Ultra Precision Reference LTZ1000

[Andreas]

Hello,

just a quick and dirty simulation of the proposal:

with best regards

Andreas

[janaf]

I attach a plot of measured output noise levels for the whole circuit. The circuit is a plain vanilla datasheet schematic but uses LTC2057 opamps.

I have varied R1 from 80R to 1000R, equivalent to between 7.4 mA and 0.6mA

The values are normalized to R1=120R as my absolute noise levels are a bit different from the LT datasheet, of a few reasons. I will get back with more when I have looked into the absolute values.

The measurements where done with a noise amp based on a LTC2057 with a gain of 1000, 1:st order HP at 0.1Hz and LP at 10Hz, followed by a 4th order LP at 10Hz, gain 1. This was enough to mains power 50Hz noise well below DUT noise.

Sampling was done in 20s blocks, 1kHz with a 24-bit ADC, data was numerically bandpass filtered at 0.1Hz and 10Hz.

Result: Noise varies linearly in a log-log plot, see attached pdf for values.

I will get back with more about the absolute noise level and noise for high R2 values later.

[Andreas]

The measurements where done with a noise amp based on a LTC2057 with a gain of 1000, 1:st order HP at 0.1Hz and LP at 10Hz, followed by a 4th order LP at 10Hz, gain 1. This was enough to mains power 50Hz noise well below DUT noise.

Sampling was done in 20s blocks, 1kHz with a 24-bit ADC, data was numerically bandpass filtered at 0.1Hz and 10Hz.

Hello,

Which ADC are you using for this?
When having the numerical filtering at 0.1 + 10 Hz and the external band pass at 0.1 + 10 Hz then you will have -6db amplification at the corner frequencies instead of -3 db. Right?
(This will play no role in comparative measurements but possibly when trying to get absolute values).

With best regards

Andreas

[Galaxyrise]

The perfect ratio would be adding 7.15 * 2 / 5 to the already existing 7.15 V giving 10.01V

For 7.2, I was able to get the perfect 1.388 ratio (25/18) using 3 ICs.  For my LM399 rig (7.06879V) I came up with 1.4166 (17/12).  The best I can do with just 2's and 3's for 7.15 is 1.40625 (13/32), which is not as good as your proposal. But I worry about splitting a single division stage across two LTC1043 modules as I think differences in when they switch will show up on the output as more noise.  I was already planning to drive them with an external clock, at least.

[janaf]

I'm using a PXI-4461 module. It has an adjustable pre-amp, I'm using the +/-0.3V range. Don't know the actual ADC inside. I could also use the DMM, think I can reach > 20Hz at 6.5 digits which is enough after the noise signal has been amplified.

Filter; I have to look into both analog and digital filtering more. That's why I'm not prepared to give absolute levels.

For example, the PXI-4461 has a built in DC-blocking which has a cut-off in the order of 1Hz, so it gives too low readings, by about 25%. When I disabled it I had problems with the grounding, the signal was drifting, giving a too high reading. The software filter; will have to dig into that too. It's very steep, brick-wall type.

Making measurements down to 0.1Hz and preferably an order of magnitude below, is very time consuming. Measurements in at 100nVrms level is challenging. So each step needs careful double-checking. I want to set up one or several signal generator that I can verify somehow, and test different ways, also at uV signal levels and something like 0.01Hz to 100Hz.

In the end, the actual noise number that comes out will be much influenced by what filtering is used. Any standards?

In any case, I can post a dump of an averaged amplitude FFT tomorrow night. Gives you an idea of the total filtering.

[Andreas]

  I was already planning to drive them with an external clock, at least.

Yes you are right:
my proposal will only work with synchronous clocks for the 3 LTC1043.

Making measurements down to 0.1Hz and preferably an order of magnitude below, is very time consuming. Measurements in at 100nVrms level is challenging.

In the end, the actual noise number that comes out will be much influenced by what filtering is used. Any standards?

True words.
Its not a easy task to amplify the 0.1 Hz to 10 Hz noise to a level where you can measure it.
You have either low noise amplifiers with large offset requiring several stages with cascaded high passes (higher order).
Or you have low offset amplifiers with high noise.
So the LTC2057 is a good compromise.

There are many different approaches for measurement so no real standard.
Some ideas might be in AN83/AN124 of Linear Technology.

I do not think that it makes sense to measure below 0.1 Hz. (10 seconds minimum measurement time).
At 0.01 Hz you will need a minimum measurement duration of 100 seconds.
If you do not have very stable environment conditions you will see temperature drifts within noise measurement.

With best regards

Andreas

[janaf]

I have looked through the whole signal path, sorted out details and re-done measurements.

Now the absolute reading I get for R1=120R is 0.213uVrms which by rule of thumb, multiplied by a factor 6 is the ptp value, in this case 1.28uVptp for the plain vanilla LTZ1000ACH circuit.

I also checked a few other current settings, R1 values, and the ratios in the diagram I posted the other day are repeatable. For example, with 249R for R1, I get 0.28uVrms noise. With the ratio from the other day, 1.33 for 249 ohm, times today's 0.213uVrms = 0.283uVrms.

The most important change from the other day was to get rid of the AC coupling for the ADC, replacing the cap with an external 50x larger value (2.2uF instead of the internal 0.047uF) which with an input impedance of 1Mohm gives a -3dB HP at 0.07Hz.

The second thing was that I changed the sample rate down to 32Hz. The instrument manual said the minimum is 1000Hz (set by a DDS) but when I query the instrument, it replies that the valid sample rate range is down to 32Hz. As it's a delta-sigma converter, now that I sample at 32Hz, the mains frequency is gone, no need for analogue LP filtering. I sill added 0.1-10Hz brick-wall filtering. Without it, the noise was about 25% higher, as expected.

Now to the strange and hopefully very positive part. Setting R2 to 5.6Mohm, did NOT increase the noise! As far as I can see.

I have checked everything I can. The instrument noise is more than 2 orders of magnitude lower than the measured value. The OP noise should be a factor 6 lower than the measured value (0.2uVptp 0.1-10Hz). The signal "looks" OK, no clipping or abnormal offset. The voltage from the circuit is the same as the other day, 6.999xxx V with the 5.6M setting, currents look OK etc.

Can someone confirm this experimentally? Is there a theoretical explanation? Could the noise cancel? Or did I simply make some lousy mistake?

The best would be if someone could confirm / reject this experimentally. IE: throw in a 5.6M resistor for R2 instead of the nominal 68K. And measure noise for both R2 values.

Of course, the problem of getting ultra stable resistors at 5+ meg remains.

[Marco]

Its not a easy task to amplify the 0.1 Hz to 10 Hz noise to a level where you can measure it.
You have either low noise amplifiers with large offset requiring several stages with cascaded high passes (higher order).
Or you have low offset amplifiers with high noise.
So the LTC2057 is a good compromise.

That's an amazingly nice hammer, I don't think there is a better all around hammer in the world right now.

AFAICS a simple low noise JFET common source amplifier even without chopping will outperform it in this case though.

[Andreas]

The second thing was that I changed the sample rate down to 32Hz.
As it's a delta-sigma converter, now that I sample at 32Hz, the mains frequency is gone, no need for analogue LP filtering. I sill added 0.1-10Hz brick-wall filtering. Without it, the noise was about 25% higher, as expected.

I have checked everything I can. The instrument noise is more than 2 orders of magnitude lower than the measured value. The OP noise should be a factor 6 lower than the measured value (0.2uVptp 0.1-10Hz). The signal "looks" OK, no clipping or abnormal offset. The voltage from the circuit is the same as the other day, 6.999xxx V with the 5.6M setting, currents look OK etc.

Is there a theoretical explanation? Could the noise cancel? Or did I simply make some lousy mistake?

Hello Jan,

I am not really shure if I understand what you are really doing. (a schematics of the setup/(filter) amplifier would be nice).
The first what I would do is measuring noise floor of the amplifier (with a 7.2 NiMh battery pack of AA cells or similar instead of the reference).
The 0.2uVpp is only the voltage noise of the LTC2057. Depending on your input cirquit/impedance there will add some current noise.

No filtering: I think you will need at least a high pass at the input of the amplifier.
I do not see how 32 Hz sample rate cancels out line frequencies (50 Hz?).
The sample rate has to be usually a exact multiple of 20 ms for this. (Might also be more dependant on the order of sigma delta modulator).

I guess in the .pdf the 22nV rms should read as 220nV rms (2*10e-7)
Are the rms values measured as true rms signal or calculated as RMS?
Is the raw signal a peak to peak (DC) measurement?

with best regards

Andreas

[janaf]

Hi,

OK, I'll make a "complete" walkthrough of hardware and software tomorrow. 

[janaf]

First step. Her's the pre-amp schematics (attached).

From input to output:
- DC blocking / 1st order HP at 0.07Hz
- Input amp with gain 1000 and first order LP at 10Hz
- Two stages with each, unity gain, 2:nd order LP at 10Hz
- Output AC coupling, again 0.07Hz first order HP

[Andreas]

Hello,

Is R9 including the input impedance of your measurement system?
Or is there another pull down (limiting the lower band width).

The upper -3dB bandwith is most probably more on the 4 Hz than the 10 Hz.

And with 1Meg input impedance the LTC2057 alone (without the resistor noise)
has a factor 15 more current noise than voltage noise (at least at 1 kHz).

With best regards

Andreas

[janaf]

Yes, R9 is the input of the instrument.
And yes, I can see on the measurements that it starts dropping off at 3-4Hz.

As-is, it looks like noise is approximately 50nVrms, input referred. Will be back with more about the measurements.

[janaf]

-3dB; dumb of me to just add the sections, all with the same -3dB. Have to change some component values....

Current noise; how did you calculate that? Remedy?

Too tired now. Searching for hours for something causing glitches now and then. It turned out to be a BNC connector...

Hello,

Is R9 including the input impedance of your measurement system?
Or is there another pull down (limiting the lower band width).

The upper -3dB bandwith is most probably more on the 4 Hz than the 10 Hz.

And with 1Meg input impedance the LTC2057 alone (without the resistor noise)
has a factor 15 more current noise than voltage noise (at least at 1 kHz).

With best regards

Andreas

[Andreas]

Current noise; how did you calculate that? Remedy?

Hello,

the datasheet parameter is 170 fA / sqrt(Hz)
* 1 Meg = 170 nV / sqrt(Hz) (neglecting the 72 Ohms at the negative input).
this compares to the voltage noise with 11 nV / sqrt(Hz)

See also figure G18 / G19 in datasheet.

So the LTC2057 is only useful for input impedances (sum of + and - input) below around 50kOhm.

Noise measurements around 1uVpp are not so easy to get reliable.
It needed me 5 tries and several months with some help until I had a working cirquit.

Battery supply for the amplifier and the DUT together with a cookie box also helps a lot.

With best regards

Andreas

[Marco]

The current noise develops a voltage across the input impedance in parallel with the source impedance, so input impedance should be irrelevant here ... only source impedance is relevant (with 72 Ohms on the other input it won't generate much voltage there).

[janaf]

Thanks Marco, that was how I saw it too. The input impedance is not the source impedance...

[splin]

Thanks Marco, that was how I saw it too. The input impedance is not the source impedance...

No, but the source impedance is still significant in this cct:

Z = sqrt((Vref output resistance)^2 + (1/2/pi()/f/2.2uF)^2)) // 1M

Ignoring the Vref o/p resistance which should be negligible in this context, I calculate Zsource to be approx 72k at 1Hz and 586kHz at .1Hz

The LTC2057 datasheet shows the input current noise spectrum is flat down to .1Hz at 150fA/sqrt(Hz) with Vcc of +/-15V. So noise due to input noise current will be approx 11nV at 1Hz rising to 88nV at .1 Hz

Looking at page 528 of my copy of the 3rd ed. of the Art of Electronics which just arrived, the AD8628 might be a better choice with voltage and current noise densities of approx 21nV and 50fA respectively at .1Hz (with no 1/f noise).

But this is all academic since the circuit is already built and its own noise is easy to measure by shorting the input - perhaps I have missed the results of this test?

Splin

[Marco]

Ehh, okay you're right ... but it drops off pretty fast. The AD8628 starts off with 500 nVpp 0.1 to 10. AFAICS you won't get close that with the LTC2057 even with the current noise.

You could also use a larger input cap on the LTC2057 with a smaller amplification, the RC time gets large, but without too much amplification it will be below clipping pretty fast and the subsequent high pass filter can get rid of the rest.

PS. if you want really low noise you'll need to go discrete of course.

[janaf]

Thanks all for the feedback and the interest! Lots of things to reply to.

First, the noise amp was for measuring the noise from the LTZ1000 circuit, which is expected to be in the order of 200nVrms 0.1-10Hz. (ie not measuring another LTC2057  ) I choose that amp because of the very low, low frequency noise, in the order of 33nVrms 0.1-10Hz. Honestly, did not give the current noise much thought because the ltz1000 circuit has a low impedance output. There are other chopper amps with lower input bias current but then of course with higher input noise. I can surely try lower input impedance as I have some big fat foil caps on the shelf. Have a 70uF polyprop, big as my fist. It's not a permanent setup anyway, was for comparing different parameters in the LTZ1000 circuit.

Even if it would be interesting, no need to go to discrete JFET input stage. I'm de-touring from the LTZ1000 all the time anyway. But in longer term I'm interested in looking at discrete JFET input stage. 

The LP filter stuff; I stared with this before realizing that the ADC I'm using can sample at rates well below the mains frequency. I'ts a Delta-Sigma so brick-wall LP anti-aliasing filtering comes in the box, I don't need to add a steep analogue LP, can chuck that out. (according to all docs, the ADC supports sample rates down to 1kHz but actually samples at rated down to 32Hz. Tested, works, attenuates mains frequency by at least 100dB . Someone lese may elaborate on how & why DS-ADCs effectively block higher frequencies without aliasing problems... I made a small paper on results only. Attached. Comments welcome.

So right now it looks ike: Larger input cap, lower input impedance, get rid of the 2nd order LP, sample at low frequency, make sure the ADC brick-walls mains, do some additional band pass filtering in software....

Yes, the most sensitive measurements are done with battery power, in the Universal Shielding Device, aka cookie box.

[MK]

try this link for a low noise amp...
users.cosylab.com/~msekoranja/tmp/04447683.pdf

also, on page 5:
www.janascard.cz/PDF/Design%20of%20ultra%20low%20noise%20amplifiers.pdf

[splin]

Ehh, okay you're right ... but it drops off pretty fast. The AD8628 starts off with 500 nVpp 0.1 to 10. AFAICS you won't get close that with the LTC2057 even with the current noise.

You could also use a larger input cap on the LTC2057 with a smaller amplification, the RC time gets large, but without too much amplification it will be below clipping pretty fast and the subsequent high pass filter can get rid of the rest.

PS. if you want really low noise you'll need to go discrete of course.

It drops off with frequency but not fast enough - I calculated the .1 to 10Hz RMS input referred noise with Rin = (2.2uF//1Mohm) to be:

LTC2057 Vn: 30nV, In x Rin: 339nV, Total: 340nV rms.
AD8628   Vn: 76nV, In x Rin: 113nV, Total: 136nV

So nearly 3x less noise. Except it turns out that the AD8628 datasheet spec for the input noise density is actually 5fA/root(Hz) not the 50fA shown in the AoE*, giving:

AD8628  Vn: 75nV, In x Rin: 11nV, Total: 76nV

So much better than the LTC2057.  The LTZ1000 noise spec is 1.2uV p-p or 181nV rms, so even 76 nV isn't brilliant, but probably good enough for this purpose. The LTC2057's 340nV is way too high but increasing the input capacitor to 11uF (5 x  2.2uF) and reducing the 1M to 200k ohms should reduce the LTC2057 input noise to approx 74nV rms.

Have a 70uF polyprop, big as my fist.

70uF/31k should reduce the total input noise further to aprox 32nV.

Splin

* Inexcusable - they've had 26 years to get it right (Ok, the AD8628 hasn't been round that long)! 

[Marco]

LTC2057 Vn: 30nV, In x Rin: 339nV, Total: 340nV rms.

Could you show me the integral? I know I shouldn't trust Spice if I can't do the math, but it still puts a doubt in my head ... with a 100 sec simulation with 100u max integration step and noise intervals I get nowhere near that (needed to do a bit of experimentation to get the noise spectrum for the current source right'ish). That's with resistor noise and significant noise energy below 100 mHz.

[janaf]

I found a opamp noise calculator that I like, it makes it easier to understand the basics and tinker with values.
http://www.dicks-website.eu/noisecalculator/index.html

[Marco]

Doesn't specifically have a calculator for this case, I have seen it said that for first order filters you should take noise bandwidth as 1.57 the -3 dB point (in which case the calculator gives me ~45 nV RMS) ... but not being able to do the math any more, I'm not sure if it's valid here