Abusing MVAR calculations

The Modified Allan Variance, MVAR, is used to tell us something about the noise-processes in time/frequency data.

But you can run other data than time/frequency measurements through the algorithm, and get informative results:


The bright red line tells us that the Fluke 732A is a pretty good voltage reference, it’s dominated by white voltage noise.

The bright blue line tells us that on the other side of the A15 constant current circuit, a lot of noise have been added, and we can tell that some of it is random-walk.

The lighter two lines show similar measurements using the HP5065 built in voltage reference, the A15CR5 zener diode.

The take away from the red curves is that the Fluke 732A is more than an order of magnitude more stable than the Zener.

The take away from the blue curves is that the constant current driver adds so much noise on top of the voltage reference that half the advantage of the Fluke is lost.

Is it kosher to send voltages through MVAR ?

Mathematically: Absolutely.

Physically: Weeeeeeell…

“Flicker Phase Noise” doesn’t make sense for voltages, does it ?

But the actual power-law behind the MVAR does make physically sense in many other domains than time/frequency, but they need to be mapped onto the laws of physics for whatever experiment provided the input data, and that takes some serious thinking.

For voltages, I have found a physical model I am confident enough to use internally, but I’m not confident enough to share them.

Running power consumption timeseries through MVAR gets you some very interesting plots but to be honest, I have no idea what they mean.

Running temperature series through MVAR can tell you things about HVAC systems and local meterology which can be very useful for tuning timeconstants of PID loops.

In other words: My experience runing “random” timeseries through the MVAR is very productive, and if nothing else the plot will usually give you a good idea at what time-scales the dominant noise-process changes.

You may not know what those processes are or what the slow means, but knowing the time-scale of relevance can help you identify them.


TVB pointed out that I’m still hung over from yesterdays party, it is of course MDEV not MVAR I have plotted.