Covariance Matrix Estimation for Radar Applications

In a typical radar system, the power of the useful signal component is tipically lower than the competing disturbance strength; consequently, radar detection becomes a very challenging problem. The SINR (Signal to Interference plus Noise Ratio) is generally the most critical figure of merit when designing a radar filter. The processor that maximizes the output SINR is a coherent, linear, transversal filter, based on the exact knowledge of the true disturbance covariance matrix. However, in real radar systems, this requirement cannot be satisfied and an estimate of the covariance matrix is adopted instead of the exact one, leading to the so-called adaptive radars. The aim of this thesis is the introduction of innovative covariance matrix estimation techniques, operating in different conditions. In particular, a covariance matrix estimator, based on statistical argumentations, is presented when homogeneous secondary data are available. Moreover, exploiting geometric considerations, two family of covariance estimators are defined and adopted for training data selection, which is useful when some outliers affect the secondary data. Finally, a family of radar receivers for extended targets in range, enforcing several structures over the disturbance covariance, is described, which is effective when it is not possible to identify secondary data free of targets.