Edge-Aware Federated Learning-Based Channel Equalization for Robust Communication in 6G-Enabled IoT Networks

Abstract

The possibilities generated by the emergence of 6G communication systems will include ultra-reliable low latency communication (URLLC), huge connectivity of devices, and high data rates that are key ingredients to the realisation of the next-generation Internet of Things (IoT) ecosystem. Although it has made such progress, it is not an easy task to provide reliable data transfer at the network edge over time-varying multipath wireless channels. This paper proposes Edge-Aware Federated Learning (EA-FL) framework of distributed channel equalization adaptable to needs and constraints of the 6G-enabled IoT networks. Proposed design uses federated learning to locally, on edge devices, train neural equalizers using locally-stored data, and thus keep the data privacy without sacrificing training efficiency due to a personalized adaption to channel impairments. The framework combines edge-specific tasks, such as client selection based on local signal quality, communication-efficient model updates and adaptive learning schedules. Over a large number of simulations on Rayleigh fading channels with QPSK and 16-QAM systems, EA-FL equalizer has better performance at symbol error rate (SER) and training convergence as well as robustness compared to centralized deep learning and conventional adaptive equalizers. The findings indicate the usefulness of edge-aware intelligence in ensuring quality communication in non-stationary contexts that support the use of EA-FL as a scalable and privacy-preserving means of real-time signal processing in ultra-dense 6G IoT networks.