Abstract:

To detect the number of signals correctly is the prerequisite for many algorithms with high resolution. The conventional approaches usually focus on eigenvalue of covariance matrix to set the threshold according to some criterion, in order to estimate the number of signals in a white noise background. However, there exists no white noise actually. This paper starts from the perspective of onedimensionreduced projection matrix of eigenspace and obtains the difference sequence of projection of the part covariance matrix, which effectively avoids the impact of the principal noise energy, thus restrains the colored noise in some sense. The simulation results illustrate the new method can estimate the correct number of multisignals in the background of colored noise.