SIGNaL PROCESSING

algorithms, architectures, arrangements, and applications

SPa 2024
THE INSTITUTE OF ELECTRICAL AND ELECTRONICS ENGINEERS INC. September 25th - 27th, 2024, Poznań, POLAND

EMG Signal Acquisition and Processing for Muscle

Contraction Classification

Bianca-Alexandra Zîrnă, Denis Mihailovschi Mădălin Corneliu Frunzete

Faculty of Medical Engineering Applied Electronics and Information
National University of Science and Technology National University of Science and Technology
POLITEHNICA Bucharest POLITEHNICA Bucharest
Bucharest, Romania Bucharest, Romania
[email protected], [email protected] [email protected]

Abstract—Nowadays, the focus is on portable, non-invasive The most common method of recording signals for
devices with a variety of functions, such as continuous monitoring muscle contractions is electromyography, a
monitoring through smartwatches or biologic signal-controlled diagnostic test that evaluates the condition and function of the
prosthetic limb control. The signal acquisition and processing motor and sensory neurons that control the muscles. A muscle
steps are the underlying principles behind all of these at rest typically has no electrical activity. Electrical activity is
applications. This study proposes a low-budget Arduino-based generated by just a small muscular contraction, and this activity
acquisition system for recording forearm surface increases with greater muscle contraction force. To generate
electromyography signals, which are then analyzed using Matlab
high-quality EMG signals, electrode selection is also essential.
software. The signals are recorded from four right-handed
Electrodes are available in two different forms: surface
healthy individuals, following an established protocol consisting
of ten 30-second recordings per hand, with movements that
electrodes that record surface EMG signals and intramuscular
alternate between five seconds of contraction and five seconds of electrodes like needles [3],[4]. The surface electrodes will be
relaxation. The classification results are then achieved by utilized to record the surface electromyography (sEMG) since
processing the data using the Support Vector Machine algorithm the intramuscular electrodes are invasive.
and GridSearch. There are four experiments conducted. The first Numerous disorders can affect the muscular system,
one distinguishes between contraction and relaxation, then three starting with the mildest (muscular fatigue) to the most severe
additional experiments are performed following the same
(several myopathies). Muscle cramps, or painful contractions,
procedure (discriminations between gender, hand, gender and
may occur in exhausted muscles frequently. After a time of
hand). Subsequently, a graphical user interface is designed to
display the uploaded signal together with some details about the inactivity, practicing an intense physical effort leads to
patients, the experiment, and the classification outcomes. The delayed-onset muscle soreness, which means strong, persistent
results achieved with the Arduino microcontroller are confirmed pains at the level of the muscle groups. Resuming the same
by performing the same steps using the Biopac system and type of effort leads to its reduction. By increasing physical
comparing the outcomes in terms of metrics. activity, it might be possible to prevent the earliest signs of
fatigue and the loss of muscular strength. Therefore, to prevent
Keywords-EMG signals; Muscular contraction; Arduino signal muscular strains or ruptures, periods of intense workouts must
recording, Matlab signal processing. be alternated with rest periods that restore the capacity for
effort. A muscle may expand or tear because of excessive
I. INTRODUCTION contraction, as well as the adjacent connective tissue [1],[5].
The active organs of movement that generate force or Currently, the focus is on portable, non-invasive devices
movement in response to a physiological stimulus are the with various purposes: using various artisanal methods, such as
muscles that make up the muscular system. A muscle is smartwatches, Arduino, Biopac, etc., to control prosthetic
composed of interconnected muscular fibers that contract to limbs (as described in [6]), monitor effort through muscle
produce force. Skeletal, smooth, and cardiac muscle are the contraction and relaxation, and continuously monitor patients
three forms of muscle that are adapted to specific functions. with various pathologies. These techniques exclusively record
About 40% of the body's mass consists of skeletal muscles, sEMG signals. Articles [7], [8], and [9] propose different
which are responsible for voluntary bone movements. These approaches to processing EMG signals and classifying
motions provide the basis for both mechanical work and different hand postures using Arduino Uno or Arduino Mega
locomotion [1],[2]. They include the skeletal muscles of the and a MyoWare sensor. Papers [10], [11], and [12] aim to build
upper limb; electrodes will be placed on the anterior forearm to a prototype hand posture recognition system for bionic hand
record electromyographic (EMG) signals, where the forearm control by analyzing the recorded sEMG signals and also using
and hand flexors, pronators of the hand, and finger flexors are different Arduino boards and EMG sensors, such as the EMG
located, as will be described later. Gravity Sensor or Muscle Sensor Toolkit.

177

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on April 05,2025 at 12:41:53 UTC from IEEE Xplore. Restrictions apply.
 After a review of the most popular approaches, the decision
was made to split the work into three sections, as will be
further described.

II. MATERIALS AND METHODS

This work proposes an EMG signal acquisition and
preprocessing method. The block diagram with the three main
directions is shown in Fig. 1. The initial stage involves
developing the signal acquisition device while taking the
quality-cost ratio into consideration. This is essential for the
device to be functional, reliable, and have results as close as
possible to those of a professional device, but it also must
remain within an affordable price range in order to serve as an
example of a low-budget system. Thus, an Arduino Uno Figure 1. The block diagram of the proposed method.
microcontroller and an EMG SEN0240 sensor compose the
EMG Device. The Arduino Uno microcontroller, DFRobot
chosen device. Next, using the Data Streamer toolbox in EMG SEN0240, Dupont wires, and a USB connector are
Excel, the recorded signals are saved as.csv files. The sensor needed for building the portable device. The connections are
with the three metal electrodes is placed on the subject's shown in Fig. 2. The EMG sensor consists of two boards: the
forearm muscle in order to record the signals. Ten sessions of signal transmission board includes two integrated circuits: an
thirty seconds are recorded for both hands, with each amplification circuit and a filter circuit, while the other board
recording alternating between two movements—a relaxation has three dry, metal electrodes integrated and a sleeve attached
and a contraction—every five seconds. All four subjects [13]. No additional processing is required for signals read from
perform this process, resulting in a total of eighty signals. an analog pin.
In the second phase, the signals are processed to achieve
multiple classifications, being divided into 5-second epochs in
order to extract the features using several techniques. After
labeling each epoch properly, the Support Vector Machine
(SVM) classification algorithm is fed the union of the prior
features, and GridSearch is used to achieve the best possible
outcomes. The first experiment distinguishes between
relaxation and contraction. Following the same steps for each
experiment, three experiments are performed: right hand vs.
left hand, female vs. male, and female right hand vs. female
left hand vs. male right hand vs. male left hand.
Figure 2. Connections and electrode placement.
The final step involves developing a graphical user
interface with two distinct objectives and potential directions: Subjects. Four patients, two male and two female, who were
hardware through data integration in a bionic prosthesis between the ages of 24 and 25 and did not have upper limb
muscle diseases, had their signals recorded. All subjects were
system or software through an e-health application. The two
right-handed dominant, and the signals were recorded from
main pages of the interface are one dedicated to multiclass
both hands. Before beginning the movement series, the
classification and the other to binary classification relaxation participants signed a written consent form and had their data
vs. contraction. The uploaded signal, some details about the anonymized after receiving a detailed description of the
patients, the experiment, and the classification results are experiment.
displayed on both pages, as is the classification result: the
signal's epoch along with its label. By performing the identical Experimental procedure. An initial experiment has been
signal acquisition and processing steps with the Biopac system proposed for this work in order to distinguish between the two
and comparing the results in terms of metrics, the results states: contraction and relaxation. Three more experiments are
obtained with the Arduino microcontroller are validated. suggested based on the results, with the same protocol being
followed for each trial. The proposed experiments are listed in
A. EMG Signal Acquisition Table I. After the initial classification, the data is compressed,
The acquisition of the EMG signals is the first step of this keeping only the contraction epochs. This process optimizes
work, which consists of five main sections: building and the algorithm and reduces compute time.
connecting the wearable device, creating an anonymized TABLE I. PROPOSED EXPERIMENTS
database with the consent of the subjects, establishing the No. Type Details
series of movements performed by the subjects, and finally, E0 Binary Relaxation vs. contraction
recording the signals and storing them in a format that is E1 Binary Hand: right vs. left
compatible with Matlab software. E2 Binary Gender: feminine vs. masculine
E3 Multiclass Gender and hand combinations (4 classes)

178

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on April 05,2025 at 12:41:53 UTC from IEEE Xplore. Restrictions apply.
 The three-electrode sensor is placed parallel to the muscle classification algorithm and GridSearch. The other three
fibers inside the forearm, along the muscle, to record the experiments are performed after the first discrimination
sEMG signals. Before replacing the sensor, the new area of between contraction and relaxation, following the same steps.
skin is cleaned with sanitary alcohol. As previously mentioned, the SEN0240 sensor includes two
circuits built into the board: filtering and amplification;
As illustrated in Fig. 3a, each subject performs thirty-
therefore, the pre-processing step is no longer necessary.
second sets of movements, alternating between five seconds of
contraction and five seconds of relaxation, resulting in three Epoch Segmentation. EMG signals are divided into epochs,
contraction and three relaxation sections. Both the relaxation or time frames of fixed duration, according to sets of
and the contraction should be recorded because a prolonged movements and to the number of samples, as they cannot be
contraction could lead to muscle pain. Furthermore, by segmented into short, known physiological units. The signal is
alternating the states, the contractions do not decrease in divided into epochs of five seconds duration (1215 samples),
amplitude. Ten recordings are performed for each hand, from alternating between contraction and relaxation based on the
both hands, for all the subjects, resulting in a total of eighty acquisition methodology. Therefore, each signal is divided
records. Both hands were recorded, not just the dominant into six epochs.
hand, since it would be necessary to employ a myoelectrically
Feature extraction. Then, fourteen features are extracted
controlled prosthesis in the event that one hand is amputated.
from every epoch, with each outcome stored in a separate
Signal processing and recording. The Arduino board is vector. Selecting the right features is essential since different
connected to the computer via the USB connector once the features perform well with different kinds of signals. The
sensor has been placed on the subject and the previously following fourteen features have been chosen based on the
specified connections have been made. The Arduino IDE literature:
program is used at this point. After the signal has been filtered
and amplified, it can be observed that the signal is almost  MAV (Mean Absolute Value): the average of the
continuous, with appropriate values of zero when the hand is amplitudes' absolute values
relaxed and increases in amplitude when the muscle contracts  WL (Waveform Length): the waveform's total length
(Fig. 3b). throughout time
 SSC (Slope Sign Change): the number of times the sign of
a. the signal slope changes
 VAR (Variance): measure of dispersion
 DAMV (Difference Absolute Mean Value): mean absolute
difference of two independent and identically distributed
b. values
 MAD (Mean Absolute Deviation): the average distance
between each data point and the mean
 MFL (Maximum Fractal Length): measure of contraction
strength
 WA (Willison Amplitude): the number of changes in
signal amplitude that exceeds a predefined threshold
 COV (Coefficient of Variation): the relative dispersion of
the data
 KURT (Kurtosis): the degree of flattening or sharpening
Figure 3. Acquisition protocol (a) and the corresponding signal (b). of a distribution
 SKEW (Skewness): the deviation of the empirical
Signal transmission. Once the code has executed, the Arduino
distribution in relation to a symmetrical distribution
IDE software is closed and the Data Streamer toolbox from the
around the mean
Excel program is opened in order to transfer the
microcontroller data to Excel. In order to give the signal  SD (Standard Deviation): the dispersion from the mean of
enough time to stabilize, the thirty-second recording begins a the values of a set
few seconds after the data loads. The data is automatically  RMS (Root Mean Square): the square root of the
saved to a new .csv file when the recording ends. To process arithmetic mean of the squares of the values
the signals further, they are later imported into Matlab.  ZC (Zero Crossing): the number of times the signal
changes its sign, passing through 0.
B. Signal processing and recording
Epoch Labeling. After the extraction of each of the
The signal processing phase consists of multiple steps: aforementioned features as independent vectors, a matrix is
transforming the files from .csv to vectors, dividing the signals generated by concatenating all of these vectors, where each
into five-second epochs, and extracting features. The best line represents a feature and each column an epoch. A starting
results are obtained by labeling each epoch properly and line is added to this matrix, denoting the epochs: 0 represents
feeding the union of the prior features into the SVM

179

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on April 05,2025 at 12:41:53 UTC from IEEE Xplore. Restrictions apply.
 relaxation and 1 represents contraction. Lastly, a 15 × 480 III. RESULTS
matrix (one line of labels and fourteen features and four A. Classification Results
patients, two hands, ten records and six epochs) will serve as
the classifier's input. The contraction and relaxation epochs are After the signals have been acquired using the previously
displayed in Fig. 4a. For the rest of the experiments, the mentioned procedure, fourteen specific features are extracted
contraction epochs will be labeled again, depending on the from each epoch. The MAV and ZC values for a few random
type of classification being performed. epochs of contraction versus relaxation are displayed in Fig. 5.
As a measure that computes the mean of the amplitudes'
Classification. After the extraction of all fourteen features, the absolute values (MAV) and the number of times the signal
values are normalized within the interval [0, 1]. Because each changes its sign (ZC) [10], contraction epochs have higher
individual and hand has a unique kind of contraction and values than those with relaxation.
relaxation, the values for the classification of relaxation vs.
contraction are normalized individually for each signal, and
for the rest of the classifications, the values are all normalized
at the end (because all contractions are normalized in relation
to others, according to gender and hand).
The epochs are randomly split into 80% training and 20%
testing data for classification, with the same number of epochs
in each set. As will be described later, 16% of the 80% training
set will be used for validation in turn (Fig. 4b).
a. b. Figure 5. MAV and ZC values for contraction and relaxation epochs.

Every experiment analyzes all fourteen features, with the

ones generating unsatisfactory results being removed (Table
II). Parallel coordinates plot was used to provide a graphical
Figure 4. Epoch labeling for E0 (a) and split for classification (b). illustration of each feature's individual results. After that, the
overall accuracy improved for all the experiments. The
The SVM algorithm, which is a useful technique for accuracy for each class in every experiment, both before and
medical signal classification, is used for classification, as it after feature selection, is presented in Fig. 6. It's also
was earlier used for EEG epileptic seizure classification in noticeable that all outcomes have improved.
[14]. The selection of hyperparameters, such as the kernel
function, is an essential step. For this reason, a cross- TABLE II. BEST FEATURES FOR EACH EXPERIMENT
validation model named GridSearch—which uses five values Experiment Features
per division—is applied. Its purpose is to automatically E0 MAV, WL, SSC, DAMV, MAD, ZC
identify the optimal pair of hyperparameters. E1 MAV, WL, DAMV, MAD, MFL, RMS
E2 WL, SSC, DAMV, MFL, ZC
The Matlab Classification Learner toolbox is used for E3 MAV, WL, SSC, ZC
classification. To evaluate the models' predicted accuracy, a
validation method is chosen. Validation helps select the
optimal model while avoiding overfitting by assessing the
model's performance on new data in comparison to the
training set. The data set is split into five subsets as cross-
validation with k = 5 is selected; each subset is used in turn as
validation data, while the other four are used as training data.
The average training accuracy and average validation error are
computed at the end of five iterations of this process. After
that, the test set is used, the same classification model is
applied to it, and the outcomes are evaluated. Figure 6. Results before and after feature selection.

C. Graphical Interface E0: Relaxation epochs were fully classified correctly, and
A Matlab Graphical User Interface (GUI) is designed and after feature selection, contraction epochs were also classified
developed in order to show the results in a user-friendly at 100%.
manner. The interface's purpose is to display a randomly E1: It results in the right-hand class having weaker detections
uploaded signal along with information about the subjects, the than the left-hand class, taking into account that the right hand
experiment, and the classification result on two main pages, is the dominant hand in all subjects. Female participant left-
one for binary relaxation vs. contraction classification and the hand signals have the weakest contractions, which makes them
other for multiclass classification. These panels appear when easier to differentiate. However, since the strength of a male
certain buttons are pushed. subject should be higher than that of a female subject, the

180

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on April 05,2025 at 12:41:53 UTC from IEEE Xplore. Restrictions apply.
 discrimination between a female right-hand and a male left- drawback of using this function, it would be effective to
hand contraction may be slightly inaccurate, leading to generalize an ideal approach.
misclassifications. Muscle fatigue also occurs after a specific
A GUI called App_EMG is developed and implemented in
amount of repetition of the contraction-relaxation series, and it
Matlab that displays the outcomes. Two primary tabs are
is characterized by a decrease in contraction strength. In
created, one for E1 and the other for E4 (Fig. 7), since E4
subjects who do not regularly participate in sports or in non-
combines E2 and E3 into a more complex multiclass
dominant hands, this occurs quicker.
classification.
E2: Compared to the female gender class, the male gender
class had poorer detections. This might be due to the fact that,
typically, a male subject's force is greater than a female
subject's, causing discrimination between a female right-
handed and a male left-handed contraction to be slightly
inaccurate. However, female subjects' left-hand signals show
the weakest contractions, which makes them easier to
separate, while some of their right-hand signals might show
contractions that are nearly as powerful as those of male
participants. In addition to the differences in the subjects,
particularly the variation in strength, there could also be
another contributing factor: muscle fatigue, which would
lessen the force of the contractions.
E3: As it can be seen, the male gender, right hand class
achieves the best results since male participant right-handed
signals show the highest strength of contraction. Due to the
fact that these epochs can be easily mistaken for another class,
such as female gender, right hand, the male gender, left hand
class achieves the lowest results.
In addition to the outcomes obtained from the
classification, several additional factors additionally influence
these results. Normalizing the data is an essential step,
whereby the data is normalized within the interval [0, 1]
before the classification process. The proper timing of data
normalization is critical for achieving optimal results. For E0,
the values are normalized individually for each signal after Figure 7. App_EMG tabs.
features are extracted from it because each subject and each
hand have a typical kind of contraction and relaxation. For all B. Comparison
other experiments, the values are normalized at the end after The same acquisition procedure is performed using the
all features from all signals are extracted, since all Biopac system in order to validate the results, and the
contractions need to be normalized relative to each other by outcomes for each experiment are compared in terms of
gender and hand. accuracy. The signals are recorded from the same subjects,
The number of balanced epochs in each class is another and the same features are kept. However, the Biopac system
significant factor. An asymmetric distribution increases the uses three disposable electrodes, which are also placed on the
likelihood that the majority class will obtain higher results. forearm but in a different configuration. Other differences
The splitting into training and testing sets is directly tied to between these two systems are detailed in Table III.
this aspect of balanced sets. In this work, an 80%-20% split
TABLE III. COMPARISON BETWEEN ARDUINO AND BIOPAC RESULTS
ratio was selected, choosing more training data since the
database's overall size is not particularly large (it has been Arduino Biopac
reduced by half in the experiments E1, E2, and E3). Reusable electrodes Disposable electrodes
Default 150 Hz sampling frequency Default 2000 Hz sampling frequency
SVM is a common classification algorithm in medical Low-cost system High-cost system
signal processing, but its main benefit comes from the Portable device Depends on the biopac module
GridSearch function, a cross-validation model that helps select Possibility of programming Standard functions
Online processing Offline processing
the best hyperparameters by automatically determining the Compatible with different softwares Only works with Acqknowledge
pair of parameters that generates the best results.
Consequently, the optimal hyperparameter pairings differed The overall accuracy of each experiment is computed to
for every experiment. Since the runtime increase is the main determine whether the results are comparable to those from a
professional device. Similar results can be observed in Fig. 8.

181

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on April 05,2025 at 12:41:53 UTC from IEEE Xplore. Restrictions apply.
 Both systems achieved 100% accuracy in discriminating software resources, all of the proposed experiments could be
between the two classes for E0 and the same accuracy for E1. performed efficiently. Finally, the results show good
Biopac's outcomes are better for E2, but poorer for E3. discrimination between the classes. The graphical interface's
basic display makes the approach easy to understand.
However, the small size of the database and the inability of
the proposed approach to classify in real-time due to the data
transfer between hardware and software are some of its
limitations. The proposed work could proceed in two
alternative long-term directions: either hardware, through data
integration into a bionic prosthesis system, or software,
through an e-health application.

Figure 8. Accuracy obtained with both systems.

ACKNOWLEDGMENT
This work was supported by a grant from the National
C. Resources Program for Research of the National Association of
Hardware. After researching a variety of articles and Technical Universities - GNAC ARUT 2023.
components available on the market, the most efficient version
REFERENCES
is selected in order to balance quality and cost, as the device
needs to be functional, reliable, and deliver results comparable [1] W.F. Boron, W. F., E.L. Boulpaep, “Medical physiology E-book”,
Elsevier Health Sciences, 2016.
to those of professional devices, all while fitting into a low-
[2] J.M. Vojtech, C.E. Stepp, “Electromyography”, Manual of Clinical
price range. Apart from being affordable, the device also has Phonetics”, 2021, pp. 248-263.
the advantage of recording high-quality, preprocessed EMG [3] W. Sae-lim, P. Phukpattaranont, K. Thongpull, “Effect of electrode skin
signals that are comparable to those obtained using specialized impedance on electromyography signal quality”, IEEE 15th
equipment. This can be observed in both the outcomes and the International Conference on Electrical Engineering/Electronics,
Computer, Telecommunications and Information Technology (ECTI-
comparisons with the Biopac system's results. Furthermore, CON), 2018, pp. 748-751.
connections are made quite easily, and only a few components [4] V. Gohel, N. Mehendale, “Review on electromyography signal
are used. Consequently, the device is extremely easy acquisition and processing”, Biophysical reviews, Vol. 12, 2020, pp.
to assemble and use. 1361-1367.
[5] E. Farago, D. MacIsaac, M. Suk, A.D. Chan, “A review of techniques
Software. Due to the equipment used, the running times are for surface electromyography signal quality analysis”, Biomedical
extremely effective for implementing the proposed algorithms Engineering, Vol. 16, 2022, pp. 472-486.
and performing all the experiments in a relatively short [6] Ș. Dumitru, A. Craifaleanu, “Kinematic and dynamic analysis of an
amount of time. Removing some features also shortens the anthropomorphic hand-arm system”, U.P.B. Sci. Bull., Series D, Vol.
85, Iss. 4, 2023, pp. 27-42.
runtimes. As previously mentioned, the algorithm shows
[7] M.A. Abu, S. Rosleesham, M.Z. Suboh, M.S.M. Yid, Z. Kornain, et al.,
extremely short running times, being efficient in both time and “Classification of EMG signal for multiple hand gestures based on
outcomes. Therefore, it is highly suitable for real-time online neural network”, Indonesian Journal of Electrical Engineering and
classification in the future. Computer Science, Vol. 17, 2020, pp. 256-263.
[8] Y. Omama, C. Haddad, M. Machaalany, A. Hamoudi, M. Hajj-Hassan,
IV. CONCLUSIONS et al., “Surface EMG classification of basic hand movement”, IEEE
Fifth International Conference on Advances in Biomedical Engineering
This study followed three main directions: acquiring the (ICABME), 2019, pp. 1-4.
signals, processing them, and displaying the outcomes in a [9] O. Kainz, A. Mak, D. Cymbalák, S. Kardoš, P. Fecil'ak, et al., “Low-
graphical user interface. The results are then validated by cost assistive device for hand gesture recognition using sEMG
comparing them with those obtained with the Biopac system. (Erratum)”, SPIE First International Workshop on Pattern Recognition,
Vol. 10011, 2019, pp. 265-271.
An advantage of this work is the database, since it [10] W.T. Shi, Z.J. Lyu, S.T. Tang, T.L. Chia, C.Y. Yang, “A bionic hand
facilitates an individual acquisition protocol. Based on the controlled by hand gesture recognition based on surface EMG signals: A
preliminary study”, Biocybernetics and Biomedical Engineering, Vol.
signals that were recorded, other studies were then proposed. 38, 2018, pp. 126-135.
As mentioned earlier, an important consideration is when data [11] H. Wu, M. Dyson, K. Nazarpour, “Real-time myoelectric control with
normalization is required. Better results and shorter runtimes an Arduino”, 27th IEEE International Conference on Electronics,
are also achieved by removing less effective features. Using Circuits and Systems (ICECS), 2020, pp. 1-2.
the GridSearch function has the advantage of assisting in [12] M. Salama, A. Bakr, “Six prosthetic arm movements using
determining the most suitable set of hyperparameters for every electromyogram signals: A prototype”, IEEE UKSim-AMSS 18th
International Conference on Computer Modelling and Simulation
classification, but it also has drawbacks, such as increasing (UKSim), 2016, pp. 37-42.
running times and failing to produce a generalized model. As a [13] Analog EMG Sensor by OYMotion SKU:SEN0240 Datasheet, 2017.
consequence of using relatively few hardware resources and [14] B.A. Zîrnă, D. Mihailovschi, M.C. Frunzete, “EEG signal processing for
having very basic connections, the device is easy to use and epileptic seizure detection”, E-Health and Bioengineering Conference,
affordable. Because the running times are minimal in terms of 2023, unpublished.

182

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on April 05,2025 at 12:41:53 UTC from IEEE Xplore. Restrictions apply.