DIY low frequency noise meter and some measurement result of voltage references

[splin]

The low frequency noise can vary quite a lot from one sample to another. Its also rather slow an thus expensive to measure. Even with expensive parts like the LTZ1000, there are good ones an bad ones with way more LF noise than typical specs. So it really makes sense to have a system for LF noise measurements.

True but the LTZ1000s are *very* expensive and given that noise is a very important characteristic, a close second to stability, then I don't think it is unreasonable that they should be 100% tested for noise. And since LT specify a maximum of 2uVpp, 1.2uV typical it looks like they have the same opinion. Of course you could still get parts exceeding the maximum but according to TI:

"All data sheet specs are usually obtained using a +/-3 sigma truncation of a typically Gaussian distribution of parts over process variations".

If your application demands parts that are more tightly specced than the datasheet maximum, or 3-sigma probability is not adequate of course you will need to test each part.

The LTC6655 are a lot cheaper, but still relatively expensive, so given the major headline feature is its very low noise, I don't think it would be unreasonable for some sort of noise screening to be performed - even if it were a very quick, and hence cheap, test for HF noise and LF noise at say 10Hz. LT on the other hand (like the vast majority of voltage references from all manufacturers) don't even bother to specify a maximum so in this case they don't agree. One would hope that in reality that the manufacturing process is well enough controlled, along with periodic QA testing, to ensure that the majority of parts do not exceed a reasonable multiple of the typical figure. The problem is what is reasonable? Given Andreas's noise measurements so far I personally would be a bit wary of the LTC6655 - but it is a very small sample.

Besides electronic LF noise, there can also be thermal noise from turbulant air flow and thermal EMF and similar. This can look rather similar to 1/f noise.

True, but Andreas's results for various other references have been roughly in line with the datasheet specifications which suggest that his procedures and test setup are good enough and have the above issues under control. It is just the 6655 results which are unexpected hence the question as to whether they have a particular problem in his setup such as inadequate decoupling, instability/HF oscillations etc.

[David Hess]

The low frequency noise can vary quite a lot from one sample to another. Its also rather slow an thus expensive to measure. Even with expensive parts like the LTZ1000, there are good ones an bad ones with way more LF noise than typical specs. So it really makes sense to have a system for LF noise measurements.

True but the LTZ1000s are *very* expensive and given that noise is a very important characteristic, a close second to stability, then I don't think it is unreasonable that they should be 100% tested for noise. And since LT specify a maximum of 2uVpp, 1.2uV typical it looks like they have the same opinion. Of course you could still get parts exceeding the maximum but according to TI:

"All data sheet specs are usually obtained using a +/-3 sigma truncation of a typically Gaussian distribution of parts over process variations".

If your application demands parts that are more tightly specced than the datasheet maximum, or 3-sigma probability is not adequate of course you will need to test each part.

I remember seeing a note from Linear Technology about contacting them for special noise (or drift?) grading of LTZ1000 references but a quick search did not find it.

Back when popcorn or burst noise was a problem, they could not test for it because it took too long;  I have seen this myself in parts where testing for it would have required hours to days.  Low noise testing for operational amplifiers is also a problem so at least for low cost parts, if they test for it at all they rely on its correlation to high frequency noise which is much faster to test.

The warmup time for the LTZ1000 would require low frequency noise testing to wait 100s of seconds but given the price, I wonder why this is not economical.  Aren't these burned in anyway?

True, but Andreas's results for various other references have been roughly in line with the datasheet specifications which suggest that his procedures and test setup are good enough and have the above issues under control. It is just the 6655 results which are unexpected hence the question as to whether they have a particular problem in his setup such as inadequate decoupling, instability/HF oscillations etc.

The LTC6655 datasheet discusses some pretty strict requirements for output capacitance with a 10uF film capacitor being about optimum but their noise test example uses 1uF which is below the 2.7uF minimum that they recommend.  It is not clear what if any effect this has on low frequency noise.

[Andreas]

The LTC6655 datasheet discusses some pretty strict requirements for output capacitance with a 10uF film capacitor being about optimum but their noise test example uses 1uF which is below the 2.7uF minimum that they recommend.  It is not clear what if any effect this has on low frequency noise.

Hello,

yes I also read the special requirements of special capacitors.
Between the lines: manufactured by virgins with golden hair in a full moon night from Teflon and oxygen free copper.

Edit: Unfortunately I am using only poor mans combination of a 10uV Ta (size A) in parallel with a 100nF 1206 X7R capacitor.

But also the supply voltage has a large influence.
I did a comparison with stabilized voltage of 10.2V and 5.66V at the input of the reference on my sample with the LS8 package.

This gives a factor 1.32 difference when averaging over 10 measurements with 10 seconds each.
10.2V gives 2.717 uVpp
5.66V gives 2.044 uVpp
and also the AC rms voltage (nVeff) has around factor 1.32 difference.

With best regards

Andreas

[David Hess]

The LTC6655 datasheet discusses some pretty strict requirements for output capacitance with a 10uF film capacitor being about optimum but their noise test example uses 1uF which is below the 2.7uF minimum that they recommend.  It is not clear what if any effect this has on low frequency noise.

yes I also read the special requirements of special capacitors.
Between the lines: manufactured by virgins with golden hair in a full moon night from Teflon and oxygen free copper.

Edit: Unfortunately I am using only poor mans combination of a 10uV Ta (size A) in parallel with a 100nF 1206 X7R capacitor.

The requirements did not strike me as quite that bad but film capacitors of that size are annoyingly large.  The low ESR requirement could be met with a polymer aluminum electrolytic or maybe a polymer tantalum capacitor.

I wonder what about the LTC6655 made for such high capacitance and low ESR requirements.  If a normal tantalum and ceramic combination was suitable, I think they would have said so.

[Kleinstein]

I think the combination of two different caps might actually be a good idea. It's a little like with lab power supply circuits: a certain low ESR capacitance is needed to prevent high frequency (e.g. 100 kHz) instability. For the bulk capacitance the very low ESR should is not be really needed and some ESR (e.g. 0.1 Ohms range) might even help. At least this is what the output impedance curve from the data-sheet suggest. It's only a large capacitance with high ESR (e.g. 100 µF with  more than 1 Ohm ESR)  that is more of a trouble.

However I would avoid both tantalum and X7R: the tantalum caps might cause noise spikes, at least some of them do. The X7R can be slightly piezo electric and thus pic up mechanical noise. I would prefer low ESR Al (e.g. 100 µF), maybe polymer and a small film cap (e.g. MKS 220 nF).

Anyway the capacitors should not have that much influence at LF noise - it might make a difference in the kHz range.

[zlymex]

.......
Besides electronic LF noise, there can also be thermal noise from turbulant air flow and thermal EMF and similar. This can look rather similar to 1/f noise.

That's especially true for references with very low LF noise such as 2DW234.
When I first measured this Zener, the noise is higher. The noise only reached to around 0.3uVpp when I wind shielded it with a lot of tissues.

[zlymex]

......
zlymex measured the LTC6655 noise, in reply 2, as approx .33ppm compared to the .25ppm spec, (33% above spec) but I assume that was using the .16Hz - 6.5Hz filter (as calculated by Alex) - the schematic has been updated since but presumably the results are still from the original circuit. I estimate that the bandwidth reduction should reduce the measured noise by about 15% which means the .1 - 10Hz measured noise would be 33%/.85 = 39% above the spec. The .25ppm spec figure is as usual a typical figure so the part still meets the specs.

My measurement of LTC6655 in reply 2 is new.
I used light-blue remarks before(for the .16Hz - 6.5Hz filter), then I changed to purple remarks for the correct bandwidth of  .1 - 10Hz.
Since the result of both LTC6655-1.25 and LTC6655-2.5 are purple color marked, it is the right bandwidth.

[zlymex]

.....
The LTC6655 datasheet discusses some pretty strict requirements for output capacitance with a 10uF film capacitor being about optimum but their noise test example uses 1uF which is below the 2.7uF minimum that they recommend.  It is not clear what if any effect this has on low frequency noise.

The output capacitor of a reference normally has two uses, one is to suppress HF noise, the other is to prevent oscillation. Either way, it has not much to do with LF noise especially below 1Hz.
For function of suppress HF noise, the capacitor can be omitted.
For function of preventing oscillation, the capacitor in the datasheet often has a very large allowance.
For function of preventing oscillation, it needs the ESR of the output capacitor to form a zero(as against pole), therefore, it is usually not "the lower the ESR the better".

[David Hess]

...

For function of preventing oscillation, the capacitor in the datasheet often has a very large allowance.
For function of preventing oscillation, it needs the ESR of the output capacitor to form a zero(as against pole), therefore, it is usually not "the lower the ESR the better".

But in this case it is "the lower the ESR the better" which I found to be very unusual.  What about the LTC6655 requires such a large and low ESR capacitor?

[zlymex]

...

For function of preventing oscillation, the capacitor in the datasheet often has a very large allowance.
For function of preventing oscillation, it needs the ESR of the output capacitor to form a zero(as against pole), therefore, it is usually not "the lower the ESR the better".

But in this case it is "the lower the ESR the better" which I found to be very unusual.  What about the LTC6655 requires such a large and low ESR capacitor?

It perhaps not "the lower the ESR the better" in the case of LTC6655. When Cout=100uF is used, there is a noise peak at about 3Hz which inferior than 10uF where the curve is flat below 30Hz. Presumably the ESR of 100uF capacitor at 3Hz is smaller than 10uF.

[Kleinstein]

The noise peak is at 3 kHz, not 3 Hz - so not really a problem with LF noise.
This noise peak comes from a resonance from the output capacitance and the output impedance of the reference chip. To reduce / dampen the resonance, a higher loss /  ESR (but still not that much) of the capacitor in the kHz range is desirable. However to prevent trouble at higher frequencies  a low ESR at higher frequency is needed. A singe simple capacitor can not provide this, but a combination of a possibly small low ESR cap (e.g. 1 µF foil, ceramic) and a large cap with moderate ESR (e.g. 100 µF with  0.1-0.5 Ohms ESR).

[zlymex]

The noise peak is at 3 kHz, not 3 Hz - so not really a problem with LF noise.
....

Oh yes.

[David Hess]

The noise peaking is exactly what I would expect if the ESR is too low but that still does not answer the question.  What effect on the low frequency noise or other performance characteristic led LT to recommend a large capacitor with such a low ESR?

If a good solid tantalum or aluminum electrolytic capacitor in parallel with a smaller film or ceramic capacitor was suitable, then why didn't they recommend that less expensive solution?

[Edwin G. Pettis]

Ceramic capacitors are noisy, voltage sensitive and have a host of ills, they should not be used in any noise sensitive circuit.  I suggest a review of LT's notes, particularly from Jim Williams on ceramic capacitors.

A short listing of ills:

They are one of the noisiest capacitors made.
They are voltage sensitive, most type's capacitance varies with applied voltage and frequency.
They are mechanically sensitive, generating spurious noise spikes.

While most capacitor types exhibit an inverse noise vs. capacitance curve, most ceramics do not follow this curve and are noisy all over the place.

There are 50V polypropylene capacitors available which are reasonably small, 63V might be slightly more common.  Yes they are more expensive than aluminum or tantalum but do not have spurious noise spikes if that is important to the circuit.

If you have ceramics in your reference circuit, remove them!

[acbern]

This only applies to ferroelectric material capacitors, the use of COG caps is ok for this purpose. Getting SMDs in 0.47uF is no problem. For the higher capacitances, Oscons are recommendable to their supperior AC behaviour.

[zlymex]

Update.
One of my friends(enjoydiy) made some compact noise meters based on this.
Magnification: 10,000
Band: 0.1Hz - 10Hz (-3dB)
Main Apamp: ADA4528-2ARMZ
HPF: 2nd order
LPF: 4th order Butterworth
Input capacitor: Nichicon UKL 1000uV/50V(low leakage type electrolytic)
Power: one 14500 rechargeble Lithium battery(AA size), 3.0-4.2V
Charge port: micro USB
Case: 25*40*83.5mm
Floor noise: 100nVp-p

[lukier]

Very nice compact design. Is your friend planning to sell some by any chance?

[zlymex]

Very nice compact design. Is your friend planning to sell some by any chance?

Those units in the first photo have already divided up among his friends. I didn't hear anything about his selling plan yet.

[Alex Nikitin]

I would buy one as well.

Cheers

Alex

[pelule]

I would buy one as well.

Cheers
PeLuLe

[TiN]

And I would buy 2.
Does your friend plan to release gerbers? I could make boards :-]

[Andreas]

Input capacitor: Nichicon UKL 1000uV/50V(low leakage type electrolytic)

Floor noise: 100nVp-p

Hello,

nice design.
What input impedance do you have?

(seems to be above 3 k-Ohms)

With best regards

Andreas

[zlymex]

Hmmmm...  A new op-amp from Linear Tech might be useful for this task:

LT6081

-Ken

Thanks for the info. This opamp seems to have the same or better voltage noise performance than LME49990 especially below 1Hz. But the  low frequency current noise is worse than that(Unbalanced).

[zlymex]

Input capacitor: Nichicon UKL 1000uV/50V(low leakage type electrolytic)

Floor noise: 100nVp-p

Hello,

nice design.
What input impedance do you have?

(seems to be above 3 k-Ohms)

With best regards

Andreas

This is the input and amp part, plus bandwidth simulation(of the whole meter).

[splin]

If you only care about the DC..10Hz noise, then you could have up to about 10K [total] input resistance without spoiling the 30nVpp noise specs.

I make it < 500 ohms @ 10Hz (2.5pA/rt(Hz) v 1.2nV/rt(Hz)),  < 175ohms @ 1Hz (8pA/rt(Hz) v 1.4nV/rt(Hz)) and < 100ohms @ .1Hz providing you balance the inputs.[/quote]