Detection of the number of signals in noise with unknown, non-white covariance matrices

An approach to detecting the number of signals in an unknown noise environment based on applying canonical correlation analysis to the outputs of two spatially separated arrays is presented. The method, called the canonical correlation test (CCT) method, works very well in a wide variety of colored noise situations and gives a quantitatively controllable performance. Compared with other methods, it shows robust performance even when both AIC and MDL fail completely. Thus, it gives a very attractive, simple, statistically rigorous approach.<>