Latency optimization for Federated Learning over Wireless Power Transfer

作者：Li, Ruijie; Xu, Hongbo; Zhang, Guoping^*; Chen, Yun; Chen, Xue

来源：Physical Communication, 2022, 53: 101765.

DOI：10.1016/j.phycom.2022.101765

• Federated Learning
• Wireless Power Transfer
• Data security
• Optimization

Federal Learning (FL) is an emerging technology in the field of machine learning (ML). Compared with traditional ML, FL is an attractive method to deal with data security issues of the user-side. So that FL can realizes its full potential in terms of low latency and high energy efficiency (EE), this paper introduces a new framework: In the wireless communication network scenario, we propose an FL architecture based on Wireless Power Transfer (WPT). By combining WPT technology and FL, we can realize green wireless communication under the premise of ensuring user privacy. We formulate a joint calculation and communication optimization problem to optimize the latency of local calculation, uplink and downlink transmission without consuming user-side energy. The problem formulas listed according to the optimization problem are non-convex. They are first transformed into convex form, and then a low-complexity iterative algorithm is used to solve them optimally. Simulations show that our proposed FL method design has achieved a significant performance improvement over other benchmark tests.