GCN-Integrated Multi-Agent Reinforcement Learning (MARL) with Modified Firefly Algorithm for Cross-Layer Optimization in Cognitive Sensor Networks

CSN s are integrating cognitive abilities, making it challenging to achieve optimized cross-layer performance with limited resources. Non-coordinated techniques typically demonstrate lower proficiency in dynamic spectrum management and network synchronization. We introduce Graph Convolutional Networks (GCN) and Multi-Agent Reinforcement Learning (MARL) in our architectural solution, with the help of the specially designed Firefly Algorithm. The framework implements a distributed learning process, utilizing GCN for local topology perception and MARL for cooperative spectrum and optimal routing decisions. The results of experimental analysis show large gains compared to traditional methods. The proposed model provided a 94.2% recognition rate in spectrum hole identification, an 89.7% enhancement in network throughput, and a 78.5% diminution in end-to-end delay. The energy efficiency surpasses baseline techniques by 67.3%, while the raw packet delivery reliability remains at 92.8%. Integrating GCN with MARL and the improved Firefly Algorithm can significantly improve cross-layer optimization, offering a promising solution for the next generation of sensor networks.