AI enabled RF sensing of Diversified Human-Centric Monitoring

Ge, Yao (2024) AI enabled RF sensing of Diversified Human-Centric Monitoring. PhD thesis, University of Glasgow.

Abstract

This thesis delves into the application of various RF signaling techniques in HumanCentric Monitoring (HCM), specifically focusing on WiFi, LoRa, Ultra-wideband (UWB) radars, and Frequency Modulated Continuous Wave (FMCW) radars. Each of these technologies has unique properties suitable for different aspects of HCM. For instance, 77GHz FMCW radar signals demonstrate high sensitivity in detecting subtle human movements, such as heartbeat, contrasting with the capabilities of 2.4GHz/5GHz WiFi signals. The research extends to both large-scale and small-scale Human Activity Recognition (HAR), examining how ubiquitous communication signals like WiFi and LoRa can be utilized for large-scale HAR, while radar signals with higher central frequencies are more effective for small-scale motions, including heartbeat and mouth movements.

The thesis also identifies several unresolved challenges in the field. These include the underutilization of spatial spectral information in existing WiFi sensing technologies, the untapped potential of LoRa technology in identity recognition, the sensitivity of millimeterwave radar in detecting breathing and heartbeat against minor movements, and the lack of comprehensive datasets for mouth motion detection in silent speech recognition. Addressing these challenges, the paper proposes several innovative solutions:
• A comprehensive analysis of methodologies for RF-based HCM applications, discussing challenges and proposing potential solutions for broader healthcare applications using wireless sensing.
• Exploration of communication signals in HCM systems, especially focusing on WiFi and LoRa sensing. It introduces the continuous AoA-ToF maps method to enhance HCM system performance and the LoGait system, which uses LoRa signals for human gait identification, extending the sensing range to 20 meters.
• Development of a FMCW radar-based structure for respiration detection, incorporating an ellipse normalization method to adjust distorted IQ signals, reducing the root mean square error by 30% compared to baseline methods.
• Collection and analysis of a large-scale multimodal dataset for silent speech recognition and speech enhancement, including designing experiments to validate the dataset’s utility in a multimodal-based speech recognition system.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: T Technology > T Technology (General); T Technology > TA Engineering (General). Civil engineering (General)
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Abbasi, Professor Qammer, Imran, Professor Muhammad and Cooper, Professor Jonathan
• Date of Award: 2024
• Depositing User: Theses Team
• Unique ID: glathesis:2024-84202
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 03 Apr 2024 14:49
• Last Modified: 03 Apr 2024 14:54
• Thesis DOI: 10.5525/gla.thesis.84202
• URI: https://theses.gla.ac.uk/id/eprint/84202

Related URLs

• Data DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding
• Conference proceeding