biosensors
Article
An Ultra-Low Power Surface EMG Sensor for Wearable
Biometric and Medical Applications
Yi-Da Wu 1 , Shanq-Jang Ruan 1






Citation: Wu, Y.-D.; Ruan, S.-J.;
Lee, Y.-H. An Ultra-Low Power
Surface EMG Sensor for Wearable
Biometric and Medical Applications.
Biosensors 2021, 11, 411. https://
doi.org/10.3390/bios11110411
Received: 4 September 2021
Accepted: 19 October 2021
Published: 21 October 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Biosensors 2021, 11, 411. https://doi.org/10.3390/bios11110411
and Yu-Hao Lee 2, *
1 Department of Electronic and Computer Engineering, National Taiwan University of Science and Technology,
Taipei 106, Taiwan; m10802114@mail.ntust.edu.tw (Y.-D.W.); sjruan@mail.ntust.edu.tw (S.-J.R.)
2 Department of Physical Medicine and Rehabilitation, Shuang Ho Hospital, Taipei Medical University,
Taipei 106, Taiwan
* Correspondence: 15411@s.tmu.edu.tw
Abstract: In recent years, the surface electromyography (EMG) signal has received a lot of attention.
EMG signals are used to analyze muscle activity or to evaluate a patient’s muscle status. However,
commercial surface EMG systems are expensive and have high power consumption. Therefore, the
purpose of this paper is to implement a surface EMG acquisition system that supports high sampling
and ultra-low power consumption measurement. This work analyzes and optimizes each part of
the EMG acquisition circuit and combines an MCU with BLE. Regarding the MCU power saving
method, the system uses two different frequency MCU clock sources and we proposed a ping-pong
buffer as the memory architecture to achieve the best power saving effect. The measured surface
EMG signal samples can be forwarded immediately to the host for further processing and additional
application. The results show that the average current of the proposed architecture can be reduced
by 92.72% compared with commercial devices, and the battery life is 9.057 times longer. In addition,
the correlation coefficients were up to 99.5%, which represents a high relative agreement between the
commercial and the proposed system.
Keywords: EMG acquisition system; biosensor devices and interface circuit; wireless transmission;
power consumption
1. Introduction
An increasing amount of research on combining bioelectrical signals with the Internet
of Things (IoT) has been undertaken in the last decade. More biometric sensor products
have been developed, with the most popular product being the surface electromyogram
(EMG). Surface EMGs have been used extensively in various applications such as games,
rehabilitation medicine, analysis of motion, analysis of muscle fatigue, and prosthesis
control [1–4]. The surface EMG sensor is able to record muscle activity by using EMG
electrodes to measure the changes in the electrical potential between two points of a
muscle [5]. Furthermore, the sensor drastically decreases bacterial infection risk because of
its non-invasive measurement [6].
Comparisons of commercial products have found that although many surface EMG
sensors on the market claim to be low-power and high-sampling, the lowest power con-
sumption among these products is up to 46.25 mW [7–10]. The battery life depends on the
battery capacity, and the maximum battery life is only 8 h. Therefore, many researchers have
investigated low-power and high sampling surface EMG systems for long-term recording
and applied them in different fields. Brunelli et al. [4] developed a wireless multi-channel
surface EMG prosthetic sampling measurement system, using 240 Kbps speed Bluetooth
technology. The research used 32-channel surface EMG sensors to sample gesture signals.
The gesture signals are extracted and classified as features of hand movement and surface
EMG signals are used to control the complex gestures of prosthetic hands. However, the
power consumption was found to be as high as 160 mW. Giorgi et al. [11] implemented
https://www.mdpi.com/journal/biosensors
Biosensors 2021, 11, 411 2 of 15

Biosensors 2021, 11, 411 However, the power consumption was found to be as high as 160 mW. Giorgi2et al. [11]
of 15
implemented a wireless surface EMG and accelerometer signal sensing system. The sys-
tem combines surface EMG sensors with accelerometers and is used for muscle fatigue
adetection. The results
wireless surface EMG of andthe research show
accelerometer that
signal when system.
sensing athletesThe wear this system
system combines device,
they obtain
surface EMGbetter
sensorstraining results by correcting
with accelerometers the muscle
and is used for muscle angle to obtain
fatigue more strength.
detection. The
The device uses ZigBee wireless transmission technology and
results of the research show that when athletes wear this system device, they obtaintransmits data atbetter
a speed of
2 Ksps, results
training but thebypower is as high
correcting as 169.3
the muscle mW.
angle toThe twomore
obtain examples above
strength. Thedemonstrate
device uses that
the surface
ZigBee EMG
wireless sampling technology
transmission system hasand been successfully
transmits data atimplemented.
a speed of 2 Ksps,Theybut have
the been
optimized
power for use
is as high in different
as 169.3 mW. The fields, and haveabove
two examples greater application
demonstrate forthe
that daily use EMG
surface than com-
sampling
merciallysystem has products.
available been successfully
However, implemented. They from
they still suffer have thebeenproblem
optimized for use
of high-power
in different
consumption. fields, and have greater application for daily use than commercially available
products.
The However,
purpose of they
thisstill
studysuffer
wasfrom the probleman
to implement ofultra-low-power
high-power consumption.surface EMG acqui-
The purpose of this study was to implement an ultra-low-power
sition system for high-sampling measurement. In the circuit, we designed surface EMG acquisi- sur-
a wireless
tion
face EMG acquisition system that consists of an instrumentation amplifier, ansurface
system for high-sampling measurement. In the circuit, we designed a wireless analog fil-
EMG acquisition system that consists of an instrumentation amplifier, an analog filter, and a
ter, and a wireless Soc module. In regard to the analog filter, we studied and optimized
wireless Soc module. In regard to the analog filter, we studied and optimized the common
the common filter architecture. We used a Bluetooth low energy (BLE) module as the wire-
filter architecture. We used a Bluetooth low energy (BLE) module as the wireless SoC
less SoCand
module, module, and embedded
embedded a ping-pong a buffer
ping-pong buffer mechanism
mechanism on the memory on the memory The
architecture. architec-
ping-pong buffer is a power-saving mechanism; it also improves the BLE data transmission data
ture. The ping-pong buffer is a power-saving mechanism; it also improves the BLE
transmission
efficiency. efficiency.
The data The data
is transmitted by is
thetransmitted
BLE moduleby to the BLE moduleThe
the smartphone. to the smartphone.
smartphone
The smartphone uses an infinite impulse response filter (IIR filter)
uses an infinite impulse response filter (IIR filter) to remove noise and display the data to remove noise
on and
display
the screenthe data on
(Figure 1). the screen (Figure 1).

Figure1.1.Overview
Figure Overviewofof the
the proposed
proposed surface
surface EMG
EMG acquisition
acquisition system.
system.

Theremainder
The remainderofof this
this paper
paper is organized
is organized as follows.
as follows. Section
Section 2 describes
2 describes the details
the details
and functionality of various components used in the sensor. We design the instrumentation
and functionality of various components used in the sensor. We design the instrumenta-
amplifier and theand
tion amplifier analog
the filter
analog forfilter
the surface
for theEMG signal
surface processor
EMG signalanalog circuit
processor and we
analog circuit
propose
and weapropose
power-saving method based
a power-saving on a ping-pong
method based on abuffer as the buffer
ping-pong memory asarchitecture
the memory ar-
in Section 3.in
chitecture The experimental
Section results are presented
3. The experimental results and discussed in
are presented Section
and 4. Finally,
discussed the
in Section 4.
conclusion is provided in Section 5.
Finally, the conclusion is provided in Section 5.
2. Surface EMG Architecture
2. Surface EMG Architecture
Surface EMG measurement usually uses a preamplifier to output the voltage difference
betweenSurface EMG voltage
the muscle measurement usuallytouses
at two points a preamplifier
the next stage. Whentothe output
EMGthe voltage
signal differ-
passes
through various tissues, various noises will interfere with the signal. Therefore, a band-passsignal
ence between the muscle voltage at two points to the next stage. When the EMG
passes
filter through
(BPF) various
and notch tissues,
filter various
are used noises
to remove will
this interfere
noise. with
Lastly, the signal. Therefore,
a microcontroller unit a
band-pass
(MCU) with filter (BPF) and notch
an analog-to-digital filter are
converter used
(ADC) to remove
is used this
to collect thenoise. Lastly,
surface EMG asignals
microcon-
(Figure
troller 2).
unit (MCU) with an analog-to-digital converter (ADC) is used to collect the surface
EMG signals (Figure 2).
2.1. Instrumentation Amplifier
The preamplifier is a key element in the bioelectric signal measurement system, and
an instrumentation amplifier (In-Amp) is usually used as the preamplifier. The In-Amp
is improved by a differential amplifier so that it has low DC offset (Vos ) and low noise. In
the surface EMG measurement system, a common-mode rejection ratio (CMRR) of more
than 100 dB, a high bandwidth (BW), and an input impedance (Zin ) greater than 100 GΩ
are required, at least [12–14].
 Biosensors 2021, 11, 411 3 of 15
Biosensors 2021, 11, 411 3 of 15

Figure 2. Surface EMG measurement sensor block diagram. The EMG signal is represented b
and the noise signals by ‘n’.

2.1. Instrumentation Amplifier

The preamplifier is a key element in the bioelectric signal measurement system
an instrumentation amplifier (In-Amp) is usually used as the preamplifier. The In-Am
improved by a differential amplifier so that it has low DC offset (𝑉𝑜𝑠 ) and low noise. I
surface EMG measurement system, a common-mode rejection ratio (CMRR) of more
Figure 2. Surface100
Figure 2. Surface dB,
EMG
EMG ameasurement
high bandwidth
measurement (BW),
sensor
sensor and
block
block anThe
input
diagram.
diagram. impedance
The
EMG EMG is
signal (𝑍𝑖𝑛is
signal ) greater
by than
‘M’ 100
represented
represented GΩ
by ‘M
and the noise required,
signals by at
‘n’.
and the noise signals by ‘n’. least [12–14].

2.2. Processing Surface EMG Noise of Filter

2.1. Instrumentation
2.2. Processing Surface Amplifier
EMG Noise of Filter
2.2.1. The
Active 2.2.1. Active Filter
Filter
preamplifier is a key element in the bioelectric signal measurement system, and
The active filterThe
an instrumentation has active
good filter
amplifier
has good
isolation
(In-Amp) between isolation
the two
is usually
between
usedstages. the
as the
twoprovide
It can stages. Ithigh
preamplifier.
can input
provide high i
The In-Amp is
impedance by impedance
anda low and low
output impedance, output andimpedance, and
it isitconvenient it is convenient
for offset
connecting for connecting
with low
the front with the
improved differential amplifier so with
that has low DC (𝑉𝑜𝑠 ) and noise. In the
and rear circuits. Compared with passive filters, the active filter does not have to take into to take
and rear circuits. Compared passive filters, the active filter does not have
surface EMG measurement
account system, a common-mode
each stage’s frequency-dependent rejection ratio (CMRR) of more thanf
account each stage’s frequency-dependent loading of theloading preceding of the
stage.preceding stage. Active
Active filters
100 dB, a highcan bandwidth
adjust of
can adjust the sensitivity
(BW),
thethe
sensitivityandofan
frequency theinput
frequency
response
impedance
response
by simply
(𝑍by
setting𝑖𝑛 )simply
greater
the Quality
than
setting 100Quality
the
Factor
GΩ are F
required, at least
(Q [12–14].
Factor). Usually, the Q factor is designed to
(Q Factor). Usually, the Q factor is designed to be 0.707 because the signal output will have be 0.707 because the signal output
the best flatteninghave the best
effect flattening effect [15–17].
[15–17].
2.2. Processing
Many documentsSurfaceMany
orEMG documents
Noise
articles oforFilter
suggest articles
using asuggest
Sallen–Keyusingfiltera Sallen–Key
to designfilter to design
surface EMG surface
sensors [18–20].sensors
The [18–20]. The
Sallen–Key Sallen–Key
filter has simple filter has simplean
architecture, architecture,
uncomplicated an uncomplicated
design, de
2.2.1. Active Filter
and can be constructed with a few electronic
and can be constructed with a few electronic components (Figure 3a). However, high- components (Figure 3a). However, high
The active
frequency signals filter
quency has
that signals good
are higher thatisolation
are higher
than between the the
than frequency
the cut-off two(Fc)
cut-off stages.
frequency It(Fc)
can easily canpassprovide
can easily
through highthroug
pass inpu
impedance and low
feedback output impedance,
capacitor (C2) and and
directly it is convenient
output to
the feedback capacitor (C2) and directly output to the next stage. High-frequency signalsthe for
next connecting
stage. with
High-frequency the fron
signal
and rear circuits.
that cannot cannot be
Compared
be eliminated eliminated will
with passive
will increase increase unnecessary
filters, power
unnecessary the active power consumption.
filter doesTherefore,
consumption. not have to Therefore,
wetake intowe
account
proposed each posed
stage’s
the use the use of the multiple
frequency-dependent
of the multiple feedback
feedback filter (MFB loading filter (MFB
filter)oftothe filter)
preceding
effectively to effectively
solvestage. solve the
Active filters
the problem pro
of high frequencyof high
signal frequency
transmission signal
to transmission
the next stage to the
(Figure
can adjust the sensitivity of the frequency response by simply setting the Quality next
3b). stage
Then, (Figure
we 3b).
experimented Then, Factor
we ex
with the mented
frequency with the
response of frequency
the response
Sallen–Key of and
filter the Sallen–Key
MFB filter filter
in and MFB
low-pass modefilter in low
(Q Factor). Usually, the Q factor is designed to be 0.707 because the signal output wil
(Fc = 500 mode (Fc = 500 the
Hz).confirmed
Figure 4 shows the frequency
confirmed signals and high frequencyFc ofsignals
the beyon
have the Hz).
best Figure 4 shows
flattening
of the effect filter,
Sallen–Key [15–17].which
and high
cannot be filtered. On the
beyond
other hand, the MFB filter is
Sallen–Key filter, which cannot be filtered. On the other hand, the MFB filter is able to filter
Many documents
to filter out orthe
articles
high suggest using
frequency signal a Sallen–Key filter to design surface EMG
perfectly.
out the high frequency signal perfectly.
sensors [18–20]. The Sallen–Key filter has simple architecture, an uncomplicated design
and can be constructed with a few electronic components (Figure 3a). However, high-fre-
quency signals that are higher than the cut-off frequency (Fc) can easily pass through the
feedback capacitor (C2) and directly output to the next stage. High-frequency signals tha
cannot be eliminated will increase unnecessary power consumption. Therefore, we pro-
posed the use of the multiple feedback filter (MFB filter) to effectively solve the problem
of high frequency signal transmission to the next stage (Figure 3b). Then, we experi-
mented with the frequency response of the Sallen–Key filter and MFB filter in low-pass
mode (Fc = 500 Hz). Figure 4 shows the confirmed and high frequency signals beyond Fc
of the Sallen–Key filter, which cannot be filtered. On the other hand, the MFB filter is able
Figure 3. filter
Figure
to 3. out
filter the high (a)
(a) Sallen–Key Sallen–Key
frequency filter
architecture (b)architecture
signal MFB (b) MFB filter architecture.
filter architecture.
perfectly.

Figure 3. (a) Sallen–Key filter architecture (b) MFB filter architecture.

Biosensors 2021, 11, 411
Biosensors 2021, 11, 411 4 of 15 4

Frequency
Figure 4. Figure response response
4. Frequency of the MFB filter
of the architecture
MFB and Sallen–Key
filter architecture filter. filter.
and Sallen–Key
2.2.2. Notch Filter
2.2.2. Notch Filter
The range of the EMG effective frequency is low (about 20 Hz~500 Hz) [21,22], so the
The range of the EMG effective frequency is low (about 20 Hz~500 Hz) [21,22], s
noise will easily affect the EMG signal. Further, the surface EMG signal will generate a lot
noise will easily affect the EMG signal. Further, the surface EMG signal will generate
of noise when passing through different muscle tissues; these noises can be divided into
of noise when passing through different muscle tissues; these noises can be divided
the following types:
the following types:
• Power line noise
• Power line noise
The most serious interference in surface EMG measurement is electromagnetic inter-
The most serious interference in surface EMG measurement is electromagnetic
ference, which is caused by AC power lines. Its main frequency component is 60 Hz [23]
ference, which is caused by AC power lines. Its main frequency component is 60 Hz
because power line noise is within the effective range of surface EMG. Some scholars
because power line noise is within the effective range of surface EMG. Some scholar
advocate using a notch filter to filter this interference [24,25]. However, other scholars have
vocate using a notch filter to filter this interference [24,25]. However, other scholars
different opinions because 30 Hz~300 Hz is the densest frequency band of EMG signals [26];
different opinions because 30 Hz~300 Hz is the densest frequency band of EMG si
thus, if a notch filter works, important signals may also be filtered out. Therefore, they do
[26]; thus, if a notch filter works, important signals may also be filtered out. There
not advocate such an approach [18,27,28]. As a result, IIR and power spectral density (PSD)
they do not advocate such an approach [18,27,28]. As a result, IIR and power spectral
compensation have been designed to separate power line noise and surface EMG signals
sity (PSD) compensation have been designed to separate power line noise and su
more effectively.
EMG signals more effectively.
• Inherent instability of signal
• Inherent instability of signal
The EMG signalThe can EMG
be affected
signalbycan
thebe
motor units.byInthe
affected most conditions,
motor these
units. In most frequency
conditions, thes
regions are 0 toquency
20 Hz. In this case,
regions are 0the
to noise
20 Hz.isInconsidered as unnecessary,
this case, the so it is as
noise is considered essential
unnecessary, s
to remove it cleanly.
essential to remove it cleanly.
2.3. Storage Method of Wireless
2.3. Storage Embedded
Method System
of Wireless Embedded System
2.3.1. The Use of Wireless Transmission Technology
2.3.1. The Use of Wireless Transmission Technology
Surface EMG signal detection can be achieved by utilizing wireless technologies.
Surface EMG signal detection can be achieved by utilizing wireless technolo
Many wireless transmission technologies are available today, such as 2G/3G wireless
Many wireless transmission technologies are available today, such as 2G/3G wireles
networks, LoRa, Zigbee, etc. However, 2G/3G wireless networks are costly and consume
works, LoRa, Zigbee, etc. However, 2G/3G wireless networks are costly and consume
high amounts of power, and LoRa and Zigbee have low data rates. Therefore, BLE and
amounts of power, and LoRa and Zigbee have low data rates. Therefore, BLE and W
Wi-Fi are the most popular wireless technology used in our daily life. In this paper, Wi-Fi
are the most popular wireless technology used in our daily life. In this paper, Wi-Fi
modules and BLE modules with SoC are investigated and evaluated based on their data
ules and BLE modules with SoC are investigated and evaluated based on their
throughput. Wi-Fi technology, for example, ESP8266EX [29] has a data transmission speed
throughput. Wi-Fi technology, for example, ESP8266EX [29] has a data transmission s
of 4.5 MB/s and an average current of 80 mA, while the power consumption is much
of 4.5 MB/s and an average current of 80 mA, while the power consumption is much l
lower for BLE technology than Wi-Fi. As a result, we used nRF52832 [30] as a wireless
for BLE technology than Wi-Fi. As a result, we used nRF52832 [30] as a wireless tran
transmission platform. When the DC–DC low power consumption mode is enabled, the
sion platform. When the DC–DC low power consumption mode is enabled, the av
average current is only 1.32 mA and the data transmission speed can reach 171.5 KB/s.
current is only 1.32 mA and the data transmission speed can reach 171.5 KB/s.
Biosensors
Biosensors
Biosensors 2021,11,
Biosensors2021,
2021,
2021, 11,411
11,
11, 411
411 555ofof
of1515
15

2.3.2.
2.3.2.
2.3.2. Improvement
2.3.2.Improvement
Improvementin
Improvement in
in Data
inData
Data Processing
DataProcessing Efficiency
EfficiencyBased
ProcessingEfficiency
Processing Efficiency Basedon
Based
Based onPing-Pong
on
on Ping-PongBuffer
Ping-Pong
Ping-Pong Buffer
Buffer
Buffer
Figure
Figure5555describes
Figure
Figure describes
describesthe
describes the
the common
thecommon
commonADC
common ADC
ADChardware
ADC hardwaredesign
hardware
hardware designfor
design
design forthe
for
for thefirst
the
the firstinput
first
first inputfirst
input
input firstoutput
first output
output
(FIFO)
(FIFO)
(FIFO) process
process
process
(FIFO) process of
ofof data
data
data allocation.
allocation.
allocation. After
After
After the
the
the MCU
MCU
MCU finishes
finishes
finishes processing
processing
processing
data allocation. After the MCU finishes processing the surface EMG the
the
the surface
surface
surface EMG
EMG sig-
sig-
nal
nal
signaldata,
naldata,
data, the
the MCU
MCU
thethe
data, MCU MCU calls
calls
calls on
on
calls the
onthe ADC
ADC
thethe
on ADC ADChardware
hardware
hardware
hardware to
to start
tostart
start collecting
collecting
collecting
to start collectingdata,
data,
data, and
and
andand
data, MCU
MCU
MCU MCUwaits
waits for
for
waits
the
the Interrupt
Interrupt
thethe
for Interrupt Service
Service
InterruptService
ServiceRoutine
Routine
Routine
Routine(ISR)
(ISR)
(ISR) to
to
(ISR) be
tobebe triggered.
totriggered.
triggered.
be triggered. When
When
When the
When theADC
the ADC
ADC
the hardware
ADChardware
hardware
hardware finishes
finishes
finishes ac-
ac-
finishes
quiring
quiring
acquiring data,
quiringdata,
data, the
the
data, MCU
thethe
MCU
MCU MCU processes
processes
processes
processes the
the ADC
theADCADC
the data,
data,
data,
ADC and
andthe
and
data, theADC
the
and ADC
ADC
the ADChardware
hardware
hardware waits
hardware waitsfor
waits forthe
for
waits MCU
thefor
MCUthe
toto
MCU finish
tofinish
finish processing
processing
processing
to finish data
data
processing again,
dataagain,
again,
data repeating
repeating
repeating
again, this
thisaction
this
repeating action
action
this indefinitely.
indefinitely.
indefinitely.
action indefinitely. However,
However,
However,
However, this kind
thisthis
this kind ofof
kind
situation
situation
situation
of situation will
will
will lead
lead
lead
will lead to
to discontinuous
totodiscontinuous
discontinuous collection
discontinuouscollectioncollection
collectionof of
of signals,
signals,
ofsignals, lower
lower
signals,lower system utilization
system utilization
lower system and
utilization and
wasted power
wastedpower
wasted consumption.
powerconsumption.
consumption.In In order
Inorder
orderto to make
tomake
makesuresure data
sure data will
data will not
will not be
not be lost,
be lost, the
thesurface
lost, the surfaceEMG
surface EMG
measurement
measurementsystem
measurement system
systemneeds
system needs
needshigh
needs high
high stability
highstability
stabilityand
stability and timely
andtimely
and transmission.
timelytransmission.
timely transmission.Hence,
transmission. Hence,
Hence,we
Hence, we embedded
weembedded
we embeddeda
aaping-pong
ping-pong
ping-pong
aping-pong buffer
buffer
buffer
buffer mechanism
mechanism
mechanism
mechanism [31–34]
[31–34]
[31–34]
[31–34] in
inininthe
thethe
the MCU
MCU
MCU
MCU architecture
architecture
architecture
architecture (Figure
(Figure
(Figure
(Figure 6).6).
6).
6).

Figure5.5. The
Figure The original
original time
time graph
graph with
with the
the FIFO.
FIFO.
Figure5.5.The
Figure Theoriginal
originaltime
timegraph
graphwith
withthe
theFIFO.
FIFO.

Figure6.6. Improved
Figure Improved time
time graph
graph with
with the
the ping-pong
ping-pong buffer.
buffer.
Figure6.6.Improved
Figure Improvedtime
timegraph
graphwith
withthe
theping-pong
ping-pongbuffer.
buffer.
3.3. Implementation
Implementation Method
3.3.Implementation
ImplementationMethod
In
Method
Inthis
thispaper,
paper,we wepropose
proposeaasurface
surfaceEMGEMGmeasurement
measurementmodulemoduleand andaasmartphone
smartphoneas asa
Inthis
In thispaper,
paper,we wepropose
proposeaasurface
surfaceEMG
EMGmeasurement
measurement module
module andand aa smartphone
smartphone asas
host
a host (Figure
(Figure7).7).The
Thesurface
surfaceEMGEMGmeasurement
measurement module
module captures the the analog
analogEMG
EMGsignal
signal
a
aon host
host (Figure 7). The surface EMG measurement module captures the analog EMG signal
on the(Figure
the skin, 7). The
skin,which
which surface
isissent
sentvia EMG
viathe measurement
thebiceps
bicepsbrachii module
muscles
brachii muscles captures
to to
thethe theelectrodes.
surface
surface analog EMG signal
Then,
electrodes. the
Then,
ononthe
EMG the skin,
skin,
data which
which
are isissent
transmitted sentvia
viathe
to the
the biceps
biceps
host brachii
brachii
using muscles
muscles
wireless for to the
to
further theprocessing.
surface electrodes.
surface electrodes. Then,
Then,
the EMG data are transmitted to the host using wireless for further processing.
theEMG
the EMGdatadataarearetransmitted
transmittedto tothe
thehost
hostusing
usingwireless
wireless for
for further
further processing.
processing.

Figure7.7. Diagram
Figure Diagram of
of the
the proposed
proposed surface
surface EMG
EMG measurement
measurement module
modulesystem.
system.
Figure 7. Diagram of the proposed surface EMG measurement module system.
Figure 7. Diagram of the proposed surface EMG measurement module system.
 Biosensors 2021, 11, 411 6 of 15
Biosensors 2021, 11, 411 6 of 15

3.1. Measurement Module Design

3.1. Measurement
3.1.1. Module
Instrumentation Design with Passive High-Pass Filter
Amplifier
3.1.1. We
Instrumentation Amplifier with
searched the literature to Passive
compare theHigh-Pass
commonly Filter
used low-power In-Amps, and
choseWethe
searched
INA333the literature
device to compare
[35,36] the commonly
as our In-Amp. used low-power
The INA333 In-Amps,
is much better thanand
the
chose
AD620 [37,38] and INA128 [39,40] in 𝑍𝑖𝑛 . Although the quiescent current (𝐼𝑄 ) and 𝑍𝑖𝑛the
the INA333 device [35,36] as our In-Amp. The INA333 is much better than of
AD620 [37,38]
the INA333 areand
10 INA128
μA and [39,40]
10 GΩ in Zin . Although
higher thethe
than that of quiescent
AD8236current
[41–43],(Ithe
Q ) and Zinhas
IN333 of
the INA333
better are 10 µA
performance and
with 10 GΩ
regard tohigher
BW, CMRR,than that of and
noise the AD8236 [41–43],
𝑉𝑜𝑠 (Table 1). the IN333 has
better performance with regard to BW, CMRR, noise and Vos (Table 1).
Table 1. Comparison of In-Amp parameters (gain = 100).
Table 1. Comparison of In-Amp parameters (gain = 100).
BW CMRR 𝒁𝒊𝒏 Noise 𝑽𝒐𝒔 𝑰𝑸
Chip BW CMRR Zin Noise
√ Vos IQ
Chip (kHz) (dB) (GΩ) (𝐧𝐕/√Hz) (μV) (μA)
(kHz) (dB) (GΩ) (nV/ Hz) (µV) (µA)
INA333 3.5 115 100 50 1 50
INA333 3.5 115 100 50 1 50
AD8236
AD8236
0.8
0.8
110
110
110
110
76
76
2.5
2.5
40
40
AD620
AD620 120
120 130
130 10
10 28
28 50
50 900
900
INA128
INA128 200
200 120
120 10
10 88 50
50 700
700

The INA333 requires a high resistor (R2 and R3) on the input pin to form the input
The INA333
bias current requires
return a high 8).
path (Figure resistor (R2 and R3results
This approach ) on theininput pinhigh-frequency
a better to form the input bias
CMRR
current return path (Figure 8). This approach results in a better high-frequency
and lower 𝑉𝑜𝑠 [35,44,45]. In regard to the In-Amp power supply, the In-Amp is designed CMRR and
lower V
for single supply mode. Since the EMG signal has both positive and negative signals, for
os [35,44,45]. In regard to the In-Amp power supply, the In-Amp is designed the
single
input supply mode.return
bias current Since path
the EMG signala has
provides biasboth positive
voltage andand
(Vcc/2) negative
allowssignals,
the EMG the signal
input
bias current
to float on return path
the bias provides
voltage a bias
[44,46]. Wevoltage (Vcc/2)
increased theand allows the
capacitors in EMG signal
the input to float
path and
on the bias voltage [44,46]. We increased the capacitors in the input path
formed a first-order passive high-pass filter by adding resistance. The first-order passive and formed a
first-order passive high-pass filter by adding resistance. The first-order passive
high-pass filter’s Fc is designed to 20 Hz. The equation for the first-order passive high- high-pass
filter’s Fc isisdesigned
pass filter shown intoEquation
20 Hz. The (1): equation for the first-order passive high-pass filter is
shown in Equation (1):
1
Fc =Fc =1 . . (1)
(1)
2𝜋RC
2πRC

Figure 8. In-Amp
Figure8. In-Amp with
with passive
passivehigh-pass
high-passfilter
filterarchitecture.
architecture.

The In-Amp’s gain is determined by the value of R1. Since the maximum frequency of
The In-Amp’s gain is determined by the value of R1. Since the maximum frequency
EMG is 500 Hz, and according to Nyquist sampling theorem, the stable signal needs to be
of EMG is 500 Hz, and according to Nyquist sampling theorem, the stable signal needs to
at 1 KHz. When the gain of INA333 is at 100 (Figure 9), signals within 1 KHz can be kept
be at 1 KHz. When the gain of INA333 is at 100 (Figure 9), signals within 1 KHz can be
stable as shown in the Equation (2):
kept stable as shown in the Equation (2):
Gain ==
Gain 1+
1+(100 KΩ/R1
(100 ). .
KΩ/𝑅1) (2)
(2)
Biosensors 2021, 11, 411 7 of 15
Biosensors 2021, 11, 411 7 of 15
Biosensors 2021, 11, 411 7 of 15

Figure 9. Gain vs. frequency.

Figure 9.
Figure 9. Gain vs. frequency.
frequency.
3.1.2. Design of the MFB Low-Pass Filter
3.1.2.The
3.1.2. appropriate
Design
Design of
of the
the MFB
MFBbandwidth
Low-Pass
Low-Pass ofFilter
an EMG signal is maximized to 500 Hz, as mentioned
Filter
in Section
The 2.2.2. We chose a 1 μF capacitor for 𝐶3 isand 0.1 μF capacitor for 𝐶mentioned
in a low-
The appropriate
appropriate bandwidth
bandwidth of of an
an EMG
EMG signal signal is maximized
maximized to to 500
500 Hz,
Hz, asas 4mentioned
pass filter,2.2.2.
in then the chose
Okawa Electric Designfor calculator
C3 𝐶andand was µFutilized to compute
for C the param-
in Section
Section 2.2.2.We We chosea a1 1µFμF capacitor
capacitor for 3
0.1 0.1 capacitor
μF capacitor 4 in𝐶a low-pass
for 4 in a low-
eters
filter, and the
then the results for 𝑅 and 𝑅 were 506.605 Ω and 𝑅 was 2 kΩ (Figure 10). Because
pass filter, thenOkawa Electric
the Okawa 4 Design
6
Electric calculator
Design was utilized
calculator 5
was utilizedto compute the parameters
to compute the param-
the
and closest
etersthe
to
andresults
506.6 Ω
for Rfor
the results
on
4 and
the
𝑅4 Rand
market
6 were was
𝑅6 506.605
510
were 506.605 Ω andΩRand
Ω, we chose
5 was 510
𝑅52was
Ω to replace
kΩ 2(Figure 506.6
10). Because
kΩ (Figure
Ω. After
the
10). Because
closest to 506.6 Ω on the market was 510 Ω, we chose 510 Ω to replace 506.6 Ω. After
re-calculating, we designed the actual Fc of the filter to be 498.33 Hz. We used the error
the closest to 506.6 Ω on the market was 510 Ω, we chose 510 Ω to replace 506.6 Ω. After
percentage of our
re-calculating, desired values
we designed and Fc
the actual compared
of the filter them to to
be our
498.33actual
Hz. values.
We used Finally, we
the error
re-calculating, we designed the actual Fc of the filter to be 498.33 Hz. We used the error
found that our
percentage Fc had
of our an error
desired of 0.334%
values and comparedas calculated themby to Equation
our actual (3):
values. Finally, we
percentage of our desired values and compared them to our actual values. Finally, we
found that our Fc had an error of 0.334% as |𝑎 calculated
− 𝑏| by Equation (3):
found that our Fc had an error of 0.334% 𝛿 = as calculated ∙ 100%by , Equation (3): (3)
|𝑎|
| a|𝑎−−b𝑏| |
where 𝛿 is the percent error, 𝑎 is the δ 𝛿==theoretical ∙ 100%
·100%,
value, and 𝑏 is the experimental value. (3)
(3)
| a|𝑎|
|
In this paper, the error rate of resistance and capacitance is 1%. Therefore, an error per-
where 𝛿 is the percent error, 𝑎 is the theoretical value and 𝑏 is the experimental value.
where
centageδ is the percent
below error, a is the
1% is reasonable. theoretical
According value and
to Section b is
2.2.1, thethe
Qexperimental
factor of 0.707value. In
is calcu-
In
thisthis paper,
paper, the error
the error(4): rate of resistance and capacitance is 1%. Therefore,
rate of resistance and capacitance is 1%. Therefore, an error percentage an error per-
lated by Equation
centage1%
below below 1% is reasonable.
is reasonable. According According
to Section to 2.2.1,
Section the2.2.1, the Qoffactor
Q factor 0.707 of
is 0.707 is calcu-
calculated by
lated by Equation
Equation (4): (4): √
𝐶3
𝐶
Q factor = q C𝑅3𝐶 4 𝑅 𝑅 .
𝑅
(4)
3
𝑅4𝐶
√ 6 +√C√ 5 +√ 5 6
Q factor = q = 𝑅q
Q factor 5 6 4 q 𝑅4
. . (4)
(4)
R6 𝑅6 R5𝑅5 𝑅5R𝑅56R6
The result of the Q factor is 0.708 (the +
√
R5error+√ ++√
𝑅5 percentage
R6𝑅6 𝑅4R4 = 0.141%).

The result of the Q factor is 0.708 (the error percentage = 0.141%).

Figure 10. Architecture of the MFB low-pass filter.

Figure 10. Architecture of the MFB low-pass filter.
Figure 10. Architecture of the MFB low-pass filter.
Biosensors 2021, 11, 411 8 of 15

Biosensors 2021, 11, 411 8

The result of the Q factor is 0.708 (the error percentage = 0.141%).

The filter outputs an unstable signal in the initial state because it takes some time
The
before the capacitor is filterTo
stable. outputs an the
calculate unstable signal
instability in the
time, we initial state
used the because
Laplace it takes some
transfer
before the
to analyze the transient capacitorThe
response. is stable.
LaplaceTotransfer
calculatefunction
the instability
of the time,
MFB we used the
low-pass Laplace tra
filter
to analyze
is shown in Equation (5): the transient response. The Laplace transfer function of the MFB low-pass
is shown in Equation (5):
Vo −1 𝑉𝑜
1
= ·  −1 1 . . (5)
C3 C4 R4 R5 s2 + s𝑉𝑖 = 1 𝐶 𝐶 𝑅 𝑅 ∙
 
Vi 3 R4 4+5 R 2 1 + 1
R4
+ 1𝑅
4 1
C3 R4 R5 4 𝑠5 +𝑠 R (𝑅4 +𝑅R55(1+
R6 C3 C))+
𝐶3 𝑅46𝑅5 𝑅6 4 𝑅5 𝑅6 𝐶 3 𝐶 4

Figure 11 showsFigure
that the
11 capacitor
shows thatisthe
stable after 2ismilliseconds.
capacitor stable after 2 In order to ensure
milliseconds. a to ens
In order
clean signal, theclean
surface EMG signal in our work is collected after 10 milliseconds.
signal, the surface EMG signal in our work is collected after 10 milliseconds.

Figure 11. Transient analysis of the proposed MFB filter.

Figure 11. Transient analysis of the proposed MFB filter.

3.1.3. Central Control Unit

3.1.3. Central Control Unit
Saving-Power Mechanism
Saving-Power Mechanism
The MCU clockTherate andclock
MCU the square
rate andofthe
thesquare
MCUofvoltage arevoltage
the MCU proportional to the
are proportional to th
dynamic power,namic
and what they
power, consumed
and what theyisconsumed
as shown in Equation
is as (6):Equation (6):
shown in
Pd = CV 2 f , 𝑃𝑑 = 𝐶𝑉 2 𝑓, (6)

where 𝑃𝑑power,
where Pd is the dynamic is theCdynamic power,load
is the switched 𝐶 iscapacitance,
the switched load
V is thecapacitance,
supply voltage, 𝑉 is the su
and f is the MCU voltage,
adjustable 𝑓 is the
and clock rate.MCU adjustable clock
To simultaneously rate. the
reduce To simultaneously
power consumption reduce the p
and finish data consumption
transmission,and it isfinish data to
essential transmission,
minimize the it isMCU
essential to minimize
frequency. Hence, theitMCU
is freque
proposed to use Hence, it is proposed
two different frequencytocrystal use two different
oscillator frequency
sources crystal oscillator
in the MCU—the 64 MHzsources in
high-frequency MCU—the
clock (HFCLK) 64 MHzusedhigh-frequency clock (HFCLK)
in BLE transmission, and the used
16 MHz in BLE transmission, and th
low-frequency
MHz low-frequency clock (LFCLK) is used by
clock (LFCLK) is used by the hardware ADC in the acquisition. Besides, the LFCLK the hardware ADC in the acquisition
is
sides, the LFCLK is enabled when the ADC signal
enabled when the ADC signal is collected to keep the power consumption at a minimum.is collected to keep the power consu
When the EMGtion at a fill
signals minimum. When HFCLK
up the buffer, the EMGwillsignals fill up theand
be activated, buffer, HFCLK
the data willwill
be be activ
and the data will be sent to the host by BLE. This
sent to the host by BLE. This design method can effectively save power. In addition, we design method can effectively
power.regulators
used two low dropout In addition, we used
(LDO) two
in our low dropout
system, regulators
one is always (LDO)
enabled forinthe
our system, one
MCU
power supply, andways enabled
the other is forcontrolled
the MCU power by the supply,
MCU. When and the theother
hostisstarts
controlled
to getby the MCU. W
data,
the GPIO of the MCU becomes high to enable the LED and start to collect EMG signals.LED and
the host starts to get data, the GPIO of the MCU becomes high to enable the
Conversely, when to collect
the hostEMG signals. Conversely,
command does not need when the hostdata
to obtain command
and the does
GPIO notisneed
at ato obtain
low level, the disabled LDO forcibly stops collecting EMG signals, and the MCU enters the EMG sig
and the GPIO is at a low level, the disabled LDO forcibly stops collecting
Biosensors 2021,11,
Biosensors2021, 11,411
411 99 ofof1515

andmode.
idle the MCU enters
After the idle
the MCU mode.
enters the After the MCU
idle mode, enters the
the overall idle mode,
average currentthe overall
will remainaver-
at
age current
1.9 uA. will remain at 1.9 uA.

StoringSurface
Storing SurfaceEMG
EMGData Databy bythe
thePing-Pong
Ping-PongBuffer
BufferMechanism
Mechanism
Accordingto
According tothe
theNyquist
Nyquist sampling
sampling theorem,
theorem, the minimum samplingsampling frequency
frequencyisis1
kHz. In this paper,
1 kHz. In this paper, the MCU was designed to acquire a single channel of 2.049
to acquire a single channel of 2.049 kHz, kHz,the
the
samplingresolution
sampling resolutionwas was14 14bits,
bits,and
andthe
theeasy-to-use
easy-to-usedirect
directmemory
memoryaccess
access(EasyDMA)
(EasyDMA)
interrupttrigger
interrupt triggerwas
wasused.
used.InInorder
ordertotoavoid
avoiddata
dataloss,
loss,the
thedigital
digitalEMG
EMGdata
datawill
willbe
bepushed
pushed
into
intoone
oneofofthe
theping-pong
ping-pongbuffers
buffersinstead
insteadofofbeing
beingdirectly
directlytransmitted
transmittedtotothe
thehost
hostvia
viaBLE.
BLE.
When
Whenthethebuffer
bufferisisfull,
full,an
aninterrupt
interruptwill
willbe
betriggered,
triggered,andandthe
thebuffer
bufferaddress
addresswill
willalso
alsobe
be
switched to other buffers. The full buffer transmits data to the host by going
switched to other buffers. The full buffer transmits data to the host by going through the through the
BLE
BLE(Figure
(Figure12).
12).

Figure12.
Figure 12.The
Theproposed
proposedtwo
twoMCU
MCUclock
clocksources
sourceswith
withping-pong
ping-pongbuffer
bufferarchitecture.
architecture.

3.2.The
3.2. TheHost
HostProcessing
ProcessingwithwithIIRIIR
ToToensure
ensurethe the data
data frequency
frequency is concentrated
is as as concentrated as possible
as possible in thein EMG
the EMG signal
signal range,
range, we
we used two IIR filters to process the EMG signals. One is an eight-order
used two IIR filters to process the EMG signals. One is an eight-order digital Butterworth digital Butter-
worth
IIR IIR band-pass
band-pass filter thatfilter that to
is used is used to eliminate
eliminate noise, andnoise,
the and
otherthe is aother is a fourth-order
fourth-order digital
digital Butterworth
Butterworth IIR band-stop
IIR band-stop filter thatfilter thattoiseliminate
is used used to eliminate the main frequency
the main frequency component com-
of
ponent
the power of line
the power
at 60 Hz.lineTheat 60 Hz. Therange
passband passband range
and the and therange
stopband stopbandare 20range
Hz toare
50020HzHz
to 500
and Hztoand
55 Hz 55 Hz
65 Hz, to 65 Hz, The
respectively. respectively.
equation The equation
for the foris the
IIR filter shownIIR in
filter is shown
Equation (7): in
Equation (7):
K L
y[n] = ∑ 𝐾bk x [n − k] − ∑ a𝐿l y[n − l ], (7)
𝑦[𝑛] =
K =0∑ 𝑏𝑘 𝑥[𝑛 − 𝑘]l =
−1∑ 𝑎𝑙 𝑦[𝑛 − 𝑙], (7)
𝐾=0 𝑙=1
where y[n] is the output signal, K is the feedforward filter order, bk is the feedforward filter
where 𝑦[𝑛]
coefficient, x [nis] is
thethe
output
input signal, 𝐾 isisthe
signal, L thefeedback
feedforwardfilter filter and al𝑏is
order,order, is the
𝑘 the feedforward
feedback filter
filter coefficient, 𝑥[𝑛] is the input signal, 𝐿 is the feedback filter
coefficient. The utilization of FIR-based filters can avoid arithmetic divisions by order, and 𝑎 𝑙 is the feed-
using the
back filter coefficient.
least-square strategy, but The utilization
they of FIR-based
need a higher order tofilters
get thecansame
avoid arithmeticslope
attenuation divisions
as theby
using
IIR theHence,
filter. least-square
the IIR strategy,
filter wasbut theyasneed
chosen a higher
the noise ordermethod
filtration to get the
forsame attenuation
its lower power
slope as the IIR
consumption [47,48]. filter. Hence, the IIR filter was chosen as the noise filtration method for its
lower power consumption [47,48].
4. Results
4. Results
In this paper, the ultra-low power surface EMG signal acquisition system was designed
In this
as shown paper,13.
in Figure theToultra-low power
verify the surface
reliability EMG
of the signal acquisition
proposed system, the system was de-
signal-to-noise
signed
ratio as shown
(SNR), linear in Figure 13.and
correlation To power
verify the reliability were
consumption of thecalculated,
proposed and
system, the signal-
a commercial
to-noiseEMG
wireless ratiodetection
(SNR), linear correlation
system was usedand power
as the consumption
criterion were calculated, and a
reference [7].
commercial wireless EMG detection system was used as the criterion reference [7].
Biosensors 2021, 11, 411 10 of 15
Biosensors 2021, 11, 411 10 of 15

Figure
Figure 13.
13. (a)
(a) Front
Front side
side and
and (b)
(b) back
back side
side of
of the
the ultra-low power surface
ultra-low power surface EMG
EMG signal
signal acquisition
acquisition system.
system.

4.1.
4.1. SNR
SNR
SNR
SNR isis the
the ratio
ratio of
of signal
signal power
power to to the
the noise
noise power,
power, and
and it
it is
is aa form
form of
of measurement
measurement
used
used in certain applied sciences. If the SNR ratio
If the SNR ratio is high, it indicates that a better signal
quality is obtained. ForFor example,
example, a high SNR ratio in audio systems means better sound
quality. Therefore, we used it as a measure of the quality of the EMG signal. To calculate
SNR,
SNR, we we defined
defined 2020to
to500
500Hz
Hzasasthe
thedesired
desiredsignal,
signal,other signals
other signals as as
noise,
noise,and input
and 10
input
mVp-p
10 mVp-p Gaussian
Gaussian white
whitenoise toto
noise the
thedevice
devicechannel
channeltotosimulate
simulatethethenoise
noiseimmunity
immunity of the
sensor. Previous studies reveal that an adequate SNR should be at least 18 dB in a surface
EMG
EMG sensor
sensor[49,50].
[49,50].The
Thecalculation
calculationforfor
thethe
SNRSNRof our proposed
of our proposed system was was
system 23.1 dB.
23.1This
dB.
proves that our
This proves thatproposed system
our proposed not only
system not reduces costs,costs,
only reduces but also
butachieves adequate
also achieves noise
adequate
noise immunity.
immunity.

4.2. Linear
4.2. Linear Correlation
Correlation Coefficient
Coefficient
In this
In this part,
part, we
weused
usedthe
theEMG
EMGsignal
signalofof
thethe
measured
measuredbiceps as aascomparison.
biceps a comparison.First, the
First,
mean absolute value (MAV) was obtained. The MAV ensures that the
the mean absolute value (MAV) was obtained. The MAV ensures that the commercial de- commercial device
signals
vice and and
signals proposed device
proposed signal
device scales
signal are consistent.
scales Then,
are consistent. the EMG
Then, the EMGsignals from
signals the
from
proposed system and the commercial system were measured and processed
the proposed system and the commercial system were measured and processed using the using the same
procedure,
same and compared
procedure, to obtain
and compared the relative
to obtain consistency
the relative betweenbetween
consistency the two the
systems [51]
two sys-
(Figure 14). Finally, a comparison of the EMG signal for each system using
tems [51] (Figure 14). Finally, a comparison of the EMG signal for each system using the the Pearson
product-moment
Pearson correlation
product-moment coefficient
correlation (PPMCC)(PPMCC)
coefficient was usedwas
to illustrate the correlation
used to illustrate of
the cor-
the waveforms. The correlations between the proposed system and the
relation of the waveforms. The correlations between the proposed system and the com-commercial system
were as high as 0.995 (Figure 15).
mercial system were as high as 0.995 (Figure 15).
4.3. Power Consumption
In this part, the current of the sensor was measured in three different states. First,
when the sensor is not connected to the host and the EMG signal is not collected, the
BLE will be in the advertising mode. In the advertising mode, BLE continuously sends
advertisement packets through the 2402 MHz, 2426 MHz and 2480 MHz channels, and the
host can quickly find the sensor and the connection. Consequently, the current is going to
be higher; it will be approximately 5.3 mA. Second, after the host is connected to the sensor,
it will fix the channels and change to the scan mode. Since the host has not yet started to
collect EMG signals at this moment, the current is about 60.64 uA. Finally, when the host
asks the sensor to collect EMG signals, the analog circuit and ADC will start operation, the
current will increase to about 1.269 mA (Figure 16). Figure 17 depicts the reduced time
scale and clearly shows the current change. We also measured the average current of the
surface EMG sensor without the MCU, the average current is about 72.14 uA. Compared to
the literature on surface EMG sensors without wireless communication, we achieved lower
power consumption in our surface EMG sensors [52,53].
Biosensors 2021, 11, 411 11 of 15
Biosensors 2021, 11, 411 11 of 15
Biosensors 2021, 11, 411 11 of 15

Figure14.
Figure 14.The
Theelectrode
electrodeplacement
placementofofthe
thetwo
twosystems.
systems.
Figure 14. The electrode placement of the two systems.

Figure 15. EMG signals from the commercial (blue) and the proposed device (orange).
Figure 15.
Figure 15. EMG
EMG signals
signals from
from the
the commercial
commercial(blue)
(blue)and
andthe
theproposed
proposeddevice
device(orange).
(orange).
4.3. Power Consumption
4.3. Power Consumption
In this part, the current of the sensor was measured in three different states. First,
In this part, the current of the sensor was measured in three different states. First,
when the sensor is not connected to the host and the EMG signal is not collected, the BLE
when the sensor is not connected to the host and the EMG signal is not collected, the BLE
will be in the advertising mode. In the advertising mode, BLE continuously sends
will be in the advertising mode. In the advertising mode, BLE continuously sends
advertisement packets through the 2402 MHz, 2426 MHz and 2480 MHz channels, and
advertisement packets through the 2402 MHz, 2426 MHz and 2480 MHz channels, and
the host can quickly find the sensor and the connection. Consequently, the current is going
the host can quickly find the sensor and the connection. Consequently, the current is going
to be higher; it will be approximately 5.3 mA. Second, after the host is connected to the
to be higher; it will be approximately 5.3 mA. Second, after the host is connected to the
 sensor, it will
started to fixEMG
collect the channels
signals atand
thischange
moment, to the
thecurrent
scan mode. Since
is about theuA.
60.64 host has not
Finally, yet
when
started to collect EMG signals at this moment, the current is about 60.64
the host asks the sensor to collect EMG signals, the analog circuit and ADC will startuA. Finally, when
the host asks
operation, the the sensor
current willtoincrease
collect EMG signals,
to about 1.269the
mAanalog
(Figurecircuit and ADC
16). Figure will start
17 depicts the
operation, the current will increase to about 1.269 mA (Figure 16). Figure
reduced time scale and clearly shows the current change. We also measured the average 17 depicts the
reduced
current oftime
thescale
surfaceandEMGclearly shows
sensor the current
without change.
the MCU, the We also measured
average current is the average
about
Biosensors 2021, 11, 411 1272.14
of 15
current of the surface EMG sensor without the MCU, the average current
uA. Compared to the literature on surface EMG sensors without wireless communication, is about 72.14
uA. Compared
we achieved to the
lower literature
power on surface
consumption inEMG sensorsEMG
our surface without wireless
sensors communication,
[52,53].
we achieved lower power consumption in our surface EMG sensors [52,53].

Figure 16.
Figure 16. Current
Current consumption
consumption of
of the
the surface
surface EMG
EMG system.
system. The
The blue
blue line is only
line is only ADC
ADC conversion
conversion and
and the
the purple
purple line
line is
is
Figure 16. Currentand
BLE transmission consumption of the surface EMG system. The blue line is only ADC conversion and the purple line is
ADC conversion.
BLE transmission and ADC conversion.
BLE transmission and ADC conversion.

Figure 17. Current consumption of the surface EMG system for small time scales. The blue line is only ADC conversion
Figure 17.
and the17.
Figure Current
purple consumption
lineconsumption
Current ofthe
thesurface
is BLE transmission
of surface
and ADC EMG
EMG system
forfor
conversion.
system small
small time
time scales.
scales. TheThe blue
blue lineline is only
is only ADC ADC conversion
conversion and
and the purple
the purple line
line is is BLE
BLE transmission
transmission and ADC
and ADC conversion.
conversion.
4.4. Advantages
4.4.
4.4. Advantages
Advantages
Table 2 lists the key techniques used in the proposed design compared toother
Table
Table 22 lists
architectures. lists thekey
the
Compared key techniques
techniques
with used
used
the previous in proposed
in the the [54],
works proposeddesign
the design
power compared
compared toother
consumption toother
ofarchi-
our
architectures.
tectures. Compared
Compared with with
the the
previousprevious
works works
[54], the [54],
power the power
consumption consumption
proposed architecture can be reduced by 92.46%. In addition, the battery life is 5.66 times of our of our
proposed
proposed
architecture
longer thanarchitecture
canprevious
the canwork
be reduced be by
reduced
92.46%.
with by In
the 92.46%. InmAh
addition,
same 300 addition,
the lithiumthe battery
battery battery life
life is 5.66 iscontinuous
5.66longer
in atimes times
longer than the
than the previous
wireless previous
connection. work work with
with the same
Compared the
with 300same 300
mAh lithium
commercial mAh lithium
battery
devices battery
[7],inour in a
a continuous continuous
proposedwireless
system
wireless
connection.
reduces connection.
the Compared
power Compared
consumption with
with commercial
by commercial
92.72% devices
and the[7], devices life[7],
our proposed
battery oursystem
is 9.057 proposed
times reducessystem
longer the
than
reduces
power the power
consumption consumption
by
the commercial devices [7]. 92.72% andby 92.72%
the and
battery the
life is battery
9.057 life
times is 9.057
longer times
than the longer than
commercial
the commercial
devices [7]. devices [7].

Table 2. Comparison of various surface EMG hardware architectures.

Architecture This Work [54] [7]

Wireless Technology BLE BLE BLE
PCB Size (cm2 ) 1.4 × 3.1 3.8 × 4.45 2.7 × 3.7
Battery (mAH) 300 300 N/A
Total Power Consumption 4.735 mW 62.7 mW 65 mW
Sampling Rate (Hz) 2k 1k 2.148k
SNR (dB) 23.1 N/A 65
Battery Life (hours) 63.4 11.2 7
Biosensors 2021, 11, 411 13 of 15

5. Conclusions
In this paper, an ultra-low power surface EMG signal acquisition system was de-
veloped and we focused on optimizing each part of the surface EMG signal acquisition
system. We confirmed that the power consumption and stability of the MFB filter was
much better than the Sallen-Key filter suggested in the literature. Regarding the MCU
power saving method, the system uses two different frequency MCU clock sources to
satisfy these system requirements. In addition, we proposed a minimization method based
on using a ping-pong buffer as the memory architecture to achieve the best power saving
effect. Finally, the EMG signal was preserved by more effective noise removal through IIR.
The experimental results showed that the proposed surface EMG sensor can significantly
reduce power consumption. Compared with commercial devices, our proposed system
reduced the power consumption by 92.72%. In addition, the collected EMG signals had a
99.5% high correlation with the commercial systems. Our proposed system was proven to
be easily available and effective; it could be used for further research on other surface EMG
applications, such as EMG posture recognition and control of a robotic arm.

Author Contributions: Conceptualization, Y.-D.W.; methodology, Y.-D.W.; software, Y.-D.W.; vali-

dation, Y.-D.W.; formal analysis, Y.-D.W.; investigation, Y.-D.W.; resources, Y.-H.L.; data curation,
Y.-D.W.; writing—original draft preparation, Y.-D.W.; writing—review and editing, S.-J.R.; visualiza-
tion, Y.-D.W.; supervision, S.-J.R. and Y.-H.L.; project administration, Y.-D.W.; funding acquisition,
Y.-H.L. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the National Taiwan University of Science and Technology—
Taipei Medical University Joint Research Program under the project “The design of musical canter
algorithm based on deep learning” (Grant No. NTUST-TMU-110-02).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Liu, Y.; Huang, H. Towards a high-stability EMG recognition system for prosthesis control: A one-class classification based
non-target EMG pattern filtering scheme. In Proceedings of the 2009 IEEE International Conference on Systems, Man and
Cybernetics, San Antonio, TX, USA, 11–14 October 2009; pp. 4752–4757.
2. Huang, P.; Wang, H.; Wang, Y.; Liu, Z.; Samuel, O.W.; Yu, M.; Li, X.; Chen, S.; Li, G. Identification of Upper-Limb Movements
Based on Muscle Shape Change Signals for Human-Robot Interaction. Comput. Math. Methods Med. 2020, 2020, 1–14.
3. Su, Y.; Fisher, M.H.; Wolczowski, A.; Bell, G.D.; Burn, D.; Gao, R. Towards an EMG Controlled Prosthetic Hand Using a 3D
Electromagnetic Positioning System. In Proceedings of the 2005 IEEE Instrumentationand Measurement Technology Conference
Proceedings, Ottawa, ON, Canada, 16–19 May 2005; pp. 261–266.
4. Brunelli, D.; Tadesse, A.M.; Vodermayer, B.; Nowak, M.; Castellini, C. Low-cost wearable multichannel surface EMG acquisition
for prosthetic hand control. In Proceedings of the 2015 6th International Workshop on Advances in Sensors and Interfaces
(IWASI), Gallipoli, Italy, 18–19 June 2015; pp. 94–99.
5. Gilmore, K.; Meyers, J. Using surface electromyography in physiotherapy research. Aust. J. Physiother. 1983, 29, 3–9. [CrossRef]
6. Al-Mulla, M.R.; Sepulveda, F.; Colley, M. A review of non-invasive techniques to detect and predict localised muscle fatigue.
Sensors 2011, 11, 3545–3594. [CrossRef] [PubMed]
7. Trignotm Wireless Biofeedback System User’s Guide Delsys Incorporated. 2021. Available online: https://www.delsys.com/
downloads/USERSGUIDE/trigno/wireless-biofeedback-system.pdf (accessed on 30 March 2021).
8. Mini Wave Infinity. 2004. Available online: http://www.h-elmar-ms.pl/helmar-ms/plik/cometa-systems_wavetrack-inertial-
system_nn4776.pdf (accessed on 30 March 2021).
9. Ultium EMG Brochure. 2019. Available online: https://www.noraxo-n.com/noraxon-download/ultium-emg-datasheet/
(accessed on 30 March 2021).
10. Shimmer User Manual Revision 3p. 2017. Available online: http://www.shimmersensing.com/images/uploads/docs/Shimmer_
User_Manual_rev3p.pdf (accessed on 30 March 2021).
11. Biagetti, G.; Crippa, P.; Falaschetti, L.; Orcioni, S.; Turchetti, C. Human activity monitoring system based on wearable sEMG and
accelerometer wireless sensor nodes. Biomed. Eng. Online 2018, 17, 132. [CrossRef]
12. Bhandarkar, A. Design Considerations for a Robust EMG Amplifier. Int. J. Sci. Res. (IJSR) 2016, 5, 1933–1937.
Biosensors 2021, 11, 411 14 of 15

13. Ho, T.-Y.; Chen, Y.-J.; Hung, W.-C.; Ho, K.-W.; Chen, M.-S. The Design of EMG Measurement System for Arm Strength Training
Machine. Math. Probl. Eng. 2015, 2015, 1–10. [CrossRef]
14. Gargiulo, G.; Calvo, R.; Jin, C. Giga-ohm high-impedance FET input amplifiers for dry electrode biosensor circuits and systems.
In Integrated Microsystems: Electronics, Photonics, and Biotechnology; CRC Press: Boca Raton, FL, USA, 2017; pp. 165–194.
15. Laghari, W.M.; Baloch, M.U.; Mengal, M.A.; Shah, S.J. Performance analysis of analog butterworth low pass filter as compared to
Chebyshev type-I filter, Chebyshev type-II filter and elliptical filter. Circuits Syst. 2014, 5, 209–216. [CrossRef]
16. Kabanov, A.A.; Nikonova, G.V. Development of Analog Filtering Circuit for Electromyography Signals. In Proceedings of the
2019 Ural Symposium on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT), Yekaterinburg,
Russia, 25–26 April 2019; pp. 55–58.
17. Elamvazuthi, I.; Zulkifli, Z.; Ali, Z.; Khan, M.K.A.A.; Parasuraman, S.; Balaji, M.; Chandrasekaran, M. Development of
Electromyography Signal Signature for Forearm Muscle. Procedia Comput. Sci. 2015, 76, 229–234. [CrossRef]
18. Wang, J.; Tang, L.; Bronlund, J.E. Surface EMG signal amplification and filtering. Int. J. Comput. Appl. (IJCA) 2013, 82, 15–22.
[CrossRef]
19. Rozaqi, L.; Nugroho, A.; Sanjaya, K.H.; Simbolon, A.I. Design of Analog and Digital Filter of Electromyography. In Proceedings
of the 2019 International Conference on Sustainable Energy Engineering and Application (ICSEEA), Tangerang, Indonesia,
23–24 October 2019; pp. 186–192.
20. Çakar, H.I.; Kara, S.; Toker, O. Design of a Portable Electromyography Device for Back Herniated Patients. In Proceedings of the
3rd International Conference on Biomedical Electronics and Devices; SciTePress: Setúbal, Portugal, 2010; pp. 20–23.
21. Go, S.A.; Coleman-Wood, K.; Kaufman, K.R. Frequency analysis of lower extremity electromyography signals for the quantitative
diagnosis of dystonia. J. Electromyogr. Kinesiol. 2014, 24, 31–36. [CrossRef]
22. Van Boxtel, A. Optimal signal bandwidth for the recording of surface EMG activity of facial, jaw, oral, and neck muscles.
Psychophysiology 2001, 38, 22–34. [CrossRef]
23. Reaz, M.B.I.; Hussain, M.S.; Mohd-Yasin, F. Techniques of EMG signal analysis: Detection, processing, classification and
applications. Biol. Proced. Online 2006, 8, 11–35. [CrossRef]
24. Li, X.; Rymer, W.Z.; Li, G.; Zhou, P. The effects of notch filtering on electrically evoked myoelectric signals and associated motor
unit index estimates. J. NeuroEng. Rehabil. 2011, 8, 64. [CrossRef]
25. Ladrova, M.; Martinek, R.; Nedoma, J.; Fajkus, M. Methods of Power Line Interference Elimination in EMG signal. J. Biomim.
Biomater. Biomed. Eng. 2019, 40, 64–70. [CrossRef]
26. Kamata, K.; Aho, A.J.; Hagihira, S.; Yli-Hankala, A.; Jäntti, V. Frequency band of EMG in anaesthesia monitoring. Br. J. Anaesth.
2011, 107, 822–823. [CrossRef]
27. Tankisi, H.; Burke, D.; Cui, L.; de Carvalho, M.; Kuwabara, S.; Nandedkar, S.D.; Rutkove, S.; Stålberg, E.; van Putten, M.J.A.M.;
Fuglsang-Frederiksen, A. Standards of instrumentation of EMG. Clin. Neurophysiol. 2020, 131, 243–258. [CrossRef]
28. Soedirdjo, S.D.H.; Ullah, K.; Merletti, R. Power line interference attenuation in multi-channel sEMG signals: Algorithms and
analysis. In Proceedings of the 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology
Society (EMBC), Milan, Italy, 25–29 August 2015; pp. 3823–3826.
29. Esp8266ex Datasheet. 2020. Available online: https://www.espressif.com/sites/default/files/documentation/0a-esp8266ex_
datasheet_en.pdf (accessed on 15 April 2021).
30. nRF52832 Datasheet. 2017. Available online: https://infocenter.nordicsemi.com/pdf/nRF52832_PS_v1.4.pdf (accessed on
15 April 2021).
31. Dieffenderfer, J.N.; Kalla, R.N. Ping-Pong Data Buffer for Transferring Data from One Data Bus to Another Data Bus. U.S. Patent
US5224213A, 29 June 1993.
32. Joo, Y.-M.; McKeown, N. Doubling memory bandwidth for network buffers. In Proceedings of the IEEE INFOCOM ‘98, the
Conference on Computer Communications. Seventeenth Annual Joint Conference of the IEEE Computer and Communications
Societies. Gateway to the 21st Century (INFOCOM), San Francisco, CA, USA, 29 March–2 April 1998; Volume 2, pp. 808–815.
33. Zhang, C.; Li, P.; Sun, G.; Guan, Y.; Xiao, B.; Cong, J. Optimizing FPGA-Based Accelerator Design for Deep Convolutional Neural
Networks. In Proceedings of the 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, Monterey,
CA, USA, 22–14 February 2015; pp. 161–170.
34. Mustafin, M.; Turar, O.; Akhmed-Zaki, D. Testing of Vulkan Visualization for Geo-Models on Mobile Devices and Desktop
Systems with Ray Tracing GPUs. In ECMOR XVII; European Association of Geoscientists & Engineers: Houten, The Netherlands,
2020; Volume 2020, pp. 1–11.
35. Joshi, S.; Wakankar, A.; Khambete, N. Design and implementation of low power compact amplifier circuitry for wearable
biosignal device. In Proceedings of the 2016 International Conference on Computing Communication Control and Automation
(ICCUBEA), Pune, India, 12–13 August 2016; pp. 1–6.
36. INA333 Datasheet. 2015. Available online: https://www.ti.com/lit/gpn/INA333 (accessed on 16 April 2021).
37. AD620 Datasheet. 2011. Available online: https://datasheet.octopart.com/AD620ANZ-Analog-Devices-datasheet-9660204.pdf
(accessed on 16 April 2021).
38. Salman, A.; Iqbal, J.; Izhar, U.; Khan, U.S.; Rashid, N. Optimized circuit for EMG signal processing. In Proceedings of the 2012
International Conference of Robotics and Artificial Intelligence, Rawalpindi, Pakistan, 22–23 October 2012; pp. 208–213.
Biosensors 2021, 11, 411 15 of 15

39. Yang, G.; Sun, Z.; Wang, Z.; Sun, G. Detection and Identification the Human Surface Electromyogra Signal. IOP Conf. Ser. Earth
Environ. Sci. 2020, 440, 022016. [CrossRef]
40. INA128 Datasheet. 2019. Available online: https://www.ti.com/lit/gpn/ina128 (accessed on 16 April 2021).
41. AD8236 Datasheet. 2009. Available online: https://www.analog.com/media/en/technical-documentation/data-sheets/AD8236.
pdf (accessed on 16 April 2021).
42. 3-lead Muscle / Electromyography Sensor for Microcontroller Applications Datasheet. 2015. Available online: https://cdn.
sparkfun.com/datasheets/Sensors/Biometric/MyowareUserManualAT-04-001.pdf (accessed on 16 April 2021).
43. Roland, T.; Wimberger, K.; Amsuess, S.; Russold, M.F.; Baumgartner, W. An Insulated Flexible Sensor for Stable Electromyography
Detection: Applicationto Prosthesis Control. Sensors 2019, 19, 961. [CrossRef] [PubMed]
44. Importance of Input Bias Current Return Paths in Instrumentation Amplifier Applications. 2021. Available online: https:
//www.ti.com/lit/pdf/sboa503 (accessed on 1 May 2021).
45. Casas, O.; Pallas-Areny, R. Basics of analog differential filters. IEEE Trans. Instrum. Meas. 1996, 45, 275–279. [CrossRef]
46. Designing Amplifier Circuits: How to Avoid Common Problems. 2007. Available online: https://www.analog.com/media/en/
technical-documentation/application-notes/AN-937.pdf (accessed on 2 May 2021).
47. Tabassum, F.; Islam, M.I.; Amin, M.R. Comparison of FIR and IIR Filter Bank in Reconstruction of Speech Signal. Int. J. Comput.
Sci. Inf. Secur. 2016, 14, 864.
48. Introduction to Filters: FIR Versus IIR. 2020. Available online: https://community.sw.siemens.com/s/artic-le/introduction-to-
filters-fir-versus-iir (accessed on 1 June 2021).
49. Chang, J.; Phinyomark, A.; Scheme, E. Assessment of EMG Benchmark Data for Gesture Recognition Using the NinaPro Database.
In Proceedings of the 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC),
Montreal, QC, Canada, 20–24 July 2020; pp. 3339–3342.
50. Fraser, G.D.; Chan, A.D.C.; Green, J.R.; MacIsaac, D.T. Automated Biosignal Quality Analysis for Electromyography Using a
One-Class Support Vector Machine. IEEE Trans. Instrum. Meas. 2014, 63, 2919–2930. [CrossRef]
51. Heywood, S.; Pua, Y.; McClelland, J.; Geigle, P.; Rahmann, A.; Bower, K.J.; Clark, R. Low-cost electromyography—Validation
against a commercial system using both manual and automated activation timing thresholds. Electromyogr. Kinesiol. 2018, 42,
74–80. [CrossRef] [PubMed]
52. Jani, A.B.; Bagree, R.; Roy, A.K. Design of a low-power, low-cost ECG & EMG sensor for wearable biometric and medical
application. IEEE Sens. 2017, 1–3. [CrossRef]
53. Seguna, C.; Buhagiar, S.; Scerri, J.; Scicluna, K. Development of a New EMG Wearable Sensor for Myoelectric Control.
In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies, Valletta, Malta,
24–26 February 2020; pp. 160–164.
54. Chen, P.-C.; Ruan, S.-J.; Tu, Y.-W. Power-Management Strategies in sEMG Wireless Body Sensor Networks Based on Computation
Allocations: A Case Study for Fatigue Assessments. IEEE Access 2020, 8, 181366–181374. [CrossRef]