Albert Gural

From Murmann Mixed-Signal Group

Jump to: navigation, search

Albert Gural's Portrait

BSEE, California Institute of Technology, 2016

Admitted to Ph.D. Candidacy: 2017

Email: agural (AT) stanford (DOT) edu

Power Reduction Techniques for Robust Heart Rate Monitoring

Background

Continuous heart rate (HR) monitoring is becoming increasingly important for health tracking and diagnostics. Modern approaches use photoplethysmography (PPG), but typically consume a lot of power. Our work aims to reduce the power of PPG-based HR sensors to a level sufficient for continuous monitoring (eg, 100 μW) while still maintaining robustness.

PPGs measure the light reflected off of, or transmitted through, blood vessels in the body. They are primarily used to measure dissolved oxygen (SpO2) and HR [1].

  • SpO2: The absorption spectrum of blood depends on its oxygenation level. The ratio of light received at two different wavelengths can thus be used to determine SpO2.
  • HR: Blood vessels dilate and contract with every pulse, changing their absorption properties according to Beer’s Law. By monitoring the changing levels of light received at the detector over time, the heart rate can be deduced.

Power Reduction Techniques

Agural ppg schematic.png

A typical driving scheme comprises an LED emitter, photodiode detector, transimpedance amplifier, and an ADC. For regions of operation of interest, it can be shown that photodiode shot noise necessitates a trade-off between three critical parameters: power, sample rate, and signal-to-noise ratio (SNR). For a fixed SNR specification, power is proportional to sample rate, so recent research has looked at techniques to reduce sample rate.

One promising technique is to use compressive sampling. Compressive sampling makes use of the fact that the signal is sparse in some domain (eg, the frequency domain for a PPG heart rate signal) so that it can be sampled at a rate much lower than Nyquist would traditionally allow. Recent research has shown impressive power reductions of close to 7x by reducing the sampling rate 30x [2]. Their setup to achieve these results was done in a controlled environment with the LED output modulated to simulate a PPG signal.

One issue with the experimental setup of [2] is that it ignores the challenges faced by real-life PPG sensors - especially ones designed for continuous monitoring, including external light polution and motion artifacts. These additional noise sources affect the robustness of compressive sampling techniques since they reduce the sparsity of the signal. Another issue is that compressive sampling reconstruction is computationally intensive, potentially cancelling the original power savings.

Our work examines other ways of reducing power while maintaining robustness. In particular, we look at machine learning (ML) techniques to answer two questions: (1) can intelligent features be designed that capture the idea of signal sparsity, while still being robust to common interferers, and (2) can low power ML algorithms reconstruct the original signal (or more importantly, infer the heart rate) from these features.

[1] Analog Devices. “An Introduction to Pulse Oxymetry.” 2016.

[2] Rajesh, Pamula Venkata, et al. “22.4 A 172μW compressive sampling photoplethysmographic readout with embedded direct heart-rate and variability extraction from compressively sampled data.” Solid-State Circuits Conference (ISSCC), 2016 IEEE International. IEEE, 2016.

Personal tools