Mixed Hardware and Software Embedded Signal Processing Methods for

in-situ Analysis of Cardiac Activity

Bertrand Massot

, Tanguy Risset

, Gregory Michelet

and Eric McAdams

Lyon Institute of Nanotechnology, CNRS - INSA Lyon, University of Lyon, Villeurbanne, France

Centre of Innovation in Telecommunications and Integration of Service, INRIA - INSA Lyon, Villeurbanne, France

Keywords:

Wearable Sensors, Heart Rate, Heart Rate Variability, Biomedical Signal Processing, Body Sensors Network.

Abstract:

This paper presents the implementation of a combination of hardware and software signal processing methods

on a wearable device for the continuous and long-term monitoring and analysis of cardiac activity during in-

situ experiments. Heart rate assessment and heart rate variability parameters are computed in real-time directly

on the sensor, thus only few parameters are sent via wireless communication for power saving. Hardware

method for heart rate measurement, and software methods for the calculation of time-domain and frequency-

domain parameters of heart rate variability are described, and preliminary tests for the evaluation of the sensor

are presented.

1 INTRODUCTION

The continuous and long-term monitoring of an indi-

vidual’s vital signs enables a real-time and more rel-

evant health diagnosis, in order to set up appropriate

preventive measures, and to undertake rapid remedial

action in the case of early detection of symptoms; as,

for example, the latent anticipation of sudden cardiac

arrest (424 000 annual out-of-hospital cardiac arrests,

with an overall survival rate of only 5.2 % (Go et al.,

2014)) by evaluating the risks of cardiovascular dis-

ease and by detecting any cardiac abnormalities. This

in turn will result in more effective healthcare deliv-

ery, both ﬁnancially and therapeutically (Van Hoof

and Penders, 2013) by avoiding untimely hospitaliza-

tion while ensuring patient safety and autonomy.

A promising solution is the development of wear-

able systems which assess relevant indicators en-

abling a direct, on-body cardiac diagnosis. Several

research projects’ results in this area have highlighted

a range of various technical challenges that must be

overcome (McAdams et al., 2011), (Massot et al.,

2013). The achievement of a suitable device for con-

tinuous, long-term monitoring of heart rate activity

will enable the detection of cardiac abnormalities in

the electrocardiogram signal (ECG), for example to

prevent ventricular ﬁbrillation (VF), and to monitor

the instantaneous heart rate (HR), from which can be

derived several parameters regarding heart rate vari-

ability (HRV). HRV provides information on auto-

nomic nervous system (ANS) activity, a relevant indi-

cator for several pathologies (Malik et al., 1996) and

more generally on an individual’s stress and arousal.

New wearable devices for the monitoring of heart

rate activity can exploit the beneﬁts of recent techno-

logical advances in electronics and wireless commu-

nication systems in order to overcome the challenges

previously cited. Prototypes developed in laborato-

ries already show really promising results in terms of

wearability, robustness and autonomy, as for exam-

ple the wearable patch developed at the Holst Centre

which beneﬁts from both elaborated hardware (Altini

et al., 2011) and software (Romero et al., 2009), and

more recently from a new kind of dry electrode for

comfortable measurements (Chen et al., 2013). There

are already commercially available products for per-

sonal monitoring of one’s own cardiac rhythm, but

they are mainly aimed at well-being and ﬁtness ap-

plications rather than being suitable for medical pre-

vention and diagnosis. Most of these systems suf-

fers from a lack of accuracy, depending on the sens-

ing method used : for example, plethysmography ap-

pears to be still questionable for instantaneous HR

and short-term HRV assessment and is highly sen-

sitive to motion artefacts in ambulatory conditions

(Sch

afer and Vagedes, 2013). Also, ﬁltering and in-

terpolating HR due to motion artefacts induces distor-

tion in frequency content of subsequent HRV param-

eters.

In this paper, we presents an optimized combi-

nation of robust and accurate methods for on-board

Massot, B., Risset, T., Michelet, G. and McAdams, E.

Mixed Hardware and Software Embedded Signal Processing Methods for in-situ Analysis of Cardiac Activity.

DOI: 10.5220/0005843703030310

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 4: BIOSIGNALS, pages 303-310

ISBN: 978-989-758-170-0

303

HR detection and HRV calculation. The methods

were implemented on a programmable system-on-

chip which provides hardware analog and digital pro-

grammable functions as well as a 32-bit ARM Cor-

tex M3 micro-controller unit. The objective was to

optimize the selected methods in order to beneﬁts

from the ultra-low power consumption of the hard-

ware part, which is used for real-time HR detection

and period measurement, and to reduce the calcula-

tion time of frequency-domain parameters of HRV,

which is done together with time-domain parameters

by the micro-controller unit.

In section 2, the different methods for ECG mea-

surement, HR detection and HRV calculation, as well

as their implementation on the targeted device are de-

scribed. Then the evaluation of the accuracy of the

methods, both on test-bench and in real-life condi-

tions is presented in section 3.

2 MATERIALS AND METHODS

2.1 Targeted Wearable Device Overview

(a) (b)

Figure 1: Device overview (a) and example of placement on

the body using three disposable electrodes (b).

The REC Heart Activity sensor is a wearable de-

vice which has been developed to enable the continu-

ous monitoring and analysis of cardiac activity during

long-term experiments in real life conditions.

The device comprises a small electronic board

and a 300 mAh Lithium-Ion battery encapsulated in

a small plastic enclosure (50 mm x 35 mm x 15

mm) and can be connected to ECG electrodes by the

means of three snap connectors, thus the sensor is in-

tended to be connected to various electrode conﬁgu-

rations, directly on the body (Figure 1). Possible elec-

trode conﬁgurations include for example a disposable

patch of gel electrodes, or a chest belt with dry elec-

trodes, depending on the requirements regarding the

experiment conditions (resting or effort during short

periods, long-term monitoring during several days,

etc.). The electronic board includes a Bluetooth Low

Energy (BLE) interface for wireless communication,

and the sensor can be integrated in a Wireless Body

Sensor Network (WBSN). In the frame of the RE-

CAMED project, an Android application has been

developed to collect data from a WBSN composed

of various wearable sensors including the REC Heart

Activity sensor.

The electronic architecture of the board is based

on a PSoC 5LP (Cypress Semiconductors). This

mixed-signal Programmable System-on-Chip con-

sists of a Cortex M3 ARM micro-controller unit, but

also includes analog programmable functions as well

as programmable logic device (PLD) based functions.

This component can thus carry out all the steps from

the conditioning of the ECG signals to the transmis-

sion of high level heart activity indicators through

the BLE interface, including signal processing, ana-

log to digital conversion, heart beat detection, heart

rate measurement, and heart rate variability calcula-

tion. All these functions are integrated within a low-

power, single chip with a highly reduced size (8 mm

x 8 mm) as described in the next sections.

2.2 Integrated Signal Processing of

Electrocardiogram

Figure 2: ECG processing chain using internal PSoC 5LP

analog hardware components.

The ampliﬁcation and digitization of ECG is done by

using the integrated and programmable analog func-

tions of the PSoC5 LP. A differential ampliﬁer is real-

ized by combining programmable gain ampliﬁers and

the outputs are directly connected to a 12-bit succes-

sive approximation differential ADC (Figure 2) . Two

additional operational ampliﬁers used as buffer and

inverting ampliﬁer respectively are chained to imple-

ment a right-leg drive (RLD) circuit to provide ad-

ditional common-mode noise rejection (Winter and

Webster, 1983). The overall differential gain is set to

24 and the SAR ADC has an input range set to ±1.024

V so the resolution is 20.8 µV/bit, and the signal is

sampled at 8192 samples per second.

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

304

2.3 Integrated Hardware Digital

Measurement of Heart Rate

Figure 3: RR-Interval detection chain using internal PSoC

5LP digital hardware components (PLDs).

The objective of using hardware digital components

for the measurement of Heart Rate is to beneﬁt for

the highly reduced consumption of PLDs available

within the PSoC. Consuming processing time of the

MCU for R-peak detection would lead to a high us-

age of battery power for a real-time detection peaks

on the ECG signal at a 8192 sps sampling rate. Al-

ternatively this hardware chain have an average con-

sumption of 140 µA and enables the MCU to remains

into an idle state, as the data is also directly trans-

mitted from the analog chain to the digital chain by a

DMA channel. R-peaks detection is done by imple-

menting a digital ﬁlter and a local maximum detector

in the PSoC’s PLDs (Figure 3).

The digital ﬁlter block (DFB) acts as an ultra-low

power DSP in which is implemented a non-linear ﬁl-

ter to enhance power in the higher frequency of ECG

signal (Pan and Tompkins, 1985). The processing

includes a double buffering of the ECG to calculate

a smoothed derivative which is then squared, and a

moving-average ﬁlter is applied on the ﬁnal signal.

The local maximum detector uses an adaptive

threshold to take into account the variability of ECG

amplitude between individuals and also aver time.

The dynamic threshold T is dynamically computed

within the PLD components by an exponential ﬁlter

as in Equation 1, where β is the amplitude ratio of the

current peak detected P, and α is the smoothing factor

of the ﬁlter.

n+1

= α · β · P

+ (1 − α) · T

(1)

The value of a 1024 Hz up-counter is copied into

the MCU memory by a DMA channel and an interrupt

is triggered each time a peak is detected on the sig-

nal. Additionally a “fail-safe” down-counter disables

detection when false peaks have been detected until

three consecutive peaks are properly detected. The

condition of false detection is true when the variation

of RR-interval is higher than 25% of the last one. This

method guarantees reliable RR-intervals value for the

proper calculation of HRV parameters which is more

sensitive to false detection than missing values (cf.

section 2.4).

2.4 Software Calculation of Heart Rate

Variability Parameters

When the hardware detector triggers an interrupt

which wakes up the MCU, the new RR-interval is

copied into a buffer which keep in memory the values

of the last 5 minutes, the standard period for evaluat-

ing short-term HRV according to the Task Force of the

European Society of Cardiology and the North Amer-

ican Society of Pacing and Electrophysiology (Malik

et al., 1996). This signal (tachogram) buffered is then

used to calculate both time-domain and frequency-

domain parameters of HRV. The Task Force has se-

lected a large number of parameters to evaluate HRV,

from which four parameters has been retained to be

calculated in this application, based on their common

usage in the analysis of ANS activity (see Table 2 at

the end of the section).

The calculations are performed directly by the

MCU with ﬂoating point values, due to the precision

needed within these operations. This induces periods

of intensive occupation of the MCU which need to

be reduced at the minimum if one wants to optimize

consumption of the device.

2.4.1 Time-domain Parameters

The calculation of time-domain parameters of the

HRV is quite straight-forward and the two most used

parameters are computed in this application, i.e. the

standard deviation of intervals in the buffer (SDNN)

and the quadratic mean of differences between suc-

cessive intervals (RMSSD). These parameters are cal-

culated used the formulas given by Equations 2 and 3.

SDNN =

n − 1





∑

i=1

(RR

)

−

∑

i=1





(2)

RMSSD =

n − 1

n−1

∑

i=1

(RR

i+1

− RR

)

(3)

Most devices commercially available which pro-

vide real-time HRV monitoring, calculate only one

time-domain parameter (one among the two cited),

and usually without naming it. In this application, we

Mixed Hardware and Software Embedded Signal Processing Methods for in-situ Analysis of Cardiac Activity

305

apply to provide precise information about the param-

eters calculated, whose variations can differ depend-

ing on the situation and thus modify the interpretation

of HRV regarding the ANS activity.

2.4.2 Frequency-domain Parameters

Calculation of frequency-domain parameters of HRV

requires an evaluation of power spectral density

(PSD) of the tachogram as it evaluates the distribu-

tion of energy of the signal in separated frequency

bands. The main frequency bands are usually de-

ﬁned as ultra-low (ULF), very low (VLF), low (LF)

and high frequencies (HF) (Table 1).

Table 1: Separation of power spectral density of the

tachogram in frequency bands.

Name Frequency range

ULF ≤ 0.003 Hz

VLF 0.003-0.04 Hz

LF 0.04-0.15 Hz

HF 0.15-0.4 Hz

In this application, the method of evaluating the

PSD is critical due to the embedded electronic ar-

chitecture used in the sensor, which provides limited

resources in performance and time. As RR-intervals

vary in time, the tachogram is composed of unevenly

sampled values; thus a traditional approach for spec-

tral analysis consists of a combination of (i) an inter-

polation, in order to recover an evenly sampled signal,

and (ii) a subsequent Fast Fourier Transform (FFT) to

obtain the PSD. However this approach, depending on

the method of interpolation, the sampling rate and the

number of points, is known to introduce distortion in

the high-frequency domain where re-sampling acts as

a low-pass ﬁlter, leading to an overestimation of HRV

parameters (Clifford and Tarassenko, 2005). Also this

method is known to be very sensitive to both errors in

detection and measurement of RR-intervals as well as

missing values in the tachogram.

Another approach for spectral analysis of an un-

evenly sampled signal is the least square analysis,

commonly termed the Lomb-Scargle periodogram,

which provides (in a normalised form), the estimated

power P of the angular frequency component ω. The

estimated power is given by Equation 4, where σ =

SDNN, the standard deviation of all R-R intervals, RR

is the mean value, and τ is an angular quantity deﬁned

by Equation 5.

P(ω) =

2σ



∑

i=1

(RR

− RR)cos(ω(t

− τ))



∑

i=1

cos

(ω(t

− τ))



∑

i=1

(RR

− RR)sin(ω(t

− τ))



∑

i=1

sin

(ω(t

− τ))

(4)

tan(2ωτ) =

∑

i=1

cos(2ωt

)

∑

i=1

sin(2ωt

)

(5)

This method, originally proposed by Lomb

(Lomb, 1976) and further elaborated by Scargle

(Scargle, 1982), was proposed as a surrogate for HRV

calculations for the ﬁrst time by Shin et al. (Shin et al.,

1994) in 1994 (to the best of authors’ knowledge).

This method provides better accuracy and lower noise

levels in the estimation of the density power spectrum,

but unfortunately it also has the major drawback of in-

volving much more calculation complexity, and thus

MCU time consumption, even when the algorithm is

optimized with classical trigonometric recurrences.

Press and Rybicki (Press and Rybicki, 1989) have

proposed a much faster computation of this parame-

ter by combining the accuracy of the periodogram and

the efﬁciency of FFT, resulting in an algorithm which

is as fast as two FFT calculations and a N log N order

instead of N

. In this case the FFT is not used for

the direct evaluation of the periodogram, but rather

to calculate approximately (but to any desired preci-

sion), both main terms of Equation 4. To evaluate

trigonometric sums of the equation, which can not be

calculated with FFTs due to the unevenly spaced data,

the method involves reverse interpolations, call extir-

polation. As the interpolation evaluates one value at

an arbitrary point upon several values from a regularly

sampled function, the extirpolation evaluates several

value of a regularly sampled function from the value

of an arbitrary point. The precision, and also the dura-

tion of this evaluation depends on the number of extir-

polated points per 1/4 cycle of the highest frequency

(MACC parameter). The raw algorithm and several

values of the MACC parameter of its fast implemen-

tation have been tested, and a performance compari-

son in accuracy and gain of time is presented in the

Results section.

As stated above, the Lomb-Scargle periodogram

being dedicated to the evaluation of PSD for unevenly

signals, it is far less sensitive to missing data than FFT

where interpolation can lead to large differences de-

pending on the interpolation method. Also both meth-

ods are sensitive to false detections, therefore an ad-

ditional “fail-safe” digital circuit has been added to

the R-peak detector as described in section 2.3. This

circuit gives a higher prevalence to correct R-peak de-

tections at the cost of additional missing values.

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

306

Table 2: Summary of short-term HRV parameters calculated by the REC Heart Activity sensor.

Variable Unit Domain Description

SDNN ms Time Standard deviation of all R-R intervals

RMSSD ms Time Quadratic mean of differences between successive R-R intervals

LF/HF n.u. Frequency Ratio between LF and HF components of the PSD of all R-R intervals

LF norm % Frequency Ratio (expressed as a percentage) between LF and LF+HF

3 EVALUATION AND RESULTS

The objective of the evaluation of the HR measure-

ment method and the HRV parameters calculation

method is primarily to optimize the different parame-

ters (gain, sampling rate, ratios of the dynamic thresh-

old, extirpolation of the fast periodogram, etc.) to en-

sure that the device will provides both the mandatory

robustness and accuracy of the signals (and thus de-

rived data) for the use of the sensor in clinical ex-

periments and applications. On an other hand, it is

necessary to maintain a suitable autonomy for long-

term experiments by reducing power consumption of

the overall device. In the proposed implementation,

where the consumption of HR measurement method

is already highly optimized by the use of dedicated

hardware functions, further reduction of power con-

sumption relies on the optimization of the calculation

time of HRV parameters.

For this purpose, the REC Heart Activity sensor

was evaluated both on a workbench in laboratory con-

ditions as well as on individuals in real-life conditions

as described in the next sections.

3.1 Accuracy of HR Assessment

Accuracy of the detection of R-peaks and the mea-

surement of RR-intervals upon the ECG signal was

evaluated using a hardware generated ECG. The Ag-

ilent 33220A is function / arbitrary waveform gener-

ator which provides a cardiac waveform. The ampli-

tude, common-mode and frequency can be varied to

verify the proper operation of the device in various

conditions as those three parameters depends highly

on the environmental and physical conditions of the

individual (resting, effort), and also varies with the

change of the electrode/skin interface over time (par-

ticularly when using dry electrodes).

The error in RR-intervals measurement was cal-

culated as the mean difference between the period set

on the generator and the period measured by the sen-

sor. As the latter uses a counter with a frequency of

1024 Hz, the precision of RR-intervals measured is

0.98 ms. A dataset of 1000 RR-intervals was col-

lected where intervals’ length was linearly varied on

the generator from 400 ms (150 BPM) to 1200 ms (50

BPM), which is representative of most common heart

rates. The mean value of all difference was -0.105 ms

and the standard deviation of the differences was of

1.027 ms over all the range of RR-intervals. This dif-

ference corresponds to an error of 0.1 BPM whan HR

is 60 BPM, and 0.4 BPM when HR is 120 BPM which

is lower than the usual 1 BPM resolution in standard

devices.

3.2 Accuracy of PSD Estimation

To simulate heart rate variability and to evaluate

the accuracy of the different implementation of the

Lomb-Scargle periodogram, a known frequency mod-

ulation was applied to the ECG signal generated by

the Agilent 33220A. The base HR was set at f

base

1.25 Hz (75 BPM). The modulating signal was a tri-

angular shape, with a frequency f

mod

of 0.05 Hz and

an amplitude of frequency deviation f

dev

of 0.2 Hz.

As the PSD is computed over RR-intervals in units of

time (ms), the theoretical continuous function RR(t)

corresponding to the tachogram is given by Equations

6 and 7. The time variations of this continuous and pe-

riodic signal as well as the normalized PSD are shown

on Figure 4.

tri

(t) = 2





t ∗ f

mod

−



t ∗ f

mod





− 1 (6)

RR(t) =

base

+ f

dev

∗ x

tri

(t)

(7)

As shown on the normalized PSD, the use of a tri-

angular shape as a modulating signal for HR has the

advantage of inducing predictable harmonics in the

PSD at multiple frequencies of the fundamental f

mod

all over the range of interest 0.015-0.4 Hz. Addition-

ally, the inverting relationship between RR-interval

values and HR breaks the vertical symmetry of the

triangular signal and thus adds even harmonics to the

odd harmonics of the original triangular shape.

It therefore possible to analyse directly PSD ob-

tains with different methods in order to compare

both quality of PSD estimation and time of calcula-

tion. For this evaluation, the original Lomb-Scargle

Mixed Hardware and Software Embedded Signal Processing Methods for in-situ Analysis of Cardiac Activity

307

Time (s)

0 10 20 30 40 50 60

RR-Interval (ms)

700

800

900

1000

Frequency (Hz)

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

Norm. PSD (/Hz)

-7

-5

-3

-1

Figure 4: Theoritical continuous tachogram of generated

RR-intervals and corresponding normalized PSD.

periodogram was implemented and optimized using

trigonometric recurrences (Press, 2007). The fast im-

plementation of the periodogram (Press and Rybicki,

1989) was also implemented, and tested with three

different values for the MACC factor (1, 2 and 4).

Figure 5 shows the result of the different methods

applied on RR-intervals during 5 minutes. Also Ta-

ble 3 presents the calculation time for each method

with the MCU set at its lower frequency (3 MHz) for

reducing current consumption.

Table 3: Calculation time for each method of PSD estima-

tion of RR-intervals for 5-minute long segments.

Type Time (s)

Original LP > 60

Fast LP (MACC = 4) 4

Fast LP (MACC = 2) 2.5

Fast LP (MACC = 1) 1

This results clearly shows that the original Lomb-

Scargle implementation is not usable due to the espe-

cially long time of calculation (over 1 minute). How-

ever, the fastest implementation (MACC = 1) which

takes only 1 second to calculate, adds considerable

noise to the original PSD with a level around -70 dB.

Finally the fast implementation with a MACC factor

of 2 seems to be the best compromise between calcu-

lation time and noise level as it does not excess the

level of the original one at -100 dB.

3.3 Evaluation of Power Consumption

Together with robustness and accuracy, one of the

main objectives of the implementation of mixed hard-

ware and software method for HR and HRV measure-

ment is the optimization of power consumption of the

LP (/Hz)

-5

-3

-1

LPF 1 (/Hz)

-5

-3

-1

LPF 2 (/Hz)

-5

-3

-1

Frequency (Hz)

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

LPF 4 (/Hz)

-5

-3

-1

Figure 5: Normalized periodogram over of a generated 300-

second buffer of RR-intervals using original Lomb-Scargle

periodogram (LP) and its fast implementation for a MACC

factor of 1,2 and 4 (LPF 1, LPF 2 and LPF 4 respectively).

REC Heart Activity Sensor. In this section we present

and discuss the result of power consumption of the

different parts of the system, composed of the analog

ECG processing, analog to digital conversion of the

signal, digital HR measurement, and software HRV

calculation. Also additional power consumption due

to wireless communication must be taken into account

to estimate the overall consumption of the system.

Table 4 summarizes the power consumption of the

different modules with the conﬁguration used, by tak-

ing into account the duty cycle of module active time

over a period of 5 minutes. The consumption of the

RF module corresponds to a BLE connection with an

Android device where every new HR and HRV values

are sent in real time. The system is powered at 3.3

V using a linear voltage regulator having a negligible

quiescent current and a 300 mAh one-cell Lithium-

Ion battery, which enables to estimate the global au-

tonomy of the sensor device.

The total resulting current consumption leads to

a theoretical autonomy of 54 hours. This results is

already large enough for very long term monitoring

of cardiac activity during in-situ experiment. Indeed

charging the battery, thanks to embedded micro-USB

connector on the device, only less than 1 hour by us-

ing an ordinary USB charger. This can be done once

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

308

Table 4: Average current consumption of the different part

of the REC Heart Activity sensor.

Module Average current

Device base 1.83 mA

Radio module 0.82 mA

Analog Front-end 2.44 mA

Analog to digital conversion 0.25 mA

Hardware HR measurement 0.14 mA

Software HRV calculation 0.02 mA

Total 5.5 mA

every two days during a short period when the de-

vice is not used (for example it can be done during

the daily time spent in the bathroom, where the de-

vice has to be removed).

On the other hand, the results show that a impor-

tant contribution to the actual current consumption is

due to analog front end which is composed of the in-

tegrated ampliﬁers for the differential ampliﬁcation as

well as the RLD circuit. This could be reduced by us-

ing existing discrete components which are optimized

for low-power applications and then extends further

the autonomy of the device with equal signal quality.

In conclusion, regarding the hardware HR mea-

surement and HRV calculation methods, the evalua-

tion has validated the advantage of combining avail-

able PLDs for real-time detection and measurement

of HR with an optimized method for the calculation

of short term HRV parameters, both in time and fre-

quency domains, directly on the embedded system.

4 CONCLUSION

The objective of this study was to evaluate possibil-

ities of taking advantage of a programmable system-

on-chip in order to combine optimized methods for

a complete, real-time monitoring and analysis of car-

diac activity directly on a wearable sensor. This was

done by using a PSoC5 LP, which combines :

• Integrated, programmable analog components,

which were used to build the analog ECG front-

end;

• Integrated digital ﬁlter components for a hardware

R-peak detection and RR-interval measurement;

• 32-bit ARM Cortex M3 micro-controller unit

for an embedded calculation of time-domain and

frequency-domain HRV parameters.

The main advantage of using a PSoC5 LP was to

have the entire ECG process, HR and HRV calcula-

tions fully integrated in a small, single chip. The Pan

and Tompkins’ method for R-peak detection was im-

plemented as a non-linear ﬁlter to beneﬁts from the

ultra-low power digital ﬁlter block, combined with a

local maximum detector using a dynamic threshold

for robust detection. The Press and Rybicki’s fast

algorithm for spectral analysis was adapted to pro-

vide a better estimation of PSD by the use of method

dedicated to unvenly sampled data rather than FFTs,

with fast enough calculation time compared to the

original implementation of the Lomb-Scargle peri-

odogram. A future optimization could be the use of a

dedicated analog front-end rather than the integrated

programmable-gain ampliﬁers which get higher cur-

rent consumption than commercially available dis-

crete components or ECG ampliﬁers.

However the REC Heart Activity sensor is already

proposed as solution for a better real-time assessment

of cardiac activity by providing not only HR mea-

surement but also both time-domain and frequency-

domain HRV parameters, calculated according to in-

ternational standards for HRV analysis.

Moreover this device can be used within a wire-

less body sensor network, together with the sensors

designed in the frame of the RECAMED project, as

well as a software platform on smartphone for col-

lecting, storing, and passing on data securely. This

WBSN is proposed as a solution for the increasing

clinical need of automated collection of health data

from multiple patients, both inside and outside of a

medical environment (hospital or nursing home).

ACKNOWLEDGEMENTS

The REC Heart Activity sensor is developed in the

frame of the RECAMED project, funded by the

BQR’s program at INSA Lyon.

REFERENCES

Altini, M., Polito, S., Penders, J., Kim, H., Van Helleputte,

N., Kim, S., and Yazicioglu, F. (2011). An ecg patch

combining a customized ultra-low-power ecg soc with

bluetooth low energy for long term ambulatory moni-

toring. In Proceedings of the 2nd Conference on Wire-

less Health, page 15. ACM.

Chen, Y.-H., Op de Beeck, M., Vanderheyden, L., Mi-

hajlovic, V., Grundlehner, B., and Van Hoof, C.

(2013). Comb-shaped polymer-based dry electrodes

for eeg/ecg measurements with high user comfort.

In Engineering in Medicine and Biology Society

(EMBC), 2013 35th Annual International Conference

of the IEEE, pages 551–554. IEEE.

Mixed Hardware and Software Embedded Signal Processing Methods for in-situ Analysis of Cardiac Activity

309

Clifford, G. D. and Tarassenko, L. (2005). Quantifying er-

rors in spectral estimates of hrv due to beat replace-

ment and resampling. Biomedical Engineering, IEEE

Transactions on, 52(4):630–638.

Go, A. S., Mozaffarian, D., Roger, V. L., Benjamin, E. J.,

Berry, J. D., Blaha, M. J., Dai, S., Ford, E. S., Fox,

C. S., Franco, S., et al. (2014). Heart disease and

stroke statistics–2014 update: a report from the amer-

ican heart association. Circulation, 129(3):e28.

Lomb, N. R. (1976). Least-squares frequency analysis of

unequally spaced data. Astrophysics and space sci-

ence, 39(2):447–462.

Malik, M., Bigger, J. T., Camm, A. J., Kleiger, R. E.,

Malliani, A., Moss, A. J., and Schwartz, P. J. (1996).

Heart rate variability standards of measurement, phys-

iological interpretation, and clinical use. European

heart journal, 17(3):354–381.

Massot, B., Noury, N., Gehin, C., and McAdams, E. (2013).

On designing an ubiquitous sensor network for health

monitoring. In 2013 IEEE 15th International Confer-

ence on e-Health Networking, Applications and Ser-

vices (Healthcom) (IEEE Healthcom 2013), Lisbon,

Portugal.

McAdams, E., Krupaviciute, A., Gehin, C., Grenier, E.,

Massot, B., Dittmar, A., Rubel, P., and Fayn, J. (2011).

Wearable sensor systems: The challenges. In Engi-

neering in Medicine and Biology Society,EMBC, 2011

Annual International Conference of the IEEE, pages

3648 –3651.

Pan, J. and Tompkins, W. (1985). A real-time qrs detection

algorithm. IEEE transactions on bio-medical engi-

neering, 32(3):230–236.

Press, W. H. (2007). Numerical recipes 3rd edition: The art

of scientiﬁc computing. Cambridge university press.

Press, W. H. and Rybicki, G. B. (1989). Fast algorithm

for spectral analysis of unevenly sampled data. The

Astrophysical Journal, 338:277–280.

Romero, I., Grundlehner, B., and Penders, J. (2009). Robust

beat detector for ambulatory cardiac monitoring. In

Engineering in Medicine and Biology Society, 2009.

EMBC 2009. Annual International Conference of the

IEEE, pages 950–953. IEEE.

Scargle, J. D. (1982). Studies in astronomical time se-

ries analysis. ii-statistical aspects of spectral analysis

of unevenly spaced data. The Astrophysical Journal,

263:835–853.

Sch

afer, A. and Vagedes, J. (2013). How accurate is pulse

rate variability as an estimate of heart rate variabil-

ity?: A review on studies comparing photoplethysmo-

graphic technology with an electrocardiogram. Inter-

national journal of cardiology, 166(1):15–29.

Shin, K., Minamitani, H., Onishi, S., Yamazaki, H., and

Lee, M. (1994). The direct power spectral estimation

of unevenly sampled cardiac event series. In Engi-

neering in Medicine and Biology Society, 1994. En-

gineering Advances: New Opportunities for Biomedi-

cal Engineers. Proceedings of the 16th Annual Inter-

national Conference of the IEEE, pages 1254–1255.

IEEE.

Van Hoof, C. and Penders, J. (2013). Addressing the health-

care cost dilemma by managing health instead of man-

aging illness: an opportunity for wearable wireless

sensors. In Proceedings of the Conference on De-

sign, Automation and Test in Europe, pages 1537–

1539. EDA Consortium.

Winter, B. B. and Webster, J. G. (1983). Driven-right-leg

circuit design. Biomedical Engineering, IEEE Trans-

actions on, 30(1):62–66.

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

310