Next-Level Connectivity: Embedding Communication Coding for Efficient Beyond 5G Wireless Federated Learning
Vijay Ganesh, D. Ashwanth Raj, R. Venkata Krishnan · 2024
Wirelessly Federated Learning (WFL) holds immense promise for collaborative knowledge-sharing among distributed edge devices in the Internet of Things (IoT). However, challenges arise in unstable wireless channel conditions, where conventional WFL struggles due to transmission noise and interference. This results in inaccurate model parameters and potential model loss. Additionally, the limitations of CPU-accelerated edge devices in encoding and decoding processes hinder efficiency. To address these issues, we propose a novel software-defined architecture integrating LDPC communication coding, empowered by GPU-CPU integration. Our approach enhances resilience to disruptions by implementing wireless channel coding in both server-side weight aggregation and client-side local training. The GPU-CPU acceleration significantly boosts computing efficiency. Experimental results showcase a remarkable 100x speedup and a 10x improvement in error reduction compared to state-of-the-art WFL schemes. Our method presents a promising solution for robust and efficient WFL in IoT environments, with broad implications for applications relying on collaborative learning among distributed edge devices.