Cyclostationarity-based versus eigenvalues-based algorithms for spectrum sensing in cognitive radio systems: Experimental evaluation using GNU radio and USRP

Spectrum sensing is a fundamental problem in cognitive radio systems. Its main objective is to reliably detect signals from licensed primary users to avoid harmful interference. As a first step toward building a large-scale cognitive radio network testbed, we propose to investigate experimentally the performance of three blind spectrum sensing algorithms. Using random matrix theory to the covariance matrix of signals received at the secondary users, the first two sensing algorithms base their decision statistics on the maximum to minimum eigenvalue ratio and the sum of the eigenvalues to minimum eigenvalue ratio, respectively. However, the third algorithm is based on cyclostationary feature detection and it uses the symmetry property of cyclic autocorrelation function as a decision policy. These spectrum sensing algorithms are blind in the sense that no knowledge of the received signals is available. Moreover, they are robust against noise uncertainty. In this paper, we implement spectrum sensing in real environment and the performance of these three algorithms is conducted using the GNU-Radio framework and the universal software radio peripheral (USRP) platforms. The results of the evaluation reveal that cyclostationary feature detector is effective in finite sample-size settings, and the gain in terms of the SNR with respect to eigenvalues-based detectors to achieve Pfa (probability of false alarm) = 0.08 is at least 4 dB.