Battery estimate calculation, nRF9160. Community discussion welcomed!

[Torje]

Hello everyone,

I'm trying to do some battery calculations for our coming asset tracker based on the nRF9160. I would very much like some discussion on the topic, as a sanity check!
As a background, all the app is doing is waking up, getting a fix, sending it, going to sleep. Once per day.

I've used the Power Profiler kit 2 (PPK2) to measure:
- Current floor (PSM)
- Stuff happening before GPS search
- GPS search
- Stuff happening after GPS fix

I'm trying to make a worst case calculation, using 300 seconds of GPS searching. Connecting to LTE and cloud and getting AGPS before, sending data and disconnecting from cloud after. I'm basing my calculations on a 5000mAH battery, and doing the calculations both in AH and Coulomb.

The calculations now looks like this:

Simplifications: I am allowing overlap with current floor time as I consider it negligible. "Per min-AH" cells are in ampere minutes.

If I assume 30 seconds per fix the device lasts for 21 years. This seems too good to be true, so I think I've missed something.

My main concerns are: Am I way off, or onto something? Is there something I haven't taken into consideration? Are the estimates "valid"? Does anyone have any experience or knowledge to share? Based on this, it seems communication protocols etc. matters very little compared to actually getting the fix faster

Please fire any questions at me, and look at this post with the most critical eyes. It would be a great help.

Thank you all!

Kind regards,

Torje

[Achim Kraus]

If you calculate with a 4.66µA "current floor", I guess you should also consider the self-discharging of the battery. Especially, if you run over a couple of years, even 2-3% per year maybe more than nothing. (Just to mention: using a LiPo Accu comes with 2-3% per month and so runtime longer than a year discharge the battery easily by a third.)

Your calculation didn't consider "LTE radio times" (NB-IoT or LTE-M). Once you consider that as well, you need also to take "network searches" into your account. E.g. in my experience, an message exchange may take about 100mC, but a network search easy up to 1800mC (depends on a lot of parameter). If your device is moving, I guess you need to consider such network searches.

As longer your device is in the field, it may requires to consider FOTAs, that also drains the battery of such a device.

Anyway, if your device is frequently exposes some light, there are also ideas to use a solar panel. I don't have experience with that and the long-term usability.

[Torje]

Hello Achim,

Thank you for some important considerations I had not done. I will include estimates for discharge as well.

The radio times are included in the "pre" and "posts". Its until psm floor is reached again, so this should be taken into account.

My main concern is perhaps that my estimates are way too optimistic. I'm getting tens of years of battery life, and it doesn't seem realistic. I will also include measurements for FOTAs, thanks for the tip.

[Achim Kraus]

> The radio times are included in the "pre" and "posts". 

Unfortunately, I guess, this still doesn't consider "network searches", especially, if the device is moving.

> I'm getting tens of years of battery life, and it doesn't seem realistic.

There are a lot of calculations, which also results in years with one wakeup a day. Until you run the device for a couple of weeks, I would not relay too much on the results.

[Torje]

Hello Achim,

And thanks again for your perspectives. We did some measurements where we sent the battery voltage every day for a month or two. It was hard to estimate something based on these, because we don't know exactly where on the non-linear curve we are. Do you have any suggestions as how to estimate battery life based on a few weeks of data? We will get our custom hardware and batteries in a couple of weeks, and then we will test (among other things) the battery life. But I find it hard to consider the validity of estimated battery life based on just a few weeks, but t here might be some tricks to this I don't know about?

On a different note, it seems to me that which protocols (mqtt, rest, coap etc) we use isn't really that decisive, as they amount for quite a small bulk of the total charge used (especially in my scenario). I have noticed that you are quite into these matters. Do you have any wisdom to share regarding this?

Thank you again!

Kind regards,

Torje

[Achim Kraus]

> It was hard to estimate something based on these, because we don't know exactly where on the non-linear curve we are.

Yes. I tried to compensate that with measuring the curves with the GNSS close to always on. That drains the battery of a Thingy:91 in almost a day and you get your chart. There still may be a difference, but using that chart is a first step to a better result.

> Do you have any suggestions as how to estimate battery life based on a few weeks of data?

Batteries are designed to offer a equal voltage for the most time. And forecasts based on the voltage in that period are therefore not that precise. What may help is to use much smaller batteries of the "same type". Or other "equipment", which is able to measure the current over a longer time. Assuming a moving device, that may be not too easy (see Count those Coulombs for Lithium Battery Fitness! ).

Anyway, I personally assume, that before a product runs for 10 years, it runs for 2. And I guess, many products will stop to run, because of other issues. Including surprising software bugs, which then also may drain the battery. I would also not stick too much to something as "grant to run for 10 years", you never know, what the MNOs will do. Or, if a future bugfix comes with an different "power consumption" in your case.

[Achim Kraus]

One final consideration:

If your device is moving, it may also be affected by different environment conditions, temperatures and humidity. You may also check, if that environment has influence on the lifetime.