EMG/EEG Controlled Prosthetic
Fangyi Lu Mary Nguyen Yen Nguyen Dana Steinberg Zoe Tcheng
University of California, San Diego
Abstract—This project study focuses on a variation of a onset of an action [3]. The P300 component is isolated from
prosthetic design using a combination of electromyography the beta frequency band of brain activity.
(EMG) and electroencephalography (EEG). This set-up
primarily covers prosthetics that limb loss that begins below
the elbow, such that the individual still has their bicep. To
cover different issues that arise with prosthetic device control
that use one or the other, the complement of both procedures
can be set up to provide more precise signals. The project
design encompasses receiving outputs from both filtering
circuits to process using Arduino hardware. From there, the
filtered signals will pass through specific logic gates before
being sent to a motor-controlled prosthetic device.
Keywords—prosthetic, EEG, EMG, Arduino, filtered signals
I. INTRODUCTION
Nearly two million people in the United States alone live
with limb loss. Vascular disease accounts for about 82% of
limb loss, which includes blood vessel disease (known as
peripheral vascular disease), diabetes, blood clots, and
osteomyelitis. 22% of limb loss is due to trauma, with trauma
also causing 75% of upper extremity loss. 4% of limb loss is Figure 1. P300 Component: The positive-going potential can be seen starting
due to congenital disease and another 4% is due to tumors [1]. between t = -1.5 and t = -0.5 with reference to when the action is initiated (t=0).
70% of upper limb amputations are below the elbow, which is [3]
the area of focus for the prosthetic design. To minimize the cost and maximize the ease of use, non-invasive
methods are preferred.
A long-standing goal of neuroprosthetic research is the
optimization of prosthetic control. The two main methods of II. EMG
obtaining the biological signals for control are
electroencephalography (EEG) and electromyography (EMG) A. Overview
coupled with machine learning. EEG tests the electrical The EMG signals are the electric potentials generated by
activity of the brain by using electrodes. The output is the the activation of muscle fibers in neuromuscular activities, and
summation of the electric field generated by synaptic the patterns of EMG signals are recognized to control the
potentials. EMG tests the electrical activity of skeletal muscles prosthetic devices for the lower arms in this case. The EMG
caused by their activation. data serves as the basics to the analysis of correlation between
Current issues of EMG controlled prosthetic devices muscular activity and central nervous system control. The
include noise that arises from movement, changing electrical process designed for the EMG part is: raw data collection,
activity due to muscle fatigue, and the fact that each amputee amplification, filtering, and digital processing.
has different muscle characteristics. The major issue with EEG B. Electrode placements
control is the lack of spatial resolution that can be attributed to
the ex vivo collection of data. To address these problems, the The triceps and biceps muscles control the position and
P300 potential recorded from EEG will be used to confirm the movement of the forearm, and the triceps connect the shoulder
intention of movement that is indicated by the contraction of and the joint to enable the slightly bent flexion of the forearm,
muscle. As previously observed between separate trials and which is considered to be the common status after the loss of
individuals, the P300 is a stable and consistent enough input to forearm. Therefore, this experiment will focus on the main
control the Brain Computer Interface (BCI) of a prosthetic [2]. muscle groups of the triceps brachii: long-head, medial-head,
and lateral-head. In the experiment that is referenced, the
The P300 is an event related potential that arises due to researchers applied 8 electrodes on one arm: 2 electrodes on
one’s intention to make a movement. It is a positive-going the belly of each head, and the reference electrodes at elbow
potential that occurs one to two seconds before the physical joint and shoulder joint.
Compared to the experiment focused on the speed
variations above, this experiment proceeds in the relatively
stationary status with only two available inputs to the
processing circuit, so the number of electrodes is reduced to
match the designed circuit. Since the long-head of the triceps
brachii extends all the way from the scapula to the elbow joint,
and it functions in the extension of the forearm, it is assumed
that patients with prosthetic will have higher activity here
because they may find easier to begin practice with the help of
shoulders. Hence, both the positive and negative input
electrodes will be placed on the long-head of the triceps and a
reference electrode placed below the elbow.
Figure 3. Simulation circuit for EMG.
The three operational amplifiers on the left form a standard
biopotential amplifier with a high gain as shown in Equation
1, a high input impedance, and a high common-mode rejection
ratio.
2R 2(10k)
Gain = 1 + RG = 1 + 2k = 11 (1)
Figure 2. Simplified placements of electrodes: on the long-heads of biceps. The rest of the circuit from C1 onwards forms an active
[20]
bandpass filter where the high pass filter has a cutoff
For both measurements of EMG and ECG, the Ag/AgCl frequency of 1.6Hz (Equation 5), the low pass filter has a
wet electrodes will be used because they are low-cost and cutoff frequency of 106Hz (Equation 6), and has a gain of 20.6
generate low noise levels, and the skin should be shaved and (Equation 7).
cleaned to have better contact with the electrodes. 1
f c,low cut = 2π·C1·R1 (5)
1
C. EMG Signal Acquisition f c,high cut = 2π·C2·R3 (6)
R2
G = 1 + R4 (7)
EMG signals were recorded from the long head brachii of a 20
y/o female. Data was collected with an Arduino Mega 2560, The final output would then be inputted into an Arduino to
3M 2560 Red Dot ™ Monitoring Electrodes with Foam Tape interface with a computer where an appropriate computer
& Sticky Gel, and 3 standard electrode leads. Using C++, the program would be used to confirm the intention to make a
speed of data acquisition was set to 9600 bits per second. The prosthetic move. After this confirmation, the program would
arduino reads the signal input and outputs it onto a serial make the prosthetic move with a motor to perform a function.
monitor in units of µV. Three trials: No Flex, High Flex, and
Low Flex, which refers to the strength of tricep tension, are E. Results
recorded. The three trials are repeated for two sets of data with
When the circuit in Figure 3 was tested with an AC voltage of
each set having the signal transduced by the Arduino recorded
from either the proximal or distal electrode and the results are .001V in the right lead (green) and .003V in the left lead (light
shown in Table 1. blue) and 100Hz frequency for both leads in Multisim, the
graph produced is shown in Figure 4. As expected, it shows
Trial Signal Electrode Average Voltage (mV) significant gain of the final output signal in dark blue and an
No Flex Distal 332 ± 34 offset caused by the bandpass filter.
No Flex Proximal 311 ± 13
Low Flex Distal 305 ± 10
Low Flex Proximal 306 ± 11
High Flex Distal 309 ± 13
High Flex Proximal 306 ± 12
Table 1. Average and Standard Deviation of EMG trials
D. Circuit design Figure 4. Test simulation for EMG using AC inputs.
The circuit shown in Figure 3 is the one used for EMG When the same circuit was simulated in Multisim, this time
simulation. with the data acquired from EMG part C, the right lead (green)
and the left lead (light blue) produced the right image in
Figure 5 and the final output is shown in the left image.
Similar to the AC test, this final output shows significant gain
of the original signals. The output spikes mirror those of the
inputs and some filtering occurs.
Figure 5. Simulation for EMG using raw EMG electrode data.
III. EEG
Figure 6. Labels for points according to 10-20 electrode placement system
A. Overview [15]
An EEG recording as mentioned, is a way to understand Due to the limitation of the instrumentation amplifier only
brain activity. This is a great advantageous point to the project allowing two inputs at a time, if multiple parts of the brain
proposal as it supports the EMG reading in verification that were to be examined, then multiple electrode pairs need to be
the brain is processing the muscle movement of one's biceps to connected to their own instrumentation (IA) amplifier. For the
start the control of the prosthetic limb of the lower arm. From purposes of experimental analysis of motor function, the
an electrical circuitry standpoint, the outline to design the central region of the brain, electrodes F3 and F4 are utilized as
EEG part of the circuit is as follows: raw data collection, an electrode pair as shown in Figure 6. If possible, all areas
amplification, noise reduction, filtering, and finally should be tested to see the relative activities during different
computerized processing of signal. behaviours to choose the best electrodes placement. Both
electrodes can be isolated to measure the lateralized readiness
B. Electrodes placements potentials, which is the preceding potentials occurring at the
First, it is necessary to point out which type of brain wave motor cortex in response to the contralateral limb movement,
the circuit should be reading. There are many types of waves very similar to P300 component mentioned. For instance, the
to analyze, depending on what type of activity is being move of the left arm will cause a more negative potential on
observed. Since the target raw data from the body is more the F4 area and a less negative potential on the F3 area and
muscle movement along with spatial awareness, alpha, or a vice versa. Taking the average of differences in both left and
frequency range of 14-30 Hz is exactly what is needed. Table right limb movement and this value is known as the lateralized
X shows the possible types of brain waves that exist. readiness potential. So in this simplified 2-channel EEG
recording, the positive and negative electrodes are placed at
EEG wave type Frequency range (Hz) Amplitude (mV) F3 and F4 respectively, and the reference electrode is placed
γ 30-100 <50 on the right ear, commonly known as A2 (not pictured on
Figure 6).
𝛽 14-30 <50
C. Circuit design
α 8-14 30-50
The electrodes are wired up to a protection circuit. This
θ 4-7 50 region of the EEG circuit allows for the prevention of
δ 0.5-4 ~100-200 overvoltage or overcurrent to happen, or rather the user
bearing consequences of malfunctioning circuitry. This
portion consists of an array of diodes and resistors to allow for
Table 2: EEG brain wave signal types (frequency ranges may vary due
to multiple experiments done in measuring such an array of signals). that overvoltage/overcurrent prevention. In this case, instead
of a numerous number of diodes and resistors to keep track of,
Typically, researchers use an EEG electrode cap with the just resistors are used. The CMRR of a typical EEG circuit
10-20 electrode placement system for EEG recording. The cap ranges from 10,000 - 100,000 (80dB - 100dB). This design
ensures the Ag/AgCl wet electrodes fix precisely on the scalp utilizes the typical 80dB CMRR to help minimize noise from
during the recording without additional markings. The other objects that may interfere with the raw signal
placement system divides the brain into 5 areas, F(frontal), acquisition. Voltage outputs of the protection region will then
C(central), T(temporal), P(posterior), O(occipital), be fed into an instrumentation amplifier.
proportionally to the surface of the head for recording.
The CMRR, or the common-mode rejection ratio and the
corresponding value in decibels are calculated by the
following equations:
|A |
C M RR = | Adc |
| |
C M RRdB = 20log 10 (CM RR)
where Ad is the differential gain and Ac is the common-mode Pen and paper calculations based on the concept of output
gain. voltage of an instrumentation amplifier should be the
following:
Once the raw signal passes through the IA, it will travel
through a bandpass filter to extract the target frequency ranges
R9 R1
to pass through, which is the 14-30 Hz frequency band. The V dark blue = R8
(1 +2* RG
)
filtered EEG signal, along with the output of the filtered EMG
signal, will feed as inputs to the Arduino Mega 2560 where R1 = R2 = R7 = R8 = R9 .
microcontroller for digital processing and be able to output a
voltage to the servo that will also be connected to the Arduino With countless trial and error, the output of all probes from
Figure 8 mostly validates paper calculations. Due to the nature
board.
of not being able to have readily available EEG electrode data
to filter out necessary beta waves, the simulated AC voltage
that would be picked up by the electrodes must suffice. The
blue graph is reading the voltage right at the intersection of the
midway point of the IA and the DRL component. The green
graph measures the IA output and the pink shows output
voltage of the entire circuit after the signal passes the
bandpass filter.
Figure 7: Multisim schematic of EEG circuit with three probes for output
voltage reading..
The Arduino Mega 2560 has analog pins to allow for
digital conversion of the raw signal that was filtered by way of
an ADC or an analog to digital converter feature the board
has. The Arduino will be connected via USB 2.0 A to B cable
to a computer for digital processing and eventual output.
Some details of the digital processing part of the
circuit logic that are beyond the scope of the class. However, Figure 8: Graphical voltage output behavior from probes
they are important to consider. The following explanation
outlines the rough idea behind what should be happening in It is clear that the voltage outputs oscillate due to the
digital processing. The filtered signal from both the EMG and sinusoidal input. What can also be observed that the output is
EEG circuits travels to the Arduino and then to the computer measured in mV and uV due to the fact that the beta waves are
for processing. It is reasonable to assume that within the code, relatively less than 50 μ V in amplitude. It is obvious that this
one must have a way to recognize both EMG and EEG filtered is not a perfect representation of the model, but it gives a
signals to check if the wave will show that upward infliction relative image of what the design was attempting to
of voltage, or P300 component. Once that threshold voltage is accomplish based on concepts from lecture.
achieved, the code will need to send a value that can equate to
a voltage value to be sent to the appropriate pinout for the Because the frequency band needs to be accounted for, the
servo motor that is attached to the prosthetic. following equations will allow for that expectation to be
integrated in the band-pass filter before signal heads to the
One crucial fact to point out in the circuit design is Arduino:
that some sort of active grounding needs to be established. R4
This is where the driven right-leg (DRL) circuit is integrated. f low = R5
Instead of the DRL directly being implemented, a version of C2
that concept is applied with a reference electrode acting as that f high = C1
active ground for the purposes of the EEG raw data collection.
That is where the A2 electrode is utilized. IV. CONCLUSION
D. Results For overcoming various issues that are encountered with
prosthetic devices and their control, the EMG/EEG
combination prosthetic has been proposed and analyzed. The
processed signals through the Arduino and servo-motor [8] Galdo-Alvarez, Santiago. (2015). Re: What are the commonly used
electrode positions in 10-20 systems to acquire EEG related to primary motor
confirm that the physical implementation of the design would cortex?Retrievedfrom:https://www.researchgate.net/post/What-are-the-comm
be valid. The findings of this project design represent a proof only-used-electrode-positions-in-10-20-systems-to-acquire-EEG-related-to-pri
of concept for this expansion of bioinstrumentation. The mary-motor-cortex/55261f84d3df3e750d8b45ff/citation/download.
advantages of this design are present in its noninvasive [9] Cacioppo, John T, et al. “Chapter Three. Event-Related Brain Potentials.”
properties and accessibility. The combined medical procedures Handbook of Psychophysiology, 2nd ed., Cambridge University Press, 2000,
would synthesize together to form precise and accurate signals pp. 53–84.
for prosthetic control. [10] “Lateralized Readiness Potential.” Wikipedia, Wikimedia Foundation, 11
Further research can expand with experimenting for Dec. 2020, en.wikipedia.org/wiki/Lateralized_readiness_potential.
signals with more reliable environment conditions and lab [11] B. C. Wheeler. “BENG 152 Biosystems Engineering Laboratory:
equipment. From there, continued development would move Laboratory 7 EMG.” UC San Diego Department of Bioengineering.
to prototyping and optimizing the circuit layout for reduced [12] G. Cauwenberghs. “Biopotential Amplifiers: The Electrocardiogram.”
noise and outside interference. Prototype designs can be UC San Diego Department of Bioengineering.
adjusted to maximize comfort for the user through trial [13] “The Instrumentation Amplifier” All About Circuits.
designs and tests. Progress into the concept of combining https://www.allaboutcircuits.com/textbook/semiconductors/chpt-8/the-instrum
entation-amplifier/
EMG and EEG signals could provide other opportunities to
increase prosthetic control and lead to more accurate [14] G. Cauwenberghs, “Biopotential Amplifiers: Instrumentation Amplifiers
and Driven-Right Leg”
prosthetics that could expand to account for other types of
limb loss. [15] “10-20 System (EEG)” - Wikipedia
https://en.wikipedia.org/wiki/10%E2%80%9320_system_(EEG) [Acessed 17
REFERENCES Mar-2021]
[16] D. Driffiths, Nelo, J. Peters, A. Robinson, J. Spaar, and Y. Vilnai, The
ModularEEG Design, 2002. [Online]. Available:
[1] “Amputation,” Stanford Health Care (SHC) - Stanford Medical Center, http://openeeg.sourceforge.net/doc/modeeg/modeeg_design.html. [Accessed:
11-Sep-2017.[Online].Available: 14-Mar-2021].
https://stanfordhealthcare.org/medical-conditions/bones-joints-and-muscles/a
mputation.html. [Accessed: 15-Mar-2021]. [17] C. M. and M. Jiang, “Brain-Computer Interface,” Brain-Computer
Interface Using Single-Channel Electroencephalography. [Online]. Available:
[2] M. Alvarado-González, E. Garduño, E. Bribiesca, O. Yáñez-Suárez, and https://people.ece.cornell.edu/land/courses/ece4760/FinalProjects/s2012/cwm
V. Medina-Bañuelos, “P300 Detection Based on EEG Shape Features,”
55/cwm55_mj294/. [Accessed: 16-Mar-2021].
Computational and Mathematical Methods in Medicine, vol. 2016, pp. 1–14,
Jan. 2016. [18] E. A. E. Alem, “Understanding EEG Part8: EEG Localization and
amplifiers, What is electroencephalography (EEG)?,” YouTube, 07-May-2012.
[3] S. Coulson, “wk1W_lec1_erp_slides,” in COGS 179 UCSD ,
[Online]. Available:
06-Aug-2020. https://www.youtube.com/watch?v=MlS0dLCXlME&t=461s. [Accessed:
[4] J. Hussain, K. Sundaraj, Indra Devi Subramaniam, C. Kiang Lam, 16-Mar-2021].
“Muscle fatigue in the three heads of triceps brachii during intensity and speed [19]A. P. Simons, “Reading and Wirelessly Sending EEG Signals Using
variations of triceps push-down exercise”, Front.Physiol., 21 Feb 2020, Arduinos and XBee Radios to Control a Robot,” thesis, UA - ECE,
[5] P. de Souza “lateral head of triceps brachii” Anatomy zone Fayetteville, 2014.
[6] Manpreet S. Tiwana, Margaret A. Sinkler; Bruno Bordoni.“Analogy, [20] “Example of Electrode Placement for Muscle Sensor v3 to Measure
Shoulder and Upper Limb, Triceps Muscle” Feb 7, 2021, Bicep Activity.” Pololu, www.pololu.com/picture/view/0J5305.
[7] M. Teplan, “Fundamental of EEG Measurement,” ResearchGate,
Jan-2002. [Online]. Available:
https://www.researchgate.net/publication/228599963_Fundamental_of_EEG_
Measurement. [Accessed: 18-Mar-2021].
