7/24/2019 Wrist Movement

1/4

A measurement system for wrist movements inbiomedical applications

Bence J. Borb ely, Attila Tihanyi , Peter Szolgay Faculty of Information Technology and Bionics, P azmany Peter Catholic University, Budapest, H-1083

Cellular Sensory and Optical Wave Computing Laboratory, Hungarian Academy of Sciences, Budapest, H-1111

Abstract The design and proof of concept implementationof a biomedical measurement device specically targeting humanwrist movements is presented. The key aspects of development arethe integrated measurement of wrist kinematics and lower armmuscle activities, wireless operation and the possibility of real-time data streaming. The designed system addresses these require-ments using single chip 9 degrees-of-freedom inertial sensors forkinematic measurements, an active myoelectric electrode frontenddesign to record muscle activities and a Bluetooth communicationinterface for device control and data streaming. In addition

to design considerations and proof of concept implementation,kinematic test measurement data is presented to validate systemusability in a future wrist movement classication task.

I. INTRODUCTION

The area of health assisting technology have been moreand more active in the last few years in the eld of medicalinstrumentation, movement rehabilitation, prosthetic devicesand tness accessories, just to name a few. This processresulted in a wider spread of devices addressing differentareas on the border of life sciences and engineering, fromsimple commercial products (e.g. small pulse monitors) to verycomplex research projects like the Modular Prosthetic Limb 1

and its control interfaces.Human movement recording - a complementary area to

these elds - however, did not show this level of activity. Thepresumable reason for this is that the measurement methodsin motion tracking applications are standardized, tested andvalidated since decades and manufacturers keep providing highlevel laboratory systems to meet these requirements. The mostwidely used measurement methods are optical (e.g. Viconor OptiTrack) and ultrasound-based (Zebris) and while thesedevices are at the top of their classes and provide excellenttracking accuracy in most cases, they require a controlled lab-oratory environment for proper operation (even when outdoormeasurement is possible).

A further aspect with regard to biomedical applicationsis the possibility of recording other modalities, most impor-tantly bioelectric signals like muscle activities (electromyo-gram, EMG). Measuring EMG is a key aspect to get deeperinsight into the dynamics of movements because it gives closerinformation about muscle activation patterns. There are manyapproaches in the literature for individual EMG recording [1][4], but these does not include the possibility of simultaneouskinematic measurement. One of the mentioned systems (Ze-bris) even includes an integrated EMG measurement option,however the wired connection of sensors and electrodes to alarge base unit highly restricts the possible movement range

1http://www.jhuapl.edu/prosthetics/scientists/mpl.asp

of the subject during recording and there is no option for real-time data streaming while a measurement is in progress.

In addition to line of sight methods, an other approachusing inertial sensors has gained popularity recently, allowingmarkerless operation without the need of external accessories(e.g. cameras or microphones). As a main difference, inertialsensors give orientation information more accurately thanposition compared to marker-based methods, providing inter-segmental angles of measured body parts [5][7]. While there

are industry-grade commercial products available with thistechnology (e.g. Xsens portfolio), to the best of authorsknowledge there are no wireless and integrated systems avail-able utilizing inertial based kinematic measurements and mus-cle activity recording in the same package.

The work described in this paper is an effort towards thedevelopment of a research oriented, fully customizable, inte-grated and wearable measurement system specically targetingwrist movements in order to record and analyze movementpatterns for biomedical applications.

II. SYSTEM DESIGN

The main system requirements are the following:

At least 100 Hz recording of hand orientation.

Up to 6 channel, 1 kHz recording of forearm surfaceEMG.

Real-time data streaming and on-board data storagecapability.

Low power consumption and self-contained, wirelessoperation.

The concept of the system is depicted in Figure 1. The core of the design is a Base Unit responsible for controlling the mea-surements, collecting sensor data, performing pre-processingtasks and sending and/or storing the output data. The EMGsensors and the Base Unit itself are planned to be integratedinto a sensor ring around the lower arm to reduce measurementnoise and make usage more practical compared to standardwet-electrode EMG systems using long wires. There are twoinertial sensors in the design which are placed at the distalend of the lower arm right before the wrist joint and on theback of the hand, respectively. This arrangement allows themeasurement of the hands relative orientation with respect tothe lower arms orientation, resulting in anatomically relevant joint angles of the wrist.

Our concept and placement of the EMG sensor ring issimilar to Thalmic Labs commercial product called Myo,however we would like to address a more research oriented

978-1-4799-9877-7/15/$31.00 2015 IEEE.

2015 European Conference on Circuit Theory and Design (ECCTD)

7/24/2019 Wrist Movement

2/4

Fig. 1. Concept drawing of the measurement system. The Base Unit isplanned to be integrated into an EMG sensor ring around the lower arm.Inertial sensors are placed at the distal end of the lower arm right before thewrist joint and on the back of the hand, allowing the measurement of thehands relative orientation with respect to the lower arms orientation.

scenario providing a higher EMG data rate (the Myo has only200 Hz EMG output rate based on the companys website 2)and external inertial sensors for kinematic measurements.

A. Base Unit The block diagram of the Base Unit is depicted in Figure

2. It is designed with an STM32F407VG microcontroller unit(MCU) as its central element which is a high performanceARM Cortex-M4 core running at up to 168 MHz. The MCUhas 1 MB ash and 192 kB SRAM, built-in 12-bit 2.4 MSPSADCs, various serial peripherals (including I 2 C, SPI andUART), a dedicated SDIO interface for high speed SD cardcontrol and a 16-stream DMA controller. In the initial phaseof development an STM32F4 Discovery board was used as thecentral hardware element of the system which provides accessto almost all pins of this MCU and a debugger unit in the samepackage.

Because the system is designed to run from battery, properpower supply is planned to be provided through a dedicatedPower Management unit that generates stable 3.3V digitalsupply from battery input ranging between 1.8V to 5.5V usinga buck-boost converter (TPS63001). The 5V analog supplywill be provided using a dual-output charge pump (MAX865)and ultralow-noise positive and negative low-dropout linearregulators (TPS7A4901 and TPS7A3001, respectively). Wire-less device control and data streaming is performed by aBluetooth 2.1 transceiver connected to the MCU throughUART interface.

B. Inertial Sensors

To perform measurements of joint kinematics, singlechip 9 degrees-of-freedom MEMS inertial sensors were used(MPU-9250). Each sensor integrates an individual 3-axis ac-celerometer, 3-axis gyroscope and 3-axis magnetometer withproperties shown in Table I.

TABLE I. I NERTIAL SENSOR PROPERTIES (MPU-9250)

Sensor type Full-scale range Resolution Sampling rate

Accelerometer 2, 4, 8, 16 G 16-bit up to 1kHz

Gyroscope 250, 500, 1000, 2000 /sec 16-bit up to 8kHz

Magnetometer 4800T 14 or 16-bit up to 100Hz

2http://developerblog.myo.com/raw-uncut-drops-today/

Fig. 2. Block diagram of the Base Unit. All measurement and processing isperformed by an ARM Cortex-M4 core running at 168 MHz. The Base Unitcontains a Power Management Unit to provide the digital and analog supplies,data streaming (Bluetooth) and data storage (SD card) modules. Inertialsensors and EMG electrodes are handled through corresponding hardwareinterfaces.

Based on these properties the sensors are able to provideenough exibility to measure most common human movementtasks without saturation. Considering the planned movement

tasks (low to moderate speed wrist movements) the sensorswere used as follows:

Accelerometer: 2G, 16-bit, 200Hz

Gyroscope: 500 /sec, 16-bit, 200Hz

Magnetometer: 4800T /sec, 16-bit, 100Hz

C. EMG frontend design

The schematic of the main EMG electrode frontend is de-picted in Figure 3/A. Considering differential signal recordingfor each channel ( EM G and EM G + ), the central element of the design is the INA128 instrumentation amplier (InAmp)

with adjustable gain and dual power supply operation. Thegain can be set with an external resistor ( R G ) between 1 and10,000. The device has wide power supply range ( 2.25Vto 18V), low quiescent current (700 A ) and high common-mode rejection ratio even at lower gains ( 90dB at f = 1kHz,G = 10V/V) that all make it suitable for portable EMG appli-cations. To allow amplication tuning during development, thegain resistor section of the frontend ( R 1 to R3 ) is designedto be adjustable over the full gain range of the amplier(R 1 = 100k trimmer, R G = 50k - 5 resulting in 2 - 10,000V/V gain), while providing the average of input potentials tothe active ground driver circuit.

The ground driver feeds back the inverted and ampliedaverage input voltage to the body through a high impedanceconnection to reduce the amount of common-mode offset andnoise (including 50Hz power line noise) in the differential sig-nal. Besides amplication, offset control is an other essentialpart of signal acquisition and conditioning which is realizedby a three resistor ( R 7 to R9 ) passive voltage averagingcircuit connected to the InAmps reference pin through a unity-gain buffer. Using this solution the DC offset of amplieroutput can be altered by about 1.43V which is suitableto compensate offsets introduced by the combination of thesmall amount of constant inter-electrode voltage (betweenEM G and EM G + ) and higher gains. At the time of writingoffset control is performed manually using a trimmer resistor,however in the nal design this functionality will be integrated

7/24/2019 Wrist Movement

3/4

Fig. 3. (A) Schematic drawing of the active EMG electrode frontend for differential measurements with active body ground driver and an offset controllerunit. (B) Test measurement output of the circuit during a single channel differential recording from the area of wrist exion muscles on the forearm.

with one of the MCUs 12-bit DACs for automated offsetcompensation even on longer terms (in this case considering3.3V operation, R 7 to R 9 resistor values need to be adjusted).

Figure 3/B shows test output from the InAmp with the gain

kept at a lower level to avoid output saturation and noise ampli-cation. During the measurement a single channel differentialrecording from the area of wrist exion muscles on the forearmwas performed using standard Ag/AgCl electrodes with aninter-electrode distance of about 3cm (effects from motionartefacts were minimised using self adhesive electrodes). Thetest circuit was assembled on a breadboard including powerisolation (THB 3-0511), 5V analog supply (using MAX865,TPS7A4901 and TPS7A3001) and the frontend itself andits output was measured directly with an oscilloscope. Therecorded data shows satisfying EMG characteristics [8] witha spectral distribution between 10 and 200Hz (calculated butnot show in the gure).

At the current state of the work the nal output stage of the frontend is still being developed. This will integrate a highorder low-pass lter designed for further signal amplicationand anti-aliasing purposes (AAF), and a biasing circuit thatoffsets the InAmps zero-symmetrical output to half of theanalog reference voltage in order to utilize the whole range of the used ADC. Considering the practical bandwidth of EMGsignals of up to 500Hz [8] the AAFs cutoff frequency willbe set to this value while the stopband attenuation remainsdependent of ADC sampling frequency and resolution.

D. Firmware

1) Base Unit: Device rmware was implemented in Cusing the Eclipse IDE (version Kepler) and the GNU Tools for ARM Embedded Processors package on an Ubuntu 12.04 LTSsystem. Device programming and debugging was performedwith OpenOCD (versoin 0.8.0). As written earlier, the centralelement of the system is an STM32F407VG ARM Cortex-M4high performance 32-bit microcontroller from STMicroelec-tronics. As this device provides enough horsepower to easilyhandle the ash and RAM overhead of an operating system,the Base Units rmware was designed and implemented usingFreeRTOS TM 3, a free and industry standard real-time operatingsystem for embedded applications. FreeRTOS Task, Queue

3http://www.freertos.org/

and Semaphore structure allowed designing system function-ality at a higher abstraction level and with straightforward exe-cution scheduling. For low-level device driver implementation,STs Standard Peripheral Library for the STM32F4 Discovery

kit (version 1.1.0) was used. The MCUs DMA controller wasutilized in each scenario where it was applicable to furtherimprove execution parallelism.

2) Inertial sensors: In addition to the Base Units rmware,a dedicated driver was developed for the inertial sensors toutilize control, calibration and measurement processes overthe I2 C bus. Zero motion calibration of the accelerometerand the gyroscope was implemented as part of the sensorinitialization process, while hard and soft iron calibration of the magnetometer was performed only once for each sensorand hard-coded into the sensor driver.

However, there is a proprietary on-chip Digital MotionProcessor (DMP) in each sensor package, it was not used

during the development because it is only capable of per-forming 6-axis (accelerometer + gyroscope) sensor fusion, andthere is no publicly available documentation for this unit.Instead, a computationally efcient open source orientationlter [9] was used to provide sensor orientations in softwareusing a gradient descent based 9-axis fusion algorithm. Theapplied method provides direct quaternion output (avoiding thephenomenon of gimbal lock) and is easily capable to providestable 200 Hz output rate enabling the system to meet andexceed the requirements for orientation data measurement.

3) EMG measurements: EMG recording is performed bythe MCUs built-in 12-bit SAR-ADC. Considering the real-ization aspects of the AAF after the analog EMG frontend,

a sampling frequency of 8 kHz was implemented for 6 inputchannels. The sampling is followed by a digital decimatinglow-pass lter ( F c = 500 Hz, decimation factor = 8) usingARMs CMSIS DSP library . To assure minimal lag and jitterduring the measurement, the ADC is controlled in scanningmode using DMA with double buffering. This setup assuresaliasing-free recording of EMG signals at 1 kHz output ratefor each channel.

E. Control software

A PC-side software application is being developed fordevice control, real-time data visualization and storage, and toperform custom post-processing tasks. The control software is

7/24/2019 Wrist Movement

4/4

Fig. 4. Proof of concept implementation of the measurement device with thecore unit (STM32F4 Discovery), SD card handling, Bluetooth transceiver andinertial sensors xed to a wristband with Velcro stripes. The EMG section isnot shown as it is still under development its the interface is prepared alreadyon the perfboard.

implemented using Python and the Kivy framework 4

includingAndroid and iOS as possible target platforms in addition to thethree major desktop operating systems.

III. PROOF OF CONCEPT IMPLEMENTATION

The proof of concept implementation of the measurementdevice excluding the EMG part is shown in Figure 4. Thisalpha prototype serves system validation, design error checkingand ne tuning purposes. The main hardware elements areassembled on a perfboard to allow rapid modications if necessary, while inertial sensors are xed to a wristband withVelcro stripes. Using this setup, preliminary experimental testdata from a wrist rest-extension-exion-extension-rest task areshown in Figure 5. The depicted angles were calculated fromthe combined quaternion values of the inertial sensors, showingpromising kinematic tracking capability of the device. Thesmall resting offsets in both extension and deviation anglesare due to the imperfect sensor alignment about their corre-sponding axis, basically caused by the fact that the subjectshand did not start nor returned back perfectly in line with theforearm in the experiment.

Although the electronics of EMG recording is still beingnalized, we can say that the current implementation meetssystem requirements providing 200 Hz kinematic recording,1 kHz EMG recording, data storage and wireless operation.The current consumption of the prototype is 150 mA at full

load, but this value can even be lower considering the extraquiescent current needed by the additional components of thedevelopment board. As a result, a slim LiPo battery could beused as power source in the nal design.

IV. CONCLUSION AND FUTURE WORK

In this paper we presented the design and proof of conceptimplementation of a biomedical measurement device withpreliminary measurement data showing its capability for kine-matic and EMG measurements. In the current state the designshows promising characteristics, however further testing and

4http://kivy.org/

Fig. 5. Experimental measurement data from a wrist movement task whilethe subject performed a rest-extension-exion-extension-rest sequence. Flexionand deviation angles were calculated from the combined quaternion datacoming from the inertial sensors placed at the distal end of the forearm andat the back of the hand as shown in Figure 4.

ne tuning is necessary to assure reliable operation. Furtherwork includes nalization and integration of EMG frontenddesign, PCB design and manufacturing of the device and datavalidity testing with reference measurements using a ZebrisCMS-HS system.

ACKNOWLEDGMENT

This research project was supported by the Central FundingProgram of P azmany Peter Catholic University (KAP1.1-14/030). In addition, the project was supported by the Univer-sity of National Excellence Program and the Research Facultygrant awarded to the Faculty of Information Technology andBionics of P azmany Peter Catholic University.

REFERENCES[1] W. Youn and J. Kim, Development of a compact-size and wireless

surface emg measurement system, in ICCAS-SICE, 2009 , Aug 2009,pp. 16251628.

[2] M. Balouchestani and S. Krishnan, Effective low-power wearable wire-less surface EMG sensor design based on analog-compressed sensing.Sensors (Basel, Switzerland) , vol. 14, no. 12, pp. 24 30528, Jan. 2014.

[3] K.-M. Chang, S.-H. Liu, and X.-H. Wu, A wireless sEMG recordingsystem and its application to muscle fatigue detection. Sensors (Basel,Switzerland) , vol. 12, no. 1, pp. 48999, Jan. 2012.

[4] X. Chen and Z. J. Wang, Pattern recognition of number gestures basedon a wireless surface EMG system, Biomedical Signal Processing and Control , vol. 8, no. 2, pp. 184192, Mar. 2013.

[5] H. Zhou, H. Hu, and Y. Tao, Inertial measurements of upper limbmotion. Medical & biological engineering & computing , vol. 44, no. 6,pp. 47987, Jul. 2006.

[6] H. Kortier, V. Sluiter, D. Roetenberg, and P. Veltink, Assessmentof hand kinematics using inertial and magnetic sensors, Journal of NeuroEngineering and Rehabilitation , vol. 11, no. 1, p. 70, 2014.

[7] X. Chen, J. Zhang, W. Hamel, and J. Tan, An inertial-based humanmotion tracking system with twists and exponential maps, in Roboticsand Automation (ICRA), 2014 IEEE Int. Conf, on , May 2014, pp. 56655670.

[8] P. Konrad, The ABC of EMG, Tech. Rep. March, 2006. [Online].Available: http://www.noraxon.com/docs/education/abc-of-emg.pdf

[9] S. Madgwick, A. Harrison, and R. Vaidyanathan, Estimation of imu andmarg orientation using a gradient descent algorithm, in Rehabilitation Robotics (ICORR), 2011 IEEE Int. Conf. on , June 2011, pp. 17.