Mobile EEG Device
Research Project Fall22-Spring23
This was a funded research project, therefore no schematics will be shown here. Click any picture/image for the source/credit.
*Note:
In the lab testing the EEG device
Late in the summer of 2022 I was contacted for a hardware/software engineering research position. This research position turned out to be one of the greatest academic experiences that I had at UNC Charlotte. Through this experience I was able to work with Dr. Jeremy Holleman, to both learn and apply advanced concepts in Analog Electronics, Digital Signal Processing, PCB design and more.
What was initially supposed to be a simple integration job turned into an entire research project. Since the main components used for the previous prototype were unavailable an entirely new design with new components needed to be designed. An additional challenge was that component selection was significantly limited and proved to be the largest initial challenge.
Initially my role for this position was to bring to fruition 1.5 years of research done by a former PHD and Masters student. They both graduated leaving the project unfinished in a rough prototype state. The project was currently using a mix of evaluation boards and required someone to develop a custom main signal acquisition board as well as new electrode boards. Unfortunately over the course of the projects lifetime the components used for prototype development became unavailable and there was no indication of them being viable any time soon(Covid chip shortages).
Mobile EEG
With the backstory complete lets dive into the project. Simply put the final product was needed to be a portable electroencephalogram (EEG) device that could transmit information to the cloud. An EEG test measures electrical activity in the brain using electrodes. A use case for such a device would be to alert a user of an early detection of seizures, or to send the data to your doctor for monitoring. The final device was expected to have 32 channels (individual electrodes), and expected to run off a battery or potentially cell phone. There were multiple methods of communication, originally the device design could stream information to a computer on the same WIFI, however for redundancy Bluetooth, as well as streaming data via a physical connection were made part of the project goals. An EEG device consists of the electrodes that obtain the signal and the main device that collects those signals and makes them into usable information (I call this the acquisition board).
Typical EEG Setup
Before the acquisition board can be used the first step is to make sure that we can get a clean usable signal, therefore my research started with what is an active electrode, and why do we want it? I came across a fantastic research paper written by Porir Marwan Abed available here, which gave great insight into the world of active electrodes. An electrode is what makes contact with a persons skin in order to measure a voltage. Electrode Tissue Impedance (ETI) aptly named is the impedance between the electrode and the surface tissue, with higher ETI results in signal deterioration. A dry probe has an impedance of 100kΩ-10MΩ, however if we use a wet probe the ETI could be lower than 5kΩ. Even with the ETI values in mind we used dry probes to increase the usability of the platform. Dry probes can more easily be used on hair and require less setup/tear down when using the device.
Active Electrode
DryElectrode
Ok so we are using a dry probe, but we still haven’t discussed what an active electrode is and why we want to use one vs a conventional electrode. When the electrode is measuring the voltage at the skin it is on the order of 100µV or 1/1000 of a volt, with a frequency of anywhere between 1-80hz. An active electrode uses an amplifier to you guessed it amplify the signal/voltage before sending to the acquisition board. Ideally this very small signal would be sent from the probe through the transmission line and to the acquisition board with no problems. Unfortunately along with the desired signal there are other signals such as powerline interference (60 Hz), interference at the electrode-tissue interface, undesired signals from other sources, as well as noise. In a standard electrode the signal could be completely drown out by these undesired signals before ever reaching the acquisition board.
Illustrating an Active Electrode Location
Amplifier
Brainwave Frequency Chart
Armed with the theory now was time to build an active electrode. An amplifier was selected and its gain was set to 100 (for more information on op-amps click here). The PCB was designed using KiCad 6.0, with an emphasis on designing the smallest electrode possible. There was a large exposed pad on the underside of the PCB where we used electrically conductive paste to adhere the dry electrode to the PCB. An audio cable was used as the transmission line to supply a reference voltage, V+, V-, and the return signal.
Active Electrode Layout
Active Electrode 3D Render
Assembled Active Electrode Size vs Penny
For initial testing a standard function generator worked well to prove that the amplifier was indeed operating as expected. Unfortunately for our case the function generator that we had only went down to 200 mVpp which is well above the intended signal strength. You may be asking… why not just test the probe on yourself? Testing on a person would be a bit premature at this point, and having a known signal to compare the final amplified signal is required. The simple way to get a very small signal from a large signal is of course… a voltage divider.
Active Electrode Test
Active Electrode Test Video
If you watched the test video above you may have noticed the housing around the active electrode. The housing was required for both the electrode and the users protection. The PCB can be damaged from ESD, hair getting wrapped around components or even just wear and tear of handing an exposed PCB. Keeping in mind this was the initial prototype the user should not be touching the FR4 fiberglass body of the PCB, the conductive epoxy holding the electrode to the PCB or the solder used to assemble the PCB. In addition to safety concerns the 3D printed housing simply made it easier to physically handle the active electrode, since the final assembled size was smaller than a penny as shown previously in this article. The housing was made of two parts that clasp shut, sandwiched around the electrode. The body of the PCB and the audio jack were used as contact points for the housing. The flat top to the housing was meant to later serve as a mounting point for the head unit that would house all of the electrodes, and the hole in the bottom of the housing allows for the contact electrode to be connected to the PCB.
Active Electrode Housing Top View
Active Electrode Housing Front View
Active Electrode Housing 3/4 View
Active Electrode Housing 3/4 View
Active Electrode Housing Front View Split
Active Electrode Housing 3/4 View Split
Active Electrode Housing Interactive Viewer
The test results shown earlier showed the active electrode working via an oscilloscope, the next step was to read the electrode signal using the acquisition board. An evaluation board for our chosen ADC(will remain unspecified) was used to verify operation before moving forward. The chosen ADC was capable of handling eight channels/electrodes at once. The initial results from the ADC shown below demonstrated shows sub Nyquist artefacts specifically a sampling moiré effect. The Nyquist Criterion says that to accurately sample a signal you need to sample it at at least twice the frequency of the signal. In my initial reading the ADCs sample rate was too slow and therefore under-sampled. The artifact/sampling moiré effect results from the interaction of the periodic signal and the periodic sampling process[5]. If you are unaware of the moiré effect I encourage you to google it or click here, once you see it you’ll recognize it in many places.
Initial ADC sampled data
Comparison between (a) a case with no sampling artefacts, (b) a case with a sub-Nyquist artefact, and (c) a case with a true sampling moiré effect.
After increasing the sampling rate to an appropriate level the ADC’s output was graphed and the output strongly resembled the input sine wave therefore the ADCs functionality had been verified.
Once the ADC and the electrode were both verified to be capable of their tasks the next step was to build the acquisition board. The EDA software KiCad was used to develop the acquisition board. This was a four layer board, and my first design that used BGA parts. This board was designed as an initial prototype with all microcontroller pins broken out, ample space between components, broken out signal diagnostic points, and broken out power and ground pins.
ADC Test Signal
ADC Attenuated Test Signal
In the above figure we are running into Quantization Noise which is negatively impacting the accuracy of our digitized signal. The figure ADC Attenuated Test Signal does not take advantage of the ADCs 16-bit resolution. Below are two figures showing an example of quantization between a 2-bit resolution vs 3-bit. From these two images we can infer that the higher the ADC resolution is the more accurate the digitized signal will be.
2-bit resolution with four levels of quantization compared to analog
3-bit resolution with eight levels of quantization compared to analog
With that being said if you have an inappropriate reference voltage then you are again not taking advantage of your ADC’s resolution. Below is the equation for the digitized value of a converted analog signal where:
ADC Conversion Formula
- N = The number of bits of the ADC
- n = digitized value
- Vin = Input voltage
- V+ref = Max voltage
- V-ref = Minimum voltage
If we assume that the analog signal above is 2Vpp and our reference range for our 3-bit ADC is +1V to -1V then 111 would refer to +1V, and 000 would refer to -1V. Now if our signal is only 1Vpp with the same reference range as before then we would would not be using our ADC resolution to the fullest and would have a similar effect as if we were using a lower resolution ADC. Therefore in order to take full advantage of our ADC’s resolution we need to set the positive and negative reference voltages to the extremes of our input voltage/signal or increase our input signal until its max/min approaches our reference voltages max/min.
The ADC communicates via SPI and since there were four of them they were set up in a multi-drop configuration. The multi-drop configuration was used rather than daisy-chain to improve speed. When using a daisy-chain configuration the data from ADC1 must travel through ADC2, then ADC3, then ADC4, verses using the multidrop configuration where we select the ADC that we want the data from and then request that data. Additionally since these configurations are set in hardware the multidrop allows us to further speed up data acquisition since we can use less than all four ADC’s and go with 16 channels for example.
Multidrop SPI Bus
Daisy-chained SPI
SPI communication between the ADC and microcontroller were initially tested and verified using an oscilloscope to monitor each signal.
Initial SPI Communication Testing
Once basic ADC functionality was verified the next step was to test the ADC by measuring a smaller 100µVpp signal(This would represent an input with the active electrode configured to a gain of 1). Shown below is the digitized 100µVpp signal, with a max amplitude of ~0.3552 and minimum amplitude of ~0.3542 we have a peak to peak of 0.001Vpp or 100µVpp with corresponds to our input signal after being run through the previously mentioned attenuator.
Acquisition Board Layout
Acquisition Board 3D Render
Unfortunately I was unable to finish the Mobile EEG prototype since I graduated shortly after designing the acquisition board so that’s where my input on this project ends. As previously stated, this was an amazing learning/employment opportunity that I am grateful to have experienced. I was able to exercise and improve many of my skill such as PCB design specifically pertaining to analog circuitry, CAD design, and even some power electronic design with LIPO battery charging, and board power distribution.