Issue with my ltz1000 circuit

[julian1]

I have built several implementations of the ltz1000 reference. In most respects the circuits appear to operate correctly.

However the last digits show more noise than expected - perhaps 20-30uV over 10seconds? And bizarrely - when I move my hand or body position with respect to the device (slowly to avoid excessive air-currents) the ref voltage changes on the order of 500uV to 1 mV!

Here is a video, that demonstrates the effect,




The issue occurs,
  - in both a breadboard and pcb configuration (I've built both)
  - without a heater stabilization circuit - with the ltz1000 left to heat to operating temperature given current/ambient conditions. (rules out heater circuit as source of issue)
   
- the only circuit changes relate to parts availability,
  - dual-rail since my op-amps wouldn't otherwise handle the approx 0.5V inputs that the rail-to-rail ltc1013 can
  - currently uses a Vishay prec 10k trimpot instead of the 1k/12k divider to provide adjustable temp control (set to run at 45degs).
  - 58k pull-ups for both sense and control collectors instead of the 70k or 30k I've seen elsewhere.
 
notes,
  - I've tried with two different ltz1000 references - both purchased directly from the linear-tech website.
  - tried both op27 and opa277 op-amps
  - tried increasing the size of the 22nF cap.
  - can freely twiggle the power supply voltages and ref shows good independence - no detectable change on 34401a
 
To rule out power-supply or other measurement influence - I have another circuit built around the dw232/op27 with current turned to balance tempco. In comparison it is very stable and will peg the last digit of the 34401a.

[mimmus78]

Well first thing first is you need to insulate ltz1000 from free air. Cover the top chip side to stabilise temperature, cover also pin on the other side of the PCB to avoid EMF effects.
Than you should re-read the datasheet (it's all explained there) and the huge LTZ1000 thread on this forum ... you'll find years or research info there and almost everything you need to know to build LTZ1000 references.

[julian1]

The datasheet states,

"Air currents blowing across the leads can also cause small temperature variations, especially since the package is heated. This will look like 1ppm to 5ppm of low frequency noise occurring over a several minute period."

The change of 500uV to 1mV is two orders of magnitude difference from that suggested by the datasheet.

The effect is present when my hand or body at a foot away is completely stationary.

[doktor pyta]

Show us what power supply which You use. Is it switch mode PSU?
Does any instrument connected to the LTZ board use switch mode PSU?

Have You checked outputs of opamps using oscilloscope ( if they are not oscillating) ?

[julian1]

Show us what power supply do You use. Is it switch mode PSU?
Does any instrument connected to the LTZ board use switch mode PSU?

Two linear HP 3610A. The power supplies are capable of holding my dw232 reference steady for all digits of the multimeter - so I am not immediately suspicious of them.

I already tried a 9 volt battery to replace the supply handling +VE/gnd with the same result. I'll see about picking up another 9V battery so that both rails are on batteries.


 

[ManateeMafia]

...
The effect is present when my hand or body at a foot away is completely stationary.

I see a similar effect when measuring high value resistors, even from several feet away from the DUT. The effect is usually mitigated by wearing a ground strap.

[julian1]

Have You checked outputs of opamps using oscilloscope ( if they are not oscillating) ?

Hmmm, it's the first thing I did. There's nothing at or below the frequency of the op-amps - currently an opa277 with GBW of 1MHz.  But there is a 100MHz 200mV thing on the rails and op-amp inputs that my scope always picks up and I don't know if it's a scope issue or RF emf (I posted about it here https://www.eevblog.com/forum/beginners/trying-to-understand1054z-with-200mv-p2p-noisesignal-at-100mhz/).

I think I really need to try powering both rails on batteries.

[DimitriP]

But there is a 100MHz 200mV thing on the rails and op-amp inputs that my scope always picks up and I don't know if it's a scope issue or RF emf

Am I the only one that thinks that this issue should be resolved/eliminated or circumvented first,  before wondering why hand waving over an unshieled board and a bunch of wires changes the 3rd , 4th and 5th decimal digits ?

[julian1]

Using 9V batteries for both rails and the same issue is apparent. 

why hand waving over an unshieled board and a bunch of wires changes

I would tend to agree. On the other-hand there is no effect on my other reference circuit.

I see the signal even on *both sides* of the op-amp 0.1uF ceramic decoupling caps. How does that make any sense unless it's a problem with the scope/probes?   

[TiN]

Can you post your schematics and layout. That voltage change sounds too big for properly operating circuit...

[julian1]

schematic is here, http://s3.julian1.io/ref-ref.pdf

The pcb made from the ki-cad schematic used a symbol library that mixed up the pinout of the 2n3904 buffer for the heater (hence the through-hole transistor bodge you can see) and direction of the two 1n4148 - so its more misleading than insightful.

The 12k/1k in the schematic are optional and were not fitted - and a 10ppm/C Vishay trimpot sets the temperature control voltage. The temperature feedback works very well according to the thermocouple I've had attached to the case of the ltz.

Note, that I had the same issue of the 1mV drop without any of the extra heater control. Just a simple breadboard with the zener/120ohm and 70k/collector sense fed as inputs to the op-amp.
 

[TiN]

Add 10nF NP0 capacitor between pin2 and pin6 of U14. Lack of attention to ground path on schematic suggests that it's same on PCB layout?
You must have separate heater high current return from signal ground. Meaning that cathode of your D8 should connect directly at PSU ground terminal for best solution.
Also don't attach anything to LTZ body, as your thermocouple acts as a heatsink in such case. Monitor heater voltage instead to check oven temperature stability.

[Andreas]

Hello,

this effect is due to switchmode supply anywhere in your lab.

But how can you try to operate a LTZ1000 cirquit without any 10-100nF between pin 6 + 7 of the reference
after having read the LTZ1000-thread?

Try to find out the most sensitive pin in your cirquit by using a steel needle in your hand touching all pins in your cirquit.
(do not forget to use ESD strap).

with best regards

Andreas

[Kleinstein]

Seeing a lot of high frequency background with the scope, suggests there could be a relatively strong source around (e.g. radio station). So it might be a good idea to test the circuit inside a metal case (e.g. the famous cooky can).

Shielding against air flows is an other important point.

Having the hand near by can have two effects: one if IR radiation that could heat up the circuit (could be a problem without thermal regulation). I don't think the few 1/10 of a K should have so much effect. The other is changing the RF field by acting as a kind of shield / antenna.

p.s.:
The OP27 is likely to high in current noise and bias drift, as the impedance of the connecting points is rather high (slightly less than the collector resistors). There is no need for a super low noise OP, as the transistors already give about a gain of 200. The more important point is low bias and reasonably long term stability. The OPA277 should be ok, about as good as the LT1013.

[TiN]

Andreas
I don't have any direct capacitance at pin6-7 either .

[Andreas]

Hello Illya,

but you have a 4 layer board and guard rings around the critical lines.
This has a similar effect at least for capacitive dissipated RF. (hand sensitivity).
Probably not so good for conducted RF from the power supply line.

with best regards

Andreas

[julian1]

I appreciate very much everybody's replies. I'm working on providing the answers !

1.) Try to experiment placing DUT in Tupperware box in another room, maybe there is EMC disturber in neighbourhood.

I can't easily do that, without moving the bench-meter and probably power supplies. 

2.) Put 2 electrolytic caps of somewhere 1000µF (16V) on the powerrails - close to the circuit ... & check powerrails with DSO
(scope BW limit = 20MHz ON)

Soldered two 1000uF/50V onto the pin-headers , +VE/gnd and gnd/-VE. BW filter on, and swept time divisions. There's nothing there except
the already mentioned 100MHz.

I believe the positive powerrail requires VCC = +VE > 12V ... so +15V is perfect to start with ... a 9V battery for positive powerrail will probably not be enough for the output headroom swing of Op Amp A2 driving the 1N4148 diode.

Yes, I wasn't sure either. Also the heater sucks some current when the ltz1000 is started from cold (bench shows +VE rail drawing a total of 40mA initially). But the the zener got up to voltage and I think the test had some value - in that the issue was still present. In any case I have increased the bench supply to +-12V.

In order to give the heater Op Amp A1 enough headroom also negative powerrail -VE should be applied.

Yes, everything is dual rail.

3.) check the voltage on pin 6 of LTZ1000A - if the heater temperature is too low and ambient temperature (Australia) too high the PID controller that drives BJT transistor 2N3904 cannot do its job properly. Can you provide value of voltage on pin 6 ?

Ok, this was quite interesting. With a sharp probe positioned next to pin 6, I measure voltage of 0.532V with ltz1000 case temp of 48C (thermocouple on case). The probe seems to have some influence on regulation however. Over the course of a minute the voltage drops to 0.526V with ltz1000 case temp downregulated to 42C. I repeated this experiement several times and observed the same result.

> What's the ambient temperature in Australia ?

abmient temperature at the moment is 16C.

4.) That resistor 120R (between pin 4 -> GND) should be of very good quality not an ordinary resistor from local HW shop - but probably that one is OK ... [PWW or MFB resistor with TC < 5ppm/K].

Yes, it's VPG metal foil 0.01%. TC is 2ppm/C

5.) Use plastic shield for LTZ1000A on top and on bottom - LTZ1000A is using KOVAR leads 39µV/K.
     It is designer's duty to keep gradient of all 8 terminal leads at delta T = 0°C ... (= all same lead temperature).

This is non A version.

6.) Consider using power planes, GND planes, star points next time you re-design your PCB.

Yep, there is a solid GND plane. I'm aware of the desirability of star-grounding, but I was only trying to knock something together to rule out issues with my previous breadboard implementation.

[julian1]

Add 10nF NP0 capacitor between pin2 and pin6 of U14.

This is frustrating. I purchased some 10nF NPO caps only a few weeks ago.
http://au.mouser.com/ProductDetail/AVX/12063A103JAT2A/?qs=9C5X5N50CkE%2F69DbCtXoFw%3D%3D

In the bag I just opened are minimelf packages and the LCR meter shows 1.2 ohm resistance and no capacitance.  Otherwise I have 1nF NPO that look right. Would you suggest that - or a regular 10uF ceramic?

 

Lack of attention to ground path on schematic suggests that it's same on PCB layout?

There is a solid ground fill. power traces on the back run along the edges are are 1mm width - the same as all other traces. Op-amps have have decoupling straight through to the ground fill.

You must have separate heater high current return from signal ground. Meaning that cathode of your D8 should connect directly at PSU ground terminal for best solution.

I'm aware of the desirability of separate/star grounding from the datasheet - but discounted it as unnecessary for a simple working prototype. Do you think that could be producing the 1mV output changes seen?

Also don't attach anything to LTZ body, as your thermocouple acts as a heatsink in such case. Monitor heater voltage instead to check oven temperature stability.

The thermocouple has no fixture - it's just two exposed braids that are wound together and would weigh a fraction of a gram - and then threaded into a rubber band. That said, probably half the tests have been done without it.

[TiN]

julian1

Haste is a bad idea with LTZ1000 circuitry. When one go over 5.5+ digit resolution, you learn to be gentle and slooow...
Learn from this, or be ready to pay for mistakes. My first LTZ1000 back few years ago was killed by such mindset (oh, I just plonk circuit together. Resulted mistake in heater controller cooked zener chip after few days).

You can get my open design PCBs from OSH Park EEVBlog member, SvanGool, ordered some recently and published the project for all.

Did you thoroughly clean your board with IPA after assembly? I can trigger such big changes (hundreds uV) only if I physically touch the board. Just tried that on a module for you
Also since you using non-A, suggest to add 200-400KOhm resistor like in reference circuit from datasheet.

[julian1]

But how can you try to operate a LTZ1000 cirquit without any 10-100nF between pin 6 + 7 of the reference
after having read the LTZ1000-thread?

I feel so ashamed. I've only read the first 25 pages  . Thanks for the heads up on the cap, I'll investigate later. Happy New Year to everyone 

[TiN]

10uF will likely end up destroy your module. Try to find 10nF NPO. Good film cap can work too. Or you can stack 10 x 1nF's

I went into trouble and recorded short vid, showing behavior of my old prototype board. Big jumps happen only on touching PCB traces.
Don't try this at home though, unless you OK with killed reference... 

Video

[martinr33]

Also, you should use a logging voltmeter. That might expose some patterns that you won't otherwise see. I have a module that hops between two voltages, probably a bad LTZ1000.

[MisterDiodes]

Julian,
The other suggestions here are correct.  Although I've -never- had to add any extra capacitor over what's shown in LTZ datasheet and building hundreds - but do watch out for that Vishay pot you're using on the heater ratio.  Even if it's a foil type look at the Vishay datasheet and you'll probably see that they list the TC for the pot's body resistor - and then look at the TC of the wiper itself, and you'll see that it's probably 25 or 50 or 100ppm TC thru the wiper.  If you measure it you might find higher TC if the wiper is near one end or the other. Typical datasheet voodoo - I've never seen a pot used that doesn't make use of the wiper! As David L. Jones would say "A Trap for Young Players". 

That high TC will easily show up on the LTZ output voltage just by placing your hand near the pot - because the heater ratio resistor drift shows up strongly on the output as shown on datasheet, depending on the pot wiper position (that high wiper TC can cause very strange effects).  I would suggest using real discrete resistors there, even PTF56's will be much better than the pot and have less capacitive coupling.  Or get some good PWW resistors from Edwin Pettis of the correct values listed on datasheet.  Start with those values until you see your circuit working and then try other values from there - then you'll learn as you go.  In other words:  Build the known working circuit first with the correct values - I've never seen that fail, if you use good PCB design techniques.

12k over 1k heater ratio I have seen be too low a value for proper temperature control - but that depends on your environment and thermal flow.  Start with 13k over 1k as shown on datasheet until you learn how the circuit behaves.

Also make sure the heater control transistor isn't creating heat transfer to the pot.  It won't take much.

I know you tried running the system without a heater circuit, but there is something to watch out for there:  You're changing the temperature of the LTZ package just by the IR heat emitted from your hand!  That brings up a thought - are you seeing the heater circuit servo correctly when it's hooked up?  Does the operating point change when your waving hand is near?  What happens when your hand is wrapped inside a rag or insulated glove?  That will help tell you if its IR heat-related.

The datasheet also specifically mentions that this circuit is not suitable for breadboard construction, as you found out.

Keep the board relatively clean and -all- air drafts away from LTZ.  Especially the non-A LTZ version.  Use good star grounding techniques on your board connections.

If you're getting LTZ's from Linear Tech, I suggest getting some '1013 amps from the source also.  That is the recommended amp for this circuit with low input current noise.  It does have a small Voffset but doesn't matter here as a current driver.

You want to power from a quality linear power supply or batteries if possible and get rid of any LED or Fluorescent lamps nearby - those can cause a lot of HF hash in your circuit.  If you're running this from a switching power supply you'll need to get creative with serious power supply bypass caps as others have pointed out.

Also: Once you get it running, keep it powered for many months to see where it settles in.  It will take that long for the LTZ die to stabilize after soldering.  This is a circuit that requires patience on a very long time scale if you want to observe true performance.  By "Observe" at a serious level means you'll really need some serious gear to monitor ppm changes accurately - ideally several 732a/b (at least three), probably a KVD and several 3458a's - or similar combination.  Using a single 3458a to monitor an LTZ circuit doesn't reveal the full story since you're just comparing the drift rate of two almost identical Vref circuit topologies.

Out of curiosity, what's your application for building an LTZ Vref?  Just for nuttery fun or will it be used on an ADC or DAC or ??

[julian1]

I have tried many suggestions, including separating out the heater return path from gnd, adding a 10nF polypropylene as well as adding other pin caps to reduce noise on the heater feedback path.

At the moment, I suspect the RF issue really needs to be dealt with. 

Implementing the reference design that uses the 200ohm trimpot in series the zener cathode, but without  the signal diode (used to guarantee zener startup) on the output of the op-amp shows a stronger immunity to the hand moving thing.

The 100MHz RF likely from an FM mast, is much higher than the bandwidth of the negative-feedback paths of the op-amps. However, the ltz1000 on-die bjts are common-emitter 200x gain amplifiers and may well be fast switching. Also relevant here is that my main variation - the Vishay trimpot feeds directly into the base of the sense base tx.

If the issue is the on-die bjts picking up EMI, that would also explain how my dw232 reference is rock-solid even to direct physical touching - since there is no gain-stage apart from the internally-compensated (ie bandwidth limited) op-amp.

I have mostly ruled out radiant temperature affects since the distortions are too gross, too fast, and happen even when the board is insulated under a stack of papers.

[MisterDiodes]

Be aware that most paper is very transparent to IR, depending on how may layers that might not be doing much for insulation.

Also note that the LTZ die can respond surprisingly quickly to -any- heat variation.  Modulate heater current very slightly and the LTZ output will track it.

Out of curiosity, why did you opt for the non-A version?  Did I read that right? 

[julian1]

Out of curiosity, why did you opt for the non-A version?  Did I read that right?

A couple of *ideas* (not fully justified or thought out) I had about the non A version were,

- that non-insulated means that it can cool faster, and I felt that was likely to be more balanced/symmetrical to the heating side of regulation which is reasonably powerful. So hopefully the pid controller would have an easier job and spend equal time on both sides as it converges on the equilibrium regulation point.

- Faster cooling means it's faster to do the tempco trimming search in the version of the circuit that is not temperature regulated. That's because after applying a heating pulse, there's less time to wait before it cools and is ready for the next pulse.

- I associated the A versions with higher temperatures - at least it was my observation that many using the A version were running them hotter (including hp3458). From an engineering point of view - if there is sufficient headroom over ambient conditions to guarantee stability then that should be sufficient. Also the better long-term drift characteristics of the ltz1000 at lower temperatures.

[Andreas]

At the moment, I suspect the RF issue really needs to be dealt with. 

Hello,

now when I have seen the other thread with the scope ...
Yes you have a massive 100 MHz RF problem.
The RF gets rectified on PN-junctions and generates a "offset-voltage".
Would be interesting if the source is radiated or conducted via the mains line.


I fear a capacitor alone will not help much.
At 100 MHz a 2.2nF to 10nF might be better than a 10nF to 100nF due to self resonance of the capacitors.

But there are too many potential problems on the PCB:
Long legs of LTZ1000 which can generate a loop antenna (similar to the scope).
(so the filter cap may be too far away to filter out the RF).
The "pot" has also a large capacitance to the environment (your finger).
I´d suggest additional using of a metal housing and at least filter the power supply lines with feed through like capacitors to the housing.

The standard cirquit of LTZ is likely to oscillate when you try to filter the zener voltage too ...

with best regards

Andreas

[julian1]

At the moment, I suspect the RF issue really needs to be dealt with. 

Hello,

now when I have seen the other thread with the scope ...
Yes you have a massive 100 MHz RF problem.
The RF gets rectified on PN-junctions and generates a "offset-voltage".
Would be interesting if the source is radiated or conducted via the mains line.


I fear a capacitor alone will not help much.
At 100 MHz a 2.2nF to 10nF might be better than a 10nF to 100nF due to self resonance of the capacitors.

But there are too many potential problems on the PCB:
Long legs of LTZ1000 which can generate a loop antenna (similar to the scope).
(so the filter cap may be too far away to filter out the RF).
The "pot" has also a large capacitance to the environment (your finger).
I´d suggest additional using of a metal housing and at least filter the power supply lines with feed through like capacitors to the housing.

The standard cirquit of LTZ is likely to oscillate when you try to filter the zener voltage too ...

with best regards

Andreas

Good, thanks. I suspect there may be some rectification on the output of the op-amp by the 1n4148 as well. Out of *all the things* I have experimented with - removing it has been the only thing that had a definitely noticeable affect.

I don't know much about RF. I'll try to buy a fully enclosed aluminum project box, and dunk it inside that. It should be an interesting test anyway.

[Kleinstein]

A closed metal can definitely a good idea. Besides of keeping RF out, it also helps to keep temperature gradients constant - so you have only one temperature to care about and not additional external temperature gradients. A case also helps keep the whole circuit at a slightly elevated temperature and thus lower and slightly more stable humidity level.

With extra capacitors one has to be careful not to make the circuit unstable and this way potentially damage the reference (e.g. gets to hot).

Even the non A version is rather fast in temperature response. Usually the same temperature regulation circuit is used - though the thermal response is not the same. Even without a very exact adaption to the thermal regulation is already very good. Still this is one of the very few points with a little room for optimization (e.g. to work better at low heater power). The main disadvantage of the A version is the slightly higher minimum temperature (and thus slightly more aging). Theoretical there could be a slight effect of the case at a different temperature as the chip itself, as there could be a minute IR radiation effect, but this should be a really small effect unless the temperature is really high.

Using the LT1013 as an OP would makes things much easier as one would not need a negative supply. There is no good reason not to use it: it is way good enough and readily available.

[Andreas]

I'll try to buy a fully enclosed aluminum project box, and dunk it inside that.

A casted case or a tin plated steel case is the way to go.
but at 100 MHz also the 1 oz copper foil from a PCB will be ok if carefully soldered at the edges.

Examples:
http://www.amplifier.cd/Technische_Berichte/Spannungsreferenzen/Spannungsreferenz_verbaut.html
http://www.amplifier.cd/Technische_Berichte/Netzteil/Versorgung_Spannungsreferenz.html

with best regards

Andreas

[VK5RC]

+1 re EMI,  I got around 100uV changes with a smpsu from a battery charger or led light driver,  probably conducted via ground. 
There is a standard Ham trick of tracking EMI of putting the DUT on full battery power,  switching everything mains powered off (some even switch off the whole house) ,  then switch it back on one circuit at a time!

[julian1]

Placing the pcb in a heavy die-cast aluminium box, with a gnd run to the case helps enormously. There is little detectable change above background noise doing the hand test. The general noise also seems lower.


 

[Andreas]

Hello,

fine,

with a 10 nF capacitor at the heater setpoint you should now get below 1uV change instead of 20-30uV.

with best regards

Andreas

[nns]

I had extremely similar issues with an LTZ design. Large changes in output voltage when I was in proximity to the board. I also have local RF noise in the 100MHz range in parts of the office, so I double shielded it to no avail, leaving me to believe that it is magnetic instead of electric noise pickup. No mumetal box big enough to try, so I spent a significant amount of time minimizing loop area in the main zener path. It seems this path (particularly the grounding location of the 120 ohm) can have large effects when there is local RF noise.

A simple method that might help is a small 100p cap directly on the output voltage (pin 3 to pin 7). This may help mute the RF pickup. You could also do some hacking to the board (what I had to do on my early prototype) to adjust the grounding location of the 120 and pin 7 locations.

After optimizing layout I was able to obtain steady output to within about 1-2uV independent on board/body position and no other consideration to shielding.

As an FYI, I've also including an "Andreas" capacitor and did find that helped a little, although not nearly as much as my other comments.

[MisterDiodes]

..Minimizing loop area on critical circuits is always a good idea! Keep the connections short and optimized.

That brings to mind something else:  Be careful adding capacitors to LTZ's and other delicate analog gadgets, especially ceramic smt's - they are very microphonic and can cause surprising problems - and turn your LTZ into a seismic detector.  We have seen those pickup local mechanical cooling fan motor vibrations.

Or like the day we had a delicate circuit that was making noise only at around between 1pm to 3pm on weekdays:  Turns out it was the train locomotive a few blocks away making a local pickups or dropoff at a rail siding... It was slowiy shaking the building enough that we didn't feel it, but an LTZ's ceramic bypass caps did.

Just keep an eye out for those types of issues.

[julian1]

Minimizing loop area on critical circuits is always a good idea! Keep the connections short and optimized.

I basically did the opposite - and tried to match the schematic as closely as possible to make it easy to debug if needed. Lots of room for improvement!

[MisterDiodes]

OH yes - the more you play with LTZ-type references, the more you realize you can make all sort of stuff unintentionally: Radio receivers, thermometers, seismic and vibration detectors, board stress detectors, barometers, board cleanliness measurement units, resistor drift test equipment, etc. 

That's why I was asking about the choice of Non-A LTZ - remember those especially can be micro-barometers and board stress detectors - those internal wire bonds are more  sensitive to the outside world mechanical stresses than the 'A version.  Also you will need to adjust that extra TC resistor for your thermal flow situation - and if you're not using the datasheet resistor values you'll need to also adjust the TC resistor for best operation at -your- ambient temperature, enclosure and environment.

In the long haul over years and decades, we've never seen one LTZ type - be it A or Non A - outperform the other in terms of drift or noise.  Either one can work well and we've plenty of examples where they just run very well inside datasheet specs.  The non-A does require a bit more fiddling but can be good for very controlled environments, and the 'A is easier to get running on a production basis - and a lot more forgiving in real world apps.  That's why a lot of manufacturers stick with the 'A.

I'm looking at a hard-working 3458a that has an 'A in it, the usual 15k over 1k heater ratio and LT1013 in plastic DIP - and it doesn't move a PPM or so annually, and has worked that way for years.  The datasheet circuit values is a very good starting place for learning how these work. 

And a PPM drift is almost impossible to -accurately- measure anyway in any real-world situation, if you add up all measurement uncertainty.