Working offline - looks like very low power consumption

I’ve been testing deepsleep and wake without a server connection. So far 10,000 wakes offline and 200 online and the battery voltage of 5xAA’s has only dropped 0.4volts. The temp in my house has dropped from 38 to 30 degrees (Celsius) over the same period so that might even account for the voltage drop, I’ll know later today when the temp goes back up.

25,000 offline wakes with 550 wifi connects to report results. No drop in battery voltage. Awesome.

How long are your wakes? If you look in the nora reference design here:

…you’ll see that with 2xAA lithium (~1.6-1.7v each) we got 80,000 wifi connects. Standard alkalines were more like 60,000.

From 5xAA alkaline you should be getting in the region of 150,000 wifi connects, though obviously this depends on how long your connects are, what DCDC you’re using, etc.

Hi Hugo, I feel like I’m taking to Linus Torvalds, thanks for the IMP platform.

Using the April board and waking to wifi every 30 seconds, I am indeed getting ~150,000 wakes, which means approx two weeks, awesome but still not long enough.

The Nora tests were for waking with wifi. I’m now looking at wakingwithout wifi. Early indications are that only uploading data when required, rather than every wake will extend the battery life by a huge margin. Of course it depends on how often data is uploaded but I’m looking at 10x improvement, meaning a battery life to 20 weeks with over a million wakes w/o wifi. That’s awesome! A 32bit processor with TLS over wifi running for months on AA’s. Wow. I know your product is a platform, not hardware, but great job on the hardware nonetheless.

You might want to add a note to the Nora battery section for newbies that wake-without-wifi is an option and uses only a fraction of the power.


Wake-without-wifi is much more power-efficient for two reasons: firstly the obvious one that the wifi hardware itself, which draws tens or hundreds of milliamps, doesn’t get powered on; and secondly the perhaps-less-obvious one that a wifi-less wake-up will end up running the CPU for far less time. An imp can wake up, check a sensor, perhaps write some data to the “nv” table, and go to sleep again in a few tens of milliseconds. Connecting to wifi and then to the cloud servers, on the other hand, takes about a second absolute best case and often more like two to three seconds. So on a CPU basis alone, a wifi wake-up can cost 100x the battery life of a wifi-less wake-up.


250,000 offline wakes and 5,000 online wakes. The AA batteries are holding up fine.

Is there someway to move from deepsleep directly to server.disconnect?
Or should I wakeup + server.diconnect ?

An imp doesn’t try to connect when it wakes up from a deepsleep - or rather, it is lazy about trying to connect.

After waking from a deepsleep, the imp will wait until it needs to connect to connect (for example, a call to server.log will cause the imp to connect).

Huumm…that explains a lot of things happening here… thanks a lot!

I am about to make something that requires an imp to run on batteries too, but it will only wake 1-3 times daily or so, some days it might not even be triggered at all. Could I just connect two AA batteries to 3v3, or would it be better with more batteries and then use an April board to regulate? The project is going to be in an plastic enclosure outside, so temperature will vary a lot.

The big issue with 2xAA (alkaline) batteries is that they discharge below the required voltage really quickly. If you want to go the 2xAA route, I recommend these, which we’ve used in a few designs. The downside is that if you give the board to someone else and they replace them with regular AA batteries, it won’t last very long.

Here’s the datasheet for the batteries - it has a comparison to alkaline AAs on the first page which shows the difference :slight_smile:

Other option will provide some room is to run the vcc rail with 2.7V . With a regulator like TPS622316DRYT (Digikey 296-27850-1-ND), I am testing this option with an IMP02 working just fine, it reads 2.72V steady. In my application could not fell any speed/performance difference. Of course other sensor should be within that range…