best buffer amp and output power stage for LTZ1000

[niner_007]

Looking over various threads on LTZ1000 buffers, I've seen various amp options and usage.

3458A itself is using LT1001 for the ADC references
Illya's FX reference is using ADA4522
AN86 was using LTC1150

Can an LTZ1000 reference project do better than ADA4522? That seems just the best op-amp all around, with a few similar options available from elsewhere, like MAX, but seems these options are not better, just different, with similar TC, noise, and power figures.

Interestingly, in AN86, LT1010 op-amp buffer was used as output power driver, but one could just use a plain old transistor. On this, it does not seem like much can go bad. Thoughts?

[RandallMcRee]

Well,
This would be current state-of-the-art

[niner_007]

There is a bit of noise generated by LT1010, in theory that part could be improved.

EDIT: I had the noise of something else in mind, LT1010 has very little noise

[David Hess]

Within the bandwidth of the feedback loop, the input amplifier reduces the noise of the output buffer so output noise can generally be ignored unless you are relying on low broadband noise and little or no output filtering.

Nothing prevents using a discrete bipolar buffer or a current feedback operational amplifier as the buffer.  One function of the buffer is to remove loading effects from the output of the preceding precision amplifier to reduce thermal feedback in the monolithic precision part.  A single transistor class-a emitter follower with pull-up or pull-down resistor is often more than sufficient.

[Kleinstein]

There are not that many cases when one really wants just a buffer. Usually one has more, like some filtering, averaging or a separate sense input. The LTZ1000 already has a kind of buffered output - so in some cases no buffer is the best buffer. Hard to beat a piece of wire in noise and stability.

There quite a few OPs with lower noise / drift than the LTZ1000 reference. So there are quite a few to choose from - the choice also depends on the other functions (e.g. filter impedance) and speed needed.
EMI filtering at the OPs could be a good thing, but could as well be added externally.
Once well lower in noise / drift it does not make a big difference. So noise wise there is no significant difference whether using an ADA4522, LTC2057, MCP6V51, OPA187, max44250 or similar.
Good BJT based OPs (e.g. OPA207, ADA4077,OP27,...) could also be good enough if the impedance is right.

[David Hess]

I have seen it suggested more than once that Linear Technology included the LT1010 where an emitter follower would have served to sell more parts, but they would hardly be alone in doing this sort of thing.  It is a great product but I always avoided it because of price compared to the alternatives.  Once you are buying LTZ1000s and the precision parts to support it, the LT1010 becomes cheap and saves design time and space.

[Andreas]

Hello,

3 possible explanations:
- the LTC1150 alone can only deliver 2mA which is too low for a reference output. (typically 10-15 mA)
- the LTC2400 creates large current spikes during conversion.
  At least with a similar LTC1050 (also 2 mA) I have large non-linearities when not using a adjusted R/C filter.
- besides this the whole cirquit should be immune against capacitive loads

So I prefer a ADA4522-1: (together with some measures for capacitive loads)
- current in excess of 10 mA
- EMI hardened inputs

with best regards

Andreas

[niner_007]

Hello,

3 possible explanations:
- the LTC1150 alone can only deliver 2mA which is too low for a reference output. (typically 10-15 mA)
- the LTC2400 creates large current spikes during conversion.
  At least with a similar LTC1050 (also 2 mA) I have large non-linearities when not using a adjusted R/C filter.
- besides this the whole cirquit should be immune against capacitive loads

So I prefer a ADA4522-1: (together with some measures for capacitive loads)
- current in excess of 10 mA
- EMI hardened inputs

with best regards

Andreas

do you use a buffer in the output, after ADA4522? It sounds like no

[ArthurDent]

Ideally you wouldn't have any loading of any voltage standard although in the real world there is always a very very small load. I modified my Dr. Frank board by adding 3 components to form an emitter follower within the feedback loop (that corrects for any variation of the added components) so that the output remains at 10.00000 volts with the added circuitry.

What I added isn't a typical emitter follower because of the 470 ohm resistor which acts as a cheap current limiter if the output is shorted. I don't like the idea of adding some buffer outside the feedback loop that won't be compensated for by feedback. I also made sure that I was sampling the output directly at the output terminals (4-wire) so none of the wiring drops would influence the output voltage. There is still some drop with loading and here's what I mentioned in a post about comparing the USA Cal Club standard to my standard.

"I decided to check the 2 supplies to see how they reacted to light loads and I knew mine could survive a direct short for some time because I designed the added transistor buffer within the opamp feedback loop to handle a direct short and to also protect the opamp. With a 100K load both supplies dropped 1uV; at 10K, mine dropped 10uV, the club’s 20uv; and at 5K, mine dropped 20uV, the club’s 40uV so the output resistance of mine is less than the club’s but neither would be loaded like that in any testing situation."

 

[Andreas]

do you use a buffer in the output, after ADA4522? It sounds like no

The ADA4522 is already the buffer.

with best regards

Andreas

[dietert1]

When you made these tests, did you ever care to check what is the effective output resistance when your output buffer has a little more basic load - say 2 or 5 mA?

I have been using a JFET instead of the transistor, a proposal of Andreas if i remember right. It increased the output voltage margin of the OpAmp from positive rail by about 2 V. So the buffer could run from 13 instead of 15 V. You can then omit the collector resistor but you need a schottky diode to prevent gate current in case of short.

Regards, Dieter

[guenthert]

Ideally you wouldn't have any loading of any voltage standard although in the real world there is always a very very small load.
[..]

  Well, it depends on the application.  For a 10V lab reference I would expect typical usage to be virtually load-free (an attached high impedance DMM should draw less than 100pA, which would cause an voltage drop too small to be detected, similar for measurements using null meter / potentiometer method).  You wouldn't drive a voltage divider with it directly.

  A 7-something-something V reference otoh might be intended for monitoring a high impedance long scale DMM or (embedded) feeding e.g. an ADC, in which case you most certainly would want it be buffered.

  Said that, I like the transistor in the output/feedback loop.  If for nothing else, then to protect more valuable parts from lab mishaps.

[David Hess]

Said that, I like the transistor in the output/feedback loop.  If for nothing else, then to protect more valuable parts from lab mishaps.

I always buffer exposed signals, and there are good reason to buffer integrated precision operational amplifiers as I stated earlier, but of course without a current limit, the buffer is susceptible to damage and a soft kill is still a kill.  But it is easy enough to add a collector resistor to limit current, Vbe current limiting with or without another transistor, or beta current limiting.

[GigaJoe]

my variant:  (some filtering, haven't had a chance to measure,  due to blow up scope preamp)

[Kleinstein]

3 capacitors in series for the filtering in overkill. There is probably more lost from reduced capacity than gained from an marginal lower voltage at the final capacitor. For an electrolytic cap, I would not be sure that zero voltage is really a desirable state and lowest leakage. R7, R8 are relatively large, so the initial settling of the capacitor DA may take a long time.

The decoupling capacitor at the OP is rather small. Similar C1 is rather small.

The usual circuit for a an output that is tolerant to capacitive loading would have an extra resistor in the feedback and a capacitor directly at the OP. Usually it also helps to have some RC element (e.g. 10 Ohms + 100 nF) already at the output. This can reduce ringing a little.

[GigaJoe]

how about that:



filter seems do the job:


no i'm not so so silly , here is 3 opamps ..  (it def better then my   )
https://www.analog.com/en/analog-dialogue/articles/why-does-voltage-reference-noise-matter.html

[guenthert]

  As stated earlier, there are plenty of precision OpAmps with lower noise than the LTZ1000.  Short time noise that is, I'm not so sure about long term stability.  I'm afraid (no firsthand experience), adding more buffers and filters buys you lower short term noise in exchange for poorer long term stability.  I guess, that might be a sensible approach for some applications, but for a lab reference? 
  And if one doesn't care about long term stability, a plain Zener (serveral mA of Iz) might be a better choice or - in the extreme - a battery.

[GigaJoe]

I thought, it for a conventional opamp, but zero drift opamp are stable for a long period of time, isn't it ?

Noise canceled can be interested for a source of much large noise, comparing to LTZ
Hmm ...   would it be make sense to make it as a separate,  noise canceling box , 1:1 repeater?  It seems quite simply do 1:1 adjustment by nulling ...
Sorry ... seems I'm do shifting a topic ...

[Kleinstein]

The LTZ1000 (and also an LM399) reference is strong enough to directly drive the filter. There is absolutely no need for the extra buffer. The filter would be a 50 K load only initially later it would be much larger than 50 K (more like > 50 G required).  Because of leakage I would consider 50 K and also the 20 K from an earlier post a little on the higher side - this makes the circuit prone to drift from leakage / bias currents.

With an AZ OP some filtering is nearly required to isolate the reference from possible higher frequency signals / spikes going backwards. However this would be more like some 2 K and 1 µF (film cap). Really filtering the low frequency noise is very hard and never really successful - 1/f always wins on the long run. Electrolytic caps are somewhat tricky here. They need a long time (e.g. days to weeks) for settling -  so if at all this would be something for a circuit that is on 24/7 only.

For some applications it may still be good to have some filtering down to the 1-10 Hz range, still just possible with film caps. Some DMMs / ADCs also react to noise in this range even for slow readings.

[guenthert]

I thought, it for a conventional opamp, but zero drift opamp are stable for a long period of time, isn't it ?
[..]

  Zero drift is of course only a marketing term.  I just checked the data sheets of ADA4522, LT2057 and OPA187.  Their offset voltage does drift very little over temperature and common mode voltage, perhaps justifying the term.  I couldn't find any specification for the drift over time.  Linear has been kind enough to publish in the data sheet of the LT2057 the observed drift of offset voltage of a handful of 'typical' units over a time frame of 1000h (much less than 1uV).  The others might not have such graphs, as it is difficult (and time consuming!) to measure.  For most practical purposes, one might consider that zero as well, but wont they diminish the stability of a LTZ1000 reference, even if a little?

[Kleinstein]

For the LTZ1000 reference, not aged very long one can be lucky and get a drift of something around 1 ppm/1000h, but worst case it can be more. This would be 7 µV/1000h and thus considerably more than expected for the AZ OPs. Still if one can avoid it, it is better to avoid.

I see not real need to have the second OP. Only if one wants / needs external sense terminals also at the negative side one may need the second OP.  Here I see 2 choices:
A) use 1 OP for the positive side and 1 OP for the negative terminal.
B) use 1 OP for the positive side and current compensation for the negative side.

The current compensation would usually have more residual current at the sense line, but this can be acceptable, if the sense connection is reasonable low resistance.

[ArthurDent]

An opamp is required if you want to translate the 7.xxxx volts from the LTZ1000 voltage reference to 10.00000 volts. Read through the 120+ pages in this thread for more info.

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/

[dietert1]

I found this test of transistor noise:

http://dicks-website.eu/low_noise_amp_part3/part3.html

Does it apply to the recommended transistor output buffer? I noticed that 2N3904 isn't really a low noise transistor. I found rs components have that ZTX851, will give it a try. It's an strange one: 60 V 5 A yet in plastic package like any other small signal transistor.

Regards, Dieter

[Kleinstein]

There is the general tendency that larger transistors can have relatively low noise. This is especially true for a test at a rather high current like 10 mA. The small BC550 or 2N3904 are more like made for operation at 100 µA or maybe 1 mA.

The output buffer circuits use the transistor inside the control loop of the OP. So for the lower frequencies the noise the transistor is attenuated by the OP and only the higher frequency part (e.g. > 1 kHz , depends on the circuit)  may be directly effected by the transistor. So the transistor noise is not really relevant for the reference buffer.

For use as a low frequency amplifier to look the reference noise, one also has to look the the noise current, as the AC coupling naturally is relatively high impedance. In this area the transistor gain is also important and the operating current is usually smaller (more like 10-100 µA).

[dietert1]

The output buffer drives a 10V to 7 V divider at 10 mA plus a LTFLU reference at about 8 mA. And the circuit includes a 100 uF output capacitor, so the control loop has to run at reduced bandwidth. This is why i should select a low noise transistor. Circuit looks roughly like the one posted by GigaJoe above - the second one including a filter.

I tried some low noise JFETs but they don't work well at higher currents and only 4 or 5 V headroom. A J112 MUX JFET works well but doesn't have any specs for linear operation.

Regards, Dieter