Low-cost Active Dry-Contact Surface EMG Sensor for Bionic Arms
Asma M. Naim, Kithmin Wickramasinghe, Ashwin De Silva, Malsha V. Perera, Thilina Dulantha Lalitharatne, Simon L. Kappel
aa r X i v : . [ c s . ET ] S e p Low-cost Active Dry-Contact Surface EMG Sensorfor Bionic Arms
Asma M. Naim ∗ Dept. Electronic and TelecommunicationEngineering, University of Moratuwa
Kithmin Wickramasinghe ∗ Dept. Electronic and TelecommunicationEngineering, University of Moratuwa
Ashwin De Silva
Dept. Electronic and TelecommunicationEngineering, University of Moratuwa
Malsha V. Perera
Dept. Electronic and TelecommunicationEngineering, University of Moratuwa
Thilina Dulantha Lalitharatne
Department of Mechanical Engineering,University of Moratuwa
Simon L. Kappel
Dept. Electronic and TelecommunicationEngineering, University of Moratuwa
Abstract —Surface electromyography (sEMG) is a popularbio-signal used for controlling prostheses and finger gesturerecognition mechanisms. Myoelectric prostheses are costly, andmost commercially available sEMG acquisition systems are notsuitable for real-time gesture recognition. In this paper, a methodof acquiring sEMG signals using novel low-cost, active, dry-contact, flexible sensors has been proposed. Since the activesEMG sensor was developed to be used along with a bionic arm,the sensor was tested for its ability to acquire sEMG signals thatcould be used for real-time classification of five selected gestures.In a study of 4 subjects, the average classification accuracy forreal-time gesture classification using the active sEMG sensorsystem was 85%. The common-mode rejection ratio of the sensorwas measured to 59 dB, and thus the sensor’s performance wasnot substantially limited by its active circuitry. The proposedsensors can be interfaced with a variety of amplifiers to performfully wearable sEMG acquisition. This satisfies the need for alow-cost sEMG acquisition system for prostheses.
Index Terms —Surface electromyography, Active electrode,Flexible printed circuit, Bionic arm, Gesture classification
I. I
NTRODUCTION
Amputation is the removal of a limb by trauma, medical ill-ness or surgery. There are approximately 10 million amputeesin the world, of which nearly % are living with an upperextremity amputation [1]. Transradial amputations (forearm)account for % of all upper extremity amputations [2]. Inrecent years, diseases and accidents, both vehicular and work-related, have drastically increased the number of catastrophicinjuries, resulting in limb losses.The rejection rate for prosthetics are high among upperlimb amputees, as people sustaining upper limb amputationspresent complex rehabilitative needs. Proper rehabilitation and *These authors contributed equally to the work.© 2020 IEEE. Personal use of this material is permitted. Permission fromIEEE must be obtained for all other uses, in any current or future media,including reprinting/republishing this material for advertising or promotionalpurposes, creating new collective works, for resale or redistribution to serversor lists, or reuse of any copyrighted component of this work in other works. comfortable, affordable and functional prostheses are a hugebenefit in the facilitation of functional restoration. Due tothe high complexity of myoelectric transradial prosthetics,the commercially available prosthetics in this category arecurrently extremely costly (as high as $ 75,000) [3]. Theseprosthetics translate muscle activity into information whichis used by motors to control the movements of the artificiallimbs.The muscle activity associated with finger movements iscaused by variations in ionic currents of relevant muscle fibers,and can be measured as myoelectric potentials on the surfaceof the forearm. Surface electromyography (sEMG) is a methodof acquiring myoelectric signals from the surface of the skin.sEMG is an important tool for Human-Computer interactiontasks, such as finger gesture recognition [4]. Typically, mul-tiple sEMG sensors are placed on the forearm of a subjectto enable the characterization of movements involving severalmuscles. Ideally, the sensors are placed on the skin surfacedirectly above the muscle of interest, to obtain the highestquality sEMG signals [5].sEMG signals are recorded using active or passive elec-trodes [6]. In active electrodes, the sEMG signals are amplifiedclose to the source by the appropriate electronic circuitrylocated in the electrode assembly. In passive electrodes, noamplification is performed close to the electrode. Instead, theelectrode material is connected directly to the sEMG amplifier,with a lead wire.Passive electrodes have been widely used in previous studies[7]. They are cheap, but generally more prone to noiseinterference, because the high impedance electrode signal istransmitted in lead wires, connecting the electrode to theamplifier.The interface between the skin and the electrode, canbe either wet or dry. With wet electrodes, gel is appliedbetween the skin and electrode to improve the stability andreduce the impedance of the electrode to skin interface [8].ith dry-contact electrodes, no gel is applied between theelectrode and skin. Thus, when using dry-contact electrodes,no skin preparation is needed [9], and dry-contact electrodesare therefore very suitable for prolonged measurements of bio-electric signals. Here, an active electrode design is typicallyrequired to handle the higher impedance of the electrode toskin interface when compared to wet electrodes.A major portion of the extensive research that has been doneon sEMG acquisition electrodes focus on active electrodes,but rarely have researchers considered the mechanical designand fastening methods of the electrode for continuous useover long periods of time. Generally, the quality of theacquired bio-signals are higher when using active electrodes,as compared to passive electrodes [10, 11]. Ribeiro et al.introduced a dry-contact, active flexible electrode, for wearablebio-signal recordings, with a novel interface material that ishighly bendable and comfortable for the wearer [12]. Theelectrode was designed to have the interface material depositeddirectly on the back of the flexible printed circuit (FPC) board.Tests conducted with this dry active flexible electrode, showedthat it had better electrical characteristics than the traditionalAg/AgCl electrode, namely less power line interference andbetter response in the signal band.Guerrero et al. studied a dry-contact electrode for acquiringmulti-channel sEMG signals using three parallel gold-platedrod electrodes fixed onto a printed circuit board (PCB) [13].The common-mode rejection ratio (CMRR) of each electrodewas boosted by an independent driven right leg (DRL) circuitto obtain measurements of a higher quality. The dry, rigidelectrode could accidentally detach from the skin more easilythan a wet or flexible electrode. Thus, the failure of oneelectrode could compromise the entire set of measurements.Studies have also been performed on active sensors wheresurface mounted components were directly attached to a textilescreen-printed circuit using polymer thick film techniques,for acquisition of ECG signals [14]. Merritt et al. connectedpassive electrodes to a buffer circuit screen-printed onto thefabric, to decrease the vulnerability of the signal to externalinterference. Although these electrodes were developed toadapt to the contours of the human body and acquire bio-signals of high quality; mechanical design considerations, likeease of attachment to the limb, were not considered.This paper presents a novel low-cost, flexible, active, dry-contact sEMG sensor with a mechanical design that was op-timized to reduce motion artifacts and enable easy attachmentto the forearm. The sensor was developed to be a part ofa wearable sEMG acquisition system, which enables multi-channel recording of low noise sEMG signals. The signalquality of the developed sensor was evaluated experimentallywith the real-time finger gesture recognition algorithm de-scribed by De Silva et al. [15]. In addition, the CMRR of thesensor was characterized to ensure that there are no substantialperformance limitation of the sensor due to its active circuitry. II. M ETHODS
This section is divided into two subsections. Section II-A describes the design of a wearable sEMG sensor andsection II-B describes the experimental setup for finger gesturerecognition and for sensor characterization.
A. Overview of the Wearable Surface EMG Sensor
For the design of the dry-contact sEMG sensor, the follow-ing factors were considered, to obtain high quality wearablerecordings [16]. • Characteristics of the electrode material and the amplifi-cation circuitry. • Attachment of the sensor, to obtain a stable electrode toskin interface. • Optimal placement of the sensors with minimal distanceto the active muscle areas of interest.To obtain a focal pickup area, a bipolar configurationwas chosen over a monopolar configuration. The sensor wasdesigned according to the SENIAM (Surface EMG for Non-Invasive Assessment of Muscles) standards [17] with thefollowing characteristics: • Shape of an Electrode: Circular • Size of an Electrode (Diameter): mm • Inter-electrode distance: mm • Material: Stainless steelFor bio-potential sensors, a key parameter to obtain highquality recordings, is a stable and low impedance electrode-skin interface. Unfortunately, dry-contact electrodes are knownto have very high electrode to skin impedances ( >
100 k Ω , i.e.fewM Ω s) [18]. However, with a stable skin contact, impedancevariations can be reduced and impedance mismatches betweenthe electrodes can thereby be minimized. Stainless steel hasdominating polarizable characteristics, and is therefore appro-priate for recording sEMG, where frequency content below Hz have little relevance [6]. Moreover, stainless steel is an inertand durable material and can therefore be reused numeroustimes, without significant degradation.The active circuitry of the sEMG sensor should have ahigh input impedance to accurately measure the sEMG sig-nal. A high input impedance is necessary to obtain a high
G=1 G=1G=1 R R R CCR t R t D tvs D tvs GuardGuard S u r f a c e o f t h e b o d y V ref V ref Out E E Out Q Q Q Q G=1
Fig. 1. Schematic of the active dry-contact sEMG sensor. The transientprotection circuit and the differential high-pass filter are highlighted.
MRR when there is a mismatch between the electrode-skinimpedances of the attached electrodes [19]. The bias currentand current noise flows through the source impedance (i.e.electrode-skin impedance), and must therefore be minimizedto avoid a significant DC offset and noise contribution to themeasured bio-potential.Fig. 1 shows the schematic for the active dry-contact sEMGsensor. The high impedance signals from the two electrodes, E and E , are buffered by Q and Q and the buffered signalsare used for active shielding of the electrodes. With activeshielding, the low output impedance of the buffer drives theshield of the electrode. Thereby, only a negligible potentialdifference appears between the electrode signal and the shield,causing the displacement current to be close to zero. Thisdesign optimally protects the high impedance electrode signalfrom electrical interference [19].The buffered signals were high-pass filtered by a differentialfilter with a cut-off frequency at Hz, to stabilize the baselineof the signal. The output of the differential filter was buffered,to obtain a low impedance differential output,
Out and Out ,of the electrode. The input of the buffers, Q and Q , wereprotected by a transient voltage suppressor (TVS) diode ateach input.The use of a unity gain buffer in the active circuitrymeans that the sensors are capable of being interfaced withcommercially available, low-cost, open-source developmentplatforms, such as the OpenBCI board [20], for accurate andlow-noise signal acquisition.The buffers, Q to Q , were implemented with the AD8244(Analog devices, Massachusetts, USA). The AD8244 is a quadbuffer with a high input impedance ( T Ω | pF), a low biascurrent ( pA), and a very low current noise ( . fA / √ Hz),which makes it suitable for buffering a high-impedance source.The TVS diodes were implemented with the TPD4E1B06(Texas Instruments, Texas, USA) diode array. This diode arraywas specifically chosen due to its small line capacitance, . pF, which is smaller than the pF input capacitance of thebuffer, ensuring that the input impedance of the sensor is ashigh as possible.The design was fabricated and assembled on a double sidedFPC, with a small form-factor. The cost of a single sensor was Stainless SteelTPU PadElectrode
Fig. 2. (Left) Top side of the FPC with the components soldered. (Right)Bottom side of the FPC with stainless steel electrodes and a TPU pad around. approximately 30 USD, including the stainless steel electrodesand the electronic components. Fig. 2 (Left) illustrates the topside of FPC with the components soldered, and Fig. 2 (Right)illustrates the bottom side of the FPC with the stainless steelelectrodes attached and a thermoplastic polyurethane (TPU)pad around them.Fig. 3 depicts the layout of the FPC sensor. The dimensionsof the sensor are indicated on the figure, together with the mostimportant design features, which are numbered and markedin dashed red. The sensor was designed such that it could beflexibly mounted onto the forearm using elastic straps, Fig. 3.3.All of the main electronic components were placed on theFPC surface covering the top side of one of the electrodes,Fig. 3.4, in order to utilize the semi-rigidity of this region ofthe FPC and to reduce the amount of traces crossing the narrowjunction between the two electrodes. Fig. 3.4 and Fig. 3.5 showthat all the components were placed laterally. This was doneto reduce the risk of damaging the soldered components whenbending the FPC around the forearm. In addition, the FPChas a narrow junction between the two electrodes, Fig. 3.7,to allow bending and twisting between the surfaces of thetwo stainless steel electrodes. The narrow junction of theFPC further allowed for slight rotations of the straps withoutaffecting the area of contact between the skin and electrodes. mm
12 374 6 5
Fig. 3. The FPC layout with indications of the important design features.1 - High impedance input is actively shielded by copper filled planes.2 - Vias added to connect the shield planes on both sides of the FPC.3 - Flexible straps for bending support were excluded from the copper fill.4 & 5 - Components were placed perpendicular to the direction of bending.6 - Shielded and plated through-hole to fix the screw to the electrode.7 - Narrow junction between the electrodes improve the flexibility towardsrotation and bending.
A mechanically stable contact between the electrode andskin is especially important when utilizing dry-contact elec-trodes, where a slight movement of the electrode can lead tosignificant variations in the impedance of the electrode to skininterface, resulting in motion related artifacts in the recordedata. To reduce these artifacts, a flexible pad, made of 3Dprinted TPU, was added around the electrodes, as shown inFig. 2 . The flexible pad increased the skin contact creating amore stable electrode to skin interface.
B. Experimental Setup
Considering that the sensor was designed to be used alongwith a bionic arm, it was evaluated with a finger gestureclassification experiment. In addition, the CMRR of the sensorwas experimentally characterized.
1) Real-Time Finger Gesture Classification:
The classifica-tion accuracy of the system based on the proposed sensor, wasevaluated by placing sensors on the skin surface directly abovethe relevant muscles pertaining to the evaluated gestures andobtaining multi-channel sEMG recordings from each subject.The experiment was approved by the Ethics Review Com-mittee at University of Moratuwa (Ethics Review Number:ERN/2019/007).
Fig. 4. The experimental setup for real-time finger gesture classification
In order to perform real-time finger gesture classifica-tion, four sensors were connected to a bio-potential am-plifier, consisting of an ADS1299EEGFE-PDK Evaluation(EVM) Board (Texas Instruments, Texas, USA), and a STM32NUCLEO-F411RE Development Board (STmicroelectronics,Texas, USA) as shown in Fig. 4. In addition, a Ag/AgCl wetelectrode was attached near the elbow to act as a bias electrode,connected to the Vref, where Vref was the mid-value of thesupply voltage. The sensors were mounted using elastic straps
Fig. 5. Finger gestures. Top row: In order from left to right; the hand in theneutral position, thumb flexion, index finger flexion. Bottom row: In orderfrom left to right; middle finger flexion, ring finger flexion, hand closure. and buckles, ensuring a stable and tight electrode-skin contactand a comfortable pressure on the forearm.The five gestures illustrated in Fig. 5 were included in theexperiment, as most commonly used gestures are a combina-tion of these gestures. Raw sEMG were recorded from fourhealthy subjects (2 males and 2 females, age: ± ), at asampling frequency of Hz, using four sensors placed atoptimal forearm positions according to Crepin et al. [5].During the data collection, the subjects were asked to per-form 20 repetitions of each gesture with their dominant hand.Each gesture was held for a period of 5 seconds, followedby a resting period of 5 seconds, during which the subjectswere asked to keep their hand in a relaxed neutral position.The classification algorithm uses temporal muscle activation(TMA) maps, and was trained and tested on an individualsubject basis. The data collection protocol and classificationalgorithm are detailed in [15].
2) Electrical Characterization of the Electrode:
All sEMGsignals are measured as differential signals between twoelectrodes. Therefore, common-mode signals, caused by e.g.electromagnetic interference, are considered as noise. Thus, ahigh CMRR of a bipolar sEMG sensor is crucial to obtain agood signal recording quality [21]. For the characterization ofthe sensor’s CMRR, a signal generator was connected directlyto the active sEMG sensor inputs, as shown in Fig. 6. Ag.HIamp amplifier (g.Tec Medical, Austria) was used to recordthe differential output from the developed sEMG sensor. E E DIFFERENTIAL MODE Signal Generator Active sEMGsensor+- 5VOut1Out2GNDVref 5V Vref GND g.HIampPowerSupplyVref E E COMMON MODE Signal Generator+- 5VOut1Out2GNDVref 5V Vref GND g.HIampPowerSupplyVref ++--GNDGNDActive sEMGsensor
Fig. 6. The experimental setup for the electrical CMRR characterization
The sensor was placed in an electro-magnetic interference(EMI) shielded box, with wires drawn out for the powersupply, inputs and outputs. The EMI shielded box was placedfar from other electronic devices and the mains power supplyto further reduce the noise interference. The sensor’s electricalcharacteristics were determined via laboratory experiments.The power spectral density (PSD) at the sensor output wasdetermined for both a differential-mode and a common-modesource, and the CMRR was the ratio between these PSDs asgiven in (1). The source was a signal generator, adjusted toa sine wave output of a fixed frequency, ω . The signals wererecorded for a fixed time period, and the recorded signals wereused to determine the CMRR of the sensor, as given by (1). M RR ( ω ) = 10 · log (cid:18) P SD d ( ω ) P SD c ( ω ) (cid:19) (1)where, CM RR ( ω ) is the CMRR at the signal generatorfrequency ω and P SD d ( ω ) and P SD c ( ω ) are the PSD ofsensor output at frequency ω for the differential-mode sourceand the common-mode source, respectively.III. R ESULTS AND D ISCUSSION
A. Real-Time Finger Gesture Classification
The results obtained from the experiment are summarizedin Fig. 7 and Table I. Fig. 7 shows the sEMG signals (black)obtained for flexion and extension of the hand closure gestureusing four sensors, along with the signal envelope used forreal-time finger gesture classification (dashed red). s E M G s i gn a l a m p lit ud e ( μ V ) time (ms) CH1CH2CH3CH4
Fig. 7. Sketch of sEMG signal profiles (black) and signal envelopes (dashedred), for flexion and extension of the hand closure gesture, using four sensors.
Table I reports the accuracies obtained for each fingergesture classification from the four test subjects. The sEMGsignals obtained from four sensors could be used to classifythe five finger gestures with an average accuracy of %. TABLE IC
LASSIFICATION ACCURACY (%)
Proposed sEMG Sensor A. De CrepinFinger Subject Avg Silva et et al.A a B C D al. [15] [5] . Thumb b
80 90 80 90 85 – . Index b
90 100 80 90 90 – . Middle 80 90 90 80 85 . Ring 70 80 90 90 82.5 . Hand – 80 80 80 80 a Subject was a pilot study, and therefore gesture 5 was not measured. b The classification of the finger gesture was not considered in [15] asthe muscles pertaining to the finger motion are not easily accessible bythe sEMG acquisition device used in the study.
Using the developed sEMG sensors, we were able to obtainclassification accuracies above 80%. The accuracies weregenerally lower than comparable results obtained by De Silvaet al. [15] and Crepin et al. [5]. The lower accuracies couldbe caused by a lower signal quality, which might be related to the biopotential amplifier that was used to perform the study.The biopotential amplifier was based on an evaluation board,making it difficult to obtain optimal wiring and shielding ofelectrode cables and traces on the PCB. Future studies shouldtest the developed electrode with a commercial grade EMGamplifier.Another aspect that might be possible to improve, is themounting of the sensor. The selected mounting method wasa trade-off between signal quality, usability, and ergonomics.With additional user testing, the sensor design can most likelybe improved to obtain a better stability of the electrode to skininterface, and thereby improving the signal quality.
B. Electrical Characterization of the Sensor
The results of the CMRR characterization are summarizedin Fig. 8. The sensor was tested in the frequency range from Hz to
Hz, corresponding to the sEMG signal bandwidth.
Fig. 8. The measured CMRR of the dry-contact sEMG sensor, compared tothe CMRR of the buffer gain mismatch, differential filter, and bio-potentialamplifier (g.HIamp).
The CMRR of the sensor is limited by: • The g.HIamp amplifier: The CMRR of the g.HIamp sys-tem was characterized by connecting the signal generatordirectly to the amplifier. • Differential high-pass filter: The bias resistor, R , ofthe differential high-pass filter had a value of 10 M Ω .Increasing the value of R improves the CMRR of thefilter, and thus the value was a trade-off between theoffset caused by the bias current and the degradationof the CMRR caused by the filter. For finger gestureclassification, it was noted that a M Ω resistor wassufficient to obtain good classification accuracies. • Gain mismatch: Gain mismatches between buffers of thedeveloped sensor deteriorate the CMRR, when the buffersare used to amplify a differential signal.Here, CMRR related to the gain mismatch and the differ-ential filter were theoretical values. The gain mismatch wasobtained from the datasheet of the buffer, whereas CMRR ofthe differential filter was based on equations derived by Casaset al. [22].he CMRR obtained from the characterization experimentwas sufficient to have an accurate classification of the fiveselected finger gestures. From Hz to approximately Hz,the CMRR was limited by the differential high-pass filter.From approximately Hz and above, there is a degradationin the measured CMRR due to the bio-potential amplifier,g.HIamp.
C. Mechanical Design of the Sensor
Prior to the development of the FPC based sensor, analternative approach based on a rigid PCB with TPU padswas also tested. Although the rigid PCB costs less than FPC,the inability of the sensor to bend along the contour of theforearm often resulted in one or both of the electrodes easilylosing contact with the skin, which made the sEMG recordingsunstable.Moreover, slight movements of the arm resulted in motionartifacts in the observed signals. When one or both electrodesloses contact with the skin, there will be changes in theelectrode-skin interface, which can result in very high sourceimpedances and impedance mismatches. Thus, the FPC basedsensor was chosen over the rigid PCB based sensor.IV. C
ONCLUSION
In this paper, a novel low-cost, active, dry-contact surfaceEMG sensor was introduced for use along with bionic arms.The sensor uses stainless steel electrodes to acquire signalsfrom the surface of the skin. The proposed sensors can beinterfaced with a variety of amplifiers, and this satisfies theneed for a fully wearable, low-cost sEMG acquisition system,for prostheses development.The developed sensor was evaluated for its ability to acquirehigh quality signals, related to five selected finger gestures,that could be classified in real-time. An average classificationaccuracy of % was obtained, using four sensors placedon the skin above the muscles corresponding to the selectedgestures.The CMRR of the sensor was determined to ensure that thesensor’s performance was not substantially limited by its activecircuitry, but rather by other external factors. The CMRR forthe developed sensor was measured to an average of dB.The classification accuracy obtained using four sensorsshows that the proposed active, dry-contact sEMG sensors canbe used to obtain high quality, distinct signals for each fingergesture. The flexible sEMG sensors, along with optimal sensorplacement, can be used to obtain high accuracy individualfinger control for a bionic arm at a low cost.Looking forward, it might be possible to further improvethe sensor design to obtain a better stability of the electrode toskin interface, and thereby a better signal quality. The numberof subjects should be increased to experimentally determinea more generalised method of mounting, in order to furtherimprove the classification accuracies. Future work would alsoinclude the development of a fully wearable sEMG acquisitionsystem, with the ability to accurately classify finger gesturesand translate it to a bionic arm in real-time. A CKNOWLEDGMENT
Authors extend their gratitude to the Bionics Laboratory ofDept. of Mechanical Eng. at the University of Moratuwa.R
EFERENCES[1] M. LeBlanc, ENGR 110, Class Lecture, Topic: Give Hope - Give a Hand,Rotary LN-4 Prosthetic Hand Project. University of Stanford, Stanford,California, November 9, 2008.[2] D. W. Braza and J. N. Yacub Martin, Chapter 119: Upper LimbAmputations, Essentials of Physical Medicine and Rehabilitation (FourthEdition), 2020, pp. 651-657.[3] D. van der Riet, R. Stopforth, G. Bright, and O. Diegel, ”An overviewand comparison of upper limb prosthetics,” IEEE 2013 Africon Confer-ence, 2013, pp. 18.[4] S. Benatti, B. Milosevicy, F. Casamassima, P. Schonle, P. Bunjakuz, S.Fatehz, Q. Huangz, and L. Benini, ”EMG-based hand gesture recognitionwith flexible analog front end,” Proc. IEEE 2014 Biomedical Circuitsand System Conference, BioCAS 2014, pp. 57-60.[5] R. Crepin, C. L. Fall, Q. Mascret, C. Gosselin, A. Campeau-Lecours,and B. Gosselin, ”Real-time hand motion using sEMG patterns clas-sification,” Proc. IEEE Engineering in Medicine and Biology SocietyConference, 2018.[6] J. G. Webster, Medical instrumentation: application and design, Fourthedition, John Wiley & Sons, Hoboken, NJ, 2010.[7] Md. Tariquzzaman, F. Khanam, Md. Sohag, and M. Ahmad, ”Design andimplementation of a low-cost multichannel rectified EMG acquisitionsystem,” 19th International Conference on Computer and InformationTechnology (ICCIT), December 2016, pp. 261-265.[8] S. L. Kappel, and P. Kidmose, ”Study of impedance spectra for dry andwet EarEEG electrodes,” Proc. Annual International Conference of theIEEE Engineering in Medicine and Biology Society, EMBS, 2015, Vol.2015, November, pp. 31613164.[9] P. Laferriere, E. D. Lemaire, and A. D. C. Chan, ”Surface elec-tromyographic signals using dry electrodes,” IEEE Transactions onInstrumentation and Measurement, Vol. 60, No. 10, pp. 3259-3268, Oct.2011.[10] O. T. Inan and G. T. A. Kovacs, ”An 11 µµ