Learning Strategies for the Interference Covariance Structure Based on a Bayesian Approach

This letter addresses the adaptive classification of the Interference Covariance Matrix (ICM) structures in radar applications. This is an essential issue when the design assumptions do not perfectly match the actual operating scenario due to environment uncertainties. Thus, in this letter, we propose a classifier capable of identifying the ICM structure as either complex Hermitian or real-valued symmetric. To this end, a Bayesian approach is employed by assuming a suitable model for the probability density function of the unknown ICM. This classification problem is firstly formulated in terms of a binary hypothesis test and the posterior probability is maximized to devise the classifier. Furthermore, the classifier resorts to secondary data only which are obtained from the adjacent cells around the cell under test and share the same ICM structure as the primary data. The illustrative examples conducted on simulated data have confirmed the superiority of the proposed classifier compared with its state-of-the-art non-Bayesian counterparts.