HDAR-CNet: hybrid deep learning based spectrum sensing and resource allocation model for cognitive radio networks

Cognitive Radio Networks require accurate spectrum sensing and resource allocation to efficiently utilize dynamically available spectrum for varying traffic conditions. To address the issues, this research proposed the Hybrid Deep Learning Based Spectrum Sensing and Resource Allocation model for Cognitive Radio Networks (HDAR-CNet) model. Initially, the Multi-size Kernel Recurrent Transformer Network (MKR-TNet) is proposed for spectrum sensing that is designed using Multi-size Kernel Attention and Gated Recurrent Unit (GRU) based feed forward network of the transformer. Then, the resource allocation strategy is employed using the proposed Sinusoidal Swift Flight Optimization based Deep Reinforcement Learning (2SO-DRL) model. In the proposed DRL model reward estimation is employed by considering the throughput, delay, user priority and penalty to achieve QoS-aware spectrum allocation. Here, the sinusoidal chaotic mapping incorporated in the proposed 2SO algorithm to improve the exploration, avoid the premature convergence and enhances global optimal solution acquisition. The experimental analysis of the proposed HDAR-CNet model based on End-to-End Spectrum Utilization Efficiency and Computation Time acquired the values of 98.9% and 5 s respectively. Thus, the proposed model provides the efficient, intelligent and reliable solution for next-generation wireless communication systems and Cognitive Radio Networks.