Thermal fluctuations on PCB

[Kleinstein]

Attached is a plot of a temperature measurement via a 1N4148 diode (normal mounted  DO35 THT case) on my DVM board.
Quite often I see a quite stable oscillation with a 5-8 seconds period. The amplitude in the temperature is not very large (some 5 mK peak to peak in the example), but this really gets anoying when aiming for low noise also over a longer time scale.  I can suppress the oscillation by covering the PCB, but it needs quite some thermal isolation to make it disappear and the cloth on top of the circuit is not ideal with extra leakage and long time to stabilize. In the test shown there is a 2nd board (low power) on top with some 12 mm separation, but similar oscillations also happen without.


I see similar fluctuations with an NTC sensor in the same size case.
It this normal to get so much instability in the temperature ?
Are there good methods to avoid it ?

[Andreas]

Does this correlate with breathing?

[Kleinstein]

There is no correlation with beathing (also happens when nobody in the room) and it also happend on a very low wind day, so not infra-sound from wind power.
It seems to get a bit worse when the temperature is otherwise stable - though also hard to see when a lot of drift. 

[miro123]

In one article Bob Dobkin menshonen that "Air tight" is the must for precision voltage measurements.
My experiance is that air tight is more  important than insulation.I still equalize the air pressure with small goretex membrane.
Do you see the same picture with NTC?  NTC have 10x more sensitivity than diode

[Echo88]

Is it an oscillating process which correlates with the current consumption of a certain part on your pcb, so it could be isolated by measuring which part is the oscillating offender? I found the current measurement method quite useful when judging oven stability, not relying on thermal sensors with their delay and thermal capacity.

[Kleinstein]

I don't think there is much variation in the power consumption, especially not at that time scale and not near the sensors. The part next to the diode are 74HCT4094 seeing no update of data, DG408 with one seeing an update eveny 20 ms in a regular way (used to measure the diode signal).
Overall the mian PCB is relatively low power (~1.7 W over 90x190 mm²).

I see similar variations with the NTC, though the NTC is more to the edge of the PCB, while the diode is more to the center. Another diode at the same place also showed similar oscillations.  I have however currently no easy way to measure them both at the same time, to see if they are correlated - this looks like a good idea. So a reason to look at the SW again.

The outer case (21x23x9 cm³ size metal cooky box) is close to air tight (still some 3x2 mm² hole at a cable) quite a bit away from the sensors.

Looking a bit more in what did help and what not, there is the possibility that the thermal oscillation of starting away from the main PCB with a 5 V regualtor that produces some 1.3 W of heat, running a bit warm with an only 35x60 mm² heat sink.  Part of the problem can be heat tranfer via convection to the relatively cool top side. The bottom sitting on the desk will not loose much heat. So the box may have an unfavorable temperature gradient.

[iMo]

I saw such a turbulence (though not nice periodic one) in my meter too - what helped was to isolate the box inside with a thin foam. Try to cover the inner walls of the box with thin isolation material. I guess those are the thermal waves bouncing between the metal walls and the pcb..

[tszaboo]

I heard a story where a precision measurement was upset by a passing tram in front of the building.
When I was doing similar 24h measurements, I could tell when someone was in the same room, or when the air conditioning was on (I had an isolated box), and I was measuring voltage and not temperature. And depending on your setup, maybe it's your multimeter doing it.

[Kleinstein]

I am just running an extra test with the NTC and tilting the box a little. Just some 5 deg tilt changes a lot. One direction the oscillation stoped (concentrated heat source at the bottom) and other direction the frequency changed.

Changing the raw input voltage changes the heat at the regulator. Going from a nominal 7.5 V to 12 V (easy to switch) the osciallation amplitude goes up ( ~3-4 x).
So this points to the regulator as the driving force.

[guenthert]

Those oscillations are only 5mK pk-pk.  Might that be self-heating?  In the case of the NTC, perhaps one with higher nominal value (and hence smaller current) might show less of it.

Can the board be mounted vertically?  Ideally that would yield a laminar air flow bottom to top and a constant temperature (gradient), instead of hot-air pockets/bubbles/oscillations.  Just something I would try to gather more information.

[iMo]

The PCB has some bodges, but except for the extra HC74 this is not too bad.  One of the NOMCA arrays is already includes in the PCB layout (though not very good - a 20 K version and using 2 of the resistors in series is probably better (less excess noise)).
It would likely still need an update to the PCB to have 2 (maybe 3) x NOMCA resistors, an updated buffer, a few more SMD parts, include the bodges and a connector for an optional external reference.

..
b) then be focused at a more "thru-hole" lower density board, with more space on it, which allows easy routing and good grounding and decoupling, for example a 2 layers pcb 100x100mm (or even 100x160mm) of size (with four 3.2mm holes at the corners), without the power supply on it (ie. it means only with +15V/gnd1, -15V/gnd2 and 5V/gnd3 connectors), it also removes a significant heat source off the ADC board. You may need to make some milling as well (around the LM399, and/or around temperature/leakage sensitive parts). You need space.
..

The Bold font comes from the original post..

[thermistor-guy]

... So this points to the regulator as the driving force.

The regulator could be driving convection currents, directly and through the PCB (PCB acting as a secondary heater).

That's consistent with the oscillation period in your original plot; the suppressing effect of a cloth; and the variation with circuit orientation.

[maxwell3e10]

I have seen very similar effect here
https://www.eevblog.com/forum/metrology/beautiful-convection-oscillations/msg4371031
 
I believe it is due to a convection cell. Turning the box upside down, so that the empty side of the circuit board is on top and is close to the top wall of the enclosure seems to eliminate the effect in the most reliable way. One can calculate the Rayleigh number, if its larger than critical and heat source is on the bottom, it causes a convection instability. The Rayleigh number is proportional to L^3, so having top surface of the enclosure as low as possible reduces the effect. But due to varying height of components it's hard to have a low top surface on the component side of the circuit board.

[Kleinstein]

The unstable / oscillating convection cell seems very plausible. With some 9 cm that box is rather large and the relatively hot regulator is a strong local source, in my case about at half the hight. The disturbance than seems to get to the sensor via variations in the air flow, even over quite some distance.

Having the PCB upside down may be a good solution for the final set-up, but it is a bit unconvenient for testing, unless one has enough test points available on the other side.
With hot THT parts there could also be trouble, as more hot air could get trapped. For the current setup this would be mainly the LM399.

A first try for a fix would be to mount the regulator heat sink to the case, more on the upper end and this way start with less heat flow via convection.
Less hight, or the PCBs mouted further up could also be a good idea.

Mounting the regulator on the top cover is just a bit unconventient - otherwise probably a good solution.

[Kleinstein]

The Rayleigh number and Rayleig-bernard convection are for the occorance of a stable convection patter. This would not be that bad and in the ideal simple theroy the flow would be steady and stable.
The problem is when the flow gets unstable / turbulent.

The occourance of oscillatin seem to be a known effect but difficult and still subject of research. As an example I found an article:
https://arxiv.org/abs/0806.4882

[tszaboo]

The Rayleigh number and Rayleig-bernard convection are for the occorance of a stable convection patter. This would not be that bad and in the ideal simple theroy the flow would be steady and stable.
The problem is when the flow gets unstable / turbulent.

The occourance of oscillatin seem to be a known effect but difficult and still subject of research. As an example I found an article:
https://arxiv.org/abs/0806.4882

And so Kleinstein descended to the fluid dynamics rabbit hole, only bringing coffe and pizza to the journey...

[Henrik_V]

Just did some measurements about stability of a laser wavelength..  using a ASL F250 and two Pt100.
Found some nice correlations between room temperature +-200mK and the 9th digit of a better telco diode laser   but that was with a 45min periode ...

So: Have you monitored the surrounding conditions (temp) ?
Those little mK are very sensitive

But 5-8 seconds periode .mmh..   if you replace the NTC with a good resisitor do you have a stable reading?

[Kleinstein]

With a resistor I get a stable reading, though the last one I tested as a direct 1:1 replacement was only 100 ppm/K grade - so a little slow drift, but not much.
I also saw corresponding variations with a diode in place of the NTC, and little variations with a low TC Zener. The NTC also shows a comparable signal with a lower current (10 µA and 30 µA instead of 100 µA used for most tests).  So I have no doubt that it is really a temperature variation.

Changes from the outside (e.g. touch the case, light diretly on top) are visible too, but slow and with some delay. That is more 1 minute and longer and no easy way to get a 5-10 second periodic signal from the outside. It takes quite some warm up time to get the temperature stable enough to see the small variation directly and not have them burried in the drift.
A fast thermal effect may come via the air pressure, e.g. from wind gusts or closing of doors, but that is usually not that periodic and not much on a low wind day.


Unstable convection (as regulator oscillation or more chaotic) seens to be not that uncommon and not that easy to predict.  The amplitude of the variations is still moderate, but when looking for very low noise in the LF region it can be an issue.
Chances are other ciruits, especially when running a bit hot could have similar temperature fluctuations.

[dietert1]

Someone should look at the heater current of a LM399 or the like with a precision meter. Who knows. I have that ADR1399 evaluation kit with the heater running at about 600 mW.
Depending on the circuit NTCs and diodes can be unstable in the sense of heating more the hotter they are while a reference oven will rather damp nearby temperature fluctuations.

Regards, Dieter

[Kleinstein]

When operated at a more or less constant current the diode as a sensor would not get thermally unstable. The diodes have a thermal run-away problem when working at fixed voltage or with low resistance. As a sensor there is a large resistor (some 220 K to 15 V if I sticked to my plan) in series. In theory the NTC could get unstable, but the test current is too low (100 µA for most of the tests) for that. Unstable would be a run-away to one direction, not oscillation.

The LM399 is significan heat source - an extra cover just on the reference did not change the temperature oscillation (I have not checked the heater current).  The TO46 case version should need some 200 mW. In my case it is on an extra adapter PCB ( had the footprint upside down  ) with with an extra connector and some 15 mm higher than normal.
I had though about oscillation in the temperature regulation of the reference - but that would have also effected normal voltage measurements in a significant way.

[miro123]

In theory the NTC could get unstable, but the test current is too low (100 µA for most of the tests) for that. Unstable would be a run-away to one direction, not oscillation.

There are many way to connect NTC. I prefer to use the constant current mode provided by ADS1220/ AD1256. It makes self heating easy to compensate.

I've run a simple Jupyter notebook simulation to show the difference between serial connection of R1=10, R2=10kohm NTC
second screenshot is ntc powered @100uA.

[dietert1]

Another method is shown in the sketch appended. I built this in 2009 for a Geller AD587 board. The inner oven is the NTC itself, glued to the reference chip. The bridge operates at maximum gain and bandwidth, limited only by NTC self heating. The bridge output level serves as temperature sensor for adjusting an outer thermostat. In steady state the NTC uses its self heating to operate at constant resistance (= constant temperature).

Regards, Dieter

[MK]

One product I worked on was trying to get a millidegree or so stability, but was dissipating about 20mW in a 1-10mW thermistor, self heating was raising its temperature well above the pcb it was hoping to control. Just by dropping the drive voltage to the bridge circuit was enough to make the stability adequate

[miro123]

Another method is shown in the sketch appended. I built this in 2009 for a Geller AD587 board. The inner oven is the NTC itself, glued to the reference chip. The bridge operates at maximum gain and bandwidth, limited only by NTC self heating. The bridge output level serves as temperature sensor for adjusting an outer thermostat. In steady state the NTC uses its self heating to operate at constant resistance (= constant temperature).

Regards, Dieter

I enjoy to read this post. You build dual loop SIMO controller with analog components!.  I wonder can you build my TC oven with analog components.
 - MIMO controller - https://en.wikibooks.org/wiki/Control_Systems/MIMO_Systems
  - Feed forward component  for outer loop based on ambient temperature and humidity
  - state machine - states IDLE, SETTLE_T_UP, SETTLE_T_DOWN , STEADY_STATE - it changes the PI coefficients and notch filters
  - Debug interface to trace controller, sensors and actuators internal data.

@Kleinstein - lessons learned from TC oven - try to reduce any fluid and air gaps as much as possible Not properly designed air and oil baths resulted in temperature noise. Unfortunately that was the lessons learned mostly on the hard way. Happly HPM7177 thermal design become available and I learned a lot from it.

[MegaVolt]

Happly HPM7177 thermal design become available and I learned a lot from it.

Can you share a link? I seem to have missed it