Home

Restoration of Hand Function in Tetraplegics Using Myoelectrically Controlled Functional Electrical Stimulation of the Controlling Muscle

by  Rune Thorsen

in collaboration with

Asah Medico A/S, Denmark

Center for Spinal Cord Injured, Copenhagen University Hospital, Denmark

and Department of Mathematical Modeling,The Technical University of Denmark, Denmark

Submitted: August, 1997            Revised: April, 1998
Industrial Ph.D. Programme, EF533   supported by The Danish Academy of Technical Sciences

Table of Content

    1. Table of Content
    2. Abstract
    3. Abstract (Dansk)
    4. Preface
    5. Objective
    6. Users
    7. The Tetraplegic's Grasp
    8. The Basic Principle
    9. State of the art
    10. Tetraplegia
      1. Brain-Muscle Nerve Path
      2. C5/6 Lesion
    11. Muscle Strength
    12. Nerve and Muscle Fibers
    13. The Muscle
    14. Myoelectric signals
    15. Electrical Stimulation
      1. Stimulation Safety
      2. Control of Contraction
    16. Electrodes
      1. Stimulation Electrodes
      2. Recording Electrodes
      3. Motion Artefacts
  1. Theory, Models and Methods
    1. Test Panel
    2. Electrode Characteristics
      1. Recording Electrode Characteristics
      2. Stimulation Electrode Characteristics
    3. Stimulator Principles
    4. Components of the Recorded Signal
      1. Variation in Myoelectrical Signal
      2. Motor Unit Distribution Model
      3. Antidromic Nerve Block Theory
      4. Spasticity
      5. Motion Artefact Recordings
      6. Spontaneous Activity
      7. Stimulation Response
      8. Inherent Noise
      9. Signal Model
    5. Signal Amplification
    6. Signal to Noise Ratio Definition
    7. Signal Processing
      1. Analogue Signal Conditioning
      2. Sampling and Pre-filtering
      3. MUAP Activity Calculation
      4. Calculation of Stimulation Amplitude
      5. Summary of Signal Processing
    8. Electrode Usage
      1. Electrode Placing
      2. Connections
      3. Hand Stimulation Technique
      4. Electrode-mount
      5. Electrode Embodiment
  2. Hardware and Software
    1. Interface
    2. Amplifier Circuit
      1. Amplifier Principle
      2. Realization
      3. Common-mode Feedback
      4. Low-pass Filter
    3. Digital Signal Processor Unit        
    4. Stimulator circuits
      1. Stimulator Type 1 Design
      2. Stimulator Type 2 Design
      3. Stimulator Type 3 Design
    5. Power Supply Unit
    6. Evaluation Method
      1. Tracking Test Description
      2. Calibration Procedure
      3. Tracking Test Set-up
    7. DSP Software
    8. Host Computer Software
    9. Summary
      1. Amplifier Specifications
      2. Digital Signal Processor Specifications
      3. Stimulator
      4. Power Supply
  3. Hardware and Software
    1. Interface
    2. Amplifier Circuit
      1. Amplifier Principle
      2. Realization
      3. Common-mode Feedback
      4. Low-pass Filter
    3. Digital Signal Processor Unit        
    4. Stimulator circuits
      1. Stimulator Type 1 Design
      2. Stimulator Type 2 Design
      3. Stimulator Type 3 Design
    5. Power Supply Unit
    6. Evaluation Method
      1. Tracking Test Description
      2. Calibration Procedure
      3. Tracking Test Set-up
    7. DSP Software
    8. Host Computer Software
    9. Summary
      1. Amplifier Specifications
      2. Digital Signal Processor Specifications
      3. Stimulator
      4. Power Supply
  4. Discussion and Conclusion
    1. Résumé
    2. The Technological Context
    3. Progress of the Project
      1. Hardware and Software Evolution
      2. Model Evolution
      3. Signal Processing Evolution
      4. Amplifier Enhancement
      5. Stimulator Development
      6. New Experiments
      7. New Stimulation Approach
      8. Electrode mount
    4. Future Aspects of the MeCFES
    5. General Discussion
    6. Market needs
    7. Conclusion


Abstract

This Ph.D. dissertation treats the development of a device that can enhance the force of a pa­retic muscle. The device is a Myoelectrical Controlled Functional Electrical Stimulator (MeCFES). The MeCFES is a small portable device that can be carried in a pocket. It is the intention that the device is to be used by peoples paralyzed by a cervical spinal cord lesion (tetraplegics). The primary aim has been to re-establish a useful grip in tetraple­gics with C5/6 lesion.

The MeCFES records the myoelectrical signals (EMG) resulting from volitional con­traction of a muscle. The muscle in question is the wrist extending muscle: Musculus Extensor Carpi Radialis (longus and/or brevis) ECR. This signal is transformed into a control signal for the intensity of functional electrical stimulation of the controlling muscle. The controlling muscle, the ECR, may be paretic (partly paralyzed) to a degree where only a fraction of the volitional power is remaining.

The unique feature of the device is its capability to simultaneously stimulate the same muscle as the one which controls the stimulation. It allows the use of surface electrodes (electrodes placed on the skin) for both recording of myoelectric signals and electrical stimulation of the same muscle. An essential quality of using of surface electrodes is that no implanted electrodes are required to use the system. It is thus not involving 'modifications' of the user to apply the device. It can thus be tested by the user without any inconvenience or obligations.

A theory of myoelectrical controlled stimulation of the controlling muscle is evolved and summarized into a model of the recorded signal. This uncovers the problems in transforming the recorded signal into a control signal for the electrical stimulation. The model is used to set the technical specifications of the developed hardware. Methods for filtering of the recorded signals are discussed and a new technique for evaluation of the voluntary myoelectrical signal has been suggested and implemented in the device. A new method of suppressing artifacts in recording of bio-potentials has been developed. This has resulted in an invention of a dedicated amplifier. The features are a fast DC offsets compensation and stimulation response suppression of the input signal.

The MeCFES has been tested and evaluated on a voluntary panel of tetraplegics. It has been found that the device can increase the isometric force of the paretic muscle. It can increase the range of controlled wrist extension against gravity for extensor carpi ra­dialis muscles with strength from 1 to 3. Thus a useful grip has been achieved in some tetraplegics. Stimulation of the thumb flexion controlled by the ECR in some experi­ments has provided an enhanced hand function. The restoration of the key grip (lateral pinch grip) and the volar grip has been achieved by this use of the MeCFES.

TOC

Abstract (Dansk)

Denne Ph.D. afhandling omhandler udvikling af et apparat der kan øge muskelstyrken af en paretisk muskel. Apparatet kaldes en Myoelektrisk Kontrolleret Funktionel Elektrisk Stimulator (MeCFES). Dette er et lille bærbart apparat, der kan være i en lomme. For­målet med MeCFESen er at den skal kunne bruges af mennesker der er lammede pga. en hals rygmarvsskade (tetraplegikere). Det primære mål har været at reetablere et brugbart greb hos tetraplegikere med C5/6 læsion.

MeCFESen måler myoelektriske signaler (EMG) stammende fra viljestyret kontraktion af en muskel. Denne muskel er den håndledsløftende muskel: Musculus Extensor Carpi Radialis (longus og/eller brevis) ECR. Signalet omdannes til et kontrol signal for styrken af funktionel elektrisk stimulation af den styrende muskel. Den styrende muskel, ECR, kan være paretisk (delvist lammet) i en grad hvor kun en brøkdel af den voluntære kraft er tilbage.

Det unikke ved apparatet er dets egenskab til at stimulere den samme muskel som sti­mula­tionen er styret af. Tillige bruges der overflade elektroder (elektroder placeret på huden) til både stimulation og måling af det myoelektriske signal fra samme muskel. En væsentlig egenskab anvendelsen af overflade elektroder er at at systemet kan bruges uden behov for implantering af elektroder. Der kræves således ikke 'modifikationer' af brugeren for at kunne anvende apparatet. Det kan således uforpligtende afprøves af bru­geren uden ulemper.

En teori for myoelektrisk kontrolleret stimulation af den kontrollerende muskel er ud­viklet og samlet til en model af det målte signal. Den afdækker problemerne ved at om­danne det målte signal til et kontrol signal for den elektriske stimulation. Modellen dan­ner grundlag for specifikationerne til det udviklede elektriske udstyr. Metoder til filtre­ring af de målte signaler er behandlet og en ny teknik til evaluering af de voluntære myo­elektriske signal er foreslået og implementeret i systemet. En ny metode til under­trykkelse af stimulations responser ved måling af bio-potentialer er blevet udviklet. Dette har resulteret i opfindelsen af en dedikeret forstærker med hurtig kompensation af DC offset af input signalet og undertrykkelse af stimulations responser.

MeCFES systemet er blevet evalueret ved at teste det på et frivilligt forsøgspanel bestå­ende af tetraplegikere. Ved brug af MeCFES systemet er der opnået en øget isometrisk muskel kraft af den paretiske muskel. Det kan også give en kontrolleret øget ekstension af håndleddet mod tyngden ved muskelstyrke 1 til 3. Systemet har således givet et greb hos nogle tetrapegikere. Eksperimenter med at stimulere tommel og finger fleksion, styret af extensor carpi radialis musklen, har givet et anvendeligt greb. Både et nøglegreb og et cylinder­greb er genetableret ved denne brug af MeCFES systemet.

TOC

Preface

This thesis is part of the requirements for obtaining the Ph.D. degree as an Industrial Ph.D. Fellow, conferred by the Technical University of Denmark. The work has been carried out for Asah Medico A/S, Denmark in affiliation to The Department of Mathematical Modeling, The Technical University of Denmark (IMM,DTU) and the Center for Spinal Cord Injury, Copenhagen University Hospital (Rigshospitalet), Denmark.

During the 90'es the former Electronics Institute, DTU has been collaborating with the Center of Spinal Cord Injury, Rigshospitalet in projects with the aim of restoring motor functions in spinal cord injured.

In 1994 Asah Medico A/S initiated the project: "EMG Signals from Paretic Muscles Controlling Electrical Stimulation of the Same Muscle" in collaboration with Center of Spinal Cord Injury, Rigshospitalet (DK); Consort Engineering Ltd (UK); Jones & Hunt, Orthopaedic Hospital (UK); Roessingh Research and Development (NL) and The Danish Paraplegic Association (DK).

The present thesis describes the research and development of electronics related subjects, carried out in the position as an Industrial Ph.D. Fellow.

The work was conducted under the supervision of Steffen Duus Hansen Ph.D. M.Sc.E.E. (DTU), Ole Trier Andersen Ph.D. M.Sc.E.E. (DTU), Fin Biering-Sørensen Ph.D. M.D. (Rigshospitalet), Olav Balle-Petersen M.Sc. E.E. (Asah Medico A/S).

I wish to thank The Danish Academy of Technical Sciences (ATV) for enabling and supporting the project and their representative Dr. Bodil Norrild (The Panum Insti­tute) for attending the project.

I am deeply grateful to the following people who have been very helpful:

Finally I would like to thank Judith Jacobsen, Bjarke Rose, Ditte Wegge, Kirsten Nielsen, Stephen Clausen, Frederik Beck and Peter Rasmussen for reading the proofs of this thesis.

Hvidovre, 29. August 1997

---------------------------------------------------------------------------------

Rune Thorsen, M.Sc.EE, Industrial Ph.D. Fellow

Asah Medico A/S
Valseholmen 11-13

DK-2650 Hvidovre

Denmark

TOC

Introduction

This chapter describes briefly the objective of the project and some essentials of the established knowledge. The sections 1.1 to 1.4 explains the motivation for the project and section 1.5 reviews the state of the art. A basic description of the relevant phy­sio­logy is presented in section 1.6. Section 1.7 defines the (MRC) scale that is used to de­scri­be the muscle strength of muscles. Finally the sections 1.8 through 1.12 summa­rizes some basics of the muscle, nerve, electrical muscle stimulation and re­cording of myo­electrical signals. This knowledge will be used in Chapter 2 for the development of models and methods for recording the muscle signals and applying stimulation. Here the conditions for the development of a system named MeCFES is stated. Chapter 3 descri­bes the developed MeCFES system and the test setup used for the recordings in Chapter 4. Finally there is the conclusion of the project in Chapter 5. As part of the development of the MeCFES as a commercial device a marketing analysis is made.Appendix A is a revised version of this marketing analysis.

TOC

Objective

The aim of this project has been to develop and test an aid, the MeCFES (Myoelec­tri­cally Controlled Functional Electrical Stimulator), for use by people with a certain physi­cal disability involving the loss of hand function. This can be due to an upper motor neuron lesion, such as a cervical spinal cord lesion, causing paralysis of the hand. This can be the case for a C5/6 lesion tetraplegic which usually involves paralysis of the trunk and lower limbs, the hands and partial paralysis, i.e. paresis, of the forearms. For these people, restoration of the hand-function will provide more independence to per­form activities of daily living (eating, drinking, writing etc.) thereby increasing their quality of life significantly. The approach is to use the electrical sig­nal from that part of a paretic muscle that is under volitional control, as a control for electrical muscle stimulation. This myoelec­trical signal, is re­corded using surface electrodes. It is used to control electri­cal stimulation of the same muscle (and/or other muscles) to generate or augment the muscle contraction. When the control­ling muscle and the stimulated muscle is the same the result is an amplification of the muscle strength (Figure 1.1-1).

Figure 1.1-1 Amplifiaction of muscleforce

This method of muscle amplification is applied to the paretic wrist extensor muscle as the first step towards obtaining a feasible grasp (involving the tenodesis function).

Users

The primary target group of users are C5/6 tetraplegics (see 1.6.2 C5/6 Lesion later on). Tetraplegics are very limited in their capability to perform activities of daily living due to the paralysis of the lower limbs, trunk, hands and paresis of the forearms. They are therefore are highly dependent on help from other people. Traumatic spinal cord lesion involves 12-17 persons/million/year in Europe. The most common cause is traffic acci­dents. Tetraplegics are the spinal cord lesioned individuals in most need of personal assistance for activities of daily living. Please refer to Appendix A for a more details on the market needs. Secondary users can be patients with other damages in the central ner­vous system e.g. some patients with multiple sclerosis and cerebrovascular diseases. Since these people in addition can have cognitive disabilities that can compli­cate the expe­riments, this group has not been involved. The precondition for the MeCFES is that a paresis is due to an upper motor neuron lesion. This will be explained further in this chapter.

TOC

The Tetraplegic's Grasp

One of the most useful methods of grasp, is the key-grip or the lateral pinch grip. This is a grip where the object is held between the thumb and the index finger (Figure 1.3-1). It is useful for holding smaller objects such as paper, pencil etc. It can be used in several vari­ants as for example for holding a mug with the thumb in the handle. For holding larger objects such as a bottle, the palmar pinch grip (thumb, index and middle fingers) or a variant, the volar grip (all fingers) is used. These grasps can be used when the wrist extending muscle, extensor carpi radialis has a sufficient strength and there are proper contractures of the fingers.

Figure 1.3-1 Key grip

Under these circumstances the tenodesis function is providing a passive flexion of the fingers and thumb as a result of wrist extension (dorsiflexion), due to the counteracting force in the finger flexor tendons. [Smith 1996]. The force of this tenodesis grip can be en­hanced by shortening the finger flexion tendons. This can either be done by surgery or by fixating the fingers in a position like a clenched fist. The latter method is most often used and will cause the desired contractures of the fingers.

Once the conditions for the tenodesis function is present the next step in obtaining this grip is to establish a controlled extension of the wrist. If the tenodesis grip at full wrist extension does not have sufficient strength; it might be necessary to improve the strength of the grip, mainly by increasing the force between the index finger and the thumb. This thesis proposes stimulation of muscles in the hand, in particular the thumb flexor muscle as described in 2.8.3 Hand Stimulation Technique, using the exten­sor carpi radialis as control.

The forearm contains several muscles that control the hand. An anatomical drawing of the superficial muscles in the dorsal side of the forearm can be found in Appendix C. In normal subjects, wrist extension is controlled by coactivation of the muscles: extensor carpi radialis longus (ECRL), extensor carpi radialis brevis (ECRB) and extensor carpi ulnaris (ECU). Only the ECRB provides a pure wrist extension whereas the other two muscles gives respectively radial and ulnar flexion in addition to the wrist extension [Gray 1973]. The ECRL is partially covering the ECRB and is for this reason most easy to stimulate. There will not be distinguished between ECRL and ECRB since they can not be stimulated individually by surface electrodes. As it appears on the drawing, the finger extensors are located between ECR and ECU and partially overlapping both. For this reason it is difficult to avoid undesired stimulation of the finger extension. On the radial side of the ECR the brachio radialis muscle and the supinator muscle are lo­cated. Accidental stimulation of these will also impede a useful grip.

TOC

The Basic Principle

The MeCFES requires 5 surface electrodes placed on the skin above the muscle. The electrodes are two stimulation electrodes, two recording electrodes and one for active electrical ground (negative feedback of common mode potential). From the re­cording electrodes the signal is fed to the electronic part of the device, which will esti­mate the voluntary activity in the muscle. This estimate controls the amplitude of a stimulation signal that is fed to the stimulation electrodes. The user thus controls the stimulation intensity by the voluntary contraction of the controlling muscle.

When the muscle in control is the same as the stimulated, (extensor carpi radialis stimu­lation) the two types of electrodes are placed over the muscle belly (see 2.8 Electrode Usage).

Thechoice of surface electrodes and not implanted electrodes was decided in the pro­posal for the project. The reasons for using surface electrodes are several. The princi­pal reason is that implantation of the electrodes can be avoided. The use of surface electro­des is safe (apart from possible skin injury) and eliminates the risk of infections or other possible complications of implanted electrodes. The use of surface electrodes leaves the person participating in the test unaffected. The person does not have to be 'modified' prior to the test. Implanting electrodes and removing them again are time consuming, difficult, requires surgeons and can be troublesome for the test person with the risk of permanent injury. The use of surface electrodes eliminates these problems and facili­tates easy and quick testing on both normal and tetraplegics. On the economical side the resources needed for testing or applying the system using surface electrodes are less than when using implanted electrodes. When it comes to marketing of the product, the use of surface electrodes offers a possibility for the customer of easy testing of func­tional electrical stimulation as an aid, before deciding. The use of surface electrodes also gives rise to many problems. This is essential in the technical solution presented in this thesis. In the first instance there are the signal processing problems presented in Chapter 2. Then comes the problems of mechanical reliably placing of the electrodes, their selecti­vity and the cosmetic appearance of the electrode system. Surface electro­des are thus not an ideal replacement for implanted electrodes by rather an alter­native to or a stage before deciding for implanted electrodes.

TOC

State of the art

In 1992 Haxthausen [Haxthausen 1992;Haxthausen, et al. 1991] obtained the Ph.D. degree for the work: "Restoration of Wrist Extension using Functional Electrical Stimu­lation Controlled by the Remaining Voluntary EMG from the Stimulated Muscles". The thesis proved that it was possible to record the voluntary myoelectrical signal from a paretic muscle that simultaneously was stimulated using surface electrodes, i.e. the recor­ded signal control­led the stimulation. The target muscle was the extensor carpi radialis, which extends the wrist. With a differential amplifier made up of leaded com­po­nents and utilizing fast recovery current conveyers, he recorded the myoelectrical signal. The amplifier was basi­cally of the same topology as what is called a conven­tional ampli­fier in the chapter 2.5 Signal Amplification (Figure 2.5-3) in this thesis. Switches were used to shut down the last stage of the amplifier during stimulation. The stimulation pulse was bi­phasic, 300ms/phase with inter pulse interval 300ms (see Figure 1.11-3 in 1.11 Electrical Stimulation). The reason for this pulse type, was to avoid skin damage. After amplification the signal was sampled by a computer (PC) and proc­essed. The signal processing strategy was as follows: The stimulation response was suppressed by blanking and filtering using a 3rd order transposed elliptic comb filter with stop bands at multi­ples of the stimulation frequency. Then the average (bin integrated) rectified value (ARV) was calculated. The stimulation amplitude was directly proportional to this ARV. The amplifier was closed by the ana­logue switches during the stimulation pulse and a number of fol­lowing signal samples were blanked (put to zero). The system was tested in faraday shielded premises using a force- and angle- tracking test (similar to the later described). It was found that C6/7 lesioned tetraplegics got an enhanced force in the wrist when using the system. The system was extensive, involving 2 Personal Comput­ers, the amplifier and stimulator were both voluminous and required special surround­ings. The results were evaluated with respect to muscle force and angle of wrist exten­sion against gravity. A tracking test was used. The test is described and used later in the present the­sis. It was concluded that it is a feasible method to obtain a key grip, but that the system needed to be brought down to a portable size and that the signal processing needed improvement.

In 1994, Thorsen [Thorsen 1994], in his M.Sc.EE. graduation project, made a micro­processor based stand-alone system for research purposes based on the specifications by Haxthausen. It had enhanced noise immunity on the amplifier side. The system was en­closed in a 19" rack making it moveable and independent of external computers.

This equipment was used by Sennels [Sennels 1996;Sennels, et al. 1997] in his Ph.D. project in the investigation of the use of adaptive filters to reduce stimulation artefacts. He made optotracking recordings of the movement of the fingers during stimulation of the extensor carpi radialis and found that the stimulation had a tendency to affect the finger extensors. Thereby counteracting on the tenodesis function and thus making it difficult to obtain a useful key grip. A control strategy using finite-state control was proposed. It was assumed that the tetraplegics could only control the myoelectri­cal signal in discrete levels. For two tetraplegics it was found that only two levels (on/off) of stimulation could be controlled. A third tetraplegic could control a finite-state control of four levels. Adaptive filters were compared to fixed finite impulse response (FIR) filters. Filters with 1 coefficient had same noise power reduction of 14dB for both the adaptive and the fixed filter. Increasing the number of non-zero coefficients gave im­provement of the noise reduction for the adaptive filter. An adaptive filter with 6 coeffi­cients was found to have optimal filterlength. The noise power reduction was 23dB.

For the above described and the present project, the most essential feature difference is the control strategy combined with the use of surface electrodes. This is what differenti­ates them from other attempts to restore the hand function. The control strategies from five of the most closely related or important works are:

TOC

Tetraplegia

Total or partial paralysis of all four limbs and the trunk, denoted as tetraplegia, can have several causes. In this thesis the term will exclusively be used for paralysis caused by a cervical spinal cord lesion. The physiological cause and consequences is summa­rized in the next subsections. This should explain why some muscles, e.g. the wrist extensor muscles, are left only paretic and not totally paralyzed by the cervical spinal cord lesion.

Brain-Muscle Nerve Path

Voluntary movements are caused by contraction of the skeletal muscles under the con­trol of the brain. The normal regulation of the muscle contraction is involving a com­pli­ca­ted network of motor and sensory nerves in the body. This transmits control signals and sensory feedback signals between the brain and the muscle. These signals are neces­sary for the accu­rate and complicated movements that able bodied humans can perform. The motor neurons are transmitting the nerve signal that controls muscle contraction. (This nerve path from the motor cortex to the muscle is described in a simplified form. The interaction with sensory  nerves in the spinal cord is omitted for the simplicity). This nerve signal is trans­mitted in two tempi. From the motor cortex the signal is transmitted by the first motor neuron to a second motor neuron, also called the lower motor neu­ron. Each lower motor neu­ron innervates (i.e. is connected to) a group of muscle fibers. This is called a motor unit. The first motor neuron, also called upper motor neuron, has its nucleus in the motor cortex and the fiber, the axon, is running down through the spinal cord. The spinal cord can be divided into segments where upper motor neuron nerve ends connect to their corresponding lower motor neurons. These segments are in suc­ces­sion from the cranium: The 8 cervical segments C1-C8, the 12 thoracic T1-T12, the lumbrical L1-L5 and final the sacral S1-S5 segments [Netter 1996]. The muscles of the forearms are inner­vated by lower motor neurons having their cell bodies in segments C4 to T2. The lower motor neurons are intermingled in the brachial plexus and col­lec­ted in nerves containing motor neurons (and sensory nerves) from different segments. One of these nerves is the radial nerve. The wrist extensor muscle, extensor carpi radia­lis muscle, is innervated by lower motor neurons in the radial nerve [Netter 1996]. There are several motor units comprised within this muscle. The cell bodies of the lower motor neurons belonging to these motor units are distributed in the spinal cord segments C5 to C7, (C8) [Kendall, Kendall et al. 1983].  Some motor units are innervated from C5, others from C6 and so on. The extensor carpi radialis muscle is thus not only innervated from one segment but from more segments.

TOC

C5/6 Lesion

A damage of the signal path, blocking the motor nerve impulses will paralyze the corre­spon­ding motor unit (from now on mainly referring to only the muscle fibers). The para­lysis can be due to a lesion of the upper motor neuron and/or a lesion of the lower motor neuron. A nerve signal in an intact lower motor neuron will cause the motor units to con­­tract. Such a signal can be evoked artificially in the neuron by electrical stimula­tion. It requires that the lower motor neurons are intact to make the motor units of the muscle contract by electrical surface stimu­la­tion.

If a lower motor neuron is severe damaged, the motor unit will be dener­va­ted and paralyzed and thus not perceptive to electrical stimulation. In case of a lesion of the upper motor neuron, and if the lower motor neuron is intact, the motor unit (the muscle fibers) is innervated but paralyzed. In this case electrical stimulation can be used to con­trol contraction of the muscle fibers in the motor unit.

Figure 1.6-1 Tetraplegic with complete C6 lesion

Tetraplegia due to a spinal cord lesion is primarily a lesion of upper motor neurons. Lower motor neurons will often be damaged too. The situation is shown in Figure 1.6-1. For simplicity an example of a complete C6 lesion is illustrated. In the C6 lesion exam­ple, the segments C1 to and including C6 are intact, C7 is damaged and C8, T1 -T12 etc. are intact. Muscles innervated from C1 to C6 are unaffected and have normal function. Muscles normally innervated from C8 and down will still be innervated but totally para­lyzed due to the lesion of the corresponding upper motor neurons in C7. Both types of muscles can be stimulated since they are fully innervated. Muscles that prior to the lesion was innervated from segments above, in and below the lesion will be paretic i.e. partly paralyzed. The total available muscle strength will for that reason be reduced. Such a muscle (among several) is the wrist extensor muscle, extensor carpi radialis muscle. As illustrated schematically, the muscle will contain motor units affected by the lesion in 3 different ways. The normally innervated motor units that are under voluntary control and can be stimulated. The motor units that are denervated due to lesion of their lower motor neurons in C7. These are paralyzed and can not be stimulated. Finally there are the motor units that are paralyzed due to the lesion of the upper motor neuron, but are innervated and can thus be stimulated due to the intact lower motor neuron. The result is a muscle with some voluntary control having paralyzed and non paralyzed parts that can be stimulated electrically. This situation is a criterion for the MeCFES principle to be feasible. (Other reasons than a spinal cord lesion can cause this type of paresis as men­tioned in 1.2 Users.)

Spinal cord lesions are never identical. There is always differences in the motor function capabilities in the population of tetraplegics. Even if the level of lesion is the same and the diagnosis is for example complete C5 lesion, they might have different abilities to use their hands. The reason is presumably variations in the extend of the lesion, that the lesion anyhow is not 100% complete and maybe anatomical variations in the nerve paths. The complete C5 tetraplegic can have different strength of the extensor carpi radialis as the data in 2.1 Test Panel illustrates. The effect of the lesion is not even symmetric but the two forearms of a C5 tetraplegic can have different wrist extension force. The deter­mination of the level of lesion is clinically done by examining muscle forces and the extent of skin sensation [Biering-Sørensen]. For that reason the level of lesion does not define the exact capabilities but rather indicates which muscles that might be affected. To deter­mi­ne whether a certain tetraplegic will benefit from the MeCFES, it is thus not suffi­ci­ent to know the level of lesion, but the person must be tested using functional electrical stimulation.

TOC

Muscle Strength

When discussing the capabilities of a certain muscle the strength of a muscle is defined in Table 1.7-1 below. It is a subjective measure that can vary slightly with the persons judging the strength. It is often referred to as the medical research council (MRC) scale.  It is a judgment of the force the muscle can provide on the joint. In this case it is the wrist.

Strength

Description

0  -  none

Totally paralyzed muscle

1  -  trace

Muscle contraction is only just visible or palpable but the muscle cannot produce any movement

2  -  poor

Movement of the joint possible only when gravity is eliminated (movement perpendicular to gravity)

3  -  fair

Movement of the joint against gravity just possible

4  -  good

The muscle can move the joint against gravity and some extra force.

5  -  normal

Normally innervated muscle that can exert normal force.

Table 1.7-1 Muscle strength definition interpreted
from Kendall [Kendall, Kendall et al. 1983]

This graduation is commonly used. The C5/6 lesioned in the test panel have strength of the extensor carpi radialis in the range 1-4.

Nerve and Muscle Fibers

A fundamental characteristic of both a nerve and a muscle is that once initiated, an ac­tion potential will propagate along the fiber to the fiber endings. An action potential is a local discharge of the fiber due a local ion transport through the cell membrane (see Figure 1.8-1). In brief it consists of an absolute and a relative refractory period. In the abso­lute refractory period the nerve can not be excited by stimulation where as it in the relative refractory period can be excited by a stimulation, but with a higher threshold. (A more extensive explanation can be found in many textbooks on physiology e.g. [Schmidt and Thews 1983])

Normal action potentials in the lower motor neurons are propagating in only one direc­tion; from the cell body in the spinal cord to the motor unit. Each neuron divides into the terminal nerve branch before connecting, via the endplates, to the muscle fibers of the motor unit. Here the nerve signal is initiating a new action potential in the fibers of the inner­vated motor unit. These endplates are placed near the middle of the muscle fibers. Here the action potential will propagate in both directions towards the ends of the muscle fiber.

Figure 1.8-1 Illustration of an action potential as function of time. 

The motor nerve is myelinated, i.e. it is surrounded by a sheath that is electrically iso­lating. The myelin sheath provides a fast conduction velocity for the nerve signals. The nerve is unmyelinated in the last part before it attaches to the muscle fibres. An unmye­linated nerve will be more easy to excite by electrical stimulation than a myelinated nerve due to the resistance of the sheath. This may be considered when applying stimu­lation electrodes and finding adequate positions (the motor points).

The fiber behaves in an all or nothing fashion, which means that the nerve impulse are like a binary signal. No intermediate levels are possible. This is the 'all or nothing' law that applies both to nerve and muscle fibers. When a nerve/muscle fiber is stimulated above a certain threshold it will fire, which means that the action potential will start propagation. The 'all or nothing' law is not equivalent to a constant level of the action potential since the amplitude can vary slightly with fatiguing of the fiber or in case of neurological deficiencies [Stålberg and Trontelj 1994].

TOC

The Muscle

Besides from being the force generating element of movements, the muscle is also a source of  an electrical signal, the myoelectric signal. This is of significance for the way it can be controlled by electrical stimulation, and for the use of the myoelectric signal for control.

The muscle can be regarded as a collection of elements, the motor units comprising the muscle fibers, that behaves according to the all or nothing law. These motor units can thus either contribute to the muscle force with a twitch of full contraction or nothing. The muscle fibers can be of different types. These types behave differently with respect to their endurance and force output. The fast glycolytic fibers (FG) can be recruited at a high frequency up to 100 Hz and generate high force. The staying power of these fibers is short. The purpose of these fibers is to generate a short powerful con­traction. The fast oxidative fibers (FO) generate a smaller force but have a longer en­durance. And finally there are the slow oxidative (SO) fibers which can sustain a mod­erate force for a long time (Figure 1.9-1). These fibers are of special interest for the stimulation of hand-func­tion, since holding typically requires a moderate near constant force for several seconds up to minutes. The central nervous system normally recruits the slow oxidative units by low frequency (<10Hz) nerve signals.

Figure 1.9-1 Fiber types vs. endurance/force [Mortimer 1984]

The muscle fibers belonging to a motor unit is of the same fiber type [Schmalbruch 1985]. The fibers of different motor units are intermingled in the muscle but fibers be­longing to the same motor unit have the highest density in the center of the motor unit [Buchthal and Schmalbruch 1980].

Only few data are available on the size and topology of different types of motor units in the muscles in humans. This may be due to the fact that it is difficult to determine which muscle fibers a particular motor neuron is connected to. Motor units of human upper limb mus­cles have an average territory of 5-10 mm in diameter. In such an area there are typi­cally fibers from 15-30 different motor units [Buchthal and Schmalbruch 1980; Schmalbruch 1985]. The number of muscle fibers in the motor units varies between motor units within the same muscle and more widely from muscle to muscle [Buchthal and Schmalbruch 1980]. For human brachioradialis muscle there are around 350 motor units with an average of more than 410 fibers per unit. 1'st Dorsal interosseus and 1'st lumbricalis (muscles in the hand) have about one hundred motor units with respectively 340 and 100 fibers per motor unit [Feinstein et al. 1954; Schmalbruch 1985].

Figure 1.9-2 Tentative model of the cross-section topology of motor units in a muscle. Two motor units MU1 and MU2 are not in the same distance from the skin.

These informations about the topol­ogy of the motor units can be interpreted to form the model in Figure 1.9-2 where the motor units are intermingled but located to different compartments of the muscle. The number of motor units and their type is significant for the myoelectric signal and for the force properties of the muscle using electrical stimulation.

TOC

Myoelectric signals

The control signal of electrical stimulation is obtained from the signal recorded on the skin over the muscle. The following is a discussion of the nature of the control signal and some of the noise components. It is assumed that the voluntary contraction of the paretic muscle is under full control of the conscious mind. The problem is to extract information of the voluntary contraction from the recorded signal. The sum of  action potentials from the muscle fibres of a motor unit generates an electrical field. This is the motor unit action potential (MUAP). The myoelectrical signal is composed of such MUAPs. A MUAP can be caused by other things than voluntary contraction. It is for that reason important to distinguish between voluntary MUAPs and MUAPs with other causes. These reflections are used in the model proposed in the section 2.4 Components of the Recorded Signal.

Voluntary contraction of a muscle is controlled by the central nervous system by modu­lation of the number of motor units recruited and the firing rate, which is the frequency of the motor nerve signals. This results in voluntary MUAPs and make up the voluntary myoelectrical signal.

Surface stimulation causes MUAPs synchronized by the stimulation pulses. These corre­lated MUAPs are denoted as the compound motor unit action potential (CMUAP). The CMUAP is by na­ture non-voluntary. This is a significant component in the re­corded signal from a stimulated muscle.

Other types of MUAPs can theoretically occur as a side effect of electrical stimulation due to two phenomena: The possibility of exiting the H-wave and the F-wave by the stimulation. The H-wave is MUAPs caused by stimulation of the H-reflex. This is a monosynaptic reflex caused by stimulation of sensory nerves (the Ia afferent neurons) in the muscle. These run from the muscle to the spinal cord where they directly stimulate the motor neurons of the same muscle [Schmidt and Thews 1983]. The latency of the H-wave in the hand muscles is typically 15ms [Tarkka 1986]. The F-wave is a re­current discharge of the motor neuron due to the electrical stimulation described by [Stålberg and Falck 1993]. The F-wave follows the CMUAP (the stimulation response) and occurs only in a small fraction of the stimuli (5%). The H-wave has in contrast to the F-waves constant shape and latency.

In addition to the above mentioned spontaneous MUAPs can occur independently of voluntary contraction and stimulation. In its ex­treme this can be spasms. A recorded myoelectrical signal from a provoked spasm in a muscle is presented in section 2.4.4 Spasticity.

The CMUAPs, H-waves, F-waves and spontaneous MUAPs are non voluntary and are regarded as noise. The voluntary MUAPs is the desired signal.

The voluntary MUAPs from a single voluntary contracted motor unit are fairly regular but the intervals between them are not constant. There is a tendency of a long MUAP  interval to be followed by short [Andreassen et.al. 1980]. This interval between MUAPs  is regarded as a stochastic process and the Cauchy distribution with a standard deviation of 20% has been proposed for modeling the process . The mean interval is typically in the range of 50-150ms, depending on the voluntary contraction [Andreassen 1978].  An example of a MUAP recorded by needle electrodes can be seen in Figure 1.10-1

Figure 1.10-1 Single MUAP [DeLuca 1993]

The fibers of a motor unit discharge synchronously since they are innervated by the same motor neuron, but the action potentials are not initiated simultaneously. This is due to variations in the length and conduction velocity of the fibers in the terminal nerve branch [Schmalbruch 1985] the action potentials will be shifted in time. This gives a dilatation of the MUAP and interference be­tween the fiber action potentials. The size of the action potentials decreases rapidly with increasing distance between the generating muscle fibers and the recording electrode. Therefore for a given electrode location, the myo­electrical signal consists of large and small MUAPs with temporal dispersion. Each MUAP will have a characteristic shape [Schmalbruch 1985]. To obtain the best results, the electrodes should be placed over the middle of the muscle belly where the distance to the motor units is minimal. If the mus­cle is shortened, the motor unit action potential duration decreases and the amplitude in­creases [Stålberg and Falck 1993].

In the previous section ( 1.9 The Muscle) it was argued that the motor units were dis­tributed in different territories. Consider each motor unit as an electrical generator sur­rounded by conductive tissue as in Figure 1.10-2, where the tissue is a non-homoge­neous volume conductor that attenuates and filters the signals from the generators. The figure serves to illustrate the complexity of the field from the motor units. It illustrates that the MUAPs contribute with a different amplitude depending on the depth in the tissue and their orientation.

Figure 1.10-2Tentative illustration of the motor units as electrical
 generators in a volume conductor

The orientation of the generators will depend on the resulting electrical vector from the depolarization pattern of the muscle fibers involved. Since the currents in the tissue are very small it is reasonable to assume linearity. Thus the electric signal arising from all the MUAPs can be modeled as a sum of impulse generators. Each has a different trans­fer functions between generator and the electrodes. Figure 1.10-3 shows the model as re­viewed in [Merletti, Knaflitz et al. 1992]. The firing pattern of the motor units are repre­sented by the impulse trains. These are stochastic distributed in time. Each impulse gen­erates a MUAP. The transfer function hn(s) is representing the shape of each MUAP and the sum of these are constituting the myoelectric signal. Assuming that the shape of the action potential of each muscle fiber and conduction time in the terminal branches are not changing over time hn(s) should be an unique fixed function for each motor unit. The amplitude An of each MUAP is dependent on the attenuation of the MUAP due to dis­tance from recording electrode and the motor unit and the size of the motor unit (in terms of number of fibers). Changes in the amplitude of action potentials of the muscle fibers (e.g. fatiguing) will also result in a change in amplitude.

Figure 1.10-3 The signal at the electrodes can be regarded
as a sum of different motor units. (Tentative model).

The goal is to transform the voluntary contraction of the muscle into a well defined control signal. Based on the reflections in this section it has been decided that: The control signal for the MeCFES should be the total number of voluntary MUAPs (motor unit action potentials) in average per time unit. The problem is then, how to transform the myoelectrical signal to an estimate of such a MUAP activity measure­ment, when the signal is noisy and the power of each MUAP is differing. A method is proposed in 2.7.3 MUAP Activity Calculation.

TOC

Electrical Stimulation

In the development of a system for controlled contraction by the use of functional elec­trical stimulation, it is of value to have knowledge about some of the mechanisms in­volved. A comprehensive practical textbook on the topic of functional electrical stimu­lation is written by Benton [Benton, et al. 1981].

When stimulating the muscle using surface electrodes, the stimulation current is flowing through the skin and the underlying tissue. The current will spread in the volume be­tween the electrodes and can affect both nerves and muscle fibers. Muscle fibers require around ten times greater stimulation current to be excited [Mortimer 1984] than nerves. For that rea­son the stimulation will predominantly be nerve stimulation. For the same reason it is assumed that denervated motor units are not susceptive to surface stimulation. Since it is the muscles that are the target for the stimulation the phrase 'muscle stimulation' will be used despite it should be called transcutaneous electrical motor neuron stimulation. (This phrase is often used and abbreviated to TENS).

As a result of the 'all or nothing' law applying to the motor units, the only way to con­trol the muscle force, is by modulation of the number of motor neurons stimulated and/or by the excitation frequency. The stimulation current threshold for a neuron is depending of the diameter of the neu­ron. The larger the diameter, the lower the threshold [Mortimer 1984]. It is believed that this is the reason that the muscle contraction can be modulated by controlling the stimulation current amplitude.

Activation of the different motor units, when controlled by the central nervous system, is asynchronous in both space and time [Merletti, et al. 1992]. In this way a steady smooth contraction can be maintained at a low firing rate for the individual motor units (Figure 1.11-2). By surface  electrical stimulation it is not possible to stimulate the motor units individually. The MUAPs from the motor units stimulated will be synchronized. Each stimulation pulse will cause a twitch of force in the muscle generated by the motor units activated. To fuse the twitches to a smooth contraction it is necessary to raise the stimulation frequency above the natural level. The frequency depends on the me­chanical properties of the joint in question. For the wrist extensor it is found empirically that the stimulation frequency should be above 10Hz to obtain a smooth contraction. (The Freehand system for hand function uses 12.5 Hz with implanted electrodes). This implies fatigue of the muscle. Increasing stimulation intensity will recruit more motor units besides the motor units stimulated at the low intensity. Thus the only way to enable these motor units to recover is to turn the stimulation of the muscle off.

Figure 1.11-2 Tentative model of nerve impulses' distribution.
Voluntary impulses are non correlated (left) and stimulated impulses are
correlated (right). [Merletti, Knaflitz et al. 1992]

Stimulation Safety

When stimulating tissue the pulse shape should be considered for safety reasons. Intro­ducing a current in the electrolyte will result in electrochemical processes. Depending on the current density, the electrochemical processes can be reversible or irreversible. The irreversible region is entered when the net charge density exceeds a certain limit. When the irreversible process occurs free radicals can be created. These may be toxic. There­fore the charge density in the tissue should be kept in the reversible region.  To comply with this, it is commonly recommended to use a bi-phasic pulse shape, see Figure 1.11-3, instead of a mono-phasic [Mortimer 1984]. The bi-phasic pulse is charge bal­anced. In that way no direct current will push the processes out into the irreversible area. There is an interpulse interval (IPI) between the two phases to reduce the annihilation of the initiated action potential. As it will become clear in 2.4.7 Stimulation Response it is of importance that there are no remaining charge after the end of the pulse and that the total duration of the pulse is kept as short as possible.

Figure 1.11-3 Stimulation pulse types

For these reasons a bi-phasic charge balanced pulse is used to avoid skin/tissue irrita­tion. The shape of the pulse is chosen to be rectangular as shown in Figure 1.11-3 with a 0.3ms pulse width/phase. This is the shape that Haxthausen [Haxthausen, et al. 1991] used and is assumed adequate for minimizing pain [Gracanin and Trnkoczy 1975]. Based on preliminary experiments, 50mA is assumed to be absolute maximum stimu­lation amplitude needed for upper extremity stimulation.

Only a few reports on negative side effects of functional electrical stimulation have been found (only temporary skin burn). Shannon [Shannon 1992] has proposed a model based upon data from cortical stimulation. (Although this stimulation is very different from muscle stimulation this is used as an indicator for tissue damage. A formula for surface stimulation has not been found in the litterature).

The current limit should not exceed a maximum value in mA given by

                                                                                                      Eq.1.11

where de is the diameter of a circular electrode in cm, tph is the duration of each phase in the simulation pulse in ms and k is a constant. For k less than 2, stimulation is regarded safe and k=1.5 is regarded as a conservative limit. (This model is based upon animal ex­periments of cortical surface stimulation). Using this formula on k=[1.5;2], t=0.3ms and I=50 one obtains a stimulation electrode diameter of 17-30mm.

The recommendations and regulations in the International Standard IEC 601-2-10 (Medical Electrical Equipment Part2: Particular requirements for the Safety of Nerve & Muscle Stimulators; 1997) should be observed. Of special interest is that it is advised to keep the current density < 2mA/cm2. (It is assumed that this is the effective value of the current although it is not explicitly stated). A biphasic pulse with a 2 x 0.3ms pulse with and 16Hz repetition frequency and electrode diameter 17mm, the root mean square value of the current density will yield the current density.

 

Using electrodes with greater diameter than 17mm is then in accordance to the directive.

Control of Contraction

An important issue is to control the force and position of the joint on which the stimu­lated muscle is acting, since this is the purpose of it all. The stimulation input versus muscle output, the so-called recruitment curve, is non-linear [Hines, et al. 1992]. The recruitment has been investigated and the results can be found in 4.2 Recruitment Curve where it will be demonstrated that it is non-linear and not a constant relation. For this reason it is neces­sary to have some contraction information feedback to the controller of the stimulation. Such closed loop systems are a important topic and often discussed topic of functional electrical stimulation. A way to provide this information is by the use of the natural sen­sors. An example can be the recording of the sensory  nerve signals from the receptors [Haugland and Hoffer 1994;Haugland, et al. 1994;Haugland and Sinkjær 1995;Popovic, et al. 1993;Yoshida and Horch 1996] or as is the case with the MeCFES the visual feed­back to the user. It can be discussed whether this is an open or closed loop.

TOC

Electrodes

To transmit the myoelectric signals from the muscle to the amplifier and the stimulation current to the muscle surface electrodes are used. There are some general requirements that the electrodes must fulfill and when stimulation and recording is performed in the same area extra constraints apply to the electrodes. A discussion of general require­ments for electrodes can be found in [Crago, Peckham et al. 1974; Tam and Webster 1977; Webster 1984; Webster 1992].

The electrodes should provide the following properties:

Stimulation Electrodes

In the previous section was found a recommended size for the stimulation electrodes. A choice of 30mm in diameter is be regarded safe based upon the findings in 1.11.1 Simulation Safety. The electrodes must ensure good uniform skin contact. In theory the stimulation should be applied where the nerve enters the muscle or over the endplate zone of the muscle. This zone is where the terminal nerve branches attaches to the muscle fibers. Since the nerve fibers here are unmyelinised they are more easily stimulated. Ag-AgCl electrodes was used for stimulation in the work of Haxthausen [Haxthausen 1992]. Personal communications (name unknown) have advised not to use Ag-AgCl electrodes for stimulation since they may cause permanent tattoos to the skin. For this reason, this type of (recording) electrode can not be used for stimulation. The stimulation electrodes used in this project are silicone-rubber-carbon electrodes from ASAH Medico A/S Denmark. The electrodes are coated with a conductive gel that makes them adhesive.

Recording Electrodes

To record myoelectrical signals Ag-AgCl electrodes can be used. These are made of silver with a thin layer of silver-chloride on the surface. A discussion of the features of these electrodes can be found in [Webster 1992]. The electrodes used in this project are Ag-AgCl electrodes (Blue sensor from Medicotest A/S, Denmark). Electrode paste are applied to the electrodes. This paste and the gel on the stimulation electrodes serves as an ionic carrier between the skin and the electrode.

The type of electrodes used for stimulation and recording in the project has been chosen since they are easy available, inexpensive, easy to use and fulfilled the demands. The electrode types are very commonly used and could be supplied by second souce.

Motion Artefacts

The electrical conditions of an electrode are very complex. A simplified model can be seen in Figure 1.12-1. Especially the half-cell potential (the potential occurring from the dissolution of metal ions into the electrolyte) is significant for the recording elec­trodes. This potential depends on the equilibrium of the different ions.

Figure 1.12-1 Simplified model of the electrode body interface [Webster 1992]

When the skin or the recording electrode is stressed the electrolytic equilibrium is shifted and thus the half-cell potentials (Figure 1.12-1) are changed [Webster 1992]. This may be the case if, for instance, the electrode is tapped, twisted or changing shear forces are applied. These actions will disturb the equilibrium states of the electrode-electrolyte, electrolyte-skin and the electrolytes in the epidermis. If the reference elec­trode does not experience the exact same changes, then the result will be a rapid change in the potential between the electrodes. This signal is termed motion artefacts. The shape of the motion artefacts is dependent on the mechanical action on the electrodes, the type of electrodes and the involved electrolytes. The motion artefacts will be occurring at random times depending on the conditions un­der which the electrodes are used. Stretching the skin can give rise to up to 10mV motion artefacts [Webster 1984], which should be com­pared to a typical peak value of 1mV in myoelectric signals.

Due to the stochastic nature of the myoelectrical signal it is believed that it is not easy to filter out the motion artefacts. For this reason the electrodes must be protected from mechanical influences and stretching of the skin must be avoided.


TOC

TOC

Theory, Models and Methods

This chapter describes development of the theoretical models and methods. This is based upon the established knowledge reviewed in Chapter 1 and experimental results. Initially the test panel of voluntary participants in the experiments is presented in sec­tion 2.1. The measurements in section 2.2, of electrode impedance and noise, are used in section 2.3 for the choice of stimulator principle and in section 2.5 for specifying the amplifier constraints. A model of the recorded signal is evolved in section 2.4. This model is used when discussing signal amplification in section 2.5, defining the signal to noise ratio in section 2.6 and choos­ing the signal processing methods in section 2.7. Section 2.8 is describing the elec­trodes used, their placement and suggest a electrode-mount, making electrode appli­cation easy for the user.

TOC

Test Panel

Myoelectrical signals from normal and paretic muscles will be presented in the follow­ing sections. In Chapter 4 results from evaluation of the MeCFES performance will be pre­sented. For these tests a number of volunteers have participated. Those, which results are presented in this report are listed in Table 2.1-1 below. For the tetraplegics the selec­tion criteria has been that they were willing to participate, were spinal cord lesioned at a level that resulted in paresis of the extensor carpi radialis, had the time and lived in a reason­able traveling distance from the laboratory in Copenhagen. More than 5 other tetraple­gics have been tested but will not figure in this dissertation.

The muscle strength is not exact, but may change a single level depending on the person, who is judging the force and when the muscle is tested. As for all muscles the condition of training or fatigue is influencing on the force. Generally, the right hand is tested since the participants were right handed before injury. Subject EG had had a muscle transfer in the right hand and had a good tenodesis function in that hand. For that reason the left hand had been used in the tests. On subject RAT (the author) the left hand has been used for testing, leaving right hand free for working. Subjects OBP and CD has only been tested once for recording of the recruitment curve.

Subject ID

Lesion

ECR strength (MRC)

Sex

Date of Injury

Date of Birth

AA

C5

4 (Right)

F

95.01.17

1917

EG

C5

4 (Left)

F

84.02.13

1955

FB

C5

4 (Right)

M

59.07.29

1942

HSJ

C5

2 (Right)

M

94.05.14

1955

JBS

C5

1 (Right)

M

94.07.16

1953

LP

C5

2 (Right)

M

92.05.22

1967

KGN

C5

1 (Right)

M

84.04.14

1965

KN

C5

2-3 (Right)

F

93.12.28

1965

RAT

NONE

5 (Left)

M

NA

1967

OBP

NONE

5(Left)

M

NA

1957

CD

NONE

5(Left)

M

NA

1966

Table 2.1-1 The test panel

Testing and myoelectrical signal recording is carried out with the forearm resting on a horizontal support with the palm down. Since the MeCFES should enable the user to extend the wrist against gravity, the angle or force of this movement is often used as a measure for muscle contraction.

As with most cases of tetraplegia the subjects are influenced differently. Only a few re­marks shall be made to some of the test persons.

Subject AA needs typically more than 30mA to generate wrist extension. This is a high current compared to the other subjects. She has a weak voluntary grip and is using the tenodesis function in small extent. She is generally difficult to stimulate to a good wrist extension.

Subject EG has limited used of the left hand. It is easier to stimulate to a wrist extension than in the case of subject AA. She has a good tenodesis flexion of the fingers.

Subject KGN has much spasticity in general, and the fingers are stretched when stimu­lating the extensor carpi radialis muscle.

Subject KN has strong contractures in the supinators, which is resulting in a tendency to hold the hands in a 'begging position'. She has no grip at all. Using special tools mounted to her hand she can use a computer.

TOC

Electrode Characteristics

Since the electrodes constitutes the electrical interface to the muscle, it is of interest to have an idea of the impedance of the recording electrode as well as the stimulation elec­trodes.

The electrodes used for recording are Ag-AgCl electrodes with electrode paste (Blue sensor from Medicotest A/S, Denmark). These are a commonly used electrode type for EEG and EMG recordings. The recording electrodes have an effective diameter of 7mm, but physical diameter is 30mm due to the adhesive non conducting backing material. The stimulation electrodes are silicone-rubber-carbon (from ASAH Medico A/S Denmark) with approximately 30mm effective diameter equal to the physical di­ameter. This type are commonly used for transcutaneous electrical nerve stimulation (TENS). The electrodes are coated with a conductive gel that makes them adhesive.

These electrode types are used for all the measurements presented in this work. The following recordings are carried out at a normal subject (subj.: RAT).

Recording Electrode Characteristics

The impedance of the recording electrodes is important for the design of the amplifier. Tree sets of electrodes have been tested. Each set was applied 3 times. A total of 9 re­cordings was thus made. Prior to the measurements the skin was washed with tap water. According to the recommended use on the electrodes, electrode gel (from Medicotest A/S, Denmark) was applied before the electrodes were applied. The distance between the electrodes was varied between 4-10cm. A 300mV sine generator was used for all the measurements (Selected as the minimum allowed by the noise level and the available equipment). This impedance recordings used the setup illustrated in Figure 2.2-1.

Figure 2.2-1 Measurement set-up

As it can be seen in Table 2.2-1, the measurements have a high deviation. The impe­dance varies from placement to placement and over time. The impedance can decrease nearly a decade over the first a quarter of an hour after application on dry skin. The dis­tance between electrodes seemed not to have significant influence on the electrode im­pedance. Table 2.2-1 shows minimum and maximum values of the measured electrode impedance magnitudes |Z| at frequencies f, which are relevant for the recording of myo­electrical signals.

f(Hz)

|Zmin|(kW)

|Zmax|(kW)

10

30

100

50

22

78

100

13

68

500

11

30

1000

17

20

Table 2.2-1 Electrode impedance recordings

It should be noted that this impedance may not be the same as for recording myoelectric signals. Applying a current in the electrodes may change the impedance due to chemical reactions.

The noise from electrodes mounted on the skin above inexcitable tissue, the bony promi­nence of ulna, is 4mVRMS ± 1mV. These measurements are performed with the MeCFES amplifier which will be described in Chapter 3. When tapping the electrodes with a fin­ger, motion artefacts of 5-10mV peak can be caused (Applying firm pressure can cause greater motion artefacts). These motion artefacts are recorded direct by an oscil­loscope (10 MW input impedance). The skin has not been prepared by e.g. removing the epider­mis (outermost dead layer of the skin), since this is not realistic for daily use of the MeCFES device. Placing the electrodes on each other gives an impedance recording of the pair of elec­trodes of less than 1kW.

In section 2.8.5 Electrode Embodiment an electrode concept is presented. Wet shammy is used as contact medium. For this reason impedance of Ag-AgCl electrodes with wet shammy as contact medium has been measured. The electrodes have been placed with direct contact to each other with the leather in between. The results are shown in Table 2.2-2.

Impedance (kW)

Noise(RMS)

Motion Artefacts

0.9@50Hz

0.75@200Hz

0.6@500Hz

<8mV

<50mV

(Recorded by oscilloscope)

Table 2.2-2 Electrode pair with wet shammy leather as contact medium.

As it may be noted the motion artefacts are significantly higher. The impedance is very low compared to the recordings in Table 2.2-1. On that basis it can be concluded that the greater part of the electrode impedance is due to the skin, since the recording in Table 2.2-2 does not imply the human tissue. When applying the shammy covered recording electrodes on the skin, the impedance is in same magnitude as in Table 2.2-1. However the motion artefacts are still in the magnitude of 10-50mV

Stimulation Electrode Characteristics

When designing a stimulator it is important to know the impedance of the stimulation electrodes. For a current output stimulator the electrode impedance determines the volt­age range for the output.

The significance of chemical reactions in the electrode interface is different for the stimulation electrodes. In the recording electrodes the current in the electrolyte is van­ishing in comparison to the several milliampere of the stimulation pulse. The area and material of the stimulation electrodes also differs from the recording electrodes. The impedance is recorded using the set-up shown in Figure 2.2-2. A set of stimulation elec­trodes are applied to the skin of a test subject (subj.: RAT). A current generator, which is the stimulator Type 1 (described in Chapter 3), is connected to the terminals T1 and T2. The current is observed using the 100W resistor in series with the electrodes and the voltage over the electrodes are recorded simultaneously.

Figure 2.2-2 Test set-up

The voltage and current can be seen in Figure 2.2-3 and voltage versus the current is shown in Figure 2.2-4. A stimulation amplitude of 17mA was tolerable and resulted in full wrist extension. The current drop in the negative pulse phase (Figure 2.2-3)are due to imperfections in the stimulator. As it can be seen, the capacitance of the electrodes are not negligible, but causes a long transient after end of the stimulation pulse.

Measuring capacitance and resistance with a LCR meter (3kHz, 7V) yields a capacitan­ce in the range 10-50nF and resistance in the range 3-10kW (depending on the pres­sure on the electrodes. High pressure gives lower values). Like the recording elec­tro­des the electrode impedance is expected to be non linear, depending on the cur­rent applied.

Figure 2.2-3 Voltage and current in stimulation electrodes

Figure 2.2-4 Voltage vs. current in stimulation electrodes
(same data as Figure 2.2-3)

TOC

Stimulator Principles

The stimulator must be able to deliver the stimulation current needed to provide desired muscle contraction. The experiments have shown that a 40 mA biphasic pulse amplitude is ample for contraction of the extensor carpi radialis, where 20mA has been sufficient for most of the persons in the test panel. (The necessary maximal stimulation is found by trial in respect to the tetraplegics well-being and under advise from occupational thera­pists). From the previous section, the impedance of the stimulation electrodes was in the range 3-10kW. An impedance of 5kW requires 200V differential voltage over the stimu­lation electrodes if e.g. 40mA is to be delivered into the tissue.

As described in section 1.11 Electrical Stimulation, a bi-phasic charge balanced pulse shape is chosen. Each phase is selected to endure 300ms with a 300ms inter pulse in­terval. This complies with the work of Haxthausen [Haxthausen et al. 1991; Haxthausen 1992].

To minimize artefacts the stimulator output must be zero immediately after the end of the stimulation pulse. Furthermore, the output must be bi-phasic so that no net charge flows though the tissue. A monophasic pulse will charge the recording electrodes and thus give rise to blocking of myoelectric recording. As discussed in 2.4.7 Stimulation Response the stimulator output must be balanced to make the sum of the currents in the two stimulation electrodes equal to zero. Otherwise stimulation artefacts will be en­larged.

The stimulator can either be realized with a transformer in the output stage to provide the high voltage or it can be realized as an electronic output stage supplied with the needed high voltage. The principle of the first type, called Type 3, is illustrated in Figure 2.3-1. (Three types of stimulators are developed in the hardware section).

Figure 2.3-1 Type 3 stimulator. Output through transformer.

The advantages for the Type 3 stimulator are a galvanic separation and that the control electronics is at the low-voltage side, which simplifies the design. The galvanic separa­tion ensures that the output is fully charge balanced and that the stimulator outputs are floating with respect to the amplifier ground (re discussion in 2.4.7 Stimulation Re­sponse). The drawbacks are that a transformer adds both weight and volume to the system. It has a low efficiency if the stimulation is not maximal. Driven from a battery, the design of the driver circuit for the primary side of the transformer is critical if a good efficiency is desired. It is desired that the transformer has a small size. This implies re­duced area of the copper and consequently a resistance of the windings that can not be neglected. A high current is required in the primary side (in the range of an Ampere) and thus resistance in the primary side and the driving circuit may represent major power loss. The limited transformer size also implies that it can not be regarded as an ideal transformer, but parameters such as saturation, series inductance, parasitic capacitance etc. must be considered. This implies that the output current is not directly proportional to the input current, and thus the characteristics of the output current may differ from the expected. In 4.1.5 Type 3 Performance the output current for such a stimulator design is shown. Another drawback of the transformer coupled stimulator is that the capacitance of the electrodes and the series self-induction of the transformer will form a resonating circuit that can disturb picking up of the myoelectric signal. It has been observed, when using a stimulator with a large transformer (used by Haxthausen [Haxthausen et al. 1991]) , that oscillations are present after stimulation pulse. This has also been re­ported by personal communication [Sennels 1996].

In the concept shown in Figure 2.3-2 a step up DC-DC converter, supplied by the bat­tery, is generating an adjustable high voltage. (According to the prior calculation this voltage should be up to ±100V). The voltage supplies the output stage of a current out­put amplifier. The supply voltage is adjusted to match the actual electrode impedance and stimulation current requirements. Thus, the voltage drop over the output transistors can be minimized and thereby increase the efficiency.

The advantage, beside high efficiency, is that the transformer can be omitted. Thereby overall size and weight can be reduced. The transistor output stage gives the possibility of a better control of the output current. This is the principle in the stimulators called Type 1 and Type 2. The drawback is that there will be no galvanic separa­tion, which requires ideal matched current generators. This makes the stimulator as well as the power supply circuit more complicated than the one needed for the Type 3 stimulator.

Figure 2.3-2 Type 1 and Type 2 stimulator. Voltage controlled current source (VCCS)

To reduce the influence of a noisy stimulator output stage on the myoelectrical record­ing, diodes can put in series with the output as shown in Figure 2.3-3. Thus noise levels below the diodes on voltage will not disturb the recording. (The resistor provides a well defined voltage at the current generator output).

Figure 2.3-3 Diodes in output

Another possibility is to add switches that can short circuit the stimulator output after the stimulation pulse. The drawback of this solution is that the switches have to be reed-relays. No analogue switches can handle the high voltages, when supplied by ±3V. These relays are noisy, power consuming and voluminous. Both these methods has been tested. The reed-relay solution is used for the Type 3 stimulator. The diode solution has been tested with the Type 1 stimulator, but has not implied significant improve­ment of the myoelectrical signal recording.

TOC

Components of the Recorded Signal

The myoelectric signal consists of signals from several motor unit action potentials. The central nervous system can control the number of motor units, and the rate at which they are activated. Although there are indications that the central nervous system can control the individual motor units selectively [Rushton 1997], it is assumed that there is no vol­untary control over which motor units is activated in the muscle. For that reason the control property is desired to be the number of voluntary motor unit action potentials per time: The MUAP activity.

An example of a myoelectrical signal (subj.: KN) from a voluntarily contracted paretic extensor carpi radialis, recorded with the MeCFES amplifier using surface electrodes, is shown in Figure 2.4-1. Hum is filtered out as described in 2.7 Signal Processing.

Figure 2.4-1 Myoelectrical signal (subj.: KN)

It shows a stochastic signal over 3sec from a muscle of strength 2-3. Full range is ± 300mV.

Variation in Myoelectrical Signal

In signal processing, the question of stationarity of the signal in question is often of inter­est. To illustrate how the signal changes in time the RMS (root mean square) value of some myoelectrical signals (subj.: JBS) from a stimulated muscle is shown in Figure 2.4-2. On the upper graph the subject attempts no voluntary contraction and on the lower graph a maximum voluntary contraction is attempted. The signals are filtered by the first order transformed FIR filter presented in 2.7 Signal Processing. It shows that the change in RMS level when stimulation is applied is minor but present. It is possible that the stimulation trig­gers the reflex arch as discussed in 1.10 Myoelectric Signals. This could produce a sto­chastic component in the signal that is impossible to reduce by linear filter­ing. (The RMS window is sliding over the signal and calculated at each sample. The RMS is not valid for the first 0.3 sec.)

Figure 2.4-2 RMS values for signals from a relaxed
and voluntarily contracted muscle.

Stimulation is applied according to the dotted curve from 0-15mA. The subject has mus­cle-strength 2 in the extensor carpi radialis.

It can bee seen that the signal has not a constant RMS value and is therefore not station­ary [Møller and Sørensen 1992]. Voluntary contraction results in a doubling of the RMS value of the signal. Also it can be seen that the RMS level rises slightly when stimulation is applied.

Motor Unit Distribution Model

For the selection of a control strategy for the stimulation, it is of interest to know how many motor units that are under volitional control. Most important are the number of voluntary controlled motor units since they determines the statistical properties of the recorded myoelectrical signal, this number are determinant for the motor unit action po­tential activity. The myoelectrical signal from a normal muscle can be assumed to be band-pass filtered white noise if the number of MUAPs (motor unit action poten­tials) is large [Graupe and Kohn 1987]. This assumption might not be valid for a paretic mus­cle of e.g. force 3 since the number of motor units are expected to be few as exemplified in the following.

In a normal ECR (extensor carpi radialis muscle) the number of motor units is expected to be of the order of one hundred motor units (The extensor digitorum longus muscle has 130 motor units [Buchthal and Schmalbruch 1980]). If the strength of a paretic ECR is Q% of the normal muscle and if the paralysis has affected small and large motor units equally, then it is assumed that Q% of the motor units are under voluntary control.

The weight of a normal hand (subj.: RAT) is measured to be approximately 0.5kg. The centre of mass is estimated to be 10cm from the rotation point at the wrist. The torque needed to extend the wrist against gravity is then: Npar= 0.5Nm. A normal person should be able to exert a force of at least 10kg at center of mass. This means that the torque for the normal joint is: Nnorm=10Nm. For a subject with the ability to extend the wrist just against gravity (muscle force 3) the percentage of motor units according to the former should be approximately

         Q%=Npar/Nnorm*100%=5%                                                                              Eq. 2.4-1

Assuming that the paralysis has affected motor units of different size and type equally, it is believed that the percentage of intact motor units is equal to the percentage of re­maining force. Based on this theory the number of voluntary innervated motor units should be 5% of the total motor units before lesion for a muscle force of 3. This will then, based on the assumption of a hundred motor units in ECR, be around five motor units at muscle force 3, which probably is too low due to some degree of atrophy in the untrained paretic muscle. The conclusion is that the number of volitionally controlled motor units is very low in the weak paretic muscles. The single MUAPs should thus be recognizable in the myoelectrical signal.

Antidromic Nerve Block Theory

It can be (and has been) questioned whether the MeCFES principle is possible. The con­cern has been that the stimulation initiated nerve impulses propagates in the 'wrong' di­rection and block the voluntary nerve signals, i.e. antidromic nerve blocking. In the fol­lowing it will be discussed that this phenomena will only reduce the voluntary myoelec­trical signal slightly.

Since the nerve axon is locally symmetrical there is in theory no constraints on which di­rection an action potential in the nerve will propagate. The direction of propagation is thus determined by the way the nerve is exited. If the action potential originates in the end of the axon it will propagate towards the opposite end. If the nerve fiber is exited in-between the ends it will propagate in opposite directions towards both ends.

Figure 2.4-3 Theory of the collision block phenomena

When the nerve is stimulated near the muscle two action potentials will be initiated. One that propagates towards the muscle and will cause a twitch and the other that will propa­gate an­tidromic towards the cell body in the spinal cord. A voluntary action poten­tial will be coming from the cell body and propagate towards the muscle. If it encounters the antidromic action potential then the result will, as depicted in Figure 2.4-3, be an annihi­lation of both signals. Both the oncoming nerve signals are succeeded by an absolute refractive period, where the nerve at that location can not be stimulated or in this case pass the action po­tential on. After the refractory period has passed, other nerve signals can be transmitted again. This collision block or antidromic nerve blocking effect will thus only affect the first coming voluntary initiated nerve impulse.

The time interval where the antidromic nerve blocking will be able to annihilate a vol­un­tary nerve impulse is dependent on the nerve fiber length and the conduction velocity. The conduction velocity (CV) is defined as the speed of a propagating action potential in the nerve. The total time of propagating of the antidromic nerve impulse is the dis­tance from the origin of excitation to the nucleus of the nerve cell or the first oncoming action po­tential divided by the conduction velocity. The conduction velocities of the motor nerves n.ulnaris and n.medianus for adults typically be in the range 50-60 m/s [Stålberg and Falck 1993; Falck et al. 1994]. If the nerve is 0.5m-0.7m (estimated as the distance from the wrist to the spinal column in anatomical normal position). The maxi­mum time of an­tidromic propagation is then in the range of 8-14ms. In comparison the stimulation pulse interval at 16Hz is 60ms.

Figure 2.4-4 Tentative illustration of the influence of the
stimulation activity on the voluntary myoelectric signal

When a nerve fiber is stimulated the corresponding motor unit will contract and create a motor unit action potential. In the duration of the motor unit action potential voluntary action potentials transferred to the motor unit will have no affect. The stimulation will thus overrun later coming voluntary motor unit action potentials in the refractive period of the muscle fibers. The stimulation activity has to exceed a threshold level before nerve fibers are stimulated. From that level increasing stimulation activity will recruit more nerve fibers until the supra maximal stimulation, where all fibers are stimulated, is reached. In this interval the increasing stimulation will lead to decreasing recorded vol­untary myoelectric signal activity due to the antidromic nerve blocking effect as well as muscle response and stimulation artefacts. A tentative illustration of the situation is illus­trated in Figure 2.4-4.

The effect of stimulation and the effect of antidromic collision block can be seen by in­vestigating Figure 2.4-5 and Figure 2.4-6. The first graph shows stimulation where the subject tries to relax. The second is of the same recording at a time where the subject (subj.: JBS) tries to contract the muscle voluntarily. The subject has an extensor carpi radialis muscle of strength 1, one year after injury, when the recording took place.

Figure 2.4-5 Voluntarily relaxed with 15mA stimulation (subj: JBS)

Figure 2.4-6 Voluntary contraction with 15mA stimulation

The stimulation amplitude is 15mA, which produces a strong force (it is not evaluated whether supra maximal stimulation is achieved). On the first graph the stimulation re­sponses are seen as a regularly train of similar shape. Comparing the two graphs it can be seen that voluntary myoelectric activity is present after 20 ms. The first 20ms after the pulse is dominated by the stimulation response. This implies that the antidrome nerve­block is not preventing the voluntary nerve signals to be recorded. The conclusion is that antidromic nerve blocking does not obstruct recording of voluntary myoelectric signals (from the upper limb) at low stimulation frequency but must be considered, and that the stimulation will have an inhibiting effect on the voluntary myoelectrical signal.

Spasticity

Detecting whether the myoelectrical signal is originated by a spasm or voluntary con­traction is a severe problem. The myoelectrical signal from spasms is alike the voluntary myoelectrical signals. Signals from spasms tend to be very strong (based on two per­sons, who are able to provoke a spasm). One example from subject LP, can be seen in Figure 2.4-7 that can be compared with Figure 2.4-8, where a pure voluntary contraction is showed (The signal is blanked as described in 2.7 Signal Processing. The signals are normalized with same factor).

Spasms will therefore be interpreted by the MeCFES as a voluntary control signal and cause high stimulation. This might provoke the spasm further more resulting in an uncon­trollable stimulation. This can only be stopped by turning the device off.

Figure 2.4-7 Spasm from LP

Figure 2.4-8 Voluntary contraction from LP

There are two things that must be investigated if spasms show out to be a major prob­lem. If the myoelectrical signal from spasms is significant higher than voluntary myo­electrical signal, then the control algorithm could be modified to turn off stimulation when the myoelectrical signal activity exceeds a threshold, typically for spasms. Another possibility is that that functional electrical stimulation tends to reduce spasticity as it has been reported [Petersen and Klemar 1988].

Motion Artefact Recordings

The conditions under which the amplifier is to be used imply that the recording elec­trodes are exposed to mechanical actions. Mechanical actions on the electrodes, such as varying pressure against the skin [Webster 1984], are causing motion artefacts due to changes the half-cell potential. To illustrate this Figure 2.4-9 shows the signal from a relaxed muscle (subj.: RAT) where the one electrode is tapped with a finger (at approxi­mately time marks 900,1600 and 2400ms). The record­ing is made 5 minutes after elec­trode application using the MeCFES amplifier.

Figure 2.4-9 Motion artefacts (Ag-AgCl electrodes)

Recalling section 2.2 Electrode Characteristics, the motion artefacts can have peak value that exceeds the ±600mV input range for the amplifier by at least ten times. The recorded signal is, for that reason, clipped.

Spontaneous Activity

It has been noted that in recording signals from a relaxed muscle, (subj.: JBS), there are some single signal spikes or impulses. These seem to occur spontaneously.

Figure 2.4-10 Background noise with electrodes

As it can be seen at Figure 2.4-10 large bursts appears at the time marks 6.3 and 8.2-8.5 seconds (no blanking is used). A zoom at these bursts can be seen in Figure 2.4-11. This recording was made using the amplifier developed by Sennels [Sennels 1996]. By ob­serving the signal over longer periods it seems that these bursts appears at random inter­vals. It is not known whether or not it is external noise or signals from the subjects mus­cles is not certain. It is known that the subject has spasms and it seems most likely that it is sponta­neous activity in the muscles, although it cannot be excluded that the bursts are due to motion artefacts.

Figure 2.4-11 Zoom at first noise pulse

This phenomenon has been noticed in another tetraplegic but has not been investigated further.

Stimulation Response

The recording electrodes are located in the same area as the stimulation electrodes. For that reason the stimulation will result in a signal component at the recording electrodes known as stimulation artefacts [Merletti et al. 1992]. The stimulation initiates a syn­chronized contraction of a number of motor units the compound motor unit action poten­tial (CMUAP), which is the muscles electrical response to the stimulation. The CMUAP can be a magnitude higher than the myoelectrical signal but may vary due to fatigue. (That property can be used as a fatigue indicator [Mizrahi et al. 1994]). The simulation artefact and the CMUAP are mixed and synchronized . The combination is denoted in this text as the stimulation response.

The large ratio between the stimulation response and the myoelectrical signal is a prob­lem for the recording of the myoelectrical signal. The magnitude of the myoelectric sig­nals typically less than 0.5mV. The stimulation current can cause voltages at the record­ing electrodes up to magnitudes of 100V in the worst case. When special care is taken to minimize the fault current from the stimulation electrodes to the recording electrodes (see 2.8.1 Electrode Placing) the stimulation response can be reduced. Careful electrode plac­ing can bring it below 0.5V at 15mA stimulation. The simulation artefact is thus the dominating component in the stimulation response. The conclusion is that the stimula­tion response can be at least 60dB above the myoelectrical signal level. For that reason the amplifier must have a fast recovery from overloading of an impulse at the input.

To protect the input of the myoelectric amplifier the inputs are clamped to a level near zero (the MeCFES amplifier is clamped to ±3V) with a resistor in series. This implies the pos­sibility for the stimulation current to flow through the recording electrodes via the clamping circuit to the return ground. A model of the configuration is illustrated in Figure 2.4-12.

Figure 2.4-12 Simplified model of the stimulation current flow

The two current generators I1 and I2 are representing the stimulator output. If the voltage at the recording electrodes exceeds the clamping voltage, the amplifier input will be equivalent to impedance in the current return path. (There will be a return path for the current from one electrode to another. This path might be the common ground for the system as well as the power supply. The model is simplified by using ground as return path). Two cases are considered:

In the first case we assume that the two current generators are not perfectly matched i.e. I1¹-I2. Since the sum of currents in a closed system is zero, the current difference must be equal to the current flowing into the electrodes and through the return path to ground: I1-I2= Ie1+ Ie2. If Ie1 =Ie2 we can call it a common mode stimulation artefact.

In the second case we assume that I1=I2.but the current flow in the tissue, in conjunction with the electrode placement, results in a potential difference of the recording electrodes exceeds the clamping voltage. In this case there will flow a current through the one re­cording electrode to the return path (e.g. ground) and out of the other electrode. That means that Ie1=- Ie2. We can call this a differential mode stimulation artefact. Worst case is where the recording electrodes are placed on a line between the stimulation elec­trodes.

The Type 3 stimulator is a method to realize the current generator that avoid common mode stimulation ar­tefacts. The output is galvanic separated from common ground. (The two generators are realized as one current generator). Elimination of dif­ference stimulation artefacts, by adjusting the electrode positions, has turned out not to be possible in practice. The dimension of the electrodes and the physiological demands limits the freedom of the electrode placing. (The stimulation electrodes must be placed on the right motor points and the recording electrodes must be placed where the desired myoelectrical signal is present).

If current flows into the electrodes these will be charged due to the change of half cell potential and the capacitance (see 1.12 Electrodes). This will prolong the stimulation artefacts since the electrodes will behave like the high pass filter described in 2.5 Signal Amplification. If the part of the stimulation current that flows into the re­cording elec­trode is not charge balanced the result, as illustrated in Figure 2.4-13, re­cording of the myoelectrical signal will be impossible as long as the electrode potential saturates the amplifier.

Figure 2.4-13 Transients in electrodes caused by stimulation responses
of different shapes. Some can saturate the amplifier.

The figures show that if the stimulation artefact is monophasic the transient in the elec­trodes can saturate the amplifier and thus temporarily disable amplification of the myo­electrical sig­nal. For a perfect charge balanced biphasic stimulation artefact this effect is reduced (but not eliminated). This is an important technical argument for using a biphasic charge balanced stimulation pulse form.

If the half-cell potential of the electrodes are uneasily changed due to a low solubility product of the ions and the capacitance is low, which is the case with Ag-AgCl elec­trodes, the decay time for the transient is reduced. The properties (impedance & half cell potential) of the recording electrode pair when applied to human skin will not be identi­cal. As described in 2.5 Signal Amplification, common mode stimulation artefacts can converted to a difference signal due to different transient courses for the electrodes). For that reason common mode stimulation artefacts must be avoided de­spite the amplifier has a high common mode rejection ratio.

 In summary the stimulation current is a serious noise source. To reduce the noise and ensure proper amplification of the my­oelectrical signal the following must apply for the stimulator and the electrodes.

The stimulation impulses must be charge balanced and biphasic.

The stimulator outputs must be floating, i.e. have infinite impedance to common ground.

The stimulator outputs must be very near identical ideal current generators.

The recording electrodes must be placed in a way that differential mode stimula­tion artefacts are reduced and the recording and stimulation electrodes can not be the same.

Inherent Noise

Both the electrodes and the amplifier generates filtered white noise. For the MeCFES amplifier the in­herent noise when the inputs are short-circuited to ground is 0.4mVRMS(Figure 2.4-14). This is 20dB less than the noise from Ag-AgCl electrodes, which is approximately 4mVRMS (described in 2.2 Electrode Characteristics).

Figure 2.4-14 Inherent noise from the MeCFES amplifier when inputs are con­nected to ground (amplitude in mV).

In other words. The inherent noise from the electrodes is dominant.

Signal Model

The previous findings can be collected to a simple model of the signals included in the recorded signal. Only the voluntary myoelectrical signal is of interest why all the other signal com­ponents are regarded as noise. The signal from the electrodes can be modeled as coming from six sources (Figure 2.4-15). Some of those sources are representing different proc­esses but are having the same characteristics and can for that reason, to


simplify the model, be regarded as one source. This applies to the sources Vs, Vn and Vsp , where these can be divided into two sources in order to obtain a more elaborate model.

Figure 2.4-15 A simple signal model

The signal sources and their expected values are:

Vm:         The voluntary myoelectric signal. This is a stochastic signal controlled by the central nervous system. The typical amplitude for a paretic muscle is below 0.5mVpeak.

Vs:          Stimulation response. This is a periodic signal with an amplitude depending on the stimulation intensity and the electrode placement. It comprises both the compound motor unit action potential and the stimulation artefacts. The com­pound motor unit action potential is depending on the stimulation amplitude as well as the state of the muscle, which for example can be how fatigued the mus­cle is. This stimulation artefacts are increasing with increasing stimulation am­plitude and is also affected by the electrode mounting and the characteris­tics of the stimulation pulse (see discussion in 2.4.7 Stimulation Response. An ex­ample of magnitude is a measurement of the stimulation response amplitude of 0.5V at 15mA stimulation. Careful rearranging of the electrodes can give lower values.

Vn :         Inherent noise. This includes the inherent (thermal) noise from both the elec­tronics and the electrodes. This is a strictly stationary stochastic signal. (This implies that is has a constant level). For the given MeCFES amplifier it is re­corded to 4mVRMSwhen using Ag-AgCl electrodes.

Vsp :        Spontaneous noise. This includes all kinds of stochastic non-stationary noise from the muscle. This applies mostly to signals arising from spasms. They are expected to be a strong intermittent signal. From provoked spasms the level is far above the level of a voluntary contraction. (It is estimated that levels can be up to 10mVpeak in worst case, these levels will cause clipping of the signal by the amplifier.

VMA:       Motion artefacts. This is transients caused by mechanical influences on the electrodes. This is a intermittent signal and therefore a non-stationary signal. As discussed in 1.12.3 Motion Artefacts they can be up to 10mVpeak regardless of the electrodes.

Vc:          Common mode signal. This is a 50Hz (60Hz) distorted sine appearing at the electrodes as a common mode signal. The origin is the capacitive coupling from mains to the user. If the MeCFES's common ground is coupled direct to the earth, the level at the amplifier input will be higher than when the MeCFES is 'floating' with respect to the electrical earth. With the common ground con­nected the electrical earth an amplitude of 0.2VRMS is measured.

Noise from the mains is capacitive coupled to the person and will be present as a com­mon mode signal at recording electrodes. The common ground for the MeCFES is cou­pled to the electrical earth. (The impedance of the latter coupling is depending on which de­vices the MeCFES has electrical connection to. For example a computer). To reduce the common mode signal the signal is amplified Gc times with a 180 de­grees phase shift and fed back to the body through the ground electrode (with impedance Zagnd).

The model is simplified and does not include the input impedance of the amplifier (which can cause conversion of some of the common mode signal into a difference signal, if the electrode impedance Ze1 and Ze2 are not equal).

TOC

Signal Amplification

The amplifier is a critical component since it provides the first processing of the weak high impedance myoelectrical difference signal. Therefore some aspects in designing the stimulator will be discussed in the following.

In normal surroundings there is a powerful electrical field from the mains that sur­rounds us. This 50Hz (in Europe) field may create common mode signals, from the user, of several volts. To suppress this field it is necessary to have a high common mode rejec­tion in the amplifier. It is well known [Webster 1992] that the common mode signal can be suppressed by amplifying and feeding it back to a grounding electrode, on the sub­ject, with a phase shift of p. This is called active ground.

The input impedance (common- and differential mode) must be very high since the myo­electrical signal has a high impedance. If the electrode impedance is unequal then com­mon mode signals will be converted to difference mode signals. The amplifier is a three terminal device on the input side with the two inputs and the ground as in Figure 2.5-1 with respect to the common mode signal. For a common mode signal the electrode im­pedance, convert common mode noise to difference mode noise which can disturb myoelectrical signal measurement.

Figure 2.5-1 Simplified equivalent diagram of amplifier input

The difference signal Vd arising from the common mode signal Vc is

                                                            Eq. 2.5-1

Where Zc is the common mode impedance of the amplifier (including cables) and Ze1 and Ze2 are the electrode impedance. For the approximation it is assumed that the com­mon mode impedance is much larger than the electrode impedance, Ze<<Zc.

To amplify the less than 0.6mV myoelectrical signal to a level that can match a 3V ana­log - digital converter, the amplification of the amplifier must be in the range of at least 74dB. The inherent-noise of the amplifier related to input must be less than the inherent noise from the electrodes (less than 4mV).

Since the inputs of the amplifier is so easy accessible, these must be protected from electrostatic discharging as illustrated in Figure 2.5-2. This protection must, when in­active, have an impedance in same magnitude as the input impedance of the amplifier to leave the myoelectrical signal undisturbed.

 

Figure 2.5-2 Protecting inputs from electrostatic discharge

This will also protect the inputs against signals from the high-voltage stimulator output. The protection can be provided by clamping the inputs. Diodes are suitable for that pur­pose. Clamping to ground, using common diodes, should be avoided due to difference in dynamic resistance, Rd, given by Eq.2.5-2.

                                                                                            Eq. 2.5-2

Where Vt =26mV and I0 in the range of nA-mA are diode specific constants. If the elec­trode half cell potential Vd=1-10mV the dynamic resistance can be less than one MW. This can cause conversion of common mode noise to differential mode noise ac­cor­ding to Eq.2.5-1 using the diodes dynamic resistance as the amplifier input impe­dance. Therefore clamping to power supply is more attractive. However the reverse cur­rent in the diodes must then be considered.

As discussed in the previous section the signal at the amplifier input is mixed with pulses of far greater amplitude than the myoelectrical signal of interest. A typical elec­tro­myo­graphy amplifier is configured as shown in Figure 2.5-3 [Hall and Munday 1994;Saridis and Goothe 1982;Ylvisaker 1986].

Figure 2.5-3 Simplified schematic of a typi­cal
 conventional myoelectric amplifier

It comprises of a differential input stage, a high pass filter to remove DC-offset and a high gain output stage. Stimulation artefacts will flow into the high-pass filter and the transients created here can cause saturation of the second stage and thus extend the dura­tion of the stimulation artefacts. The recovery time is here defined as the time from the amplifier input is exposed by a pulse till the amplifier output reaches the active area (i.e. between the supplies:-3 to 3V). The test pulse, after 20dB amplification in the input stage, is defined as an 1V, 1ms wide rectangular pulse. For an amplifier as in Figure 2.5-1, with gain 74dB, the response would be determined from the equation derived from adding the RC circuit response from two step functions.

                                   Eq. 2.5-3

For A=500 this yields

                                                             Eq. 2.5-4

Therefor the recovery time would take most of the inter stimulation period. To deal with this problem a switch [Minzly, et al. 1993;Mizrahi, et al. 1994] or a sample-hold circuit [Babb, et al. 1978;Howson and Heule 1980] can be applied before the filter. As an alternative the filter cut-off frequency can be raised (1kHz) well above the stimula­tion frequency [Haugland and Hoffer 1994]. CMOS switch and sample-hold circuits causes a charge injection in the signal path which give rise to new artefacts. Furthermore a high cut off frequency is inadequate for recording surface myoelectric signals for which it is commonly accepted that the information is present in the frequencies below 500Hz [Bilodeau, et al. 1993;Kwatny, et al. 1970]. Choice of the cut off frequency is a compro­mise between minimizing the recovery time from stimulation artefacts and the re­covery time from changing the DC offset. The recovery time is here defined as the time from the end of a stimulation pulse until the circuit is no longer saturated.

Reduction of artefact transients in the amplifier circuit is essential for subsequent proc­essing. As long as the signal is not clipped or by other means distorted in the analogue circuits, further suppression of artifacts can be done by means of digital signal proc­essing. This can be done by blank­ing [Holländer 1987;Kitzenmaier and Boenick 1993;Nikolic, et al. 1994;Popovic, et al. 1993] and filtering [Haxthausen, et al. 1991;Sennels, et al. 1995].

This artefact problem has been leading to the development of the MeCFES amplifier as described in Chapter 3.

TOC

Signal to Noise Ratio Definition

To evaluate the efficiency of different signal processing methods a measure of the signal to noise ratio (SNR) is required. The dilemma is that the signal is never known because we only have access to the output of the system and the applied stimulation pulses, by which it is influenced. The input signal is the "decision" which the subject is control­ling. (This reflects the desire for a movement). This is not possible to measure objec­tively. An illustrative input-output model (Figure 2.6-1) can be seen as a black box with the inputs: mental decision and electrical stimulation. The output is the resulting myo­electric signal. The only exact measurable signals are the stimulation and the output.

As described in the 2.4.3 Antidromic Nerve Block Theory and 2.4.7 Stimulation Re­sponse the stimulation will have an inhibitory effect on the voluntary myoelectrical signal and add more noise to the output signal with increasing stimulation amplitude. It is therefore reasonable to define the signal to noise ratio at a fixed stimulation level. First we assume that the noise is non-correlated with the signal, (which not applies to the part of the stimulation that inhibits the signal i.e. the following is only an approxi­mation). Then assuming that when the subject is relaxing the muscle, the output (Mrel) is repre­senting only the noise component. When the subject concentrates on full contrac­tion of the muscle the output (Mvol) comprises the sum of the noise and the signal. Then the signal to noise ratio (S/N) can be defined as




Figure 2.6-1 Input-output model. The voluntary myoelectrical signal (MES) is generated in the voluntary controlled motor units.

                                                             Eq. 2.6-1

where Mvol is the myoelectrical signal measured under full voluntary contraction and Mrel is the signal from the voluntarily relaxed muscle. The root mean square value of the signal is denoted by RMS.

A feasible way to measure this has been found (by experiments). The participant is asked to perform maximal contraction and then the stimulation is applied. Mvol is re­corded immediately after the onset of the stimulation. Immediately after recording Mvol , the subject is asked to relax. Then Mrel can be recorded. To avoid the effect of fatigue the two measurements should be accomplished within a few seconds.

TOC

Signal Processing

This section describes the signal processing for the MeCFES.

Analogue Signal Conditioning

The signal processing starts in the amplifier where the signal is amplified and filtered using the MeCFES amplifier, which is described in detail in section 3.2 Amplifier Circuit. The amplifier is designed to minimize the stimulation responses by having a fast recovery time and clip signals levels exceeding ±600mV. A change in the electrode offset exceed­ing ±0.1mV starts the fast offset compensation. In this way both motion artefacts and stimulation responses are reduced.

The filtering of the myoelectrical signal is determined from the following observations of the frequency content. A typical frequency spectrum (FFT) of a myoelectrical signal is shown in Figure 2.7-1 using an amplifier with 0-1.5kHz bandwidth and sampling at 3.3kHz. Here it can be seen that frequencies above 500Hz are below -20dB.

Figure 2.7-1 Frequency spectrum of a myoelectrical signal
(normalized by peak value. Subj.: RAT)

Applying stimulation to the full voluntary contracted muscle results in the spectrum seen in Figure 2.7-2

Figure 2.7-2Myoelectrical signal and stimulation (Normalized)

As it can be seen the harmonics of the stimulation response are dominating at fre­quen­cies above 500Hz. To comply with the Nyquist criterion before sampling it has been chosen to provide the MeCFES amplifier with a 500Hz 2nd order low pass filter to reduce the frequency components above 500Hz. The signal is then sampled at 2kHz without sig­nificant aliasing since the signal at 1kHz (half the sampling frequency) will be further damped 12 dB by the filter. Since the recorded signals contains impulses it is important that the filter has the best possible impulse reproduction. A Bessel-filter is chosen since this filter complies with this requirement [Langvad 1987].

Sampling and Pre-filtering

After being amplified and processed by the amplifier the signal is sampled at 2kHz. The SNR (signal to noise ratio) of the amplifier can be at the most 60dB (maximal input signal level is 600mVpeak and the noise is 0.4mVRMS excluding electrodes and 4mVRMS including electrodes). For that reason a 10 bit resolution (60dB) is sufficient for the myoelectrical signal sampling.

In Figure 2.7-3 the stimulation response (subj.: JBS) from a voluntarily relaxed, but stimulated muscle is shown. The stimulation pulse starts at the time zero and has a total duration of 900ms. As it can be seen that the amplifier is saturated for approximately 10ms after the initiation of the stimulation pulse. This is a typical response. Since this interval does not contain information of the voluntary myoelectrical signal the samples are blanked in this interval i.e. values are put to zero. In the signal processing the 10ms blanking is equal to discarding 20 samples. This reduces the samples by 20% and thereby saves computing resources without loss of information.

Figure 2.7-3Stimulation response (subj.: JBS)

In Figure 2.7-4 the subject performs voluntary contraction during the stimulation. The subject has a very weak muscle of force 2. Comparing with Figure 2.7-3, one finds that the voluntary myoelectric part is much smaller than the stimulation response.

Figure 2.7-4Mixed signal (subj.: JBS).

To remove stimulation responses a linear filter is applied. Haxthausen [Haxthausen et al. 1991; Haxthausen 1992] used a transformed 3rd order elliptical filter for the suppres­sion of the stimulation response. It has been found [Thorsen 1994] that this filter does not improve the SNR much compared to a fixed FIR (finite impulse response) filter (Eq. 2.7-1). Sennels [Sennels et al. 1997] compared adaptive filters to fixed FIR filters. It was found that adaptive filters with 6 non zero coefficients gave some improvement of the SNR compared to a fixed FIR filter.

The drawback of adaptive filters is that they require significant more computation than a fixed FIR filter. Considering the SNR improvement it has been decided to use a trans­formed first order FIR filter which requires minimal computation. The filtered myo­elec­tric signal m is given by the difference equation

                                                                                      Eq. 2.7-1

where n is the sample number, x is the sampled signal and p the number of samples in the period between consecutive stimulation pulses is p. (Gain factor leaves the RMS (root mean square) value unchanged for a stochastic signal).

By selecting the stimulation frequency to be an integer fraction of the power supply net frequency (which in Europe is 50Hz) we get the double advantage of the comb filter. It thus both suppress stimulation signals and 50 Hz hum. Experiments have shown that the stimulation frequency should be above 10Hz to obtain a constant muscle contrac­tion. It is desirable to keep the stimulation frequency as low as possible, since the amount of valid samples of the myoelectrical signal is increasing with decreasing stimulation fre­quency. For these reasons the 16.6Hz stimulation frequency has been chosen , a third of the mains frequency. (For countries with 60Hz mains frequency a fourth i.e. 15Hz will be suitable).

If the signal is contaminated by a high level of hum from mains the difference in the en­ergy before and after comb filtering is high. This is used for detecting if the signal is too noisy. This will be the case if the recording electrodes does not have proper skin contact. Then one or both the electrodes will have a high impedance which may cause common mode hum to be converted to a differential signal as described in 2.5 Signal Amplification.

MUAP Activity Calculation

After the signal to noise ratio has been improved by filtering, the myoelectrical signal can be converted into the control signal. This is done by estimating the MUAP activity, defined as the average voluntary MUAPs (motor unit action potentials) from the con­trolling muscle over time. Some different techniques for estimating the control signal have been considered e.g. autoregressive models [Hefftner et al. 1988] or the use of neu­ral networks [Costa and Gander 1993]. It is the believe that the advantages of such methods does not match the need for computation. For that reason, simple methods based on the energy of the voluntary myoelectrical signal have been chosen. Assuming that the myoelectrical signal is a white noise signal from a high number of independent generators with a stochastic distribution of the firing interval, an ARV (average rectified value) of the signal is often used. The use of ARV has a historic reason, since it was simple to realize in analogue electronic circuits. With the introduction of low cost digi­tal signal processors the RMS value of an interval (sliding window) of the signal has become another often used measure. In the context of signal process­ing, the RMS of the myo­electrical signal is more attractive since it applies to the com­mon calculation of the sig­nal-noise relation. Which of the two measures are used is not important since the RMS and ARV can be considered proportional [Hermens 1991]. Haxthausen [Haxthausen et al. 1991; Haxthausen 1992] used the ARV as a measure for the MUAP activity.

In this work has been chosen to use the RMS of segments of the myoelectrical signal as one method for estimation of the MUAP activity. It is practical to calculate the RMS for a block of samples, where a block is defined as the samples between two consecutive stimulation pulses. The control signal is estimated by low-pass filtering of the RMS (in quadratic sense). A method using ARV instead of RMS has been imple­mented in the MeCFES for compatibility with the work of Haxthausen.

A different approach will be described in the following. This is an ad-hoc method for estimation of the MUAP activity. As discussed in 2.4.2 Motor Unit Distribution Model the number of MUAPs are few. From 1.10 Myoelectric Signals it is known that the dif­ferent MUAPs do not contribute equally to the myoelectrical signal. The RMS value will be differently affected by different motor units and might thus not reflect the num­ber of MUAPs per time in the muscle, but rather which motor units are currently re­cruited. The argument is thus that if the MUAPs are few and changing in origin then the RMS over a short time interval of the signal will not be an accurate estimate for the firing activity. It is desired to keep this interval short to minimize the total delay of the system.

Figure 2.7-5 Selected MUAPs

Figure 2.7-5 shows MUAPs from a normal muscle (subj.: RAT) at a very weak contrac­tion (RMS of myoelectrical signal = 6mV) selected by visual inspection. The duration of the MUAPs is in the range of 6-10ms. According to 1.10 Myoelectrical Signals, the electrical properties of the motor unit does not change over time. At this low level of MUAP activity, the individual MUAPs are visible in the electromyogram as discrete action potentials. At levels of 20mV the hand starts extension against gravity.

Let S=(s0,s1,s2,…,sm-1) be the MUAP that is going to be detected. If the MUAP occurs at different times in the signal m and if the MUAPs do not overlap, then the MUAP can be detected using a filter with the impulse response h=a(sm-1,sm-2,…,s0) [Justesen and Forchhammer 1992].

Figure 2.7-6Idealized MUAP

A standard MUAP is generated by smoothing the sequence [0,-1,0,1,0](using the MATLAB™ interpolation command interp). This representation is arbitrary chosen, based on inspec­tion of the recorded MUAPs. A time interval of 3msec between the bottom and the top of the pulse is used. The frequency response for the filter together with the spec­trum of a 20mVRMS myoelectrical signal is shown in Figure 2.7-7.

Figure 2.7-7 Frequency response for filter and signal spectrum

As it can be seen in Figure 2.7-7, the filter is a band pass filter. If the filter output us an extreme of a certain level, it is assumed that a MUAP is encountered. Finding local extremes is not a trivial computing task. To simplify the computation the MUAP acti­vity was chosen to be calculated as the number of samples in the filtered signal with a mag­nitude that exceeded a certain threshold tr according to Eq. 2.7-2

                                     Eq. 2.7-2

where y is the filtered myoelectrical signal m and N is the number of samples in the block.

The following results are based on weak contractions of a normal muscle (subj.: RAT). The signals are produced by measuring the RMS value of the myoelectrical signal from the muscle. The subject tries to retain a steady contraction with the visual feedback from a RMS meter at different RMS levels of the myoelectrical signal.

In Figure 2.7-8 the threshold counting method is compared to the RMS value. The values are calculated over 3 seconds of the most constant contraction. The left curve shows the relation between the attempted contraction and the actual root mean square value of the myoelectrical signal RMS(m) . The right curve shows the TC value (TC(m)) of the same myoelectrical signal m where the threshold tr is arbitrary chosen to two times the RMS of the inherent noise (which is 4mV).

Figure 2.7-8 RMS values compared to the threshold counting (TC)

The noise immunity of the two methods are compared at low myoelectrical signal levels. Figure 2.7-9 shows the RMS values of filtered myoelectrical signals (m) added with noise. The RMS values of m are 4,6,8,12,20,40,60 and 80 mVRMS. The noise is the re­corded inherent noise amplified to 0,10,20,30,40 mVRMS. Figure 2.7-10 shows the TC values with threshold 60mV of the same signals. The five noise levels are seen as the five curves with increasing level.

Figure 2.7-9 RMS(m+noise) vs RMS(m)

Figure 2.7-10 TC(m+noise,60)

vs. RMS(m)

At low levels, the TC is more noise suppressing than the RMS method as illustrated by Figure 2.7-11. Here the level of noise+signal relative to the signal is calculated for the 6mVRMS and 12mVRMS signals. The horizontal axis shows the noise relative to the signal. On the vertical axis the RMS of the filtered noisy signal or the TC of the noisy signal is relative to the pure signal.

Figure 2.7-11Noise immunity

The TC method is calculated with thresholds 30mV and 60mV. When the MUAP activity is low, this method is better in suppressing the noise. Finding the threshold is done by trial and error.

The threshold counting method is implemented in the developed device as another method of estimating MUAP activity . The method is applied to each block of samples.

Calculation of Stimulation Amplitude

The MUAP activity is calculated over each block of data. A block is defined as the signal between two consecutive stimulation pulses. The MUAP activity is filtered by a low pass filter. The filter used calculates the mean value over 20 stimulation periods (corresponding to 1.25sec) of the MUAP activity.

Alternatively an IIR (infinite impulse response) low-pass filter can be used. The imple­mented filter has the form:

                                                                                                   Eq. 2.7-3

(Where y is the output and x the input, n denotes the sample number, a and b are constants determining the gain and frequency response).
The stimulation amplitude is calculated by a piece-wise linear function.

                                          Eq. 2.7-4

Here Z is the filtered muscle activity, MAofs, IGain and IOfs are constants that can be adjusted by trial and IAmp is the stimulation amplitude. The amplitude is limited by IMax for safety reasons.

Summary of Signal Processing

One of the signal processing configurations is illustrated in Figure 2.7-12. The left col­umn is showing different stages of the signal processing and the right column is an corre­sponding example of the signal. The recording electrodes are placed in a way that the stimulation response are minimized. The amplifier band pass filters the signal and sup­presses stimulation responses before sampling. The comb filter removes the stimula­tion responses and the 50Hz periodic hum from the mains. Then the signal is trans­formed into a control signal for the amplitude of the stimulator, which outputs a pulse with 60ms interval. In the shown control signal, formed by RMS method the, signal to noise relation is only 5dB.

Figure 2.7-12Signal flow

TOC

Electrode Usage

This section describes the electrodes and their use in the experiments. Furthermore a suggestion for a electrode-mount is presented.

Electrode Placing

The electrodes used for recording are Ag-AgCl electrodes with electrode paste (Blue sensor from Medicotest A/S, Denmark) and the stimulation electrodes are silicone­rubber-carbon (from ASAH Medico A/S Denmark). Electrodes are placed as shown in Figure 2.8-1. The stimulation site is found by trial and error where the best compromise between a strong wrist extension and a minimal finger extension. The distance between the electrodes have been varying from 5 to 10 cm. The exact placing is individual and varies from person to person and must be found by trial and error. Precise positioning of stimulation electrodes is very important if a pure wrist extension is desired. Elec­trode displacement of more than a few millimeters can cause undesired stimulation of finger extensors.

Figure 2.8-1 Electrode position

After these motor points have been found the recording electrodes are placed perpen­dicular to a line between the stimulation electrodes. This is a compromise between plac­ing them in equal distance from the stimulation electrodes and being able to record the myoelectrical signal. This is also done by trial and error. In practice the area in which the recording electrodes can be placed is limited by the geometry of the arm and size of electrodes. If large stimulation artefacts occur the recording electrodes may be repositioned.

To reapply the electrodes an overhead transparency is used with advantage. The natural brown spots in the skin are suitable as fix-points for the transparent where both spots and electrode position is transferred to, using an overhead pen. The active ground electrode is placed anywhere away from the area bounded by the other electrodes. A suitable site is at the flexor side of the arm or the elbow.

It is very important that the recording electrodes have matched impedance. Otherwise there will be a risk that common mode signals will be transformed into difference mode signals. Especially at 50 Hz, where the common mode noise is high this can cause prob­lems. The difference signal will approximately be the difference in electrode im­ped­ance(»104W) divided by the input impedance (>10GW common mode for the used am­plifier) times the common mode signal as discussed in 2.5 Signal Amplification.

The coupling from mains to the electrode wires is assumed to be negligible compared to the coupling to the body. Thus wire impedance should be included in amplifier impe­dance.

Connections

Figure 2.8-2 Schematic drawing of the electrode placement and the cabling

It is important that the recording electrodes are connected via a shielded cable to the amplifier. Any unshielded sections of the cable is twisted in order to reduce noise. The shield of the cable can be used as lead for active reference (ground). This is negative feedback of the common mode signal from the amplifier and as such not is to be con­nected to the actual ground or frame of the electronics. The principle in the connection is outlined in Figure 2.8-2.

Hand Stimulation Technique

Stimulation of the hand can be accomplished as shown in Figure 2.8-3 whereby the hand closes as a clenched fist. One electrode is placed over the pisiform bone and the other is placed on the back of the hand over the thenar space at the dorsal side of the hand. The first electrode stimulates the palmar branch of nervus ulnaris which controls the small finger flexors of the hand. The second stimulates the flexion of the thumb (Flexor pol­lices muscle, adductor pollices muscle etc.).

Figure 2.8-3 Hand stimulation

This electrode configuration has been evolved during the work with the functional elec­trical stimulation experiments. It requires that the adductor pollices and I. interossi mus­cles, in particular, are innervated.

Electrode-mount

The concept of using adhesive electrodes is not very feasible for a device which is sup­posed to be used on daily basis. It is desirable to have a device that holds the electrodes on the right position. This electrode-mount should be easy to put on and off. Prefera­bly by the tetraplegics themselves. It should place the five electrodes and hold them firmly to the skin. The wires to the electrodes should be hidden in the electrode-mount and connected to the MeCFES by a single wire.

The hand stimulation results (see Chapter 4) have shown such good results that it has been decided to look for a solution that can place two stimulation electrodes on the hand as shown in Figure 2.8-3. One between the thumb and index finger on the back of the hand and one at the end of ulna on the palmar side of the hand.

The electrode-mount shall put the recording electrodes over the extensor carpi radialis and optionally also the stimulation electrodes for same muscle as in Figure 2.8-1.

The electrode-mount is intended as a splint that covers the electrode placement sites in­cluding part of the hand. This requires a linkage at the wrist. Designer Marianne Thorsen, Danmarks Designskole (Danish School of Design) has been a great help with the design of a concept for the electrode-mount. A model made of Aqua plast™ is shown in Appendix D. This material is available in plates. It can be formed at tem­peratures above 60 degrees Celsius. Hot water is suitable for the heating. After cooling it maintains the form.

It is composed around a straight rectangular piece, Figure 2.8-4, the back-piece that runs along ulna. At the proximal end there is a plate formed around the arm. This part con­tains the recording electrodes and optional stimulation electrodes. A strip that wraps around the arm just above the wrist serves for stabilization. The distal end of the back-piece is connected with a hinge to the hand part of the electrode-mount. This wraps around the middle part of the hand from the thenar space and into the palm, covering the pisiform bone. It locates an electrode at the thenar space and near the pisiform bone.

Figure 2.8-4 Components of electrode-mount

The construction has an important feature. When laying the electrode-mount on a table it turns with the back-piece down, due to its center of mass. In this position the arm and hand can be put into the electrode-mount since the electrode-mount is an open con­struction. The arm must be partly supinated. This movement can be done by a tetraple­gic (subj.: KN). (See test subject description later on). The electrode-mount is lined with soft water rejecting foam to avoid wetting when the electrodes are wetted. (Electrodes has not been incorporated in the prototype, but the electrodes from first subsection has been taped in for the test).

The electrode-mount needs a closing mechanism to hold the electrodes firmly to the skin and maintains the position. One solution is to lace it up with elastic lace. A ring at­tached on the lace with a size so it can be caught by a paralyzed thumb (or index finger).

The ring is used to guide the lace around knobs on both side of the opening of the elec­trode-mount and finally used for tightening of the lace around the last knob. This re­quires some control of the position of the opposite hand and the ability to catch the ring. This action has not been possible for KN for which the electrode-mount was de­signed. It has unfortunately not been possible within the economical frame of the project to have a prototype with electrodes produced.

A design student (Birgitte Bennike, Danmarks Designskole) has designed a electrode-mount with a more attractive cosmetic design. This is supposed to be produced in a stronger type of plastic. Photos of the model can be seen in Appendix D. The outline for the box containing the MeCFES is included as well.

Electrode Embodiment

A suggestion for electrode incorporation in the electrode-mount is as follows (Figure 2.8-3). Each electrode is placed between the lining and the electrode-mount shell. For this long term use of the electrodes it is important, besides the properties discussed in 1.12 Electrodes that the electrode material is bio-compatible, chemically and me­chani­cally stable (e.g. corrosion could lead to decreased performance). There must be a hole (Ø5mm recording and Ø20mm stimulation) in the lining over the electrode. The elec­trode can be covered by a water absorbing material being in between the skin and the electrode. The water serves as a conductor (electrolytic carrier) between the electrode and the skin. Tests have shown that shammy is good for that purpose. I has a long dry-up time and is a good skin interface.

Figure 2.8-5 Electrode build in

The material should perform a spring-like pressure on the skin and provide good con­tact. If the lining is not waterproof, the material should be surrounded by a rubber ring (gas­ket) to impede drying up and preventing the lining from getting wet.

Hardware and Software

This chapter describes the hardware and software developed and the test set-up

The keywords for the design of the electronic parts of the system besides functionality, are minimum size, weight and a low power consumption. Power consumption is closely connected to the size and weight since it determines volume of batteries which are among the heavy and space consuming parts. To minimize the size it has been attempted to use few components and choose the most power efficient techniques as possible. The availability of low price components limits the design possibilities.

Through several test circuits and three prototypes, the system has been evolving to comprise an amplifier, stimulator, digital signal processor (DSP), power supply/battery manager including rechargeable batteries and wires for electrodes. This system is called MeCFES (Myoelectrical Controlled Functional Electrical Stimulator). There are four different printed circuit boards which are the amplifier-, the stimulator-, the digital signal processing (DSP)- and the power supply- board. Especially for the DSP system, state of the art devices has been used causing some problems with faulty devices and supplier problems. Developing, manufacturing and testing the hardware and software has been occupying more than 2/3 of the project period leaving only half a year for systematic trials with tetraplegics. Sections 3.1 through 3.5 are describing the MeCFES hardware. Section 3.6 describes the method and set-up for the evaluating the perform­ance of the MeCFES. The software developed for the system is described in section 3.7 and 3.8. It consists of a DSP program for the MeCFES and a host program for occa­sional communication with the DSP. Only a brief description is provided for this com­prehensive work to preserve proprietary rights. Finally section 3.9 is a summary of the MeCFES specifications.

TOC

Interface

Tetraplegics have very limited possibilities to turn knobs and push buttons to operate the MeCFES. To comply with this, it has been decided that the device shall have only one push-button, by which the user can operate the MeCFES. To inform the user of changes in the states of the device a sound transducer is used. Figure 3.1-1 shows the on/off procedure using the button.

Figure 3.1-1 Start-up procedure

Sound signals with different characteristics informs the user as listed in Table 3.1-1.

Single tone

The device is turning on and initializing

Rising scale of tones

The device goes ready and operating

Descending scale of tones

The device is shutting down

Two alternating tones repeating

An error has occurred. Turn off the device. Try again, recharge batteries or call service.

Repeated buzz

Battery low. Shut off the system and recharge batteries.

Table 3.1-1 Sound signals

This restricted user interface implies that no adjustments of parameters can be done. These adjustments must be done initially with a connected host computer as described later. The demands for stimulation intensity and myoelectrical sensitivity might change from time to time. For that reason the program has been prepared for implementation of an automatic calibration procedure. This should be executed during the initialization phase after power up of the device. A suggestion for the auto calibration procedure is follows.

After the button is pressed, system powers up and starts a self-test. If the test is passed, an initialization mode is entered and the user is informed by a sound signal whereupon the stimulation increases slowly (limited by a pre-set maxi­mum value). The user ob­serves when the maximum contraction is achieved and confirms by pressing the button and the stimulation stops. This procedure will set the stimulation gain and the MeCFES will confirm this by a sound signal.

There is an option of connecting a host computer (BM PC) to the MeCFES. This re­quires a special cable from the parallel port of the host computer to the MeCFES. Com­munication with the MeCFES is then possible using the developed host computer soft­ware. After the manufacturing of the MeCFES the DSP program has to be down­loaded to the device from a host computer. The program is then stored in the FLASH memory (an erasable non volatile integrated memory circuit) of the MeCFES. Once programmed, the system will not need the host connection to operate.

The host interfacing option gives access to the following actions:

TOC

Amplifier Circuit

As described in 2.5 Signal Amplification the amplifier is the most important link in the signal processing. The developed amplifier is patent pending and a complete schematic can be found in Appendix B. The special property of the amplifier is that high-pass filtering is achieved using a non-linear feedback loop.

Amplifier Principle

Figure 3.2-1 is a model showing the principles of the amplifier. The electrodes, denoted by (1), pick up the signal. Each of the electrode signals are limited (2) to reduce stimu­lation artefacts and to protect the pre-amplification circuit (3). The pre-amplifier (3) has a difference gain of A1 and transforms the high impedance difference signal into a low impedance signal. The pre-amplifier output (3) is added (4) with the negative output (12) from the feed-back network which provides an estimate of the DC-offset at the pre amplifier output (3). The offset regulated signal from the addition (4) is fed to a post-amplifier (5) with gain A2.

Figure 3.2-1 Principle of the amplifier

The feedback network is divided in two parallel paths to accommodate different behav­ior to small signal as to large signal offsets.

High-pass Filter:

Small signal feedback (6-7-8-11-12-4) realizes the high-pass filter. The output from the post-amplifier (5) is limited by a non-linear function NL1 (6). NL1 has the input output relation illustrated in Figure 3.2-2. Clipping the signal in NL1 reduces the influence of impulses in the subsequent circuit. After clipping (6) the signal is attenuated (7) by a factor k, where k is determined by the amplification A2 and the desired overall high-pass cut off frequency. (see subsection 3.2.2 Realization for details). Hence the signal is inte­grated (11) and phase shifted p (12). (The integrator (11) is a linear element and the processing of the signals from the added (8) outputs form the attenuation (7) and the non-linear function NL2 (10) does not influence on each other). This part of the signal path provides linear high-pass filtering of small signals.

DC-offset Compensation:

Large signal feedback (9-10-8-11-12-4) provides the offset compensation. The output from the post amplifier (5) is low-pass filtered (9). The cut off frequency of the filter (9) determines the overall recovery time for the MeCFES amplifier. The filtered signal is fed to a non-linear function NL2 (10). The NL2 has the I-O relation illustrated in Figure 3.2-2. Only if the absolute value of the input of NL2 exceeds a threshold the output of NL2 is non-zero. In this case the signal will run through the adder (8) to the integrator (11) and thus fast establish the offset compensation. It is this part of the circuit that pro­vides the fast recovery time of the entire circuit.

Thresholds for NL1 (6) and NL2 (10) should be equal with characteristics as shown in Figure 3.2-2.

Figure 3.2-2 Non-linear functions

Realization

The embodiment of the amplifier is shown in the simplified schematic Figure 3.2-1 with the main component values. It is designed as a ±3V system with a total gain of 74dB and a small signal high-pass filter cut off frequency at 8Hz. DC-offset compensation is start­ing after 50ms (significantly longer than the expected muscle response). At the output there is a second order low-pass Bessel filter (not shown) with a cut off fre­quency of 500Hz.

Figure 3.2-3Simplified schematic of the MeCFES Amplifier

The signal from the electrodes is clamped between the power supplied through a resis­tor-diode network to protect the instrumentation amplifier IC16. This device is one of the critical parts since it determines the common-mode rejection ratio and the input impedance of the MeCFES amplifier. The output of IC16 is divided by 2 by the iden­tical resistors R22 and R23. Operational amplifier IC13B 'mirrors' the DC offset at the output of IC16.

The IC16 is selected to have a gain of 20dB. This allows a differential offset of the electrode potentials of up to 0.3V without saturation of the amplifier. The stimulation artefacts are saturating both IC16 and IC14 which for the same reason are chosen be fast recovery circuits (<10ms). The gain in IC14 is selected to 60dB to obtain a total signal gain of 74dB.

DC-offset Compensation

To find the time constant R32C31 of the low-pass filter (LP) it is assumed that the tran­sistors are switches that are open below the Vbe,on voltage of about ±0.6V. If a DC-offset saturates the amplifier, V3 will be clamped to ±3V, and the voltage at V4 will change ac­cordingly to the formula

                                                                               Eq.3.2-1

Setting V4 to the 0.6V Vbe,on voltage then, with the time of 50ms, Eq 3.2-1 yields a time constant of 220ms.The transistors realizes the non-linear function (NL2). It is assumed that the basis current in the transistors can be neglected. When the transistors are active they feed current into the capacitor C34. Saturation of IC14 (post amplifier) calls for at least 3V/500=6mV compensation at V2. When the transistor is on, a current of 3V/R30 flows into C3, giving a change in V2: of

                                                                                                      Eq. 3.2-2

where it is desired to have a recovery time t=50ms. The values of R30 (=R31) can calcu­lated using Eq. 3.2-2 and should not be greater than R32. With C3=1mF the result will be R30»25MW. Because of the non-linearity of this circuit there is a potential danger of in­stability but, with the chosen components, it has proven to be stable.

High-pass Filter:

The diodes (NL1) D6 and D7 clamps the input of the linear filter to the range ±0.6V. This minimizes the effects of transients from stimulation responses. Resistor network (attenu­ator), R39, R43 and R29, attenuates the signal. The transfer function of the entire high-pass filter is derived using the following three equations, derived from the circuit in Figure 3.2-3:

                                                                                               Eq. 3.2-3

                                                                                                      Eq. 3.2-4

                                                        Eq. 3.2-5

Here GIC14 is the gain of IC14 (1000 times). Combining Eq. 3.2-3, Eq. 3.2-4 and Eq. 3.2-5 yields the transfer function for the post-amplifier stage.

                                                                         Eq. 3.2-6

Multiplying this with the gain of the pre-amplifier (which is 10 times) yields the small-signal transfer function of the two first stages of the MeCFES amplifier circuit in Figure 3.2-3.

Common-mode Feedback

The common-mode signal is provided by IC16. It is amplified 40 dB and band-pass fil­tered from 5Hz to 500Hz. This suppresses the harmonics of the hum in the bandwith of the amplifier. The output from IC16 has an offset that is removed by R21and C30 before amplification.

Low-pass Filter

The amplifier is ended with a low-pass filter to comply with the Nyquist cri­terion before sampling. A simple RC high-pass filter is applied before the low-pass filter to remove offsets. The 2nd order low-pass filter is configured as a multiple feedback Sallen-Key filter, using a single operational amplifier (Figure 3.2-4). The transfer func­tion for the filter can be found to

               Eq. 3.2-7

 

Figure 3.2-4 Low-pass filter

Bessel filter type is selected and the gain is chosen to unity The denominator for the second order transfer function must be of the form s2+3s+3 [Jensen 1987] By normaliz­ing s with respect to the 3db cut off frequency being unity, the transfer function for the Bessel filter becomes

             Eq. 3.2-8

From Eq.3.2-7 the gain A can be found to be equal to

                                                                                                               Eq.3.2-9

Comparing Eq. 3.2-8 and Eq.3.2-9 the capacitors can be calculated

                                                                                      Eq .3.2-10

                                                                       Eq .3.2-11

Setting R25 »R26 , the gain becomes one and to obtain a cut-off frequency f0 close to 500Hz, selecting C33=8.2nF, C32=2.7nF (type 1% np0 SMD), the resistors can be calcu­lated to:R27=56kW (1%), R25=55kW (1%), R26=56kW (1%)

The realization of the Bessel filter is sensitive to the component values. The single op­erational amplifier configuration has been selected to minimize the number of required components.

TOC

Digital Signal Processor Unit        

Figure 3.3-1 DSP board

The digital signal processor unit (DSP) is the core of the system and is outlined in Figure 3.3-1. Special attention has been paid to power consumption and size. Schematic of the digital signal processor board can be found in Appendix B. The central unit is the TMS320LC50 signal processor from Texas Instruments. It was in 1995 the commercial available processor with least power consumption with respect to calculation capa­bilities (Figure 3.3-2). It contains 10kWords on chip RAM that can be used for both data and program memory, has sufficient calculation capacity and is available in small quantities. The processor takes care of data communication with A/D (analogue-digital) and D/A (digital-analogue) converters as well as miscellaneous controls of the sub­sys­tems (amplifier, stimulator and power supply).

Figure 3.3-2 Power consumption of common DSP's

The DSP board is equipped with an AT29LV1024 FLASH memory from Atmel. This device has 64k*16bit non-volatile memory. It is used for storage of the program, pa­rameters and miscellaneous information. It can be programmed in sectors of 128 words. The lifetime is more than 1000 write cycles in each sector and a data retention time of more than ten years. The advantage is that it can be programmed 'on board' without the need for a programming voltage other than the 3.3V power supply. The power con­sumption is low (5mA @ 1MHz) compared to other ROM (read-only memory) circuits. The FLASH is connected to the program data bus of the TMS320 so the program can be run directly. Special commands in the TMS320 provides access to both reading and writing of the FLASH.

Since the FLASH will contain no information after assembly of the DSP, it must be programmed. For this purpose the board has been equipped with a boot-load control logic. It controls the mode in which the TMS320 will start in after reset (power up). In stand-alone mode the program execution will begin from the FLASH. In boot-load mode it will execute the Texas build in boot load program to start reading from the serial port. Thus a communication program can be transmitted from the host computer to the program RAM of the TMS320, using four wire serial communication. Afterwards the entire program can be transferred via the TMS320 to the FLASH. This start up mode is determined by the host computer. If no host computer is connected the stand alone mode is automatically selected.

The choice of converters was at the time of system design very limited by the low-power 3.3V constraint with serial interface. The A/D converter is the TLV1543C from Texas Instruments with 10 bit resolution and 11 channels. This samples the myoelec­tric signal, battery level, stimulation voltages etc.

For D/A conversion the only available device fulfilling the demands was the LTC1452 from Linear Technology. It is a 12bit converter with one channel output. It generates the analogue stimulation signal. The output is besides the stimulator also fed to the A/D converter for self-testing purpose. For safety reasons to prevent erroneous function a watch dog has been implemented. If a control signal line is not toggled within 50ms power to the entire system will be shut off. This and other controls are interfaced by the 6 line output buffer. If the system thus executed invalid code the watch dog will turn the system off to protect the user against uncontrolled stimulation. (An example of such a situation can be if the program is changed due to external noise)

It will be outside the scope of this report to give an extensive description of the DSP board but a few remarks are made. The figures are subsections of the full DSP schematic in Appendix B.

The host computer interface is shown in Figure 3.3-3. All signals are ESD protected by the D1 and buffered by the IC9. The clock PClk, frame synchronization PFS and trans­mit data PX are outputs from the computer and the data receive PR is the input to the com­puter. The receive signal is buffered by a transistor to match a 5V low impedance input of the parallel port of the host computer. The actual operation of the serial ports of the TMS320, after reset, is not in agreement with the description in the user's guide Resis­tors R151 and R150 in the DSP circuit compensates for this.

Figure 3.3-3 Host computer interface subcircuit

Since the 2'nd serial port, connecting to the A/D and D/A converters, can be config­ured before it is enabled the receive (TFSR and TCLKR) can be connected directly to the transmit (TFSX and TCLKX) of the frame sync and clock signals.

The serial port is not directly compatible with the converters, why the interfacing counter (pleas refer to Appendix B, IC31) and gates are necessary to generate the chip enable signals for the converters, using the frame sync and clock signals from the 2'nd serial port.

Figure 3.3-4 Boot control subcircuit

 The system is reset when both PFS and PX are low (Figure 3.3-4). The duration of the low pulse determines whether stand-alone mode or boot-load mode is initiated. A long pulse will charge C21 and thus setting the boot signal. This signal is used for the input of IC11 and IC10 that will put the '16 bit serial boot-load mode command' (binary xxxx xxxx xxxx 0100) pattern on the data bus and set the MP low (starting the on-chip fac­tory programmed boot loader). As the last thing it shall be mentioned that patches has been made to enable the on off button to interface the signal processor and at the same time be able to turn the system on/off. These are not in full agreement with the sche­matic.

TOC

Stimulator circuits

Two different stimulator concepts (Type 1 and Type 2) have been designed based on a high voltage supplied transistor output stage. A third stimulator concept (Type 3) based on the transformer output has been developed and produced. The aim is to create the most ideal current generator according to the discussion in 2.3 Stimulator Principle, with a very low quiescent power consumption, optimal efficiency and the desired pulse form.

The concept of Type 1 and Type 2 provides the opportunity to improve efficiency and size of the stimulator. The advantages of avoiding connection, to the stimulation elec­trodes, through a transformer is described in 2.3 Stimulator Principles.

The specifications set for the stimulators are:

Stimulator Type 1 Design

The following describes the design of the Type 1 stimulator. The full schematic is in Appendix B. The following calculations applies to Figure 3.4-3.

There are four identical current generators configured as in the illustrative model Figure 3.4-1. The current generators are active in pairs. The a positive control signal controls generator number 1 and 3 while number 2 and 4 are off. In the negative stimu­lation period number 1 and 3 are off while 2 and 4 are on. (If the current in the genera­tors is not identical, the result will be the fault current in the recording electrodes as de­scribed in 2.4.7 Stimulation Response).

Figure 3.4-1 Illustrative model of stimulator configuration

Each current generator is realized as outlined as in Figure 3.4-2 where a voltage con­trolled current generator (VCCG) is determining the current in the resistor R49. If the transistors are identical then the basis emitter voltage must be identical and thereby the voltage over R51 must equal the voltage over R49. This yield a current gain when ideal­izing the transistors

                                                                                          Eq. 3.4-1

To obtain a high efficiency the resistor R51 should be much lesser than the electrode imped­ance and the ratio of R49:R51 should be high to minimize the power loss in R49. The value chosen for resistor R51 is 100w and the current gain is selected to 10.

Figure 3.4-2 Current amplification

Since high voltage transistors must be used (which are not ideal and usually have a low gain), it is of interest to include the gain (b) and the basis emitter voltage Vbe,on in the calculations. Let DVbe be the difference in the basis-emitter voltage, Vbe, for the two transistors. The current output of the transistors are given by

                                                                                            Eq. 3.4-2

                                                                                            Eq. 3.4-3

The difference basis-emitter voltage difference provides the currents in the resistors

                                                                                    Eq. 3.4-4

Combining these equations yields

                                                                            Eq. 3.4-5

From this it appears that the output current is sensitive to DVbe and difference in b for the transistors. To minimize the consequence of a DVbe the resistor R51 must not be too small. Typical (50-200) variations in b can give rise to an error on the current gain of some percent. For the selected transistor type the Vbe can vary 0.1V in worst case and thus give rise to significant offset approaching a milliampere if R51=100W.

Figure 3.4-3 shows a section of the schematic including generator 1 and 4.

Figure 3.4-3 Section of schematic

The VCCG in Figure 3.4-2 is realized by the transistor T12 (and in the negative pulse pe­riod T11) and operational amplifier IC18B which controls the current in the resistor R63. This current will (when neglecting the basis current in T12) be supplied by the transistor T6. The coupling of this transistor results in a 'shut down' feature. When the input to the stimulator is near zero (the 10 bit D/A converter will have a certain noise level at the output of at least±5mV) the current generators will effectively be shut off leaving the stimulator output in a current-less high impedance state. The stimulator will thus be 'silent' and not disturb myoelectrical signal recording and the power consumption will be reduced to the quiescent power consumption of the operational amplifiers.

This on/off threshold is determined by

                                                                                        Eq. 3.4-6

where Vbe,on of T12 can be set to 0.6V and is the input to the stimulator. In summary the transfer function for the stimulator is

                                 Eq. 3.4-7

Stimulator Type 2 Design

It was found that the stimulator Type 1 is sensitive to the matching of the transistors. In order to improve the performance of the stimulator design an approach to use a different technique is attempted. This circuit (Figure 3.4-4) makes use of a current sensor that feeds back to the control sub circuits. This makes a closed loop control of the output current. This should in theory give very accurate balanced stimulation output. A full schematic can be found in Appendix B.

Figure 3.4-4 Principal function of Type 2

The concept is build up upon the idea of letting the high level voltage across the current sensing resistor be transformed down to a level of ±3V using capacitors (see Figure 3.4-4). Then an instrumentation amplifier amplifies this differential signal, representing the current flowing into the electrode (when using FET transistors in the output). The ca­pacitors low side must have equal voltage before the stimulation pulse starts. This volt­age must not exceed the active region of the instrumentation amplifier input. This is ensured by switches. The output is switched between the two electrodes depending on the phase of the stimulation pulse. A matching circuit is providing the negative counter­part to the in Figure 3.4-5 shown circuit.

Figure 3.4-5 Principle of stimulator Type 2

The switching between the diodes are demanding two logical signals On2 and On1 con­trolling the phase (see Figure 3.4-5). Stimulation form/amplitude is send to the positive part of the circuit Vlevel. This is inverted as the actual current in the positive part VRi+ and fed to the negative stimulator part as the control signal. This ensures that the current in the negative output mirrors the positive output current. The following calculations refers to Figure 3.4-5.

Desired precision of IOut1 and IOut2 is determined by the precision of the sensing resis­tors. The voltage drop over the resistor is chosen to 0.5V at 50mA, i.e. R114=10W.

The high voltage is stored in capacitors in the power supply. Selecting a 0.5V margin to the ±3V rated input of the instrumentation amplifier the allowable voltage drop of Vhi is 2V over 2*300ms. (the selected type of integrated circuits INA118 or AD620 are inter­nally protected at the inputs to ±40V)

The impedance for the instrumentation amplifier is 1010W which can be neglected. The leakage current is 10nA. This implies a possible DC offset error on VRi of ±30mV cor­responding to an error in the stimulation current balance of 3%. This is removed by a high-pass filter. The resistor Re is limiting the current output from IC92 when VHi is low. The resistor Rcb is removing charge from the gate. Transistors T28 and T30 are ena­bl­ing the outputs and selecting which of the two electrodes the output current is ap­plied to.

Stimulator Type 3 Design

The Type 3 stimulator is made by Miguel Hermann at Asah Medico A/S and the sche­matic can be found in Appendix B. The circuit is controlling the current in the primary side of an 1:20 transformer. It needs 9V battery supply. It is thus assumed that the trans­former is ideal and that the current in the secondary side is proportional to the current in the primary side. This stimulator needs a modified power supply.

TOC

Power Supply Unit

The power supply circuit can be found in Appendix B and the functions is illustrated in Figure 3.5-1. The power supply part of the system is taking care of

It comprises the push button control, the sound source and a watchdog. The power will be shut down if the unit does not receive a signal for the watchdog .

Figure 3.5-1 The power supply unit

The high voltage is generated by a switch-mode fly back converter. The DSP can control the voltage level by turning the converter on/off. To protect the circuit components, a feedback loop in the converter ensures that the voltage does not exceed ±75V.

TOC

Evaluation Method

The first step in evaluation of the MeCFES performance is the enhancement of the mus­cle force and wrist movement. This is evaluated by the tracking test (used by Haxthausen [Haxthausen, et al. 1991]). The next step is evaluation of the functional benefits of that enhancement. This is done by a hand function test has been developed by the occupational therapists at the Center for Spinal Cord Injury, Copenhagen Univer­sity Hospital, Rigshospitalet, Denmark. The hand func­tion test has been conducted by an occupational therapist. It is described along the conclusion of the results in 4.6 Func­tional Evaluation.

Tracking Test Description

The tracking test is recording isometric wrist force or the wrist extension angle against gravity with the purpose of being an objective repeatable test of the MeCFES. The test is modi­fied and extended by an endurance measurement. The evaluation thus comprises three types of tests: A force tracking test, an angle tracking test and an en­durance tracking test. All measure­ments are using the same set-up which is described last in this section. In addition to the tracking test where the test participant is control­ling the stimulation, the set-up is used for recording of recruitment curves (the current-muscle output).

In the tracking test, a target, a course of desired muscle output is displayed on a com­puter screen. The subject being tested is supposed to track this target as close as possi­ble. The track is repre­senting either an isometric force or the angle of the wrist exten­sion against gravity. The recorded parameter will be displayed, real time, with a mov­ing marker on the computer screen together with the target. In the force and angle tests the target is a trapezoid with a duration of 20 seconds, see Figure 3.6-1. Muscle con­traction is required in 16 seconds. The target maximum value is 90% of the maximal MeCFES assisted contraction. The endurance test is a modified tracking test where the partici­pant is supposed asked to keep a  50% force for 200 seconds, see Figure 3.6-2. The vertical axis is normalized with the maximum contraction.

Figure 3.6-1 Example of a tracking test

Figure 3.6-2 Endurance Test example.

Three different means of evaluating the performance are chosen. A 10% margin for the track, is shown on each side of the target according to the precision needed to perform a given task. If the tracking is outside the margin once, then the task has failed. The 10% margin is arbitrarily chosen.


In range:           A normal subject can easily keep the track within this margin. The time the track is outside this margin is shown as the percentage of the total time. No contraction and normal tracking will yield respectively: In range 23% and In range 100%.

Fails:                 This is the number of times the track exceeds the margin. This is a measure for the reliability of the movement, since it only takes a short failure of contraction for a cup of coffee to end on the lap. For a reliable system a fail should not occur.

Error:               This is the value used by Haxthausen. It is the root mean square (RMS) value of the vertical distance between the target and the tracking (used by Haxthausen). No contraction will yield: Error 60% RMS.

Recording of the recruitment curves are using the same set-up. The current is not controlled by the myoelectrical signal but is controlled by the computer. It is direct proportional with the target used in the tracking tests. The target is providing the per­centage of maximum stimulation (determined by the participant). The curves can be recorded as stimulation-angle or stimulation-isometric force curves. The corresponding recording to endurance test is the stimulation-endur­ance curve, where a constant stimu­lation of 90% maximum stimulation is applied for 200 seconds.

Calibration Procedure

The parameters for the MeCFES are set by trial and error, using the settings from previ­ous experiments as a starting point. When this is finished the tracking test set-up is cali­brated. The offset (relaxed non stimulated) and maximal MeCFES assisted contraction is measured at the start of the test. For adjustment of the MeCFES system and train the patient the Angle Tracking Test is used. This test is used maximally 10 times before commencing the measurements.Performing the tracking test comprises of the following steps:

The procedure is not followed strictly due to practical reasons during the actual test situation. The deviations are on the order of using MeCFES and some times a test has been restarted or repeated.

Tracking Test Set-up

The set-up for the tracking tests is comprising a device illustrated in Figure 3.6-1 and a electronic circuit, the transducer interface board, that can be found in Appendix B. The circuit consists of two amplifier circuits, one for the force transducer and one for the angle transducer. These two signals are converted by two separate analog-digital con­verters, enabling sampling on a PC (IBM compatible personal computer). The con­nec­tion to the computer is using some of the free pins on the same parallel port as is used by the MeCFES. This enables simultaneous sampling of either force or angle and monitoring of the MeCFES processing. The force signal is provided by a strain-gauge bridge and the angle by a linear single turn potentiometer. Both the control of the MeCFES and the tracking test is integrated in the MeCFES host program.

The mechanical set-up consists of a plate mounted with a lever, see Figure 3.6-1. The forearm is intended to be placed on the plate in such way that the lever rests on the back of the hand over the knuckles. The plate is mounted on a flexible arm to provide the opportunity to place the plate in the most comfortable position for the test person.

Figure 3.6-1 Angle and force measurement set-up

The rotation axis of the lever is parallel to the rotation axis of the wrist and the angle is recorded by a mechanical connected potentiometer. The wrist extension angle, where gravity is the only force reacting on the hand, is recorded for the Angle Tracking Test. The movement is a dynamic movement that allows concentric and eccentric contrac­tion that is comparative to the conditions under which the hands is used when grasping and lifting objects.

The lever can be blocked by a split as seen in Figure 3.6-1. The Wrist Force Tracking Test and Endurance Test are performed by fixing the lever at a position where the wrist is parallel to the arm (normal anatomical position). The contraction will then be an isometric contraction. A force transducer is mounted on end of the lever that rests on the knuckles and by blocking the lever the isometric force is recorded providing a well defined reproducible measurement. Note that the force of gravity is not eliminated. In all tests the hand is taped to the lever. This is necessary since the attempt of wrist exten­sion in all the participants was accompanied by some supination of the wrist.

TOC

DSP Software

The DSP (digital signal processor) is taking care of: Sampling and filtering, output of stimulation pulse, calculation of stimulation output and miscellaneous control tasks.

The software for the DSP is coded in the TMS320C50 assembly language consists over 2000 codelines. Besides the signal processing it enables communication and data ex­change capabilities to the optional host computer. The program provides different signal processing strategies, which can be changed by the host computer. A combination of the following signal processing steps are possible: The functions in italics are the default, Mode1.

Signal name

Processing

X=

Sampling of amplified signal

Y=

Stimulation response suppression filter (1.order transposed FIR-filter )

Noise indication

Noise detection

MES=

Filtering the blocs of data (Changeable coefficients)

MA=

Counting samples above a threshold

Rectified mean value

Root mean square

M=

FIR low-pass filtering

IIR low-pass filter

I=

Stimulation amplitude = Piece-wise linear function

Constant stimulation

Table 3.7-1 Signal processing

The filters with changeable coefficients are using a lookup table for the coefficients. At assembly time different tables can be used/created. The sampling frequency is set to 2kHz but can be changed in the source code for the program (it then has to be reassem­bled). Of the total stimulation interval, which is 60ms, the first 10ms is blanked, i.e. they are implicitly set to zero. The block length (samples in-between stimulation) is 100 samples. This is equivalent to 50ms of the myoelectrical signal.

The stimulation pulse is given by a sequence of 10 interrupt routines with an interval of 0.1ms. This gives the opportunity to select an arbitrary stimulation pulse with a duration of 1ms with a resolution of 0.1ms. The used stimulation from is a biphasic pulse with equal positive, inter- and negative pulse have a duration of 0.3ms each. The execution of these tasks are illustrated in Figure 3.7-1

Figure 3.7-1 Timing of program

The sampling and output of the stimulation pulse have the first priority since they de­mand precise timing. Calculation of the new stimulation amplitude can be performed in parallel with the sampling. It must be completed before the stimulation starts. The signal processing flow is illustrated in Figure 3.7-2.

Figure 3.7-2 Flow of the signal processing

The control tasks can be executed asynchronously in parallel with the two other proc­esses as indicated on the timing diagram Figure 3.7-1. These three classes of processes are controlled by pointers. The pointers are: SamplingStateSel_M.which takes care of the sampling, ProgramModeSel_M selecting the signal processing of the blocks. Finally there are two background process pointers BackProcessSel_MeCFES and ComModeSel_M for the asynchronous tasks.

Figure 3.7-3 Task dispatching

The pointer, SamplingStateSel_M, is controlling the sampling process using the timer interrupt, as shown in Figure 3.7-3. After 100 samples the pointer is being changed to the stimulation process and then again after 1ms to the control process during the blank­ing of 10ms. Totally a stimulation period of 60ms. Changing the stimulation period from 16.66Hz to 15Hz it is done by extending the blanking time by 6.66ms. This can be done by correcting a fine-tuning parameter called PeriodTune_k that extends the control process.

A feature that recently has been implemented in the program is a mode for data logging of a myoelectrical signal. This can be used to register spasticity in a muscle over 24hours. In this mode, SpasmRecord, the stimulator module is not used. The program calculates the RMS value of the myoelectrical signal in blocks of 6 seconds. These values are saved in the FLASH and can be downloaded to the host computer after the end of the recording period.

TOC

Host Computer Software

Figure 3.8-1 Configuration

A range of data exchange options can be accessed by connecting a host computer (80386 IBM compatible PC or higher ) to the MeCFES. The host computer program, containing over 3000 code lines, is written in Turbo Pascal as a DOS application. The primary purpose of the program is to program and test the DSP. The features are:

In addition the program performs the tracking tests, endurance tests and the recruitment measurements.

TOC

Summary

This section gives a summary of the specifications for the hardware. The hardware is produced in surface mount technology, minimizing size and weight

The MeCFES system, comprising stimulator, DSP and power supply, is build into a 11cm x 7cm x 3.5cm box with a total weight of 200g. The box is intended to be placed e.g. in a pocket A flat cable connects it to the amplifier which is intended to be placed near the recording electrodes. The amplifier is build into a 5mm x 45mm x 35mm box. The bottom of the box is mounted with the reference electrode. It is the intention to have the amplifier enclosed in the electrode-mount (described in 2.8.4 Electrode-mount ) near by the recording electrodes. The system is powered by 4x1.2V rechargeable NiCd batteries with a total capacity of min 2Wh.The measured power consumption of the four parts is

DSP (executing signal processing.)

26mW @ 3,2MHz clock. (3.3V)

Amplifier

18mW. (±3V)

Stimulator (Type1)

10-150mW stimulation dependent

Power supply

Not available (unstable efficiency)

Some of the following measurements are described in detail in section 4.1 Hardware Performance.

Amplifier Specifications

The produced amplifier has an artefact suppressing construction and needs no shut-down during stimulation.

Power supply

±3V

Active current*

6mA

Input impedance**

>10GW common mode

>10GW differential mode

Gain* , Input range

74dB, ±600mV

Frequency range*

10-500Hz

Filter

Low pass: 2 order Bessel

High pass: Non-linear feedback

Common-mode rejection**

>110dB + active grounding of patient

Noise related to input*

400nV(RMS) or 3mV(peak-peak

Recovery time@10mV step as differential input(2)

50ms-100ms

Offset compensation activation level

±0.1mV

ESD & Stimulation artefact protected

Digital Signal Processor Specifications

Power supply

3.3V

Active current

8 mA

Microprocessor

TMS320LC50

Speed

1,6 MIPS (Million Instructions Per Sec.)

FLASH Memory

64kWords

RAM

10kWords

A/D converter

10bit >10kHz, 11 channels

D/A converter

12bit >100kHz serial

Logic outputs

6

Serial communication port for host computer connection

Digital signal processing is fully software controlled. The signal processor program has a mean execution time of : 0.788MIPS with a peak at 0.972MIPS during data acquisi­tion and host communication. The DSP takes care of all control and signal processing tasks.

Stimulator

The stimulator is converting the signal from the DSP to a current. The shape of the stimulation signal is controlled by the DSP. The Type 1 stimulator is used.

Power supply

±3V and

Variable -75V

Typical output power

14mW @ 15mA, 2kW load,16Hz

Pulse shape

Biphasic with interpulse interval. (DSP controlled).

Current forced to zero in inter stimulation period.

Pulse width

0.9mSec (DSP controlled)

Current output amplitude

<50mA

Pulse repetition rate

Arbitrary

A power efficient concept for the stimulator has been chosen. It is based upon a switch mode DC-DC converter to produce high voltage. This voltage is controlled by the DSP. In this way the voltage for the stimulator output stage is kept as low as possible in order to increase efficiency. A transistor based output stage provides the stimulation current.

Power Supply

The system is powered by build in batteries. The power supply for the Type1 stimulator is taking care of: Charging of the battery, Sound module, push button and generation of 5 different voltage levels.

Power supply

4x1,2V NiCd batteries

Output voltages

±3V for analog circuits

3.3V for digital circuits

0-±75V for stimulator circuit

Charging input

1A @ 10-20V AC or DC

High voltage output efficiency

<40% @ 70V,0.3mAmean output

High voltages are generated using switch mode power supply technique.

* Measured

** Specified by IC manufacturer

Hardware and Software

This chapter describes the hardware and software developed and the test set-up

The keywords for the design of the electronic parts of the system besides functionality, are minimum size, weight and a low power consumption. Power consumption is closely connected to the size and weight since it determines volume of batteries which are among the heavy and space consuming parts. To minimize the size it has been attempted to use few components and choose the most power efficient techniques as possible. The availability of low price components limits the design possibilities.

Through several test circuits and three prototypes, the system has been evolving to comprise an amplifier, stimulator, digital signal processor (DSP), power supply/battery manager including rechargeable batteries and wires for electrodes. This system is called MeCFES (Myoelectrical Controlled Functional Electrical Stimulator). There are four different printed circuit boards which are the amplifier-, the stimulator-, the digital signal processing (DSP)- and the power supply- board. Especially for the DSP system, state of the art devices has been used causing some problems with faulty devices and supplier problems. Developing, manufacturing and testing the hardware and software has been occupying more than 2/3 of the project period leaving only half a year for systematic trials with tetraplegics. Sections 3.1 through 3.5 are describing the MeCFES hardware. Section 3.6 describes the method and set-up for the evaluating the perform­ance of the MeCFES. The software developed for the system is described in section 3.7 and 3.8. It consists of a DSP program for the MeCFES and a host program for occa­sional communication with the DSP. Only a brief description is provided for this com­prehensive work to preserve proprietary rights. Finally section 3.9 is a summary of the MeCFES specifications.

TOC

Interface

Tetraplegics have very limited possibilities to turn knobs and push buttons to operate the MeCFES. To comply with this, it has been decided that the device shall have only one push-button, by which the user can operate the MeCFES. To inform the user of changes in the states of the device a sound transducer is used. Figure 3.1-1 shows the on/off procedure using the button.

Figure 3.1-1 Start-up procedure

Sound signals with different characteristics informs the user as listed in Table 3.1-1.

Single tone

The device is turning on and initializing

Rising scale of tones

The device goes ready and operating

Descending scale of tones

The device is shutting down

Two alternating tones repeating

An error has occurred. Turn off the device. Try again, recharge batteries or call service.

Repeated buzz

Battery low. Shut off the system and recharge batteries.

Table 3.1-1 Sound signals

This restricted user interface implies that no adjustments of parameters can be done. These adjustments must be done initially with a connected host computer as described later. The demands for stimulation intensity and myoelectrical sensitivity might change from time to time. For that reason the program has been prepared for implementation of an automatic calibration procedure. This should be executed during the initialization phase after power up of the device. A suggestion for the auto calibration procedure is follows.

After the button is pressed, system powers up and starts a self-test. If the test is passed, an initialization mode is entered and the user is informed by a sound signal whereupon the stimulation increases slowly (limited by a pre-set maxi­mum value). The user ob­serves when the maximum contraction is achieved and confirms by pressing the button and the stimulation stops. This procedure will set the stimulation gain and the MeCFES will confirm this by a sound signal.

There is an option of connecting a host computer (BM PC) to the MeCFES. This re­quires a special cable from the parallel port of the host computer to the MeCFES. Com­munication with the MeCFES is then possible using the developed host computer soft­ware. After the manufacturing of the MeCFES the DSP program has to be down­loaded to the device from a host computer. The program is then stored in the FLASH memory (an erasable non volatile integrated memory circuit) of the MeCFES. Once programmed, the system will not need the host connection to operate.

The host interfacing option gives access to the following actions:

TOC

Amplifier Circuit

As described in 2.5 Signal Amplification the amplifier is the most important link in the signal processing. The developed amplifier is patent pending and a complete schematic can be found in Appendix B. The special property of the amplifier is that high-pass filtering is achieved using a non-linear feedback loop.

Amplifier Principle

Figure 3.2-1 is a model showing the principles of the amplifier. The electrodes, denoted by (1), pick up the signal. Each of the electrode signals are limited (2) to reduce stimu­lation artefacts and to protect the pre-amplification circuit (3). The pre-amplifier (3) has a difference gain of A1 and transforms the high impedance difference signal into a low impedance signal. The pre-amplifier output (3) is added (4) with the negative output (12) from the feed-back network which provides an estimate of the DC-offset at the pre amplifier output (3). The offset regulated signal from the addition (4) is fed to a post-amplifier (5) with gain A2.

Figure 3.2-1 Principle of the amplifier

The feedback network is divided in two parallel paths to accommodate different behav­ior to small signal as to large signal offsets.

High-pass Filter:

Small signal feedback (6-7-8-11-12-4) realizes the high-pass filter. The output from the post-amplifier (5) is limited by a non-linear function NL1 (6). NL1 has the input output relation illustrated in Figure 3.2-2. Clipping the signal in NL1 reduces the influence of impulses in the subsequent circuit. After clipping (6) the signal is attenuated (7) by a factor k, where k is determined by the amplification A2 and the desired overall high-pass cut off frequency. (see subsection 3.2.2 Realization for details). Hence the signal is inte­grated (11) and phase shifted p (12). (The integrator (11) is a linear element and the processing of the signals from the added (8) outputs form the attenuation (7) and the non-linear function NL2 (10) does not influence on each other). This part of the signal path provides linear high-pass filtering of small signals.

DC-offset Compensation:

Large signal feedback (9-10-8-11-12-4) provides the offset compensation. The output from the post amplifier (5) is low-pass filtered (9). The cut off frequency of the filter (9) determines the overall recovery time for the MeCFES amplifier. The filtered signal is fed to a non-linear function NL2 (10). The NL2 has the I-O relation illustrated in Figure 3.2-2. Only if the absolute value of the input of NL2 exceeds a threshold the output of NL2 is non-zero. In this case the signal will run through the adder (8) to the integrator (11) and thus fast establish the offset compensation. It is this part of the circuit that pro­vides the fast recovery time of the entire circuit.

Thresholds for NL1 (6) and NL2 (10) should be equal with characteristics as shown in Figure 3.2-2.

Figure 3.2-2 Non-linear functions

Realization

The embodiment of the amplifier is shown in the simplified schematic Figure 3.2-1 with the main component values. It is designed as a ±3V system with a total gain of 74dB and a small signal high-pass filter cut off frequency at 8Hz. DC-offset compensation is start­ing after 50ms (significantly longer than the expected muscle response). At the output there is a second order low-pass Bessel filter (not shown) with a cut off fre­quency of 500Hz.

Figure 3.2-3Simplified schematic of the MeCFES Amplifier

The signal from the electrodes is clamped between the power supplied through a resis­tor-diode network to protect the instrumentation amplifier IC16. This device is one of the critical parts since it determines the common-mode rejection ratio and the input impedance of the MeCFES amplifier. The output of IC16 is divided by 2 by the iden­tical resistors R22 and R23. Operational amplifier IC13B 'mirrors' the DC offset at the output of IC16.

The IC16 is selected to have a gain of 20dB. This allows a differential offset of the electrode potentials of up to 0.3V without saturation of the amplifier. The stimulation artefacts are saturating both IC16 and IC14 which for the same reason are chosen be fast recovery circuits (<10ms). The gain in IC14 is selected to 60dB to obtain a total signal gain of 74dB.

DC-offset Compensation

To find the time constant R32C31 of the low-pass filter (LP) it is assumed that the tran­sistors are switches that are open below the Vbe,on voltage of about ±0.6V. If a DC-offset saturates the amplifier, V3 will be clamped to ±3V, and the voltage at V4 will change ac­cordingly to the formula

                                                                               Eq.3.2-1

Setting V4 to the 0.6V Vbe,on voltage then, with the time of 50ms, Eq 3.2-1 yields a time constant of 220ms.The transistors realizes the non-linear function (NL2). It is assumed that the basis current in the transistors can be neglected. When the transistors are active they feed current into the capacitor C34. Saturation of IC14 (post amplifier) calls for at least 3V/500=6mV compensation at V2. When the transistor is on, a current of 3V/R30 flows into C3, giving a change in V2: of

                                                                                                      Eq. 3.2-2

where it is desired to have a recovery time t=50ms. The values of R30 (=R31) can calcu­lated using Eq. 3.2-2 and should not be greater than R32. With C3=1mF the result will be R30»25MW. Because of the non-linearity of this circuit there is a potential danger of in­stability but, with the chosen components, it has proven to be stable.

High-pass Filter:

The diodes (NL1) D6 and D7 clamps the input of the linear filter to the range ±0.6V. This minimizes the effects of transients from stimulation responses. Resistor network (attenu­ator), R39, R43 and R29, attenuates the signal. The transfer function of the entire high-pass filter is derived using the following three equations, derived from the circuit in Figure 3.2-3:

                                                                                               Eq. 3.2-3

                                                                                                      Eq. 3.2-4

                                                        Eq. 3.2-5

Here GIC14 is the gain of IC14 (1000 times). Combining Eq. 3.2-3, Eq. 3.2-4 and Eq. 3.2-5 yields the transfer function for the post-amplifier stage.

                                                                         Eq. 3.2-6

Multiplying this with the gain of the pre-amplifier (which is 10 times) yields the small-signal transfer function of the two first stages of the MeCFES amplifier circuit in Figure 3.2-3.

Common-mode Feedback

The common-mode signal is provided by IC16. It is amplified 40 dB and band-pass fil­tered from 5Hz to 500Hz. This suppresses the harmonics of the hum in the bandwith of the amplifier. The output from IC16 has an offset that is removed by R21and C30 before amplification.

Low-pass Filter

The amplifier is ended with a low-pass filter to comply with the Nyquist cri­terion before sampling. A simple RC high-pass filter is applied before the low-pass filter to remove offsets. The 2nd order low-pass filter is configured as a multiple feedback Sallen-Key filter, using a single operational amplifier (Figure 3.2-4). The transfer func­tion for the filter can be found to

               Eq. 3.2-7

 

Figure 3.2-4 Low-pass filter

Bessel filter type is selected and the gain is chosen to unity The denominator for the second order transfer function must be of the form s2+3s+3 [Jensen 1987] By normaliz­ing s with respect to the 3db cut off frequency being unity, the transfer function for the Bessel filter becomes

             Eq. 3.2-8

From Eq.3.2-7 the gain A can be found to be equal to

                                                                                                               Eq.3.2-9

Comparing Eq. 3.2-8 and Eq.3.2-9 the capacitors can be calculated

                                                                                      Eq .3.2-10

                                                                       Eq .3.2-11

Setting R25 »R26 , the gain becomes one and to obtain a cut-off frequency f0 close to 500Hz, selecting C33=8.2nF, C32=2.7nF (type 1% np0 SMD), the resistors can be calcu­lated to:R27=56kW (1%), R25=55kW (1%), R26=56kW (1%)

The realization of the Bessel filter is sensitive to the component values. The single op­erational amplifier configuration has been selected to minimize the number of required components.

TOC

Digital Signal Processor Unit        

Figure 3.3-1 DSP board

The digital signal processor unit (DSP) is the core of the system and is outlined in Figure 3.3-1. Special attention has been paid to power consumption and size. Schematic of the digital signal processor board can be found in Appendix B. The central unit is the TMS320LC50 signal processor from Texas Instruments. It was in 1995 the commercial available processor with least power consumption with respect to calculation capa­bilities (Figure 3.3-2). It contains 10kWords on chip RAM that can be used for both data and program memory, has sufficient calculation capacity and is available in small quantities. The processor takes care of data communication with A/D (analogue-digital) and D/A (digital-analogue) converters as well as miscellaneous controls of the sub­sys­tems (amplifier, stimulator and power supply).

Figure 3.3-2 Power consumption of common DSP's

The DSP board is equipped with an AT29LV1024 FLASH memory from Atmel. This device has 64k*16bit non-volatile memory. It is used for storage of the program, pa­rameters and miscellaneous information. It can be programmed in sectors of 128 words. The lifetime is more than 1000 write cycles in each sector and a data retention time of more than ten years. The advantage is that it can be programmed 'on board' without the need for a programming voltage other than the 3.3V power supply. The power con­sumption is low (5mA @ 1MHz) compared to other ROM (read-only memory) circuits. The FLASH is connected to the program data bus of the TMS320 so the program can be run directly. Special commands in the TMS320 provides access to both reading and writing of the FLASH.

Since the FLASH will contain no information after assembly of the DSP, it must be programmed. For this purpose the board has been equipped with a boot-load control logic. It controls the mode in which the TMS320 will start in after reset (power up). In stand-alone mode the program execution will begin from the FLASH. In boot-load mode it will execute the Texas build in boot load program to start reading from the serial port. Thus a communication program can be transmitted from the host computer to the program RAM of the TMS320, using four wire serial communication. Afterwards the entire program can be transferred via the TMS320 to the FLASH. This start up mode is determined by the host computer. If no host computer is connected the stand alone mode is automatically selected.

The choice of converters was at the time of system design very limited by the low-power 3.3V constraint with serial interface. The A/D converter is the TLV1543C from Texas Instruments with 10 bit resolution and 11 channels. This samples the myoelec­tric signal, battery level, stimulation voltages etc.

For D/A conversion the only available device fulfilling the demands was the LTC1452 from Linear Technology. It is a 12bit converter with one channel output. It generates the analogue stimulation signal. The output is besides the stimulator also fed to the A/D converter for self-testing purpose. For safety reasons to prevent erroneous function a watch dog has been implemented. If a control signal line is not toggled within 50ms power to the entire system will be shut off. This and other controls are interfaced by the 6 line output buffer. If the system thus executed invalid code the watch dog will turn the system off to protect the user against uncontrolled stimulation. (An example of such a situation can be if the program is changed due to external noise)

It will be outside the scope of this report to give an extensive description of the DSP board but a few remarks are made. The figures are subsections of the full DSP schematic in Appendix B.

The host computer interface is shown in Figure 3.3-3. All signals are ESD protected by the D1 and buffered by the IC9. The clock PClk, frame synchronization PFS and trans­mit data PX are outputs from the computer and the data receive PR is the input to the com­puter. The receive signal is buffered by a transistor to match a 5V low impedance input of the parallel port of the host computer. The actual operation of the serial ports of the TMS320, after reset, is not in agreement with the description in the user's guide Resis­tors R151 and R150 in the DSP circuit compensates for this.

Figure 3.3-3 Host computer interface subcircuit

Since the 2'nd serial port, connecting to the A/D and D/A converters, can be config­ured before it is enabled the receive (TFSR and TCLKR) can be connected directly to the transmit (TFSX and TCLKX) of the frame sync and clock signals.

The serial port is not directly compatible with the converters, why the interfacing counter (pleas refer to Appendix B, IC31) and gates are necessary to generate the chip enable signals for the converters, using the frame sync and clock signals from the 2'nd serial port.

Figure 3.3-4 Boot control subcircuit

 The system is reset when both PFS and PX are low (Figure 3.3-4). The duration of the low pulse determines whether stand-alone mode or boot-load mode is initiated. A long pulse will charge C21 and thus setting the boot signal. This signal is used for the input of IC11 and IC10 that will put the '16 bit serial boot-load mode command' (binary xxxx xxxx xxxx 0100) pattern on the data bus and set the MP low (starting the on-chip fac­tory programmed boot loader). As the last thing it shall be mentioned that patches has been made to enable the on off button to interface the signal processor and at the same time be able to turn the system on/off. These are not in full agreement with the sche­matic.

TOC

Stimulator circuits

Two different stimulator concepts (Type 1 and Type 2) have been designed based on a high voltage supplied transistor output stage. A third stimulator concept (Type 3) based on the transformer output has been developed and produced. The aim is to create the most ideal current generator according to the discussion in 2.3 Stimulator Principle, with a very low quiescent power consumption, optimal efficiency and the desired pulse form.

The concept of Type 1 and Type 2 provides the opportunity to improve efficiency and size of the stimulator. The advantages of avoiding connection, to the stimulation elec­trodes, through a transformer is described in 2.3 Stimulator Principles.

The specifications set for the stimulators are:

Stimulator Type 1 Design

The following describes the design of the Type 1 stimulator. The full schematic is in Appendix B. The following calculations applies to Figure 3.4-3.

There are four identical current generators configured as in the illustrative model Figure 3.4-1. The current generators are active in pairs. The a positive control signal controls generator number 1 and 3 while number 2 and 4 are off. In the negative stimu­lation period number 1 and 3 are off while 2 and 4 are on. (If the current in the genera­tors is not identical, the result will be the fault current in the recording electrodes as de­scribed in 2.4.7 Stimulation Response).

Figure 3.4-1 Illustrative model of stimulator configuration

Each current generator is realized as outlined as in Figure 3.4-2 where a voltage con­trolled current generator (VCCG) is determining the current in the resistor R49. If the transistors are identical then the basis emitter voltage must be identical and thereby the voltage over R51 must equal the voltage over R49. This yield a current gain when ideal­izing the transistors

                                                                                          Eq. 3.4-1

To obtain a high efficiency the resistor R51 should be much lesser than the electrode imped­ance and the ratio of R49:R51 should be high to minimize the power loss in R49. The value chosen for resistor R51 is 100w and the current gain is selected to 10.

Figure 3.4-2 Current amplification

Since high voltage transistors must be used (which are not ideal and usually have a low gain), it is of interest to include the gain (b) and the basis emitter voltage Vbe,on in the calculations. Let DVbe be the difference in the basis-emitter voltage, Vbe, for the two transistors. The current output of the transistors are given by

                                                                                            Eq. 3.4-2

                                                                                            Eq. 3.4-3

The difference basis-emitter voltage difference provides the currents in the resistors

                                                                                    Eq. 3.4-4

Combining these equations yields

                                                                            Eq. 3.4-5

From this it appears that the output current is sensitive to DVbe and difference in b for the transistors. To minimize the consequence of a DVbe the resistor R51 must not be too small. Typical (50-200) variations in b can give rise to an error on the current gain of some percent. For the selected transistor type the Vbe can vary 0.1V in worst case and thus give rise to significant offset approaching a milliampere if R51=100W.

Figure 3.4-3 shows a section of the schematic including generator 1 and 4.

Figure 3.4-3 Section of schematic

The VCCG in Figure 3.4-2 is realized by the transistor T12 (and in the negative pulse pe­riod T11) and operational amplifier IC18B which controls the current in the resistor R63. This current will (when neglecting the basis current in T12) be supplied by the transistor T6. The coupling of this transistor results in a 'shut down' feature. When the input to the stimulator is near zero (the 10 bit D/A converter will have a certain noise level at the output of at least±5mV) the current generators will effectively be shut off leaving the stimulator output in a current-less high impedance state. The stimulator will thus be 'silent' and not disturb myoelectrical signal recording and the power consumption will be reduced to the quiescent power consumption of the operational amplifiers.

This on/off threshold is determined by

                                                                                        Eq. 3.4-6

where Vbe,on of T12 can be set to 0.6V and is the input to the stimulator. In summary the transfer function for the stimulator is

                                 Eq. 3.4-7

Stimulator Type 2 Design

It was found that the stimulator Type 1 is sensitive to the matching of the transistors. In order to improve the performance of the stimulator design an approach to use a different technique is attempted. This circuit (Figure 3.4-4) makes use of a current sensor that feeds back to the control sub circuits. This makes a closed loop control of the output current. This should in theory give very accurate balanced stimulation output. A full schematic can be found in Appendix B.

Figure 3.4-4 Principal function of Type 2

The concept is build up upon the idea of letting the high level voltage across the current sensing resistor be transformed down to a level of ±3V using capacitors (see Figure 3.4-4). Then an instrumentation amplifier amplifies this differential signal, representing the current flowing into the electrode (when using FET transistors in the output). The ca­pacitors low side must have equal voltage before the stimulation pulse starts. This volt­age must not exceed the active region of the instrumentation amplifier input. This is ensured by switches. The output is switched between the two electrodes depending on the phase of the stimulation pulse. A matching circuit is providing the negative counter­part to the in Figure 3.4-5 shown circuit.

Figure 3.4-5 Principle of stimulator Type 2

The switching between the diodes are demanding two logical signals On2 and On1 con­trolling the phase (see Figure 3.4-5). Stimulation form/amplitude is send to the positive part of the circuit Vlevel. This is inverted as the actual current in the positive part VRi+ and fed to the negative stimulator part as the control signal. This ensures that the current in the negative output mirrors the positive output current. The following calculations refers to Figure 3.4-5.

Desired precision of IOut1 and IOut2 is determined by the precision of the sensing resis­tors. The voltage drop over the resistor is chosen to 0.5V at 50mA, i.e. R114=10W.

The high voltage is stored in capacitors in the power supply. Selecting a 0.5V margin to the ±3V rated input of the instrumentation amplifier the allowable voltage drop of Vhi is 2V over 2*300ms. (the selected type of integrated circuits INA118 or AD620 are inter­nally protected at the inputs to ±40V)

The impedance for the instrumentation amplifier is 1010W which can be neglected. The leakage current is 10nA. This implies a possible DC offset error on VRi of ±30mV cor­responding to an error in the stimulation current balance of 3%. This is removed by a high-pass filter. The resistor Re is limiting the current output from IC92 when VHi is low. The resistor Rcb is removing charge from the gate. Transistors T28 and T30 are ena­bl­ing the outputs and selecting which of the two electrodes the output current is ap­plied to.

Stimulator Type 3 Design

The Type 3 stimulator is made by Miguel Hermann at Asah Medico A/S and the sche­matic can be found in Appendix B. The circuit is controlling the current in the primary side of an 1:20 transformer. It needs 9V battery supply. It is thus assumed that the trans­former is ideal and that the current in the secondary side is proportional to the current in the primary side. This stimulator needs a modified power supply.

TOC

Power Supply Unit

The power supply circuit can be found in Appendix B and the functions is illustrated in Figure 3.5-1. The power supply part of the system is taking care of

It comprises the push button control, the sound source and a watchdog. The power will be shut down if the unit does not receive a signal for the watchdog .

Figure 3.5-1 The power supply unit

The high voltage is generated by a switch-mode fly back converter. The DSP can control the voltage level by turning the converter on/off. To protect the circuit components, a feedback loop in the converter ensures that the voltage does not exceed ±75V.

TOC

Evaluation Method

The first step in evaluation of the MeCFES performance is the enhancement of the mus­cle force and wrist movement. This is evaluated by the tracking test (used by Haxthausen [Haxthausen, et al. 1991]). The next step is evaluation of the functional benefits of that enhancement. This is done by a hand function test has been developed by the occupational therapists at the Center for Spinal Cord Injury, Copenhagen Univer­sity Hospital, Rigshospitalet, Denmark. The hand func­tion test has been conducted by an occupational therapist. It is described along the conclusion of the results in 4.6 Func­tional Evaluation.

Tracking Test Description

The tracking test is recording isometric wrist force or the wrist extension angle against gravity with the purpose of being an objective repeatable test of the MeCFES. The test is modi­fied and extended by an endurance measurement. The evaluation thus comprises three types of tests: A force tracking test, an angle tracking test and an en­durance tracking test. All measure­ments are using the same set-up which is described last in this section. In addition to the tracking test where the test participant is control­ling the stimulation, the set-up is used for recording of recruitment curves (the current-muscle output).

In the tracking test, a target, a course of desired muscle output is displayed on a com­puter screen. The subject being tested is supposed to track this target as close as possi­ble. The track is repre­senting either an isometric force or the angle of the wrist exten­sion against gravity. The recorded parameter will be displayed, real time, with a mov­ing marker on the computer screen together with the target. In the force and angle tests the target is a trapezoid with a duration of 20 seconds, see Figure 3.6-1. Muscle con­traction is required in 16 seconds. The target maximum value is 90% of the maximal MeCFES assisted contraction. The endurance test is a modified tracking test where the partici­pant is supposed asked to keep a  50% force for 200 seconds, see Figure 3.6-2. The vertical axis is normalized with the maximum contraction.

Figure 3.6-1 Example of a tracking test

Figure 3.6-2 Endurance Test example.

Three different means of evaluating the performance are chosen. A 10% margin for the track, is shown on each side of the target according to the precision needed to perform a given task. If the tracking is outside the margin once, then the task has failed. The 10% margin is arbitrarily chosen.


In range:           A normal subject can easily keep the track within this margin. The time the track is outside this margin is shown as the percentage of the total time. No contraction and normal tracking will yield respectively: In range 23% and In range 100%.

Fails:                 This is the number of times the track exceeds the margin. This is a measure for the reliability of the movement, since it only takes a short failure of contraction for a cup of coffee to end on the lap. For a reliable system a fail should not occur.

Error:               This is the value used by Haxthausen. It is the root mean square (RMS) value of the vertical distance between the target and the tracking (used by Haxthausen). No contraction will yield: Error 60% RMS.

Recording of the recruitment curves are using the same set-up. The current is not controlled by the myoelectrical signal but is controlled by the computer. It is direct proportional with the target used in the tracking tests. The target is providing the per­centage of maximum stimulation (determined by the participant). The curves can be recorded as stimulation-angle or stimulation-isometric force curves. The corresponding recording to endurance test is the stimulation-endur­ance curve, where a constant stimu­lation of 90% maximum stimulation is applied for 200 seconds.

Calibration Procedure

The parameters for the MeCFES are set by trial and error, using the settings from previ­ous experiments as a starting point. When this is finished the tracking test set-up is cali­brated. The offset (relaxed non stimulated) and maximal MeCFES assisted contraction is measured at the start of the test. For adjustment of the MeCFES system and train the patient the Angle Tracking Test is used. This test is used maximally 10 times before commencing the measurements.Performing the tracking test comprises of the following steps:

The procedure is not followed strictly due to practical reasons during the actual test situation. The deviations are on the order of using MeCFES and some times a test has been restarted or repeated.

Tracking Test Set-up

The set-up for the tracking tests is comprising a device illustrated in Figure 3.6-1 and a electronic circuit, the transducer interface board, that can be found in Appendix B. The circuit consists of two amplifier circuits, one for the force transducer and one for the angle transducer. These two signals are converted by two separate analog-digital con­verters, enabling sampling on a PC (IBM compatible personal computer). The con­nec­tion to the computer is using some of the free pins on the same parallel port as is used by the MeCFES. This enables simultaneous sampling of either force or angle and monitoring of the MeCFES processing. The force signal is provided by a strain-gauge bridge and the angle by a linear single turn potentiometer. Both the control of the MeCFES and the tracking test is integrated in the MeCFES host program.

The mechanical set-up consists of a plate mounted with a lever, see Figure 3.6-1. The forearm is intended to be placed on the plate in such way that the lever rests on the back of the hand over the knuckles. The plate is mounted on a flexible arm to provide the opportunity to place the plate in the most comfortable position for the test person.

Figure 3.6-1 Angle and force measurement set-up

The rotation axis of the lever is parallel to the rotation axis of the wrist and the angle is recorded by a mechanical connected potentiometer. The wrist extension angle, where gravity is the only force reacting on the hand, is recorded for the Angle Tracking Test. The movement is a dynamic movement that allows concentric and eccentric contrac­tion that is comparative to the conditions under which the hands is used when grasping and lifting objects.

The lever can be blocked by a split as seen in Figure 3.6-1. The Wrist Force Tracking Test and Endurance Test are performed by fixing the lever at a position where the wrist is parallel to the arm (normal anatomical position). The contraction will then be an isometric contraction. A force transducer is mounted on end of the lever that rests on the knuckles and by blocking the lever the isometric force is recorded providing a well defined reproducible measurement. Note that the force of gravity is not eliminated. In all tests the hand is taped to the lever. This is necessary since the attempt of wrist exten­sion in all the participants was accompanied by some supination of the wrist.

TOC

DSP Software

The DSP (digital signal processor) is taking care of: Sampling and filtering, output of stimulation pulse, calculation of stimulation output and miscellaneous control tasks.

The software for the DSP is coded in the TMS320C50 assembly language consists over 2000 codelines. Besides the signal processing it enables communication and data ex­change capabilities to the optional host computer. The program provides different signal processing strategies, which can be changed by the host computer. A combination of the following signal processing steps are possible: The functions in italics are the default, Mode1.

Signal name

Processing

X=

Sampling of amplified signal

Y=

Stimulation response suppression filter (1.order transposed FIR-filter )

Noise indication

Noise detection

MES=

Filtering the blocs of data (Changeable coefficients)

MA=

Counting samples above a threshold

Rectified mean value

Root mean square

M=

FIR low-pass filtering

IIR low-pass filter

I=

Stimulation amplitude = Piece-wise linear function

Constant stimulation

Table 3.7-1 Signal processing

The filters with changeable coefficients are using a lookup table for the coefficients. At assembly time different tables can be used/created. The sampling frequency is set to 2kHz but can be changed in the source code for the program (it then has to be reassem­bled). Of the total stimulation interval, which is 60ms, the first 10ms is blanked, i.e. they are implicitly set to zero. The block length (samples in-between stimulation) is 100 samples. This is equivalent to 50ms of the myoelectrical signal.

The stimulation pulse is given by a sequence of 10 interrupt routines with an interval of 0.1ms. This gives the opportunity to select an arbitrary stimulation pulse with a duration of 1ms with a resolution of 0.1ms. The used stimulation from is a biphasic pulse with equal positive, inter- and negative pulse have a duration of 0.3ms each. The execution of these tasks are illustrated in Figure 3.7-1

Figure 3.7-1 Timing of program

The sampling and output of the stimulation pulse have the first priority since they de­mand precise timing. Calculation of the new stimulation amplitude can be performed in parallel with the sampling. It must be completed before the stimulation starts. The signal processing flow is illustrated in Figure 3.7-2.

Figure 3.7-2 Flow of the signal processing

The control tasks can be executed asynchronously in parallel with the two other proc­esses as indicated on the timing diagram Figure 3.7-1. These three classes of processes are controlled by pointers. The pointers are: SamplingStateSel_M.which takes care of the sampling, ProgramModeSel_M selecting the signal processing of the blocks. Finally there are two background process pointers BackProcessSel_MeCFES and ComModeSel_M for the asynchronous tasks.

Figure 3.7-3 Task dispatching

The pointer, SamplingStateSel_M, is controlling the sampling process using the timer interrupt, as shown in Figure 3.7-3. After 100 samples the pointer is being changed to the stimulation process and then again after 1ms to the control process during the blank­ing of 10ms. Totally a stimulation period of 60ms. Changing the stimulation period from 16.66Hz to 15Hz it is done by extending the blanking time by 6.66ms. This can be done by correcting a fine-tuning parameter called PeriodTune_k that extends the control process.

A feature that recently has been implemented in the program is a mode for data logging of a myoelectrical signal. This can be used to register spasticity in a muscle over 24hours. In this mode, SpasmRecord, the stimulator module is not used. The program calculates the RMS value of the myoelectrical signal in blocks of 6 seconds. These values are saved in the FLASH and can be downloaded to the host computer after the end of the recording period.

TOC

Host Computer Software

Figure 3.8-1 Configuration

A range of data exchange options can be accessed by connecting a host computer (80386 IBM compatible PC or higher ) to the MeCFES. The host computer program, containing over 3000 code lines, is written in Turbo Pascal as a DOS application. The primary purpose of the program is to program and test the DSP. The features are:

In addition the program performs the tracking tests, endurance tests and the recruitment measurements.

TOC

Summary

This section gives a summary of the specifications for the hardware. The hardware is produced in surface mount technology, minimizing size and weight

The MeCFES system, comprising stimulator, DSP and power supply, is build into a 11cm x 7cm x 3.5cm box with a total weight of 200g. The box is intended to be placed e.g. in a pocket A flat cable connects it to the amplifier which is intended to be placed near the recording electrodes. The amplifier is build into a 5mm x 45mm x 35mm box. The bottom of the box is mounted with the reference electrode. It is the intention to have the amplifier enclosed in the electrode-mount (described in 2.8.4 Electrode-mount ) near by the recording electrodes. The system is powered by 4x1.2V rechargeable NiCd batteries with a total capacity of min 2Wh.The measured power consumption of the four parts is

DSP (executing signal processing.)

26mW @ 3,2MHz clock. (3.3V)

Amplifier

18mW. (±3V)

Stimulator (Type1)

10-150mW stimulation dependent

Power supply

Not available (unstable efficiency)

Some of the following measurements are described in detail in section 4.1 Hardware Performance.

Amplifier Specifications

The produced amplifier has an artefact suppressing construction and needs no shut-down during stimulation.

Power supply

±3V

Active current*

6mA

Input impedance**

>10GW common mode

>10GW differential mode

Gain* , Input range

74dB, ±600mV

Frequency range*

10-500Hz

Filter

Low pass: 2 order Bessel

High pass: Non-linear feedback

Common-mode rejection**

>110dB + active grounding of patient

Noise related to input*

400nV(RMS) or 3mV(peak-peak

Recovery time@10mV step as differential input(2)

50ms-100ms

Offset compensation activation level

±0.1mV

ESD & Stimulation artefact protected

Digital Signal Processor Specifications

Power supply

3.3V

Active current

8 mA

Microprocessor

TMS320LC50

Speed

1,6 MIPS (Million Instructions Per Sec.)

FLASH Memory

64kWords

RAM

10kWords

A/D converter

10bit >10kHz, 11 channels

D/A converter

12bit >100kHz serial

Logic outputs

6

Serial communication port for host computer connection

Digital signal processing is fully software controlled. The signal processor program has a mean execution time of : 0.788MIPS with a peak at 0.972MIPS during data acquisi­tion and host communication. The DSP takes care of all control and signal processing tasks.

Stimulator

The stimulator is converting the signal from the DSP to a current. The shape of the stimulation signal is controlled by the DSP. The Type 1 stimulator is used.

Power supply

±3V and

Variable -75V

Typical output power

14mW @ 15mA, 2kW load,16Hz

Pulse shape

Biphasic with interpulse interval. (DSP controlled).

Current forced to zero in inter stimulation period.

Pulse width

0.9mSec (DSP controlled)

Current output amplitude

<50mA

Pulse repetition rate

Arbitrary

A power efficient concept for the stimulator has been chosen. It is based upon a switch mode DC-DC converter to produce high voltage. This voltage is controlled by the DSP. In this way the voltage for the stimulator output stage is kept as low as possible in order to increase efficiency. A transistor based output stage provides the stimulation current.

Power Supply

The system is powered by build in batteries. The power supply for the Type1 stimulator is taking care of: Charging of the battery, Sound module, push button and generation of 5 different voltage levels.

Power supply

4x1,2V NiCd batteries

Output voltages

±3V for analog circuits

3.3V for digital circuits

0-±75V for stimulator circuit

Charging input

1A @ 10-20V AC or DC

High voltage output efficiency

<40% @ 70V,0.3mAmean output

High voltages are generated using switch mode power supply technique.

* Measured

** Specified by IC manufacturer

Discussion and Conclusion

TOC

Résumé

The development of a small portable, battery powered device, the MeCFES has been described. The primary goal has been to establish a grip in cervical spinal cord lesioned with pa­ralysis of the hand and paresis of the wrist extensor muscles. By enhancing the wrist extension force the tenodesis function can be used for the key grip. The MeCFES can record the voluntary myoelectrical signal from a muscle and use it for control of functional electrical stimu­la­tion of the same muscle. The size is 11cm x 7 cm x 3.5cm with a weight of 200g and the device is rechargeable. For a paretic wrist extensor muscle the MeCFES will provide an amplification of the muscle contrac­tion. Both for recording of the myoelectri­cal signal and for the electrical stimulation surface electrodes attached to the skin are used. A model of the recorded signal has been developed identi­fying the signal and the sources of noise. It is used to specify the demands to the hard­ware and software. The system has been tested by C5 spinal cord lesioned tetra­ple­gics and the perform­ance has been evaluated by tracking tests and functional tests on 5 tetraplegics.

A pair of surface recording electrodes placed over the muscle picks up the signal in­cluding the voluntary myoelectric signal. This is fed to the MeCFES that amplifies, filters and converts the signal to a control for the amplitude of a 16Hz biphasic stimu­la­tion current. The stimulation cur­rent is applied to the same muscle using a pair of stimulation electrodes. The muscle must have innervated motor units. A part of these can be paralyzed. The muscle must be superficial so it can be reached electrically by the surface electrodes.

TOC

The Technological Context

By the start of the project in 1994 there were found no commercialized products for restoration of the hand function in tetraplegics. Since then, the Handmaster™ from NESS ltd. Israel and the Freehand™ system from NeuroControl Corp.,USA has been launched. The Bionic Glove from University of Alberta, Canada is in approach. None of these systems are using the myo­elec­tric signal as control.

The Freehand™ system is using implanted electrodes, which give an accurate selective stimulation. Another advantage is that it is an "invisible" system. The MeCFES ap­proach can bee seen as a safe method of testing a functional elec­tri­cal stimulation (FES) system without having implants. This applies for the user prior to selec­ting or pur­chas­ing a FES system and to the investigation of control/stimulation strategies. The MeCFES is thus an alternative or supple­ment to an implanted system. It may also be seen as the first step towards a MeCFES similar control strategy for an implanted system.

The Handmaster™ and Bionic Glove are both non-invasive systems using surface elec­tro­des for stimulation. Both uses on/off control of the stimulation but, where the Handmaster™ uses a grasp program started by a push button, the Bionic Glove uses a mechanical wrist extension sensor for the control. The control movement for the Bionic glove is then similar to the MeCFES apart from the control being not linear. The MeCFES approach is from the user point of view assumed to be the most convenient control method offering a con­trol of the force. The drawbacks are that the myoelectric signal as a con­trol signal is more unreliable than the movement as a control signal and as it appears from the experiments only few tetraplegics have the type of muscle pare­sis that is required for the functionality of the MeCFES.

TOC

Progress of the Project

The MeCFES concept was developed and proved feasible by E-U. Haxthausen at The Technical University of Denmark (DTU) in 1992 [Haxthausen 1992]. In continuation S. Sennels, DTU has refined the signal processing and control strategy [Sennels 1996]. Both have provided test results from tetraplegics using the concept. This projects uses the same principles as developed by Haxthausen. The tracking test method used by Haxthausen, has been used as a part of the evaluation of the MeCFES.

The contributions from this project have been:

Hardware and Software Evolution

The system used by Haxthausen was non portable since it was designed for research only. The system comprised an amplifier, stimulator as separate units powered by mains supplied power supplies. Signal processing was taken care of by a personal computer. The developed MeCFES is a system where amplifier, stimulator, signal processor unit and power supply is mini­mized in size. This has required a totally different design of all parts. Special attention has been paid to minimizing the power consumption. The system is now portable and powered by batteries. The software program for the device has been written and control/programming software running on a PC platform has been developed.

Model Evolution

In the minimization of the hardware design, there has been a need for knowing the speci­fications that the system has to meet. Since the aim is a system for functional use the reliability of the system should be optimal. These items have led to the development of a model of the recorded signal from a stimulated muscle. This reveals the possible problems and demands to the signal processing.

Signal Processing Evolution

The digital signal processing is using a more simple filter than Haxthausen proposed. A new method (threshold counting) for converting the voluntary myoelectric signal to a control for stimulation is proposed and used. No objective measurements on whether this method is better than the average rectified value method used by Haxthausen (and many others), has been performed by experiments on tetraplegics.

Amplifier Enhancement

Typical amplifiers as well as the one designed by Haxthausen or Sennels have a high-pass filter that can prolong the stimulation artefacts as described in section 2.5 Signal Ampli­fi­ca­tion. Since a low voltage amplifier is more sensitive to this problem it has been neces­sary to find a solution. A novelty in the hardware design is the MeCFES amplifier. It is different from other amplifiers in the way the high pass filtering is achieved. The MeCFES amplifier is a stimulation artefact suppressing fast reco­very amplifier that enables the recording of the myo­electrical signal in the presence of short pulses that are several orders of magnitude larger than the desired signal. Another feature of the ampli­fier is the fast recovery from DC offset changes in the input signal. This can be caused by a change in the half-cell potential of the recording electrodes due to mecha­ni­cal actions. SPICE™ simulations of the conventional amplifier and the MeCFES ampli­fier have shown that the MeCFES amplifier is better suited for the pur­pose. Recordings on the amplifier have verified the simulation result.

Stimulator Development

The low size and low power consumption demands could not be met by the stimulator design used by Haxthausen. For that reason the stimulator has been totally redesigned Three different concepts have been developed and tested.

New Experiments

The tracking test performed on the 5 tetraplegics has the same concept as used by Haxthausen but has been expanded by the endurance test.

The tracking tests showed that there is a better control of an increasing force or angle than of a decreasing force or angle, where the angle is the angle of wrist extension against gravity. The participants with high muscle strength have better control than those with low muscle strength. A participant (subj.: KGN) with the weakest voluntary unas­sisted wrist exten­sion which was less than a 2° angle against gravity obtained a 23° angle against gravity by use of the MeCFES. For the same person the MeCFES amplified the isometric muscle force from 1N to 13N. The force/angle amplification ranged from 1 (i.e. no im­provement) for the strongest participant to a magnitude of 10 times for the weakest participant.

The results in form of the root mean square of the error are summarized in Table 5.1. An error of 57% is corresponding to no movement at all. First number is the tracking error without the MeCFES and second number is with use of the MeCFES.

Subj:

AA

EG

FB

KGN

KN

Angle range

32°

43°

38°

23°

48°

Error % RMS

9  |  9

28  |  9

8  |  7

60  |  32

39  |  14

Force range

16N

52N

41N

13N

28N

Error % RMS

19  |  17

17  |  17

9  |  13

57  |  18

16  |  8

Endurance

50% 16N

50% 52N

50% 41N

50% 13N

50% 28N

Error % RMS

10  |  10

see 4.5.2

10  |  7

50  |  35

5  |  5

Table 5.1 Summary of the tracking test results. Results without | with the MeCFES.

Only in the subjects EG, KGN and KN the MeCFES gives an improvement of the move­ment. As found by Haxthausen and Sennels it is difficult for the tetraplegics to control the stimulation of an accurate movement. Sennels has shown that a reason for this is that the myoelectric signal is not a reliable control signal. Control appears to be particularly difficult for a decreasing muscle contraction.

The degree of precision depends on the subject, the remaining volun­tary muscle strength and how well the various signal processing parameters are adjusted. The condition of the subject (concentration, fatigue etc.) affects the performance and must thus be con­si­de­red. A conclusion of the tracking tests is that the precision of the MeCFES assisted move­ment is history dependent. An increasing contraction is more accurately con­trolled than a decreasing contraction. This is explained by the non-linearity of the muscle. To further clarify the reasons for control difficulties, recordings of the recruit­ment curves (stimulation vs. muscle contraction) have been performed. It is found that the recruit­ment curve has a hysteresis-like shape, where the gradient of the current-output rela­tion for an increasing current differs from a de­creasing current. The decreas­ing slope is very steep. Less current is required to generate a certain muscle contraction from a previous high contraction than from a low contrac­tion. This phenomena is as­sumed to be the main problem in controlling a moderate contraction. The control prob­lem is not only the reliability of the myoelectric signal but also due to this recruitment curve.

This project presents the first evaluation of the MeCFES principle by functional tests. The conclusion of functional tests is that stimulating the wrist extensor solely provides no pinch grip due to finger extension but can provide a volar grip in 2 out of the 5 tetra­plegic participants. This is in accordance to the movement analysis performed by Sennels, where it was concluded that the pinch grip is not feasible to obtain only by surface stimulation of the extensor carpi radialis muscle.

New Stimulation Approach

As a spin off from the experiments, an efficient method of stimulating the hand muscles to obtain both key grip and volar grip has been found. This stimulation method has been possible on all participants. It has been found that the stimulation is suited for control by the wrist extension. Letting the wrist extensor controlled MeCFES stimulate the hand muscles, a useful grip was established in 3 out of 3 tetraplegic participants.

Electrode mount

A concept for a electrode mount, allowing easy placement of the electrodes has been designed. Together with the MeCFES the electrode mount will comprise the complete system that can be used at home by tetraplegics. The electrode mount is not tested.

TOC

Future Aspects of the MeCFES

Letting the MeCFES control stimulation of the muscles in the hand provides a good controlled grip. This stimulation can be a supplement to the stimulation of exten­sor carpi radialis to augment wrist extension.  A way to compensate for this is to stimu­late thumb flexion and finger flexion controlled by the MeCFES. Wrist extension will then control a functional electrical stimulation en­hanced tenodesis function, which pro­vides the user with a useful grasp. The conclusion is that the MeCFES should  addi­tionally stimulate selected muscles in the hand. In general the MeCFES can be used for all muscles where the voluntary myoelectrical signal can be recorded and the muscle can be stimulated by use of surface electrodes. The device has in its pres­ent state one channel for recording and one for stimulation, but can be extended to more channels by adding stimulator modules, amplifier modules and a minor software modifi­cation.

The principle of the MeCFES has advantages and disadvantages. It is a supplement to existing and future devices for hand function restoration. For the Handmaster™ that is using surface electrodes and has a well functioning electrode mount system, the MeCFES principle would be well suited. Instead of the trigger button, the wrist extensor muscles could control the stimulation. The muscles stimulated are near the con­trol­ling muscle, which requires the same stimulation artefact suppressing features as in the origi­nal MeCFES principle. This would provide the user with better control of the grasp and faster grip release opportunity. For the Bionic Glove, which already uses the wrist exten­sion, the MeCFES approach will be less attractive, but can maybe be used as an alterna­tive to the mechanical wrist angle transducer.

The system has been developed for restoration of wrist extension but there is reason to assume that the methods are applicable to all paretic muscles. It may therefore in addition be used in paraplegia and hemiplegia to assist lower limb movements. Such applications could be a foot-drop stimulator control and hip extension/flexion in stand­ing and walking. If the method shows feasible in longer terms of use, it can be used for control of implanted systems as well. For example the Freehand™ implant system could be controlled by use of the MeCFES principle, forming a hybrid using both surface elec­trodes and implanted electrodes.

Besides for the MeCFES application, the amplifier relates generally to the field of re­cording biopoten­tial signals from tissue that is simultaneously stimulated and thus ex­tends to other applica­tions than the MeCFES. An example could be recording of elec­trocardiograms during electroshock and evoked potentials e.g. from brain stimulation.

As a further result of the MeCFES development, the system can be used for acquisition of myoelectrical signals in general. The system can thus be used for a recording of spas­ticity in a muscle. This is an application that might find its use in clinical evaluation of a range of patients besides the spinal cord lesioned.

TOC

General Discussion

For a reliable system it might be inadequate to maintain the linear control, since this is not a robust control as found by Sennels or Saxena. These and other works have pro­posed a finite state control i.e. the myoelectric signal is classified in different levels, where the most simple is the on/off control. Functional tests of a MeCFES using linear control and a MeCFES using on/off (maybe intermediate states) control should be compared.

The MeCFES must be expanded to stimulate the hand to the grip. This approach should be tested on several tetraplegics and the fraction of the population of tetraplegics that can benefit from the device must be found. The tests must be compared to the tests of  the Handmaster™.

In this project the threshold count method of processing of the myoelectric signal should have been compared systematically to the use of average rectified value. The choice of signal processing is important to obtain best possible result and must therefore be investi­gated further.

The hardware needs too many corrections in its present form. The efficiency of the power supply and the performance of the stimulator are insufficient. The software is not easy to use for non-engineers and calls for a user-friendly interface.

The design and testing of an cosmetic acceptable and easy-to-use electrode mount is essential for the success of the system.

TOC

Market needs

The tentative marketing analysis shows a need for 400 devices/year in Europe and 500 in USA. The function of the device is very attractive for the users. The use of surface electrodes is essential and provides an alternative to surgical implanted electrodes. It will be relatively simple to apply to the customer and easy to service since it is a non-invasive system.

TOC
TOC

Conclusion

The development of a myoelectrical controlled functional electrical stimulator has re­sulted in a functional prototype, a portable MeCFES device. Its ability to enhance wrist extension has been tested on 5 tetraplegics with muscle strength in the range from 1 to 4 (MRC scale). The tests have shown that the MeCFES gives an enhanced force and movement range of the wrist to which it is applied. The technique is most feasi­ble for tetraplegics with a weak wrist extension. Tracking tests have shown that a MeCFES assisted movement can be controlled, but not with the same precision as a normal vol­untary move­ment. The precision ranges from nearly normal control to on/off control.

The conclusion is that the MeCFES is a feasible concept for restoration of hand function in spinal cord lesioned persons. It is a technological platform from which there are possibilities for further evolution of functional electrical stimulation aids for disabled. It offers a complimentary solution to implanted systems, allowing the po­ten­tial recipients of this device to test the benefits from functional electrical stimula­tion, in a simple non-invasive way, before making a decision for a permanent func­tional elec­tri­cal stimulation system. There are on the other hand still many practical problems that have to be solved before the MeCFES can be commercialized. This applies espe­cially to the design of the electrode mount and the robustness of the device.


Home