Home // AICT 2023, The Nineteenth Advanced International Conference on Telecommunications // View article

Federated Learning for Distributed Sensing-aided Beam Prediction in 5G Networks

Authors:
Adwitiya Pratap Singh
Chitwan Arora
Abheek Saha

Keywords: Wireless Technology; Artificial Intelligence; Deep Learning; Federated learning.

Abstract:
The increasing demands for higher data rates have caused newer communication systems to move towards higher frequency bands. However, during the initial network access, the user faces a problem of high beam selection, due to the rich scattering environment and the large number of possible beams. For high mobility and low latency applications, such as vehicular communications, high beam selection overhead is a very big problem. Sensing-aided beam prediction using environmental sensing information as well as telemetry data can be a possible solution to this issue. In this paper, a novel approach is suggested that combines real-time series Global Positioning System (GPS) data, as well as terrain related data for beam selection. Using the DeepSense dataset, we demonstrate that distributed machine learning algorithms, while being computationally tractable, can choose the top N beams with an accuracy that is comparable to that of centralized learning, but faster than it. The novelty of our work lies in the usage of this data set to simulate federated learning and trying different techniques to increase accuracy.

Pages: 8 to 13

Copyright: Copyright (c) IARIA, 2023

Publication date: June 26, 2023

Published in: conference

ISSN: 2308-4030

ISBN: 978-1-68558-068-1

Location: Nice, France

Dates: from June 26, 2023 to June 30, 2023