PSN-L Email List Message
Subject: Re: AD7745 Capacitative sensors
From: ChrisAtUpw@.......
Date: Sat, 15 Apr 2006 00:24:42 EDT
In a message dated 14/04/2006, Brett3kg@............. writes:
> However, they are difficult to construct and have a working gap=
=20
of
> less than a millimetre, and are prone to off axis sensitivity.
>
> This is largely incorrect for amateur applications. If you demand 1nm=20
> resolution or better, you do need special materials and construction.
I'm not sure pure resolution is that hard to come by. Noise & temperature=20
sensitivity are another matter, though.
Hi Brett,
=20
There are ways of reducing the effects of thermal expansion without=20
using exotic materials. Noise can be greatly reduced using a dual FET front=
end=20
on a chopper amplifier. The latest types give about 1 nV rms / Root Hz.=20
> There are three basic types of capacitor sensor.
>You can have a pair of parallel plates excited by sine or square waves=20
>with a central sensor plate which moves perpendicular to the plane.
There are two major disadvantages with this type.
Movement of the sensor plate displaces air between the plates on both=20
sides. This damping + spring cushion effect depends on the plate size, thei=
r=20
separation and on the plate velocity - it varies a lot. The traditional met=
hods=20
of reducing these problems involve boring air holes in the capacitor plates=
=20
and evacuating the seismometer case.=20
This sort of sensor is only linear over the whole movement range if you=
=20
use either voltage sensing or a special feedback circuit with 'floating'=20
power supplies. However, you are critically dependant on stray capacity eff=
ects.=20
These can be largely compensated for sine wave excitation, but with some=20
difficulty. If you use a charge amplifier, it will only be approximately li=
near=20
for 1/3 to 1/4 the available movement range.
Since other designs without these problems are available, why do things=
=20
the hard way?
> You can use two pairs of parallel plates with a central sensor plate=20
> moving parallel to the plane
> You can use basically parallel circuit board plates with a pair of=20
excitation strips on one side, a cross coupled square / rectangular sense a=
rray=20
on the other and a plate with vertical 'shadow strips' moving parallel in=20
between. It is easy to make these out of double sided glass circuit board.=20=
The=20
central shadow plate does not need to be earthed. Only the fixed plates nee=
d=20
to be wired up, which is a considerable advantage. The maximum movement is=20
half the square 'cell size'. Again this can be quite large. See Randall Pet=
ers'=20
SDC sensor at _http://physics.mercer.edu/petepag/sens.htm_=20
(http://physics.mercer.edu/petepag/sens.htm) Charge detection is usually u=
sed. An array of=20
coupled cells can be used to increase the sensitivity. It is an advantage t=
o=20
make the shadow plate out of etched double sided glass board. The 'electric=
al=20
thickness' is the actual thickness divided by the dielectric constant. Havi=
ng=20
all three boards made from the same material greatly reduces any thermal=20
drift. There is no air flow problem with plate movement.
I'd always assumed that only the first type had adequate displacement=20
sensitivity, a couple of orders of magnitude greater than the=20
others. Sounds like I need to go back and check the numbers.
They mostly come down to dL / L considerations. The larger you make the=
=20
total range L, the more you need to amplify the signal. It is quite possibl=
e=20
to maintain a very high sensitivity by detecting the whole signal range and=
=20
then adding a high pass filter and more amplification. My LVDT Lehman senso=
r=20
allows the first stage to give +/-10V for the allowed +/-10 mm range. This=20
signal is then put through a high pass filter and the 'AC' component is=20
amplified further. This allows resolution of a few 10s of nm on my 16 bit AD=
C over=20
the whole 10 mm range. =20
> In a message dated 13/04/2006, Brett3kg@............. writes:
>Biggest VRDT problem seems to be its low drive frequency. In a feedback
>design the large demod filters are prime contributors to loop oscillation=20
>problems.
> So reduce the filtration and apply a DC + pulsed feedback? Use=20
> another method?
Can you amplify on this? Not exactly sure what your'e proposing, but it=20
sounds interesting.
If you look at the circuit in=20
_http://psn.quake.net/info/bb13OperManual.pdf_ (http://psn.quake.net/info/bb=
13OperManual.pdf) two stages of RC bandpass=20
filter are used starting with the square wave from a quartz oscillator. One=
=20
filter is on the input line to the differential capacitor sensor and the=20
other on the amplified output from the sensor. The output of the final ampl=
ifier=20
has a DC component with a large AC ripple on it at twice the oscillator=20
frequency. This is applied to the inductive winding of the feedback coil th=
rough a=20
parallel RC link to give a low phase error feedback signal. If you fully=20
smoothed the output, you would have a greater phase delay. You then add a l=
ow=20
pass filter to smooth the signal for the A/D converter.
> The feedback phase delay is only a problem if you do it this way!
> >I'm not sure about the noise. Does the VBB measure displacements in the
> >=B11 nm range?
>
>With the sensor plates above, 1LSB=3D0.08nm. But I think noise is what
>determines the useful resolution. However 0.3nm / sqrt-Hz and 2.1nm RMS a=
t
>50 SPS isn't too shabby. It would be interesting to assume a seismic-mass
>system and model how this would compare with commercial instruments and
>earth-noise models. I'm betting it won't look so bad.
> Have you measured your environmental noise level? Is 2.1 nm a=20
> realistic target? The amplitude of the 6 second ocean microseisms may be=20
> from 500 to 15,000 nm!
Actually 2.1nm was no target. That's just what I calculated you could get=20
using the AD7745. I agree that it is a good bit better than a typical home=
=20
site would justify, which is why it looked so interesting.
The lower limit for LVDT measurements is about 0.1 nano m. With=20
capacitative sensors you can reduce this by 100. This is far smaller than t=
he=20
background noise limits.=20
> He then goes on to describe the VRDT. I suppose for the VBB sensor this=20
> would greatly simplify the electronic design if one can deal with small=20
> sensor gaps. I'm not sure about the noise. Does the VBB measure=20
> displacements in the 1 =B1 nm range? --- Just thinking out loud. I thin=
k=20
> it greatly depends on what type of sensing one wants to do local,=20
> regional or teleseismic.
> Amateur seismometers are usually limited by either microseisms or =20
> by environmental noise - we can't usually choose a quiet remote site. I=20
> managed to reduce the noise of my LVDT to about 7 nm for a 6 mm range at=20
> 10 Hz, but my environmental noise is much greater than this.
>
> There are 'problems that you do not need to have' :-
>
> The velocity feedback damping does not need to be generated that way=
!
>
> Neither do we need to use that troublesome design of capacitative=20
sensor!
>
> *** You can use JUST position + integral current / coil feedback if=20
> you ALSO have a quad magnet + Cu plate for the velocity damping! Trying=20
> to provide velocity damping by differentiation and coil feedback is=20
> likely to very significantly increase the overall circuit noise! ***
I'm now thinking that's where I was heading, except for retaining the=20
"perpendicular" capacitance sensor. =20
I suggest that you reconsider the capacitative sensor design. You reall=
y=20
don't need pneumatic damping problems and it is helpful to be able to choos=
e=20
your range. If you are trying to measure a few parts in a 15,000 nm signal,=
=20
you need the lowest noise most linear system that you can get. A considerab=
le=20
reduction in expansion coefficient is possible using either conducting=20
paint, evaporated metal, or etched metal on thin pyrex sheet glass. You can=
stick=20
metal foil onto sheet glass with acrylic adhesive - don't try epoxy unless=20
you prebake the glass to over 150 C. The electrode plates can be made by ph=
oto=20
etching.
True chopper amplifier circuits are available. These are immune to 1/f=20
noise. The SDC type sensor offers a great constructional advantage in ease=20=
of=20
wiring - no electrical connections are required to the seismic mass - and y=
ou=20
can use coaxial screened cable with a charge amplifier. The design is not=20
too critical on electrode spacing, which can be quite small.
Since the AD7746 goes directly from=20
capacitance to digital, I was hoping to use either a PC or commercial DSP=20
chip or FPGA to do most of the Position-Velocity derivative and other=20
shaping. I like that because that needs minimal analog circuitry and what=20
you do need (integral current feedback) is working at virtually DC. And,=20
yes, for that to work you would need a well-damped and stable spring=20
mass. First-order displacement linearization could be done digitally. Also=
=20
you'd need to be sure that your displacement sensor range was adequate.
There are two problems associated with either analogue or digital generatio=
n=20
of a velocity feedback signal from a position signal. The generated signal=20
is inherently noisy. You can also run off the end of the voltage or count=20
scale --> system failure. The phase errors / delays need to be considered /=
=20
compensated.=20
It is relatively easy to provide a very quiet precision velocity=20
feedback signal within the stop limits of the mass movement using either ma=
gnet /=20
coil / resistor damping or a quad magnet / variable area damping plate. Thi=
s=20
uses NO external power OR electronics! Why 'do things the hard way'?
The _http://gravity.ucsd.edu/research/OFSEIS/opt_seis.html_=20
(http://gravity.ucsd.edu/research/OFSEIS/opt_seis.html) account claims a r=
eduction in the=20
feedback noise using magnet / coil / resistor damping.
24 bit DACs are available for positional feedback / integral feedback.
=20
Another factor which can significantly effect the overall seismometer=20
performance lies in the design of the suspension system, as you outlined in=
=20
your rolling foil design.=20
The 'art' of being successful lies largely in not making mistakes and i=
n=20
avoiding unnecessary problems and limitations.
=20
Regards,
=20
Chris Chapman
In a message dated 14/04/2006, Brett3kg@............. writes:
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000=20
size=3D2>> However, they are=20
difficult to construct and have a working gap of
> &nbs=
p;=20
less than a millimetre, and are prone to off axis=20
sensitivity.
>
> This is largely incorrect for=20
amateur applications. If you demand 1nm
> resolution or better, you=
do=20
need special materials and construction.
I'm not sure pure resoluti=
on=20
is that hard to come by. Noise & temperature
sensitivity are=
=20
another matter, though.
Hi Brett,
There are ways of reducing the effects of therm=
al=20
expansion without using exotic materials. Noise can be greatly reduced using=
=20
a dual FET front end on a chopper amplifier. The latest types give abou=
t 1=20
nV rms / Root Hz.
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000=20
size=3D2>> There are three basic types of capacitor=
=20
sensor.
>You can have a pair of parallel plates excited by sine or=20
square waves
>with a central sensor plate which moves perpendicular=
to=20
the plane. <clip>
There are two major disadvantages with this=20
type.
Movement of the sensor plate displaces air betw=
een=20
the plates on both sides. This damping + spring cushion effect dep=
ends=20
on the plate size, their separation and on the plate velocity - it vari=
es a=20
lot. The traditional methods of reducing these problems involve boring air h=
oles=20
in the capacitor plates and evacuating the seismometer case.
This sort of sensor is only linear over the who=
le=20
movement range if you use either voltage sensing or a special feed=
back=20
circuit with 'floating' power supplies. However, you are critically dependan=
t on=20
stray capacity effects. These can be largely compensated for sine wave=20
excitation, but with some difficulty. If you use a charge amplifier, it will=
=20
only be approximately linear for 1/3 to 1/4 the available movement range.
Since other designs without these problems are=20
available, why do things the hard way?
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000=20
size=3D2>> You can use two pairs of parallel plates=20=
with a=20
central sensor plate
> moving parallel to the plane=20
<clip>
> You can use basically=20
parallel circuit board plates with a pair of excitation strips on one side=
, a=20
cross coupled square / rectangular sense array on the other and a plate wi=
th=20
vertical 'shadow strips' moving parallel in between. It is easy to=20=
make=20
these out of double sided glass circuit board. The central shadow pla=
te=20
does not need to be earthed. Only the fixed plates need to be wired=
up,=20
which is a considerable advantage. The maximum movement is half t=
he=20
square 'cell size'. Again this can be quite large. See Randall Peters' SDC=
=20
sensor at http://physics.mercer.=
edu/petepag/sens.htm Charge=20
detection is usually used. An array of coupled cells can be used to increa=
se=20
the sensitivity. It is an advantage to make the shadow plate out of etched=
=20
double sided glass board. The 'electrical thickness' is the actual thickne=
ss=20
divided by the dielectric constant. Having all three boards made from the=20=
same=20
material greatly reduces any thermal drift. There is no air flow problem w=
ith=20
plate movement. <clip>
I'd always assumed that only the=
=20
first type had adequate displacement
sensitivity, a couple of orders o=
f=20
magnitude greater than the
others. Sounds like I need to go back=20=
and=20
check the numbers.
They mostly come down to dL / L considerations.=
The=20
larger you make the total range L, the more you need to amplify the sig=
nal.=20
It is quite possible to maintain a very high sensitivity by detecting the wh=
ole=20
signal range and then adding a high pass filter and more amplification. My L=
VDT=20
Lehman sensor allows the first stage to give +/-10V for the allowed +/-10 mm=
=20
range. This signal is then put through a high pass filter and the 'AC' compo=
nent=20
is amplified further. This allows resolution of a few 10s of nm on my 16 bit=
ADC=20
over the whole 10 mm range.
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000=20
size=3D2>> In a message dated 13/04/2006, Brett3kg@bnordgre=
n.org=20
writes:
>Biggest VRDT problem seems to be its low drive frequency. I=
n a=20
feedback
>design the large demod filters are prime contributors to l=
oop=20
oscillation
>problems.
> So reduce the=20
filtration and apply a DC + pulsed feedback? Use
> another=20
method?
Can you amplify on this? Not exactly sure what your'e=
=20
proposing, but it
sounds interesting.
If you look at the circuit in
http://psn.quake.net/i=
nfo/bb13OperManual.pdf two=20
stages of RC bandpass filter are used starting with the square wave fro=
m a=20
quartz oscillator. One filter is on the input line to the differential capac=
itor=20
sensor and the other on the amplified output from the sensor. The output of=20=
the=20
final amplifier has a DC component with a large AC ripple on it at twice the=
=20
oscillator frequency. This is applied to the inductive winding of the feedba=
ck=20
coil through a parallel RC link to give a low phase error feedback sign=
al.=20
If you fully smoothed the output, you would have a greater phase delay. You=20=
then=20
add a low pass filter to smooth the signal for the A/D converter.
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000=20
size=3D2>> The feedback phase delay is only a proble=
m if=20
you do it this way!
> >I'm not sure about the noise. Does the VBB=
=20
measure displacements in the
> >=B11 nm range?
>
>Wit=
h the=20
sensor plates above, 1LSB=3D0.08nm. But I think noise is=20
what
>determines the useful resolution. However 0.3nm / sqrt-H=
z=20
and 2.1nm RMS at
>50 SPS isn't too shabby. It would be interes=
ting=20
to assume a seismic-mass
>system and model how this would compare wi=
th=20
commercial instruments and
>earth-noise models. I'm betting it=
=20
won't look so bad.
> Have you measured your=20
environmental noise level? Is 2.1 nm a
> realistic target? The=20
amplitude of the 6 second ocean microseisms may be
> from 500 to 15=
,000=20
nm!
Actually 2.1nm was no target. That's just what I calculat=
ed=20
you could get
using the AD7745. I agree that it is a good bit be=
tter=20
than a typical home
site would justify, which is why it looked so=20
interesting.
The lower limit for LVDT measurements is a=
bout=20
0.1 nano m. With capacitative sensors you can reduce this by 100. This=20=
is=20
far smaller than the background noise limits.
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000=20
size=3D2>> He then goes on to describe the VRDT. I suppose for th=
e VBB=20
sensor this
> would greatly simplify the electronic design if one c=
an=20
deal with small
> sensor gaps. I'm not sure about the noise. Does t=
he=20
VBB measure
> displacements in the 1 =B1 nm range? ---=20=
Just=20
thinking out loud. I think
> it greatly depends on what type of sen=
sing=20
one wants to do local,
> regional or teleseismic.
> &nb=
sp;=20
Amateur seismometers are usually limited by either microseism=
s or=20
> by environmental noise - we can't usually choose a quiet remote s=
ite.=20
I
> managed to reduce the noise of my LVDT to about 7 nm for a 6 mm=
=20
range at
> 10 Hz, but my environmental noise is much greater than=20
this.
>
> There are 'problems =
that=20
you do not need to have' :-
>
> The=20
velocity feedback damping does not need to be generated that=20
way!
>
> Neither do we need to use that=20
troublesome design of capacitative sensor!
>
> =20
*** You can use JUST position + integral current / coil feedba=
ck=20
if
> you ALSO have a quad magnet + Cu plate for the velocity dampin=
g!=20
Trying
> to provide velocity damping by differentiation and coil=20
feedback is
> likely to very significantly increase the overall cir=
cuit=20
noise! ***
I'm now thinking that's where I was heading, excep=
t=20
for retaining the
"perpendicular" capacitance sensor. =20
I suggest that you reconsider the capacitative=20
sensor design. You really don't need pneumatic damping problems and it is=20
helpful to be able to choose your range. If you are trying to measure a few=20
parts in a 15,000 nm signal, you need the lowest noise most linear system th=
at=20
you can get. A considerable reduction in expansion coefficient is possible u=
sing=20
either conducting paint, evaporated metal, or etched metal on thin pyrex she=
et=20
glass. You can stick metal foil onto sheet glass with acrylic adhesive=20=
-=20
don't try epoxy unless you prebake the glass to over 150 C. The electrode pl=
ates=20
can be made by photo etching.
True chopper amplifier circuits are available.=20
These are immune to 1/f noise. The SDC type sensor offers a great constructi=
onal=20
advantage in ease of wiring - no electrical connections are required to the=20
seismic mass - and you can use coaxial screened cable with a charge=20
amplifier. The design is not too critical on electrode spacing, which=20=
can=20
be quite small.
<=
FONT=20
style=3D"BACKGROUND-COLOR: transparent" face=3DArial color=3D#000000 size=
=3D2>Since=20
the AD7746 goes directly from
capacitance to digital, I was hoping to=20=
use=20
either a PC or commercial DSP
chip or FPGA to do most of the=20
Position-Velocity derivative and other
shaping. I like that beca=
use=20
that needs minimal analog circuitry and what
you do need (integral cur=
rent=20
feedback) is working at virtually DC. And,
yes, for that to work=
you=20
would need a well-damped and stable spring
mass. First-order displacem=
ent=20
linearization could be done digitally. Also you'd need to be sure th=
at=20
your displacement sensor range was adequate.
There are two problems associated with either=20
analogue or digital generation of a velocity feedback signal from a position=
=20
signal. The generated signal is inherently noisy. You can also run off the e=
nd=20
of the voltage or count scale --> system failure. The phase errors /=
=20
delays need to be considered / compensated.
It is relatively easy to provide a very quie=
t=20
precision velocity feedback signal within the stop limits of the mass=20
movement using either magnet / coil / resistor damping or a quad magnet /=20
variable area damping plate. This uses NO external power OR electronics! Why=
'do=20
things the hard way'?
24 bit DACs are available for positional feedba=
ck /=20
integral feedback.
Another factor which can significantly effect t=
he=20
overall seismometer performance lies in the design of the suspension system,=
as=20
you outlined in your rolling foil design.
The 'art' of being successful lies largely in n=
ot=20
making mistakes and in avoiding unnecessary problems and limitations.
Regards,
Chris Chapman
[ Top ]
[ Back ]
[ Home Page ]