Deep Learning With Harmonic Elk Herd Optimization for Spectrum Sensing With Cyclostationary in Cognitive Radio Network

In Cognitive Radio (CR), effective spectrum utilization is regarded as vital to enhance the spectrum efficacy and to accommodate the need for wireless communication services. Spectrum sensing is more essential in CR networks for permitting spectrum prospects without any harmfulness to Primary Users (PUs). However, existing spectrum sensing approaches depend on energy detection, which leads to various disadvantages, like noise sensitivity, ambiguity in detecting weak signals, and fluctuation in background noises. Hence, this paper introduces a new technique termed Harmonic Elk Herd Optimization (HEHO)-PyramidNet+ Kernel Least Mean Square (HEHO-PyramidNet+KLMS) for spectrum sensing in CR networks. First, the signal is collected from the simulated CR system network. Next, the cyclic spectrum is extracted, then the spectrum sensing is carried out by Kernel Least Mean Square (KLMS) filter. On the other hand, the extracted cyclic spectrum is subjected to spectrum sensing, which is performed using PyramidNet. Here, the PyramidNet is tuned using Harmonic Elk Herd Optimizer (HEHO). Afterwards, the attained spectrum sensing outcomes are integrated using the Average Fusion approach. The HEHO-PyramidNet + KLMS measured a maximum probability of detection, throughput, and energy efficiency of 0.919, 91.77 Mbps, and 94.88 bits/J, and a minimum probability of false alarm of 0.089 and a detection time of 21.54 ms.