Spectrum sensing framework for wireless information transmission with a novel LCGT based DL approach for cognitive radio sensor networks
E. Vargil Vijay, K Aparna, K M V Madan Kumar · Engineering Research Express · 2025
Abstract Spectrum sensing is essential for identifying radio frequency bands available for interference-free communication in today’s cognitive radio systems context for providing efficient spectrum management. It is very important to ensure seamless wireless device operation in the dynamic and congested spectrum environments. In this paper, we propose a novel deep learning (DL) based LCGT (LMS-CNN-GRU-TN) approach combining a Least Mean Squares (LMS) equalizer, Separable Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and Transformer Network (TN), leading to an improved spectrum sensing scheme. LMS equalization mitigates the distortions induced by the channel to improve performance. The temporal signal features are extracted by the separable CNN, GRU learns the sequential dependency from the signal pattern, while the TN utilizes attention mechanisms to model the complex signal dynamics. The proposed integrated architecture is robust, adaptable, and performs better with regard to spectrum sensing in noisy environments. Results from the proposed method show improvements in probability of detection, false alarm, sensing error, F1 score, and Cohen’s Kappa Coefficient (CKC), indicating that proposed method can effectively analyse and sense the spectrum in wireless sensor networks (WSNs).