LT1021CMH based mini-reference

[Alex Nikitin]

I've worked on that project slowly over half a year now and at last the reference is boxed and working quite well. It is built on a bit of a stripboard and as I only have enough bits to build one more, I am not planning to make a pcb for it. Most components are from eBay, the main exception is Pomona CuTe connectors. The reference is very simple, containing the LT1021CMH chip, a 20K multi-turn pot, few resistors and diodes for an additional temperature compensation (without it the tempco was over 10ppm/C, now it is around 1ppm/C), a reverse polarity protection diode on the P/S connector, a P/S capacitor, an LED, a 3-position switch and a nice VBF 10K/1.111111K divider in a hermetic case. This unit can provide 10V, 1V, 10K and 1K reference values with a reasonable accuracy and (hopefully) a decent long-term stability (better than 50ppm looks about right). The switch connects the output terminals to the voltage reference in one position, disconnects it in the middle and shorts the terminals in the third position. The divider is wired permanently to the output terminals providing 1V reverence from 10V output, so with the voltage reference disconnected, the resistance between two red terminals is either 10K or 1K. I will now measure the voltage and resistance values over a period of time (probably a couple of weeks) and my plan is to use this unit as a travelling reference. There is a space left for an output buffer to give the reference better load capabilities and a better short-circuit protection (right now it will survive a short, but most likely would suffer a hysteresis shift after that) so I will install it as a next step.

Cheers

Alex

[TiN]

Your DMM is unhappy though.

Nice little box. How about 10V noise? Also it's small enough to be fit into little heater box, do you plan running tempco tests as well?
I'm in for travel comparisons, if you need victims. VHD divider from hifi-guy? I remember he had this value dividers in his shop.

[Alex Nikitin]

Your DMM is unhappy though.

Error 103 was left after a remote operation    .

Nice little box. How about 10V noise? Also it's small enough to be fit into little heater box, do you plan running tempco tests as well?
I'm in for travel comparisons, if you need victims. VHD divider from hifi-guy? I remember he had this value dividers in his shop.

The noise (at NPLC100  ) is below 1ppm p-p, the tempco is estimated from the overnight run - see the graph below - the temperature change was about 6C max-min, so the tempco is about +1ppm/C). And yes, the divider is from that source. I've bought several items from him, including two of these dividers specifically for that box.

Cheers

Alex

[Alex Nikitin]

I've added a buffer using OP97 (so now short circuits are not a problem) and run a quick test on NPLC1 to look at the noise. Here is the result for a 90 seconds run, the vertical scale 0.2ppm/div . I suppose it is OK for this kind of a reference. The current consumption is less than 4mA unloaded, the load regulation on the output terminals with a buffer is less than 3ppm/mA (<3mohm output resistance).

Cheers

Alex

[Alex Nikitin]

And here is another measurement for 80min at NPLC10, in the lab at about 23C room temperature, with 10K load placed on the output terminals for 15min.

Cheers

Alex

[Andreas]

Is this battery powered or mains connected?

With best regards

Andreas

[Alex Nikitin]

Is this battery powered or mains connected?

With best regards

Andreas

Right now it is powered from mains through a simple switcher adapter  . It can certainly run from a couple of 9V batteries in series. For what it should do it is good enough as it is. I might add an internal 15V LDO regulator though.

Cheers

Alex

[Andreas]

Hello,

At least in the 2nd Diagram the noise seems to be rather high to me. (10uVpp).
After my understanding the 0.1-10 Hz noise should be 6uVpp typical.
With 10 NPLC (5 Hz) it should be even sligthly below.

With best regards

Andreas

[Alex Nikitin]

Hello,

At least in the 2nd Diagram the noise seems to be rather high to me. (10uVpp).
After my understanding the 0.1-10 Hz noise should be 6uVpp typical.
With 10 NPLC (5 Hz) it should be even sligthly below.

With best regards

Andreas

Hi Andreas,

The power supply right now is not very good, without question, just something that was at hand in the lab. I will try a battery run and a different (linear stabilized) supply possibly tomorrow. The "typical" values in the datasheet however are not a guarantee of performance. The "typical" tempco for the LT1021CMH is shown as 5ppm/C with maximum 20ppm/C in the temperature range. In fact the one I'm using in this unit has a tempco over 10ppm/C around 23C, fortunately quite linear so I could correct it with a simple diodes chain.

Cheers

Alex

[Alex Nikitin]

I am running the reference now from a nice linear 24V regulated supply however the 0.5ppm "jumps" are still present and I suppose that is the LT1021 chip itself. Still OK for the application intended. The VBF divider 10K/1K111111 is reasonably stable, below the results from the 68hours run over the weekend on the divider output giving 1V reference, the temperature dropped from +26C at the start to +20C at the lowest point, making the tempco a bit over +1ppm/C, plus a zoomed-in 1h graph to show the "jumps". Also attached the current schematics. The five diodes chain is selected for the particular reference (each diode gives about -2ppm tempco correction in this configuration, with the resistor R7 changed accordingly to keep the output at 10V). VR1 provides a fine trim. Output terminals switched by a three position switch, in one position only SW1.1 and SW1.2 closed, in the second position all switches are open and in the third only SW1.3 is closed, providing a short on the output to connect 10K and 1K111111 resistors in parallel, giving 1K reference value.

Cheers

Alex

[Alex Nikitin]

One more graph - 10K output measured over 15min (NPLC100 on HP3458A).

Cheers

Alex

[Andreas]

I am running the reference now from a nice linear 24V regulated supply however the 0.5ppm "jumps" are still present and I suppose that is the LT1021 chip itself.

Hello,

looks like popcorn noise.
I had also 1 extreme candidate (besides some other) in my LT1027DCLS8 thread.

What NPLC did you use on the measurement?

With best regards

Andreas

[Alex Nikitin]

I am running the reference now from a nice linear 24V regulated supply however the 0.5ppm "jumps" are still present and I suppose that is the LT1021 chip itself.

Hello,

looks like popcorn noise.
I had also 1 extreme candidate (besides some other) in my LT1027DCLS8 thread.

What NPLC did you use on the measurement?

With best regards

Andreas

Hi Andreas,

It is a pop-corn noise  , however I think it is acceptable for this unit  . I use NPLC100 most of the time for long-term logging, with some earlier graphs taken at NPLC1 and NPLC10 (noted in the relevant messages) . Here is another graph taken (at NPLC100) during a warm-up from a non-powered state on 10V output - 5ppm/div vertical and 1min/div horizontal. So far the stability is OK, I may try to run the unit in my improvised "Happy Reptile" temperature chamber to look at the tempco.

Cheers

Alex

[Kleinstein]

Even with some popcorn noise, the noise level is very good (e.g. 0.6 µV for the 1 h run).
The steps are only like 200-300 nV - so hard to tell with a less sensitive meter.
You also have to take into account that some of the observed noise is from the reference inside the meter.

[Alex Nikitin]

I did a 23C-15C-45C-23C run in my "temperature chamber" on the 1010B reference and will post some graphs later. Few things transpired during that run.

First, there is a noticeable lag between the reference itself and the temperature compensating diodes, as the diodes are placed on the board and not in a direct thermal contact with the case of the reference. It is only noticeable during reasonably fast temperature changes, as the diodes react faster than the reference, so if the temperature rises fast, the voltage can actually drop by 5-10 ppm before rising again some minutes later.

Second, the tempco measured after the voltage is settled at a particular temperature is reasonably constant at +1...+1.5ppm/C over the whole 15C-45C range, so with a better temperature compensation circuit this could be adjusted closer to zero. I won't bother with this unit though, it is good enough for it's intended application as a travelling reference that could be posted cheaply.

Third, the hysteresis was not readily measurable over that run, less than 1ppm variation after settling down at the same 23C temperature as at the start. It is also worth noting that the reference was transported at an outside temperature below 5C and did not exhibit any voltage shifts. The control run will be after the weekend when I'll bring it back to HP3458A.

I will also run a proper temperature test for the internal 10K reference resistor. A preliminary estimate from few points is below 2ppm/C.

Cheers

Alex

[Alex Nikitin]

Here is the second run in a "temperature chamber", this time with few "stops" - again, there is a visible lag on the thermal compensation, essentially no hysteresis, +1ppm/C tempco and a bit of a popcorn noise, still visible though the voltage was measured with the Keysight 34465A.

Cheers

Alex

[Kleinstein]

The popcorn noise look much stronger than in the earlier measurement. So maybe it is from a different source, like the DMM.

[Alex Nikitin]

The popcorn noise look much stronger than in the earlier measurement. So maybe it is from a different source, like the DMM.

Not really, the popcorn noise is about the same 0.5ppm p-p. Vertical scale is 2ppm/div. Other "jumps" are from the thermal lag between the tempco correcting diodes and the reference IC.

Cheers

Alex

[Alex Nikitin]

I've measured the reference again with the HP3458A after three weeks and some temperature changes powered and not powered between 5C and 40C - the 10V output has drifted less than 3ppm. I suppose it is good enough for that reference.

Cheers

Alex

[Alex Nikitin]

Here is the latest from a 48 hours run with the room temperature variations between 21C and 24.5C. On a longer scale the popcorn noise type variations are about 1ppm p-p, which is still acceptable for this kind of a travelling reference, I think, as the stability with time, temperature and humidity variations appears to be good.

Cheers

Alex

[Alex Nikitin]

I've done a little update on the 1010B last week, adding a DMOS FET to bootstrap the supply voltage on the LT1021. The main reason was the supply voltage variation affecting the output too much due to the self-heating of the LT1021. With the FET in the circuit the supply voltage on the reference chip is 11.3V and does not change with the external voltage supplied from 13 to 28V. As a nice side effect the power dissipation in the LT1021 is halved and so is the warm-up drift (from 40ppm to 20ppm). The rest of the parameters seems to be unaffected.

Cheers

Alex

[Gyro]

Just looking at the LT1021 datasheet, I hadn't realised there was an LT1021-7 which does not include internal scaling resistors (or trim pin). The thing that surprised me is that its long term drift is quoted at a mere 7ppm/kh typ versus 15ppm for the other variants, marginally lower than the LT LM199 spec (8ppm/kh). Presumably this is for the H package only.

I'm just curious that it's the lowest drift I've seen quoted for any IC reference. It's also quieter than the LM199 and with the benefit of a buffered output. Ok, the TC is higher than the LM199 but much lower than an unheated LM329. It looks as if it could work very well in a low temperature oven. Anybody tried it?

[Kleinstein]

It looks like there is no (more) LT1021-7 in a metal case. At least LT does not have is for sail in small quantity.

[Gyro]

Hmm, a good point (sadly). It's still just about possible to find genuine LM329H on ebay (and other LT1021H variants) but no sign of LT1021-7. An idea behind its time then. 

[Andreas]

It looks like there is no (more) LT1021-7 in a metal case. At least LT does not have is for sail in small quantity.

How can you tell that?
Arrow has more than 1000 LT1021DCH-7 stocked.
And sells in single quantities.

Besides this:
The stated long term drift (the first 1000 hrs after soldering) usually is much higher than that what you can measure after a 10000 hrs run in phase.
I normally measure 1-2 ppm/year for good hermetically references after that.
So for me a T.C. larger than 1 ppm/K hurts much more since a compensation of T.C. is never perfect (perhaps a factor 10 improvement) when the compensation is outside the reference chip.

With best regards

Andreas