Design and implementation of a myoelectric control system for a printable robotic hand (PDF Download Available)
5.05University Carlos III de MadridAbstractIn this project, the design and implementation of a myoelectric control system to control a robotic hand is presented. The aim of the project is to serve as a basis for the development of a low-cost myoelectric prosthesis.
The myoelectric control system uses myoelectric signals to generate control commands that can be used to control several kinds of devices. A myoelectric signal is an electric biological signal generated by the motor neurons connected to the muscle fibers during the contraction of a muscle. To measure these signals so as to use them in the control system, an electromyography (EMG) circuit has been designed. The EMG circuit detects the signal with two electrodes, and then it filters and amplifies the signal. Two units of the EMG circuit have been manufactured, to have two control channels. The myoelectric control system has been implemented in an Arduino, an open source electronic platform. The Arduino digitizes the signal, processes it and generates the control commands. Finally, to test the control system, a three-finger robotic hand has been designed. The robotic hand has been manufactured using a low-cost 3D printer.Discover the world's research14+ million members100+ million publications700k+ research projects
UNIVERSIDAD CARLOS III DE MADRIDDEPARTAMENTO DE INGENIER?IA DE SISTEMAS Y AUTOM?ATICATESIS DE M?ASTERDESIGN AND IMPLEMENTATION OF A MYOELECTRICCONTROL SYSTEM FOR A PRINTABLE ROBOTIC HANDAutor:?Alvaro VillosladaDirector: Mohamed AbderrahimM?ASTER OFICIAL ENROB?OTICA Y AUTOMATIZACI?ONLEGAN?ES, MADRIDJULIO 2012
UNIVERSIDAD CARLOS III DE MADRIDMASTER OFICIAL EN ROB?OTICA Y AUTOMATIZACI?ONEl tribunal aprueba la tesis de Master titulada “DESIGNAND IMPLEMENTATION OF A MYOELECTRIC CONTROLSYSTEM FOR A PRINTABLE ROBOTIC HAND” realizado por?Alvaro Villoslada.Fecha: Julio 2012Tribunal:Francisco Jos?e Rodr??guez UrbanoAlberto Brunete Gonz?alezLuis Hern?andez Corporales
ContentsList of Tables viiList of Figures xAgradecimientos xviiAbstract xixResumen xxi1 Introduction 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Types of prostheses . . . . . . . . . . . . . . . . . . . . . . 31.1.1.1 Cosmetic prostheses . . . . . . . . . . . . . . . . 31.1.1.2 Body-powered prostheses . . . . . . . . . . . . . 41.1.1.3 Electrically powered prostheses . . . . . . . . . . 61.1.2 Myoelectric control systems . . . . . . . . . . . . . . . . . . 151.1.2.1 Non-pattern recognition-based myoelectric control 171.1.2.2 Pattern recognition-based myoelectric control . . 191.1.3 EMG circuits . . . . . . . . . . . . . . . . . . . . . . . . . . 211.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23v
2 Design and implementation 252.1 EMG circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.1.1 Differential electrode configuration . . . . . . . . . . . . . . 272.1.2 Preamplification stage (active electrode) . . . . . . . . . . . 272.1.3 Common mode voltage feedback (elbow drive) . . . . . . . 312.1.4 Filtering stage . . . . . . . . . . . . . . . . . . . . . . . . . 332.1.5 Amplification stage . . . . . . . . . . . . . . . . . . . . . . . 362.1.6 Precision power stage . . . . . . . . . . . . . . . . . . . . . 372.1.7 Electrical isolation and user protection . . . . . . . . . . . . 382.1.8 PCB design . . . . . . . . . . . . . . . . . . . . . . . . . . . 392.2 Myoelectric control system . . . . . . . . . . . . . . . . . . . . . . 402.2.1 Analog-to-digital conversion . . . . . . . . . . . . . . . . . . 412.2.2 Arduino control software . . . . . . . . . . . . . . . . . . . . 422.2.2.1 Zero level setting . . . . . . . . . . . . . . . . . . . 422.2.2.2 Activation threshold setting . . . . . . . . . . . . . 432.2.2.3 Hand control . . . . . . . . . . . . . . . . . . . . . 452.2.2.4 Data storage . . . . . . . . . . . . . . . . . . . . . 462.2.3 Interface shield . . . . . . . . . . . . . . . . . . . . . . . . . 462.3 Robotic hand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472.3.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472.3.1.1 Fingers . . . . . . . . . . . . . . . . . . . . . . . . 482.3.1.2 Gear set and servo motor . . . . . . . . . . . . . . 512.3.1.3 Casing . . . . . . . . . . . . . . . . . . . . . . . . 532.3.2 Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . 543 Experimental tests 573.1 EMG circuit tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.1.1 Frequency spectrum . . . . . . . . . . . . . . . . . . . . . . 583.1.2 CMRR measurement . . . . . . . . . . . . . . . . . . . . . 603.1.3 SNR measurement . . . . . . . . . . . . . . . . . . . . . . . 61
3.1.4 Common mode voltage feedback . . . . . . . . . . . . . . . 623.2 MCS tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643.2.1 Digital signal processing . . . . . . . . . . . . . . . . . . . . 643.2.2 Generation of the control signals . . . . . . . . . . . . . . . 663.2.3 Servo noise . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.3 Robotic hand performance . . . . . . . . . . . . . . . . . . . . . . 694 Conclusions 75Bibliography 81A Project budget 87B Layouts of the designed circuits 89B.1 EMG circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89B.2 Interface shield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
List of Tablesix
List of Figures1.1 Prosthetic hand used in the Middle Ages and modern hand cosme-sis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2 Current and old body powered prostheses . . . . . . . . . . . . . . 51.3 Switch controlled prosthesis . . . . . . . . . . . . . . . . . . . . . . 71.4 Myoelectric signal formed by the superposition of several MUAPTs 91.5 iLimb Pulse prosthetic hand . . . . . . . . . . . . . . . . . . . . . . 121.6 Otto Bock MyoHand . . . . . . . . . . . . . . . . . . . . . . . . . . 141.7 Block diagram for a pattern recognition-based MCS . . . . . . . . 202.1 Active electrode schematic . . . . . . . . . . . . . . . . . . . . . . 282.2 Active electrode PCB design . . . . . . . . . . . . . . . . . . . . . 302.3 Common mode bias and noise cancellation with the RLD circuit . . 312.4 Elbow drive schematic . . . . . . . . . . . . . . . . . . . . . . . . . 322.5 Comparison of the gain response of different types of filters . . . . 342.6 High pass and low pass filters . . . . . . . . . . . . . . . . . . . . . 352.7 Amplification stage . . . . . . . . . . . . . . . . . . . . . . . . . . . 372.8 Power section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382.9 EMG circuit schematic . . . . . . . . . . . . . . . . . . . . . . . . . 392.10 EMG circuit prototype . . . . . . . . . . . . . . . . . . . . . . . . . 402.11 Arduino SD card shield . . . . . . . . . . . . . . . . . . . . . . . . 46xi
2.12 Interface shield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472.13 Grip types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492.14 Dimensions of the upper finger and its 3D design . . . . . . . . . . 492.15 Dimensions of the lower finger and its 3D design . . . . . . . . . . 502.16 Finger slits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502.17 Nut hole for the set screw . . . . . . . . . . . . . . . . . . . . . . . 512.18 Gear set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522.19 Casing parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532.20 Fused deposition modeling operating principle . . . . . . . . . . . 542.21 Detail of the robotic hand fingers . . . . . . . . . . . . . . . . . . . 552.22 Detail of the set screw of a gear and the gear set . . . . . . . . . . 552.23 Right part of the casing . . . . . . . . . . . . . . . . . . . . . . . . 562.24 Assembled robotic hand . . . . . . . . . . . . . . . . . . . . . . . . 563.1 Typical frequency spectrum of a myoelectric signal . . . . . . . . . 583.2 Frequency spectrum of the myoelectric signal from muscle longuspalmaris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.3 Frecuency spectrum of the myoelectric signal from muscle bicepsbrachii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.4 Noise frequency spectrum . . . . . . . . . . . . . . . . . . . . . . . 593.5 Configuration for measuring the CMRR . . . . . . . . . . . . . . . 603.6 Two signal and noise samples used for the calculation of the SNR 623.7 SNR values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623.8 Myoelectric and noise signals, and signal and noise frequencyspectra, acquired without the ED circuit . . . . . . . . . . . . . . . 633.9 Raw myoelectric signal . . . . . . . . . . . . . . . . . . . . . . . . . 643.10 Myoelectric signal at the zero volt level . . . . . . . . . . . . . . . . 653.11 Rectified myoelectric signal . . . . . . . . . . . . . . . . . . . . . . 653.12 Smoothed myoelectric signal . . . . . . . . . . . . . . . . . . . . . 653.13 Threshold set at 0.067 . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.14 Threshold set at 0.12 . . . . . . . . . . . . . . . . . . . . . . . . . . 663.15 Generation of the control signals . . . . . . . . . . . . . . . . . . . 673.16 Myoelectric signals with noise induced by the servo motor . . . . . 683.17 Control errors caused by servo motor noise . . . . . . . . . . . . . 683.18 Power supply for the servo motors . . . . . . . . . . . . . . . . . . 693.19 Grasping a soft object . . . . . . . . . . . . . . . . . . . . . . . . . 703.20 Grasping a can . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703.21 Grasping an apple . . . . . . . . . . . . . . . . . . . . . . . . . . . 713.22 Grasping a Rubik’s cube . . . . . . . . . . . . . . . . . . . . . . . . 713.23 Grasping a thin object: a tin welder . . . . . . . . . . . . . . . . . . 723.24 Grasping and holding a filled bottle . . . . . . . . . . . . . . . . . . 723.25 Grasping and holding a box by its handle . . . . . . . . . . . . . . 73B.1 Top layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89B.2 Bottom layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90B.3 Top layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91B.4 Bottom layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Cyborgs are not less human, as they can use all their human power.
AgradecimientosEn primer lugar quiero agradecer a mi director de Tesis, Mohamed Abderrahim,el haberme permitido desarrollar este proyecto, y el haber podido hacerlo contotal libertad. Aunque podr??a haberme encomendado alguno de los proyectos enlos que ya est?a trabajando, me dio v??a libre para trabajar en mi propia propuesta,siempre pensado en mi futuro como investigador.Tambi?en quiero expresar mi gratitud hacia mi compa~nero y amigo Alejan-dro Escalante, quien ha estado trabajando conmigo en el dise~no de la manorob?otica. Sin su gran experiencia en el manejo de SolidWorks y sin sus ideas yconsejos de dise~no, este proyecto no ser??a lo que al final ha sido.A los t?ecnicos de laboratorio,?Angela y Fernando, quiero darles mis m?assentidas gracias. Siempre est?an ah?? cuando necesitas un componente, unaherramienta o solucionar cualquier problema. Y lo hacen siempre con toda suamabilidad y simpat??a (y con buena m?usica).Finalmente quiero agradecer a todos los compa~neros del Robotics Lab queme han apoyado, dado consejos, consolado, que se han tomado ca~nas y quehan perdido su tiempo haciendo descansos para fumar y desconectar. En defini-tiva, a todos los que hacen que el Robotics Lab sea un gran lugar humano paratrabajar.xvii
AbstractIn this project, the design and implementation of a myoelectric control system tocontrol a robotic hand is presented. The aim of the project is to serve as a basisfor the development of a low-cost myoelectric prosthesis.The myoelectric control system uses myoelectric signals to generate controlcommands that can be used to control several kinds of devices. A myoelectricsignal is an electric biological signal generated by the motor neurons connectedto the muscle fibers during the contraction of a muscle. To measure these signalsso as to use them in the control system, an electromyography (EMG) circuit hasbeen designed. The EMG circuit detects the signal with two electrodes, andthen it filters and amplifies the signal. Two units of the EMG circuit have beenmanufactured, to have two control channels. The myoelectric control systemhas been implemented in an Arduino, an open source electronic platform. TheArduino digitizes the signal, processes it and generates the control commands.Finally, to test the control system, a three-finger robotic hand has been designed.The robotic hand has been manufactured using a low-cost 3D printer.xix
ResumenEn el presente proyecto se presenta el dise~no e implementaci?on de un sistemade control mioel?ectrico para el control de una mano rob?otica. El objetivo delproyecto es servir como base para el desarrollo de una pr?otesis mioel?ectrica debajo coste.El sistema de control mioel?ectrico utiliza las se~nales mioel?ectricas para generarcomandos de control que pueden ser utilizados para controlar varios tipos dedispositivos. Una se~nal mioel?ectrica es una se~nal el?ectrica biol?ogica generadapor las neuronas motoras conectadas a las fibras musculares durante la con-tracci?on de un m?usculo. Para medir estas se~nales con el fin de utilizarlas enel sistema de control, se ha dise~nado un circuito de electromiograf??a (EMG).El circuito de EMG detecta la se~nal mediante dos electrodos, y despu?es filtray amplifica dicha se~nal. Se han fabricado dos unidades del circuito de EMG,para tener dos canales de control. El sistema de control mioel?ectrico se ha im-plementado en un Arduino, una plataforma electr?onica open source. El Arduinodigitaliza la se~nal, la procesa y genera los comandos de control. Por?ultimo, paraprobar el sistema de control, se ha dise~nado una mano rob?otica de tres dedos.La mano rob?otica se ha fabricado mediante una impresora 3D de bajo coste.xxi
Chapter 1IntroductionThe human hand is the most versatile tool we use in our daily lives. It is a highlydexterous organ that gives us a wide range of manipulation capabilities: its largenumber of degrees of freedom allows us to perform many tasks. Besides mak-ing us able to manipulate, operate or deform objects, its sensory ability allows usto, among other things, identify objects just by touch and shape, without seeingthem. Our hands have played a major role in our own evolution and the develop-ment of our intelligence. They were our very first tools, and with them, we madeour first artificial tools. Art (music, painting, writing) would be nothing withoutour hands. We give love and comfort with them. The hands are essential in theway we interact with our world as they are involved in almost every action weperform.1.1 MotivationPeople suffering a limb amputation are forced to face their daily life tasks with thedisadvantage of not having all their limbs. In the case of upper limb amputations,not having one or both hands is a major barrier in carrying out the daily tasksfor those who suffer the amputation. Actions as simple as getting dressed, tying
2 Introductionshoelaces or pouring water into a glass, have an added difficulty which restrictsthe autonomy and independence of the amputee.The problem of limb amputations is specially serious in developing countries.There are several factors that increase the risk of suffering a limb amputation inthese countries. War is present in a great number of developing countries: thereare so many armed conflicts that is difficult to have an statistic on how manyamputees are in these countries. A person can lose his limb not only as a com-batant, but also because of the dangers derived from war, such as land minesburied in conflict zones that can be activated by anyone. Another important fac-tor are the safety measures at work. In sectors such as the industrial or theagricultural, defective and obsolete machinery is used, which increases the riskof suffering an accident. Besides the state of the machinery, factors like the lackof training of workers and supervisors, or the scarcity of resources invested inrisk prevention, increase even more the probabilities of suffering an amputation.To all this is added the inadequate health care system existing in developingcountries, which makes that treatable diseases like diabetes, or curable infec-tions, can worsen and end in problems so severe that the only possible solutionis the amputation of the affected limb. This situation makes the number of am-putees in developing countries increase alarmingly. Thus, is important to offerthese people a tool that allow them to retake their normal lives.Given this scenario, there is a clear need for a tool to partially restore thefunctionality of the missing upper limb. That is why researchers and companiesaround the world have developed prostheses that help these people on livingtheir lives in a more independent and simpler way. This can be achieved throughthree types of prostheses: cosmetic prostheses, body-powered prostheses andelectrically powered prostheses. In the next section, the characteristics of thesethree types of prostheses will be reviewed, pointing out their advantages anddisadvantages.
1.1 Motivation 31.1.1 Types of prosthesesThe different types of upper limb prostheses can be classified according to theirfunctionality and actuation system. At the lower functionality end, there are thecosmetic prostheses, prostheses which replicate the appearance of the missinglimb but have no additional functionality. Somewhat more practical are the body-powered prostheses, prostheses that have no motors or other actuators and areactuated by the user’s force. These prostheses are ahead of cosmetic prosthe-ses in functionality, as they can perform simple manipulation tasks. However,this type of manual actuation is not entirely intuitive, because it requires the userto use his/her good hand or other body parts, like the shoulders or chest, tooperate the prosthesis. Seeking to resolve this problem, there is a third type ofprostheses: electrically powered prostheses. This kind of prostheses are actu-ated by electric motors, freeing the user from exerting the actuation force. Theyhave two main types of control: by switches or by myoelectric signals, bioelectricsignals generated by muscle contractions. Among these three types, electri-cally powered prostheses, and more specifically myoelectric prostheses, are themost versatile, intuitive and comfortable, but they are also the most expensivebecause of the technology they use.1.1.1.1 Cosmetic prosthesesCosmetic prostheses, or cosmeses, are the oldest type of prostheses. Themost ancient prosthesis found to date is a wooden toe from the Egyptian NewKingdom era (from around 1000 BC). Prostheses were also used in the ancientGreece and in the Roman Empire. In the Middle Ages, knights wore prosthesesto hold their shields. Richer knights used more advanced prostheses, like themechanical arm of G¨otz von Berlichingen. The hand was actuated by a systemof catches and springs, and could hold different objects, like a sword or a featherpen [1].
4 IntroductionFigure 1.1: Prosthetic hand used in the Middle Ages and modern hand cosmesisThe function of cosmeses is to replace the missing limb only in appearance,not in functionality. Thanks to the advances in the materials used in their man-ufacture, it is now possible to find prostheses that realistically reproduce theveins, hair or wrinkles of the affected limb. Although there are prostheses thatallow the user to manipulate objects, there are people who consider the activerole of the prosthesis to be much less important than its appearance, simplicityand comfort.1.1.1.2 Body-powered prosthesesFunctional prostheses, with some manipulation capabilities, date back to thesixteenth century. Body-powered prostheses used nowadays differ little fromthe APRL hand designed after World War II, as can be seen in figure 1.2. Thistype of prostheses are designed with usability and function as a central pur-pose. They can carry out simple manipulation tasks, such as holding objects,lifting moderate weights or typing. The term body powered refers to the fact thatthe force to operate such components comes from mechanical transmission ofmuscular effort generated elsewhere in the body, remote from the amputationsite [2].A below-elbow body-powered prosthesis is composed by a plastic socketwhere the amputee’s remaining limb is inserted, a wrist joint to fix the terminal
1.1 Motivation 5device to the prosthesis, and the terminal device which can be a hook or a hand.The most common way of actuating these devices is by a steel cable whichgoes from a shoulder harness to the terminal device. The shoulder harness isplaced on the shoulder opposite to the amputated limb. When the user movesthe shoulder in certain ways, the cable tightens or loosens, and the terminaldevice opens or closes. Due to this actuation mechanism, the terminal deviceusually can only perform the opening/closing movement.Figure 1.2: Current and old body powered prosthesesBody-powered prostheses are the most used type of prosthesis due to threemain reasons: they are cheap, lightweight and they have a simple mechanicaldesign which makes them very reliable.However, they have some important disadvantages too. The actuation mech-anism is uncomfortable for the user: the shoulder harness used to control theterminal device restricts the movements the amputee can perform with that partof his body. Another major drawback of body powered prostheses is that they re-quire the user to make great efforts to operate the device. Therefore, prolongeduse of these prostheses can be tiring. In the case of high-level amputations, thiskind of actuation is not suitable as the amputee may find it impossible to gener-ate enough motion or strength. The artificial appearance of some body-powered
6 Introductiondevices is also a problem for some amputees.1.1.1.3 Electrically powered prosthesesThe last type of prostheses are externally powered prostheses. As the mostused actuators in this type of prostheses are electrical motors, they are alsocalled electrically powered prostheses. During the last years, the number ofamputees using these prostheses has increased considerably. This is becauseelectrically powered prostheses are the most versatile type of prostheses. Theyhave manipulation capabilities, in some cases exceeding the manipulation capa-bilities of body-powered prostheses, and thanks to the electrical actuators theycan exert more force than body-powered prostheses, with much less effort fromthe user.There are several types of control interfaces for electrically powered prosthe-ses, depending on the control signal used. These signals can be classified intobiomechanical and bioelectric signals [3]:1. Biomechanical(a) Motion of a part of the body, e.g., shoulder movement, chest expan-sion, muscle bulge, skin movement.(b) Change in properties of a body component as the result of mechanicalactivity, e.g., hardness of muscle, electrical resistivity of muscle.(c) Voice and air-flow control, e.g., exhaling and inhaling, humming, whistling,speaking.2. Bioelectric(a) Electromyogram (EMG): muscle electric signals.(b) Electroencephalogram (EEG): brain electric signals.(c) Electroneurogram: electric signals directly from nerves.
1.1 Motivation 7(d) Electrooculogram: electric signals derived from the eyes.Body motion, and more specifically shoulder motion, is the most commoncontrol source in the area of biomechanical signals. To detect body motion andturn it into electrical signals suitable to control a device, there are many kinds oftransducers. Among them, three are used in commercial prostheses [4]. Theseare: mechanical switches that require both force and muscle displacement toturn them on or off, pressure-sensitive transducers that change their impedancewith force applied but with essentially no displacement, and displacement trans-ducers that measure distance but with essentially no force required.Among these three types, mechanical switches, and more specifically push-buttons, are the most frequently used. An electrically powered prosthesis con-trolled by a push-button system can be easily operated by pressing the trans-ducers through body movements. Switches have the advantages of being easyto use, simple and inexpensive. However, switch control is not always feasible,depending on the degree of amputation of the user. In addition, there are timeswhen the switches are moved from their optimal zone (due to movements of theprosthesis with respect to the arm) and the amputee has to make greater effortsto activate them.Figure 1.3: Switch controlled prosthesis
8 IntroductionIn the case of bioelectric signals, the most widely used signals in prostheticcontrol are the myoelectric signals, although neuroelectric signals are gainingmore prominence. Myoelectric signals are bioelectric signals generated by mus-cle contractions. In the following section, the physiology of the myoelectric signalwill be briefly explained.The myoelectric signalThe basic functional unit in the neural control of the muscular contraction pro-cess is the motor unit. It is composed of an α-motoneuron and all the musclefibers that are innervated by the multiple branches of the motoneuron’s axon[5]. When the motoneuron activates its associated muscle fibers, a small elec-tric signal is generated: the motor unit action potential (MUAP). This signal isthe fundamental component of the myoelectric signal. MUAPs are the result ofa depolarization and polarization process caused by an ionic flow across themembrane of the muscle fibers. Starting from the motoneuron end plates, theaction potential spreads along the muscle fiber through a tubular system. Thisexcitation leads to the release of calcium ions in the intra-cellular space. Linkedchemical processes (electro-mechanical coupling) finally produce a shorteningof the contractile elements of the muscle cell, and thus, the muscle contraction.This process is called ”excitation-contraction coupling” [6].To maintain a muscle contraction, the nervous system has to continuallysend activation signals to the motor units. This continuous activation of mo-tor units generates a sequence of MUAPs, called motor unit action potentialtrain (MUAPT) [5]. The myoelectric signal measured on the surface of the skin isformed by the superposition of all MUAPTs under the detection zone (figure 1.4).
1.1 Motivation 9Figure 1.4: Myoelectric signal formed by the superposition of several MUAPTsThere are two key factors affecting the magnitude and density of the myo-electric signal: the recruitment of MUAPs and their firing frequency [7]. Theseare the main mechanisms that control the contraction process and modulate theforce output of the muscles. Because of this two factors, the raw (unprocessed)myoelectric signal has a random nature, meaning that the same movement willnever have the same signal associated. The actual set of recruited motor unitsconstantly changes within the matrix/diameter of available motor units, and thefiring rates of single motor units are randomly distributed with a firing frequencyin the order of ten per second [8].The myoelectric signal has an amplitude range varying from -5 to 5mV or0 to 1.5mV(rms), with a mean value of 0V. The usable energy of the signal islimited to the 0 to 500Hz frequency range, with the dominant energy being in the50-150Hz range [9].The use of electrodes is necessary to measure the myoelectric signal. Twomain types of electrodes are used to detect the myoelectric signal: invasive andnoninvasive [5]. The first ones measure the myoelectric signals directly fromthe user’s nervous system, using needle or wire electrodes implanted into themuscles. This kind of electrodes deliver high quality signals because they aregathered exactly from the right spot and they are very little affected by noise. Thesmall detection area allows the detection of individual MUAPs during relatively
10 Introductionlow force contractions. The use of needle electrodes is painful, because theyconsist of a cannula with one or more wires inside, and the cannula must remaininside the muscle throughout the duration of the measurements. Wire electrodesare less painful as they have a diameter of 25-100uV and the cannula is onlyused to insert them into the muscle. The maintenance of these electrodes is verydelicate, and their invasive nature makes them not suitable for their use in thecontrol of myoelectric prostheses. Invasive electrodes are used in exploratoryclinical EMG, in kinesiological and neurophysiological studies of deep muscles,to study MUAP characteristics, or to study control properties of motor units (firingrate, recruitment, and others).On the other hand, noninvasive electrodes are easier to use and maintainand they do not require painful procedures to be used. Noninvasive electrodesare also called surface electrodes, because the myoelectric signal is measuredon the surface of the skin. For this reason, it is impossible to measure individualMUAPs; the measured signal is the sum of all MUAPTs generated under theelectrodes. Surface electrodes are classified in two categories: passive andactive electrodes.Passive electrodes consist of a conductive detection surface, typically anAg/AgCl disc, surrounded by an adhesive surface to attach them to the surfaceof the skin. Usually, a conductive gel is used between the electrode and skin,to improve electrical contact. Removing the dead cells and protective oils fromthe surface of the skin also improves conductivity. This is done by light abrasionof the skin, using alcohol or other specialized substances. Active surface elec-trodes eliminate the need of skin preparation and conductive
that is why theyare often called dry electrodes. Active electrodes have some of the electronicsrequired to process the myoelectric signal in the same housing as the detec-tion surfaces. The key element of an active electrode is a high input impedanceamplifier.
1.1 Motivation 11Surface electrodes have some disadvantages:1. Non-selective detection of the signal: as said before, surface electrodescan be used effectively only with superficial muscles and they cannot beused to detect signals selectively from small muscles.2. Physiological cross-talk: as the detection area covers all the underlyingmuscles, those muscles in the vicinity of the muscle of interest may pro-duce myoelectric signals that are also detected by the electrodes. Thisphenomenon is known as ”cross-talk”.3. Electrical noise: the signal acquired by noninvasive electrodes has to befiltered and conditioned because, being measured on the surface of theskin, it is much more affected by different kinds of electric noise.4. Tissue characteristics: the electrical conductivity of the human body varieswith tissue type, thickness of the subcutaneous fat tissue, physiologicalchanges, sweat and temperature.5. Changes between the muscle belly and electrode site: the myoelectricsignal changes with the variation of the distance between signal origin anddetection site. This variation may be caused by dynamic contractions or byexternal pressure.Unlike invasive electrodes, surface electrodes are used to interface a personwith an external device, for example an electrically powered prosthesis. Theirother uses include the study of time-force relationship of myoelectric signals, ki-nesiological and neurophysiological studies of surface muscles, or psychophysi-ological studies. They are also used with children in the same cases as invasiveelectrodes, because they are reluctant to the use of implantable electrodes.
12 IntroductionMyoelectric prosthesesThe use of myoelectric prosthesis has been increasing over recent years. How-ever, the technology in which these prostheses are based exists since 1944,when the first myoelectric control system was built in Germany [10]. In thosedays, the transistor was not inv the system was designed with vacuumtube technology. For this reason, the system was big and heavy, and hence,impractical for its use as a prostheses controller. The first transistor-based my-oelectric system was developed in the USSR [11]. Its dimensions and weightmade possible its use in prostheses control. Since then, and thanks to advancesin technology, several commercial prostheses have been developed, such as the”Otto Bock MyoHand” [12] and the ”iLimb Pulse” [13].Figure 1.5: iLimb Pulse prosthetic handMyoelectric signal as control signal is so broadly used in the field of pros-thetic devices because it has some advantages over other control signals [14]:the user is freed of straps and har the signal is noninvasively detectedon the surface of the
the muscle activity which provides the control signalis relatively small and can resemble the effort required from an intact
thesignal can be adapted for proportional speed or force control in a relatively sim-ple w the electronic circuits that acquire, filter and process the signal can becontinuously improved and miniaturized, allowing the design of less bulky, lighter
1.1 Motivation 13and more
and with the increasing speed of microprocessors,more complex algorithms that allow the prosthesis to perform more complexfunctions can be used.Another great advantage of myoelectric control is that the signal is generatedby voluntary muscle action. This means that the myoelectric prosthesis will onlywork when the amputee generates the control signal, by voluntary muscle action.Such a system is immune to influence from external forces, prosthesis location,or body position/motion, problems that other types of prostheses have. Theprosthesis should neither be affected by the influence of environmental electricalnoise [4].Despite these advantages, myoelectric control has some drawbacks thatmust be overcome in order to be used in commercial applications [15]. Oneof the major weaknesses is that the interface between the user and the device isnot bidirectional, that is, there is no sensory feedback (except visual feedback)to help the user to adjust the muscle contraction in fine control. Another prob-lem with myoelectric control is the inability to individually control some muscles,even on intact limbs. Achieving individual control of all muscles requires lots ofexercise, and sometimes is impractical. This makes it very difficult to accomplishfine control in multifunction devices. In addition, there is the problem caused bythe need to concentrate and continuously physically react during manipulation,which would affect the daily life of the user.One of the biggest disadvantages of myoelectric prostheses is their price.Because of their mechanical and electronic complexity, they are more expen-sive than the other types of prostheses. The cost of a commercial myoelectricprostheses can reach up to 20000e. This problem is exacerbated in the case ofdeveloping countries, since the purchasing capabilities of their inhabitants is farfrom what is needed to acquire one of these prostheses.Thanks to myoelectric control interfaces, more functional prostheses can bedesigned. It is possible to control terminal devices able to perform different types
14 Introductionof grasps, and other joints such as wrists or elbows. However, there is a majorconstraint when designing terminal devices for prosthetic applications. The me-chanical and sensory complexity of the human hand makes it very difficult toreplicate. In the case of prosthetic hands, the number of degrees of freedomis one of the major constraints, because replicating a large number of degreesof freedom would make the hand very bulky and heavy due to the required ac-tuators [16]. For this reason, current commercial myoelectric prostheses havevery limited functionality, usually with one or two active DOFs, in order to keepthe device small and lightweight. These prostheses replicate the appearanceof the human hand, but not its functionality: they act like simple grippers [17].This is the case of one of the most used myoelectric prostheses, the MyoHandfrom Otto Bock, that is simply a robotic gripper with two degrees of freedom (fig-ure 1.6). In addition, the low number of degrees of freedom cause an unnaturaland non-cosmetic grasping movement. These drawbacks can be solved by de-signing the hand prosthesis using microactuators [18], new actuators such asartificial muscles [16], or underactuated mechanisms [19].Figure 1.6: Otto Bock MyoHandThe use of more dexterous robotic hands makes it more difficult for the userto accurately control the device. Fine control of the hand requires, if possible,individual position and force control of the single fingers. Precise position control
1.1 Motivation 15is needed because, to carry out our daily life tasks, we must perform differentgrasping types which involve the activation of specific joints in particular posi-tions. Force control is important too because, in order to manipulate differentobjects or tools, we have to adjust the exerted force: we do not apply the sameforce to firmly grab a hammer than to take an egg from the fridge. With the latestresearch in myoelectric control systems, precise control of prosthetic hands isbecoming a reality.1.1.2 Myoelectric control systemsTo control a device with a myoelectric signal, the signal has to be processed inorder to extract those features that are useful to perform the control. The rawmyoelectric signal is not suitable as a control signal, but several of its featurescan be used as control inputs. The system that extracts the features of the sig-nal and uses them to generate the control commands for the device is calledmyoelectric control system, or MCS. In the field of prosthetic devices, the MCSuses myoelectric signals from the amputee’s remaining limb muscles to controlthe different movements and actions of the prosthesis. Usually, the myoelectricsignal is acquired in a noninvasive way, by using surface electrodes. The valu-able information of the muscle contractions provided by the signal is used by theMCS to send commands to the active prosthesis to perform movements.Although active prosthesis is the most important and only commercial ap-plication of MCS, myoelectric signals can be used for purposes other than thecontrol of active prostheses. For example, in [20] a system capable of recogniz-ing five Japanese vowels by observing the myoelectric signals from the musclesassociated with speech is described. There are other examples of the use ofmyoelectric control in various applications such as wheelchair control [21], ex-oskeleton control [22] or grasping control [23]. Not only disabled people canbenefit from MCS. This technology can be used to design new input methodsfor applications in flight control, space, or the video game industry. An example
16 Introductionof this can be seen in [24] where the authors used EMG technology to design avirtual joystick and keyboard.The evolution of myoelectric control systems, since the first clinically viableprosthesis was presented by USSR engineers in 1965 [11], has been incre-mental. This evolution is a consequence of the parallel advances on the keyaspects of myoelectric prostheses: the control algorithms and the device hard-ware. The necessity for better and more complex control algorithms increaseswith the prosthesis functionality, and more functional prostheses are only use-ful if the control algorithms are able to control all the device functions properly.Improvements in myoelectric control systems can be summarized in three dis-tinct generations [15]. The first generation comprises on/off control schemesto actuate finger or wrist motors with a single speed or a single rate of actua-tion. The devices using this type of control have very simple functions, activatedwhen the myoelectric signal goes above a certain threshold. The control sys-tems composing the second generation include a state machine, large-scalethreshold manipulation, signal amplification, the adjustment of muscle contrac-tion rate (in order to minimize effort required in the first generation co-contractiontype switching), and proportional control. Also, these control systems lower themuscle thresholds allowing more users with upper limb deficiency to take advan-tage of myoelectric prosthetic technology. Finally, the third generation of MCSincorporates programmable microprocessors that allow an infinite range of ad-justment of myoelectric characteristics for the enhancement as well as ease inprosthetic control.The use of microprocessors in myoelectric control results in an increase inprosthesis functionality as well as a decrease in the device cost. One of thebiggest advantages of microprocessor technology is the ability to reprogramcontrol options and input characteristics in a quick and easy way during theclinical setting, without purchasing or changing the device components. They al-low more complex filtering of the EMG signal, which results in a better response
1.1 Motivation 17of the prosthesis. As microprocessors can be reprogrammed, control thresh-olds and sensitivity of the prosthesis can be changed easily as the user strengthand ability evolves. Probably the most important advantage of using micropro-cessors in myoelectric control systems is the possibility of using artificial intelli-gence algorithms that allow pattern recognition-based control schemes, makingit possible to control highly functional devices.Myoelectric control systems can be classified in two groups: pattern recog-nition and non-pattern recognition-based [15]. In the latter group, the controllersare mainly constructed on threshold control and/or finite state machines, andtheir output are limited and predefined control commands based on a sequenceof input signal patterns. Pattern recognition-based controllers are more com-plex: they are able to discriminate the desired classes of functions from signalpatterns, using classifiers.1.1.2.1 Non-pattern recognition-based myoelectric controlNon-pattern recognition-based MCS are used to control simple terminal devices,usually with no more than two degrees of freedom. These MCS include propor-tional control, threshold control and finite state machines [15]. They are simplesystems and they are normally implemented in ON/OFF control or navigation.Proportional controlIn proportional control, the amplitude of the myoelectric signal, that is, the strengthof muscle contraction, controls the speed or the force exerted by the motors.This kind of control can be implemented alongside pattern recognition-basedor non-pattern recognition-based controllers, to achieve precise positioning andaccurate force control.
18 IntroductionThreshold controlThreshold control systems detect the activation and deactivation of the muscle.Before the system detects whether the muscle is active or not, the myoelectricsignal is rectified and smoothed. With this simple signal processing, the averageactivity of the muscle over a time interval is obtained. The theory of operation ofa threshold system is very simple: if the value of the processed signal exceedsa predefined threshold value, the MCS generates a control signal.The simplest form of threshold control is the two-state system [25]. In thiskind of control, the terminal device is normally closed, when there is no mus-cular activity. When the myoelectric signal from the control muscle exceeds thethreshold value, the terminal device opens. A two-state system requires only asingle control muscle to control two functions.Two two-state systems can be used to independently control two functionswith two different control muscles. With this combination, one muscle controlsthe opening and the other muscle controls the closing of the prosthetic hand. Ifthe myoelectric signals of both systems are below the threshold value, the handis in a fixed ”rest” position with the motor off. When the myoelectric signal ofone of the two systems is greater than the threshold value, the hand closes.If the second system is activated, the hand opens. Another option for two-sitetwo-function controllers is to use one of the muscles to control the opening andclosing of the hand while the other is used to control the rotation of the wrist orelbow.Finally, three-state systems, also called double-threshold systems, use onlyone control muscle and compare its myoelectric signal with respect to two thresh-old values [26]. If the value of the signal is less than the value of the lowerthreshold, the motor is off and the hand is in a stationary position. When thesignal is between the lower and upper threshold, the hand closes. Finally, if thesignal surpass the value of the upper threshold, the hand opens. To avoid theclosing of the hand in the transition from the ”off” state to the ”opening” state, a
1.1 Motivation 19time delay is applied in the ”closing” state.Finite state machinesControl systems using finite state machines have a finite number of predefinedcontrol commands associated to the different states of the system. The transitionbetween those states is done by a sequence or a combination of myoelectricsignals, or features of these signals. Finite state machines need to be tunedbefore
by defining states, state transition rules, and output commands[15].Finite state machines can be used to control a prosthesis, or other deviceslike a wheelchair. In [27], the authors implemented a two-state finite state ma-chine to control a simple prosthetic hand with two control muscles. The twostates are ”opening” and ”closing”. The transition from the ”opening” to the ”clos-ing” state occurs when the myoelectric signal of one of the muscles exceeds acertain threshold. The system remains in the ”closing” state until the myoelectricsignal of the other muscle reaches its threshold and the systems changes backto the ”opening” state.A four-state finite state machine is used to control a wheelchair in [21]. Acombination of the myoelectric signals from the right and left shoulders is usedto change the state of the system. If the right shoulder signal is active and the leftshoulder signal is inactive, the ”turn right” state is activated. With the oppositecombination of signals, the ”turn left” state activates. If both signals are active,the system switches between the ”stop” and the ”forward” states. If both signalsare inactive, the previous state remains active, if this state was either ”stop” or”forward”.1.1.2.2 Pattern recognition-based myoelectric controlPattern recognition-based MCS are able to control more complex, multifunc-tional robotic hands, with several degrees of freedom. There is a wide variety
20 Introductionof pattern recognition-based MCS, depending on the pattern recognition algo-rithms used. Pattern recognition-based controllers share the same structure,which can be seen in figure 1.7.Figure 1.7: Block diagram for a pattern recognition-based MCSThe first module acquires the myoelectric signal and performs the analogsignal processing and analog-to-digital conversion. The data segmentation blockdivides the sampled signal in small time segments, to improve accuracy andresponse time. During the feature extraction stage, the information contained inthe time segments is processed so as to emphasize the relevant structures inthe data, while rejecting noise and irrelevant data. Instead of using the raw EMGsignal as the input of the classifier, a feature set is used, improving classificationefficiency. Sometimes, a reduction of the dimensionality of the feature set isneeded to simplify the task of the classifier. Finally, the classification modulerecognizes signal patterns, and classifies them into predefined categories. Theclassifier should be robust and capable of learning from examples how to classifysimilar signals in the same category. Another desirable feature is the capabilityto adapt itself to changes during operation, by the means of online training.There are several types of classifiers used to identify myoelectric signals inthe literature. A fuzzy logic system was used in [21] to control a wheelchair usingthe muscles located near the neck. Neural networks are widely used in the clas-sification of myoelectric signals. Hudgins et al. [8] used a Multilayer Perceptronto classify four different movements to control an upper limb prosthesis. The au-thors in [28] also used a Multilayer Perceptron, this time to diagnose neuromus-cular disorders. Probabilistic methods are also very common in the classifica-tion of myoelectric signals. In [29] the authors implemented a Gaussian MixtureModel to classify six different motions to control an upper limb prosthesis. They
1.1 Motivation 21compared its performance with other classifiers, and showed that it yielded thehighest classification rate, with a 96.3% of success. The most recent classifiersunder research are the Support Vector Machines. Liu et al. [30] designed aCascaded Kernel Machine comprising a Generalized Discriminant Analysis al-gorithm and a Support Vector Machine, to classify eight different types of grasp.They tested its performance compared to the results obtained with other clas-sification methods, such as k-Nearest Neighbor or the Support Vector Machinealone. The Cascaded Kernel Machine obtained the highest classification rate, a96.76%. When using only the Support Vector Machine, they obtained a 93.74%of success.1.1.3 EMG circuitsTo obtain the myoelectric signal in order to use it in a MCS, it is not enoughto measure it using two electrodes. As said before, the amplitude of the signalis very small. Also, the signal is riddled with electrical noise that distorts theinformation contained in it. That is why a system is needed to process the signalso as to increase its amplitude and eliminate the noise. This is the task of anEMG circuit, or electromyograph.The biggest problem with EMG systems is that it is an expensive technol-ogy. The cost of commercial electromyographs ranges from 2000e to more than8000e. Devices using EMG technology also have a high price, as is the case ofrobotic prosthesis, whose prices can be up to 20000e. This high cost of EMGtechnology prevents its arrival to the mass market with applications such asmedical diagnosis at home, robotic prostheses or new forms of human-machineinteraction.There are very few commercial low-cost EMG systems. One of them is theMuscle Sensor from Advancer Technologies, a one channel EMG device pricedat 45$ [31]. The drawback of this system is that the output myoelectric signalis rectified and smoothed, it is not the raw myoelectric signal. This means that
22 Introductionthe Muscle Sensor can be used in a control system, but discards its use asan electromyograph to perform clinical analyses of the signal. Another low-costEMG circuit is the one developed in the open source project OpenEXG-2 [32].The aim of this project is to develop a circuit to measure EEG, ECG and EMGsignals. Being a multipurpose circuit, some of the specific features of the my-oelectric signal have not been taken into account. For example, the bandwidthof the circuit has the upper cutoff frequency at 270Hz, thus losing almost half ofthe information contained in the myoelectric signal. The cost of this two-channelEXG device is 77$.Some examples of EMG circuit design can be found in the literature. In [33],the authors designed an EMG circuit to perform an Autoregressive analysis ofmyoelectric signals. This circuit has a notch filter to eliminate the 50Hz noise,and a stage to rectify the myoelectric signal and compute its RMS value. Choiet al. developed in [34] a surface myoelectric sensor to specifically control ahand prosthesis. They designed the surface active electrode with a notch filterto get rid of the power line noise at 60Hz. In [35], the design of a double differ-ential electrode to control myoelectric prostheses is described. This circuit has abandwidth wider than recommended, uses a notch filter to suppress power linenoise, and outputs a rectified and smoothed myoelectric signal. The authors in[36] designed a very simple myoelectric circuit for educational purposes. Sim-plicity and low cost are the key objectives of this circuit. Because of this, it usescheap electronic components that decrease the quality of the measured signal.The above circuits have been designed for very specific applications, typicallythe control of robotic prosthesis. Because the EMG circuit is not the central partof these projects, they are simple circuits in which no priority has been given tothe quality of the obtained signal, or in which it is not important that the measuredsignal contains all the information. The filtering of the signal is insufficient, andthey usually use a notch filter to reject the power line noise, suppressing animportant part of the signal. Therefore, the use of these circuits is restricted
1.2 Objectives 23to very specific applications and they usually are not suitable for clinical use.However, simplicity does not have to be at odds with quality: it is possible todesign a circuit with a simple architecture and at the same time able to obtainvery stable signals with very little noise.1.2 ObjectivesThe main objective of this work is to create the foundations for the developmentof a low-cost myoelectric prosthesis. This system may also be used as a re-search platform for experimentation with new myoelectric control algorithms. Toachieve this, there are several sub-objectives that have to be accomplished:1. Design of an active electrode and the electronics needed to filter and am-plify the myoelectric signal. To test the validity of the active electrode de-sign, its electronic components will be implemented with the rest of theelectronics in a single PCB. The main design objective of this circuit is theminimization of the influence of electrical noise, by maximizing its signal-to-noise ratio (SNR) and common-mode rejection ratio (CMRR). This cir-cuit may be used in any clinical application and not only in the control ofexternal devices, so the output signal will not be rectified and smoothed.2. Design of a simple MCS to control a 2 DOF robotic hand. The MCS willuse two control muscles, each of them to control each DOF of the hand.The MCS will be implemented in a microcontroller to take advantage of itsbenefits in prosthetic control, described above.3. Design of a 2 DOF robotic hand. One DOF will correspond to the open-ing/closing of the hand, while the other DOF will correspond to the rotationof the wrist. This hand should be as simple as possible, and to keep itscost as low as possible, it will be manufactured using low-cost 3D printing.
24 IntroductionThe next chapter describes the design and implementation of the whole sys-tem, and it is divided in three main parts: the EMG circuit, the MCS and therobotic hand. The third chapter comprises the experimental tests performedto characterize the EMG circuit and its main quality values: the SNR and theCMRR. Also, the operation of the MCS with the robotic hand is described. Fi-nally, the last chapter discusses the conclusions of the work as well as the pos-sible future developments provided by the project.
Chapter 2Design and implementationDown in this chapter, the requirements that the whole system (EMG circuit, MCSand robotic hand) must meet are analyzed. Next, the design and operation ofthe system according to these requirements are explained. The design andimplementation includes both the hardware and the control software.On the hardware side, two parts can be distinguished:1. The electronic system that acquires the myoelectric signals and generatesthe control signals. This system is composed by:o The EMG circuit that measures the myoelectric signals of the user’smuscles. The circuit detects, filters and amplifies the signals.o The microcontroller that converts the analog myoelectric signal intoa digital one and generates the control signals. The microcontrollerincludes a SD card to store the myoelectric signals, as well as aninterface to select the different modes of the control software.2. The robotic hand that is the terminal device of the prosthesis. It is a simplethree-finger robotic hand able to perform basic grips that serve as an aidin the user’s daily life.
26 Design and implementationThe myoelectric control system is the software part of the system. It is re-sponsible for processing the digitized myoelectric signals, and depending ontheir amplitude, it generates the control signals to operate the servomotor of therobotic hand.2.1 EMG circuitLike any other circuit for measuring biopotentials, an EMG circuit must be care-fully designed, paying particular attention to electrical noise that can distort themeasurements. The EMG circuit has been designed following the guidelinesgiven by De Luca in [9]. These guidelines can be summarized as follows:1. Differential electrode configuration. The detection surfaces can be eithertwo parallel bars with dimensions 1cm long by 1mm wide, or two circularsurfaces with a diameter of 1cm. In both cases the spacing between thetwo detection surfaces should be 1cm.2. Active electrode configuration. The differential amplifier should be mountedas close as possible to the detection surfaces. Input impedance & 100MOhm.CMRR = 90-120dB.3. Feedback of the common mode signal of the differential electrode withan electrode attached to the elbow. The use of a notch filter should beavoided.4. Band-pass filter. Bandwidth of 20-500Hz with an attenuation of at least40dB/dec.5. Electrical isolation between the user and the power source.Next, it will be explained how these guidelines have been applied in the circuitdesign.
2.1 EMG circuit 272.1.1 Differential electrode configurationThe EMG signal have to be detected with a differential configuration, becauseof the low amplitude of the signal with respect to the electrical noise. This typeof configuration consists in two detection surfaces in contact with the skin, con-nected to an instrumentation amplifier which subtracts the two detected signalsand amplifies the difference. The consequence is that the signals common toboth detection surfaces are eliminated, as is the case of signals that are gener-ated far from the detection points (like power line noise signals). Signals that aredifferent at the two sites, like local EMG signals, are subtracted and amplified.In this project, commercial ECG electrodes have been used to detect themyoelectric signal. This type of electrode consist of a round Ag/AgCl detectionsurface with 1cm diameter, surrounded by an adhesive material to attach it tothe skin. The detection surface has conductive gel to improve the quality of thedetected signal. The electrode is attached to the input cable by means of a snapfastener. These electrodes have the advantage of being cheap. On the otherhand, they are disposable electrodes, so that a new pair have to be used ineach session.2.1.2 Preamplification stage (active electrode)The preamplifier, or differential amplifier, is one of the key components of anEMG circuit. This is the component that subtracts the signals detected by theelectrodes, and amplifies the resulting signal so it can be used with standardresolution analog-to-digital converters. Hence the name of differential ampli-fier. There are two important parameters to have in consideration when choos-ing a preamplifier: the common mode rejection ratio (CMRR) and the inputimpedance.The CMRR is the measurement of the accuracy with which the signals aresubtracted by the amplifier. It is advisable that this value be at least 80dB. Themost common values for the CMRR are 90 or 120dB.
28 Design and implementationThe preamplifier input impedance should be as large as possible to avoid theattenuation and distortion of the detected signal. These negative effects are dueto the impedance at the junction between the skin and the detection surfaces,whose value may range from several thousand ohms to several megohms. Theinput impedance should not be as big as to cause additional problems to theworkings of the differential amplifier. A value of 100MOhm or greater is advised.De Luca points out that the closer is the preamplifier to the electrodes, thebetter the quality of the detected signal. This is due to the small capacitancebetween the input wires and the power line, that introduces a power line noisesignal into the amplifier. Thus, the optimal solution is to have the preamplifiermounted above the electrodes.Taking into account all this information, an active electrode has been de-signed. The circuit schematic can be seen in figure 2.1. The highlighted partcorresponds to the active electrode itself, while the rest of the circuit is the elbowdrive, which will be explained later.Figure 2.1: Active electrode schematicThe differential amplifier selected for this circuit is the INA333 from TexasInstruments. This component has been chosen because of its electrical charac-teristics. The most important ones for our purpose are:
2.1 EMG circuit 291. CMRR = 100dB (min).2. Input impedance = 100GOhm.3. Low noise density = 50nV/√Hz.4. Low offset voltage: 25uV (max).5. Low input bias current: 200pA (max).6. Minimum supply voltage of 1.8V, which ensures a very low power con-sumption.7. RFI filtered inputs.The INA333 is powered by a single power supply, so the circuit can be usedwith batteries making it a portable device. For this reason, a reference voltagehas to be used because the amplifier’s output voltage can only take values thatare between the value of its positive supply voltage and the value of its negativesupply voltage (5V and 0V in this case). If the signal takes positive and negativevalues (as in the case of EMG signals) and a reference voltage is not used, thenegative half of the output voltage is lost, since the minimum value that it cantake is 0V. By using a reference voltage, the signal oscillates around the thisvoltage value, obtaining the full signal at the output.The gain is set by the two resistors between the RG pins. Using the gaincalculation formula from the INA333 datasheet, the gain value is: G = 1 +100KOhmRG= 1 +100KOhm50KOhm+50KOhm= 2V/V. This low value has been selected because ahigh gain value decreases the CMRR of the circuit, and therefore the quality ofthe signal.Between the output of the INA333 and its reference pin, there is a feedbackintegrator circuit. The task of this circuit is to remove the DC offset voltage at theoutput of the INA333, and force the DC component of the differential amplifier’soutput to always be at the reference voltage (2.5V). Integrating an AC signal
30 Design and implementationresults in the mean (average) of the signal, which is just its DC component, oroffset. This integrator can also be seen as a single pole low-pass filter, with acutoff frequency of ωc= 0rad/s.The operational amplifier chosen for the feedback integrator circuit is theOPA333. This component was chosen because of its low noise in the 0.01Hzto 10Hz band (1.1uVpp), its low offset voltage (max 10uV) and its low powerconsumption.A PCB for the active electrode circuit has been designed, using SMD com-ponents to reduce its size. The resulting circuit has a size of 2x2cm, and has twodifferent versions: a first one with the detection surfaces integrated in the PCB(figure 2.2), and a second one with two snap fasteners than can be used withcommercial Ag/AgCl electrodes.Figure 2.2: Active electrode PCB designThe manufacturing of the active electrode is one of the future developmentsof the project, because in order to validate the circuit design, it has been in-tegrated in a single PCB with the rest of the system’s blocks: the elbow drive,filtering, amplification and power stages.
2.1 EMG circuit 312.1.3 Common mode voltage feedback (elbow drive)One of the biggest problems in EMG is the power line noise signal. The humanbody functions as an antenna that picks up electromagnetic radiation from dif-ferent sources. The amplitude of power line noise may be up to three orders ofmagnitude greater than the EMG signal. It completely overlaps the informationcontained in the signal.To eliminate this kind of noise, there are two approaches. The first one is touse a notch filter to eliminate the signal in the 50Hz or 60Hz band (dependingon the country). This has a great disadvantage: the EMG signal has most ofits information in the 50-60Hz portion of the spectrum, so by eliminating thesefrequencies an important part of the EMG signal is also being eliminated. Thus,to do proper EMG signal analysis, notch filters should be avoided.The second approach is the one used in this project. It is based in usinga circuit similar to the right leg drive (RLD) circuit used in ECG. The task ofa RLD circuit is to minimize the common mode voltage, the voltage of the userwith respect to the differential amplifier’s common, and thus, improve the circuit’sCMRR.Figure 2.3: Common mode bias and noise cancellation with the RLD circuit
32 Design and implementationThe basic topology of a RLD circuit can be seen in figure 2.3. By connectingthe amplifier’s common to the user’s body, the common mode voltage is ide-ally 0V. Feeding back the inverted common mode signal cancels the signalscommon to both electrodes. That means that}