Brian Do and the Bionic Bunnies Myoelectric Prosthesis Johns Hopkins Applied Physics Lab, Baltimore, MD Alex Sollie |Callie Wentling | Michael LoNigro | Kerry Schmidt | Elizabeth DeVito | Brian Do Objectives • Create a myoelectric interface device • Apply current technology in medical prosthetics Brian Overview Electromyography (EMG): is a technique for observing the electrical activity produced by skeletal muscles. Myoelectric signals: Signals caused by contraction of skeletal muscles. Prosthetic: Artificial device extension that replaces a missing body part. Brian Objectives Figure 1: [1] pg 454, Relationship between normal and myoelectric control, CNS – central nervous system Brian Feasibility Myoelectric Signals Brian Feasibility Brian Division of Labor Division of Labor Signaling Electrode Design Brian/Callie Analog Filtering Brian/Callie/Elizabeth Analog Signal Processing Brian/Kerry Noise Reduction Callie/Elizabeth Signal Amplifications Brian/Elizabeth Buffer System Elizabeth Wireless Transmission Michael/Alex/Kerry Analog Digital Converter Michael/Alex Computer FPGA Michael/Alex Digital filtering Michael/Alex Device Drivers Michael/Alex Code/Processing Michael/Alex Input/Output Module Michael/Alex/Callie Control System Michael/Alex/Brian Printed Circuit Board Michael/Alex/Kerry Amplifier Mechanical Power Motor control Prosthetic Design Wireless Interface Packaging Signals - Brian/Elizabeth/Callie Computer - Michael/Alex/Callie Mechanical – Kerry/Brian/Elizabeth Mechanical Kerry/Callie/Elizabeth Kerry Kerry/Alex/Michael Kerry/Brian Kerry/Michael/Alex Kerry/Elizabeth Brian Division of Labor Brian Goals levels Base Level: • Basic myoelectric control, single channel, output to LEDs Mid Level: • Multi-channel myoelectric control, 4 set heuristics, embedded, simple prosthetic High Level: • Compatible with amputee anatomy, wireless electrode design, multi-channel control Brian Physiology Action Potential (AP): the chemical depolarization of a muscle cell Myoelectric Signal (MES): the resulting electrical activity of AP propagation through the muscle Callie Action Potential Callie Callie AP Propagation Callie Callie Electrodes • Detects electrical potential of muscle cells • General picture of muscle activation • Muscle contraction AP Callie Callie Bipolar Electrode Technique • • • • • 3 electrodes / signal Differential amplifier between two electrodes Reference electrode Negates transducer noise Maximize SNR Callie Callie Electrodes Callie Human Interface Concerns • • • • • • • • • Impedances Differentiation Cross talk Normalization Dry vs. Gelled Electrodes Fiber Density Electrode Distances Temperature Physiological Conditions Callie Callie Calibration • Repeat or new users • Response to impedance and normalization • Initialization system: detects min and max for each muscle system based on electrode placement and differences between users • Affects software base values Callie Signal Flowchart Sensing Processing Output Elizabeth Signal Sensing User’s muscle signals Electrodes Buffer High Gain Amplification Stage Initial Filtering (SNR) Elizabeth Noise • Our myoelectric signals are expected to be very noisy; we will filter out the noise. • Sources for the noise include heartbeat and other muscle movements. – Can’t isolate one muscle – 60 Hz from environment • Need good reference points for filtering. • Want maximum signal-to-noise ratio (SNR) . Elizabeth Safety Concerns • Need to ensure no current is able to travel through the electrode to the user. – Buffer circuit. – High impedance during the amplification stage – Lower power • Wires dangling from subject – Wireless Implementation Elizabeth Schematics For Signal Sensing The Instrumentation Amplifier to the left, provides a buffer as well as high gain. 4-pole low pass filter Elizabeth Risks and Contingencies • Weak Signals – Group members are working out to increase signal strength – backup plan • Broken Parts • Time – Work effectively as a team • Cost – Try not blowing chips – Order backup parts – ESD safety Elizabeth Computing FPGA - Overview Why FPGA? • Use signals to control a variety of things. • Need an IC that can be easily re-programmed for different tasks. • Can also re-purpose pins for extra analog to digital capabilities. Michael FPGA – Inputs/Outputs Input • Myoelectric signal (~60 Hz) • Sample waveform Functionality • Analyze digital waveform Output • Corresponding analog signal to control motor Michael FPGA Possibilities • By using the re-programmable FPGA, we can control a variety of devices. • Simple LEDs for testing. • We can output arm movement information to a computer screen. If a robotic arm design falls through, we can try to design a virtual arm. • Final goal: a semi-realistic robotic arm Michael FPGA Controls • Most important FPGA task: – Determine what arm motion should occur based on the myoelectric signals from multiple electrodes. • This is based on signal amplitude (minus the noise) and also signal shape and approximate frequency. Michael FPGA Controls Some different signal shapes that we’ll have to take into consideration. Michael FPGA Controls The speed of the arm movement can be deduced from the relative amplitude of the signals. Michael FPGA Controls • We would also like to program some easy realistic arm movements using heuristic rules. • These are educated decisions on how some motors should operate based on operations of other motors. Michael More FPGA Information • It is highly likely that we will need to utilize frequency information of the myoelectric signals to make control decisions. • On the FPGA we will need to implement some sort of FFT algorithm. • We may need to utilize the Altera FFT MegaCore for this task (compatible with the Cyclone II FPGA). Michael FPGA – Risks and Pitfalls • The entire project is dependent on successful sampling and digital processing of the myoelectric signal. • Processing times: how long is the sampling and processing going to take? • The FFT implementation could become incredibly complex. If frequency analysis falls through, we can try to glean all the information we need from the amplitudes of the different electrodes. • We need to sample 5+ signals simultaneously. We may need to use multiple FPGA boards to achieve this (depending on how many A/D conversions we can squeeze out of one board. Michael Risk Analysis • Even an ideal electromyogram will be around 6mV at its maximum amplitude. • If we determine the movement type based on signal frequency, we will need a clean strong signal, to avoid mistaking noise for a waveform. • Notch filtering should be avoided, so noise needs to be minimized. Alexander Sampling Spectrum Alexander Risk Reduction • Noise reduction will be crucial – One way to reduce noise will be by using Bipolar electrode arrangements – Essentially a pair of electrodes, which use sample, then subtract out signals common to both with a differential amplifier – The idea is to eliminate noise present at all points on the surface of the skin Alexander Signal Isolation • Minimize lead lengths at all costs - even house the preamp on the sensor – This is important to minimize coupling with environmental AC power, as well as control signals present in the device • It is important that pre-amplifier circuits have strong DC component suppression circuitry. – Even a small DC component would drown out the signal after amplification • There are DC components caused by factors involving skin impedance and the chemical reactions between the skin and the electrode and gel. Alexander Optimizing the Usable Signal • It is very important that EMG pre-amplifiers have high input impedance. • Input (i.e. source) impedance is typically less than 50 kOhms with gel electrodes and proper skin preparation • To avoid input loading, the preamp needs a very high input impedance – 10s of MOhms for gel electrodes – 1000s kOhms for dry electrodes Alexander Scheduling • So lets talk for a moment about how all of this will be completed • There are three main parts to this project – Sensing and Analog Signal Processing – Digital Signal Processing and Control Logic – Device Hardware Alexander Prosthetic Arm Input Functionality Output • FPGA-Processed analog signal • Magnetic energy spins the rotor • Rotation speed dependent on amplitude and duration of signal • Motor swings the forearm appropriately Kerry Prosthetic Arm (Higher Level Design) Fore-arm twisting motion • Activated by pulsecontrol • Would require a specific, alternate signal from FPGA Kerry Prosthetic Arm (Higher Level Design) Clamping motion • Also activated by pulse-control • Would allow for pinching and grasping actions Kerry Bill of Materials Part Cost ($) Mechanical Hardware 250 Surface electrodes and gel 50 Motors and drivers 150 PCB fab (2 revisions) 100 FPGA 50 Hardware: op-amps, wires, resistors 150 Wireless transmitters and receivers 175 Clamp 20 IC chips 60 Printing 130 Total 970 Kerry Questions ??? No? GOOD.