Unexplained daily cyclic variation in an electrical measurement

[Alan_B]

I have set up an experiment which measures the variation of small currents (10 to 100 nA) with time. My samples are some treated diodes held under reverse bias at elevated temperature (50 C). The data shows a clear daily cyclic variation in all the measured 'leakage' currents (I have 8 diodes measured sequentially) and I am stumped to know why. The lekage current peaks at around 10 PM and reaches a minimum at around noon. The maximum current is up to 8% greater than the minimum. The "waveform" of the variation is broadly sinusoidal but distorted by the fact that the rise time is ~10 hours whilst the fall time is ~14 hours.

 

The diode leakage current is particularly sensitive to temperature so the most likely explanation is that there is some crosstalk influencing the thermocouple reading causing the actual temperature to vary although the controller 'thinks' it is held constant. But what could do this and have a daily rhythm? I suppose the room temperature might have a weak influence but I believe this varies much more randomly according to the outside temperature and whether or not the door is open.

 

I know that demand for electricity peaks during working hours and is minimum overnight but even if this were to influence the mains voltage and this could somehow influence my data, it is hours out of phase with what I see. So any ideas about what might be causing my "circadian rhythm" in something that is not biological?

[swansont]

You don't mention the circuit details, which might be important.

 

The earth's magnetic field fluctuates over the course of a day. And while I doubt that a simple circuit would be sensitive enough, I know of experiments where the effect of the number of cars in the parking lot mattered. An experiment in Paris had to be run early in the morning, when the subway wasn't running. (both owing to magnetic effects, but vibration could also have been an issue)

 

If it was the earth's field variation, I suppose one could check this by changing the orientation of the circuit, so that it did not envelop field lines from the earth. Or by bringing an external magnet close by, if one isn't as into subtle measurements.

 

Similarly, you could check temperature directly. Humidity could also be a possible culprit, as it can affect capacitances.

[Klaynos]

Are you in a residential area?

 

10 pm peak, 12 noon minimum is similar to domestic power usage in since areas.

 

Details on diernal variation of the earth's electric field are easily found in the literature.

 

Do you see any affect in convective weather?

[Sensei]

IMHO you should:

- start recording voltage, to see if there is correlation.

- try using voltage stabilizer (Zener diode f.e.)

- start recording temperature (f.e. place thermometer and digital camera record its value for a start).

- try using your circuit with batteries/accumulators to check whether there is difference or not.

- move your apparatus to different room/apartment/city to see if there is difference.

 

The more you will gather data, the less guessing.

[John Cuthber]

How are you measuring the voltage?
If you have a computer set up to measure the current regularly, and there's any mains hum on that signal you could end up with an aliasing error that reflects the drift in mains frequency with power use and thus time of day.

Is it possible to run the system that's providing teh bias from a battery?

If you are doing, is it possible to run as near as possible, everything from batteries and screen the system.

[Enthalpy]

Such questions are typically difficult to answer remotely... (Educated) experimenting tends to show the cause.

 

I don't believe the geomagnetic field can have much influence. But RF interferences yes, if the preamplifier doesn't reject them enough, and at 10nA this happens easily. My preferred one. And mains noise too, as JC said. 10pm and noon could correspond to fluorescent lights. How is the experiment shielded? Did you try to add a ground plane or aluminium foil?

 

The 10-14h response time is unexpected. Is it a software integration time? At the analog input, it would mean 1mC stored somewhere, not likely. And I don't guess the usefulness of a very long response time to measure 10nA; a few seconds use to suffice.

 

Depending on the diodes' package, you may pick some light at the diode or somewhere at the amplifier. This was critical when germanium diodes had just black paint on a glass package; nowadays it shouldn't happen. But try to direct a flashlight on your experiment.

 

Seebeck effect (rather at the amplifier than at the diode) can spoil small voltages easily. Though, if you measure the current with a 10Mohm transimpedance amplifier, 2nA variation means already 20mV error, too much for Seebeck.

 

How do you stabilize the +50°C? A thermocouple measures only temperature differences, so if the room's temperature fluctuates, and the thermocouple drives the oven, the 50°C will fluctuate too. At 0.55eV activation energy, 8% needs 1K variation. And possibly, the thermocouple or its amplifier pick up noise, since the voltage is very small; any silicon temperature sensor is better, when the temperature range permits it.

 

You might have some leakage current at the experiment, especially where the +-50V conductors are near, and such a leakage is sensitive to moisture which varies over a day, but for instance 2nA is already an important leakage. If you did take standard precautions (degreasing, varnish, guard rings) then leakage should be 100-1000* smaller than that. But if you didn't, this explanation gets credible.

[Alan_B]

Thank you for these comments; I was not able to reply more promptly. Unfortunately the test equipment was built by somebody else so I am not intimately acquainted with the details of its construction.

 

The key question is what could cause a diurnal variation. Yes, lots of things could be wrong but what could be diurnally wrong? The diodes have opaque covers and are inside a closed miniature oven so we can eliminate light. Likewise moisture in the atmosphere inside the chamber will not not, surely, vary in a diurnal manner. If you recall the characteristics of a diode you will remember that the reverse bias current is not very sensitive to voltage so even if the voltage were badly controlled (It is set to -10V) it should not have very much effect. Reading are taken once every two minutes so I don't see that it is a software problem. The temperature is maintained via a PID controller but unless that has some odd firmware I see no reason why it should develop a daily rhythm. My laboratory is on the outskirts of a residential area, becoming rural, with no unusual vibrations. However I have just found some evidence to suggest my data may be sensitive to the room temperature.

 

There is some work going on elsewhere in the building today and the outside door down the corridor is open. The leakage currents I am measuring have been dropping steadily from the time I arrived and opened the door to the room where the measurements are taking place; this door is otherwise closed. This provides the best evidence to date that there is some residual sensitivity to room temperature as the room has been cooling since I arrived. I measure and control the temperature in the device chamber with a thermocouple using electronic cold junction compensation so perhaps that gives me some small residual sensitivity to the ambient room temperature? And at this time of year there is no control of the room temperature (neither air conditioning nor heating).

 

I am in a single storey building with a pitched roof. The roof warms up during the day but the insulation between the roof and the loft space and then between the loft space and the rooms below might mean that the interior temperature of the building varies out of phase with the external temperature of the roof. So perhaps my daily cyclic variation is just a reflection of a small variation in the ambient room temperature and a flaw in my equipment causes the actual controlled temperature to be slightly sensitive to room temperature?

 

To test this hypothesis I need to set up more equipment to log the room temperature and, if possible, get a more accurate reading of the temperature in the chamber.

[John Cuthber]

"Reading are taken once every two minutes so I don't see that it is a software problem. "

I don't see it as a software problem either.

But I still wonder if you have an aliasing error.

Did you write the S/W that's making the measurements?
If so, could you humour me and have it randomly add or subtract a fraction of a second to the interval from one sample to the next?

 

[Enthalpy]

Add one sensor in the oven to monitor the temperature there. If you see 1K variation, it explains the current change, and the regulation is broken or badly done.

 

"the reverse current is not very sensitive to voltage", that's true only for well-built PIN high-voltage diodes far from breakdown. The reverse current is often quite sensitive to the voltage. In case you imagine the reverse current to be I_s as in I_s*exp(qV/kT), forget it, it's b**cks.

 

I don't see any reason to take hours nor minutes to measure 10nA: <<1s suffices. While such a measure isn't the most difficult one Mankind has done, it does need an engineer with decent experience for analog electronics. My best bet up to now is that the setup is botched, and you pick up radio noise and powerline noise. So:

- Could you post a picture of the experiment, so we can assess how good shielding is?

- Could be useful too: the electric diagram of the experiment.

 

And do you really need a computer and software to measure a reverse current? This introduces more noise and errors.

[Enthalpy]

A circuit to measure 10 to 100nA leakage is easy. It takes a banal BiCmos op amp at room temperature (not in the oven) and a handful of components. Only cabling matters.

 

Everything can be cabled on a Veroboard with a ground plane (without sockets of course). After welding, clean the circuit with trichoroethylene and don't put the fingers on the tracks; varnish isn't necessary for such currents. Weld the coax cable or a coax connector on the board. No box is necessary. Check with a scope that moving the cable leaves the output calm; polyethylene cables are better for that (the ones that melt at soldering iron's heat). The ceramic 100nF must be chosen and cabled with low inductance. The circuit's output must be clean on a scope. Decouple the supply - and I like voltage regulators near the op amp.

The TLC081's input bias current makes <50pA error, the offset voltage <200pA and it's constant so the measure can zero it. The voltage and current noise of the op amp are negligible, the 10Mohm resistor contributes 0.2pA rms with the 0.01s response time here, and 30nA measured leakage contribute 0.7pA rms noise over 0.01s (without avalanche), so there's no need to integrate over hours.