Deep Learning-CT based spectrum sensing for cognitive radio for proficient data transmission in Wireless Sensor Networks

Abstract

Spectrum sensing is a vital element of cognitive radio networks. It's the basis for unlicensed users to access underutilized bands without disturbing licensed or primary users (PUs). Deep learning has given a better way to analyze signals compared to energy detection and other conventional methods that were present earlier. This research introduces a CNN-TN-based spectrum approach that integrates both CNNs and transformer networks (TNs) to maximize efficient spectrum utilization. CNNs are good at finding features in signals, which helps them pick out important features from the IQ-based spectrum data. Transformer networks, on the other hand, are able to take into account long-range dependencies over time, which can help improve and fine-tune these features even more. This method achieves low sensing errors and a better probability of detection by capturing both local and global patterns in the spectrum data. Performance measures like Cohen's Kappa coefficient and F1 score show that this method's performance is better than earlier methods when it comes to performance. These findings also suggest that real-world wireless communications systems could potentially use the proposed methodology, thereby improving overall usage and indicating its potential for practical applications in the field.