Designing and Implementing a High Density Electromyogram Sensor Array

Joshua Hernandez, EEConor O'Reilly, EE Advisor - Professor Hanson

Our Goal is to… Extract data representing activity of individual

fingers from surface EMG data, by

Developing and implementing a sensor array, to be worn on the forearm, capable of providing both spatial and temporal EMG data, and

Developing a processing algorithm to utilize the spatial and/or temporal data to decompose the raw EMG data into its SMU constituents and interpret finger activity.

Electromyogram Fundamentals There are two basic types of electromyogram:

Intramuscular EMG▪ Needle electrode is inserted directly into the desired muscle▪ Direct electrical connection established to muscle tissue▪ Records only local motor unit (MU) action potentials (MUAPs)▪ High signal to noise ratio

Surface EMG▪ Electrodes are placed on skin over the desired muscle▪ Indirect electrical connection - signal must pass through the

skin▪ Records a summation of MUAPs over a large area▪ Potential for a low signal to noise ratio

Why Use a Surface EMG? Intramuscular EMG requires needles. Needles are no fun. A surface EMG (sEMG) is non-

invasive Therefore, it is fun.

Inherent Problems with sEMG

Time Delay Nerve impulse takes time to propagate

through the MU▪ Conduction Velocity (CV) is typically 1 - 10

meters/sec▪ Muscle fibers within an MU contract at slightly

different times Delay due to CV causes distortion of the

recorded signalSignals from multiple MUs present

in signal A single muscle often has multiple MUs

Crosstalk Electrode may record signals from nearby

muscles

So what are we dealing with?

We have:0.5 s of sEMG data using a normal

double differential (NDD) electrode configuration recorded from the

tibialis anterior of a healthy subject during a 30% maximumvoluntary contraction

We want:A bunch of these

Sensor Array Placement

Inter-electrode Distance Motor Unit Innervation Zones

A 61-electrode array recording biceps contraction. (A) is filtered raw data, (BCD) have been decomposed.

Illustration showing the relation of electrode position to MUAP characteristics.

Sensor Array (cont.) Design

Materials▪ Pyralux ▪ Stainless Steel Pads▪ Circuitry

CAD▪ Need CAD software that supports large boards

Cost Health Efficiency Signal Quality

Sensor Array (cont.)

Construction Printer/Pyralux Stainless Steel Pads Amplifiers

Testing Placement Spatial Resolution Signal Quality Determine Proper Sampling Rate

Printing & Etching wet chemical etching process:

used a Xerox™ Phaser® 8650N solid ink printer to print wax directly onto a sheet of DuPont™ Pyralux® AP9121R

Block Diagram

A/D μC Wireless Link

Processing Algorithm

SMU Classification

Decomp. of Signals

μC Board Computer

Control

Wireless Link

MATLAB IMPORT

Determine finger value

Sensor & Amplifiers

Filtering

Computer Connection A/D Hardware

Number of Channels Needed Bit Rate/ Depth Multiplexing?

Connection Microprocessor Microproccessing USB connection

▪ Possible Future radio Link Drivers

▪ Import to MATLAB

Receiving the Signals

The Vernier EKG (Electrocardiogram or ECG) Sensor

Processing etc… Digital Filter

60 Hz notch Additional Noise Filtering (20Hz < Signal <2kHz)

SMU Decomposition MUAP Classification Decompose Signal Assign Finger Values

Processing Algorithm Wavelets

▪ Mexican Hat▪ Reqs. Unipolar input

Amplitude/C.V.▪ Reqs. Bipolar i/p

Support Vector Machine▪ linearly classifies the features into related groupings

Processing of Data

Read Raw EMG

Signals

Divide Into 250

ms Segments

Filter Noise

Identify Root Mean

Square Features

Identify Frequency Energy Features

Identify Phase

Coherence

Features

Send to Classifica

tion Algorith

m

Import Features Data into the WEKA

ToolkitOr

MATLAB

Specify a Data Set

to be Used as a Training

Set

Run the Training

Algorithm

The SVM Generates Classifica

tion Function

to Identify Future

Data Sets

Generate a Data Set

to be used as a Test Set

SVM Uses Classifica

tion Function

to Classify MUs

Function Returns Which

Finger is Moving

Data Prep and Feature identification:

Identification Algorithm :

Questions?

Special Thanks to…

Konstantin Avdaschenko, EE Demarcus Hamm, EE Travis Hoh, Neuroscience Prof. Hanson, EE Prof. Hedrick, EE Prof. Catravas, EE Prof. Olberg, Neuroscience Prof. Rice, Biology