Optimal Spectrum Sensing Framework for Cognitive Radio Networks Using Attention-Based Autoencoder With Multi-Scale Capsule Network

Wireless communication industry's explosive growth over the last 10 years caused a shortage of resources since its demand has
greatly increased. A technology called cognitive radio (CR) was created to make efficient utilization of the spectrum from radio
sources. The effectiveness of CR is significantly influenced by the spectrum sensing (SS) function, and it is the primary function
of CR, which helps to discover available spectrum for better spectrum utilization and reduce detrimental conflict with approved
users. In addition, the conventional models cause high computational complexity in the SS. In this work, an adaptive SS system
for CR networks is developed to identify unused bands of frequencies in order to get outside of these constraints. Initially, the
essential synthetic data are gathered manually, and these data are used by the suggested adaptive and residual hybrid network
(A-RHN) for SS. The A-RHN network is developed by using an attention-based Autoencoder with a multi-scale capsule network (AA-MCN). Moreover, the effectiveness of this model is enhanced by optimizing parameters via the revised uniform variable￾based addax optimization algorithm (RUV-AOA). The proposed model enables more efficient use of the available spectrum by
avoiding transmitting on frequencies that are already in use.