[Personal Project] ECG(EKG) authentication (biometrics) system

[Project Name] : ECG Biometrics System with Mouse Device Type Electrodes

[Project Period] : DEC 2015 - JUL 2016

[Author] : Junhyun Lee, Prof. Sanghoon Lee$^{\dagger}$
Department of Biomedical engineering, Korea University, Seoul, Republic of Korea

This personal project was conducted in 2016 when I was undergraduate student researcher at Intelligent Bio-MEMS Laboratory(iBML, Korea Univ., Advisor: prof. Sanghoon Lee). At this lab, I participated in research about the fabrication of elastic electrode material, CNT/PDMS (Composite of carbon nanotube and poly-dimethylsiloxane). I’ve wanted to apply this material at the application of medical instrumentation. However, unfortunately because of time limitation, the integration of CNT/PDMS electrodes and biometrics system was not done in this work.

The purpose of primary stage was just building whole process for ECG(ElectroCardioGram, or EKG) biometrics system. So, There are lots of rooms for improvement in each part.


These days, electrocardiogram (ECG) is used for not only diagnosis of the heart’s condition, but also authorization of individual access. But ECG usually can be measured through some specific machine. Because ECG measurement requires a few electrodes, amplifiers, filters and a data acquisition module. Especially, the electrodes should be connected to the person’s body and the wire from electrode also should be connected at the data acquisition machine. Therefore, the movement of a person is restricted. So the biometrics based on ECG were difficult to be used in daily life. In this work, user-friendly device for ECG measurement is proposed. This device is computer peripherals with electrodes for measurement of ECG. With Bluetooth communication and this device, ECG biometrics can be applied in daily life. For the classification, we use correlation coefficient and neural network.


Biometrics became a remarkable method of certification system these days[1-5]. In this project, we propose a convenient device for biometrics. There are many personal features used in biometrics (fingerprint, iris, retina, vein, etc.)[2]. Authorization system based on ECG has several advantages over the others. One of the biggest problems with fingerprint recognition technology is the theft of fingerprints. Another problem is that the remaining fingerprints of fingerprint scanners can be copied from the glass. And It is not applicable if the fingerprint is damaged or missing at all. Retinal scanners are inefficient for people like blindness and cataract patients. Retinal scanners are also very annoying to users because they require close contact with the user’s eyes on the authentication terminal. While costly machine is required to measure most of them, electrocardiogram (ECG) can be measured with just a few chips and electrodes.
But during the measurement of ECG, user’s motion is restricted by the wire and electrodes. Because leadⅠ, Ⅱ, Ⅲ are used at usual ECG measurement, four electrodes (left arm, left leg, right arm, right leg) is required at least. So the ECG signal was cumbersome as a feature for biometrics[6-9]. In this work, just lead Ⅰis used for biometrics[10-13]. The number of electrodes and wires is reduced. Also, type of electrodes is comfortable for user. We propose mouse device type electrode for ECG measurement. In modern days, people are familiar with computer and peripherals. With proposed electrodes, people feel less discomfort during the measurement. Keyboard and mouse are convenient device for computer users. Other author designed keyboard type electrodes without mouse for ECG measurement at fingertip, but mouse type electrode can get more stable signals. Because mouse case can support user’s hand [11].
ECG signal is measured through lead I electrodes and filtered by analog filter. Then, it is transmitted using Bluetooth communication and filtered again by digital filter. To reduce base noise, pick the 15 PQRST-waves and get the mean wave. And for the classification, compare measured signal with database by calculating correlation coefficient. Finally, user’s name will be founded from database with highest correlation coefficient. Later on, we categorize them using a simple neural network.


ECGbiometrics-01 Figure 1. block diagram of our ECG biometrics process

Our process is depicted in Figure 1 as block diagram. This is composed of hardware part and software part. The left fingertip is signal source, right thing is reference source, and the right palm is ground. At these points, ECG signal wave lead Ⅰ can be measured. Table 1 shows ECG acquisition condition.

Sampling rate
131 Hz
Analog filtering
0.72 Hz HPF, 59.54 Hz LPF
Digital filtering
1.5 Hz HPF, 58 Hz LPF
Electrical potential
Lead Ⅰ
State of ECG recording
Normal heart rate (60~80/min)

Table 1. ECG acquisition condition.


Hardware part is composed of electrodes, mouse body, wire, circuits, and data acquisition module (DAQ) with Bluetooth communication chip.

  • Electrodes : As shown in Figure 2 (left and center), there are the yellow electrode for left hand’s finger and gray electrodes for right hand’s thumb and palm. The electrodes are fabricated by nickel tape with copper wire. The electrodes can be replaced by CNT/PDMS material at the next stage project. ECGbiometrics-02 Figure 2. Conceptual schematic of electrodes(left and center) and actual experiment setting(right)

  • Circuit : We used the buffer, the differential amplifier, the secondary amplifier and the some filters. Figure 2 are just proposal design, which contains integrated circuit in the mouse body for signal processing. But the circuit for experiment is constructed on the external breadboard. The circuit can be replaced by integrated circuit such a ADS 1292.
    ECGbiometrics-03 Figure 3. Buffer (1) and differential amplifier (2), image from wikipedia
    At the Figure 6, blue box of 1 indicated buffer. This used for high impedance. Because of high impedance of the buffer, we can get the signal bodily. And at the red box 2-1, the gain of differential amplifier is 1. That is because of ratio of $15 k\Omega : 15 k\Omega$. But at the 2-2, there are resistors as the $R_1$ and the $R_{gain}$. We can get the gain $1+ \frac{200k}{4.7k} = 43.55$. Further, the blue box 1 + red box 2 is called implementation circuit which can be replaced by integrated circuit such a INA 118. ECGbiometrics-04 Figure 4. High pass filter, image from wikipedia
    Blue box 3 in Figure 6, there are high pass filter that has cutoff frequency $\frac{1}{2 \times \pi \times R \times C} = \frac{1}{2 \times \pi \times 10^6 \times 10^{-6}} = \frac{1}{2 \times \pi}$ about 0.16Hz. The cutoff bandwidth is very narrow but it has significant roles. At first, when user moves so noise is generated, it is saturated from the range of our monitor. If there are no high pass filter, saturated signal doesn’t recovery to stable state. So high pass filter get rid of DC component, and its time constant make saturated signal to be stable state. ECGbiometrics-05 Figure 5. Active low pass filter, image from wikipedia
    At the red box 4 in Figure 6, the active low pass filter and secondary amplifier allow us to get the amplified signal. Its gain is $1 + \frac{300}{15} = 21$.
    Last, at the black box 6, we used driven right leg (DRL) for get rid of common voltage using negative feedback. So we can get higher Common-Mode Rejection Ratio(CMRR) and get rid of 60Hz noise using DRL circuit.

ECGbiometrics-06 Figure 6. Schematic of our signal filter circuit

  • Data acquisition module : Arduino UNO is used as the DAQ. DAQ contains analog-to-digital converters(ADC) part and bluetooth module (HC-06, China). The DAQ can be replaced by integrated daq chip such a ADuCM360.

There are all-in-one solution chip for hardware part(except electrodes), such a S3FBP5A samsung bio processor.


The data acquisition and the entire digital process are controlled by MATLAB graphic user interface (GUI) in Figure 7. First, insert the name of bluetooth module and push connect button. After the module is connected, put the subject’s hand on the electrodes. Subjects received the notice about the purpose of this experiment and study. When the scan button is clicked, ECG measurement is started. Finished the scanning during the duration, two lines show up in the axes like Figure 7 (b). The upper line is raw signal just filtered by analog filter. Another line is filtered signal by digital filter. Its peaks are detected by MATLAB function ‘findpeaks’. Set the region of interest (ROI). Then the scale of axes is modified depending on ROI setting. Select the fifteen R-peaks manually. This part can be replaced by deep learning technique of object detection or segmentation. Choice the final behavior (enrollment or classification), then the fifteen ECG pulse is summed and processed depending on selected behavior. For the classification, measured ECG data is compared with database signal by calculation correlation coefficient.

ECGbiometrics-07 Figure 7. MATLAB GUI for biometrics

But there are limitations at the correlation method. If we have a lots of subjects, error rate should be high. Because correlation value can not reflect complex information or patterns of ECG signals. So we need more accurate method for classification, like an artificial neuronal networks. So we used the simple neuronal network, multilayer perceptron that is depicted at the Figure 8. The multilayer perceptron and the input signals can be replaced by convolutional neural network and others signal format such a power spectrum, respectively.

ECGbiometrics-08 Figure 8. Structure of multilayer perceptron

For the input, the signals were chopped by a heartbeat. It is depicted at Figure 9. In this way, we got 125 signals from 7 subjects. 100 signals were used as training set, 25 signals were used as test set.

ECGbiometrics-09 Figure 9. ECG signals for multilayer perceptron input


ECG signal is passed through the analog filter. Its pass frequency band is 0.72 Hz ~ 59.54 Hz. We can obtain the primarily filtered signal using bluetooth module. At the software part, primary signal shows up as upper signal at Figure 7 (b). It should be passed digital filter because of signal baseline. The signal is secondarily filtered at 1.5 Hz ~ 58 Hz frequency band. So we can set the baseline. It shows up as below signal at Figure 7 (b). Also, the peaks is detected by MATLAB function ‘findpeaks’. The standard setting for peaks detection is ‘MinPeakProminence’. Because each person has different peak value and wave shape, the setting value can’t apply uniformly. So we should pick the peaks manually at the software GUI. Then, for the rejection the base noise, get the average signal of picked signal.
Figure 10 is the result of comparison between database signal using our electrodes and measured signal using commercial Ag/AgCl electrodes (Covidien, Canada). It is for validation of our electrodes. Correlation coefficient is 0.9565.

ECGbiometrics-10 Figure 10. ECG signal measured on the Ni tape and the Ag/AgCl electrodes

As shown at Figure 7, There are two radio buttons for final behaviors at this GUI, enrollment and classification. If user’s ECG signal is not contained at the database, user should select enrollment. Then, measured ECG signal is enrolled at the database with user’s name. For the classification, user should select classification radio button. The final state of GUI for classification is depicted in the Figure 7 (d). The red line is measured signal, and the black line is a few days old database signal. The result of classification is represented in the red box at Figure 7 (d). It is correct result.
As mentioned above, the experiment using neural network for classification is also conducted. The result is shown at Figure 11. There are totally 125 signals from 7 subjects. 100 signals were used as training set, 25 signals were for test set. We got 100% accuracy from both training set and test set. Because of the size of the dataset, the result does not mean that our model is perfect. It is just shown feasibility.

ECGbiometrics-11 Figure 11. The result of classification on multilayer perceptron.

Also, we suggest dry and flexible type electrode in this study. So we fabricated electrode by using carbon nanotube(CNT) with polydimethylsiloxane(PDMS). Because of its physical property, it can be attached to mouse. To reduce contact resistance, copper wire is inserted to CNT/PDMS. PDMS makes the electrode more durable. CNT/PDMS is depicted at Figure 12.

ECGbiometrics-12 Figure 12. CNT/PDMS electrodes.


In this study, we build the prototype of biometrics system and propose the mouse type electrodes for biometrics based on ECG. Because people are accustomed to computer and peripherals, it will be convenient device for biometrics users. In other author’s paper, keyboard type electrodes were designed without mouse for ECG measurement at fingertip, but mouse type electrode can get more stable signals. Because mouse case can support user’s hand. Also, data is transmitted based on bluetooth communication.
As we can see in Figure 9, ECG signals of subjects are discriminated from each other. So users who is enrolled at the database can be identified by comparing correlation coefficient. In this work, dry type electrode is used. The wet type electrode is improper to use at daily device like the keyboard and the mouse. So our proposed electrodes is fabricated by nickel tape. At the result of comparison between proposed electrodes and commercial electrodes, in Figure 10, there are insignificant difference. Correlation coefficient is 0.9565. It is enough to accept as a reliable signal for biometrics. Based on this data, classification by proposed electrodes acquire the reliability. So it can be used as real-life application of ECG biometrics.
However, there are some challenging problem in this paper. At first, comparison of correlation coefficient is not perfect evaluation method. For rejection of noise, we apply the filter and summation to measured signal. So the signal is modified and detail features is get rid of with noise. To avoid this phenomena, wavelet transform is recommended [17-19]. If we use the wavelet transform, extraction the detail feature and the automatic classification would be possible. For the classification with detail features, support vector machine classifier is usually used [11, 20-22]. we used simple neural network, multilayer perceptron.
Second challenge problem is electrode oxidation. Because of limitation of wet type electrodes in daily life, the electrodes are fabricated with Ni tape in this paper [23]. But the metal tends to be oxide. Then it can trigger the error at classification thereby generating noise. If we use the polymer type electrodes, for example the carbon nanotube with polydimethylsiloxane (CNT/PDMS), this problem would be prevented [24].


  1. Silva H, Lourenço A, Canento F, Fred AL, Raposo N, editors. ECG Biometrics: Principles and Applications. BIOSIGNALS; 2013.
  2. Jain AK, Ross A, Prabhakar S. An introduction to biometric recognition. Circuits and Systems for Video Technology, IEEE Transactions on. 2004;14(1):4-20.
  3. Jain A, Bolle R, Pankanti S. Biometrics: personal identification in networked society: Springer Science & Business Media; 2006.
  4. Jain AK, Ross A, Pankanti S. Biometrics: a tool for information security. Information Forensics and Security, IEEE Transactions on. 2006;1(2):125-43.
  5. Wayman J, Jain A, Maltoni D, Maio D. An introduction to biometric authentication systems: Springer; 2005.
  6. Han Y-H. A Study on Monitoring of Bio-Signal for u-Health System. Journal of the Korea society of computer and information. 2011;16(3):9-15.
  7. Chung W-Y, Lee Y-D, Jung S-J, editors. A wireless sensor network compatible wearable u-healthcare monitoring system using integrated ECG, accelerometer and SpO 2. Engineering in Medicine and Biology Society, 2008 EMBS 2008 30th Annual International Conference of the IEEE; 2008: IEEE.
  8. Muhlsteff J, Such O, Schmidt R, Perkuhn M, Reiter H, Lauter J, et al., editors. Wearable approach for continuous ECG-and activity patient-monitoring. Engineering in Medicine and Biology Society, 2004 IEMBS’04 26th Annual International Conference of the IEEE; 2004: IEEE.
  9. Led S, Fernández J, Serrano L, editors. Design of a wearable device for ECG continuous monitoring using wireless technology. Engineering in Medicine and Biology Society, 2004 IEMBS’04 26th Annual International Conference of the IEEE; 2004: IEEE.
  10. Silva H, Gamboa H, Fred A. One lead ECG based personal identification with feature subspace ensembles. Machine Learning and Data Mining in Pattern Recognition: Springer; 2007. p. 770-83.
  11. Da Silva HP, Fred A, Lourenco A, Jain AK, editors. Finger ECG signal for user authentication: usability and performance. Biometrics: Theory, Applications and Systems (BTAS), 2013 IEEE Sixth International Conference on; 2013: IEEE.
  12. Shen T-W, Tompkins W, Hu Y, editors. One-lead ECG for identity verification. Engineering in Medicine and Biology, 2002 24th Annual Conference and the Annual Fall Meeting of the Biomedical Engineering Society EMBS/BMES Conference, 2002 Proceedings of the Second Joint; 2002: IEEE.
  13. Shen T-W, Tompkins WJ, Hu YH. Implementation of a one-lead ECG human identification system on a normal population. Journal of Engineering and Computer Innovations. 2011;2(1):12-21.
  14. Aihua F, Chunhua B, Xinbao N, Aijun H, Jianjun Z, editors. Portable electrocardiogram monitor based on ARM. 2008 International Conference on Information Technology and Applications in Biomedicine; 2008.
  15. Ng KA, Chan PK. A CMOS analog front-end IC for portable EEG/ECG monitoring applications. Circuits and Systems I: Regular Papers, IEEE Transactions on. 2005;52(11):2335-47.
  16. Agarwal R, Sonkusale S, editors. Direct Analog-to-QRS detection front-end architecture for wearable ECG applications. Engineering in Medicine and Biology Society (EMBC), 2010 Annual International Conference of the IEEE; 2010: IEEE.
  17. Li C, Zheng C, Tai C. Detection of ECG characteristic points using wavelet transforms. Biomedical Engineering, IEEE Transactions on. 1995;42(1):21-8.
  18. Sahambi J, Tandon S, Bhatt R. Using wavelet transforms for ECG characterization. An on-line digital signal processing system. Engineering in Medicine and Biology Magazine, IEEE. 1997;16(1):77-83.
  19. Bahoura M, Hassani M, Hubin M. DSP implementation of wavelet transform for real time ECG wave forms detection and heart rate analysis. Computer methods and programs in biomedicine. 1997;52(1):35-44.
  20. Odinaka I, Lai P-H, Kaplan AD, O’Sullivan JA, Sirevaag EJ, Rohrbaugh JW. ECG biometric recognition: A comparative analysis. Information Forensics and Security, IEEE Transactions on. 2012;7(6):1812-24.
  21. Raj PS. ECG Biometrics using Intuitive Bases and Support Vector Machines: University of Toronto; 2014.
  22. Lourenço A, Silva H, Fred A, editors. ECG-based biometrics: A real time classification approach. Machine Learning for Signal Processing (MLSP), 2012 IEEE International Workshop on; 2012: IEEE.
  23. Gruetzmann A, Hansen S, Müller J. Novel dry electrodes for ECG monitoring. Physiological measurement. 2007;28(11):1375.
  24. Jung H-C, Moon J-H, Baek D-H, Lee J-H, Choi Y-Y, Hong J-S, et al. CNT/PDMS composite flexible dry electrodesfor long-term ECG monitoring. Biomedical Engineering, IEEE Transactions on. 2012;59(5):1472-9.