Ultra Precision Reference LTZ1000

[janaf]

The attached, from that site, is the same, apart from no DC blocking and filtering?

Doesn't specifically have a calculator for this case,

How do you get images inline in the posts here?

[Marco]

I used the first one though, can't even set bandwidth or specify a coupling capacitor in any of the others.

[rf-design]

The LTC2057 is kind of strange due to it's input arrangement.  Usually for a "normal" op-amp, you want to keep the input resistance the same on both inputs to minimize drift.  Well, with the LTC2057, most of the input leakage is due to the switching action of the chopper disciplined amp, and so [up to about 70C] is doesn't help to balance the input resistances.

Could you explain why an impedance balance on the input of a chopper amplifier does not help!

I thought that the chopper switch arrangement internal to the OP is symmetric. So the switch charges should be balanced up to the internal mismatch and the effect on the external network impedance is minimized if for instance the capacitance is equal.

[janaf]

The calculator the way I used it:

1) Specify source signal characteristics. I set it as expected from the LTZ1000 circuit; 0.2uVrms, to 10Hz, pessimistically 1 ohm (it's actually much less).
2) Hit the "Calculate Signal Source Properties"
3) Enter the op-amp data, I used 11nV/sqrtHz and 0.17pA/sqrtHz for both inputs
4) Enter resistor values.
5) Hit the "Calculate..." button.

Result attached. Shows all noise contributions on the output and, among a lot of things, a 15dB SNR. As expected. (1.2uV/0.2uV=15.6dB)

[janaf]

I did measurements now on a new board:

- 2.2uF in
- 1.3M input impedance
- Ri/Rf 1ohm/100K
- Output 2.2uF to 1M in instruments
No other components except decoupling caps.
- Sample rate 49Hz / 24.5Hz bandwidth
- It runs on 2x9V batteries, all in a metal can.

The OP is rated 11nV/sqrtHz
11*srt(24.5) = 54nVrms
The GBW of the amp is 1.5MHz so with gain 100000 it should start dropping off at 15Hz.

I'm measuring around 70nVrms 0.1-24.5Hz, i.e. a bit high as the overhead noise from resisotrs etc should be quite small. With the noise calculation being a bit fuzzy to me.

However, noise is still less than 1/4 of the LTZ1000 circuit noise.

It looks to me like there may be a low frequency problem, air convection? I put the amp in a sock , helped a bit, it was worse before.

[Marco]

Have you sanity checked yet? (ie. use a signal generator with a 1 Mega Ohm/1 Ohm divider to test the transfer function.)

[janaf]

No sanity check with this one. I did it with the previous one with 1000:1 gain. But I really should do it with this one too.

What I just did though; shorted the input on the OP side of the input cap. The measurements I posted an hour ago, the input was shorted on the "outside" of the input cap. Then, the results where a bit high and very sensitive to movements in the room etc.

Now, it's solid as a small rock, the signal LOOKS like white noise, the FFT is flat, the noise is stably at 55nVrms.

[Marco]

What I just did though; shorted the input on the OP side of the input cap.

That's not really fair though, it's not equivalent to having a low impedance source attached.

[janaf]

"Fair", no it's the baseline, for the OP-amp and measurements.

Now I'm measuring with 40K input impedance, not shorted. Then measuring around 68nVrms.

Next: with 40K and 70uF cap shorted. Back in a few minutes.

 

[splin]

Mea culpa - my calculations were wrong - despite checking it multiple times, I still managed to miss out a sqrt(f) term 

I used the (corrected) attached spreadsheet to numerically integrate the noise. Note that the LTC2057 noise current is specced at 150fA/sqrt(Hz) at 5V Vcc - the 170fA figure is for Vcc = 2.7V.

The noise due to the noise current now works out to be 32nV rms to give a total of 44nV so Marco was right. With a 70uF capacitor the current noise becomes negligible at 1nV to give a total of 30nV rms.

If you want the exact integral I reckon it is:

Vn^2 = In^2 x R^2 x Integral[sqrt(f)/(1 + (2 x pi() x R x C x f)^2) df] over the interval .1 to 10 Hz

So the integral is of the form   f^(.5)/(1 + af^2) df which is beyond my maths. However an excellent site for this problem is:

http://www.integral-calculator.com/

I've attached a picture of most of its output. (Note that I adjusted the constant a to scale the graph appropriately).

Splin

[Kleinstein]

For good performance one should not use so much amplification for the single OP amp stage - this will give extra noise, as the input is not close to zero at the chopper frequency or even at 50 Hz / 60 Hz. Even an amplification of something like 100 times should be OK to bring the signal to a level that is easy to handle.

[rf-design]

The LTC2057 is kind of strange due to it's input arrangement.  Usually for a "normal" op-amp, you want to keep the input resistance the same on both inputs to minimize drift.  Well, with the LTC2057, most of the input leakage is due to the switching action of the chopper disciplined amp, and so [up to about 70C] is doesn't help to balance the input resistances.

Could you explain why an impedance balance on the input of a chopper amplifier does not help!

I thought that the chopper switch arrangement internal to the OP is symmetric. So the switch charges should be balanced up to the internal mismatch and the effect on the external network impedance is minimized if for instance the capacitance is equal.

It's all in the data-sheet for this part.  The "normal" behavior that you would expect [where it helps to have balanced resistance on the inputs] doesn't begin until about 70C.  Before that temperature, the total input leakage is dominated by chopper switch action and not op-amp input leakage, and so the direction of the leakage currents is wrong to get cancellation with a balanced input resistance.  The Good News is that up to about 70C, the input leakage current remains relatively flat.  So, the bottom line is for this part, don't worry about balanced input resistances.

Also of note on this part, the input current noise appears to be linked with power supply voltage [charge injection from the switches?].  There is no chart for it, so we don't know if it's linear; and it is only specified at certain fixed rail voltages, but not for a total Vs of 15V [or 12V], which is what most people will use with the LTZ1000(A).

Thank you for the information. I am surprised that the switching-charge based input bias current of the inputs is nearly opposite in polarity. I expect that a textbook chopper switch arrangement generate some common mode bias current which could be common-mode voltage dependend and also have bias current offset which of lower magnitude because of matching.
In this sense the LTC2057 could not optimize the bias current contribution to offset error and noise by matching the impedance. LT possible prefer a different switch arrangement for possible better offset voltage or drift.

[janaf]

Not sure I follow you. You mean the ripple at mains frequency and switching frequency would fold down into the low frequency? It does not seem to be the case as the GBW and gain limit the bandwidth to 15Hz.

But sure I can decrease the gain, but LTspice tells me that there is a gain range approaching instability without capacitance over Rfb. I was considering that as a sanity check; just decrease gain ratio by increasing the Rin to 10K. That would also be a sanity check for input noise as Rin would give significant noise contribution, a raise by sqrt (10.000) = to the contribution by Rin.

For good performance one should not use so much amplification for the single OP amp stage - this will give extra noise, as the input is not close to zero at the chopper frequency or even at 50 Hz / 60 Hz. Even an amplification of something like 100 times should be OK to bring the signal to a level that is easy to handle.

[janaf]

Moderator: Are we steering off-topic? Start a new thread on LF noise and move some posts there?

[janaf]

I got a PM on other stuff and a comment about the 7-digit realm this stuff is in.

No pun at all to the poster but for noise, I'd be happy if I can establish it accurately with two digits, maybe even one digit accuracy.

It's not important to me to establish if the 0.1-10Hz noise is 49.8 or 49.7nVrms. It be excellent to know with some certainty if it's 49 or 50nVrms, two digits, probably quite enough for me if I know with some certainty if its 40 or 50 or 60nVrms, ie one digit.

With the right amplifier and an 8-bit ADC one could do these measurements with enough accuracy. Amp gain accuracy of 1% is enough. 
 
I looked at the LT1028 and it seems (datasheet and LTSpice) it's noise can actually be lower than the LTC2057, all the way down to 0.1Hz. I'll give it a try as I still have problems with the LTC2057 below 1Hz, when increasing input impedance and adding DC blocking. Makes no reals sense.

[paulie]

How do you get images inline in the posts here?

If you mean expanded image mixed in with the text here's one way:

1. Click on the thumbnail to expand it then right click and select image info.
2. Cut and paste the highlighted part (or an image URL elsewhere) into your post.
3. Add  [ img] and [ /img] tags to the HTML.

There's probably an easier way like some other sites have but this only takes a few seconds with practice and works on virtually all forums.

ps. What I would like to know is what do those yellow squares mean in everybodies avatar? They seem to increase and decrease  randomly.

[paulie]

With the right amplifier and an 8-bit ADC one could do these measurements with enough accuracy. Amp gain accuracy of 1% is enough. 

Funny you mention that. My own voltage reference logger uses one of the low res AVR analog inputs fed from the peak detector fed from the noise amp fed from the 10x reference amp:



No point wasting one of the 6 digit meter channels on noise because high precision not required and noise only once every 65536 voltage samples. I scaled the total gain so that full scale about 4x my worst voltage reference noise (one of the penny zeners) hoping no future devices would be much more than that. Surprisingly noise from bandgaps did not seem to be a lot higher than the zeners including my LM399 as previously suggested. But the amp noise was about 1/10th that of quietest reference and everything fits quite nicely into an 8 bit byte.

See? Inline image EASY!

[janaf]

I'm stuck. Defeated by low frequency noise.

DC coupling, no problem. Nice and stable, easy to measure, measurement match spice and datasheet. High gains, not a problem either, within the usual stability limits & cures. Fore example the LT1028 is stable with 10nF over the Rfb, it has a phase margin of 60 deg.

But with A/C coupling it all goes sour, i.e. follows the theory. My bad, I did not quite have a grip, still don't entirely have one.
 
The bottom line; with A/C coupling, the input impedance noise source is equivalent to the Rin noise. I think.

To get the kind of noise levels I need to compare to LTZ1000 with some SNR, Rin should preferably be no more than 100ohm-ish. Matching input impedance, we end up with a ginormously humongous input cap. A 10.000uF cap + 150ohm would give a 0.10Hz cutoff but I suspect that cap leakage current would have the offset going all over the place.... Verifying things take an awful time. Maybe 1000uF +150ohm and just measure from 1Hz...

Actually, with the LTC2057, the zero correcting did work decently even with the wrong input impedance. I guess nulling took care of some.... but the output signal was weird, probably due to the zeroing going on. In all, I could produce some useful results, even if the noise was higher than expected, SNR was fairly poor.

I'm depressed. I need a new approach. 

One good approach to problems is to ignore them, estimate noise, move on with the LTZ1000

The reason for the noise side-track was the R2, increasing it to 5.xMeg nulled the temperature sensitivity and my crude noise measurements could not detect any noise increase which caused some discussions....

[Marco]

Not sure, if I agree with your diagnosis ... but if you want you can DC couple it. Just throw a big capacitor in series with Rg, and add a switch from the input to the capacitor with some resistance to quickly charge it.

[janaf]

Not sure if I agree either :-)

Problem 1) With DC coupling, I can have a gain of 1 or a little more.
Problem 2) With AC coupling, I get the noise down to uV level because of current noise. With JFET, current noise s low enough but voltage noise and 1/f noise is prohibitive? Modern choppers seem to be best compromise so far.

It also made me remember this AN by Jim Williams on the subject, in this case measurig noise of the 6655 reference, which is almost as low as the LTZ1000....

Title: 775 Nanovolt Noise Measurement for A Low Noise Voltage Reference
Subtitle: Quantifying Silence

http://cds.linear.com/docs/en/application-note/an124f.pdf

[Kleinstein]

The LT1028 is good for very low impedance sources (e.g. 50 Ohms), but just the wrong chip after an AC coupling capacitor. There are several options for lower noise at higher Impedance so that the AC coupling capacitor stays reasonable. Still you have to take care of things like thermal EMF and turbulances: so cover it.

A more conventional option would be a classical low noise OP like a OP27 or LT1037 possibly also LT1013 or OP177 if smaller caps are used. The noise specifications are better than the LTZ1000. They need a resonable size AC coupling cap (e.g. a few hundred µC). Not all electrolytic types will work and they may need some time (e.g. 12 h) of settling before leakage currents (and thus noise) come down. A simple short as a ref. may not work because of missing polarization - one may have to uses something like a 3 or 9 V battery as a low noise Ref.

An auto zero Amp like the LT2057 should work too - still the coupling capacitor should not be to small, as there is some bias and current noise. It also needs some care not to get to high current during power up or when connecting to a source. The Auto zero Amp may also be more sensitive to EMI - so good shielding is needed. A very high amplification and thus a low BW may also cause trouble.

Another option may be using some batteries in series to bing down the voltage from the ref., than have a DC coupled amplification (e.g. 10 fold with an OP27) and have AC Coupling only after this.

[Marco]

Problem 1) With DC coupling, I can have a gain of 1 or a little more.

As I said, you can just put a big cap in series with Rg (resistor from negative input to ground). If you make Rg 1 kOhm (just 13 nV RMS to 10 Hz) then for a 0.1 Hz "-3 dB" (it's not really a first order filter, but close enough) point you need 1.66 mF ... that's not really a problem, 8 220 uF lytics in parallel for instance. The only problem is that it will take a huge time to stabilize after turn on, so you need to speed that up a bit.

PS. hmmm, leakage might be a bit of an issue with electrolytics I guess.

[janaf]

Marco, you mean A/C couple the inverting side to ground and DC-couple the non-inverting? Interesting!

[janaf]

Another option may be using some batteries in series to bing down the voltage from the ref., than have a DC coupled amplification (e.g. 10 fold with an OP27) and have AC Coupling only after this.

I like the battery idea :-) But wouldn't batteries generate noise, even if current is virtually zero?

[Edwin G. Pettis]

Yes batteries produce noise by the chemical reaction, while they can be 'cleaner' than some power supply sources, they are as often not.  They are very good for isolation of course if that is needed.

That tantalum capacitor that Jim used in his ap note is still available for a rather large amount, the last time I priced it it was around $1.100,00 and a special order.  I have found large polypropylene capacitors on ebay, a 1400uF once for a very reasonable price, leakage is probably as good or better than the tantalum but you are still looking at least 24 hours settling time for a measurement.

By the way, an oscilloscope is the only accurate method to measure P-P noise in real time, this is how Linear Tech and everybody else does it.  Using an ADC integrates the noise which produces an inaccurate result as the integral of gaussian noise is constantly changing and does not repeat itself.  The data sheets for these devices all use scopes for the measurements.  It is very easy to verify this, call your LT rep and ask, he'll (or she'll) be happy to tell you all about it.