Experimental assessment of eigenvalue-based detection for cognitive radio

This paper presents an experimentally-driven assessment of eigenvalue-based detection (EBD) for cognitive radio (CR), which puts such techniques to a realistic test beyond the assumptions of Gaussian target signals and the availability of large numbers of samples for detection purposes typically adopted in purely theoretical analysis. It is shown that these approximations lead to significantly optimistic results compared to those obtained in laboratory employing standard-compliant signals. Instead of relying on asymptotically large sample sizes, it is also found that better performance is obtained by splitting samples into smaller blocks (multiple covariance matrices), followed by an adequate combining of results based on corresponding test statistics. This approach, referred to as EBD with partial decision combining, is shown to outperform the standard (single-matrix) approach.