A Selective Machine Learning Approach for Double-Threshold Spectrum Sensing

Spectrum sensing based on energy detection is
widely used due to its low complexity, but its performance degrades
under low signal-to-noise ratios and noise uncertainty.
Double-threshold energy detection introduces an uncertainty
region between two thresholds, where reliable decisions are
difficult to obtain using conventional methods. This paper
proposes a selective machine learning assisted spectrum
sensing method in which a simple classifier is applied
only when the sensed energy falls within the uncertainty
region, while energy detection is used for confident decisions.
Simulation results show that the proposed method achieves
detection accuracies between approximately 71% and 99%
under different signal-to-noise ratio and noise uncertainty
conditions. Machine learning is invoked only for a subset
of sensing instances, resulting in low average classification
times below 1 ms.