Characterising low frequency noise?

[alanambrose]

Hi,

I thought it would be helpful to gather together a thread with the various pre-amp designs for measuring noise, plus the thinking and characterisation techniques focusing mainly on 0.1-10Hz voltage ref noise, but also bearing in mind ~10-20MHz PSU/regulator noise. This is a new area for me, so I feel only qualified enough to ask some of the dumb questions I'm hoping to get people's thoughts on:

+ the pluses and minuses of the various available pre-amp designs
+ techniques for measuring / characterising noise
+ instruments and techniques

As always I'm hoping that we might sometimes wander off topic to relevant sub-topics. Following Rumsfeld, here are some things I think I know, but am pretty sure I don't understand well...

Some of the various amplifiers / filter designs for noise measurement:

Linear (Jim Williams) - http://cds.linear.com/docs/en/application-note/an124f.pdf
TI - http://www.ti.com/tool/TIPD122
Andreas - https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg834013/#msg834013 (see last attachment in message)
Cosylab - http://users.cosylab.com/~msekoranja/tmp/04447683.pdf
Blackdog - https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/msg684852/#msg684852
c4757p - https://www.eevblog.com/forum/projects/op-amp-for-amplifiying-5uv-of-noise-to-200uv/msg365606/#msg365606 (see png at end of message)
tangentsoft - http://tangentsoft.net/elec/lnmp/misc/schematic3.pdf
Gerhard Hoffmann - http://www.hoffmann-hochfrequenz.de/downloads/lono.pdf
Linear - Measuring 2nV/?Hz Noise and 120dB Supply Rejection on Linear Regulators - http://cds.linear.com/docs/en/application-note/an159fa.pdf

Some noise measurement and low noise design techniques:

Intersil - http://www.intersil.com/content/dam/Intersil/documents/an15/an1560.pdf
Electronic Design - http://electronicdesign.com/analog/build-your-own-capacitor-free-high-pass-filter

Some previous discussions:

Measuring amplifier for 1 / f noise 0.1 - 10 Hz http://translate.google.com/translate?hl=&sl=de&tl=en&u=http%3A%2F%2Fwww.mikrocontroller.net%2Ftopic%2F207061&sandbox=1
Low frequency, very low level, DC biased, noise measurements https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/50/
Op amp for amplifiying 5uV of noise to 200uV? https://www.eevblog.com/forum/projects/op-amp-for-amplifiying-5uv-of-noise-to-200uv/msg364788/#msg364788
Question about preamp for measuring noise. https://www.eevblog.com/forum/beginners/question-about-preamp-for-measuring-noise/msg204478/#msg204478

Filtering voltage ref noise:

TI SBVA010 - "Improved Voltage Reference Filter Has Several Advantages" http://www.ti.com/lit/an/sbva010/sbva010.pdf
Walt Jung - "Build An Ultra Low-Noise Voltage Reference" http://waltjung.org/PDFs/Build_Ultra_Low_Noise_Voltage_Reference.pdf

Calculating op amp noise:

Nat Semi - "Noise Specs Confusing?" http://www.electro.fisica.unlp.edu.ar/temas/pnolo/p1_AN-104.pdf
AD MT-047 - "Op Amp Noise" http://www.analog.com/media/en/training-seminars/tutorials/MT-047.pdf
AD MT-048 - "Op Amp Noise Relationships: 1/f Noise, RMS Noise, and Equivalent Noise Bandwidth" http://www.analog.com/media/en/training-seminars/tutorials/MT-048.pdf

References:
Noise Reduction Techniques in Electronic Systems, 2nd Edition 2nd Edition by Henry W. Ott http://www.amazon.com/Noise-Reduction-Techniques-Electronic-Systems/dp/0471850683/ref=tmm_hrd_swatch_0?_encoding=UTF8&qid=&sr=
Low Level Measurements Handbook - 7th Edition http://www.tek.com/sites/tek.com/files/media/document/resources/LowLevelHandbook_7Ed.pdf

As always, cost-effectiveness is helpful. For instance, I read that a 'dynamic signal analyser' (which appears to be a lf spectrum analyser) is great for making 0.1-10Hz spectral noise plots - but this is not easily available technology. Personally, these questions are bugging me atm:

+ what are the pluses and minuses of the various noise amplifier designs?
+ what are the effective lowish-cost measurement techniques?
+ what about <0.1Hz?
+ is there a cost effective technique for spectrum plots?
+ what is the good instrumentation / software for this use - analog scopes?, adcs? adc-style signal acquisition with 3458As? Tektronix 7A22? something else?
+ why do all the commercially available amplifiers and dynamic signal analysers seem to be available only as old boat anchors - does nobody do this stuff these days, or do they use different techniques/instruments?

I hope this is useful / interesting. As always, I also hope that some of our more experienced members will give us the benefit of their knowledge & thinking. Does anyone care to set the context by describing a 'map' of where we are now?

Regards, Alan

[Andreas]

Hello Alan,

thanks for collecting all that info onto one site.

The design depends very much on the purpose for which you want to use the pre-amplifier.
There is no easy way which fits all.
So what application do you want to do with this amplifer.
Which noise levels and which DC-offsets do you want to cover?

Most of the designs are only valid if you have no DC-bias into your cirquit.
The input impedance is a design criteria.
Most OP-amps with low voltage noise have a large current noise.
So a low voltage noise needs a low input impedance.
With a low input impedance you cannot measure directly a unbuffered LTZ1000.
The charging pulse of the input capacitor+input impedance will degrade the LTZ1000 for many months,
since the setpoint of the temperature controller goes to infinity.

So my design is done with the constraint of being able to measure references with DC-Bias offset.
A maximum reference differential resistance of up to about 10 ohms without much affect on the measurement.
A noise << 1.2uVpp (< 0.2 uVpp) for a LTZ.
This is possible when selecting the input resistance of around 1K.
But during charging the input capacitor you will need a extra resistor in order not to kill the LTZ.

+ what about <0.1Hz?

For 0.1 Hz you will need a minimum observation time of 10 sec.
Below 0.1 Hz you have to be very patient.
And how do you distinct thermal drift from noise below 0.1 Hz?

with best regards

Andreas

[T3sl4co1l]

If you've got AoE3, there's a lot in there about noise and measurement, too.

Tim

[Kleinstein]

The instruments one could use these days would be a reasonably quality bench DMM, likely a low noise amplifier and software on the PC. The software could be just something like excel or a package to do FFT. There should also be some free versions available.  The main part of interest for the DMM would be a mode to get continuous 1PLC readings with more than about 3-4 digit resolution - so nothing really that demanding. Higher resolution might be helpful if the amplifier does not have fine adjustment.  Alternatively something like a 12 -16 bit ADC card for the PC should work too. Sound-cards are usually missing the DC and LF  part - otherwise they would be an option.
Digital scopes can be tricky and have too much low frequency-noise, and 8 Bit resolution might need a limited frequency-band. Some USB scopes with higher resolution converters might be a very good option.

50 Hz readings should be just enough to get the noise up to 10 Hz - the lower limit is possibly set by the amplifier or the time used to take date. Also thermal fluctuations get more imports at very low frequencies, especially with amplifiers based on discrete transistors.

For very low values the classical 2 channel method with 2 FET based amplifiers , ADCs and than correlation in software is also worth looking at. It may just take some time to find suitable software. The nice thing is that one could trade in longer measurement times for lower noise relatively easy. It's somewhat an advantage as thermal fluctuations are not that reproducible to allow accurate subtraction of single channel measurements - as a first approximation this still works.

For amplifier one has the choice of BJT based (high current noise), low noise AZ OPs (possibly high current noise, essentially no 1/f noise) and JFET based amplifiers (often relatively high 1/f noise and thermal effects). As far as I know, there is no clear winner here. The low current noise of JFET based versions can be essential even at low frequencies. Its just expensive to make them work well with low impedance. Usually modern amplifiers are good enough for all but extreme cases (e.g. resistors at low temperature).

Very low frequencies (e.g. < 0.1 Hz) might need DC compensation instead of AC coupling at the input.

[TiN]

Nice idea, Alan, subbed to thread. 

I don't have anything to fill in this thread as of this moment, but I think Keithley's 7th Edition Low Level Measurements Handbook is relevant.

Also while ago one of EEVBlog members started thread about Keithley 1801 nV preamp for 2001/2002 DMMs. I had finish design of preamp supply plugin board for it with isolated PP SMPS and ultralow noise LDO's, but did not had much time to play with that setup and EM A10 preamp head yet. Thread owner and few other volt-nuts have that board too  .

Hopefully I'll have something worthy to show here later on.

[alanambrose]

>>> If you've got AoE3

Good point, just ordered, I was being cheapskate over $70.

>>> So what application do you want to do with this amplifer. Which noise levels and which DC-offsets do you want to cover?

For me, low impedance voltage refs and 'low noise' voltage PSUs/regulators/sources - all <25V. I don't yet understand what anyone uses low noise current sources for, I would like to know though My interest is OCD/nuts motivated.

>>> Below 0.1 Hz you have to be very patient. And how do you distinct thermal drift from noise below 0.1 Hz?

Can 'drift' with time be thought of as vlf noise? 'Thermal drift' I'm hoping to correlate out by regressing general drift vs temperature and then compensating for it (which begs the question 'temperature of what ).

>>> 50 Hz readings should be just enough to get the noise up to 10 Hz

Is there a general formula for sampling 'random walks'? My guesstimate was 2 x 10Hz (i.e. 2.5 NPLC in 50Hz land) plus a bit for luck, so I've been using 2NPLC.

>>> For very low values the classical 2 channel method with 2 FET based amplifiers , ADCs and than correlation in software

Apologies, but I have not come across that technique before - do you have a reference/link anywhere?

>>> Keithley 1801 nV / EM A10

Ah interesting, I hadn't seen the 1801 and you just reminded me about the the EM A10. More choices

Ho ho ho, you've already given me several weeks of homework I'll undertake to update the first post so it has a ready reference set of links. There must be a better way

Alan

[Andreas]

Hello,

I mentioned the "current noise" of a pre-amplifier.
Which is a design criteria.
You cannot have both: low voltage noise and low current noise.
So you have to optimize for a certain input impedance.

See also DN140, DN6 and DN3 of Linear Technology.

With best regards

Andreas

[Kleinstein]

The two channel method is a classical method from old times when amplifier noise was a real problem and FFT was not jet available. As far as I know it is still the standard method of choice for very low noise level, e.g. with DUTs with lower noise than the amplifiers used.  Could not find a really good free source, here a few links, with the 1st beeing likely the best:

http://tf.nist.gov/general/pdf/1133.pdf
http://www.felicecrupi.it/wp-content/uploads/2011/04/IMTC07_A.pdf
http://home.deib.polimi.it/sampietr/spectrum/SPECTRUM1.html

The main idea is, that the power spectrum is equal to the fourier-transformed auto-correlation function. The correlation of the output of two amplifiers that measure the same source gives the auto-correlation plus the cross correlation of the noise of the two amplifiers. As the noise of separate amplifiers is usually not correlated the amplifiers noise is ideally eliminated. Especially longer averaging can reduce the background very effectively.

[branadic]

There are two very good german threads for two different types of noise amps:

0.1 - 10Hz:
http://www.mikrocontroller.net/topic/207061#2046034

10Hz - 100kHz:
http://www.mikrocontroller.net/topic/250656#2573926

Is there a general formula for sampling 'random walks'?

No, not as far as I know. But modified allan deviation or overlapping allan deviation is a nice tool to indicate at what point your values are stable and what noise types are present in your system.

[alanambrose]

Hi,

>>> There are two very good german threads

The 1st I've seen but the 2nd is new to me, thanks. Quite hard going if you're a Google translate user but patience and persistence is rewarded (apparently)

>>> The two channel method

Very interesting, I think I get it - and it looks quite doable.

>>> modified allan deviation

Ahhhh, Allen deviation for noise measurement, very interesting. Would there be some way of comparing standard and Allen deviation on the same data and therefore isolating the noise term from the drift term?

What would be a good way of setting up an Allen deviation measurement in the 0.1-10Hz voltage noise domain? Something like: noise amp -> ADC at say 1 NPLC -> analysis in R? Should we worry about the 1 NPLC integration time not picking up the full P-P data?

p.s. would it be reasonable try and automate the switching in and out of the big input capacitor in most of the pre-amp designs? Or maybe DC couple and null out the DC offset with a DAC - using say a battery as the ref for the DAC i.e. a kind of roll-your-own chopper with a variable null point?

Alan

[Kleinstein]

Using a 1 PLC integration still gets most of the 10 Hz signal. So this will still be able to capture the xx-10 Hz peak-peak noise. This definition also assumes a low pass filter at 10 Hz - so there is some attenuation in this range to. If you really want it exact like definition one might need to adjust this filter a little or use a faster sampling. To really look at noise it the spectrum (from FFT) is often more useful than peak-peak values.

In a certain way (auto-) correlation is similar to the Allan deviation. It both comparing values a slight time apart.

Input coupling for sources with DC offset is a difficult part. And might be limiting, especially below 0.1 Hz or with low noise sources. Switching in and out of capacitors might not really help (there is still current noise of the amplifiers acting on the caps) and cause extra spikes. So I don't think this a really viable path.

Using batteries as a compensation for DC offset is more viable: even with only 1.2 V steps (NiCd Cells) one could use some DC coupled amplification. A DAC might need to be really low noise, so possible, but not easy. For very low frequencies DC coupling is essentially the only option. Higher resolution ADCs could allow measurements even with quite some DC offset at the ADC. So with DC coupling it really might make sense to use something like a 6 digit DMM or 24 Bit ADC, to ease on the DC compensation.

[alanambrose]

I'm having a very nice time reading the new AoE, luckily SWIMBO is pretty relaxed about it. These already raised my interest:

+ dc-coupled auto-nulling amplifier for picking off a small signal riding on a DC level - Fig 5.3 / Page 298
+ op amp noise/frequency curves - Fig 5.54 / Page 338
+ output noise filter for voltage refs - Fig 9.95 / Page 683
+ 'capacitor multiplier' for reducing power supply noise - 8.15.1 / Fig 8.92 / Page 578
+ noise spectrum of common bench power supplies - Fig 8.123 / Page 580

Any thoughts on the 1st item above - the auto-nulling amp?

Alan

[alanambrose]

OK I'm going to keep documenting what I find for others tracking the same path.

+ picking small signals off drifty dc levels is often encountered in bio-medical applications e.g. EEG. There are some clever cmos implementations of very low frequency low pass filters.
+ I didn't realise but Dave has done some teardowns on old Agilent DSAs e.g. 35660A and 35670A.
+ the reason that dynamic signal analysers (i.e. low frequency spectrum analysers) are largely discountinued old boat anchors is that the processing has moved to dsp and pc-based solutions. With these low frequencies (unlike at GHz) an fft can be done very easily on any old processor. e.g. http://www.crystalinstruments.com/dynamic-signal-analyzers/. You could see any old adc plus pc combination with enough dynamic range being used in this fashion (e.g. maybe a 3458A with a preamp).
+ DSAs are used in lots of mech eng applications e.g. vibration, cars, strain gauges, seismic etc
+ alternatives to conventional low noise/low impedance 0.1Hz coupling (e.g. 1,500uF wet tantalum / 1.2K low noise resistor) appear to be:

- switched capacitor filters
- DSP filters
- ADC/processor/DAC methods,

see e.g. this conversation http://www.electronicspoint.com/threads/ultra-low-frequency-low-pass-filters.26609/

[Kleinstein]

If you want to use correlation of two inputs, it is essential to have input amplifiers with low current noise (e.g. JFET) . In this case one can use film capacitors (50 µF range) to rather low frequencies, well below 0.1 Hz. Discrete JFETs (e.g. SK369) can be rather low noise, not much higher than good BJT OPs.

Even if only the 0.1 Hz - 10 Hz noise is of interest, it is a good idea to have the input AC coupling with a much lower frequency limit, with a much higher resistor. The resistor to ground does not directly adds voltage noise. The contribution of the resistor is more a kind of current noise (thus a higher resistor is better). So the initial AC coupling could be something like 20 µF (Film type) and 1-10 M Ohms, with possibly a switch for initially lower values to speed up settling.  The final 0.1 Hz filtering is better done in a later analog stage, or digitally after the ADC. High resolution ADCs have plenty of room for extra LF signal, so it won't matter to have the analog part down to lower frequencies.

The other option I see is rough compensation of the DC offset with a series of batteries and/or a DAC (very low noise !) and than have DC coupled amplification (e.g. 10 times) for the first stage.The DAC to compensate does not have to be high resolution or linear or fast, it's just to keep the input stage from saturation. The main requirement is ultra low noise - so normal DACs won't really work.  If the initial DC coupled amplifier can be build to handle a +-0.8 V signal one could get away with just a string of batteries and relays.  After that the final high pass filter is far less critical, as the signal is low impedance and noise levels can be higher.

I would consider planing with moderate quality amplifier(s) first, and plan with the option to have an external amplifier that fits special needs in case the standard one is not sufficient.

[T3sl4co1l]

If you want to use correlation of two inputs, it is essential to have input amplifiers with low current noise (e.g. JFET)

Why?

How is voltage noise also not a concern?

The lowest noise (as noise figure) amplifiers seem to be BJT, assuming your signal source impedance is low enough to match.  JFETs are most suitable for higher impedances (kohms+).

In principle, you can use a lossless transformer to achieve a constant noise figure for any arbitrary source impedance.  Good transformers are hard to make, so it's usually better to design your circuit around the source impedance in the first place.  This dictates the choice of device: not current or voltage noise alone, but the ratio.

Tim

[Kleinstein]

When doing the correlation type measurement, the voltage noise of the amplifiers still matters, but it can be suppressed by choosing a longer measurement time. The current noise is more critical, as current noise of the two amplifiers will add at the source source resistance and thus produces correlated noise, just like the noise of the source itself. So current noise is  suppressed by correlation.

The second reason current noise is nasty is that with AC coupling capacitors this will cause quite some noise voltage, as the impedance of the coupling caps may not be so low. At 0.1 Hz, even a 1000 µF capacitor has about 1.5 K of impedance.

Voltage noise of BJTs is generally lower than that of FETs. This is especially true for OPs. However this comes with a high current noise and an 1/f component in the current noise. So the low noise of BJTs is usable only for low impedance sources.
As the 1/f corner for the current noise is generally higher than for the voltage noise, this is especially important in the low frequency range. The optimum input impedance can be lower by something like a factor of 10-100 in the LF range.

Good, large area discrete JFETs also can reach noise levels below 1 nV/Sqrt(Hz) and even the 1/f noise is not that bad with good ones. As the current noise with FETs is very low, its a question of price on how many JFETs or how large ones one can use to achieve low voltage noise.

[T3sl4co1l]

I don't get the "cross correlation FFT" either: surely, this simply shows a peak near zero, because the two amplifiers have similar delays?  (Referring to the http://tf.nist.gov/general/pdf/1133.pdf diagram.)  How does that give you a spectrum?

Surely the only useful processing would be to average the signals together, which can be done in analog or digital, and therefore is equivalent to simply connecting the amplifiers in parallel (thus reducing noise as 1/sqrt(N) as usual)?

Tim

[Kleinstein]

The Fourier transformed of the auto-correlation function gives the power spectrum of a signal. With a bandwidth limited noise or 1/f noise, the ACF will be more than just a peak at zero delay, but will show a more or less slow decay and possibly "ringing" for pink noise. For computational reasons one might still compute the ACF via a method using FFT - thus the other way around: doing FFT of the signal, "square" and than FFT back to get the ACF.

With two amplifiers sensing the same signal, one can use cross - correlation to calculate the auto-correlation of the signal. This can suppress the voltage noise of the amplifiers with longer averaging times.  Measuring of the correlation function can be done over long times without the need to store all the data - just the data for the delay window (gives lower frequency limit) need to be stored. So it's not just the 1/sqrt(2) reduction from two amplifiers in parallel, but another reduction from doing one long measurement.  It's more convenient and less prone to drift (e.g. temperature dependence of noise) / nonlinear effects than measuring the noise of the amplifier separately and than subtract it from the noise of the system + amplifier.

[branadic]

AN159 by Linear Technology is worth reading.

www.linear.com/docs/47682

[alanambrose]

Yes v interesting although only >10Hz.

Attached is a link to a pdf of all the LNA schematics I could dig up. I'm planning to build something based on Gerhard Hoffmann's 2014 design to begin with.

Alan

http://anagram.net/nuts/LNA/Noise%20Amplifier%20Designs%20Summary.pdf

[splin]

Yes v interesting although only >10Hz.

Attached is a link to a pdf of all the LNA schematics I could dig up. I'm planning to build something based on Gerhard Hoffmann's 2014 design to begin with.

Alan

http://anagram.net/nuts/LNA/Noise%20Amplifier%20Designs%20Summary.pdf

Don't use Hoffmann's preamp as is - he cocked it up by forgetting, or miscalculating the noise resulting from the opamp input current noise multiplied by the source impedance - especially the input decoupling capacitor. You'd either need to use opamps with much lower input noise current or use a much larger input capacitor. See reply #27 here:

https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/25/

[EDIT] corrected link from email to url

[alanambrose]

OK thanks splin - it's taken me a while to get round to chewing through that Can I ask some dumb-ish questions? - I can't quite replicate your calcs atm - I'm probably missing something...

Firstly, where does the sqrt(20) factor come from?

Also is there some easily available arithmetic which approximates total rms/p-p noise at the 1/f end given a couple of points on the 1/f graph, a bandwidth (or the upper end of the bandwidth anyway), and an assumption that the line is straight-ish? Or alternatively could you get the noise results like this out of spice?

TIA, Alan

p.s. I've added links in the 1st post to voltage ref noise filtering and op amp noise calcs.

>>>
He measured his amplifiers's voltage noise density to be 220 pV rms/sqrt(Hz) at 1kHz, rising to 1nV rms/sqrt(Hz) at .1Hz. The current noise will be rather high though with sqrt(20) x the noise of one op-amp (1/f characteristic – 2.4pA/sqrt(Hz) at 1kHz rising to 10pA/sqrt(Hz) at 1Hz) so only suitable for low impedance sources as he notes in his article. However I believe his amplifier to be seriously flawed – I calculated that the noise voltage generated by the current noise flowing through the 10k//160uF input filter will be around 45nV rms/sqrt(Hz) at 1Hz rising to 800nV at .1Hz!
...
Note that his article has a chart showing the very respectable noise floor with the amplifier input grounded; however the short is applied after his 160uF, 10kohm DC blocking filter so only the voltage noise is shown. The significant current noise can be seen in the same chart showing the noise floor with the input grounded via a 50 ohm resistor – again after the input filter. At .1Hz this is 18nV/sqrt(Hz) compared to 10nV for4 the voltage noise alone; this implies the noise due to the current noise in the 50 ohm resistor is sqrt(18nV^2 – 10nV^2 – 1nV^2), or 14.7nV/sqrt(Hz). The DC filter has a resistance of 7kohms at .1Hz so the current noise totally dominates.

He’s not alone though – looking at the data sheet for the ADA4898-2, on page 14 Analog Devices seem to have made exactly the same mistake! Their test circuit shows an AD743 used to amplify the noise x1000 using an input filter of 1uF/1612kohms!! The current noise of the AD743 will swamp the ADA4898-1’s noise - as shown in their 500nV pk-pk result shown in Figure 45.

The real value .1 to 10Hz AD4898-1 noise I calculate to be around 82nV pk-pk. Since AD generally know what they are doing, this may be a simple mistake showing the wrong values in the schematic etc. but I’m not convinced given that their 500nV pk-pk noise is almost exactly what I calculate it would be from the AD743 current noise spec into the 1uF//1612k input resistance. Don’t believe everything you see in the datasheets.
<<<

[Kleinstein]

The input current noise is exactly why Hoffmanns circuit is not good and only of limited use with very low impedance sources.
At low frequencies the input filter will not be useful - but at higher frequencies (e.g. > 100 Hz)  it can work. The problem is not the 10 K resistor but to small a capacitor.

I can't find the circuit in the AD4898 datasheet (maybe newer version), but usually the amplifier under test is used to do a first amplification (e.g. 100 or 1000 fold). So the following stage is far less critical.  The AD743 is well suited for such a filter with 1 µF and 1.5 M, as it has the low current noise of a FET amplifier: 7 fA at 1.2 M (which is about the impedance at the crossover frequency) give some 8.5 nV /Sqrt(Hz) of noise at that low frequency. At higher frequency it gets less.  Already at 3 times the lower limit the current noise is less critical than the voltage noise. One could have used a larger cap though.

With an amplifier like the AD743 it does make sense to use a few in parallel. It still does not need low impedance sources.

[alanambrose]

OK I think I'm beginning to understand. Below is a spice analysis with an OP27 and a 10mHz (that is 750uF (planned to be 16 x 47uF 25V MLCC) // 20K) HP filter. Gives 17nV RMS total from 10mHz to 10Hz. The application is for voltage ref noise measurement < 10V.

What do you think?

BTW I did run some stats on the AF743 but I have not been able to figure out how to get non-Linear Tech parts into LTspice yet. I'm also thinking that the 10Hz-10MHz and 0.1Hz-10Hz requirements are different enough that they need to be different designs (or at least different front-ends).

Alan

[Kleinstein]

The spice models for many OPs do not include noise. So the noise seems to be from the resistors only.

The OP27 like other BJT based low noise OPs has quite some current noise. This really gets a problem at low frequencies. So it is likely not a good choice here.

So I would think about a difference stage based on discrete JFETs like Sk170, BF862.