AI-driven spectrum sensing: An in-depth meta-analysis of trends, challenges and opportunities

• This study explores how AI-driven approaches, particularly ML and DL, are transforming spectrum sensing in 5G and 6G networks, enabling more efficient and intelligent resource management. • A deep dive into AI methodologies for spectrum sensing, categorizing their applications, strengths, contributions and limitations to adaptive and intelligent spectrum utilization; as well as providing a taxonomy based on the findings. • The findings underscore the urgent need for energy-efficient, scalable AI models capable of handling real-time processing and dynamic network conditions in future wireless communication systems. Artificial Intelligence (AI) is playing a crucial role in transforming Spectrum Sensing (SS) and Cognitive Radio Networks (CRNs), especially for next-generation wireless communication systems. This study presents a meta-analysis of 13 survey articles, also analyzing a total of 113 primary studies, to synthesize the applications of AI, specifically Machine Learning (ML) and Deep Learning (DL), in spectrum sensing. Key models identified include Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Graph Neural Networks (GNN), among others. The analysis reveals measurable performance improvements and the main metrics to measure it. Despite these advancements, challenges persist, including computational complexity, adaptability to real-time environments, and model generalization. The study also highlights promising future directions like energy-efficient AI architectures, federated learning for decentralized CRNs, and cooperative spectrum sensing methods. Addressing these challenges and pursuing open research areas is critical to fully realize AI-powered CRNs. Such progress is expected to enable autonomous and intelligent spectrum management in beyond-5G and 6G networks, ultimately enhancing system reliability, scalability, and spectrum utilization efficiency.