Compressive Wireless Pulse Sensing CTS 2015 Internet of Things - PowerPoint PPT Presentation

Compressive Wireless Pulse Sensing CTS 2015 – Internet of Things Harvard University Kevin Chen Harnek Gulati HT Kung Surat Teerapittayanon Tracking reliable pulse waves for long term health diagnostics 1

Motivation Classification of Heart Health Classification of heart conditions derived from heart rate over time [1] Peng, Chung-Kang. "Toward a General Principle of Health and Disease." Toward a General Principle of Health and Disease. Harvard Medical School, Cambridge. 26 Mar. 2015. Lecture.

Motivation Diagnostics based on pulse Apnea Apnea Heart Rate Sleep apnea diagnosis based on changes in heart rate Time (Minutes) Blood pressure calibration - Wrist Blood Volume - Finger Time (Minutes) from phase change of PPG signals in two locations 3 Time (Seconds)

Message With the recent availability of low-power wireless chips, for the first time, we can monitor pulse waves over a long period of time for applications such as measuring heartrate variability. However, we are still limited by the power budget available on wearables. In this paper, we will show how we can use compressive sensing to reduce power consumption. 4

Problem to Solve Power Consumption of Wearables Battery life of heartrate watches Battery consumption of Apple Watch Mio Link wearables restricts its ability to Mio Alpha Garmin Forerunner continuously monitor pulse wave 0 2 4 6 8 10 12 Lifetime (hours) With new low-power wireless chips like BLE and additional power-saving compressive sensing techniques of this paper, it is now feasible for battery- powered wearables to monitor pulse wave continuously for days or even weeks. 5

Overview of System Tracking reliable pulse waves for long term health diagnostics Signal Acquisition 6

Video Demo of Pulse Wave Reconstruction 7

Outline of Presentation 1. Signal Acquisition – Compressive sensing for pulse waves 2. Wireless Data Transmission – Forward error correction by interleaving and randomization Adaptations in response to channel quality – 3. Signal Recovery – Reconstruction of pulse wave through sparse coding – Noise removal 8

Part One: Signal Acquisition Compressive sensing for pulse waves 9

Compressive sensing formulation 1. Compression by linear projection 2. Finding sparse representation of x z y S D y S x Sensing = Sensing = Dictionary Matrix Matrix Obtain Trained Givens Givens Solve 3. Reconstruction of x x D z = Dictionary Trained Given 8

Uniform subsampling to reduce sensor wake-up time We use uniform subsampling as the sensing matrix y U x The measurements 1 = 1 1 y is a linear projection of x 1 Obtain subsampling 11

Finding the sparse representation of x x Givens y z U D 1 = 1 1 1 Uniform subsampling matrix subsampling Trained Dictionary 12

Reconstructing the signal from sparse representation x D z = Dictionary Solved Trained Simple Matrix Multiplication 13

Experimental Results With a dictionary trained on pulse waves, uniform subsampling performs better than classic compressive sensing methods. Low construction error and very efficient to implement 14

Wireless Data Transmission — Forward error correction by interleaving and randomization — Adaptations in response to channel quality 15

Naïve transmission scheme Putting a whole signal segment in one packet is not ideal, because there is no way to recover information without resending #1 #2 #3 #4 #p … Batch of packets A whole segment of signal is lost 16

Packet interleaving By interleaving packets, we can recover the information of lost packet from neighboring received packets. #1 #2 #3 #4 #p … Batch of packets Packet 3 Packet n Packet 1 segment 1 1 2 3 4 5 6 n segment 2 2n n+1 n+3 segment 3 … segment p 17

Problems with burst packet loss However, consecutive packet loss still results in consecutive sample loss in each segment #1 #2 #3 #4 #p … Batch of packets Packet 3 Packet n Packet 1 segment 1 1 2 3 4 5 6 n segment 2 2n n+1 n+3 segment 3 … segment p 18

Randomizing packet sending order We can avoid consecutive sample loss by sending packets in randomized order #1 #2 #3 #4 #p … Batch of packets #10 #3 #1 #2 #21 Randomized … sending order 1st Pkt 2nd Pkt 3rd Pkt 4t h Pkt p-th Pkt Sent Sent Sent Sent Sent 19

Reconstruction with updated packet transmission scheme y z U D We can represent 1 = 1 the packet 1 1 interleaving as a Uniform projection subsampling matrix Dictionary Before After z w C U D 1 1 = 1 1 1 1 1 1 Channel Uniform matrix subsampling matrix Matrix that represents how Dictionary we interleave packets 20

Reconstruction error with varying packet loss rates Transmission rate is adaptive to packet loss Channel is bad, but we get loss tolerance by simply increasing sampling rate Channel is good, so we can sample at a very low rate. 21

Signal Recovery — Reconstruction of pulse wave through sparse coding — Noise Removal 22

Reconstructing the signal 1. Use sparse coding to recover z z w U C D 1 1 = 1 1 1 Solve 1 1 1 Channel Uniform matrix subsampling matrix Dictionary Known 2. Reconstruct x x z D = Dictionary 23

Cleaning the signal from outliers There can be outliers caused by movements, sensor voltage change, etc. 24

Augmenting the dictionary for noise removal With a little tweak, we can even tolerate corrupted measurements z w Corrupted 1 1 1 1 = CUD measurement Dictionary Identity matrix 25

Reconstruction error at different noise levels We can deal with corrupted samples by increasing sampling rate 26

Implications of Our Results Pulse Diagnostics Readily Available Long term health monitoring made possible Classification of heart conditions derived from heart rate over time Heart Rate Sleep Apnea diagnosis based Apnea Apnea on changes in heart rate Time (Minutes) - Wrist Blood Pressure Calibration Blood Volume - Finger from phase change of PPG signals in two locations 27 Time (Seconds)

Summary With new BLE chips, continuous health monitoring is possible for the first time Lower wakeup frequency Signal Acquisition 28

Conclusion Due to the recent availability of pulse sensing chips, and low-power wireless chips, for the first time we can monitor pulse waves over along period for applications such as measuring heartrate variations. But we have a challenge of coping with limited power budget available on wearables. We have shown in this paper that we can use compressive sensing to reduce power consumption. 29

Training a dictionary with pulses* (remove?) Trained on earlier samples x z D = Dictionary 30

31

Naïve transmission scheme Putting a whole signal segment in one packet is not ideal, because there is no way to recover information without resending #1 #2 #3 #4 #p … Batch of packets c A whole segment of signal is lost 32

33

Compressive Wireless Pulse Sensing Kevin Chen Harnek Gulati HT Kung Surat Teerapittayanon Signal Acquisition 34