Incorrect voltage measurements - Hardware Issue

[vinzenzm]

We have a fleet of many Crazyflie2.1's, which are equipped with QI chargers.

In our setup we currently wait until the voltage of a crazyflie rises above a certain threshold, before we allow it to take off.
For landing we have another threshold which decides, if we can keep flying or have to land.

However while some crazyflies fly for several minutes with this technique, others merely fly for a few seconds.

I did quite elaborate testing on this, and have managed to find out the following:

The voltage measurement of the NRF51 on the problematic crazyflies is incorrect.

• Its not the batteries, as did loads of testing with new ones/used ones, as well as swapping them from "good" crazyflies to "bad" crazyflies.

• I know the BQ24075 charge controller is fine on all of them, because after it stops (CHG-Pin) I can measure the battery with a multimeter which reports 4.2V, crazyflie reports 4.17V ( Good crazyflies report 4.2V in this case). I also charged a battery with a "bad" crazyflie and then used it in a good crazyflie, which worked without problems and gave me very good flight times.

• I measured all Voltage divider Resistors around the nrf51 they are also fine.

• I currently have a deck with an ATTIN3226 on it, which is directly powered by VCOM (which is the Battery Voltage) and was able to measure the input Voltage and report it back to the crazyflie. I managed to produce the following graph with it:

In orange: the voltage measured by the ATTINY; In blue: the voltage measured by the NRF51 (pm.vbat)
In the recording I took off, landed and then took off again and stayed airborne until the battery "died".

You can clearly see, that the NRF reports a lower voltage than the ATTINY.
The difference is lower during no-load and increases when load is applied.

• I also used a lab-bench power supply and tested this no-load offset. This particular crazyflie had an offset of 0.03 Volt. I wasn't able to test under load, however the 30mV seem to align with the ATTINY measurement, which lets me trust its measurements.

• The ADC on the nrf seems to be fine as well, because the charge current is measured with the same peripheral, and both good and bad crazyflies report 0.90 A when QI-fastcharging (both with minor fluctuations)

• The crazyflies which have this issue are the oldest ones / most flown ones, and might even had the overtemperature protection threshold (on NRF51) higher than usual for some time (I cant remember setting it to more than 80.0 however).

I am therefore pretty sure this is not a software issue.
However which component on the PCB could be faulty here?
The no-load vs load difference hints at additional resistance somewhere.
But this has to be close to the NRF, because the general power is fine.

Maybe you have had similar experiences with old crazyflies?
Maybe someone with more electrical engineering experience can explain this phenomenon?

[gemenerik]

Thanks for sharing, we'll try to get back to you next week

[tobbeanton]

Thank you for a very extensive report!

It is a very strange behavior so a lot of guessing now. I'm thinking that the nRF input pin/ADC is broken/damaged and is effecting the reading. The resistor divider used to measure the battery has a very high resistance and any abnormality on the pin could effect the reading.

If you would like to investigate it further you could try to reduce the sampling rate in the code and check if that has any affect. You could also try to replace the divider resistors with lover values to get a less sensitive measuring point. A side effect is that the Crazyflie will consume more energy in off mode.

[vinzenzm]

Thanks for your input and thoughts on this. I investigated further and actually found answers.
As always its not a single reason, but multiple effects which together produce these results:

1. The main culprit under load was the battery connector. After cleaning them with isopropanol, and opening up the contact (Remove plastic housing, rub contacts with alcohol, split the tip (which should not be pressed together tightly), reassemble). This improved things a lot, and is even measurable in the battery test. Improving the test by sometimes more than 200mV.
   So actually it should be called a battery and connector test ^^.
   This finding also aligns with the "bad" ones being the oldest, because back in the days we did not use QI-Chargers but rather swapped batteries a lot, which leads to degradation of the connector (Especially the "springy tip").
2. The voltage divider has a tiny impact on static measurement offset (for the ones i tested this was up to 3 counts on the adc, which is 0.01V in the end)
3. Some crazyflies measure higher than the actual voltage, for these its actually the internal voltage reference of the NRF or the ADC that is slighty off. For this I configured the ADC to read VDD/3 against VGB. This should result in 852/853 ADC counts. However some measure up to 10 counts more here. (I measured the VDD (3V regulator on board) and it was spot on). These +10 counts can increase the reading by 0.04V.

That being said we will check our battery connectors now regularly/ not swap batteries before we know the connector is fine.
As for the 2nd effect (voltage divider) there is probably nothing we can do about, the error is small anyways.
For the 3rd effect we might implement a correction into our nrf-firmware (read VDD/3 against VBG regularly, when we are sure VDD is stable (VBAT > 3.5V), then apply this offset to vbat (and iset) measurement.

Regarding my Attiny measurements, the Tiny seems to read high in certain scenarios, didn't investigate further.

You also suggested lowering the sampling rate, this didn't seem to affect the measurement. I tested this on the oscilloscope and for the 100Hz it is actually pretty close, that the capacitor has enough time to recharge (But that's also mentioned in the firmware).

[tobbeanton]

Well investigated and thanks for sharing! I'm don't know your usecase but if cleaning the contacts is a haste it might be possible to switch to the gold plated ones and the 350mAh battery.