Neuroevolutionary Optimization of Shannon’s Capacity in Edge AI Protocols for M2M Communication

The efficient operation of edge devices in Machine-to-Machine (M2M) networks hinges on their ability to communicate reliably under highly variable conditions. However, conventional application-layer protocols such as CoAP, MQTT, AMQP, and HTTPS often fail in constrained environments due to static configurations and lack of adaptivity. Here, we present a Neuroevolutionary Algorithm (NEA) framework designed to autonomously optimize the configuration of these edge AI protocols to approach Shannon’s capacity. The framework employs a Genetic Algorithm (GA) based Neural Architecture Search (NAS) to evolve Artificial Neural Networks (ANNs) that predict protocol performance and tune key parameters accordingly. We evaluated this system across edge network using temperature sensor data transmitted from the Raspberry Pi-based edge device to the laptop using a simulated approach. Our results demonstrate substantial improvements in throughput, protocol efficiency, and adaptivity over baseline configurations under fluctuating Signal quality and constrained bandwidth. This study illustrates the potential of neuroevolution as a strategy to enable self-optimization and communication-efficient edge protocols.