Signal detection based on linear trend in the noise-originated eigenvalues of the correlation matrix for cognitive radio

Cognitive Radio introduces the opportunistic usage of unoccupied frequency bands by unlicensed users without disturbing the Quality-of-Service of the licensed users. The first problem to be solved by a Cognitive Radio terminal is sensing the signals of the licensed users. In our study, we offer a new sensing method based on the eigenvalue decomposition of the covariance matrix of received signals. The method defines a test statistic on the distribution of eigenvalues to give a “Yes or No” decision about the signals in the band-of-interest. The main idea behind the method is that eigenvalues belonging to the noise subspace of the covariance matrix tend to exhibit a linear trend of decrease in case of limited number of samples of the received signals. The proposed method presents a realistic approach, and due to its robustness to sensing without suffering from noise power uncertainty, it has advantages against the energy level comparison based detectors that are highly influenced by the noise.