Adaptive Spectrum Sensing and Management in Cognitive Radio Networks Using Federated Deep Reinforcement Learning

The dynamic and unexpected character of settings in wireless communication calls for sophisticated spectrum sensing techniques for cognitive radio networks. Building on the work of earlier ensemble machine learning approaches, this study presents a state-of-the-art framework for real-time spectrum management using federated deep reinforcement learning (FDRL). The combination of reinforcement learning's strategic decision-making process with Deep Belief Networks' (DBN) and Long Short-Term Memory's (LSTM) architectures is at the heart of this methodology. This approach, which operates inside a federated learning paradigm, gives user privacy and data locality, guaranteeing a reliable and private solution. Through processing signal vectors under different noise situations, the FDRL model repeatedly learns the best spectrum allocation strategies, improving its comprehension over time. This novel approach offers effective adaptability to the ever-changing wireless environment, improving network speed and spectrum utilization while protecting user privacy. Effectively separating idle from active channels, it continuously adjusts to variations in signal-to-noise ratios and user demands. This sophisticated technology is shown through thorough simulations to provide a significant improvement in both spectrum efficiency and user throughput. Because of its scalability and decentralization, it presents a viable answer to the changing wireless network environment, which is marked by an increasing need for autonomy and data-driven operations. This approach's proven ability to reduce interference and improve service quality indicates a major step forward for intelligent and autonomous spectrum sensing methods, which are critical in the age of ubiquitous wireless communication. The suggested FDRL-DBN-LSTM approach was implemented in Python and achieves an accuracy of 98.4%.