| Feature-based spectrum sensing techniques have emerged as good balance between energy-based techniques and coherent-based techniques, where the former require minimal prior information of the observed signal, and the latter have robust detection performance when the observed signal is very weak. In this thesis, we focus on pilot tone-aided detection as a feature-based detection class. We propose an improved pilot tone-aided spectrum sensor that utilizes the presence of the pilot tone and the overall energy of the received signal. We show that the optimal Neyman-Pearson detector is a weighted summation of a feature-based component and an energy-based component. The former provides coherent gains at the low signal-to-noise ratio (SNR) regime, whereas the latter provides non-coherent gains at moderate SNR levels. The proposed detector intelligently adapts its weights based on the SNR of the observed signal and the power allocation factor of the pilot tone. This helps it attain significant performance gains compared with the conventional pilot tone-aided detectors.;In addition, we present suboptimal detectors that reduce the computational complexity. For instance, we demonstrate that moment estimators are effective techniques for spectrum sensing. Motivated by insights gained from the derivations of these moment estimators, we present a selective mean-variance estimator that performs well in the absence of the prior knowledge about the pilot tone. |
