THE AVERAGE age of people living in industrialized countries. Design and Implementation of a Portable Long-Term Physiological Signal Recorder

Transcription

1 718 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 14, NO. 3, MAY 2010 Design and Implementation of a Portable Long-Term Physiological Signal Recorder Timo Vuorela, Ville-Pekka Seppä, Jukka Vanhala, and Jari Hyttinen Abstract This paper describes a design and implementation of a portable physiological signal recorder. The device is designed for measuring electrocardiography, bioimpedance, and user s activity. The bioimpedance measures the dynamic changes in the impedance, and its main application is monitoring user s respiration. Activity is measured with three-axis acceleration sensor. During the design, a special attention is paid on the device s power consumption and the target has been set to a 24-h operating time. Functionality of the implemented measurement device is proven with test measurements, which include, e.g., comparison of measurement signals against reference signals, testing the device operation under vigorous upper body movements, and during a light exercise. In order to verify the device operation during real-life activities, one full day, 24-h long, measurement is carried out. The measurement system is tested with both commercial Ag/AgCl gel-paste electrodes and custom-made textile electrodes. Device is proven to be operational with both electrodes, but textile electrodes are found to be more sensitive for movement artifacts. This paper also gives a small review of other existing portable and wearable physiological measurement devices and discusses some general requirements of these devices. Index Terms Bioimpedance, electrocardiography, long-term portable, physiological measurement. I. INTRODUCTION THE AVERAGE age of people living in industrialized countries is continuously increasing. It has been predicted that in Europe the percentage of people older than 65 years is going to almost double from 11% in year 1960 to 21% by the year The same trend can be seen in other countries as well. For example, in Japan the number of people older than 65 years was about 12% of population in the year 1960 and is going reach 28% in the year With growing age the prevalence of chronic diseases generally increases [1]. This together with growing living standards and increase in amount of life stylerelated diseases sets high requirements for the future healthcare and medical services. Manuscript received June 29, 2009; revised November 23, 2009 and January 22, First published February 17, 2010; current version published June 3, This work was supported by the Finnish Funding Agency for Technology and Innovation, TEKES. T. Vuorela and J. Vanhala are with the Department of Electronics, Tampere University of Technology, Tampere, Finland ( timo.vuorela@ tut.fi; jukka.vanhala@tut.fi). V.-P. Seppä and J. Hyttinen are with the Department of Biomedical Engineering, Tampere University of Technology, Tampere, Finland ( ville-pekka.seppa@tut.fi; jari.hyttinen@tut.fi). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TITB Even though many elderly people are capable for independent living, the amount of people needing daily assistance or surveillance is going to increase in the near future. Also the rate of hospitalization and nursing visits are going to increase. One solution to these problems is to offer remote monitoring and surveillance services. These services enable monitoring in peoples natural environment, e.g., in home. It has been shown [2] that people who are equipped with remote monitoring devices are less keen to utilize nursing services. A key component in telemonitoring and remote health care systems is a physiological measurement device. The operation principle of the device can be twofold. The device can either measure physiological signals and analyze them in real-time or only record the signals and send them to a medical center for further analysis. In the implementation of the analysis algorithms also plain data recording devices are needed. Commercial devices for recording physiological signals in nonclinical environment do already exist. One example is the wrist worn heart rate monitors [3], [4] or wrist worn activity monitors [5]. An example of a sport computer is the FRWD [6], which measures heart rate and location related signals such as distance, speed, and altitude. A SenseWear armband from Bodymedia is a measurement device worn near a biceps [7]. Device measures skin temperature, galvanic skin response (GSR), threeaxis accelerations, and heat flux from body. A LifeShirt from Vivometrics [8] is a measurement system integrated into an undershirt. LifeShirt measures, e.g., electrocardiograph (ECG), and respiration, and tracks posture and physical activity of the user. Another garment like approach is the adistar Fusion products [9]. AdiStar products are developed in cooperation of Adidas and Polar. Products contain, e.g., a running shoe, which has a place for Polar S3 Stride sensor and a running shirt, which has integrated textile electrodes for heart rate measurement. Many research groups have developed wearable and portable devices for measuring physiological signals. AMON is a wrist worn device, which monitors blood oxygen saturation and pulse (SpO 2 ), electrocardiography, blood pressure (BP), acceleration, and temperature [10]. Clothing-like measurement device is the VTAMN garment. The undergarment measures ECG, respiration frequency, temperature and acceleration. The system also includes a global system for mobile communications (GSM) and global positioning system (GPS) modules for location awareness and wireless connection [11]. The Wealthy measurement system consists of a garment integrated textile sensors and a portable patient unit (PPT), which records, processes, and sends forward the measured signals. Wealthy system is capable of monitoring five lead ECG, respiration with both strain sensors and impedance pneumography (IP), and activity based on /$ IEEE

2 VUORELA et al.: DESIGN AND IMPLEMENTATION OF A PORTABLE LONG-TERM PHYSIOLOGICAL SIGNAL RECORDER 719 movements measured with strain sensors [12]. Healthwear project is a continuum of the Wealthy project [13]. The measurement system developed in the Healthwear is capable to measure same signals than the Wealthy system and in addition skin temperature, SpO 2, and three-axis accelerations. Other examples of gadget type measurement system are VMote [14], wrist-worn device [15], HealthGear [16], and Smart Headband [17]. Rest of this paper is organized as follows: Section II lists some general requirements for portable physiological measurement devices. Sections III and IV present the implemented measurement device and performed test measurements, respectively. Finally, Section V draws some conclusions and presents future work. Fig. 1. Block diagram presenting implementation of bioimpedance, electrocardiography, and acceleration measurements. II. REQUIREMENTS OF PORTABLE PHYSIOLOGICAL MEASUREMENT DEVICES A. Skin Contact In physiological measurement devices, the contact to the skin is an essential requirement. In order to measure signals related to or generated in a body at least electrodes must have a skin contact. Due to this a natural location for a measurement device is as close to the body as possible. A convenient place for a device is clothing, which usually covers almost 90% of the skin area and, therefore, provides an easy way to guarantee the skin contact. Another important aspect is the electrodes utilized in the measurement. In order to get high-quality measurement signals, electrodes should be stable and stay in place firmly, which is difficult to guarantee especially during movements. In clinical measurements, gel-paste electrodes are normally utilized. In these electrodes, gel is used to improve the contact between the skin and electrode. A problem is that the gel tends to dry over the time, which poses a problem in long-term measurement. Another possibility is textile electrodes, manufactured from conductive fibers. Textile electrodes usually contain no gel, so drying is not a problem with them. Furthermore, the contact area of the textile electrodes is usually larger than in gel-paste electrodes, which improves the skin contact especially during a motion. Because textile electrodes have no glue for attaching them onto the skin they do not irritate the skin as much as gel electrodes. However, a challenge in using the textile electrodes is to guarantee proper pressure between the electrode and the skin so that the impedance in this interface stays stable and low enough. This can be done, e.g., with an elastic band and Velcro [18]. B. User-Related Requirements Portable measurement devices should be reliable, robust, durable, and easy to use [19]. These requirements yield from facts that users are rarely specialists and, therefore, not aware of the technical limitations of the device. Furthermore, a device must be designed to produce accurate measurements and not to provide false signals in any situation. Accuracy is especially important if a device is utilized in clinical measurements. Portable measurement devices are often targeted for longterm measurements in nonclinical environments, and therefore, they should be comfortable to wear and carry. This means that either the devices must look like an essential part of user s outfit or that they must disappear into the clothing [19]. An essential part of wearing comfort is that the measurement device is small in both weight and size. Device should also require as little maintenance as possible, e.g., power consumption should be small enough to enable operating time of at least one working day so that batteries can be charged overnight. An important aspect to pay attention is biocompatibility. Biocompatibility is especially important in implantable devices, but also a device located on the skin should cause as little irritation as possible. Many measurement devices have a communication channel through which they communicate with a data terminal device, e.g., a computer or a cell phone. This communication link should be designed to be as automatic as possible so that it does not pose burden to a user [19]. Also the security of the data transfer must be taken into account. Security is important if data is transferred through a radio link, which can be eavesdropped. Important aspects in the radio communication are energy efficiency and safety of the radio frequency (RF)-radiation. Communication link must also be able to operate even if other devices are present on the same frequency band. III. IMPLEMENTED MEASUREMENT DEVICE For recording the physiological signals, a portable measurement device is implemented. According to the best of our knowledge, this is at the moment the only portable device that can measure and store continuously bioimpedance, ECG, and acceleration data with high sample rates for 24 h. The implemented measurement device can be divided into two main parts, a measurement part and a control and communication part. Measurement part can be further divided to three blocks, each responsible for measuring one physiological signal, i.e., bioimpedance, electrocardiography, and acceleration. Block diagram of the measurement part is presented in Fig. 1. The measurement device is encapsulated into a commercially available casing, manufactured from acrylonitrile-butadienestyrene (ABS) plastic. Size of the casing is mm and the weight of the measurement device excluding the measurement wires is 48 g. Encapsulated measurement device is presented in Fig. 2.

3 720 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 14, NO. 3, MAY 2010 Fig. 2. Portable measurement device in the enclosure. Total weight of the device excluding measurement wires and electrodes is 48 g. A. Bioimpedance Measurement Block Purpose of the bioimpedance measurement in the measurement device is to produce information about changes in the volume of the thorax, which is mainly caused by respiration activity. Measuring the bioimpedance requires feeding a small alternating current, excitation current, into a tissue under measurement. Voltage generated by the current is measured and the bioimpedance is formed as a ratio between voltage and current Z measured = U measured /I excitation. Bioimpedance can be measured with two, three or four electrodes. In order to decrease measurement errors caused by electrode skin interface a four electrode, tetrapolar, measurement setup is preferred. In this setup, two electrodes are used to feed the current and other two to measure the generated voltage. In the implemented measurement device, the bioimpedance is measured with a custom designed measurement block, presented in Fig. 1. The measurement block can be divided to two parts, excitation current generation and voltage measurement. In measurement device prototypes [20], [21], a sinusoidal excitation current is generated with a direct digital synthesis (DDS) component. This integrated circuit requires a relatively large amount of power, which is not desirable in portable devices. A benefit of using DDS is a possibility to change the excitation currents frequency during the measurement. However, according to measurements performed with prototypes, changing the frequency plays only a small role in respiration measurement as long as the frequency is high enough, i.e., over 100 khz. In order to reduce the energy consumption of the measurement device, the DDS is replaced with a simple oscillator circuit implemented with a second order low pass filter, tuned to be unstable and oscillate with a frequency of 130 khz. This implementation reduces the power consumption from 21 mw drawn by the DDS to 3 mw required by the operational amplifier in the oscillation circuit. The generated signal is not a pure sinusoidal voltage and its dc (direct current) level is not correct. Therefore, the signal is filtered before converting to current and feeding to the tissue. Voltage generated by the excitation current is measured with an instrumentation amplifier (IA). A problem with the instrumentation amplifier arose from a relatively high bandwidth requirement. Because the base frequency of the measured signal is 130 khz the bandwidth of the IA should be at least one decade higher in order to prevent signal s attenuation. However, according to survey about commercially available IAs it turned out that common mode rejection ratio (CMRR) of amplifiers with a corner frequency this high is quite poor. Therefore, the IA is implemented with three separate operational amplifiers and tuned manually to achieve high CMRR. In order to minimize the energy consumption in the bioimpedance measurement block, all filters and the peak detection circuit are implemented with passive topology. The gain stage is implemented with one operational amplifier and its amplification can be adjusted through a digital potentiometer. As can be seen from Fig. 1, the bioimpedance measurement contains a high pass filter with corner frequency of 44 mhz. This filter removes the dc level from the demodulated signal and, therefore, makes it impossible to measure absolute value of the impedance and to detect changes in the impedance, which are beyond the corner frequency. This is a conscious decision, enabling high amplification of impedance s dynamic part without signal saturation. Even through in some measurements, the low frequency band and impedance s absolute value contain useful information, they accurate measurement is challenging and affected by many uncontrollable variables, e.g., body posture, body movements, electrode skin interface, and variances in the measurement electronics. Accurate measurement would require, e.g., continuous calibration of the measurement system with a stable reference value, as has been done in other bioimpedance measurement systems [22]. However, in order to keep the measurement electronics simple and the power consumption low, a decision has been made to reject the low frequency band and to measure only the dynamic chances in the bioimpedance s absolute value. B. Electrocardiography Measurement Block Electrocardiography signal is measured with same electrodes than bioimpedance signal and conditioned with a custom designed measurement block, presented in Fig. 1. Because the bandwidth of the electrocardiograph signal is much narrower than the frequency of the bioimpedance excitation current, a commercial instrumentation amplifier can be utilized. Before instrumentation amplifier, the 130-kHz signal caused by excitation current is removed with a low pass filter. Energy consumption of the ECG measurement block is kept as low as possible and there is only one active gain block, whose gain can be adjusted with a digital potentiometer. C. Acceleration Measurement Acceleration sensor is added to the measurement device in order to measure user s activity. This makes it possible to, e.g., detect whether a high heart or respiration rate is caused by physical activity or a fit. The acceleration is measured with a single accelerometer (SCA3000-E04) manufactured by VTI Technologies. This sensor measures accelerations in three dimensions with 6 g measurement range. Accelerometer has an on chip analogue to digital converter (ADC) and it is connected to the microcontroller through a digital serial interface. Sensor s power consumption is very low only 0.5 mw.

4 VUORELA et al.: DESIGN AND IMPLEMENTATION OF A PORTABLE LONG-TERM PHYSIOLOGICAL SIGNAL RECORDER 721 D. Control and Measurement Storage Digital part of the measurement device controls the measurements, stores measurement results and handles the communication between measurement device and a base station, connected to a computer. The measurement device is controlled with a Texas Instruments microcontroller (MCU) MSP430F1611, which is a 16-bit ultralow power circuit. Main reasons for selecting MSP430F1611 are its low energy consumption and relatively high amount of memory, 48 KB nonvolatile flash memory for program and 10 KB of volatile memory for runtime data. Bioimpedance and ECG signals are converted to digital with MCU s internal ADC, whose resolution is 12-bits. In conversion program special attention is paid on minimization of the jitter between consecutive measurements. This guarantees that the measured signals can be reconstructed afterwards with high time domain accuracy. Measurement results are stored onto a microsecure digital (µsd) memory card. A drawback in µsd is its relatively highenergy consumption. However, the removable memory card is selected because it enables easy transfer of measurement results from measurement device to a computer with a memory card reader. Measurement data transfer through a universal serial bus (USB) and radio link has also been tested during the implementation process. Due to relatively low computation power of the measurement device s MCU, transferring of all measurement data with these methods requires as long time as the measurement session itself. Therefore, these data transfer methods are rejected. E. Radio Communication For reviewing the signals and for configuring the measurement device before measurement recording, a radio communication interface is implemented. Possible radio frequency communication techniques at the moment for low power devices are Bluetooth (BT), ZigBee, and ANT. Selection between different technologies has been made according to technologies key-properties [23] [25], which were defined to be power consumption, data rate, amount of software required, and link s range. Bluetooth consumes more energy than other two technologies. However, Bluetooth has also a higher data rate, which enables relatively low duty cycles and, therefore, decreases the average energy consumption. Drawbacks in Bluetooth are the relatively high peak currents, which must be taken into account in selection of a power source, and a large amount of program memory required for a software stack. An easy way to utilize Bluetooth is to purchase a commercial module in which the radio and the microcontroller with a software stack are integrated on the same printed circuit board (PCB). However, this causes some hardware overhead, because another processor or a microcontroller is required for performing the measurements. ZigBee and ANT require considerably less energy than Bluetooth, but their data rates are also lower which increases the duty cycle and the average energy consumption. In the implemented portable measurement device, the operating range of the radio link is not an important factor, because the link is not utilized during the actual measurements recording. Eventually, the ANT link was selected. Main reasons for the selection are extremely small amount of software required for the stack and low energy consumption versus data rate. F. Power Supply and Energy Consumption The measurement device is powered from a lithium polymer (LiPo) rechargeable battery. The utilized battery is PLF from Varta s ( Poliflex battery series. Capacity of the battery is 340 mah, and size and weight mm and 8 g, respectively. Energy consumption of the measurement device is relatively small. In measurement mode, when the device is sampling measurement signals and storing data onto the memory card, the current consumption is 13 ma. In the tracking mode, when the device is not storing measurements but transferring data through the radio link to a computer, current consumption is 14 ma. In theory, this yields to approximately one-day operating time in both modes. This has been verified with a 24-h measurement test, presented in the verification section. G. User Interface User interface (UI) of the measurement system can be divided into two parts. The first part is a hardware UI on the portable measurement device and the second part is a graphical user interface (GUI) running in a computer. Dividing the user interface into two parts makes it possible to create only a minimal easy to use interface on the portable device and then to implement all complicate operations, like setting sample rates, with the graphical user interface through a radio link. Before a measurement session, the graphical interface is utilized to verify that the measurement signals quality is high enough and after the session measurement data is converted into a readable format with the GUI program. During the measurement storing, the GUI and radio link are shut down and only the hardware UI is utilized. User interface on the portable measurement device consists of three LEDs and two push buttons as can be seen from Fig. 2. At the moment naming of UI components does not describe the functionality of the device at all and in the future the naming must be redesigned. With this UI, it is only possible to start and stop measurements and to active the radio link. LEDs A, B, and C indicate the battery status, radio link status, and the measurement activity, respectively. IV. VERIFICATION MEASUREMENTS In order to verify the measurement device operation, some tests are performed. Purpose of these measurements is not to produce useful medical results but only to proof that device is working as designed. One target is to test how device performs with movement artifacts and normal daily activities. A. Electrodes The data recorder has been tested with two different electrodes. The first ones are commercial gel-paste Ag/AgCl

5 722 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 14, NO. 3, MAY 2010 Fig. 3. Electrode placement in the test measurements. Fig. 5. (Top) ECG signals measured with the implemented measurement device (black) and the Biopac measurement system (gray). (Bottom) Computational difference of measured signals (S.D µv). Fig. 4. (Top) Impedance pneumography signal measured with developed measurement device (black) and respiration signal measured with a pneumotachograph connected to a Biopac measurement system (gray). (Bottom) Computational difference of measured signals (S.D ml). electrodes, designed for clinical ECG measurement. These electrodes are attached to the skin with an adhesive included in the electrode. Another approach in portable physiological measurement devices is to utilize textile electrodes as described in the section two. Therefore, the measurement device has been also tested with a custom made set of textile electrodes. These electrodes are manufactured from a conductive silver yarn with an embroidery machine. Textile electrodes are square shaped and size of one electrode is 225 mm 2. Electrodes are described in more detail by Seppä et al. [18]. In all verification tests, the electrode configuration is selected according to the previous measurement results obtained by Seppä et al. [18]. In the selected configuration, the volume changes in thorax, caused by respiration, correlate well with the pulmonary flow. The utilized electrode configuration is illustrated in Fig. 3. B. Comparison With Reference Signals In order to verify accuracy of the implemented measurement device, both the bioimpedance and the ECG signals are compared against reference signals acquired with a Biopac measurement system. During the measurement the test subject has been sitting steadily in a chair so that movement artifacts are minimized. ECG and bioimpedance signals are measured with commercial gel-paste electrodes located in aforementioned places. Comparison signal for the bioimpedance is respiration airflow, i.e., tidal volume (TV), signal acquired from a pneumotachograph. All measurement signals are mathematically processed in the MATLAB and comparison results for respiration and ECG signals are presented in Figs. 4 and 5, respectively. Upper part in both figures presents the reference signal and the signal measured with the developed measurement device. Lower part in the figures presents a difference signal calculated from measurement signals. In Fig. 4, the voltage signal from the implemented device is calibrated to match the liter values acquired from the Biopac device. The calibration is done by attenuating the voltage signal 44.6 db. For removing a possible offset both signals are high pass filtered with a second order Chebyshev filter, which corner frequency is 20 mhz. The reference signal is also low pass filtered with a filter which corner frequency is 65 Hz. As can be seen from Fig. 4, the maximum difference in signals is about 50 ml. A calculated standard deviation of the difference signal is 22.4 ml. One potential cause of differences between the respiration signals is changes in the heart volume caused by cardiac activity. The bioimpedance measures all volume changes in the measurement zone, whereas the pneumotachograph measures only changes in the lungs volume. This difference can be minimized through selecting the electrode placement so that heart is not in the measurement zone. It must be mentioned that according to our preliminary tests the calibration of bioimpedance pneumography signal and the tidal volume signal from pneumotachograph is valid only for a certain test subject in a certain position. Therefore, in long-term measurements it is difficult to obtain absolute volumetric lung values from bioimpedance signal if the test subject is moving. However, it has been shown that at least the respiration rate can be accurately measured with the impedance pneumography even during an exercise, e.g., running [26]. In the verification test, the respiration signals and ECG signals have been measured simultaneously. The EGC signals presented in Fig. 5 is a snapshot of the whole test and locates between 10 and 20-s time stamps in Fig. 4. The ECG signal from the implemented measurement device is attenuated 82.9 db in order to calibrate it with the reference ECG signal. A possible dc-offset is removed from both signals by high pass filtering them in the MATLAB with a fourth order Chebyshev filter, which corner frequency is 0.5 Hz. As can be seen from Fig. 5, the ECG signal measured with the implemented measurement device and the reference signal have a very similar shapes and the difference between signals is less than 20 µv. Standard deviation of the difference signal is 2.88 µv. The largest differences in the ECG signal are on R-peak voltages. This difference is caused by Biopac s higher sample rate, which enables more accurate measurement of R-peak s amplitude.

6 VUORELA et al.: DESIGN AND IMPLEMENTATION OF A PORTABLE LONG-TERM PHYSIOLOGICAL SIGNAL RECORDER 723 Fig. 6. Comparison measurement of textile and Ag/AgCl gel-paste electrodes. Upper curves present the raw unconditioned measurement signal and lower curves presents the same signal averaged for 1-s period. Fig. 7. Light exercise test performed with Ag/AgCl gel-paste electrodes. TABLE I ACTIVITIES PERFORMED DURING THE COMPARISON TEST OF TEXTILE AND Ag/AgCl GEL-PASTE ELECTRODES C. Movement Artifacts A major source of error in portable physiological measurements is artifacts caused by motion. In order to test the behavior of the measurement device during motion with different electrodes, a comparison test is performed. In this test a test person is sitting in a chair in front of a desk and the measurement device is set on the desk. Electrodes are in the aforementioned places. The bioimpedance signal is recorded with 256 Hz sample rate. Test is preformed first with textile electrodes and then with gel-paste Ag/AgCl electrodes. During the test different upper body activities, i.e., swinging arms, are performed. Because performed motions cause a lot of high frequency disturbance raw measurement signals are averaged for 1-second period. Averaging is done with a central moving average filter, which calculates an average from 256 samples and, therefore, produces a cutoff frequency of 1.5 Hz. These values for the filter are chosen according to earlier experience in bioimpedance data analysis and according to results obtained during the analysis of these exact signals. We have previously detected that averaging for a relatively long period removes disturbances caused, e.g., by steps during walking [22]. Both the raw measurement data and averaged data signals are presented in Fig. 6. In Fig. 6, time span during which an activity is performed is marked with vertical dash lines and Roman numeral. Performed activities and corresponding Roman numerals are listed in the Table I. Both test measurements start with a period during which the test person is sitting still and breathing normally (see segment I in Table I and in Fig. 6). As can be seen from Fig. 6, during this period the movement artifacts are negligible and the respiration activity is clearly visible in raw measurement signal and in averaged signal. During a next time period, the test person is holding breath and sitting still (see segment II in Table I and in Fig. 6). During this period both electrodes give a relatively smooth and flat signal with no sign of movement artifacts. When the test person is swinging arms and holding the breath (see segments III and IV, respectively) very strong artifacts can be seen in the raw measurement data. However, because the frequency of arm movements is high, it can be effectively averaged out as can been in Fig. 6. The most interesting results are obtained from the periods when arms are moved and the test person is breathing normally (see segments V and VI in Table I and in Fig. 6). Moving arms causes high artifacts, which are removed through averaging. However, swinging the arms vertically (see segment VI in Table I and in Fig. 6), like, e.g., when running, produces smaller artifacts than swinging the arms horizontally (see segment V in Table I and in Fig. 6), like, e.g., when swimming a breaststroke. Due to this the breathing signal is more readable after averaging during the vertical movements than during the horizontal movements. D. Light Exercise The measurement device is targeted to measurements performed during normal daily activities. Therefore, the device should produce high-quality signals during normal movements and during light exercise like walking and climbing stairs. In order to test this operation a light exercise test is performed with both the Ag/AgCl and textile electrodes. Result signals are presented in Figs. 7 and 8, respectively. The different test periods are listed in Table II. The test starts with a 1-min period during which the test person is standing still (see segment A in Table II, Fig. 7, and Fig. 8). This period provides comparison data because no movement artifacts are present. After the first period four floors of stairs are walked downwards (see segment B in Table II, Fig. 7, and Fig. 8). Then a 35-m long corridor is walked from one end to another (see segment C in Table II, Fig. 7, and Fig. 8).

7 724 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 14, NO. 3, MAY 2010 Fig. 8. Light exercise test performed with textile electrodes. TABLE II ACTIVITIES PERFORMED DURING THE LIGHT EXERCISE TEST Fig. 9. Results of the 24-h real-life measurement, implemented with the Ag/AgCl gel-paste electrodes. Annotations are as followed: 1, working in an office; 2 and 8, walking a route between home and work; 3 and 5, evening routines like eating and watching TV; 4, outdoor activities with a 2-year old child; 6, sleeping; and 7, morning routines. bioimpedance signal and confirmed with comparison to the increased heart rate. However, as can be seen from Fig. 8, the ECG signal measured with textile electrodes is very sensitive to motion artifacts. This problem requires further research and testing with different shape textile electrodes Next four floors of stairs are walked upwards (see segment D in Table II, Fig. 7, and Fig. 8) and another 35 m long corridor is walked from end to end, which returns the test person to the starting point. The test route is walked twice. On the first run hands are kept in a pocket, which decreases motion artifacts in the upper body. In Figs. 6 and 7, this is marked with lower index p. On the second run, hands are swinging freely, which should produce more artifacts. Lower index f indicates the second test run. Between the test-runs and after the second run there is a 1-min standing still period, which provides some rest for the test person and decreases the heart rate. The heart rate signal presented in Figs. 7 and 8, respectively, is calculated from the measured ECG signal with a simple algorithm. The ECG signal measured with 512 Hz sample rate is first averaged with a central moving average filter of eight samples. Next, the signal is differentiated, which amplifies the fastest changes, i.e., r-peaks. After this the r-peak s position is defined with a threshold method in which the reference level is defined as two thirds of a difference between an average and a maximum value inside a one second window. In order to reduce an amount of false r-peak detections an assumption is made that the heart rate cannot exceed two hundred beats per minute. The bioimpedance signal is averaged for a 1-s period to reduce motion artifacts in similar manner than in movement artifacts tests. Acceleration signals are averaged for a period of one quarter of second. Results reveal that after averaging the bioimpedance signal is readable during the movements with both electrodes. Also the depth of the breathing can be detected from the signal. During and after the exercise sessions the test person is breathing more deeply than before the session, which can be seen in the E. Real-Life 24-h Test In order to evaluate the measurement device operation during normal daily activities, a 24-h measurement is carried out. In this test, gel-paste electrodes were utilized. Test started in the morning, when the test person arrived to the office and continued until the next morning. Test results are presented in Fig. 9. In order to make the measurement data more readable, every signal is presented as an average of 5-min values. Respiration rate signal is derived from the bioimpedance signal in order to illustrate that the bioimpedance signal can produce useful data. Activity and posture signals are derived from the acceleration signals. Activity is formed as a variance of the acceleration vectors absolute value and the posture signal is obtained from the X-axis acceleration signal with a threshold method. The posture signal indicates whether the subject is in standing or sitting E (erect) or lying down S (supine). A similar 24-h test has been carried out with textile electrodes. However, during the measurement session, a measurement device suffered some kind of malfunction and the measured signals are, therefore, unreadable. This problem must be further studied. V. CONCLUSION AND FUTURE WORK This paper has presented the design and implementation of a portable physiological measurement device. Operation of the device has been verified through test measurements and the device has been found to workout as designed. In the future, more tests are going to be performed with the device and with different kind of textile electrodes. Electronic and software modules designed during the process are going to be utilized in implementation of other forthcoming measurement device projects.

8 VUORELA et al.: DESIGN AND IMPLEMENTATION OF A PORTABLE LONG-TERM PHYSIOLOGICAL SIGNAL RECORDER 725 ACKNOWLEDGMENT The authors would like to thank the people working in the Wisepla project for the cooperation. REFERENCES [1] H. Steg, H. Strese, J. Hull, and S. Schmidt. (2005, Sep.). Europe is facing a demographic challenge ambient assisted living offers solutions. [Online]. Available: [2] R. J. Rosati, Evaluation of remote monitoring in home health care, ehealth, in Proc. Int. Conf. Telemed., Soc. Med. (etelemed), Feb. 1 7, 2009, pp [3] Polar heart rate monitors. (2009, May 26). [Online]. Available: [4] Suunto wrist computers. (2009, May 26). [Online]. Available: [5] Vivago well being products. (2009, April 16). [Online]. Available: [6] FRWD sport computers. (2009, May 26). [Online]. Available: [7] SenseWear R armband homepage. (2009, May 26). [Online]. Available: [8] Homepage of LifeShirt R. (2009, May 26). [Online]. Available: [9] Homepage of Adidas and Polar adistar R Fusion products. (2009, May 26). [Online]. Available: [10] U. Anliker, J. A. Ward, P. Lukowicz, G. Troster, F. Dolveck, M. Baer, F. Keita, E. B. Schenker, F. Catarsi, L. Coluccini, A. Belardinelli, D. Shklarski, M. Alon, E. Hirt, R. Schmid, and M. Vuskovic, AMON: A wearable multiparameter medical monitoring and alert system, IEEE Trans. Inf. Technol. Biomed., vol. 8, no. 4, pp , Dec [11] N. Noury, A. Dittmar, C. Corroy, R. Baghai, J. L. Weber, D. Blanc, F. Klefstat, A. Blinovska, S. Vaysse, and B. Comet, VTAMN A smart clothe for ambulatory remote monitoring of physiological parameters and activity, in Proc. 26th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (IEMBS), Sep. 1 5, 2004, vol. 2, pp [12] R. Paradiso, G. Loriga, and N. Taccini, A wearable health care system based on knitted integrated sensors, IEEE Trans. Inf. Technol. Biomed., vol. 9, no. 3, pp , Sep [13] R. Paradiso, A. Alonso, D. Cianflone, A. Milsis, T. Vavouras, and C. Malliopoulos, Remote health monitoring with wearable non-invasive mobile system: The Healthwear project, in Proc. 30th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBS), Aug , 2008, pp [14] E. Hughes, M. Masilela, P. Eddings, A. Raflq, C. Boanca, and R. Merrell, VMote: A wearable wireless health monitoring system, in Proc. 9th Int. Conf. E-Health Netw., Appl. Serv., Jun , 2007, pp [15] D.-W. Ryoo, C.-S. Bae, and J.-W. Lee, The wearable wrist-type gadget for healthcare based on physiological signals, in Proc. Dig. Tech. Papers Int. Conf. Consum. Electron. (ICCE), Jan. 9 13, 2008, pp [16] N. Oliver and F. Flores-Mangas, HealthGear: A real-time wearable system for monitoring and analyzing physiological signals, in Proc. Int. Workshop Wearable Implantable Body Sens. Netw. (BSN), Apr. 3 5, 2006, p. 4. [17] S. Kim, D. Ryoo, and C. Bae, Implementation of smart headband for the wearable healthcare, in Proc. Dig. Tech. Papers Int. Conf. Consum. Electron. (ICCE), Jan. 9 13, 2008, pp [18] V.-P. Seppä, J. Viik, A. Naveed, J. Väisänen, and J. Hyttinen, Signal waveform agreement between spirometer and impedance pneumography of six chest band electrode configurations, in Proc. IFMBE, vol. 25 of VII, New York: Springer-Verlag, 2009, pp [19] I. Korhonen, J. Parkka, and M. Van Gils, Health monitoring in the home of the future, IEEE Eng. Med. Biol. Mag., vol. 22, no. 3, pp , May Jun [20] T. Vuorela, V.-P. Seppä, J. Vanhala, and J. Hyttinen, Two portable longterm measurement devices for ECG and bioimpedance, in Proc. 2nd Int. Conf. Pervasive Comput. Technol. Healthcare 2008, Pervasive Health 2008, Tampere, Finland, Jan. 30 Feb. 1, p. 4. [21] T. Vuorela, V.-P. Seppä, J. Vanhala, and J. Hyttinen, Wireless measurement system for bioimpedance and ECG, in Proc. 13th Int. Conf. Electr. Bioimpedance 8th Conf. Electr. Impedance Tomogr. (IFMBE), Graz, Austria, Aug. 29 Sep. 9, 2007, vol. 17, pp [22] T. Vuorela, K. Kukkonen, J. Rantanen, T. Järvinen, and J. Vanhala, Bioimpedance measurement system for smart clothing, in Proc. 7th Int. Symp. Wearable Comput. (ISWC), White Plains, NY, Oct , 2003, pp [23] G.-Z. Yang, Ed., Body Sensor Networks. New York: Springer-Verlag, May 2006, p ISBN-10: [24] Homepage of ANT technology. (2009, Jun. 11). [Online]. Available: [25] J.-S. Lee, Y.-W. Su, and C.-C. Shen, A comparative study of wireless protocols: Bluetooth, UWB, ZigBee, and Wi-Fi, in Proc. 33rd Annu. Conf. IEEE Ind. Electron. Soc. (IECON), Nov. 5 8, 2007, pp [26] V.-P. Seppä, O. Lahtinen, J. Väisänen, and J. Hyttinen, Assessment of breathing parameters during running with a wearable bioimpedance device, in Proc. IFMBE, vol. 22, New York: Springer-Verlag, 2008, pp Timo Vuorela received the M.Sc. degree in electrical engineering from Tampere University of Technology (TUT), Tampere, Finland, in He is currently working toward the Ph.D. degree in the Personal Electronics Group in the Department of Electronics, TUT. His current research interests include embedded systems, e.g., portable physiological measurement devices and smart clothing. Ville-Pekka Seppä received the M.Sc. degree from Tampere University of Technology (TUT), Tampere, Finland, in He is currently working toward the Ph.D. degree in the Department of Biomedical Engineering, TUT. His current research interests include physiological measurement, pulmonary function, physiological signal processing, and impedance pneumography. Jukka Vanhala received the M.Sc. degree in 1985, the Licentiate of Technology degree in 1990, and the Doctor of Technology degree in computer science in 1998, all from Tampere University of Technology, Tampere, Finland. He is currently a Professor in the Department of Electronics, Tampere University of Technology, where he has been engaged in teaching embedded systems. He is also the Head of the Kankaapää Research Unit for Wearable Technology. His research interests include ambient intelligence, smart garments and embedded systems. He has been engaged in the academics for 15 years and in the industry for 5 years both in Finland and in the United States. Jari Hyttinen received the M.Sc. and Ph.D. degrees from Tampere University of Technology (TUT), Tampere, Finland, in 1986 and 1994, respectively. He is currently a Professor of biomedical engineering at TUT and the Director of the Department of Biomedical Engineering, TUT. He is the author or coauthor of more than 200 scientific papers as well as inventor or coinventor in some patents. Dr. Hyttinen is a former Chairman and current member of the Board of the Finnish Society of Medical Physics ( and Biomedical Engineering (affiliate of the IFMBE). He is a member of the general council of the European Alliance for Medical and Biological Engineering and Sciences (