Add to Google Calendar
Date:  Thu, October 01, 2020
Time:  3:00pm - 5:00pm
Location:  online conference
Speaker:  Shekh Md Mahmudul Islam, PhD Candidate

Please contact EE Office eeoffice@hawaii.edu for online meeting connection information.

An unobtrusive and non-contact continuous authentication system can potentially improve security throughout a login session. Traditional user authentication procedures such as fingerprint, password, or facial identification typically provide only an initial spot-check of identity at the start of a user session, potentially allowing undesired user changes or subsequent access to personal information. The research to be presented is focused on creating a non-contact continuous authentication system based on Doppler radar, which analyzes back-scattered RF signals which carry body motion information indicating a human subject’s vital signs (breathing rate, heart rate) and associated unique patterns. A key advantage of this radar technique is that continuous authentication is achieved without intrusive video imaging. While prior results focused on the use of respiratory motion to identify a single isolated subject, the challenge of resolving and identifying multiple subjects within the radar field of view is addressed here.

This research introduces an SNR-based decision algorithm which coherently combines two separation methods to overcome multiple subject monitoring limits. The hybrid-approach system manages well-spaced subjects. those at the edge of the antenna beamwidth or beyond, using a Direction of Arrival (DOA) approach, and more closely spaced subjects by employing Independent Component Analysis with the JADE Algorithm (ICA-JADE) to isolate individual respiratory signatures. Additionally, highly distinguishable breathing dynamics related to hyper-features (inhale area, exhale area, and breathing depth) can be extracted from the radar-captured respiration patterns to facilitate individual subject identification. Extracted hyper feature sets for 20 subjects measured with a 24-GHz radar system were used to train and test two different popular machine learning classifiers (K-nearest neighbor (KNN) and Support vector machine (SVM)). SVM with quadratic kernel outperformed the other classifiers, with an accuracy of 97.5%. The empirical entropy of the breathing-related hyper-feature sets was calculated, and it was determined that the empirical entropy (approximately 3 bits) is insufficient for secure identification.

To increase the security of the proposed system, a Fuzzy Extractor with linear coding was included to transform the hyper-feature set into a strong biometric key compatible with machine learning classifiers. The feasibility of using the proposed radio-based identity verification system with an in-home sleep apnea monitoring system has been examined. Sleep studies can be complicated when multiple subjects are present or when there is some threat of a surrogate substitution. A modality switch compliance tracking protocol has been developed which avoids disambiguation and resists attacks (mix-up, surrogate, eavesdrop, etc.). This is the first known reported attempt to achieve secure radio-based multi-subject identity verification for in-home obstructive sleep apnea screening by harnessing the power of Doppler radar and Fuzzy key extractors.