Doppler Radar Techniques for Distinct Respiratory Pattern Recognition and Subject IdentificationDate: 2017-03-09 Add to Google Calendar
Location: Holmes Hall 389
Speaker: Ashikur Rahman, EE PhD Candidate
Stationary continuous wave Doppler radar has been used for displacement measurement and vital signs detection in many state-of-the-art works. However, further improvement, i.e., accurate radar characterization of respiration may allow sleep diagnostics, and unique identification. A low distortion DC coupled system with high signal to noise ratio is required for such characterization and classification. This is especially critical with small signals as with through wall measurements with poor signal to noise ration (SNR). This thesis proposes techniques to improve signal to noise ratio by DC offset management and using the method of zooming in the fractions of respiratory cycle waveform. A month-long study on 6 human subjects were performed and the developed Doppler radar system and classification algorithms have shown promising results in unique identification (vital sign based fingerprint) with more than 90% accuracy with measurement time as low as 30 seconds. Neural networks, minimum distance classifiers, and majority vote algorithms were fused on multi feature spaces to make classification decisions. Training and testing were performed on the extracted features such as, variation in their breathing energy, frequency and patterns captured by the radar. The system has shown the advantages of non-chat unique identification where camera based system is not preferred or incapable. This study also has impact on radar-based breathing pattern classification for health diagnostics.
This research also investigates poor SNR problem on a mobile platform as measurements become challenging due to motion artifacts induced by the platform. To implement a feasible field applicable solution low intermediate frequency (IF) techniques for non-invasive detection of vital signs from a mobile short-range Doppler radar platform were proposed and validated through mechanical and human experiments. A low IF radar architecture using RF tags is employed to extract desired vital signs motion information even in the presence of large platform motion. Upon researching SNR improvement and developing algorithms this research took one step further for unique identification of human subjects behind the walls. Many potential applications such as, security, health monitoring, IoT, virtual reality, and health diagnostics will have a broader positive impact on human society and sustainable existence of human and nature.