Application of Autoencoders in pilot-based Channel Estimation for 5G Physical Uplink Shared Channel
Loading...
URL
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
2021-01-26
Department
Major/Subject
Communications Engineering
Mcode
ELEC3029
Degree programme
CCIS - Master’s Programme in Computer, Communication and Information Sciences (TS2013)
Language
en
Pages
86+4
Series
Abstract
5G New Radio (NR) is expected to achieve high-speeds/low-latency capabilities that enable the increasing demand for data rates and low latency in 5G systems, especially to meet the unprecedented requirements in the fifth generation era. Thus, to cope with the complexity and requirements in 5G NR, it is necessary to implement efficient and high-performance algorithms in the physical layer processing. The aim of this thesis is to optimize he channel estimation block in the 5G NR Physical Uplink Shared Channel (PUSCH) by developing a CNN-based Autoencoder that estimates the wireless channel based on DMRS pilots.This solution is compared with the traditional Least Squares channel estimation method. The Autoencoder was trained under multiple channel conditions defined by five different 3GPP channel models and a wide range of SNR values. For this purpose, three different experiments scenarios were specified considering for each of them a different input sample to feed the model. In addition, obtaining the data for the models is an essential part of the process, thus, a 5G-compliant link level simulator and a newly invented 5G Data Generation Tool were employed to produce high quality datasets, which proved to efficiently train the Autoencoder to reach a level of generalization where the model was able to accurately estimate the channel regardless of the SNR values and the channel model. Taking the Least Squares channel estimation method as a baseline for comparison, the Autoencoder outperformed this traditional method of estimation, proving to be a promising solution in the 5G uplink physical layer processing. Moreover, it was found that the Autoencoder achieved an average of up to 90% less estimation error compared with the LS method under certain conditions over all the channel types and SNR values.Description
Supervisor
Jäntti, RikuThesis advisor
Medeiros, LuizKeywords
5G NR, PUSCH, channel estimation, autoencoders, DMRS