A Deep Learning Approach for Channel Estimation in 5G Wireless Communications

Abstract

This paper studies the applications of Deep Learning (branch of Machine Learning) in 5G wireless communication systems. Deep Learning requires having large sets of data for training and testing purposes. Obtaining such large datasets through measurement campaigns is a challenging and costly task. Therefore, it is a common practice to use synthesized data. The aim of the current paper is to survey different applications of deep learning in 5G systems, and more specifically, that implementing the massive multiple input multiple output (mMIMO) using deep learning, for estimating the wideband mMIMO channel, and for beamforming co-operative transmission. Then, methods for dataset generation for such systems are covered and reviewed. The required datasets for deep-learning may be constructed via ray-tracing simulators. Illustration of the methodology with some deployment scenarios relying on Remcom wireless InSite® ray-tracing simulator are introduced. These scenarios are used to provide the required datasets to be used for deep learning. Then, several directions are proposed for future work that are based on sensing the environment type, (Indoor/Outdoor), and jointly estimating the channel vector, Direction-of-Arrival (DoA), Direction of Departure (DoD), and Rx point power allocation. In a manner similar to adaptive beamforming, the RF radiation pattern can be adapted according to the time-varying DoA and DoD, and estimated channel.