Ultra Precision Reference LTZ1000

[MegaVolt]

LTZ at -260°C ? Coming soon  .

What for?

[BU508A]

LTZ at -260°C ? Coming soon  .

What for?

Because it is much cooler.

[MegaVolt]

Because it is much cooler.

Cool and then hit with a hammer )))

[exe]

LTZ at -260°C ? Coming soon  .

I guess internal heater should be disabled...

[Dr. Frank]

LTZ at -260°C ? Coming soon  .

Hello Illya,
Below about -15°C the LTZ1000 chip will pick up kind of non reversible hysteresis / shift / damage of its output voltage.
Even the 7000 can not handle this state any more by its de-'gaussing' method.

Therefore I'm curious, what's your intention (@liquid/pumped H₂?) ... lower noise?

When I see this cryogenic apparatus, have you meanwhile bought a JJA, indeed?
 
Frank   

[Andreas]

LTZ at -260°C ? Coming soon

and do not forget to measure 1/f noise and wideband noise.

with best regards

Andreas

[Andreas]

By the way:

I measured wideband noise on my buffered LTZ´s.
Mainly to see wether the EMI-capacitors C16-C19 have a bad influence on wideband noise.
(transferring some voltage regulator noise via the shield to the output of the LTZ).

I first thought that my AN83 LNA is defective because the noise was really low.
But in comparison with the other LTZ´s the trend is that my LTZ1000 based references with ADA4522 buffer have significantly lower noise than my LTZ1000A based references with LTC2057 output buffer. So who is the guilty: The LTZ variant (where I would estimate around 5-10% lower noise for the non-A Variant) or the buffer?

Any other experiences?

The 1/f noise has no correlation to A or non-A type.
And since my first 1/f noise measurements LTZ#3 now has increased 1/f jumps (some kind of very short popcorn noise).

with best regards

Andreas

[Kleinstein]

The measured noise is quite a bit lower than the white noise specs in the LTZ1000 datasheet (some 40 nV/sqrt(Hz) in the graph for 4 mA zener current).

8 µV RMS for the 100 kHz BW corresponds to some 25 nV/sqrt(Hz) white noise.

If there is an extra buffer, one could easily add some higher frequency filtering it does not take that much in the kHz range. So the acural reference noise may not be that relevant. The buffer amplifier noise should be lowert than the measurend noise. However some of the AZ OP can show higher noise when used as a simple buffer (more higher frequency spikes going in to out and the other way round). The noise specs for the OPs are usually with a higher gain (e.g. > 10). So it may be worth to just check the noise of the buffers alone.

The higher frequency reference noise can effect some multislope or similar ADCs, as it can get mixed back to near DC and for some odd reason some DMMs don't include much filtering there, though it would be quite simple. Most other uses should not really care much about the higher freuquency noise.

[Andreas]

Hello,

I finally managed to summarize also some 1/f (0.1 .. 10 Hz) noise measurements.
As already mentioned there is no correlation to -A or non A type of LTZ.

I usually do 100 second measurements to catch more popcorn noise events.
But of course these have a little higher peak-peak noise than the standard 10 second measurements since more "rare" noise amplitudes are recorded.

LTZ#3 has moved into a "little jumper" over the time.
Where I have no explanation for it.
But also at the beginning the noise was not very low.

According to my calibration measurements LTZ#4 is the device with lowest drift over time.
Interestingly it is also one of the lower noise devices.

with best regards

Andreas

[SigurdR]

Interesting results!

Noise is actually increasing over the 5-year span.

Or is it the noise of the measurement apparatus that increases, I wonder, but I guess a reference measuurement is made to verify this is not the case.

[Andreas]

Noise is actually increasing over the 5-year span.

Or is it the noise of the measurement apparatus that increases, I wonder, but I guess a reference measuurement is made to verify this is not the case.

Mhm,

I have looked up the LTZ4 measurement:

Actually there is one change in measurement evaluation.
2016 I have generally used resolution enhancement from 16 to 20 Bit
instead of now (2021) using a 1 kHz FIR filter to reduce the oscilloscope bandwidth and noise.
The 20 Bit filtering effectively gives together with 50 kHz sample rate a low pass frequency somewhat below 200 Hz instead of 1 kHz.
But in average this reduces only the Vpp values by ~10nV so around 1%.
Another reason is that there is one "outlier" with 1.77uVpp in the 2021 measurements.

The other thing that I have changed is a EMI capacitor 100nF between output ground and power supply ground of the LTC2057 buffer on LTZ#3 to LTZ#6. I will to have check if there is a influence.

On the other side: Noise is statistical. You never get the same results so 0.1-0.2uVpp difference in 1.2uVpp values is not a "significant change".

The noise floor is of course checked before the measurement series started. I usually check that it is below 0.2 uVpp (leakage current of the input capacitor low enough with a 7-10V NiMh battery) and that the 50 Hz mains hum is not significantly high. (no mains transformer too near to the cookies box).

with best regards

Andreas

[SilverSolder]

Don't forget it is important to replace the cookie box frequently, necessitating a statistically significant consumption of cookies!   

[SigurdR]

You are thorough!

Checking the noise floor and calibrate out that level is a good thing to do.
When we calibrate our DVM for ex, we use Voltage/etc references. Is there such a thing as a noise reference? a noise reference that gives out a lot of noise specified within a given band width so we can use that to check our own noise measurement gear?

Indeed, noise is statistical in nature, and 0,1-0,2uV for a 1,2uVp-p is almost 10-20%. When I do my 20-20k noise measurements I am happy to be +-10% "correct". Lately, I try to do FFT noise measurements downto 0.1Hz, and it is not trivial to get accurate results!


Kind regards,
Sigurd

Noise is actually increasing over the 5-year span.

Or is it the noise of the measurement apparatus that increases, I wonder, but I guess a reference measuurement is made to verify this is not the case.

Mhm,

I have looked up the LTZ4 measurement:

Actually there is one change in measurement evaluation.
2016 I have generally used resolution enhancement from 16 to 20 Bit
instead of now (2021) using a 1 kHz FIR filter to reduce the oscilloscope bandwidth and noise.
The 20 Bit filtering effectively gives together with 50 kHz sample rate a low pass frequency somewhat below 200 Hz instead of 1 kHz.
But in average this reduces only the Vpp values by ~10nV so around 1%.
Another reason is that there is one "outlier" with 1.77uVpp in the 2021 measurements.

The other thing that I have changed is a EMI capacitor 100nF between output ground and power supply ground of the LTC2057 buffer on LTZ#3 to LTZ#6. I will to have check if there is a influence.

On the other side: Noise is statistical. You never get the same results so 0.1-0.2uVpp difference in 1.2uVpp values is not a "significant change".

The noise floor is of course checked before the measurement series started. I usually check that it is below 0.2 uVpp (leakage current of the input capacitor low enough with a 7-10V NiMh battery) and that the 50 Hz mains hum is not significantly high. (no mains transformer too near to the cookies box).

with best regards

Andreas

[Kleinstein]

I resistor is a kind of noise reference  60 ohms gives 1 nV/sqrt(Hz). There is however still the difficulty that any amplifier will have its own noise and usually also current noise.
The other usual option is to us a resitor or similar only as a white noise source to check the frequency band and than use a sine or similar signal to test the actual voltage gain.

There are some noise sources thoug the stability is often not so great to call it a reference and most are for the RF range. Some AWGs can generate pseudorandom noise that may be good enough, but the quality varies.

For the scattering character of noise, the directly measured peak to peak values are by nature scattering a lot. The RMS reading is usually much less scattering.
For getting the noise from a FFT it is normal that there is quite some scattering for the lower decade or so. So to get good noise reading down to 0.1 Hz it take more like 100-100 seconds, so one has averaging over several 10 scond windows.

[niner_007]

I took a little bit of a different take on my implementation of an LTZ1000 module. Instead of focusing so much on the 3458A reference module form factor, I do want a smaller module, as long as performance is not impacted, but size and form factor are unrestricted. Much inspiration and learning for the layout of TiN's FX ref, Dr. Frank's implementation, and the CERN ADC, excellent and invaluable information folks! Not done yet, but I realized I got lots of intuition on how to implement a better 3458A reference module as well from this. So I'll probably spawn a few form factors. Not done yet, and far from final layout, I'm trying different layouts.

[TiN]

Using VHP resistors for 70kOhm bias resistors is waste of money. Also ditch the slots
Suggest you to make smaller board, not bigger, it will be helpful for performance of the module, if the placement is good.

[niner_007]

Using VHP resistors for 70kOhm bias resistors is waste of money. Also ditch the slots
Suggest you to make smaller board, not bigger, it will be helpful for performance of the module, if the placement is good.

Datron loved the slots though, and they look cool

The 70K resistors (and 120 resistor) are actually VHP101, these are the best resistors VPG can make, with a max absolute TCR of 0.4ppm within a 20C to 30C window (binned), you can get these binned from Vishay, but I am using the garden variety ones, so they are hardly fancy, just a mere 2ppm/C or 5ppm/C, don't remember. The VHD200 divider isn't your Digikey resistor order, it's binned by Vishay and comes from Vishay with a guaranteed maximum tracking TCR of 0.5ppm within a 20C to 30C window, anything higher than that they sell to digikey Part number is 303391 if anyone wants to order, for a cool $90 per resistor.

Just to sit down for a bit and think, super cool that Vishay can produce these, bin them and guarantee the spec, not your datasheet spec or marketing talk, or typical figures talk, where you never know what you'll be getting. You don't have to buy 50,000 and select 50 best ones. Wish Analog Devices offered something like that for the LTZ1000.

[mrk]

Thought about replacing the voltage divider R4/R5 for the tempreature with a PWM divider. Hoping to get at least some advantages from that.

• Cheaper?
• Adjustable Tempreature
• Higher stability?

What do you think?

[Kleinstein]

A PWM divider could be resonable stable, but there is still some possible residual drift (different on/off times, leakage, switch resistance drift). The other challange is avoiding noise / ripple. The start phase may be a challange too. One may need a sparate startup sequence to avoid excessive overshoot apon start up.  A possible added point may be a better defined ramp up phase - up to the point of including special exursions to lower turn off/on hysteresis.

It sounds possible and interesting, but not necessary easy. Likely it would be about building a seprate PWM DAC first and see how good it works - no extra need for very high linearity, but low ripple and low drift.

[bsdphk]

A PWM divider could be resonable stable, but there is still some possible residual drift (different on/off times, leakage, switch resistance drift).

Those are pretty much solved problems.  See Figure 7 on page 12 in HP Journal April 1989 (the 3458 issue) and the HP3245 which does precisely that with its PWM divider (as far as I have been able to tell.)

[bsdphk]

I should add: The PWM facilities in good microcontrollers can easily be programmed to do the "balanced pulse" trick.

[Kleinstein]

A PWM divider could be resonable stable, but there is still some possible residual drift (different on/off times, leakage, switch resistance drift).

Those are pretty much solved problems.  See Figure 7 on page 12 in HP Journal April 1989 (the 3458 issue) and the HP3245 which does precisely that with its PWM divider (as far as I have been able to tell.)

The HP figure 7 solves a different point, that I thing would not apply to a constant PWM ouput anyway. The On/off time drift is usually quite small, unless the frequency is really high. So just make sure the frequency is no too hight. In the DMMs the small drift is part of the offset that is corrected by AZ mode together with other larger effects.
Also the PWM DACs in the something like the Fluke 5700 usually uses a zero adjustment from time to time. I has a few additional sources of drift and the switch time drift may not be the main issue.

Anyway the PWM DAC part is an interesting part but at least initially it would be not very LTZ1000 sprecific. This thread is long enough - so better start a new thread on a PWM DAC, there  are already a few, but the main part there is more linearity and ripple.
Because of the attenuation factor the error from the temperature set voltage one effect the result to a small fraction, so the demands may not be too extreme.

[KT88]

A PWM DAC would be nice for the gain stage that amplifies Vz to 10V as well.
But this would not only reguire high strability but also high resolution, if we want the 10V to be spot on...
I agree with Kleinstein that another thread for this deep rabbit hole makes more sense than keeeping it here.

Cheers

Andreas

[bsdphk]

The HP figure 7 solves a different point, that I thing would not apply to a constant PWM ouput anyway.

Implement your PWM DAC with two pins instead of one, one pulling up and one pulling down, and it solves the ailments you listed above (if you calibrate it).

Dont tell me it doesnt work: I've done it and it works.

[RoGeorge]

You mean like this?
https://www.eevblog.com/forum/projects/fun-circuit-to-play-with/