Skip to main content

Hardware Friendly Machine Learning Integrated Circuits and System for Low Power Wearable Wireless Electrocardiogram Sensors

07-01-2020 - Future wearable wireless biomedical sensors demand novel technologies to overcome the increasing challenge in implementing intelligent signal sensing and processing, the shortage of battery lifetime, as well as latency and security issues. The proposed research focuses on a novel computing modality with low-power analog-to-feature converter integrated circuit and patient-specific machine learning framework for on-sensor machine learning with energy-efficient wireless data communication. The key idea is to combine the advantages of the oversampling modulators for feature extraction, and the rotation linear kernel support vector machine for learning and classification, with oversampling encoder and asynchronous ultra wideband impulse radio for data communication. We aim to study the system design principles and fundamental performance limits of the novel computing framework in arrhythmic heartbeat classification. The goal of this proposal is to ultimately address the challenges of next-generation wireless wearable biomedical sensors by systematical efforts, which include interrelated studies in low power circuit design, hardware-friendly algorithm design, and communication system analysis. The proposal demonstrates broader impacts in the forms of technological, societal, and educational impact based on the comparative advantage of New Mexico. In particular, the examples of applications of the proposed system are presented for elderly care and health care assistant programs. Educational activities are also planned to attract and train the next-generation of scientists and engineers at NMSU, which is one of the best Hispanic-Serving Institutions in research.