The transmission nature of Primary User (PU) is often unpredictable, which leaves certain frequency bands or geographic regions temporarily unoccupied—these underutilized portions, known as “spectrum holes,” represent the opportunity for dynamic spectrum access. Accurate Spectrum sensing is crucial that enables the cognitive radio (CR) system to optimize spectrum utilization efficiently. On the other hand, conventional sensing approaches face performance degradation, in low signal-to-noise ratio (SNR) conditions, due to high false alarm and low detection accuracy. To overcome these challenges, this study introduces a multi-feature-based hybrid deep learning model named DeepMLLHSNet, designed to enhance reliability, detection accuracy and robustness even under adverse situations. It integrates four distinct features of signal like, I/Q components, cyclostationarity, power spectrum and energy statistics—into a feature matrix facilitating efficient feature learning along with robust classification. For performance estimation, RadioML2016.10b dataset was used in simulation with special focus on low SNR condition (e.g., -20 dB to 0 dB). Proposed framework shows improvement in sensing than both traditional models (CNN, Inception, ResNet, LeNet, LSTM, CLDNN) and recent hybrid approaches (CNN-LSTM, DetectNet, ResNet-LSTM, DeepSenseNet) hence confirming its efficiency in spectrum sensing in challenging condition. DeepMLLHSNet has achieved 98.89% prediction accuracy( P a ) with 97.65% precision and 97.76% recall.
Mondal et al. (Thu,) studied this question.