Frequency Offset Estimation for OTFS Systems Based on Deep Belief Networks

  • Yuchen Zhang
  • , Rao Fu
  • , Yufei Wang
  • , Xiaohui Dong
  • , Xinxin Yang
  • , Michel Kadoch

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In this study, we introduce a Carrier Frequency Offset (CFO) estimation technique based on Deep Belief Networks (DBN) in the context of deep learning principles. We leverage unsupervised learning methods to pre-train the layers of a Restricted Boltzmann Machine (RBM), keeping the initial weights and biases within an ideal range. Subsequently, supervised learning is employed to finetune the network parameters, mitigating issues related to random parameter initialization and local optima. Experimental analysis indicates that the CFO estimation accuracy significantly improves with the DBN-based CFO estimation technique in high-dynamic environments. Moreover, there is a notable enhancement in Bit Error Rate (BER) performance.

Original languageEnglish
Title of host publicationProceedings - 2023 International Conference on Information Processing and Network Provisioning, ICIPNP 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages495-499
Number of pages5
ISBN (Electronic)9798350363470
DOIs
StatePublished - 2023
Event2023 International Conference on Information Processing and Network Provisioning, ICIPNP 2023 - Beijing, China
Duration: 26 Oct 202327 Oct 2023

Publication series

NameProceedings - 2023 International Conference on Information Processing and Network Provisioning, ICIPNP 2023

Conference

Conference2023 International Conference on Information Processing and Network Provisioning, ICIPNP 2023
Country/TerritoryChina
CityBeijing
Period26/10/2327/10/23

Keywords

  • Carrier Frequency Offset Estimation
  • Deep Belief Network
  • OTFS System

Fingerprint

Dive into the research topics of 'Frequency Offset Estimation for OTFS Systems Based on Deep Belief Networks'. Together they form a unique fingerprint.

Cite this