ADS1299 electrode impedance measurement algorithm
[Editorial note added in 2015: the forum posts labeled below as "K" and "deborahgalloway" are actually the same engineer, and NOT the person Deborah Galloway. This engineer asked us to delete his forum account, so the forum software somewhat arbitrarily reassigned it to Deborah. The label "K" can be associated whimsically as the character in the Men in Black movies who forgets his own identity.]
On higher end commercial EEG amps, you can get built-in electrode impedance measurements for each sensor. This is helpful, especially when you have a lots of sensors on a cap. Because it's likely that on your first pass you may not apply enough gel or paste to get the best possible connection. (Loosely defined as impedance somewhere in the range of 5 to 20K ohms, the lower the better.)
The ADS1299 supports this with what is called "lead-off detection". a guy in Serbia has actually developed code to do this, which is explained in his post on the TI Forum. Below is a quote from that thread. He is willing to help us out in recreating this. I am in touch with him via email and gave him links to the OpenBCI site and Kickstarter. His original code was for a commercial customer, so cannot be posted.
Most likely this would be implemented not continuously, but for a brief period of time during cap setup. Continuous monitoring can be supported, but then the final app consuming the EEG data stream must filter out the injected signal after the impedance has been calculated.
----
Hi Graham,
I've succeeded to online measure electrode impedance by using ADS1299 internal current source. You can measure differential impedance by turning on current sources on both inputs of one or more channels, or you can separately measure impedance of electrode on positive input, or on negative input. You must use AC current source, with amplitude of 6nA (because it is small enough and it will not disturb EEG signal) with frequency of Fdr/4. For example is sampling rate is 500Hz, then current source frequency will be 125Hz. When you collect EEG data, you must convert AD data to voltage, and do FFT. In amplitude spectrum of the FFT you will see one component of frequency of 125Hz. When you divide amplitude of that component with amplitude of the current source, it will give you electrode impedance. To try this, you can attach resistors on ADS1299's inputs. I've tried it with fixed resistors of 1Kohm, 10Kohm and 33Kohm. The results is 1.1Kohm, 11Kohm and 30kOhm respectively. The difference of the real impedance and measured impedance is because of the error of the current source amplitude. If you calculate back amplitude of the current source with measured impedance and amplitude in spectrum on frequency of 125Hz, the result will be about 7.5nA. It is small error, and for EEG measurements this error is negligible.
If you need any help with this, please contact me, and I will help you.
Best regards,
K
On higher end commercial EEG amps, you can get built-in electrode impedance measurements for each sensor. This is helpful, especially when you have a lots of sensors on a cap. Because it's likely that on your first pass you may not apply enough gel or paste to get the best possible connection. (Loosely defined as impedance somewhere in the range of 5 to 20K ohms, the lower the better.)
The ADS1299 supports this with what is called "lead-off detection". a guy in Serbia has actually developed code to do this, which is explained in his post on the TI Forum. Below is a quote from that thread. He is willing to help us out in recreating this. I am in touch with him via email and gave him links to the OpenBCI site and Kickstarter. His original code was for a commercial customer, so cannot be posted.
Most likely this would be implemented not continuously, but for a brief period of time during cap setup. Continuous monitoring can be supported, but then the final app consuming the EEG data stream must filter out the injected signal after the impedance has been calculated.
----
Hi Graham,
I've succeeded to online measure electrode impedance by using ADS1299 internal current source. You can measure differential impedance by turning on current sources on both inputs of one or more channels, or you can separately measure impedance of electrode on positive input, or on negative input. You must use AC current source, with amplitude of 6nA (because it is small enough and it will not disturb EEG signal) with frequency of Fdr/4. For example is sampling rate is 500Hz, then current source frequency will be 125Hz. When you collect EEG data, you must convert AD data to voltage, and do FFT. In amplitude spectrum of the FFT you will see one component of frequency of 125Hz. When you divide amplitude of that component with amplitude of the current source, it will give you electrode impedance. To try this, you can attach resistors on ADS1299's inputs. I've tried it with fixed resistors of 1Kohm, 10Kohm and 33Kohm. The results is 1.1Kohm, 11Kohm and 30kOhm respectively. The difference of the real impedance and measured impedance is because of the error of the current source amplitude. If you calculate back amplitude of the current source with measured impedance and amplitude in spectrum on frequency of 125Hz, the result will be about 7.5nA. It is small error, and for EEG measurements this error is negligible.
If you need any help with this, please contact me, and I will help you.
Best regards,
K
This discussion has been closed.
Comments
Wondering if you have looked into the DC method? If so, why did you decide to go for the AC method?
In the BCI2000 video I just posted on the other thread, they just check the impedances at the START of the session, by putting the amp into a special mode. I think they were using a Guger amp in that.
I'm hoping that K will setup a login here and jump in.
appears to be pretty straightforward, and it does look like the impedance can be measured in absolute terms (with some acceptable error) which is what we call around here 'Awesome!'
I will write you now a little letter about impedance measurement of the electrode-skin interface in EEG measurements.
First I would like to say biomurph something about impedance measurement. You cannot measure impedance by using DC current source. Impedance is a complex term which consists of two parts, real and imaginary. Like a series or parallel connection of the resistor and capacitor. Electrode-skin impedance can be modeled like parallel connection of the resistor and the capacitor. If you use DC current source, than you cannot impedance of that parallel connection. DC current source can be used only to say if the electrode is connected or not. That method is more used in ECG measurements, where the amplitude of the signal is 100x greahter than EEG signal amplitude.
ADS1299 posses AC signal sources with three different frequencies of the signal (7.8Hz, 31.2Hz and Fs/4). First two frequencies (7.8Hz and 31.2Hz) cannot be used for online impedance measurements, because this two frequencies are in range of the EEG signal spectrum. So I more prefer using the current source on the frequency of the Fs/4. I'm using sampling frequency of the 500Hz, so the current source frequency will be 125Hz.
If you short connect all INM inputs of the ADS1299, then we can measure impedance in two ways. First if we want to measure impedance of the all electrodes on INP inputs without measuring the impedance of the reference electrode on INM input. This impedance can be measured by calculating FFT of the every signal and extracting component at 125Hz. This component we can divide with the current of the current source, and we get impedance.
Second way is to measure sum impedance of the reference electrode and electrode on the INP input.
I've implemented this two methods and it works great.
I will send to you an equation which you can use for extraction component at 125Hz from FFT spectrum, and impedance calculation.
At the time, I was not prepared to increase the sample rate, so I stopped pursuing the idea of continuous impedance monitoring. Since we're now smarter with the system and how it responds, maybe now is the time to start re-considering this choice. If my only reason was fear of too much data, maybe that was a bad decision. Maybe increasing the data rate would work out just fine. Hmmm...
Chip
when using sample rate of the 250Hz, you cannot implement good impedance measurement. This sample rate is too low for EEG. What about person who have epileptic disease? EEG of the epileptic person is above 60Hz (70-90Hz). If you sample with 250Hz, you will not be able to see this spikes in EEG signal.
I'm using 500Hz sampling rate for 24 EEG channels (three ADS1299 IC's) which is 2mS. On every 2mS I sample all 24 channels and transfer that data to the PC over bluetooth (83 bytes) in real-time. The data is plotted and processed in real-time on PC + 24 FFT calculations for the impedance measurement.
In OpenBCI system you use only 8 channels, you can easily increase sampling rate to the 1024Hz. That must work properly without any interruption. That is my expirience.
K
In my situation, at 250 samples per second at 24-bits per sample times 8 channnels, I'm at 48 kbps of raw data. Given the overhead of the serial link, that is about as fast as a "56kpbs" link will go.
If you're doing 500Hz with 24 channels, you're pumping at least 500 Hz x 24bits x 24channels = 288 kpbs, which is quite a lot of data.
You're really doing that over Bluetooth? Which module are you using? How do you have it configured? Have you tried this in multiple locations, or just at one location in your home/lab?
I'm very curious!
Chip
I'm using Bluegiga WT12 Bluetooth module. Baudrate between MCU and Bluetooth module is 921600b/s. I'm using SPP prifile, so max baudrate in the air can 500kb/s. For this application you must use handshaking and ring buffer.
K
V = \sum_{n=0}^{N-1}(x(n)*e^(-j*2*pi*f_r*n/F_s))
If the frequency of interest is Fs/4, then the complex exponential in the above equation reduces to the repeating sequence { 1, -j, -1, +j}. That turns the multiply operation into a sign changing and data steering operation. This might not be all that useful with processors with hardware floating point multiply capability, but it can save a significant number of ops vs. an FFT.
-- Mike
first I would like to say to James that he done a great job, bu implementing impedance measurement. James, according to your schematic, you have internal serial resistors of 10Kohm on every input. The big problem is that real electrode-skin impedance at EEG measurements is not pure resistor. It also have a capacitor in parallel with resistance. In your case, you can simply subtract input resistance from the total calculated impedance and that will be the electrode impedance (modeled as a pure resistor)!!! In real EEG measurements we have a complex impedance (real + imag part). If you want to have a precise impedance measurement you must know starting phase angle of the current source. The problem is because we dont know phase angle of the current, so when I implemented impedance measurement I was assumed that impedance is modeled as a capacitor, so my formula for impedance measurement is:
Zelectrode = sqrt((V125/I)^2 - R^2);
where V125 is the amplitude of the voltage across electrode impedance + input resistance at 125Hz, I is the 6nA current source, and R is input resistance, which is in your case 10K, according to your schematic.
Can you please try to implement this, instead of:
float impedance = amp * 1000.0f / 6.0f; // your peace of code
All comments are welcome.
K
I checked out your code on github. Thanks for sharing.
There were a few things that I couldn't find...for instance, where are the header files that you #include? Specifically, I'm interested in looking at the one that does the EEG setup. Do you have it shared?
Thanks,
Chip
One question that I have is regarding the AC current source built into the ADS1299. I'm using the 6nA setting at ~31Hz frequency. What is the actual current waveform? Sine wave? Square wave? This affects how one scales the calculation of impedance.
For example, with my hardware using known resistors, I seem to get the best answer if I do my calculation as:
Impedance = Vrms * sqrt(2) / 6 nA
where Vrms is after filtering the signal (10 Hz pass band) around the expected ~31 Hz frequency. If this equation really is correct (especially the sqrt(2) scale factor), it implies that the waveform of the injected current has an amplitude (not RMS) of 6 nA. Or, said another way, that the injected current waveform is a sine wave with a peak-to-peak amplitude of 12 nA.
Is this correct?
Chip
If you're interested in a little more detail about how (I think) EEG impedance measuring works, you can check out by post:
http://eeghacker.blogspot.com/2014/04/openbci-measuring-electrode-impedance.html
Chip
its great to hear that you have measured impedance. Measuring frequency of 31Hz is not good because it is in range of the EEG signal spectrum. You must increase this frequency to Fs/4. In real EEG measurements you will have a error when measuring impedance because of the EEG signal at this frequency. Also, real EEG impedance is not pure resistive. It also have a capacitive component. Try to measure parallel connection of the known resistor and capacitor. Also try to measure series connection of the resistor and capacitor. You will see that this is not trivial task. You cannot simply subtract input resistor value.
I concur with your statements regarding the complications of the simplistic method that I've used...but I'd like to reply with the following views:
1) The purpose of the impedance measuring is to try to get it "low enough", so absolute accuracy is not as important as "is it low or is it high". Therefore, any real EEG signals near 31 Hz are only a problem if they are large. In this context "large" would mean EEG signals of a magnitude of (10 kOhm * 6 nA) = 60 uV, which is highly unlikely.
2) Your suggestion to move to Fs/4 is excellent if I want to monitor impedance continuously while doing my EEG recordings. While this is an excellent way of operating (and I know that you've achieved it), it is not my goal. For me, running at 250 Hz, fs/4 would be too close to my 60 Hz line noise. I know that it is possible to run faster, but we discussed that earlier.
3) I concur that the capacitive component of the skin impedance greatly complicates things and that my "subtract 5K" is inappropriate in a number of cases. As a first step, it is better than nothing. I need to build experience before making refinements. I bet others are in a similar place...they need to get started, see the deficiencies, and then make improvements. The purpose of my post was to get myself and others started on the first step.
I saw all your posts over on TI's community forum, too. You're really helping everyone understand how this is done!
Chip
http://eeghacker.blogspot.com/2014/04/impedance-of-electrodes-on-my-head.html
Chip
I've mentioned before that several prominent experts in the field of EEG neurofeedback have argued that skin impedances even in the range of 100K+ ohms are acceptable. This is because the input impedance of amps these days is so high, in the hundreds of megohms or even gigaohm. So considering the effect of that difference between 10K and 100K in how the EEG microvolt signal is effected -- the actual microvolt difference is miniscule. Yes you can see large differences in microvolts with the constant current injected signal used by the ADS1299 lead off circuit topology. But real EEG measured from the scalp is not operating in this constant current paradigm.
Here is a page from the book, Clinical Neurotherapy, the chapter by Dr. Richard Soutar. A very clear explanation of what is often a contentious issue. And it cites Robert Thatcher's painstaking numerical calculation as well, as further evidence.
http://www.lightfield.com/jpg/thatcher-impedance.png
As some more data points, I have done some DC impedance measurements of my Ag-AgCl saline sensor system, and also get impedances in the 100K range. This is completely normal for a saline system which uses no skin abrasion. Skin abrasion in fact these days is being called into question, since you want to avoid any possibility of breaking through the skin barrier. A number of other EEG headsets are using simple saline pads, such as the Emotiv EPOC, so impedances in this range are quite normal and functional.
I do see the value of the ability to check our impedances so we can see when they have gotten significantly out of range. Or when the ADS1299 is being used in say a medical / clinical setting where they might have strict rules about acceptable ranges for their applications or research.
Thanks again to Chip and everyone who is making this measurement possible, and adding to our knowledge of ADS1299 EEG practice.
commenter labeled "deboragalloway", is not really that person. This was
a side effect of importing the forum from the old bbPress software into
Vanilla forums. The account / person who generated the
"deboragalloway" posts, deleted their own forum account on the bbPress
site when their management complained that they may be divulging too
much information. And thus the internal database "account number" got
reused by the later user "deboragalloway". No relation to the posts in
this thread AT ALL.
Some of the posts above have also been edited
to remove that person's original name. That name was replaced with the
string "K", the amnesiac character in the Men in Black movie series.
:-) We want to thank K for all his contributions and suggestions. A
really brilliant engineering professor.
----
A more user-friendly tutorial on how to implement the impedance measurement algorithm, is on Chip's blog, at these posts,
http://eeghacker.blogspot.com/2014/04/openbci-measuring-electrode-impedance.html
http://eeghacker.blogspot.com/2014/04/impedance-of-electrodes-on-my-head.html
----
This September 21 addendum to this thread was prompted by another thread discussing the possibility of GSR measurements with the ADS1299. I referred back to this thread for a historic context.
http://openbci.com/forum/index.php?p=/discussion/156/gsr-openbci