this paper, we combine the weighted sample covariance matrix and a weighted knowledge-aided (KA) covariance matrix. We present a KA-Conjugate Gradient (KA-CG) algorithm that processes the enhanced covariance matrix estimate. Simulation results show that the proposed KA-CG algorithm substantially improves the probability of resolution of unknown close sources in the system, especially at middle low signal-to-noise ratios (SNR), requiring a reasonable number of samples for this aim.