6G-Enabled Mobile Access Point Placement via Dynamic Federated Learning Strategies

Abstract

Advanced Indoor Positioning Systems (IPS) based on Received Signal Strength (RSS) fingerprints have been paramount in 6G network research and commercial exploitation due to their costeffectiveness and simplicity. Despite their popularity, the advent of 6G has prompted a shift towards exploring Deep Learning algorithms to further enhance their performance and precision. Deep Learning research typically demands large datasets, leading to reliance on data augmentation and crowdsourcing techniques for data collection. However, the traditional centralization of data in crowdsourcing poses privacy risks, and here is where Federated Learning (FL) comes into play. In light of this, our study introduces FL to bridge this divide in a decentralized way, eliminating the need for servers to acquire labeled data directly from users. This approach aims to minimize localization error in RSS fingerprints, preserve user privacy, and reduce system latency, all key goals for 6G networks. Moreover, we explore the use of power transmission techniques to further decrease the latency in the FL system. Our simulation outcomes confirm the superiority of FL over traditional Stochastic Gradient Descent (SGD) methods considering critical evaluation metrics like localization error and global loss, paving the way for efficient 6G implementation.