Hi.
Since the battery capacity is measured in mAh, and we observ no correlation between capacity drops and voltage:
Wouldn't it be reasonable to try to find the correlation between the capacity drops and the current consumption?
Regards.

Yes it would, of course. But since the current is highly variable over time, it is not useful to just record it every five minutes like the other variables. One would have to average over a certain period of time. Even there I doubt that five minutes are not too much.

Anyway, my slow and bulky R script is not the best way to achieve this. I need to rewrite it a little to not use so much RAM. Currently it reads the whole data file just to append one line. This was the easiest way to save the file properly and add new variables on the fly.
Maybe I'll write an R script that runs in the background, collects all the data and now and then spits out a .png graph for Billboard...

We need to take a look at the current, that's crucial, right!
Some integration will then show us the battery's true capcity.
And hopefully give us a hint for the drops.

Hi.
Maybe a variable sampling rate could help: you can establish a threshold for the current so that if the current is higher than the threshold, the sampling rate should be higher.... doesn't look easy. Also, to determine the electric current threshold some kind of 'trial and error' is required....
Regards.

Yes, I thought about that as well. My approach would be so far to have a relatively high frequent loop that checks all significant parameters (charge, voltage, current, etc...) around every couple of seconds. Until one of the parameters changes (significantly), average up the current and then write one single line of data containing all variables.
This has the advantage to be able to average the highly variable electric current and hold the data file compact while maintaining all infomation.

Hi.

I had also thought about the possibility to monitor the CPU load (it should be easier than the electric current) however I have also noted that while the radio traffic is very high (WiFi and/or GSM/GPRS) the CPU load in the system doesn't grow much, although the battery certainly drains very quickly... so measurements of CPU load may not be accurate enough.
I also think you should keep in mind that file I/O processes may drain more current and influence the measurement results...

Regards.