analog switch in EEG front-end circuitry ?

[Bill]

Hi,

I'm designing the analog front-end for a device that measures EEG
(approximately 0.5 to 50 Hz, 10 to 1000 uV peak-to-peak). I'm using a
good instrumentation amplifier, and a good input RC network that does
not degrade the CMR.

Once in a while, I also need to measure (estimate) the impedance of
the electrodes (silver/silver-chloride).

It would be perfect if I could disconnect the input RC network (the
one just in front of the instrumentation amplifier) while I measure
the electrode impedances.

Would you guys dare to insert, in series with the input lines (the
ones that carry microvolts of differential voltage), a good analog
switch like the ADG1636?
(http://www.analog.com/en/switchesmultiplexers/analog-switches/adg1636/products/product.html)

If not, why?

Mismatch (that could degrade the CMR) between Ron's can't be a
problem. I'll have 1 % 6.04 kohm in series with each input line. Not
even the Ron itself is a problem. The ADG1636 has around 1 ohm of on
resistance.

Off resistance can't be a problem, either.

Linearity of the on resistance as a function of the analog voltage
across the switch can't be a problem either, because Ron is much lower
than 6.04 kohm.

I'm only worried about noise. Would it be a problem? The datasheet
says very little about noise (it mentions 0.007 % THD+N and shows a
graph). My instrumentation amplifier is an INA121UA (with 20
nV/sqrt(Hz) @ 1 kHz).

Would you dare to use such an analog switch in this application?
Any of you has experience in this field?

Thanks a lot.
Bill

 

[David L. Jones]

Hi,

I'm designing the analog front-end for a device that measures EEG
(approximately 0.5 to 50 Hz, 10 to 1000 uV peak-to-peak). I'm using a
good instrumentation amplifier, and a good input RC network that does
not degrade the CMR.

Once in a while, I also need to measure (estimate) the impedance of
the electrodes (silver/silver-chloride).

It would be perfect if I could disconnect the input RC network (the
one just in front of the instrumentation amplifier) while I measure
the electrode impedances.

Would you guys dare to insert, in series with the input lines (the
ones that carry microvolts of differential voltage), a good analog
switch like the ADG1636?
(http://www.analog.com/en/switchesmultiplexers/analog-switches/adg1636/products/product.html)

If not, why?

Mismatch (that could degrade the CMR) between Ron's can't be a
problem. I'll have 1 % 6.04 kohm in series with each input line. Not
even the Ron itself is a problem. The ADG1636 has around 1 ohm of on
resistance.

Off resistance can't be a problem, either.

Linearity of the on resistance as a function of the analog voltage
across the switch can't be a problem either, because Ron is much lower
than 6.04 kohm.

I'm only worried about noise. Would it be a problem? The datasheet
says very little about noise (it mentions 0.007 % THD+N and shows a
graph). My instrumentation amplifier is an INA121UA (with 20
nV/sqrt(Hz) @ 1 kHz).

Would you dare to use such an analog switch in this application?
Any of you has experience in this field?

Thanks a lot.
Bill

Any particular reason why you don't simply use a nice small reed relay?
A solution with far fewer gotcha's.

Dave.

 

[Bill]

Reed relays are nicer in this respect, and if you can control the
orieintation of the board carrying the switches, mercury wetted reeds
last longer than dry reeds (10^8 operations as opposed to 10^7) and
offer slightly lower and more stable on-resistance).

I did take a look at many reed relays, but the problem is their size
(specially their height). All the parts used should be as flat as
possible.

And no, I cannot control the orientation of the switches, since this
will be a portable device.

Thanks.

 

[Bill]

You may need to worry about stray capacitances - the data sheet lists
a bunch of channel capacitances from 68pF to 220pF - and you may want
to isolate the positive and negative supply pins from the power rails
with LCR filter networks to keep supply ripple out of the signal
inputs.

Yes, that sounds critical, too. Supplies should be clean, although
Fig. 21 in the ADG1636 datasheet suggests good power supply rejection
(by the way, I wonder what happens below 50 Hz).

I'm not too worried about the 220 pF, since the input RC network has 1
nF and 10 nF capacitors.

Thanks.

 

[David L. Jones]

I did take a look at many reed relays, but the problem is their size
(specially their height). All the parts used should be as flat as
possible.

How does 3.4mm sound?
http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=374-1039-1-ND

Dave.

 

[Bill]

How does 3.4mm sound?
http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=374-1039-1-ND

Doesn't sound bad. A little pricey, though (and I would need four of
these, because they are 1xSPST and I need 1xDPDT). I'll think about
it.

Thank you.

 

[Nicholas Kinar]

Sure. A cmos switch has zero offset and just Johnson noise. As Bill
says, keep the power supplies clean.

Which part numbers of CMOS switch have you used? Could you recommend a
manufacturer?

Do you buffer the switch with an op amp? Is the buffer placed before or
after the switch?

 

[Nicholas Kinar]

Thanks, John!

Just an HC4051 type is good. Agilent uses them in their 6.5 digit
DVMs. There are lots of fancier/more expensive parts around too.

It is very neat to know that such parts are used by Agilent.

Depends on what you're driving. One caution is that most CMOS switches
kick out a bunch of charge when they switch, and that can mess up some
opamps for a bunch of microseconds.

Perhaps the best way to deal with this would be to simply ignore
measurements taken immediately after each switch? I can see that this
wouldn't be much of a problem for calibration, which is occasionally done.

I would imagine that good CMOS switches would have low ON resistance and
low capacitance.