Fwd: Facial muscles and EEG-ECG-EMG (Stefan Jung)

Well, another wonderful rant by the self-appointed EEG guru Stefan Jung.
Thanks Stefan, almost as entertaining as some of your previous anti-Emotiv
efforts. I really don't understand why you hold such a grudge, but maybe
there are some personal or financial motives I just don't
understand. Please DO NOT refer to Emotiv products and staff in such a
derogatory way in public again. You are misinformed and clearly know less
about EEG than you profess to. Your comments are slanderous and will not be
tolerated. In answer to your many erroneous points (and some valid but
irrelevant ones):

1. You are conflating EMG artefacts with movement and slippage artefacts.
These aretwo completely  different issues. EMG is an inherent body signal
and is regarded as an EEG artefact when arising from the activation of
muscles in the vicinity of an EEG sensor. EVERY EEG system, from OpenEEG to
the $60,000 clinical systems, gathers exactly the same data. It is less
apparent in a clinical setting because the poor subject is strapped down
and prevented from normal movement, speech or interaction in order to
obtain a 'clean' signal for a specific clinical purpose. This is perfectly
acceptable in a clinical setting, the subject wants to know if they have
some kind of damage or illness. In the real world where EPOC and many other
systems live, users are free to walk, talk, eat and laugh while wearing the
equipment. This is not bad design, it is DESIGN FOR USE. Emotiv uses the
muscle signals (inherent EMG) to determine which muscle groups are
activating, and by inference to calculate the user's facial expression. Far
from being a marketing take on a bad design, this is a very intelligent use
of a resource which is normally excluded from EEG. On top of that, the
classification of these muscle signals actually allows us to better analyse
the real brain signals - we already have a heads-up on the expected nature
of the artefact so we can extract the real brain data for use in our other
detection suites. MOTION and SLIPPAGE artefacts are more problematic -
these contribute to a sudden change in contact impedance, generating pops
and jumps in the electrical signals. These happen because the shape of the
head changes or the skin stretches and contracts as the face or head move
around. EPOC deals with these items mechanically - they are not completely
eliminated but they are WAY less prevalent in a real setting than EEG caps
and many other EEG solutions. EPOC sensor arms are independently sprung, so
even if one sensor is affected by movement there is no resultant pressure
or movement on the remaining sensors. We do not see a significant amount of
data loss to this effect. It is NO WORSE than most other systems except
where sensors are literally glued to the skin on independent wires. EPOC is
WIRELESS and LOW MASS to minimise this effect and this has many other
advantages (like a willingness of the subject to be seen in public).

2. We use a relatively low input impedance (1M is not exactly tiny by the
way) because we can. We extensively tested a wide range of input circuits
(yes, we even tested active sensors, they were NOT deleted by the bean
counters but by the engineers, because they didn't contribute anything
useful to the mix, especially for the additional cost and complexity).
Saline sensors are superior to gel pasted solutions in a consumer setting.
Nobody wants to get prepped and gelled. Saline also has the advantage that
it penetrates even very difficult hair. It provides EXACTLY THE SAME
electrical signals as a proper gelled sensor - perhaps the impedance is a
little higher on average, and the saline evaporates over a couple of hours,
other than that it is perfect (and adding glycerin to the saline helps
reduce impedance and evaporation rate). We benchmarked EPOC sensors against
gelled sensors using a clinical EEG system and we found NO DIFFERENCE in
real world use. The typical contact impedance with green indicators for
EPOC is in the range 5-15k. An amplifier input impedance of 1M completely
dominates that level of contact impedance, higher input resistance is
complete overkill. EEG amplitude is barely affected until the contact
impedance exceeds 100k - and at this level the EPOC contact quality
indicators are BLACK. You know you need to fix it.

So why did we choose 1M instead of more? Two reasons. One is the shot
noise, which increases significantly as the resistance increases. It
contributes to the noise floor of the instrument. We have been able to
build an economically viable EEG sense circuit for 14 CHANNELS - I repeat,
14 CHANNELS, by paying attention to noise sources and this was one of them.
The OTHER reason is related to interference. We wanted to put the
amplifiers on the main circuit board to avoid the complexity of active
amplifiers. If you do this, you effectively hang a long antenna with a
terminating resistance of ~10k ohms off a high impedance amplifier input.
In my day we used to call that an antenna. Electrostatic pickup from people
walking around the user pours into the antennas (don't forget there are 14
of them spread around the head) and this perturbs the body potential. By
keeping the input impedance low enough, we can cancel that effect using the
DRL circuit and other magic tricks. If we wanted a higher
amplifier impedance, we would be forced to move the amplifiers onto the
sensors to minimise the antenna length - which complicates the design and
definitely costs more. Note that we have a different design brief with the
new Insight product, it does indeed  use a much higher input impedance with
active amplifiers, because the semi-dry sensor material is harder to work
with than saline. I look forward to your whining about why this is a stupid
mistake as well.

3. "It doesn't matter what anybody says, publishes or writes"??? I wish I
was so vain that my opinion outweighs about 50 independent EEG and
neuroscience research groups who have published data, including direct
comparisons of EPOC with $60,000 clinical EEG systems, in reputable
refereed journals. Stefan Debner's group were able to extract single-sample
P300 responses, Nick Badcock's group showed IDENTICAL N1 auditory mismatch
negativity responses to a Neuroscan clinical device FITTED AT THE SAME
TIME. Nearly every major EEG research group on the planet has purchased and
is using EPOC EEG model for their research, usually alongside more
expensive equipment they already have. It is supported by OpenViBE, BCI2000
and BioExplorer and endorsed by neurofeedback providers. I won't go on,,,

Get over it Stefan, you have set yourself up as a guru on this site but you
are working from a very biased perspective and you seem to think nothing
much has changed since the OpenEEG project began. It is a fantastic design
and all kudos should go to the originators, yourself included, but it is
not the only way to solve the problem. Believe it or not, there are
engineering schools all over the world teaching exactly the same subjects
as you studied, and who knows, some of those millions of engineers may even
be smart enough to come up with something you didn't think would work. EPOC
has sold tens of thousands of units to researchers worldwide. They can't
all be stupid!

Daniel, if you're having trouble getting decent signals from EPOC there are
usually only a few causes. Inadequate wetting, incorrect fitment and dirty
or damaged sensors. You may also not have noticed there are two potential
locations for the reference detector set - some head shapes benefit from
using the alternates, or relocating the comfort pads.

Regards,
Geoff Mackellar
CTO Emotiv


On Tue, Jan 21, 2014 at 3:04 AM,
<[email protected]<javascript:_e({}, 'cvml',
'[email protected]');>
> wrote:

> Message: 1
> Date: Mon, 20 Jan 2014 14:58:48 +0200
> From: Stefan Jung <[email protected] <javascript:_e({}, 'cvml',
> '[email protected]');>>
> Subject: Re: [Openeeg-list] Facial muscles and EEG-ECG-EMG
> To: [email protected] <javascript:_e({}, 'cvml',
> '[email protected]');>
> Message-ID:
>         <CAC3TAzM5e6W5AxsYYB59b8-cfCaNmN3aqkkt1Ku=
> [email protected] <javascript:_e({}, 'cvml',
> '[email protected]');>>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Grant and further down my msg, for Daniel
>
> Any EEG signal is full of artefacts.  If your name is Emotiv and you use
> shitty saline electrodes with an EEG front-end that only has a 1 mega-ohm
> input impedance, then the signal quality will be 10x worse than what the
> professionals get with prepped and conductive paste applied Ag/AgCl
> electrodes.
>
> The Emotiv EPOC is a toy, intended for gamers, doesn't matter what anybody
> says, publishes or writes.  The low input impedance is just beyond
> laughable.
>
> Rather than try and wish the artefacts away, Emotiv turned this around and
> made it a selling point for their so called expressive detection software.
>
> Here is your problem: Emotive only uses electrodes on the scalp.  If you
> want to reliably pick up facial muscles, it stands to reason that you need
> to apply electrodes to the face (EMG) and around the eyes (EOG).
>
> What do you mean with a "simple" ECG or EMG circuit?  I hope you don't mean
> simpler than openeeg.  Well, actually you could build a very good device
> with a much reduced component count, using modern ADCs such as the ADS1xxx
> family from TI.  But doing such will not be lighter on the pocket, in fact,
> it's likely to cost more.  Not sure if I equate your simple as also meaning
> low cost.
>
> If we ignore the possible requirement of many channels, the monopolar
> versus bipolar input design and ignore the possible concurrent sampling
> requirement for EMG (the latter is not applicable for you Grant, but to our
> other friend (Daniel) looking for a ECG/EMG/EEG circuit it might be), then
> the circuit will look the same.
>
> Seeing that the EEG is the biometric signal with the lowest amplitude, a
> circuit such as openeeg could very easily be modified to work with ECG or
> EMG - you simply reduce the gain.  In fact, an adjustment on that gain
> pre-set might be enough.
>
> Granted, the EEG has the lowest bandwidth requirement, while ECG and EMG
> are higher.  But this is really only applicable to professional equipment
> used in a professional application.  The typical sampling rate of 256SPS is
> enough for general ECG monitoring.
>
> EMG has a VERY, VERY wide possible application range, so there it is more
> difficult to comment.  You could argue that EOG is a subset of the EMG and
> for EOG a professional would need to go down to DC, i.e. 0Hz.  For general
> muscle flexing pickup such as Grant wants to do, the 256SPS is no problem.
> If on the other hand, you want to check nerve impulse propagation over a
> specific area, you would want concurrent sampling at a minimum of say, 4Khz
> sampling, for 16/32/64/ etc. channels, making such an EMG device rather
> expensive, in comparison.
>
> Because 256SPS is SOOOO pedestrian with todays technology, a modern
> amplifier that samples at 4 or 8 KHz is absolutely no problem to build and
> design, and that would then be enough for any and all EEG, ECG and EMG
> work.  The very small disadvantage with a wideband amplifier would be that
> for EEG, where this wider bandwidth is not required, it would increase the
> noise floor slightly.  But even then, you could filter it out in the
> digital domain.
>
> That brings us to the different gain/sensitivity requirements of EEG, ECG
> and EMG.  If you use a modern 24-bit low noise ADC, you can expect a real
> world resolution of perhaps in the order of 18 or so bits.  That should be
> enough dynamic range for the amplifier to handle all type of signals.
> Obviously, if you go cheap and don't even use a separate ADC, but instead
> use the internal ADC of some micro (as openeeg does), well then forget it.
> Most micros have 10-bit ADCs and some manage 12-bits.  That is not good
> enough.  In that case one has no option but to implement gain switching.
>
> For the benefit of the Daniel crowd over here; gain switching would best be
> done using CMOS switches sitting in the negative feedback loop of an
> opamp.  Preferably a bit further up the signal path, after the signal has
> already been amplified somewhat.  I did such on my own EXG2 design, see
> here: *sites.google.com/site/ <http://sites.google.com/site/>openexg/*
>
> At the end of the day, whatever you want to do, you don't only need a
> reasonable performing amplifier, but a good electrode setup.  If you use a
> shitty pickup such as is used in the Emotiv EPOC, then it is a matter of
> crap in, crap out.  No amplifier can create a good signal from garbage.
> Ask any RF engineer.  You start with a GOOD aerial if you want good RF
> reception.  Ditto with biometric signals.
>
> Finally, as much as I have been bashing the Emotiv EPOC for the last few
> years, considering the cost restrain, the design is not that bad at all.
> Patent applications show that the Emotiv crowd wanted to implement active
> electrodes.  Emotiv's bean-counters must have stopped the engineers from
> implementing the extra 14x unity gain opamp buffers due to costs.  This is
> the decision that torpedoed that product into a heap of junk.  Just my
> view.
>
> Regards
> Stefan
> -------------- next part --------------
> An HTML attachment was scrubbed...
>
> ------------------------------
>
>

------------------------------------------------------------------------------
CenturyLink Cloud: The Leader in Enterprise Cloud Services.
Learn Why More Businesses Are Choosing CenturyLink Cloud For
Critical Workloads, Development Environments & Everything In Between.
Get a Quote or Start a Free Trial Today. 
http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk