Technical Paper Session - tronshow.org

P R O C E E D I N G S 2021 TRON Symposium −TRONSHOW−

December 8-10, 2021

Tokyo, Japan

Technical Paper Session

Organizer: TRON Forum

Co-sponsor: collaboration Hub for University and Business (cHUB),

Faculty of Information Networking for Innovation And

Design (INIAD), Toyo University / Institute of

Infrastructure Application of Ubiquitous Computing

(IAUC), Interfaculty Initiative in Information Studies, The

University of Tokyo

Technical Co-Sponsor: IEEE Consumer Technology Society

Proceedings of 2021 TRON Symposium -TRONSHOW-

Tokyo Midtown, Tokyo, Japan

December 8-10, 2021

Published by TRON Forum

ISBN 978-4-89362-375-1

Part Number: CFP2164X-USB

Copyright 2021 @TRON Forum

It is prohibited by law to copy, duplicate, reproduce or translate all or any part of this document by any means, including electronic, mechanical,

optical or chemical, without written permission.

The names of hardware and software products mentioned in this book are trademarks or registered trademarks of the respective companies.

TRON, ITRON, μITRON, BTRON, CTRON, MTRON, and eTRON are not the names of particular commercial products or product line. T-Kernel, T-

Kernel 2.0, T-Kernel 3.0, T-Kernel 2.0 Extension, μT-Kernel, μT-Kernel 2.0, μT-Kernel 3.0, MP T-Kernel, T-Kernel Standard Extension, TRON Safe

Kernel, IoT-Engine and IoT Aggregator are the names of open real-time OS specifications and/or computer system architectures that TRON Forum

has been promoting.

Proceedings of 2O21 TRON Symposium -TRONSHOW-

Table of Contents

TRON Symposium Outline

Greeting from the Chair of Technical Program Committee

Papers

TRON 2021: 2021 TRON Symposium (TRONSHOW)

S1: Paper session

Chaired by Fahim Khan (Toyo University, Japan)

01:30 pm: Arrhythmia Classification Using EFFICIENTNET-V2 with 2-D Scalogram

Image Representation

Reza Fuad Rachmadi (Institut Teknologi Sepuluh Nopember, Indonesia); Supeno Mardi Susiki

Nugroho (Sepuluh Nopember Institute Of Technology, Indonesia); Muhammad Furqon (Sepuluh

Nopember Institute of Technology Surabaya, Indonesia); Arief Kurniawan (Institut Teknologi

Sepuluh Nopember, Indonesia); I Ketut Eddy Purnama (Institut Teknologi Sepuluh Nopember,

Indonesia); Mpu Aji (Sepuluh Nopember Institute of Technology Surabaya, Indonesia)

01:45 pm: Harnessing IoT Technology for the Development of Wearable Contact

Tracing Solutions

Rex Acharya (Sam Houston State University, USA); Amar A Rasheed (Sam Houston State

University, USA); Hacer Varol (Stephen F. Austin State University, USA); Mohamed Baza (College

of Charleston, USA); Louanne Mozer Sallo (Sam Houston State University, USA); Rabi Mahapatr

(Texas A&M University, USA)

Appendix

Technical Program Committee, and other staff members

The total program of 2021 TRON Symposium -TRONSHOW-

ISBN978-4-89362-375-1 (C) 2021 TRON Forum

Arrhythmia Classification Using EFFICIENTNET-V2 with 2-D Scalogram Image Representation

1st Muhammad Furqon Dept. of Electrical Engineering Faculty of Intelligent Electrical

and Informatics Technology Institut Teknologi Sepuluh Nopember

Surabaya, Indonesia 60111 [email protected]

4th Arief Kurniawan Department of Computer Engineering Faculty of Intelligent Electrical and

Informatics Technology Institut Teknologi Sepuluh Nopember Surabaya, Indonesia 60111

[email protected]

2nd Supeno Mardi Susiki Nugroho Department of Computer Engineering Faculty of Intelligent Electrical and


[email protected]

5th I Ketut Eddy Purnama Department of Computer Engineering Faculty of Intelligent Electrical and


[email protected]

3rd Reza Fuad Rachmadi Department of Computer Engineering Faculty of Intelligent Electrical and


[email protected]

6th Mpu Hambyah Syah Bagaskara Aji Department of Computer Engineering Faculty of Intelligent Electrical and

Informatics Technology Institut Teknologi Sepuluh Nopember Surabaya, Indonesia 60111 [email protected]

Abstract—Cardiovascular disease is part of global death's main cause. It is the term for all types of diseases that affect the heart or blood vessels. Heart disease is a type of cardiovascular disease. It can be detected early by examining the arrhythmia presence. Arrhythmia is an abnormal heart rhythm that is commonly diagnosed and evaluated by analyzing electrocardiogram (ECG) signals. In classical techniques, a cardiologist/ clinician used an electrocardiogram (ECG) to monitor the heart rate and rhythm of patients then read the journal activity of patients to diagnose the presence of arrhythmias and to develop appropriate treatment plans. However, The classical techniques take time and effort. The development of arrhythmias diagnosis, toward computational processes, such as arrhythmias detection and classification by using machine learning and deep learning. A convolutional neural network (CNN) is a popular method used to classify arrhythmia. Dataset pre-processing was also considered to achieve the best performance models. MIT-BIH Arrhythmia Database was used as our dataset. Our study used the EfficientNet-V2 which is a type of convolutional neural network to perform the classification of five types of arrhythmias. In pre-processing, the ECG signal was cut each 1 second (360 data), signal augmentation is applied to balance the amount of data in each class, and then the Continues Wavelet Transform (CWT) is employed to transform the ECG signal into a scalogram. The dataset is then distributed into subsets by using modulo operation to get variants of data in each subset. The colormap is applied to convert scalograms into RGB images. By this scheme, our study achieved superior accuracy than the existing method, with an accuracy rate of 99.97%.

Keywords— arrhythmia, Continues Wavelet Transform (CWT), scalogram, colormap, EfficientNet-V2.

I. INTRODUCTION Cardiovascular (CVD) is mentioned by WHO (World

Health Organization) as the leading cause of global death. An estimated 32% of the world's population or 17.9 million people died from CVDs in 2019. Out of the 17 million of those who died are being under the age of 70. Early detection of cardiovascular disease is important for management with counseling and effective treatment [1]. It can be detected by examining the presence of arrhythmias. Arrhythmia is an abnormal heartbeat rhythm, which is analyzed in an electrocardiogram (ECG) signal [2]. The heartbeat rhythm

classification is a crucial step for arrhythmia diagnosis during electrocardiographic (ECG) analysis [3]. In the clinical setting, a cardiologist/ clinician uses an ECG to monitor the patient's heart rate and rhythm and reads the patient's activity journal to diagnose the presence of an arrhythmia. The classical clinic setting, on the other hand, needs time and expert review [4]. To solve this problem, computerized-based analysis has been applied [5]. The disruption of digital technology has led researchers to develop cardiac detection methods based on machine learning by using feature extraction such as morphological features and RR interval [6,7]. To classify heartbeats based on ECG signals, some algorithm was proposed, such as particle swarm optimization (PSO), rule-based rough sets, support machine vector (SVM), linear discriminant analysis (LDA), neural network [8], and Extreme Learning Machine (ELM) [9] which has good performance.

However, the ECG waveform and their morphological characteristics of each patient have different variations, and at different times taken, the ECG signal of patients will be different. The fixed features used in these methods have low accuracy in distinguishing arrhythmias of different patients. Recently, deep learning-based methods have attracted more attention because they can automatically extract discriminant features from the training data. Studies [10,11] showed that deep learning-based methods have provided significant performance in medical image analysis, and several studies [12-15] have also showed that they can extract more abstract features and resolve variations between patients in ECG classification.

The other problem of ECG classification is different frequency components even noise composed in ECG signal which difficult to extract discriminant features by a deep learning based method. The known method to solve this problem is to transform the ECG signal into a time-frequency domain to avoid the effects of aliasing of different frequencies components. As the authors know, Short-Time Fourier Transform (STFT) [16], High-Order Synchro squeezing Transform (FSSTH) [17] and Wavelet Transform (WT) [18, 30] are widely used time-frequency techniques. In some studies, WT provides better time-frequency domain analysis results than STFT [19,20]. The output of wavelet

transformation are wavelet coefficients, then they are arranged to form a scalogram (coefficient value). The conversion of scalogram into RGB images make the classification easier. The way of the coefficient values are mapped into colors has an influence on the visual interpretation of the RGB Images. The selection of the appropriate colormap is an important step of data exploration. The color mapping is used to represent ECG signals from the scalogram (coefficient value) to RGB images. How data are represented visually has a powerful effect on how the structure in those data is perceived and analyzes interpretation of the data [21]. Dataset pre-processing and distribution also play an important role in training performance. For developing machine learning and deep learning model, it is common to split the dataset into training and testing sets to protect the model against unexpected issues that can arise due to overfitting [22].

Motivated by these challenges and the many pieces of information that could potentially be used to improve the performance of the cardiac arrhythmia classification on electrocardiogram, the authors developed an automatic arrhythmia classification method based on setting in pre-processing including employing Continuous Wavelet Transform (CWT) to convert ECG signal into scalogram, after then used a modulo operation to divide the dataset into subsets to get variants of data in each subset. The colormap was applied to convert the scalogram into RGB images and classified the 2D RGB diagrams. The Convolutional Neural Network (CNN) is used to perform the training and classification process. The EffecientNet-V2 architecture was chosen as a convolutional neural network (CNN) model to classify the types of arrhythmias in this term of study. Our method achieved an accuracy rate of 99.97%.

II. DESIGN AND IMPLEMENTATION This research study is implementing some techniques in

the pre-processing dataset and used EfficientNet-V2 to classify arrhythmia into 5 classes, including Fusion of Ventricular and Normal (FVN), Left Bundle Branch Block (LBBB), Normal Beat (NOR), Premature Ventricular Contraction (PVC), Right Bundle Branch Block (RBBB). Our study aims at the pre-processing of arrhythmia dataset, where the 1D ECG signals will be converted into 2D scalogram RGB images, and then fed the images into EfficientNet-V2 to classify the types of arrhythmias. The diagram block of the proposed system shows in Figure 1.

Fig. 1 Diagram Block of the Proposed System.

A. Preprocessing - Signal Segmentation The dataset was obtained from the MIT-BIH Arrhythmia

Database (https://physionet.org). These include 48 records with 30 minutes of each record (650,000 data) and frequency 360 Hz. It is used as data in the training process, validation process, and testing process. The dataset of ECG signal is cut for 1 second with a fixed-size of 360 data by taking 180 samples before and after the R-peak. The type of arrhythmic disease is known as a label at the annotation point in the dataset. The cut Signals are shown in Figure 2.

(a)

(b)

(c)

(d)

(e)

Fig. 2. ECG Signals, (a). FVN, (b). LBBB, (c). Normal Beat, (d). PVC, (e). RBBB.

The number of samples in each class has different values, Table I shows the type and number of samples from arrhythmia dataset.

TABLE I Type and samples of arrhythmia.

Arrythmia Number of Samples

FVN 802

LBBB 8071

NOR 75011

RBBB 7255

PVC 7129

B. Preprocessing - Dataset Augmentation and Partition As shown in Table I, FVN has the least amount of data

(802 data). Augmentation was applied to increase the variation of the data by manipulating the transformation of the data for the balance of the dataset and avoiding overfitting of the model trained [31,32,33]. The process of augmentation in FVN is shifting the middle point (R-peak) ± 5 samples to the right and ± 5 samples to the left of each data and has result 8020 total sample data. In the end, PVC has the fewest samples of the classes (7129 data). To balance the class of our dataset, samples of PVC are used as a reference for equalizing the amount of all classes data. The dataset's partition into three subsets, with a split ratio: 80% of the training set, 10% of the validation set, and 10% of the testing set. Table II shows the partition of ECG signal samples for the dataset.

TABLE II Partition of ECG Samples After Augmentation.

Arrythmia Dataset

Total Testing Training Validation

FVN 713 5703 713 7129

LBBB 713 5703 713 7129

NOR 713 5703 713 7129

RBBB 713 5703 713 7129

PVC 713 5703 713 7129

C. Preprocessing - Dataset Distribution The study [2] used MIT-Dataset with spectrogram as

representation of Frequency Spectrum from ECG signal and CNN as classification. It classified five classes of arrhythmia. The total data of each class is 7129. It used the split ratio of the dataset with the following: 80% of set Training, 10% of set Validation, and 10% of set Testing. There are several methods of distributing data into a subset of testing, training, and validation. To the best of our knowledge, one of them is by using random/shuffle in partition data into subsets [39,40]. In the study [2], they distributed the data into the subset of testing, training, and validation by regular partition. They divide the data from range 0 to 712 into the testing sets (713 data), then the data from range 713 to 6415 as the training sets (5703 data), and data from range 6416 to 7128 as validation sets (713). Instead of using the same method, the

authors try to divide data into subsets by using the modulo operation. The distribution of data into subsets of testing and validation used modulus by 9 and for the subsets of training used modulus by 3. Figure 3 shows the diagram block of distribution data into subsets.

Fig. 3 Diagram Block of data distribution into subsets using a modulo

operation.

The distribution of data on the subsets for each class with the modulo operation was shown in Table III.

TABLE III Distribution of Data into Subsets of Each Class.

Subset Data Total Data Testing Set 0, 9, 18, . . ., 6408 713 Training Set 2, 3, 4, . . ., 7128 5703 Validation Set 1, 10, 19, . . ., 6409 713

Figure 4 shows an example how the ECG plot (before

transformed into scalogram RGB images) distributed to the training subset of the FVN class using the modulo operation.

Fig. 4 Sets of Data from FVN Class used modulo operation distribution.

To compare the used of modulo operation in distributing data into subsets, the authors also applied the distribution method such as the study [2]. The diagram block of distribution data into subsets with regular partition was shown in Figure 5.

Fig. 5 Diagram Block of data distribution with regular partition.

The distribution of data on the subsets for each class with regular partition was shown in Table IV.

Table IV Distribution of Data into Subsets with Regular Partition.

Subset Data Total Data Testing Set 0, 1, 2, . . ., 712 713 Training Set 713, 714, 715, . . ., 6415 5703 Validation Set 6416, 6417, 6418, . . ., 7128 713

In the Figure 6 showed the example of ECG plot data in the subsets of training from FVN class before transformed into scalogram RGB images with regular partition.

Fig. 6 Sets of data from FVN class used regular partition.

D. Preprocessing - Transformation of ECG Signal into Scalogram

The Time-frequency domain was used to facilitate feature

extraction. In this study, the ECG signal was transformed using Continuous Wavelet Transform (CWT) to decomposed signals in the time-frequency domain. Continuous Wavelet Transform (CWT) is defined as:

𝐶𝐶𝑎𝑎(𝑏𝑏) =1√𝑎𝑎

� 𝑥𝑥(𝑡𝑡) .𝜑𝜑 �𝑡𝑡 − 𝑏𝑏𝑎𝑎

� 𝑑𝑑𝑡𝑡∞

−∞ (1)

where a is the scale of parameter, b is the translation parameter, and φ(t) is the wavelet function (mother wavelet). The scale can be converted to frequency by:

𝐹𝐹 =𝐹𝐹𝐶𝐶 ∗ 𝑓𝑓𝑠𝑠𝑎𝑎

(2)

where Fc is the center frequency of the mother wavelet, fs is the sampling frequency of signal x(t) [4].

In this study, continuous wavelets transform has been implemented, and the authors experimented with some different mothers of wavelets, such as morlet (morl), complex morlet (cmor0.2-1.8), frequency B-spline wavelets (fbsp2-0.9-0.5), Gaus8, Gaus4, and mexican hat wavelets (mexh) [38].

The Morlet wavelet (morl) are defined as:

𝜑𝜑(𝑡𝑡) = exp − 𝑡𝑡2

2 cos(5𝑡𝑡) (3) While the complex morlet wavelet (cmorB-C) is given by:

𝜑𝜑(𝑡𝑡) = 1

√𝜋𝜋𝜋𝜋 exp − 𝑡𝑡

2

𝐵𝐵 exp j2π𝐶𝐶𝑡𝑡 (4)

Where B is the bandwidth and C is the center frequency. In this study, the authors set a complex-valued Morlet wavelet with a bandwidth parameter (B) of 0.2 and a center frequency (C) of 1.8.

Another one is frequency B-spline wavelets (fbspM-B-C) which are defined as:

𝜑𝜑(𝑡𝑡) = √𝜋𝜋 �sin �𝜋𝜋𝜋𝜋 𝑡𝑡

𝑀𝑀�

𝜋𝜋𝜋𝜋 𝑡𝑡𝑀𝑀

�

𝑀𝑀

exp 2jπ𝐶𝐶𝑡𝑡 (5)

Where M is the spline order, B is the bandwidth and C is the center frequency. In this study, the authors set the spline order (M) of 2, with a bandwidth parameter (B) of 0.9 and a center frequency (C) of 0.5.

For the Gaussian wavelets (gausP) correspond to the following wavelets:

𝜑𝜑(𝑡𝑡) = 𝐶𝐶 exp −𝑡𝑡2 (6) Where P is an integer between 1 and 8 corresponds to the

order derivative of the gaussian wavelet. The authors employ order 4 and order 8 of the Gaussian wavelet [38].

And the mexican hat wavelets (mexh) are defined as:

𝜑𝜑(𝑡𝑡) = 2

√3 √𝜋𝜋4 exp − 𝑡𝑡2

2 (1 − 𝑡𝑡2) (7)

Different Continuous Wavelet Transform (CWT) scale factors were used to obtain the signal wavelet coefficient at different scales. These wavelet coefficients can be represented as 2D ECG scalogram signals in the time-frequency domain. The last is to convert the wavelet coefficient of the scalogram into corresponding images with pseudo-color processing. Figure 7 are the results of the ECG signal in 2D images representation.

(a)

(b)

Fig. 7 (a). Scalogram image of Normal Beat, (b). Scalogram image of FVN

E. Preprocessing - Color Mapping Continuous Wavelet Transform (CWT) is utilized to

transform One-Dimensional electrocardiogram signals (1D-ECG) into corresponding images. It has output wavelet coefficients, and they are arranged to form a scalogram. The wavelet coefficients (scalogram) were converted into images by using pseudo-color with some different types of colormaps.

In this study, the authors convert ECG scalograms using some different types of colormaps to find the corresponding colormaps for our scalogram. The following colormap is Jet, Viridis, Paired, Pastel2, HSV, Tab 20c, Turbo, Inferno [41]. Different colormap representations for Normal Beat are shown in Figure 8, and for FVN is shown in Figure 9.

Jet

Turbo

Viridis

Inferno

Paired

Pastel2

HSV

Tab20c

Fig. 8. Scalogram of Normal Beat in different colormap

Jet

Turbo

Viridis

Inferno

Paired

Pastel2

HSV

Tab20c

Fig. 9. Scalogram of FVN in different colormap

F. EfficientNet-V2 A convolutional neural network (CNN) is a deep learning-

based algorithm which widely used to detect or classify arrhythmias [23]. The EfficientNet-V2 is the type of convolutional neural network algorithm, new series, and has a faster training speed and better parameter efficiency than the last family of EfficientNet (EfficientNet-V1). The previous version (EfficientNet-V1) also has been used to handle ECG data [42]. However, the EfficientNet-V2 model was developed with the combination of training-aware neural architecture search and scaling, to jointly optimize training speed and parameter efficiency. The model was searched from the search space enriched with new operations such as Fused-MBConv [27].

III. DISCUSSION This section presents the result of our experiment, and the

design system analyzes. Tests were carried out to determine the error rate and draw a conclusion from the system proposed. There are three implementations and tests carried out, including:

1. The result of using different mother wavelet implementations.

2. The result of the used modulo operation on data distribution into subsets.

3. The result of different colormap (CMAP) implementation.

In this section of the experiment, training and validation were performed by efficientnetv2-b0 model, 1k checkpoints, and (224 x 224) input images. The number of epochs [35] is 100 and 64 of batch size [36]. The early stopping function is used to stop the training process when the performance saturates or begins degrading [37]. At the end of this section, the report of the best-performed model in this study is presented.

A. Testing the used of different mother of wavelet This part presents the result from the use of different

mother wavelets in ECG signal time-frequency transformation. The jet colormap was applied to uniform the scalogram RGB images representation. In this experiment, the 1D ECG signals were transformed into a scalogram (wavelets coefficient) by using different mother wavelets, and then the scalogram (wavelets coefficient) was converted into 2D RGB images by using a jet colormap. The mother wavelet shown in Figure 10 includes (a). Mexican Hat Wavelet (mexh), (b). Morlet Wavelet (morl), (c). Gaussian Wavelet (gaus8), (d). Complex Morlet Wavelet (cmor0.2-1.8), (e). Gaussian Wavelet (gaus4), and (f). Frequency B-Spline Wavelet.

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 10. Wavelet Functions.

Table V shows accuracy detection from each class with a different wavelet. The mexican hat achieved a superior result than other wavelets.

Table V Accuracy detection from each class with different wavelet

Wavelet Accuracy (%) Acc.

Rate (%) FVN LBBB NOR PVC RBBB

Mexican Hat 100 99.86 100 100 100 99.97

Morlet 99.86 99.72 100 99.16 99.72 99.69

Gaus8 99.86 99.44 100 99.16 99.86 99.66 Complex Morlet 99.86 99.30 100 99.16 99.86 99.64

Gaus4 99.72 99.16 100 99.30 99.72 99.58

B-Spline 99.72 99.02 100 98.74 99.72 99.44

B. Testing result in different colormap In this part, we used the mexican hat (mexh) as the mother

wavelet to transform the ECG signals into a scalogram, so all the experimental data used the same transformation method. The different colormap was applied in this experiment to present the effect of using different colormaps as the representation of wavelet coefficient (ECG signal in time-frequency) in RGB Diagrams. Table 6 shows the type of colormap and accuracy in each class of arrhythmia. The authors used some different types of colormaps, with the following: Jet colormap, Turbo colormap, Viridis colormap, Inverno colormap, Paired colormap, Pastel2 colormap, HSV colormap, and Tab 20c colormap. The code and type of colormap can find in matplotlib library [41].

Table VI Colormap and accuracy of arrhythmias

CMAP Accuracy (%) Acc.

Rate (%) FVN LBBB NOR PVC RBBB

Jet 100 99.86 100 100 100 99.97

Turbo 99.72 99.72 100 99.16 100 99.72

Viridis 100 99.58 100 99.16 99.72 99.69

Inverno 99.72 99.72 100 99.16 99.72 99.66

Paired 99.86 99.30 100 99.02 99.86 99.61

Pastel2 100 99.86 100 98.32 99.72 99.58

HSV 99.58 99.58 100 98.88 99.72 99.55

Tab 20c 99.86 99.30 100 98.32 99.72 99.44

C. Testing the result of using modulo operation in data distribution. This part presents the result of the experiment which used

modulo operation and regular partition in distributing data into subsets of training, validation, and testing. The data distribution is shown in Table III and Table IV. In this experiment, the authors used the mexican hat (mexh) as the mother wavelet to uniform the experimental data in the scalogram representation. Then used three types of colormaps and two models of data distribution (used modulo operation and regular partition). The result of the used modulo operation in dataset distribution achieved better accuracy than regular

partition. Table VII shows the accuracy rate of this experiment.

Table VII Accuracy detection with Variants and Regular of data distribution.

Dataset Partition

Accuracy (%) Acc. Rate (%) FVN LBBB NOR PVC RBBB

Jet with Variants 100 99.86 100 100 100 99.97

Jet regular 74.33 99.86 100 97.33 99.72 94.25

Viridis with Variants

100 99.58 100 99.16 99.72 99.69

Viridis regular 73.49 100 100 98.18 99.72 94.28

Hsv with Variants 99.58 99.58 100 98.88 99.72 99.55

Hsv regular 67.04 99.86 100 96.77 99.44 92.62

Table VII shows the use of modulo partition in data

distribution (Jet with Variants, Viridis with Variants, Hsv with Variants) and the regular partition (Jet Regular, Viridis Regular, Hsv Regular) in the same colormap representation. The use of jet as colormap with variants distribution achieved an accuracy rate of 99.97%. The use of Viridis as colormap with variants distribution achieved an accuracy rate of 99.69% (this part was presented in ICAST2021 [34]), and the use of HSV as colormap with variants distribution achieved an accuracy rate of 99.55%. However the use of jet as colormap with regular partition achieved an accuracy rate of 94.25%, the use of Viridis as colormap with regular partition achieved an accuracy rate of 94.28%, and the use of HSV as colormap with regular partition achieved an accuracy rate of 92.62%.

D. The report of the best-performed model in this study. This part describes the best method obtained from the

result of our experiments. After conducting an experiment with different mother of wavelets to transform the ECG signals into scalogram (wavelet coefficient), then used the different colormap to convert the scalogram (wavelet coefficient) into RGB images and distributed the data into subsets by using modulo operation and regular partition method. Finally, we found the best combination model that can improve the accuracy results of the arrhythmia classification. By experimenting with several different mother wavelets, the best results were obtained by mexican hat as the mother of wavelets. In the experiment with different colormap representations, the best result is achieved by jet colormap, and the best method in distributing data into a subset of training, testing, and validation is achieved using modulo operation.

Our study was performed by efficientnetv2-b0 model, with 1k checkpoints, and (224 x 224) input images. The number of epochs was set to 100 and used 64 of batch size. Early stopping function is applied. The testing is carried out

using a dataset with the pre-processing method (ECG Signals Transformed by Mexican hat and jet colormap), with a total of 5703 training samples, 713 validation samples, and 713 testing samples (distributed by modulo operation). The best accuracy result of our study is 99.97%. The graph of accuracy and loss value are shown in Figure 11.

(a)

(b)

Fig. 11 (a). Graph of Accuracy Value, (b). Graph of Loss Value

In our study, the best performance was achieved by using Mexican hat (mexh) as wavelet, variants in data distribution into subsets of training, validation, and testing, and jet as a colormap. With this schema, our proposed method approaches an accuracy rate of 99.97% with the following details FVN class correctly detecting 713 samples (100%), LBBB class as many as 712 samples (99.86%), Normal class as many as 713 samples (100%), PVC class as many as 713 samples (100%), and RBBB class as many as 713 samples (100%). The confusion matrix of arrhythmia classification was shown in Figure 12.

Fig. 12 Confusion Matrix from Testing Result.

The confusion matrix in Figure 12 shows the arrhythmia classification on the test set. The total number of samples in each class is indicated inside parenthesis. Numbers inside blocks are the number of samples classified in each category. To evaluate the model from our study in more depth, we perform calculations of precision, recall, and F1-score. Precision is used to measure the degree of precision/accuracy of the model in terms of the values predicted positive, how many of them are positive. The precision is defined as:

Precision =𝑇𝑇𝑇𝑇

𝑇𝑇𝑇𝑇 + 𝐹𝐹𝑇𝑇 (8)

Where TP is True Positive, and FP is False Positive. Recall is used to measure the degree of accuracy of the model of predicting the actual positives with respect to the values which are positive. The recall is defined as:

Recall =𝑇𝑇𝑇𝑇

𝑇𝑇𝑇𝑇 + 𝐹𝐹𝐹𝐹 (9)

Where TP is True Positive, and FN is False Negative. F1-Score is used to seek balance between, previously mentioned metrics, Precision and Recall. The F1-Score is defined as:

F1 − Score = 2 xPrecision x Recall

Precision + Recall (10)

The result of Precision, recall, and F1-Score from our study is shown in Table VIII.

Table VIII Classification Report of Our Proposed Method.

Class Precision Recall F1-score

FVN 1.0000 1.0000 1.0000

LBBB 1.0000 0.9986 0.9993

Normal 0.9986 1.0000 0.9993

PVC 1.0000 1.0000 1.0000

RBBB 1.0000 1.0000 1.0000

Accuracy 0.9997

In this study, the authors also compared the classification accuracy result with existing method. Table IX compares the classification accuracy between the existing work and our method.

Table IX The classification accuracy of existing works and our method.

It is difficult to choose the best comparison method. It is only possible to compare reports that classified the same type of arrhythmias and the same number of classes from the same data set [2]. We can only report the average accuracy from elated studies.

IV. CONCLUSION This study used the EfficientNet-V2 method to perform

the classification of five arrhythmias, including FVN, LBBB, Normal, RBBB, and PVC. The ECG database was obtained from MIT-BIH Arrhythmia Database. The ECG signal was cut each 1 second (360 data). Continuous Wavelet Transform (CWT) is employed to transform the 1D-ECG signal into a scalogram (wavelet coefficient). The dataset was distributed with a modulo operation to get variants in each subset. The colormap is applied to convert scalograms into RGB images. From the results of the tests that have been carried out, several conclusions can be drawn as follows:

1. Mexican hat (mexh) wavelet was achieved a superior accuracy rate among our testing mother wavelets.

2. The use of modulo operation in distribution of datasets, has an impact on accuracy rate. The subset with variant of data (distributed with modulo operation) achieved better accuracy than regular partition distribution.

3. The jet colormap achieved a superior accuracy rate among our testing colormaps.

This study achieved superior accuracy results than previous studies [2] with an accuracy rate of 99.97%. This method can consider used in arrythmia classification system in the future.

ACKNOWLEDGMENT The authors would like to thank the Computer Engineering

department Faculty of Intelligent Electrical and Informatics Technology (F_ELECTICS- ITS) for supporting this research.

REFERENCES [1] World Health Organisation, “Cardiovascular Diseases (cvds),” 2019,

https://www.who.int/en/news-room/fact-sheets/detail/cardiovascular-diseases-(cvds), Last accessed on 2021-10-30.

[2] A. Kurniawan et al., “Arrhythmia Classification on Electrocardiogram Signal Using Convolution Neural Network Based on Frequency Spectrum,” 2020 International Conference on Computer Engineering, Network, and Intelligent Multimedia (CENIM), 2020, pp. 29-33.

[3] A. Y. Hannun et al., “Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network,” Nature medicine, vol. 25, 2019, pp. 65-69.

[4] T. Wang et al., "Automatic ECG Classification Using Continuous Wavelet Transform and Convolutional Neural Network," Entropy (Basel), 2021, 23:119.

[5] E. Svennberg et al., “Safe automatic one-lead electrocardiogram analysis in screening for atrial fibrillation,” EP Europace, vol. 19, 2017, pp. 1449–1453.

[6] M. Shen et al., “Atrial Fibrillation Detection Algorithm Based on Manual Extraction Features and Automatic Extraction Features,” IOP Conference Series: Earth and Environmental Science, vol. 428, 7th Annual International Conference on Geo-Spatial Knowledge and Intelligence 20–21 December 2019, Guangzhou, China.

[7] S. Sahoo et al., “Machine Learning Approach to Detect Cardiac Arrhythmias in ECG Signals: A Survey,” IRBM 41, 2020, pp. 185–194

[8] U. Satija, B. Ramkumar and M. S. Manikandan, "A New Automated Signal Quality-Aware ECG Beat Classification Method for Unsupervised ECG Diagnosis Environments," in IEEE Sensors Journal, vol. 19, no. 1, pp. 277-286, 1 Jan.1, 2019.

[9] Yuefan Xu, Sen Zhang, Zhengtao Cao, Qinqin Chen, Wendong Xiao, "Extreme Learning Machine for Heartbeat Classification with Hybrid Time-Domain and Wavelet Time-Frequency Features", Journal of Healthcare Engineering, vol. 2021, Article ID 6674695, 12 pages, 2021.

Ref. Dataset Class Pre-processing

Method Accuracy

(%) Segmentation Transformation

A. Kurniawan [2] MIT-BIH Normal, RBBB, LBBB, FVN, PVC. Signal was cut into pieces for 1 second (360 data)

Spectrograms using Overlap STFT

CNN 98.60

T. Wang [4] MIT-BIH AAMI Classes (N, V, S, F) Signal was cut into pieces for 200 data

Scalogram Using CWT CNN 98.74

J. Huang [24] MIT-BIH Normal, RBBB, LBBB, PVC, APC. Signal was cut into pieces for 10 second

Spectrograms using STFT CNN 99.00

A. Ullah [25] MIT-BIH PVC, PAB, RBBB, LBBB, APC, VFW, VEB. - Spectrograms

using STFT CNN 99.11

M. Naz [26] MIT-BIH, CUDB Ventricular arrhythmias (VA)

Signal slice into 1,024-sized segment

Reshaped into 32 x 32 binary images

AlexNet, VGG-16 Inception-v3 + Cubic SVM

97.60

Enbiao Jing [28] MIT-BIH Ventricular ectopic (V) Signal sliced each 3 seconds long

Daubechies Wavelet (db8) ResNet-18 96.50

M. Wu [29] MIT-BIH Normal, RBBB, LBBB, PVC, APC. Signal was cut into pieces for 360 data

- CNN 97.41

Our method MIT-BIH Normal, RBBB, LBBB, FVN, PVC. Signal was cut into pieces for 1 second (360 data)

RGB Images Scalogram Using CWT and Color mapping

EfficientNet-V2 99.97

[10] I. Irmakci, S. M. Anwar, D. A. Torigian and U. Bagci, "Deep Learning for Musculoskeletal Image Analysis," 2019 53rd Asilomar Conference on Signals, Systems, and Computers, 2019, pp. 1481-1485.

[11] S. M. Anwar et al., ”Medical image analysis using convolutional neural networks: A review,” J. Med. Syst., 2018, 42, 226.

[12] M. Salem, S. Taheri and J. Yuan, "ECG Arrhythmia Classification Using Transfer Learning from 2- Dimensional Deep CNN Features," 2018 IEEE Biomedical Circuits and Systems Conference (BioCAS), 2018, pp. 1-4.

[13] Ö. Yıldırım et al., “Arrhythmia detection using deep convolutional neural network with long duration ECG signals,” Comput. Biol. Med., 2018, vol. 102, pp. 411–420.

[14] Y. Wu et al., “A comparison of 1-D and 2-D deep convolutional neural networks in ECG classification,” Conference proceedings : Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society, Annual Conference 2018, 2018, pp. 324-327.

[15] J. Zhao, X. Mao, and L. Chen, “Speech emotion recognition using deep 1D & 2D CNN LSTM networks,” Biomedical Signal Processing and Control, vol. 47, pp. 312–323, Jan. 2019.

[16] T.Virtanen, E. Vincent, S. Gannot, “Time-frequency processing - Spectral properties,” Audio source separation and speech enhancement, Wiley, 2018, 978-1-119-27989-1. hal-01881426.

[17] P. Bing, W. Liu and Z. Zhang, "DeepCEDNet: An Efficient Deep Convolutional Encoder-Decoder Networks for ECG Signal Enhancement," in IEEE Access, vol. 9, pp. 56699-56708, 2021.

[18] Barmase, Swapnil, S. Das and S. Mukhopadhyay, “Wavelet Transform-Based Analysis of QRS complex in ECG Signals,” ArXiv abs/1311.6460, 2013.

[19] P. S. Addison, ‘‘Introduction to redundancy rules: the continuous wavelet transform comes of age,’’ Philos. Trans. Ser. A, Math. Phys. Eng. Sci., vol. 376, Aug. 2018, Art. no. 20170258. [Online]. Available: doi: 10.1098/rsta.2017.0258.

[20] M. Guo, X. Zeng, D. Chen and N. Yang, "Deep-Learning-Based Earth Fault Detection Using Continuous Wavelet Transform and Convolutional Neural Network in Resonant Grounding Distribution Systems," in IEEE Sensors Journal, vol. 18, no. 3, pp. 1291-1300, 1 Feb.1, 2018, doi: 10.1109/JSEN.2017.2776238.

[21] Rogowitz, Bernice E. et al., “How not to lie with visualization,” Computers in Physics 10, 1996, pp. 268-273.

[22] V. R. Joseph and A. Vakayil, Split: An optimal method for data splitting, arXiv preprint arXiv:2012.10945, 2020.

[23] Ali Haider Khan, Muzammil Hussain, Muhammad Kamran Malik, "Arrhythmia Classification Techniques Using Deep Neural Network", Complexity, vol. 2021.

[24] J. Huang, B. Chen, B. Yao and W. He, "ECG Arrhythmia Classification Using STFT-Based Spectrogram and Convolutional Neural Network," in IEEE Access, vol. 7, pp. 92871-92880, 2019.

[25] A. Ullah, S. M. Anwar, M. Bilal, and R. M. Mehmood, “Classification of Arrhythmia by Using Deep Learning with 2-D ECG Spectral Image Representation,” Remote Sensing, vol. 12, no. 10, p. 1685, May 2020.

[26] M. Naz et al., “From ECG signals to images: a transformation based approach for deep learning,” PeerJ. Computer science, vol. 7 e386, 10 Feb. 2021.

[27] M. Tan, Q. V. Le, “EfficientNetV2: Smaller Models and Faster Training,” arXiv:2104.00298v1 [cs.CV], 2021.

[28] Enbiao Jing, Haiyang Zhang, ZhiGang Li, Yazhi Liu, Zhanlin Ji, Ivan Ganchev, "ECG Heartbeat Classification Based on an Improved ResNet-18 Model", Computational and Mathematical Methods in Medicine, vol. 2021, Article ID 6649970, 13 pages, 2021.

[29] M. Wu et al., “A Study on Arrhythmia via ECG Signal Classification Using the Convolutional Neural Network,” Frontiers in computational neuroscience, vol. 14 564015, 5 Jan. 2021.

[30] Z. Zhao et al., "Noise Rejection for Wearable ECGs Using Modified Frequency Slice Wavelet Transform and Convolutional Neural Networks," in IEEE Access, vol. 7, pp. 34060-34067, 2019.

[31] P. Cau et al., “A novel data augmentation method to enhance deep neural networks for detection of atrial fibrillation,” Biomedical Signal Process and Control, vol. 56 101675, 2020.

[32] S. Alladi, “Augmenting Electrocardiogram Datasets using Generative Adversarial Networks,” MSc Thesis, Faculty of The Graduate School of The University of Minnesota, 2020. [Online]. Available:https://irvlab.cs.umn.edu/sites/irvlab.dl.umn.edu/files/santhosh_thesis_final.pdf

[33] P. Silva et al., “Towards better heartbeat segmentation with deep learning classification,” Sci Rep 10, 20701, 2020. [Online]. Available: https://doi.org/10.1038/s41598-020-77745-0

[34] Muhammad Furqon et al., “Classification of Arrhythmia by Using EFFICIENTNETV2 with 2-D Electrocardiogram Spectral Image Representation,” The 16th International Student Conference on Advanced Science and Technology (ICAST 2021), no. 10-4, 2021.

[35] K. Kaur, R. Dhir and K. Kumar, "Transfer Learning approach for analysis of epochs on Handwritten Digit Classification," 2021 2nd International Conference on Secure Cyber Computing and Communications (ICSCCC), 2021, pp. 456-458, doi: 10.1109/ICSCCC51823.2021.9478102.

[36] G. Beatty, E. Kochis and M. Bloodgood, "Impact of Batch Size on Stopping Active Learning for Text Classification," 2018 IEEE 12th International Conference on Semantic Computing (ICSC), 2018, pp. 306-307, doi: 10.1109/ICSC.2018.00059.

[37] C. Corneanu, M. Madadi, S. Escalera and A. Martinez, "Explainable Early Stopping for Action Unit Recognition," 2020 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2020), 2020, pp. 693-699, doi: 10.1109/FG47880.2020.00080.

[38] Lee et al., “PyWavelets: A Python package for wavelet analysis. Journal of Open Source Software,” 2019, 4(36), 1237, https://doi.org/10.21105/joss.01237.

[39] Qi Meng et al., “Convergence analysis of distributed stochastic gradient descent with shuffling,” Neurocomputing, vol. 337, pp. 46-57, 2019, doi: 10.1016/j.neucom.2019.01.037.

[40] L. Fassina et al., “A Random Shuffle Method to Expand a Narrow Dataset and Overcome the Associated Challenges in a Clinical Study: A Heart Failure Cohort Example,” Frontiers in Cardiovascular Medicine, vol. 7, pp. 263, 2020, doi: 10.3389/fcvm.2020.599923.

[41] J. D. Hunter, "Matplotlib: A 2D Graphics Environment," Computing in Science & Engineering, vol. 9, no. 3, pp. 90-95, 2007.

[42] N. Nonaka and J. Seita, "Electrocardiogram Classification by Modified EfficientNet With Data Augmentation," 2020 Computing in Cardiology, 2020, pp. 1-4, doi: 10.22489/CinC.2020.063.

ISBN978-4-89362-375-1 (C) 2021 TRON Forum

HARNESSING IOT TECHNOLOGY FOR THE

DEVELOPMENT OF WEARABLE CONTACT

TRACING SOLUTIONS

Acharya Rex

Dept. of Computer Science

Sam Houston State University

Huntsville, TX, USA

[email protected]

Baza Mohamed


College of Charleston

Charleston, SC, USA

[email protected]

Rasheed Amar



Huntsville, TX, USA

[email protected]

Mozer-Sallo Louanne



Huntsville, TX, USA

[email protected]

Varol Hacer

Dept. of Physics, Engineering, and

Astronomy

Stephen F. Austin State University

Nacogdoches, TX, USA

[email protected]

Mahapatra Rabi

Dept. of Computer Science and

Engineering

Texas A&M University

College Station, TX, USA

[email protected]

Abstract— Advancements in mobile computing and

embedded system technologies show superiority in controlling

the transmission of a virus during the COVID-19 pandemic.

They provide rapid contact data compared to manual contact

tracing and medical monitoring methodologies. Data tracing

capabilities were achieved by existing technologies via the

utilization of smartphone-based exposure notifications systems

and proximity sensing tools based on Bluetooth/GPS. Such

systems lack user privacy and data anonymization capabilities,

it also requires that users have continuous access to

smartphones. This paper proposes the development of a

lightweight wearable prototype for contact tracing. The

proposed system is based on the deployment of an IoT

development board, incorporated with a Bluetooth chipset for

proximity sensing. To overcome the problem of user-to-

smartphone accessibility, the processing of contact data by the

exposure notification system has been migrated from mobile

computing to a contact tracer management server that is

managed by the community. Daily contact exposures to the virus

are recorded by each user’s contact tracing system and stored

in the user’s blockchain. Data integrity and immunity against

data injection attacks were achieved via the implementation of

lightweight block chaining and validation schemes. Data

anonymization was also supported via the utilization of the

onboard AES crypto engine. Meanwhile, support for user

anonymity was incorporated into the proposed system through

the implementation of an anonymous authentication protocol

based on the Zero-Knowledge Proof (ZKP) approach. Finally,

we have analyzed the performance of the system in terms of

power consumption under various systems settings, such as

encryption key size and the required number of ZKP validation

attempts per authentication session.

Keywords—IoT, Zero-Knowledge Proof, Contact tracing

I. INTRODUCTION

The latest Coronavirus pandemic has reshaped the way we

do business, a new era of workplace that supports a healthy

workforce and business operations continuities has been

established. Technologies that support contact tracing

capability [1], [2], [3] and automated social distancing

tracking have become an integral part of today’s work

environment. Two existing technologies that offer

comprehensive solutions for incorporating contact tracing

services into today’s workplace are (i) proximity tracking

technology [4] (ii) GPS-enabled technology [5]. Both

technologies rely on users’ accessibility to existing resources,

such as smartphones and cellular communication

availabilities. These technologies utilize the sensing

capability of a smartphone to compute proximity and identify

any exposure to an individual with COVID-19 positive. They

use Bluetooth and GPS features to measure individual-to-

individual proximity.

Contact tracing solutions based on proximity tracking

tools have been implemented in several countries. Such

technologies are effective to decrease manual contact

processing, accelerating the identification of positive cases

within a community, and helping facilitate rapid

dissemination of contact tracing data to individuals.

Additionally, temporal data augmented with proximity

sensing data could be fully utilized by contact tracer

management systems to locate, detect, and rapidly isolate

areas within communities and workplaces that have a high

exposure rate to the virus.

Although, empirical data that investigates the

performance of these technologies shows the usefulness and

effectiveness of such systems. They still have some of the

following disadvantages:

Socioeconomic and technology acquisition biases. Users

need to have access to a smartphone and cellular data

communication services. Individuals require to know

how to install and run apps.

The optimal result could not be achieved without the

support of widespread adaptions of these contact tracer

technologies by everyone in a community.

All individuals in a community or a workforce must

maintain continuous access to their smartphone devices

and cellular communication services.

Specific functions on the smartphone must be always

enabled to provide accurate contact tracing results. This

includes Bluetooth and GPS features. The latter might

create conflict with other applications settings,

specifically settings related to user privacy and location

sharing.

Data privacy, data confidentiality, and unauthorized

access. Sensitive user data including user’s location data

are shared with an online contact tracer management

system. Users’ temporal data stored on the management

systems could be comprised and utilized to track user

locations [6], [7].

This research effort presents a wearable IoT platform that

enables contact tracing among individuals of a community

while preserving users’ data privacy and users’ anonymity. In

addition to providing users’ anonymity and data privacy, data

integrity for contacts exposure was achieved through the

integration of a block chaining approach. The proposed

system offers a cost-effective, energy-efficient, and phone-

independent contact tracing mobile solution. It provides

contact tracing capability for the workspace that cell service

is inaccessible (e.g., underground mining field, offshore oil

and gas operations, etc.). Unlike existing smartphone-based

exposure notification systems, the proposed technology will

be able to provide contact tracing services for controlled

communities and workspace. Specifically, monitoring

employees’ and workers’ exposure to the virus in their

workplaces, factories, students, and staff members in

educational institutions, and including senior home

communities.

Our main contribution in this paper is the development of

a wearable IoT-based prototype that is capable of tracking

user daily contacts and determining possible exposures to the

virus within a community or a workplace (see Fig. 1). The

proposed system is capable of logging hundreds of contact

tracing data using the device's internal memory. We utilize a

proximity sensing approach that makes use of the device’s

built-in Bluetooth feature to generate contact tracing data. A

user of the system will be able to discover possible exposure

to the virus by streaming the device's daily logs into an online

management tracer system. In this research, we have

considered an online contact tracer management system that

is capable of obtaining daily health status related to Covid-19

positive cases in a community/ workplace. Data collected via

the IoT-based tracer system is cross-referenced with data

obtained via the online contact tracer management system.

An exposure notification system is implemented to alert the

user for any possible exposure to the virus that occurred

during the last 24 hours.

The following technological capabilities were

implemented and integrated into the proposed IoT system:

User Anonymity: A lightweight interactive Zero-

Knowledge Proof scheme [8] is developed and

implemented on the proposed testbed. The scheme

supports the non-disclosure of devices’ IDs during data

generation. Devices’ IDs were obfuscated via the ZKP

scheme to provide resiliency against the device’s

spoofing attacks.

Anonymous Community Based Authentication: We

achieve anonymous authentication among users in a

community through the support of the ZKP scheme.

Devices’ IDs are used by the ZKP scheme to establish

unique authentication parameters for each device

deployed in the community. During devices’ contacts,

two devices will be able to authenticate each other

without revealing their authentication parameters.

Data Privacy: Resiliency against device-specific attacks

is supported in the proposed system. We have considered

the case where a malicious actor will be able to map a

unique ZKP parameter to a specific user in a community

(e.g., utilizing ZKP proofs data during device

authentication for device identification). Captured

information could be used to induce the current health

status of a user under attack and whether he/she is Covid-

19 positive/ negative.

To prevent the above attack, devices’ IDs are randomly

assigned by an online contact tracer management system

daily. This will present the attacker with a moving target.

Data Confidentiality: Contact tracing data is protected

via the utilization of AES encryption/Decryption engine.

Data computed during proximity exposure were ciphered

using the onboard AES-128, AES-192, or AES-256

encryption/decryption engine. The encrypted data is then

stored in the device’s internal memory. AES

encryption/decryption keys are preloaded into users’

devices before deployment. AES encryption/decryption

keys are shared with the online contact tracer

management system.

Data Integrity via Block Chaining Approach: To

maintain data integrity, we have developed and

implemented a lightweight block chaining scheme [9].

The proposed scheme uses a similar approach to the

blockchain when generating contacts tracing data blocks.

In our implementation, each block i was developed using

the device �� , timestamps �� , Hash of the last block in

the chain �� , user’s health status related to Covid-19

{�� ∈ | � �1, 0�� , where �� 1 indicate the

user is Covid-19 positive and �� 0 refers to Covid-

19 negative. Blocks are hashed via SHA-256 and stored

in the device's internal memory.

We have analyzed the performance of the proposed

system by measuring the device’s power consumption during

runtime under various encryption/decryption key lengths. We

have estimated the device’s dynamic power dissipation with

predefined ZKP protocol configuration parameters.

This paper is organized as follows: Section II provides

preliminaries and related works. Section III describes

hardware components that are used to design and implement

Fig. 1 Contact tracing in a workplace

Contact tracing

the proposed contact tracing system. Section IV presents

threat models. Section V introduces security primitives and

techniques used to support user anonymity and

authentication, data privacy, data anonymization, and data

integrity. Section VI describes the performance analysis of

the proposed system under various device settings. Section

VII concludes the paper and describes future research

directions.

II. PRELIMINARIES AND RELATED WORKS

This section introduces two main techniques which are

fully designed and implemented on the proposed testbed (i)

an interactive ZKP protocol (ii) a lightweight block chaining

scheme.

A. The Interactive ZKP Protocol

Zero-Knowledge Proof [8],[9], [10] is an interactive

scheme that allows two systems, one called the prover P, and

the other system is called the Verifier V. ZKP system enables

a system acting as prover to prove his identity many times to

a verifier V using polynomial validation approach. ZKP

system prohibited a verifier V from misrepresenting himself

as a prover to someone else. During the ZKP approach,

multiple proofs are computed by the prover and

communicated to the verifier. The number of proofs that are

required for successful validations is predefined in the ZKP

protocol. ZKP system offers a tradeoff between scheme

efficiency and security requirements. For each system

participating in the ZKP protocol, a secret key �� is

established and �� is computed and published

over the network where m is chosen to be the product of two

large prime numbers of the form 4� + 3. In the ZKP scheme,

�� serves as wittiness for whether a certain number is a

quadratic residue or quadratic non-residue (�� ) without

revealing a signal bit of information related to the secret key

�� . In this research, anonymous authentication and user

privacy were achieved via the implementation of ZKP on an

IoT development board. Dynamically assigned devices’ IDs

were established via the deployment of ZKP secret keying

information.

ZKP provides a lightweight identification mechanism, it

is suitable for power-limited devices, smart card

technologies, and IoT platforms. For a preselected set of ZKP

configuration parameters, the ZKP protocol will be able to

reach about two orders of magnitude of the processing speed

of an RSA-based authentication scheme. Section V provides

a full description of the implemented ZKP scheme. It presents

how the ZKP system can be used to support authentication

and user anonymity for the proposed wearable contact tracer

platform.

B. Chaining Technique for Data Blocks

Blockchain is a decentralized system [11], [12], [17-20]

that’s composed of shared and immutable public or private

ledgers. Transaction records are computed and stored on

public/private ledgers. Blocks or transaction records are

timestamped, encrypted, and linked together. Each block is

linked to the previous block in the chain through the

implementation of a cryptographic hash function. Since

blocks are cryptographically linked together in the chain, it

makes the system immutable against data injection and data

modification attacks. To support resiliency against data

injection and data modification attacks on the proposed

system, we have adopted a blockchain-based approach. In

this research effort, contact tracer management servers will

act as private ledgers. Blocks generated from each user’s

device will be timestamped, secured, and cryptographically

bonded to the previous data block. The timestamp of a block

is calculated based on the recording of the time for which two

devices become close to each other. In the proposed system,

a user's daily contact exposure data will be added to the user

contact’s tracing blockchain. Users’ blockchains are stored

and maintained by the contact tracer management system.

Daily contact exposure data blocks are recorded by the user’s

blockchain. A user’s blockchain captures all contact exposure

blocks that occurred within a 14-days period, which is based

on the Covid-19 virus incubation period. Users’ blockchains

are dynamically updated by the private ledger (contact tracer

management system). Users’ blockchains are verified daily

to capture contact exposure data blocks that might be expired

within the last 14-days. All expired contact exposure data

blocks will be removed from the user’s blockchain, the

blockchain will get updated by the contact tracer management

system.

In this paper, a lightweight blocks chaining algorithm is

implemented on the proposed testbed. A total of 1000 daily

contact exposure data blocks can be generated and logged

into the device's internal memory. To ensure the integrity of

users’ chains, new data blocks will be verified by the private

ledger via reconstruction of the blocks before adding them to

the chains. The contact tracer management systems will

recompute each block transmitted from the user’s contact

tracing device using the block raw data and a hash value of

the previous block in the chain. Section V illustrates the

proposed implementation for the block chaining technique.

III. HARDWARE AND SOFTWARE DEVELOPMENT TOOLS FOR

THE PROPOSED TESTBED

A. Hardware and Data Acquisition Tool

Bluetooth Module for Proximity Sensing: We achieved

proximity sensing exposure measurement capability

through the integration of the PmodBT2 Bluetooth

module. PmodBT2 module [13] employs the Roving

Network RN-42. It is compatible with Bluetooth

2.1/2.0/1.2/1.0. The module provides wireless

capabilities based on Class 2 low power Bluetooth radio

Streaming of data blocks and proximity sensing

measurement features is provided by the PmodBT2

module. Data transmitted via the PmodBT2 is secured by

a 128-bit hardware-based encryption engine. As

illustrated in figure 2, we have utilized the 12-pin port

with a UART interface to achieve serial communication

between the IoT development board and the PmodBT2

module.

IoT Development Board: The proposed contact tracing

device (see Fig. 2) will be based on the deployment

Freedom-K64 development board. FRDM-K64

development board [14] was selected to develop our

proposed system due to being an ultra-low-cost

development board.

The board is capable of supporting multiple serial

communication ports which enables rapid prototyping

and integration for the development of IoT technologies

[21]. In addition to serial communication, the board

comes integrated with 6-axis digital accelerometer and

magnetometer sensing engine. The storage capability is

another attractive feature for us to consider when

developing the contact tracing system. Storage is

expandable through the integration of a microSD card

slot, which means data blocks generated during contact

exposure could be logged into the device's flash memory.

Meanwhile, connectivity for the FRDM-K64 is

supported using the onboard Ethernet port and headers

for the integration of Bluetooth or a 2.4 G radio module.

In this research, our testbed is based on Kintetis K24-120

MHz with 256KB SRAM Microcontroller based on the

Arm Cortex-M4 Core. Block generations, hash value

computation based on SHA-256, and storing of blocks

information will be handled via the proposed FRDM-

K64 development board. Techniques that enable the

chaining and generation of new data blocks are

developed and deployed on the proposed IoT

development platform.

Dynamic Power Consumption Measurement Platform:

To measure the performance of the proposed system, we

have measured the device’s power consumption during

runtime. Power consumption data were collected under

various configuration parameters of the ZKP protocol

and different sizes of the encryption keys. We have used

a power profiling and power optimization system for

measuring our testbed’s power consumption during

runtime. Power Profiler Kit II (PPK2) [16] is a current

measurement tool for embedded development. The kit

provides us with power debugging capability to estimate

the battery lifetime of the Device Under Test (DUT). The

PPK2 was able to supply and measure the current of the

proposed DUT. It has a programable voltage regulator

with 0.8V to 5V supply voltage and up to 1A current

supply. PPK2 ensures a current range between 200nA to

1A and with a resolution that varies between 100nA and

1mA.

B. Software Development Tools

Mbed OS [15] provides us with a rich development

environment for coding a deploying energy-efficient and

computational-constrained solutions. It enables rapid

prototyping for new technologies deployed on low-ended

devices. Since the wearable contact tracing solution proposed

herein must be capable of logging and streaming hundreds of

data blocks daily, our system must be able to provide energy-

efficient computation and at a lower device cost.

Anonymous authentication based on the ZKP protocol and

chaining and generation of the user’s data block was designed

and developed using embedded C/C++. Algorithms proposed

in this research were developed with the support of the Mbed

OS open-source operating system for the IoT Cortex-M

development board. The application that implements contact

tracing on the proposed testbed was coded using the Arm

Mbed OS APIs. We have utilized the Arm Mbed online

toolset for software development and deployment.

IV. THREAT MODELS

The proposed contact tracing system was incorporated

with multiple defense mechanisms that protect the following

threat models:

Device Cloning Attacks: A user’s contact tracing device

that is captured by a malicious actor is replicated during

these attacks. Contact exposures daily logged including

the device’s ID can be copied to create multiple clones

of the compromised device. Clone devices could be

placed in areas where human traffic flow is high (e.g.,

dining halls, classrooms, meeting rooms, etc.). Using of

cloned devices enables a malicious actor of injecting

faulty data into the user’s contact tracing data logs,

making the system unreliable. Additional to injecting

false data, a cloned device could be optimized by the

adversary to perform replay attacks. We achieved

PmodBT2 (Bluetooth module) pinout [13] FRDM-K64 development board pinout

Fig. 2: Contact Tracing development platform

resiliency against device cloning attacks via the

development of a block chaining approach. Contact

exposure data blocks captured during the day are

validated one by one and inserted into the user’s

blockchain by the contact tracer management system.

Data generated from the last contact exposure is

concatenated with a hash of the previous block. The

latter is hashed by the contact tracer management system

and loaded into the device’s internal memory. Section V

provides a full description of the implemented block

chaining algorithm.

Device Spoofing Attacks: We have considered the case

where an adversary spoofs a device ID to inject false

contact tracing data into other users’ blockchains. To

prevent such attacks, the proposed testbed is integrated

with a mechanism that allows each device to be assigned

to a unique ID. Devices’ software IDs are randomly

chosen from a pool of precomputed IDs. In addition to

randomly assigned device software IDs. Contact Tracing

devices are randomly picked by users to prevent users’

tagging based on hardware ID discovery. Assignments

of random device’s ID provide the adversary with a

moving target with a probability of guessing the correct

device’s ID of 1/n where n represents the pool size. We

assumed " ≫ $, where k indicates the total number of

contact tracer systems deployed in a community.

Data Modification Attacks: Data blocks that captured the

contact exposure between two users are vulnerable to

data modification attacks. An attacker will be able to

modify the data in transit. To protect against data

modification attacks, data blocks were stored in each

device's internal memory in encrypted form. We have

utilized the AES-256 encryption engine to provide data

privacy and resiliency against data modifications.

V. THE PROPOSED SECURITY SCHEMES

The following security schemes were developed and

deployed on the proposed testbed:

A. User Anonymity and Authentication

We achieve users’ anonymity via the implementation of a

lightweight interactive ZKP approach. To prevent users from

identifying other users in the community, the following

cryptographic techniques were developed:

User Anonymity: Users’ devices IDs are dynamically

chosen from a pool of precomputed IDs. To prevent a

malicious actor from identifying a subject based on

discovering his/her device hardware ID or Bluetooth

MAC address. We have considered the case where

contact tracing devices are assigned randomly to users

daily. Before devices assignments, software-based IDs

are computed and preloaded into each user contact tracer

system by the contact tracer management system. To

defend against spoofing attacks, devices IDs are updated

daily.

Anonymous authentication: An authentication protocol

was developed and implemented on the proposed

testbed. The protocol enables two users’ devices to

communicate with each other without revealing their

identification information. To prevent the disclosure of

devices IDs during the contact exposures data process,

IDs were used by the ZKP protocol to compute ZKP

proofs. In the proposed protocol, a device IDj is

composed of concatenating m-secrets,

�%,�||, … , ||�',� .During the ZKP protocol initialization

phase, a pool of n-devices IDs is computed by the contact

tracer management system. In the proposed system, we

have considered " ≫ $ , where k refers to the total

number of contact tracing devices deployed in a

community. Quadratic residues mod N for each valid

device’s IDj is computed by the contact tracer

management system. Computed quadratic residues for

daily assigned devices’ IDs are published to each contact

tracing device. The following steps will take place during

the ZKP protocol initialization phase:

( � ±��*�� )

+,� � ( ∙ . ��,∗01

01

*�� )

(� � +,� ∙ . 2∗,�01

01

*�� )

Binary vector (b0, …, b

m-1)

Fig. 3: ZKP protocol for anonymous authentication and

systems intialization

2

1

3

4

�2%%, … , 2%'�

32�%, … , 2�'4 325%, … , 25'4

326%, … , 26'4

…

32%%, … , 2%'4 32�%, … , 2�'4 325%, … , 25'4

326%, … , 26'4 …

3�%,� , … , �',� 4 3�%,7 , … , �',7 4

…

…

Assigned secrets

for device x

Assigned witnesses

Assigned secrets

for device y

Contact Tracer server

with blockchain

Contact exposure

ZKP

Authentication

Distribution of

device’s IDs

Devices are randomly

assigned by the server

�8 �� 9 ��7 ��: ;<8 ;<9

=� � >?�*;<8)

=7 � >?:*;<9)

@� � �*;<8||�*;<��)

@7 � �*;<9||�*;<�7) ��7

��

Fig. 4: Data exchanges during contact exposure

Data

block

The contact tracer management system randomly

chooses an ID pool of n-precomputed IDs for each

device deployed in the community (see Fig. 3). Each

device IDj is composed of m-secrets.

Multiplicative inverses of modular squares for each

of the m-secrets for a device IDj is computed as 2�,� �±1/��,��*�� ) for i =1, …, m;

For every assigned device IDj, witnesses 2�,% ,

…,2�,' are computed for the m-secrets.

For k number of devices deployed, IDs witnesses’ �2%%, … , 2%'�, �2�%, … , 2�'�, … . , �26% , … , 26'� are

published and stored into each user device.

When two users’ devices are near each other, the two

devices rely on the ZKP protocol for anonymous

authentication. To achieve mutual authentication, the ZKP

protocol will be executed twice. During a single ZKP session,

one device will act as a prover and the second device will act

as a verifier. A prover tries to prove his identity to the verifier

by effectively supporting evidence that it holds the correct set

of m-secrets. A device that was acting as a prover during the

first ZKP session will be serving as a verifier during the

second ZKP session. The following steps are performed

during a single authentication session:

1. A device acting as a prover generates a random value

( � ±��*�� ) and send the value R to the

verifier.

2. The verifier node produces a randomly generated

binary string of m-bits, one bit per secret �∗,�. It sends

the binary string to the prover.

3. A device acting as the prover computes the response

+,� � ( ∙ ∏ �∗,�0D0D *�� ) and transmits the +,�

to the verifier.

4. The verifier solves and verifies that (� � +,� ∙∏ 2∗,�0D0D *�� )

5. The above steps will be iterated E times, where 1≤G ≤ E . Parameter E represents the total number of

ZKP iterations required to authenticate a device.

Since each device’s ID is being randomly chosen from a

pool of precomputed IDs. These daily updates of the devices’

IDs will prevent a malicious actor from utilizing the ZKP

witness data used during the authentication process to learn

the identity of another system in the community.

B. Data Privacy

The proposed system support data privacy through the

deployment of an AES encryption/decryption scheme. AES

encryption along with the device ID is randomly assigned by

the contact tracer management system. Data related to Covid-

19 infection status will be encrypted by each user’s tracing

device using the onboard AES encryption engine. Encrypted

data will be transmitted to other users’ devices during a

contact exposure process. User’s devices won’t be able the

learn the virus infection status of other users in the

community. The proposed system ensured that each user will

only be able to learn whether he/she was exposed to the virus

or not while keeping users’ with Covid-19 infection status

protected. This will prevent a malicious actor in a community

from injecting or modifying the health statuses of other users

in the community.

C. Chaining of Data Blocks

The proposed system was integrated with a lightweight

block chaining scheme. The onboard block chaining scheme

provides full immunity against data modification and data

injection attacks. The scheme also supports resiliency against

device replication and data replays attacks.

Contact Exposure: During the contact exposure process,

two devices, such as x and y securely exchange data blocks

related to the users’ virus infection status. The following steps

were implemented for secure data exchange between two

devices (see Fig. 4):

1. Device x generates a data block ;<8 . Block ;<8 is

composed of the device IDx, Timestamp �8, a hash of

the last block ;<� , and the user’s virus infection status

{�� ∈ | � �1, 0��, where value 1 indicates the

user is Covid-19 positive and value 0 refers to Covid-

19 negative. Similarly, device y with IDy generates a

block ;<9 .

2. Device x computes =� � >?�*;<8) , where =

represents a block ;<8 encrypted with an AES-256

encryption key H� .it’s also computes @� ��*;<8||�*;<��), where @� is a hash of the current

block ;<8 concatenated with a hash value of the last

block. Similarly, device y computes @7, =7.

3. Devices x and y exchange their computed values IDx,

=�, @�, IDy, @7, =7 with each other.

Detection Mechanism for Coronavirus Exposure: In the

proposed system, identification of possible exposure to the

novel Coronavirus was achieved through the implementation

of the block chaining approach. We have considered the

utilization of a contact tracer management server to store

direct person-to-person contact activities of users. To ensure

the integrity of the tracing process, we have adopted a block

chaining methodology. Direct person-to-person events were

captured via the deployment of user-specific blockchains.

Each user’s blockchain was implemented to keep all tracing

events that were collected during the last 14 days. The

chaining of the blocks provides full immunity against data

modification and data injection attacks. Detection for replay

attacks was also incorporated into the proposed system. For

example, when user x made direct contact with other users

(e.g., users y and z) in the community, contact exposure data

blocks are generated and stored in the user’s x contact tracing

device. Hash values of the blocks are computed and verified

using blockchains for users (users y and z) who had

established close contact with the user’s x. Data blocks were

examined to ensure no block duplicates will be inserted into

the user’s blockchain. The following protocol was

implemented into the proposed testbed to protect against

block replication attacks, including data modification and

data injection threats. Data blocks collected during user x

exposure to users y and z are verified with users y and x

blockchains. During the data exchanges process between user

x and the contact tracer management system.

• User x sends =7, @7, =I, and @I to the contact tracer

management system.

• The contact tracer management system recomputed @7

by (i) decrypting =7 to obtain data block ;<7 ��?:*=7) (ii) accessing user y’s blockchain to extract

the last block ;<�7 (iii) Computing @7J ��*;<7||�*;<�7) (iv) comparing if @7J= @7 , then the

new block @7 is added to user x’s blockchain.

Similarly, @I is verified using user z’s blockchain

before adding the new block to user x’s blockchain.

• After adding all contact tracing blocks into user x’s

blockchain, the newly computed last block in the chain

is sent by the contact tracer management system to

user x.

One shortcoming of the above protocol is that it is still

vulnerable to replay attacks. For example, if user y’s contact

tracer device was not deployed in the community during the

last 24-hour. A malicious actor can still utilize a user’s y data

block to inject false data into other users’ blockchains. Since

the last block in user’s y blockchain remain the same from the

last contact exposure event in which user y had made direct

contact with other users in the community. To overcome the

above vulnerability, we have enhanced the resiliency of the

proposed protocol against block replication by injecting a

random dummy block to the user’s blockchain every time a

user’s contact tracing device is not deployed in the

community during the last 24-hours. With the enhanced

protocol, users’ blockchains are updated daily to prevent an

adversary from using clone devices to inject replicated blocks

to other users’ blockchains. Finally, in the case of hardware

/software failure of the contact tracer device, daily recorded

contact tracing data will not be encoded into the user

blockchain, and hence contact exposure information might be

irretrievable. To mitigate this issue, the blockchain server can

conduct daily maintenance operations by using users’

blockchains to identify missing data blocks. Users'

blockchains are compared with each other to detect missing

blocks.

Fig 5. The proposed testbed with the power measurement tool

VI. PERFORMANCE ANALYSIS OF THE PROPOSED SYSTEM

We have analyzed the performance of the proposed

system by measuring the device power consumption under

various configuration parameters, encryption key sizes, and

several ZKP validation attempts, E. Power consumption data

were collected in a lab environment. In our power analysis,

the effect of data retransmission on the overall device’s

power consumption due to communication noise was

considered in this paper. In the proposed testbed (see Fig. 5),

we have utilized the Power Profiler Kit II platform and nRF

connect framework, specifically the power profiler app to

measure the power consumption of the proposed testbed. Fig.

5. Illustrates our system set up to capture the device’s power

consumption under different settings. In this research, two

contact tracing platforms were employed, each device was

loaded with a set of data that is randomly generated during

the device initialization phase. To measure the device’s

power consumption, we run the Power Profiler Kit II in

ampere meter mode, in which it supplies 3.3v regulated

voltage to one of the devices. Since each device is required to

establish mutual authentication with other devices in the

community before the exchange of contact exposure data,

power consumption related to the authentication process is

critical in determining the system's overall performance. We

have analyzed the device’s power consumption when acting

as a prover and verifier during the ZKP authentication

process. Fig. 6a shows the power consumption signals in the

time domain for a device acting as a verifier and prover. Fig.

6b presents the kernel density functions of the device power

consumption for both verifier and prover with E � 5. The device power consumption data were collected during

the contact exposure process. We tested the proposed system

under various AES encryption engines. We have measured

the device’s power consumption during sending and

receiving data. Fig 7a illustrates the kernel density function

estimation for the device’s power consumption during data

reception with AES-128, AES-192, and AES-256 encryption

engines. Fig 7b presents the k-density estimation for the

device power consumption when transmitting data during the

contact exposure process. Power consumption data were

collected under various encryption engines (AES-128, AES-

192, and AES-256). Finally, we have measured the device’s

total power consumption for a single contact exposure with

AES-128 data encryption, including power consumed during

the ZKP-based authentication scheme. Fig. 8 presents the

total power consumption signals for a device sending and

receiving data.

Table I presents the device average power consumption

during contact exposure and ZKP-based authentication.

Table I: Device peak and average power consumption during contact

exposure and authentication with various AES implementation

Schemes

Device power consumption

Average power

consumption

(W)

ZKP Authentication

ZKP verifier 0.127914

ZKP Prover 0.110483

Data exchange

during the contact

exposure

process

Sending data with AES-128 0.105976



Receiving data with AES-128 0.166063



PPK2

Contact tracing device

Power

measurement tool

Fig. 7: Device’s power consumption during sending and receiving under various AES encryption engines (AES-128, AES-192, AES-256)

Fig.6: Device’s power consumption during ZKP-based authentication

a b

a b

Fig. 8: Device’s total power consumption during contact exposure and ZKP-based authentication with AES-128

ZKP prover Device receiving data

Device sending data ZKP verifier

VII. CONCLUSION AND FUTURE WORKS

This research proposed the development of a lightweight

contact tracing solution that monitored a person's daily

exposure to the coronavirus in the workplace. The

proposed testbed is capable of offering data privacy, user

anonymity, data integrity, and undependability from

smartphone solutions. Data privacy was implemented into

the proposed system via the deployment of various AES-

based encryption engines. Data accessibility was only

supported via the contact tracer management server,

restricting users of the system from learning other users’

private data. User anonymity and anonymous

authentication were integrated into the proposed testbed by

incorporating an anonymous authentication protocol based

on ZKP. To ensure that ZKP authentication parameters

cannot be utilized by a malicious actor to discover users’

identities, ZKP authentication parameters were randomly

assigned by the contact tracer management system daily.

Immunity against data injection and data modification

attacks were also supported by the proposed testbed. We

have implemented a block chaining algorithm, where

contact exposure data are encoded into the blockchain.

Data blocks that capture the person's exposure to the virus

are stored in a blockchain. Data blocks generated during

the contact exposure process are verified by the user’s

blockchain before they get added. The latter ensures

resiliency against data injection and data replay attacks.

Finally, we have analyzed the performance of the proposed

testbed by measuring the device's dynamic power

consumption while executing ZKP authentication schemes

and contact exposure protocols. Power measurement data

were collected under various AES encryption/decryption

engines (AES-128, AES-192, AES256). Also, we have

measured the device power dissipation during ZKP

authentication when a device acts as a verifier and as a

prover. In this research, we have utilized proximity sensing

based on Bluetooth capability to detect direct contact.

Future work will address the computational complexity of

the proposed blockchain algorithm. We will further

investigate the performance of the system under a various

number of users. on the integration of a new proximity

sensing solution that is based on Ultra-Wideband (UWB)

technology to provide highly accurate distance

measurement when determining possible exposure to the

virus

VIII. ACKNOWLEDGEMENT

This work was supported by the FAST 2021 summer grant,

Sam Houston State University.

IX. REFERENCES [1] Apple, and Google. "Exposure Notifications: Using Technology to

Help Public Health Authorities Fight COVID-19." (2021).

[2] E. Lee, K. Park, D. J. Park, J. Kim and C. Jo, "Locally Testable

Privacy-Preserving Contact Tracing Protocol Without Exposing Secret Seed," 2021 IEEE International Conference on Consumer

Electronics (ICCE), 2021, pp. 1-5, doi:

10.1109/ICCE50685.2021.9427587.

[3] N. Ahmed et al., "A Survey of COVID-19 Contact Tracing Apps," in IEEE Access, vol. 8, pp. 134577-134601, 2020, doi:

10.1109/ACCESS.2020.3010226.

[4] D. J. Leith and S. Farrell, "Coronavirus contact tracing: Evaluating

the potential of using Bluetooth received signal strength for

proximity detection", May 2020, [online] Available:

https://www.scss.tcd.ie/Doug.Leith/pubs/bluetoothrssistudy.pdf.

[5] J. Berglund, "Tracking COVID-19: There’s an App for That," in IEEE Pulse, vol. 11, no. 4, pp. 14-17, July-Aug. 2020, doi:

10.1109/MPULS.2020.3008356.

[6] P. H. O’Neill, T. Ryan-Mosley and B. Johnson, "A flood of

coronavirus apps are tracking us. Now it’s time to keep track of them", MIT Technol Rev, May 2020, [online] Available:

https://www.technologyreview.com/2020/05/07/1000961/launchin

g-mittr-covid-tracing-tracker/.

[7] A. Hekmati, G. Ramachandran and B. Krishnamachari, "CONTAIN Privacy-oriented Contact Tracing Protocols for Epidemics," 2021

IFIP/IEEE International Symposium on Integrated Network

Management (IM), 2021, pp. 872-877.

[8] A. Fiat and A. Shamir, "How To Prove Yourself: Practical Solutions To Identification and Signature Problems", Proceeding of CRYPTO

1986.

[9] A. A. Rasheed, R. N. Mahapatra, C. Varol and N. Karpoor, "Exploiting Zero Knowledge Proof and Blockchain Towards the

Enforcement of Anonymity, Data Integrity and Privacy (ADIP) on

IoT," in IEEE Transactions on Emerging Topics in Computing, doi: 10.1109/TETC.2021.3099701.

[10] A. Rasheed, R. R. Hashemi, A. Bagabas, J. Young, C. Badri and K.

Patel, "Configurable Anonymous Authentication Schemes For The Internet of Things (IoT)," 2019 IEEE International Conference on

RFID (RFID), 2019, pp. 1-8, doi: 10.1109/RFID.2019.8719256.

[11] H. T. T. Truong, M. Almeida, G. Karame and C. Soriente, "Towards

Secure and Decentralized Sharing of IoT Data," 2019 IEEE International Conference on Blockchain (Blockchain), 2019, pp.

176-183, doi: 10.1109/Blockchain.2019.00031.

[12] Christidis, Konstantinos, and Michael Devetsikiotis. "Blockchains

and smart contracts for the internet of things." Ieee Access 4 (2016): 2292-2303.

[13] Pmod BT2 Available: https://digilent.com/shop/pmod-bt2-

bluetooth-interface/

[14] FRDM-K64F Platform|Freedom Development Board|Kinetis MCUs

| NXP Semiconductors. Available: https://www.nxp.com/design/development-boards/freedom-

development-boards/mcu-boards/freedom-development-platform-

for-kinetis-k64-k63-and-k24-mcus:FRDM-K64F

[15] Mbed Operating System. Available: https://os.mbed.com

[16] Power Profiler Kit II. Available: https://www.nordicsemi.com/Products/Development-

hardware/Power-Profiler-Kit-2

[17] J. Song, J. Nang and J. Jang, "Design of Anomaly Detection and

Visualization Tool for IoT Blockchain," 2018 International Conference on Computational Science and Computational

Intelligence (CSCI), 2018, pp. 1464-1465, doi:

10.1109/CSCI46756.2018.00292.

[18] V. A. Gasimov and S. K. Aliyeva, "Using blockchain technology to ensure security in the cloud and IoT environment," 2021 3rd

International Congress on Human-Computer Interaction,

Optimization and Robotic Applications (HORA), 2021, pp. 1-5, doi: 10.1109/HORA52670.2021.9461397.

[19] S. Malik, V. Dedeoglu, S. S. Kanhere and R. Jurdak, "TrustChain:

Trust Management in Blockchain and IoT Supported Supply

Chains," 2019 IEEE International Conference on Blockchain (Blockchain), 2019, pp. 184-193, doi:

10.1109/Blockchain.2019.00032.

[20] A. K. Singh and N. Kushwaha, "Software and Hardware Security of

IoT," 2021 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS), 2021, pp. 1-5, doi:

10.1109/IEMTRONICS52119.2021.9422651.

[21] D. Balsamo, A. Elboreini, B. M. Al-Hashimi and G. V. Merrett,

"Exploring ARM mbed support for transient computing in energy harvesting IoT systems," 2017 7th IEEE International Workshop on

Advances in Sensors and Interfaces (IWASI), 2017, pp. 115-120,

doi: 10.1109/IWASI.2017.7974230.

Appendix

Technical Program Committee, and other staff members

Chair

Dr. Ken Sakamura, Professor and Dean of INIAD, Toyo University, Chair of TRON Forum

Vice Chairs:

Dr. Tomohiro Hase,

Professor, Ryukoku University, Japan,

Board Member, IEEE Consumer Electronics Society

Members of TPC (Alphabetical order of the family name):

• Muhamed Fauzi Bin Abbas Singapore Institute of Technology

• Heikki Ailisto VTT Technical Research Centre of Finland

• Yasuhito Asano INIAD, Toyo University

• Alessandro Bassi Alessandro Bassi Consulting

• Alex Galis University College London

• Masahiro Bessho INIAD, Toyo University

• Mu-Yen Chen National Taichung University of Science and Technology

• Kwek Chin Wee Republic Polytechnic

• Stephan Haller The Bern University of Applied Sciences

• Chiaki Ishikawa YRP Ubiquitous Networking Laboratory

• Tomonari Kamba INIAD, Toyo University

• Shindo Katsunori YRP Ubiquitous Networking Laboratory

• M. Fahim Ferdous Khan INIAD, Toyo University

• Ryoichi Kawahara INIAD, Toyo University

• Jee-Eun Kim INIAD, Toyo University

• Tatu Koljonen VTT Technical Research Centre of Finland

• Akira Matsui Personal Media Corporation

• Akihiro Nakao The University of Tokyo

• Takako NONAKA Shonan Institute of Technology

• Alok Prakash Nanyang Technological University

• George Roussos Birkbeck College, University of London

• Kentaro Shimizu The University of Tokyo

• Toru Shimizu INIAD, Toyo University

• Amiruddin Bin Jaafar Sidek Custommedia

• Mohit Sindhwani Quantum Inventions

• Juha-Pekka Soininen VTT Technical Research Centre of Finland

• Thambipillai Srikanthan Nanyang Technological University

• Kazuya Sumikoshi YRP Ubiquitous Networking Laboratory

• Junpei Tsuji INIAD, Toyo University

• Takeshi Yashiro INIAD, Toyo University

• Kenji Yoshigoe Embry-Riddle Aeronautical University

• Takeshi Ogasawara INIAD, Toyo University

• Ge Hangli The University of Tokyo

Publication co-chairs:

• Chiaki Ishikawa • Masahiro Bessho • Takeshi Yashiro • Toshihisa Muto

Publicity Chair:

• Chiaki Ishikawa

The total program of 2021 TRON Symposium -TRONSHOW- Please see the following URLs: https://www.tronshow.org/index-e.html https://www.tron.org/tronshow/2021/regist/schedule/?lang=en (technical sessions)

https://www.tronshow.org/index-e.html

https://www.tron.org/tronshow/2021/regist/schedule/?lang=en

www.tronshow.org

Proceedings of 2021 TRON Symposium -TRONSHOW- ISBN 978-4-89362-375-1

Documents

Technical Paper Session - tronshow.org