Estimating the Number of Signal Sources in Array Antenna Radar by an Empirical Threshold

For interference suppression and DOA (Direction of Arrival) estimation in radar it is necessary to know the number of signal eigenvalues, here called number of signal sources, of the antenna array covariance matrix. These eigenvalues are the largest ones and correspond to signals sources such as targets, clutter and jamming. This paper suggests to use the largest eigenvalue of a receiver noise-only spatial covariance matrix, estimated from measured data, as a threshold to separate signal and noise eigenvalues and thereby estimate the number of signal sources. This method, here called MEEV (Maximum Empirical EigenValue), was compared with six estimation methods from the literature, with or without prewhitening of the correlated noise, on simulated and measured data. Conclusions of this paper are: It is important to test on measured data. Simulated data are often too idealistic. It matters how the prewhitening is performed. On our measured data, MEEV always gives good performance and is robust and is the preferred method of the tested methods. MEEV also works when the number of data snapshots is less than the size of the covariance matrix.