CCSS: Reference-free and Spatial-aware Deep Sensor Array Decoding towards High-fidelity Remote Health Monitoring

About This Grant

Remote health monitoring is highly promising for big data-driven precision medicine, through conveniently and obtrusively tracking health conditions of people. However, when the sensing device is placed off-body for remote monitoring, the captured human signal is usually very weak. This is because the signal quickly decays when it propagates from the human body to the device. Further, the signal of the target person may be interfered if there are more than one person in the environment. Targeting these crucial challenges, this project will advance the science of high-fidelity remote health monitoring, through efforts on innovating the remote signal sensing and decoding system architecture. This project will greatly advance the national health towards pervasive, high-fidelity, and long-term big data establishment. More specifically, this project will design a novel deep senor array decoding system, which leverages the data-driven deep learning algorithm to decode the noisy and weak signal, without needing a reference signal used for propagation-induced distortion estimation. Besides, the multi-sensor spatial information will be leveraged by deep learning to boost the signal fidelity and recover the signal-of-interest from noise and interferences. The project will further contribute to research-education integration through new course development, new pedagogy practices, curriculum enhancement, and broad student training. The PI will continue broadening the participation of undergraduate students as well as K-12 students thereby effectively training the next-generation engineers and researchers. This project will innovate a novel deep sensor array decoding system, which can decode the signal-of-interest from the noisy and weak signal remotely captured, towards promising remote health monitoring and precision medicine big data. The multi-sensor signal captured by a sensor array, will be analyzed by the deep learning algorithm to learn the noise patterns, suppress the noise, and decode the high-fidelity signal. This data-driven approach does not need the reference signal that is usually used for propagation-induced distortion estimation, thereby enabling intelligent and convenient signal decoding. The spatial dynamics captured by the sensor array encode complex information about the signal-of-interest, can be effectively learned with the deep learning-empowered signal decoding. Besides, the deep learning algorithm will learn to separate the signal-of-interest if there are more than one person in the environment. The specific signal patterns for the target user will be learned and used by the deep learning algorithm to mine the target-relevant patterns in the multi-sensor signal captured. The proposed system architecture will be further evaluated with real-world experiments, to demonstrate the generalizable innovation and the effectiveness of the system. The novel system architecture will broadly contribute to various remote health monitoring applications, advance national health with pervasive and convenient big health data establishment, and promote the science on deep sensor array decoding for high-fidelity remote health monitoring. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.