A deep learning enabled adaptive threshold selection framework for efficient spectrum sensing and energy optimization in cognitive radio networks

Abstract Spectrum sensing (SS) plays a crucial role in cognitive radio (CR) networks, identifying unused primary user (PU) spectrum for efficient utilisation. SS is measured using sensing parameters, including the probability of detection (Pd), the probability of false alarm (P f ), and the probability of missed detection (P md ). The IEEE 802.22 standard is a CR standard designed to enable opportunistic access to unused licensed TV spectrum. According to the standard, the probability of false alarm should be ≤0.1, and the probability of detection should be ≥ 0.9. The probabilities of detection and false alarm are strongly influenced by the threshold value, and the selection of the threshold plays a crucial role in identifying the presence or absence of the PU. In the proposed method, a deep learning (DL)-based adaptive threshold selection approach is employed for energy detection (ED) in CR networks, using the Bidirectional long short-term memory (BiLSTM) model to achieve P d = 1 and P f = 0. In many available SS approaches, the threshold is determined by fixing P d or P f and optimising the other detection parameter. In the present study, we have considered the optimal threshold at low signal-to-noise ratio (SNR) using BiLSTM, which achieves better sensing performance compared to traditional methods. The proposed framework determines an adaptive threshold for low SNR (−30 dB to −8 dB) and demonstrates notable improvements in throughput (up to 65%) and energy efficiency (up to 70%) compared to traditional threshold-based methods.