On the performance of eigenvalue-based signal detection

Comparing to GLRT which should have some prior information, eigenvalue-based spectrum sensing is a totally blind detection technique without knowing the information of signal, channel and noise power. Existing eigenvalue-based spectrum sensing techniques are based on the approximated methods to determine the threshold. The performance of this approach is satisfactory for multipath channels in low SNR environments, however, at the cost of a long detection latency and high computational complexity due to the fact that a large number of samples are required. To tackle this problem, the thresholds can be calculated by a joint eigenvalue distribution. Unfortunately, this kind of distribution functions hardly meet the requirements of signal detection. In this paper, we explore the usage of an exact probability density function (PDF) of Demmel condition number for complex Wishart matrices. We show that the potential gain obtained by using the proposed function is rather large compared to the traditional approximated methods with smaller size of samples.