Spectrum Sensing Exploiting Guard Bands and Weak Channels

We address the problem of primary user detection in Cognitive Radio from a wideband signal comprising multiple primary channels, exploiting a priori knowledge about the primary network: channelization and spectral shape of primary transmissions. Using this second-order statistical information, a multichannel Gaussian model is formulated. In order to obtain a generalized likelihood ratio test, we first address maximum likelihood (ML) estimation of the power levels at the different channels, as well as of the noise variance. The ML conditions suggest a suboptimal closed-form estimate, which takes the form of a constrained least squares estimator whose asymptotic efficiency is shown for flat bandpass spectra in white noise, a case of practical importance. The resulting detectors exploit those frequency bins corresponding to guard bands and to primary channels perceived as weak to improve noise variance estimation. Analytical expressions for the probabilities of detection and false alarm are presented. Performance is evaluated via simulations in the setting of a terrestrial TV primary network with realistic channelization parameters.