NGC17 Talk @ Samsung Research America - June 9, 2017

1. Reconfigurable Embedded Systems Applications
for Versatile Biomedical Measurements
NECST Group Conference @ Samsung Research America
09/06/2017
Luca Cerina <luca.cerina@polimi.it>
Marco D. Santambrogio <marco.santambrogio@polimi.it>
Politecnico di Milano
Progetto Cariplo MORPHONE 2016-1010: A Challenges
Driven Design for Effective and efficient Autonomic Mobile
Computing Architecture

2. 2
Outline 2
•Context
•Proposed approach
•Related works
•Results
•Case studies
•Conclusions

3. 3
A bit of history 3

4. 4
A technological issue 4

Modular

Fault Tolerant

Powerful analysis

Connection oriented

Bulky

Energy hungry

High cost per device

Portable

Point-of-Care /
patient oriented

Low cost per device

Limited
analysis

Limited sources

Battery constrained

Limited connectivity

5. 5
A planning issue 5
“The number of mobile clinical assets
is skyrocketing, along with associated
service costs, while utilization remains
below 50%. […] While 100% utilization
is impossible, we believe 70 to 80% is
a realistic, achievable target .” [1]
[1] Out of control - how clinical assets proliferation and low utilization are draining healthcare budgets,”
General Electric Healthcare, Tech. Rep., 2012.

6. 6
Main contribution 6
The development of a device which is:

Low-power

Modular / upgradable

Versatile both on sensor sources and computational capabilities

Open to software defined networking

Power source agnostic
Exploiting FPGA-based, multi-processor System-On-Chip (MPSoC)
technologies

7. 7
MPSoC technology 7
ARM Cortex core

OS and software
execution

High-level I/O
connectivity

High-level programming
(C++/Java/Python)
Xilinx Zynq-7 FPGA

Sensor low-level I/O

Reconfigurable efficient
hardware

High parallelismXilinx Zynq-7 Series MPSoC system

8. 8
Related works 8
Reconfigurable EKG analog front-end [1]
Real-time EEG controlled stimulator [2]
1024 channel ultrasound beamformer [3]
[1] D. Morales, A. Garca, E. Castillo, M. Carvajal, J. Banqueri, and A. Palma, “Flexible ECG acquisition system based on
analog and digital reconfigurable devices,” Sensors and Actuators A: Physical, vol. 165, no. 2, 2011.
[2] A. Chemparathy, H. Kassiri, M. T. Salam, R. Boyce, F. Bekmambetova, A. Adamantidis, and R. Genov, “Wearable low-
latency sleep stage classifier,” in 2014 IEEE Biomedical Circuits and Systems Conference (BioCAS) Proceedings, Oct 2014
[3] F. Angiolini et al., "1024-Channel 3D ultrasound digital beamformer in a single 5W FPGA," Design, Automation & Test
in Europe Conference & Exhibition (DATE), 2017, Lausanne, 2017, pp. 1225-1228.

9. 9
Proposed architecture 9
Xilinx MPSoC Pynq-Z1
Board (~220$)
Power supply
Management
Programming model
Software running
on ARM core
Soft-processors code
or hardware solutions
Sensors / Actuators
Different case studies are presented to represent the versatility of the system

10. 10
Case study A: Remote EKG 10
ECG
Amplifier
HW filter
WiFi
data logging
Signal characteristics

Medium bandwidth (~250Hz for diagnostic
use)

1 to 12 acquisition channels

Complex analysis → information extracted

from morphology, interbeat intervals, and
electrodes’ location

11. 11
Case study B: Pupil dilation 11
OpenCV
Videocapture
Histogram
equalizer
HW filter
OpenCV
Pupil contour
Python on ARM Core
FPGA
Characteristics

Ease of use with Python and
OpenCV

Application specific acceleration

Parallelism and low power

Fast implementation → 33 fps after
a 3 weeks development by bachelor
students

12. 12
Case study C: Heterogenous signals 12
ECG
Amplifier
HW filter
GSR
Amplifier
Downsample
+
HW filter
Soft-core 2 (Running @ 8 MHz)
Soft-core 1 (Running @ 100 MHz)
Software running
on ARM core (@ 650 MHz)
Different hardware clock
regions enable soft parallelism
without increasing complexity
at software level
Galvanic Skin Response (GSR) tonic components are limited to 3-5Hz,
with lower requirements on the processing core

13. 13
Case study D: Fog infrastructure 13
Humans become a fundamental piece in the sensor-actuation loop:
● Low latency Observe – Decide – Act systems → safety & emergency
● Collaborative environmental control → human thermal comfort
● Secure-by-design indoor localization → better building management
● Human-in-the-loop control → low cost retrofitting towards smart ambients
All enhanced by reconfigurable, low-power, multi-task Fog nodes

14. Conclusions 14
This work proposed a device prototype for versatile biomedical
measurements, demonstrating that FPGA-based systems are valid
for applications that require:

Low power consumption, comparable with existing commercial
devices

Multi-purpose capabilities (useful for Point-of-Care devices)

Extended connectivity

High performance and efficient HW parallelism

Modularity and low costs

15. QUESTIONS?
Reconfigurable Embedded Systems Applications
for Versatile Biomedical Measurements
Luca Cerina <luca.cerina@polimi.it>
Marco D. Santambrogio <marco.santambrogio@polimi.it>
Politecnico di Milano