Low noise reference

[Ken Smith]

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference. Are there any references with a low
noise spec. A care most about the frequencies above 0.1Hz, Below that
point, the noise can increase by 20dB/decade without causing too much
trouble.

 

[GregS]

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference. Are there any references with a low
noise spec. A care most about the frequencies above 0.1Hz, Below that
point, the noise can increase by 20dB/decade without causing too much
trouble.

I was going to try to take a few references and adding them. It will work but
takes many along with the filtering.

greg

 

[[email protected]]

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference. Are there any references with a low
noise spec. A care most about the frequencies above 0.1Hz, Below that
point, the noise can increase by 20dB/decade without causing too much
trouble.

Take a look at the Linear Technology LT1031

http://www.linear.com/pc/downloadDocument.do?navId=H0,C1,C1154,C1002,P1677,D3907

The data specifies a typical 11uV rms from 10Hz to 10kHz which is
110nVper root Hz.

The voltage noise plot puts the noise corner is at 40Hz, and show the
high frequency noise as closer to 120nV per root Hz.

The LT1031 is one of the better buried-zener based references around,
and does specify broad-band noise, unlike the Thaler VRE310

http://www.thaler.com/thcpdf/vre310.pdf

If you want to do better you may have to buy an integrated array of
Josephson junctions from the National Burea of Standards, and a bucket
of liquid helium to keep them at their working temperature.

http://www.lne.fr/en/r_and_d/electrical_metrology/josephson_effect_ej.shtml

 

[[email protected]]

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference.

Yep, that sounds right. Resistor noise from the impedance of a Zener
is higher than your target (see AD587 data sheet for instance; about
100 nV/sqrt(Hz) for a Zener-based reference). I hear that the old
Weston standard cells have good low output impedance (and
therefore lower noise).

The guys down the hall from me put together a 300-cell battery for one
experiment (but they were aiming for ten or eleven decimal places...
got there, too).

 

[Ken Smith]

Take a look at the Linear Technology LT1031

http://www.linear.com/pc/downloadDocument.do?navId=H0,C1,C1154,C1002,P1677,D3907

The data specifies a typical 11uV rms from 10Hz to 10kHz which is
110nVper root Hz.

I assume you mean LT1021 not LT1031. That is a mere 22 times bigger than
I need. It is better than many. The LT1236-10 seems to be in that same
range. Most of the rest I have found are above 300nV/sqrt(Hz).

[....]

If you want to do better you may have to buy an integrated array of
Josephson junctions from the National Burea of Standards, and a bucket
of liquid helium to keep them at their working temperature.

I think the customer may object. I'm already getting rumblings about the
size of the battery cart needed for the wrist unit.

 

[[email protected]]

I assume you mean LT1021 not LT1031. That is a mere 22 times bigger than
I need. It is better than many. The LT1236-10 seems to be in that same
range. Most of the rest I have found are above 300nV/sqrt(Hz).

I knew that either the LT1021 or the Lt1031 was good, but it was from
my time at Cambridge Instruments - 1982 - 1991. The broadband noise
spec for the LT1021-10 looks identical with the spec for the LT1031.

[....]

If you want to do better you may have to buy an integrated array of
Josephson junctions from the National Burea of Standards, and a bucket
of liquid helium to keep them at their working temperature.

I think the customer may object. I'm already getting rumblings about the
size of the battery cart needed for the wrist unit.

http://www.lne.fr/en/r_and_d/electrical_metrology/josephson_effect_ej.shtml

Then you probably don't want to know about the voltage reference for
electron microscopy in Review of Scientific Instrruments in the late
1960s. Some guy took a vacuum insulated standard capacitor - about
100pF, mounted in a bell-jar - and charged it up to his reference
voltage, and monitored the charge with a Carey vibrating reed
electrometer.

Capacitors don't generate Johnson noise, so the basic idea was okay,
but they are microphonic ....

Texas Instruments still sells the Burr-Brown low noise FET input op
amps OPA-657

http://focus.ti.com/lit/ds/symlink/opa657.pdf

with 4.8nV/root Herz noise and 2pA leakage current, but that would need
a capacitor cart dragged along behind your battery cart.

 

[Fred Bartoli]

Ken Smith a écrit :

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference. Are there any references with a low
noise spec. A care most about the frequencies above 0.1Hz, Below that
point, the noise can increase by 20dB/decade without causing too much
trouble.

I had some similar requirements for really low noise PSUs two year ago
and I had to get down to 0.1Hz too but I managed to have this point
somewhat relaxed.

The best ref. you can come with (well, the best I could find) are the
XFETs one.

All the refs I found were about 100nV/rtHz but that's not hard to
filter... except for the low freq part. Also they are not all 10V, so
for you the interesting data is the ratio of low frequency noise to ref
voltage.

The good old AD586 at 4uVpp/10V is still a good value, which only
equaled by the more recent 5V ADR445 (2.25uVpp/5V)

Are you really wanting under 5nV/rtHz flat down to 0.1Hz? That's pretty
tough.
Over a 0.1Hz-10Hz BW this is about 20nVrms or 120nVpp which is just
within reach (guaranteed) of LT1028/1128 and AD797.
Unfortunately the low LPF corner frequency needed to filter the ref
voltage requires high impedances which won't do any good with these
bipolar opamps high current noise.
The OPA227 with 3nV/rtHz and 90nVpp(typ) has a 2.5 to 5 time better
current noise, but is still high at 0.4pA/rtHz and skyrockets at
20pA/rtHz at 0.1Hz which will require high value caps with their own
noise pbs (over 10000uF, no even speaking of LT1128 or AD797).

Jfet opamps are really bad at low frequency. The best is probably the
AD745, but will still miss the target by a 10 factor at 0.1Hz

 

[Steve Goldstein]

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference. Are there any references with a low
noise spec. A care most about the frequencies above 0.1Hz, Below that
point, the noise can increase by 20dB/decade without causing too much
trouble.


--

Fundamentally a very tough requirement. Excluding the exotic stuff like
Josephson junctions, the two basic reference-generation methods use
either Zener diodes (noisy) or bandgaps (noisier). The bandgap loses
because the basic voltage you generate is fairly low, so lots of gain is
needed to get where you want to go. Zeners are just intrinsically noisy
gadgets, although the so-called buried Zeners used in most IC refs these
days are much better because at least they solve the surface-noise
problem.

No matter which you use, you still need an op-amp (more noise, but
probably far from the dominant contribution). If you really really need
the noise level you mention, I think filtering is your only practical option.
Watch out for cap leakage in the filter, these can give you shot noise
across your (presumably) large filter resistor.

steve

 

[Ken Smith]

Then you probably don't want to know about the voltage reference for
electron microscopy in Review of Scientific Instrruments in the late
1960s. Some guy took a vacuum insulated standard capacitor - about
100pF, mounted in a bell-jar - and charged it up to his reference
voltage, and monitored the charge with a Carey vibrating reed
electrometer.

I could use an extra battery in the battery cart. That would give me a
fairly low noise but I'd have to regulate its temperature to prevent
drift.

Capacitors don't generate Johnson noise, so the basic idea was okay,
but they are microphonic ....

Texas Instruments still sells the Burr-Brown low noise FET input op
amps OPA-657

I was thinking of using something like the LSK389. If I make the
capacitors out of nested spheres, the microphonics will cancel to teh
first order.

http://focus.ti.com/lit/ds/symlink/opa657.pdf

Check the 1/F point on the noise graph. It really sucks when you get down
to 0.1Hz.

with 4.8nV/root Herz noise and 2pA leakage current, but that would need
a capacitor cart dragged along behind your battery cart.

It is too bad that the great big power MOSFETs don't come with a noise
spec. I'd bet the noise voltage is quite low.

In real life, I think I'm going to end up with a large capacitor as a
filter. This will prevent the noise current from making too much noise
near the 0.1Hz point. Bipolar op-amps like the LT1007, have a much better
1/F point than the TI fet input one. The typical LT1128 meets the spec
but the worst ones don't.

Note for any Linear folks who happen to read this: Because of LT spice,
all things being equal a Linear part goes in the design.

 

[Ken Smith]

Ken Smith a écrit :

I need a 10V reference with a noise level below 5nV/sqrt(Hz). My current
thinking is that the only way to get this is to add a huge amount of
filtering to a normal reference. Are there any references with a low
noise spec. A care most about the frequencies above 0.1Hz, Below that
point, the noise can increase by 20dB/decade without causing too much
trouble.

[....]
The best ref. you can come with (well, the best I could find) are the
XFETs one.

I'll look this up.

[....]

The good old AD586 at 4uVpp/10V is still a good value, which only
equaled by the more recent 5V ADR445 (2.25uVpp/5V)

Are you really wanting under 5nV/rtHz flat down to 0.1Hz? That's pretty
tough.

If it was easy I'd be done now. Lucky for me, I have room for a large
capacitor or two and no "instant on" requirement.

Over a 0.1Hz-10Hz BW this is about 20nVrms or 120nVpp which is just
within reach (guaranteed) of LT1028/1128 and AD797.
Unfortunately the low LPF corner frequency needed to filter the ref
voltage requires high impedances which won't do any good with these
bipolar opamps high current noise.

If I use a truly huge capacitor I can get around that issue. Right now
I'm thinking about using a discrete part to do the 0.01Hz to 1KHz band and
an op-amp to set the DC level. Sort of like this:


Ref in --------!+\ !---+15V
! >--+--/\/\---+---->!
---!-/ ! ! !----+------ Ref out
! === === !
! ! ! !
---------+--/\/\--------------+
! !
GND \
/
\
!
-15V