Performance Analysis of Applying Federated Learning in Wireless Networks Based on Stochastic Geometry

The future sixth-generation (6G) network architecture will have ubiquitous heterogeneous connectivity and native intelligence to support the increasingly complex and diverse intelligent services. How to deeply integrate and efficiently deploy artificial intelligence (AI) in radio access networks (RANs) has become a key concern in academia and industry. Federated learning (FL) is one of the most promising intelligent collaboration solutions in 6G networks due to its advantages in collaborating distributed data and protecting user privacy. However, when deploying FL in wireless networks, the model transmitted over a wireless channel will inevitably be affected by the interference between different wireless signals, which would result in the performance deterioration of the FL model. This paper aims to analyze the coverage probability of a wireless network using stochastic geometry and investigate the impact of coverage probability on the performance of the FL model. Simulation results show that an optimal coverage probability can be found to achieve the highest FL model accuracy and shortest model delay, which can provide a guidance for the actual deployment of FL on the RAN side.