# Computing stuff tied to the physical world

## Could a coin cell be enough?

In Musings on Jul 29, 2015 at 00:01

To state the obvious: small wireless sensor nodes should be small and wireless. Doh.

That means battery-powered. But batteries run out. So we also want these nodes to last a while. How long? Well, if every node lasts a year, and there are a bunch of them around the house, we’ll need to replace (or recharge) some battery somewhere several times a year.

Not good.

The easy way out is a fat battery: either a decent-capacity LiPo battery pack or say three AA cells in series to provide us with a 3.6 .. 4.5V supply (depending on battery type).

But large batteries can be ugly and distracting – even a single AA battery is large when placed in plain sight on a wall in the living room, for example.

So… how far could we go on a coin cell?

Let’s define the arena a bit first, there are many types of coin cells. The smallest ones of a few mm diameter for hearing aids have only a few dozen mAh of energy at most, which is not enough as you will see shortly. Here some coin cell examples, from Wikipedia:

The most common coin cell is the CR2032 – 20 mm diameter, 3.2 mm thick. It is listed here as having a capacity of about 200 mAh:

A really fat one is the CR2477 – 24 mm diameter, 7.7 mm thick – and has a whopping 1000 mAh of capacity. It’s far less common than the CR2032, though.

These coin cells supply about 3.0V, but that voltage varies: it can be up to 3.6V unloaded (i.e. when the µC is asleep), down to 2.0V when nearly discharged. This is usually fine with today’s µCs, but we need to be careful with all the other components, and if we’re doing analog stuff then these variations can in some cases really throw a wrench into our project.

Then there are the AAA and AA batteries of 1.2 .. 1.5V each, so we’ll need at least two and sometimes even three of them to make our circuits work across their lifetimes. An AAA cell of 10.5×44.5 mm has about 800..1200 mAh, whereas an AA cell of 14.5×50.5 mm has 1800..2700 mAh of energy. Note that this value doesn’t increase when placed in series!

### CR2032

Let’s see how far we could get with a CR2032 coin cell powering a µC + radio + sensors:

• one year is 365 x 24 – 8,760 hours
• one CR2032 coin cell can supply 200 mAh of energy
• it will last one year if we draw under 23 µA on average
• it will last two years if we draw under 11 µA on average
• it will last four years if we draw under 5 µA on average
• it will last ten years if we draw under 2 µA on average

An LPC8xx in deep sleep mode with its low-power wake-up timer kept running will draw about 1.1 µA when properly set up. The RFM69 draws 0.1 µA in sleep mode. That leaves us roughly a 10 µA margin for all attached sensors if we want to achieve a 2-year battery life.

This is doable. Many simple sensors for temperature, humidity, and pressure can be made to consume no more than a few µA in sleep mode. Or if they consume too much, we could tie their power supply pin to an output pin on the µC and completely remove power from them. This requires an extra I/O pin, and we’ll probably need to wait a bit longer for the chip to be ready if we have to power it up every time. No big deal – usually.

A motion sensor based on passive infrared detection (PIR) draws 60..300 µA however, so that would severely reduce the battery lifetime. Turning it off is not an option, since these sensors need about a minute to stabilise before they can be used.

Note that even a 1 MΩ resistor has a non-negligible 3 µA of constant current consumption. With ultra low-power sensor nodes, every part of the circuit needs to be carefully designed! Sometimes, unexpected consequences can have a substantial impact on battery life, such as grease, dust, or dirt accumulating on an openly exposed PCB over the years…

### Door switch

What about sensing the closure of a mechanical switch?

In that case, we can in fact put the µC into deep power down without running the wake-up timer, and let the wake-up pin bring it back to life. Now, power consumption will drop to a fraction of a microamp, and battery life of the coin cell can be increased to over a decade.

Alternately, we could use a contact-less solution, in the form of a Hall effect sensor and a small magnet. No wear, and probably easier to install and hide out of sight somewhere.

The Seiko S-5712 series, for example, draws 1..4 µA when operated at low duty cycle (measuring 5 times per second should be more than enough for a door/window sensor). Its output could be used to wake up the µC, just as with a mechanical switch. Now we’re in the 5 µA ballpark, i.e. about 4 years on a CR2032 coin cell. Quite usable!

It can pay off to carefully review all possible options – for example, if we were to instead use a reed relay as door sensor, we might well end up with the best of both worlds: total shut-off via mechanical switching, yet reliable contact-less activation via a small magnet.

The RFM69 draws from 15 to 45 mA when transmitting a packet. Yet I’m not including this in the above calculations, for good reason:

• it’s only transmitting for a few milliseconds
• … and probably less than once every few minutes, on average
• this means its duty cycle can stay well under 0.001%
• which translates to less than 0.5 µA – again: on average

Transmitting a short packet only every so often is virtually free in terms of energy requirements. It’s a hefty burst, but it simply doesn’t amount to much – literally!

### Conclusion

Aiming for wireless sensor nodes which never need to listen to incoming RF packets, and only send out brief ones very rarely, we can see that a coin cell such as the common CR2032 will be able to support nodes for several years. Assuming that the design of both hardware and software was properly done, of course.

And if the CR2032 doesn’t cut it – there’s always the CR2477 option to help us further.