Signal Quality Testing of EEG Systems

Since the Avertus H10C has been out in the market in March this year, I have received a lot of questions from researchers and application developers alike on how good the quality of the signal is, where does it stand stacked against other systems, how do we know it, etc. etc.

Acknowledging the importance of having reliable, high-quality signal when it comes to scientific experiments, as well as high-precision products, I decided that it's a good thing to write up something on this topic publicly. We also posted a segment of this study on our blog.

Conclusion: we know our signal quality is good, damn good in fact because we tested it - over and over again.

As in most cases in other industries, there are three forms of testing we apply:

1. White-box testing: this is a form of testing where we put signal into the headset using industry standard signal generator, with strict control on amplitude, frequency and wave forms. In our test, we used the Agilent 33250A signal generator commonly used in most labs to generate sin and square wave forms at controlled frequencies and amplitudes. We then connect the headset with the signal generator in a customized board and measure the output on our Avertus EEG software.

2. Black-box testing: this is a form of testing, contrasting to the white-box method, where we don't know the exact input signal values. In this case, what we are doing is comparing the recordings from the same source using two different devices: one is the H10C, another is a gold-standard device. We chose a gold-standard wired and a gold-standard wireless device each, and tested the same subject with the same, simple-to-reproduce task (eyes closed vs. eyes open in normal, seated scenario).

Below are the results for the leading dry, wireless EEG headset compared to the H10C:

Above is the H10C recording on channel O1 for eyes-closed, compared to below from a non-Avertus leading EEG device on channel O1 for eyes-closed. You can see the matching alpha oscillation mixed with some weaker power harmonics.

Now, we have also conducted an experiment on the leading wet, wired EEG device as well in a hospital setting with two wet EEG experts.

What we have found out is that the H10C demonstrates matching frequency to that of a gold-standard wet device.

This proves that the digital resolution and temporal resolution of the device being equivalent to industry leading devices with a real source of EEG.

3. Longitudinal equivalence testing: this is the final scenario where we test a large number of recordings over time and compare the signal features in terms of amplitude, frequency component presence (using power spectral analysis), other statistical methods as well as randomized visual comparisons to ensure that over the longer sessions ( > 4 hours) the signal does not deteriorate. Most recordings we did are virtually indistinguishable from leading, wet devices on the given frequency band investigated (1Hz ~ 65Hz) over long average.

I hope this helps provide some perspectives on how we treat quality here at Avertus.

Cheers,

Alex