FedBeam: Federated learning based privacy preserved localization for mass-Beamforming in 5GB

Abstract

The overall enhancement of 5G and beyond (5GB) communication accelerates the rise of humongous devices/user equipment’s (UE’s) per-unit area. Massive MIMO (mMIMO) beamforming generates highly directed beams to serve massive UE’s in any area. In dense areas, generating closely distant beams require accurate localization of UEs. Ultra-accurate localization is demanded by implementing the directional beams since even a slight deviation in location leads to significant data loss. With such escalating device density and massive resource demands, the formation of multiple directional beams causes harmful radiation and colossal interference. To optimize beam allocation, a novel idea of mass-beamforming is introduced where a group of users with similar resource demands are served through a single beam. The centroid of massive UE’s in any indoor location is used to create a beam towards a user group. Also, it is essential to maintain users’ location and data privacy. Therefore, this paper proposes a privacy-preserving federated learning-based localization framework, FedBeam, for mass-beamforming in 5GB communication. FedBeam utilizes a deep learning model to acquire precise position location while preserving users’ data privacy. A localization-specific mass-beamforming dataset is modelled to evaluate the proposed framework. The simulation was conducted to validate the accuracy achieved by the proposed framework.