From: Joerg Hansmann (info_at_jhansmann.de)
Date: 2002-01-31 16:14:48
Hi Andreas, Jim Meissner and all,
----- Original Message -----
From: sleeper75se <sleeper75se_at_yahoo.se>
To: <buildcheapeeg_at_yahoogroups.com>
Sent: Thursday, January 31, 2002 12:34 AM
Subject: [buildcheapeeg] Re: Brainmaster input stage
...
> > What do you think about the AD620 input current noise
> > (especially at low frequencies) relating to the high-impedance
> > high-pass network C4,R4 and C5,R5 ?
...
> I honestly don't know... your filter had a big resistor in series
> with the capacitor, perhaps that was the culprit? Have you tried
> removing the 10Mohm resistor in series with the 100nF capacitor?
The 10Mohm has not been in series with the 100nF C. It was the R-part
of a highpass filter. So it can not be removed.
(see attached picture, how the components were arranged)
> Jim wrote:
>
> > So by inspecting the BrainMaster input stage, I guess at some low
> > frequency the source resistance will be 10 Meg as the Xc of the
> > 0.01 mf capacitor increases with lower frequencies. I wonder if
> > you could try to simulate that stage again using larger capacitor
> > values. I would be interested to see what capacitor value will
> > give "good" noise down to 0.1 Hz? Somewhere I have the schematic
> > for the original Mind Mirror. I think they used a 10 or 100 mf
> > capacitors feeding into a discrete transistor for the lowest
> > possible noise.
>
> Jim, are you're talking about Joergs asymmetric filter?
>
> Anyway, I did some calculations.
>
> A network consisting of a 10Mohm resistor in parallel with a big
> capacitor has these impedances at 0.1Hz:
>
> C = 100uF => Z = 16Kohms
> C = 270uF => Z = 6Kohms
> C = 470uF => Z = 3.3Kohms
With these _very_ big capacitors Z would be OK and noise also,
however for these values you would need to use electroytic
(or tantalum) type Cs and they have too nasty properties
to be used in the input stage (high leakage and dielectric absorption),
and to get the right time constant you would have to use a fitting
R for the highpass in the 10th kOhms range, that would completely
ruin the input impedance of the amplifier.
> At 0.1Hz the resistance has very little influence. 10Mohms or no
> resistor at all, give approximately the same result, meaning this
> calculation applies to the BM-input as well.
>
> So with a modern instrumentation amplifier the noise above 0.1Hz
> should be negligible for C
> 270uF.
> The inevitable noise below 0.1Hz, caused by the resistor can be
> removed by a digital filter, if you want the "raw" signal to look
> more pleasing to the eye. For pure frequency anlysis, I don't see a
> need for it though.
>
> Joerg, could you try it out with your assymmetric filter, replacing
> the 100nF caps with 270uF?
It would not work because of the above reasons.
BTW: I am rather sure that the BM input stage used really
10nF caps (I have seen it on a picture and the Cs were definitely
no electrolytics).
I have done some impedance and noise estimations too and
got disastrous results, that matched with my experimental findings.
C = 10nF at 0.1 Hz => Z = 160 megOhms, 10megOhms R dominates -> 160uVp-p
C = 10nF at 1 Hz => Z = 16 megOhms, 10megOhms R dominates -> 160uVp-p
C = 10nF at 10 Hz => Z = 1.6 megOhms -> 16 uVp-p
C = 100nF at 0.1 Hz => Z = 16 megOhms, 10megOhms R dominates -> 160uVp-p
C = 100nF at 1 Hz => Z = 1.6 megOhms 16 uVp-p
C = 100nF at 10 Hz => Z = 160 kOhms 1.6 uVp-p
Now multiply the impedances by the noise current of 10pAp-p(for
0.1Hz..10Hz BW) to get the lower and the upper boundary of the
estimated noise voltage.
An exact solution would be to integrate the 1/f noise (that merges
with the white noise somewhere above 100Hz) shaped by the HP-filter
over the amplifier bandwidth (e.g. 0.16 Hz .. 75 Hz for modularEEG
or something else for the BM).
I have tried this with SPICE but was not successful until now
(Johnson noise is no problem, but how do I get/simulate 1/f noise ???
I have used a current source forward driving a diode but can not
see any 1/f noise ... :-(
)
@ Andreas:
In another posting you have shown your excellent skills in solving
noise equations. Perhaps you can comment on this...
Regards,
Joerg
This archive was generated by hypermail 2.1.4 : 2002-07-27 12:28:37 BST