Machine Learning and Deep Learning Based Spectrum Sensing (Mldls) in Cognitive Radio Networks: A Comprehensive Review of Techniques, Applications and Challenges and Future Directions

Abstract

As wireless communications grow in popularity, there is growing demand for radio spectrum, but limited spectrum resources. Traditional fixed spectrum allocation may lead to underutilized bands when there are no licensed users. The Cognitive Radio (CR) technology is a solution to this issue, since by using CR technology the secondary users can detect and use the unused spectrum without interfering the primary users. Spectrum sensing is thus an important task of Cognitive Radio Networks (CRNs). The conventional Energy Detection, Matched Filter Detection and Cyclostationary Detection are limited in low Signal to Noise Ratio (SNR), noise uncertainty, fading, interference and wideband conditions. In recent times, the field of Machine Learning (ML) and Deep Learning (DL) has seen significant progress, offering intelligent solutions for enhancing sensing accuracy and adaptability. The received signal data can be used to extract useful patterns using ML methods like SVM, KNN, DT, RF and ANN. DL models, such as CNN, RNN, LSTM, CNN-LSTM, transfer learning and Transformer-based models, can automatically extract spatial and temporal signal features. Other technologies, notably cooperative spectrum sensing and reinforcement learning reinforce reliable and adaptive spectrum access. In this paper, the works on spectrum sensing using machine learning and deep learning in the last six years (2020-2025) are reviewed. It includes conventional and intelligent sensing, wideband and cooperative spectrum sensing, reinforcement learning and emerging AI sensing techniques. Some of the main difficulties such as low-SNR sensing, noise uncertainty, limited training data, computational complexity, energy consumption and real-time implementation and model generalization are discussed. New directions like lightweight DL, Transformer-based sensing, Graph Neural Networks, prediction based sensing, Explainable AI, edge intelligence and AI enabled Cognitive Radio IoT are also mentioned. The review shows that there is great potential in the synergy between CR and ML/DL, in order to enable more accurate, adaptive and efficient utilization of the spectrum in future wireless networks.