Adaptive CFAR detection in partially correlated clutter

Abstract

The problem of adaptive constant false alarm rate (CFAR) detection in a spatially correlated background environment is studied. The clutter is modelled spatially as a first-order Markov Gaussian process, whilst the target return is assumed to be Rayleigh envelope distributed. The case where the clutter power is much higher than the thermal noise power is considered and an expression is derived for the actual probability of false alarm of the CA-CFAR detector. In the analysis, the covariance matrix of the total noise (thermal noise plus clutter) is approximated by the covariance matrix of the clutter. It is shown that the CFAR parameters of the CA-CFAR detector are dependent on the clutter covariance matrix and that the achieved probability of false alarm may be degraded up to five orders of magnitude when the degree of correlation of the clutter samples is high, i.e. the threshold derived by the conventional CA-CFAR detector, which assumes independent noise samples, is unnecessarily too high when the clutter returns are correlated. To alleviate this problem, we propose a generalised CA-CFAR (GCA-CFAR) detector that adapts not only to changes in the clutter level but also to changes in the clutter covariance matrix.