A new information-theoretic criterion for detection of the number of signals in spatially correlated noise with unknown covariance matrix

Two new information theoretic criteria for detecting the number of signals in spatially correlated noise with unknown covariance matrix are presented. The new criteria take a very simple form involving the canonical correlation coefficients of sensor data, and are applicable to the cases either two sensor arrays or a single array are available. In both cases, the new schemes are significantly superior to the conventional information criteria.<>