Real-Time Optimization of 5G/6G Signal Transmissions Using Spiking Neural Networks and Asynchronous Federated Learning

This paper introduces Time-Introduced baCKpropagation Through Optimized Cooperation (TICKTOC), a real-time approach to improve signal transmission that integrates spiking neural networks (SNNs) with Cooperative Optimization (CO-OP) from asynchronous federated learning, enabling dynamic adjustments in digital predistortion (DPD). SNNs offer energy-efficient processing suited to the power demands of 5G/6G infrastructure. Federated learning enables faster decentralized updates through inherent parallelism, while CO-OP further reduces latency by using asynchronous aggregation, allowing faster updates and improving real-time performance when accommodating heterogeneous client workloads. Simulation results demonstrate that TICKTOC achieves a 50.0% faster convergence compared to Centralized BPTT and 75.0% compared to FedAvg BPTT, significantly outperforming state-of-the-art SNN methods in real-time DPD applications.