Wireless Electroencephalogram

Keith Jamison, Benjamin Madoff, Kevin N. Rohmann, Virginia M. Woods

BioNB 440

Fall 2005

Professor Bruce Land


Disclaimer: This device is not intended for medical use including but not limited to diagnosis and treatment of neurological disease. Use extreme caution whenever connected powered circuits to the human brain.  Failure to do so may result in injury, death, or worse.



*   Table of Contents

*    Introduction

*      Neurobiology of the Brain

*      Overview of Electroencephalography

*      Design Considerations

*    High Level Design

*    Hardware Design

*      Electrode/Amplifier

*      Transmitter/Receiver

*    Results

*      Electrode/Amplifier

*      Transmitter/Receiver

*      Combined System

*      Future Considerations

*      Appendix

*      References and Datasheets

*      Additional Schematics


*   Introduction


1.  The Neurobiology of the Brain


1.A. Electrical Properties of the Neuron


The human body is able to successfully function as a single system through the integrated communication of electrical pulses known as action potentials.  The origin of action potentials can be traced to the electrical activity at the cellular level.  Neurons are cells highly specialized for their electrical function and use their membranes to maintain electrochemical gradients of sodium, potassium, calcium, and chloride ions.  The resting concentrations of each ion as well as their associated electrical potential as quantified by the Nernst equation are listed in Table 7.1.  Generally speaking, there is a higher concentration of potassium on the cell’s interior and a higher concentration of sodium in the extracellular matrix.


Table 7.1  Ion Concentrations for squid giant axon (Hodgkin, 1964; Baker, Hodgkin, et al., 1971), for frog muscle fiber (Conway, 1957), and for human erythrocytes (Davson, 1970).  The salt concentrations in both intracellular and extracellular solutions are much higher in marine animals than in terrestrial animals.


Concentration (mmol/L)





















































Table 7.1 modified from BEE 454 Phyiological Engineering Lecture Manual (Aneshansley, Fall 2005)


The electrochemical concentration gradients across a semipermeable membrane result in a local equilibrium potential called a biopotential.  This electrical property can be quantified according the Goldman Equation:



PK = permeability to K+

[K+]out = K+ concentration outside cell

PNa = permeability to Na+

[Na+]out = Na+ concentration outside cell

PCl = permeability to Cl-

[Cl-]out = Cl- concentration outside cell

R= Universal gas constant (8.31 joules/mole/deg. K)

T= temperature in degrees Kelvin (273+deg. C)

F= Faraday constant


Introduction of a transient depolarizing potential, such as an excitatory synaptic potential, will result in the generation and subsequent propagation of an action potential due to ionic channel kinetics.  This process was quantified by Hodgkins and Huxley in 1952.  The plots in Figure 1 are based on their equations and offer a visualization of the relative temporal permeabilities.


1.)  Arrival of depolarizing stimulus changes the permeability of both sodium and potassium channels.  The sodium conductance begins to increase, flowing down their electrochemical gradient into the axon.  This transfer of charge causes further depolarization as sodium drives the membrane potential closer to its Nernst potential.

2.)  This positive feedback loop continues until the two slower consequences of depolarization, sodium inactivation and potassium conductance, begin to take effect.

3.)  The efflux of potassium ions as they flow down their chemical gradient drives the membrane potential towards the Nernst potential of potassium

4.) This hyperpolarization continues until the voltage is slightly below the resting level, known as the refractory period.  Normal conditions are restored by potassium inactivation.


Figure 1. Electrical Changes at the Cellular Level with an Initial Applied Current of 20mV.  The equations used to generate these plots are from Hodgkin and Huxley, 1952.


5.)  The process of generating an action potential causes a local excitability of the membrane, which provides the excitatory stimulus for the next region.  Thus, the steps are repeated and the action potential is propagated down the axon.

6.)  Once the action potential reaches the end of the axon, it signals the release of a neurotransmitter, a chemical messenger, to continue the electrical transmission to a neighboring neuron.



*         The electrochemical gradients of a neuron induce changes in ionic channels for the generation of an action potential.

*         The action potential is the most fundamental unit of information transfer within the nervous system. 

*         The process of action potential propagation within an axon excites neighboring neurons.

*         The collective electrical activity of a specific region in the brain is known as the local field potential. 


1.B. The Electrophysiology of the Cerebral Cortex


The rational authority of EEGs as a diagnostic tool in clinical applications is due to the morphology and electrical properties of the cerebral cortex.  There are two types of cortical neurons:  pyramidal cells and interneurons.  Pyramidal cells are oriented orthogonal to the cortical surface and interneurons exist tangent to the surface.  The neurons in Figure 2 show how a single axon connects to the soma of a pyramidal cell and diffuses outward in a complex dendritic arbor.


Figure 2.  The Structure of Cortical Neurons. (Figure adapted from Ehlers 2005)


There is an extremely high level of interconnection within the cortex with an approximated 104 synapses per neuron.  The resulting intracortical connections have a relative lack of specificity with some being excitatory while others are inhibitory.  As described in Neocortical Dynamics and Human EEG Rhythms, “the high degree of integration in neural tissue motivates some neuroscientists…to view brain operation at hierarchical levels above the single neuron – the neural mass” (Nunez, 102).  Thus, because of the large density of synaptic interactions within a neural mass, macroscopic properties emerge that are relatively independent of its specific inner circuitry (ibid).  Therefore, this abstraction means that the oscillatory behavior of individual neurons due to the cyclic changes of ionic conductance that are described in Part 1A cannot be directly extracted from the rhythmic patterns of brain activity.


In our understanding of how the morphology of the cortex relates to scalp EEG measurements, Nunez states “it is often convenient to idealize this complicated current source distribution [i.e., the cortical neural mass] by a small dipole layer.  At some fixed instant in time, a slight imbalance in sources over sinks…causes a net outward macroscopic current density…and macroscopic potential difference” (Nunez, 29).  In Figure 2, the dipole he describes would be the difference in electric potentials between input synapses of the axon and the synapses of the dendritic arbor.  It is this polarization that creates the surface potentials measured by the scalp electrodes (Figure 3).  However, because the scalp electrode is separated from the neural tissue by both distance and a poorly conductive skull, the spatial resolution of the precise neural mass is typically rather poor. 


Figure 3.  Electrophysiological Origins of the Scalp Potential.  Note: The neurons pictured are one of about 104 or 105 neurons in the volume of a neural mass.  (Figure adapted from Nunez 1995)


*         The dendritic arborization of cortical neurons creates highly connected regions known as neural masses.

*         The macroscopic polarization of the neural masses provides an outward current density and a potential difference.

*         Scalp electrodes are able to measure this value with poor spatial resolution.


1.C.  Computational Models of the Cerebral Cortex


The behavior of the cerebral cortex can be quantified as a dynamic nonlinear system (Nunez, 263).  The biological evidence supporting this model stem from the following characteristics (ibid):

*         Nonlinear voltage dependency

*         Feedforward and feedback connections with distributed time delays

*         Modulatory inputs can affect different sites in different ways


The work of van Rotterdam et al formally articulated the behavior of a chain of neurons by a nonlinear partial differential equation, which is linearized to obtain an analytical solution.  Its solution yields a relationship between frequency (ω) and wave number (k).  This following analysis is adapted for Nunez, 263-264:



Where a-1 and b-1 are excitatory and inhibitory postsynaptic potential rise times, respectively (a-1 ≈ 10 msec and b-1 ≈ 20 msec), c-1 and d-1 are characteristic distances at which excitatory and inhibitory intercortical cell connections were assumed to decrease, respectively, (in the range of 30 to 500cm-1), Q is a nondimensional control parameter characterizing the strength of coupling between excitatory and inhibitory neurons, and k generally has a value less than 0.5 because of the poor electroconductivity of the skull (here, k=0). 


For Q less than but of the order of the threshold value:


Solutions to Equation 1 yield to pairs of complex roots ωi = ωRi – jγi, of which one pair is “overdamped” and one pair is weakly damped as Q → Qth.  For Q > Qth, the latter pair of frequencies become unstable.  The frequencies of the pair of modes for Q near Qth are ωR = ±.


Assuming parameters matching those of the alpha brain wave frequency, |ωR| = 71 rad/sec, so ƒ ≈ 11Hz.


Unfortunately, this linearization represents a large simplification.  A better fit of the neuronal network behavior can be achieved by applying a set of coupled nonlinear differential equations both in space and time (Nunez, 264).  The emerging behavior from this analysis would be chaotic.  Our understanding is that its steady, attractor states, which would depend on the parameter values and initial conditions, would be the predicted value for the frequency of the brain’s activity.  A more detailed survey of the published literature is warranted to be certain of this claim.  A current initiative in biological research aims at determining whether experimentally recorded brain activity is actually chaotic, particularly in epileptic patients (Nunez, 265).



*         Equation 1 quantifies the nonlinear behavior of the cortex based on its excitatory and inhibitory connections.

*         Setting the parameters to the appropriate values correctly predicts a frequency in the alpha range.

*         A more accurate computational model of coupled nonlinear differential equations would describe the chaotic behavior of the cortex.

*         A future direction in biological research aims to confirm the chaotic behavior of the brain.


2.  Overview of Electroencephalography


2.A.  What is an EEG?


Electrocephalograms (EEGs) are recordings of the electrical potentials in the brain.  The recording electrodes are positioned on the scalp and their integrated output displays the rhythmic activity of the brain.  Analysis of EEGs is often used in the clinical setting as a diagnostic tool to detect pathologies associated with aberrant electrical behavior or stimulus-directed behavior.


Table 2.  Different Types of Normal EEG Waves (Sampl waveforms adapted from Webster, 1997)


Amplitude (µV)

Frequency (Hz)

Associated Behavior

Representative Samples



8 – 13

Awake, resting state



14 – 30, can be as high as 50

Beta I –Attention

Beta II – Intense, mental activity



4 – 7

Emotional stress



< 3.5

Deep sleep, serious disease


2.B.  The Clinical EEG


The system most often used to place electrodes for monitoring is the International Federation 10-20 System as illustrated in Figure 4.


Figure 4.  The 10-20 Electrode System. (Webster, 1997)


The electrodes are fixed to the surface of the scalp with an adhesive or glue.  In between the metal electrode and the skin is a conductive gel.  Typically, the electrodes are made of ionized silver or gold.


2.C.  Application of EEGs


EEG are generally used in medical diagnostics to detect electrical abnormalities in the brain due to disease (e.g., epilepsy, parkinson’s, tremors)


The emerging field neurprosthetics utilizes the information extracted from EEG recordings to control artificial limbs.  A popular area of research involves selecting neural signals from EEGs for the control of arm position and velocity parameters in macaque monkeys.  In the near future, we will most likely see this research extend to control systems for artificial arms in humans. 


3.  Design Considerations


3.A.  Amplification of Biopotentials


The distinguishing feature of an EEG amplifier is that it must amplify very small signals of the brain.  Thus, the best instrumentation amplifier has a very high input impedance, high CMRR, high differential gain, and very low Johnson noise. 


3.B.  Circuit Enhancements for Noise Reduction


In any recording setting there is electrical interference from the environment.  Peripheral power sources can magnetically induce electrical current into the lead wires.  If ground is used as the circuit reference, then there will be a common mode interference potential.  Most of this voltage will be eliminated by a high CMRR, but this situation can be improved by using a “driven right leg circuit”, a convention in which the body is used as the circuit’s reference. 


Filtering is another method to passively reduce the effects of noise on EEG measurements.

*         To remove RF interference, place a small capacitor between each electrode lead and ground.

*         Very high frequency electromagnetic interference can be removed with small inductors or ferrite beads in the lead wires.

*         DC offsets can be corrected by the addition of an adjustable DC biasing circuit.

*         60 Hz noise can be removed either by using a notch, a Chebychev low-pass, or a series of Sallen-Key low-pass filters.


Figure 5. This graph shows the frequency and amplitude distribution of several types of biopotential signals.  Each has its own frequency characteristic, but 60Hz noise reduction is always a problem.

Figure 5. This graph shows the frequency and amplitude distribution of several types of biopotential signals.  Each has its own frequency characteristic, but 60Hz noise reduction is always a problem. 


*   High Level Design


Figure 6. High Level block diagram. Click individual components for detailed schematics.


The goal of this project was to detect EEG signals and transmit them by radio frequency.  The first stage of the setup was to amplify the voltages incident on the electrodes.  The voltages would then be converted to a frequency, transmitted, converted back to a voltage, and measured with an oscilloscope.  The signals can be easily transmitted by RF due to their extremely low frequency compared to the transmitter.  The frequency difference makes the biopotential signals appear to be changes in DC input, so any voltage controlled oscillator could be used for the voltage to frequency converter. 


*      Breaking the project up into two halves was necessary in order to complete the assignment in the time allotted and assure that four bodies were not crowded around a single protoboard.


*         Part 1: Acquiring/Amplifying/Filtering the EEG signal (Ben & Virginia)


Ben and Virginia focused on the goal of acquiring the EEG signal. This involved acquiring an EEG signal from electrodes placed on the skin, amplifying the signal, and filtering it to reduce noise. To detect an EEG signal, appropriate amplification was needed.  If the average signal can be assumed to be 100μV, or 10-4V, an amplification of 10,000 or 104 is appropriate.  Additionally, 3 Sallen-Key low pass filters with a cutoff frequency of 51Hz were used to filter 60Hz noise.  These filters were used separately or in series to attenuate as much noise as possible.  In order to test this half of the project, a linear optocoupler was used so a person could be wired to electrodes without the risk associated with being directly connected to wall power or an oscilloscope.


Although the goal of this project was to detect EEG signals, we were unable to do so.  The results shown below are EMG results measured from the bicep.  They are there for proof of concept, but we were unable to distinguish brainwave activity from background noise during this project.


*         Part 2: Transmitting/Receiving a voltage as an FM signal (Keith & Kevin)


Keith and Kevin took on the task of designing a system to transmit the EEG signal from Ben and Virginia’s circuit to an oscilloscope.  The sine wave generator on a protoboard was used to simulate EEG impulses at various frequencies within the EEG band during testing.  The signal was first converted to a frequency by a voltage to frequency converter (VFC) in order to reduce noise during transmission. Once the signal was in frequency form it was transmitted through a single chip FM transmitter to an FM radio receiver (an old boom box). The output of the radio was then sent through a frequency to voltage converter (FVC) to restore the EEG signal to voltage form. This final output was then sent to an oscilloscope to display the EEG signal.


*   Hardware Design

*      Electrode/Amplifier Design


This schematic shows the instrumentation amplifier in its test configuration.  The instrumentation amplifier receives input from two electrodes, with a third providing a ground reference.  The ground reference is set at Vdd/2, providing a virtual ground for the circuit.  The first op amp on the left adjusts the virtual ground.  Since the gain for the operational amplifier is set at 10,000, slight differences in the DC input voltage can cause the amplifier output to go to the positive or negative rail. 


The instrumentation amplifier is able to cancel out most of the noise from the electrode leads, since approximately the same noise will be present on each electrode lead.  The largest noise present on the leads is the 60Hz noise from the AC in the building.  This instrumentation amplifier, the INA118 was chosen because of its low supply voltage, high common mode rejection ration (CMRR), and high gain (maximum 10,000).  The gain for the instrumentation amplifier was set with an external resistor according to the equation Gain = 1 + 50kΩ/R. 


The IL300 is a linear optocoupler.  This provides isolation between the user and the 120V AC line.  The optocoupler is able to maintain linearity using two op amps, one on either side.  Since there are two photodiodes activated by the LED (indicated by the arrows in the IL300), the input to the LED can be adjusted using the negative feedback of the amplifier.  If the resistor between the output of the photodiodes and ground is the same for both outputs, then the current flowing through them must also be equal.  The second op amp buffers the output before being read by the oscilloscope.  In setting up the linear optocoupler, it is important to keep the input into the LED around 20mA by setting the input resistor.


*      Transmitter/Receiver Design

*      Voltage-to-Frequency Converter




The schematic on the left is a voltage offset circuit that was found to be necessary during our initial testing to keep the input voltage from going below 1V.  This kept the output frequency of the VFC above approximately 1KHz so it could be filtered with a simple low-pass RC on the receiver side (See notes for FVC).  The VFC schematic on the right is adapted from the sample schematic suggested in the LM331 datasheet, with a few changes.  Most were minor changes in component values to accommodate what could be found in the lab.  One change, however, was more significant.  The input to the suggested circuit had a built-in RC low-pass filter with a cutoff frequency of 16Hz – well within the range we had hoped to transmit.  We therefore decreased the capacitor by a factor of 10, thus increasing this cutoff beyond the range we needed.


*      FM Transmitter



The FM transmitter schematic was based on the circuit suggested on the MAX2606 website (http://www.maxim-ic.com/appnotes.cfm/appnote_number/1869).  Since we were only transmitting one channel, the stereo inputs were eliminated.  Additionally, the 10K potentiometer for input level control was changed to 100K to allow for a greater range of maximum voltages to be examined, when initially determining this parameter (The transmitter works best with a rail-to-rail input signal around 30-40mV).


*      Frequency-to-Voltage Converter




The FVC circuit on the left was modified from the suggested FVC circuit provided in the LM331 datasheet.  Initial experiments showed that the output from the FVC was actually the desired low-frequency (or DC) signal combined with the low-amplitude, high-frequency input signal to the FVC (See FVC output in results).  The isolated low pass filter shown on the right was added to the output in order to get eliminate that high frequency signal.  Our VFC circuit was such that fout changed by 1KHz for every 1V at the input.  Thus, if the input signal was 50mV, the VFC output would be 50Hz.  It would therefore be impossible to filter the VFC noise from even a very slow input signal, so the voltage offset was added that keeps the VFC output above 1KHz.  Since we are only looking to detect signals up to 100Hz, this separation is more than adequate, and the isolated low-pass filter shown on the right with cutoff 160Hz cleans the signal almost completely (See signal output in results).


*      For schematics with the amplifier and transmitter combined, see additional schematics at the end of the page.


*   Results

*      Electrode/Amplifier

      *      Breadboard assembly of the biopotential amplifier


*      EMG signal recorded by the biopotential amplifier and FFT analysis of the signal


The results in this section are for EMG signals measured from the bicep.  The burst of spikes represents the contraction of the muscle.  Each contraction will be picked up on the oscilloscope at an amplitude consistent with the strength of the contraction.  This amplitude, though, will also depend on the sensitivity of the electrodes and the gain of the amplifier.


*      Fourier analysis of amplifier signal

This picture shows the fast Fourier transform (FFT) output from the oscilloscope used to search for frequency spikes that would indicate an EEG signal.  Unfortunately, neither the direct output of the amplifier nor the FFT revealed any patterns consistent with EEG signals.  The most common signals were the 60Hz noise signals and low frequency (1-5Hz) signals that were inconsistently detected and inconsistent with the mental state of the person being recorded.


*      Transmitter/Receiver

*      Voltage-to-Frequency Converter with Voltage Offset


*      FM Transmitter


*      Frequency-to-Voltage Converter and Low-Pass Filter


*      Output from voltage offset

Input voltage is offset by 2 volts (channel 2 -- bottom).



*      Output from VFC


The input sine wave from the protoboard (channel 1) was converted to a square wave with frequency modulated by the FVC (channel 2).


*      Output from FM receiver

Signal similar to VFC output.  Notice the aliasing caused by increased frequency during the input signal’s upswing.


*      Output from FVC


The high frequency noise seen in the FVC output (channel 2) is caused by the addition of the FVC input to the signal.


*      Output from FVC after Low-Pass Filtering

When compared to the channel 2 signal on the left-hand figure above, you can see the low-pass filter was successful in removing most of the high frequency noise generated by the FVC (see channel 2 below).


*      Although our initial prototype of the FM transmitter did not require an antenna, we found we needed antennas both on the transmitter and on the radio for the final soldered circuit to work properly. Without the antennas the signal-to-noise ratio was too low. It is possible that the protoboard served as an antenna for the initial prototype of the FM transmitter.

*      With an antenna added to the FM transmitter, it works at a wider radio receiver amplitude range and also works without distortion up to ~80Hz (compared to ~35Hz without the antenna).

*      To configure the radio receiver, in our case a boom box stereo, we set the gain on the two lower EQ bands to 0 and set the high frequency EQ all the way up to reduce low frequency noise relative to the sound transmitted.


*      Combined System

      *      Electrode/Amplifier and Transmitter/Receiver systems hooked together


*      EMG output of the combined system


These two pictures show the output of the biopotential amplifier on channel 1 (top waveform) and the output of the transmitter on channel 2 (bottom).  The waveforms on the left are for several short contractions, the waveform on the right is for a longer continuous contraction. 


This picture shows a zoomed in EMG signal with RF transmitter output to show the correlation between the signal at the output of the amplifier and the signal that is transmitted. 


*   Future Considerations: What we would do differently next time

*      More experimentation with different types of electrodes would have helped us to better use the silver disc electrodes. 

*      We tried several different types of feedback, all with mixed results.  Next time, a working setup of all types of feedback should be breadboarded and tested together, so the results from each feedback method can be compared.

*      As with the feedback methods, it would be convenient to have several different types of low pass and perhaps notch filters wired in the frequency near the range of each biopotential signal.  There are some built in filtering functions in many oscilloscopes also; it would have benefited us to learn more about them during this project. 

*   We would also make sure when soldering two identical surface mount chips to a board that we solder them on in the same orientation. It makes translating and copying circuits and diagrams much easier.


*      Appendix

*      References

1.      Aidley, D.J. 1971.  The Physiology of Excitable Cells. London, UK, Cambridge University Press

2.      Bronzino, J.D.. 1995.  “Principles of Electroencephalography” IN:  Bioelectric Phenomena.  Boca Raton, FL: CRC Press.

3.    Carter, Bruce: A Single-Supply Op-Amp Circuit Collection (http://instruct1.cit.cornell.edu/courses/bionb440/datasheets/SingleSupply.pdf)

4.      Ehlers, M.. Ehlers Lab.  November 22, 2005.  Duke University School of Medicine.  December 7, 2005.  http://www.ehlerslab.org/research/neuronal_polarity/1.html.

5.      Hodgkins, A. L., and A. F. Huxley. 1952. A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. London. 117:500-544.

6.      Kandel, E.R., Schwartz, J.H., Jessell, T.M., 1991. Principles of Neural Science. New York, NY, Elsevier Science Publishing Co., Inc.

7.      Medical Instrumentation: Application and Design, 3rd edition, John G. Webster (ed.), John Wiley & Sons, New York, 1998.

8.      Nunez, P.L. (1995)  Neocortical Dynamics and Human EEG Rhythms.  New York:  Oxford University Press.

9.      Thakor, N.V.. 1999.  “Biopotentials and Electrophysiology Measurement” IN: The Measurement, Instrumentation, and Sensors Handbook. Boca Raton, FL: CRC Press. 

10.   van Rotterdam, A., Lopes da Silva, F.H., van den Ende, J., Viergever, M.A., Hermans, A.J.. (1982)  “A model of the spatial-temporal characteristics of the alpha rhythm”.  Bull. Math. Biol.  44:283-305.

11. “Single-Chip FM Transmitter Extends Home-Entertainment Systems” http://www.maxim-ic.com/appnotes.cfm/appnote_number/1869

12. Webster, John (1998). Measurement, Instrumentation, and Sensors Handbook

13. Webster, John (1997). Medical Instrumentation, Application and Design, 3rd edition

14. Woods, V.M., Serna, C. 2005. “Neural Conduction Laboratory”.  BEE 454 Physiological Engineering. Unpublished.


*      Datasheets

*      IL300 Linear Optocoupler: http://www.vishay.com/docs/83622/83622.pdf

*      INA118 Instrumentation Amplifier: http://focus.ti.com/lit/ds/symlink/ina118.pdf

*      LM331 Precision Voltage-to-Frequency Converter: http://www.national.com/ds.cgi/LM/LM231.pdf

*      LM358 Operational Amplifier: http://cache.national.com/ds/LM/LM158.pdf

*      MAX2606 FM Transmitter: http://pdfserv.maxim-ic.com/en/ds/MAX2605-MAX2609.pdf


*      Additional Schematics


This schematic replaces the linear optocoupler with an RF transmitter.  When the transmitter was used, though, a Sallen-Key low pass filter was inserted between the instrumentation amplifier and the voltage to frequency converter to reduce some of the higher frequency noise most likely caused by the transmitter.  The wires used for electrode leads act as antennas for the RF transmitter.  Fortunately, the frequency content of the RF signal is 103 – 107 times greater than any type of biopotential signal, so a low pass filter is all that is necessary to clean up any high frequency RF noise.


Voltage offset and transmitter schematic


Receiver schematic


Complete circuit.