Learning-Based Spectrum Sensing in OFDM Cognitive Radios
In this paper, spectrum sensing in OFDM-based cognitive radio systems is modeled as a pattern recognition problem. The proposed scheme uses a linear classifier to decide on when the spectrum is busy (class 1) or not busy (class 2). Two types of feature vectors are compared in this work, namely energy estimates and cross-correlation estimates using the cyclic prefix of the OFDM signal. Simulation results indicate that the energy-based linear classifier provides excellent performance in terms of detection probability over AWGN channels but suffers significant degradation if the channel undergoes flat Rayleigh fading conditions.