Zigbee V2 Version - Battery drain

[rivnut]

Hello everyone,
first of all, big thx to rbaron for this project and to be avaiable as a open source project to rebuild.
Maybe someone can help me with a problem of battery drain with a zigbee version of this sensor.
I ordered some fully assembled sensors from jlcpcb.com and flashed the firmware with j-link. The sensors works great at all but after some days, it starts to drain the battery and the capacatiy goes down from 100% to 30% within one or two days. Can anybody help me, where I should start to search for the problem?

Image attached

Best Regards, Robert

[rbaron]

Hey @rivnut,

Does this happen with a single sensor or all of them? Could you plot the other ones as well? While at it, could you plot the temperature? If you remove the battery and measure it with a multimeter, what do you get?

I'm asking because our batt discharging model (map from voltage -> %) is pretty rough and does not take temp into account. In extreme temps, the % reading will be off.

[rivnut]

Hello @rbaron,
Thx for your fast response. When I am back at home from work I can send the requested data and measurement. This behaviour is quite the same at the other sensors (till now four) but only at this sensor (is fully assembled by jlc, at the other sensors I assembled the nRF chip by myself and only the other parts were assembled by jlc) the battery went down so extremely and then the sensor stopps working.
As I read in the forum, there is no way to export the voltage instead of the percentage?

Best regards, Robert

[rbaron]

There is a trick (that works sometimes) to read the battery voltage from the zigbee samples through Home Assistant. In the device page, click the three dots then "manage zigbee device", select "PowerConfiguration" cluster, then "battery_voltage" and tap "READ ATTRIBUTE":

This is in 100 mV units, so 29 means 2900 mV = 2.9 V in this case, and it reads 91% in the top right of the screenshot.

In this other example, 2.8-ish V (rounded) => 38%

Now a little backstory...

This is because the discharge curve of a CR2032 looks a bit like this (from energizer, assumes constant draw, constant temp):

If I remember correctly I based the CR discharge model I set up was based on an example of an old SDK version (now deprecated). It tries to linearize the curve piecewise.

In reality this mapping is not really good. If you look at a typical multi-month discharge pattern in %:

You can see a steep drop in the beginning. I think this is more a consequence of a bad battery curve model (that doesn't take into account battery current during measurement, temperature etc) than a battery drain issue. This is still at 2.8-ish V, which indicates it's working okay.

So, this is somewhat expected within reason given the current simple battery modeling. Still, a sharp drop from 100% to 20% in a couple of days is concerning.

Let me know if you manage to gather some more data from the other sensors, and also from the temperature.

[rivnut]

Hello @rbaron,
thx for your fast and large reply.
Attached you will find a chart from the last two days (battery vs temp).
Till 21.06.25 - 21:30 the sensor was inside the house on a table without contact to soil. After that I put it outside into a flower box. You can see the peak at 23.06.25 at 14:30 were the sun was shining on the sensor directly. As you mentioned you see a correspondence between temp and battery status.
After 23.06.25 - 16:30 I put it out of the flower box and put it inside the house without contact to soil anymore. The temp and batt values stabilized.
Thx for the hint with reading the power cfg cluster. Attached you will find the values of this sensor:

{
    "battery": 36.5,
    "humidity": 52.04,
    "illuminance": 0,
    "illuminance_lux": 0,
    "last_seen": "2025-06-24T08:03:18+02:00",
    "linkquality": 152,
    "soil_moisture": 0,
    "temperature": 24.71,
    "voltage": 2800
}

So the battery has 2,8V at the moment. I will have a look in the next days at the performance of the battery and will setup a test setting with 6 b-parasites - 3 indoor and 3 outdoor and capture the data for analysis.

The next week I am on holiday, but I will post the results when I am back home :-)

Maybe then we can develop a better model for the battery percentage calculation taking account of the sensor temperature.

Best regards, Robert

PS: here is the graph of the last seven days of another sensor outside of the house with sunlight impact

[rbaron]

I have one of these cheap LYWSD03MMC sensors flashed with the popular custom pvvx/ATC_MiThermometer firmware.

Out of curiosity I put it outside yesterday, and I also see a similar variation in battery voltage and %:

In their case, we can see a pretty much linear mapping from temperature, voltage and battery %. I found where it is handled in code, and it really is a linear function in the range 3.0 V - 2.2 V.

This is even "simpler" than what we do here with 4 piecewise linear approximations, but I would argue it looks smoother and doesn't have the sharp drop as our model shows in practice for our current pattern. So maybe simpler is better?

[rivnut]

Hello @rbaron,
I think it would be worth a try.
Till tomorrow I will setup my four test devices (new devices, with new batteries) - then two of them will be placed outside (with sun) and two of them inside the house.
Then I record the values for one week (because I am on vacation in alto adige (south tyrol) :-)). The I post the raw data.
If you want I can add four other devices flashed with the new variant for battery percentage calculation (but maybe I need help adapting the code).

Best regards,
Robert

[rivnut]

Testsetup ready, Batteries checked - 3,22 V

[rivnut]

Hello @rbaron,

attached you will find the results from the tests.
Sensors t1 and t2 were placed outside, t3 and t4 where placed inside the house. t1 and t2 next to each other, t3 and t4 too.

In the Excel file you will find the raw data.

Best regards,
Robert

Raw data file