Freaklabs - Open Source Zigbee Blog - 2009-11-29 Status Update

Home

Blog

Misc

2009-11-29 Status Update

| Print |

Written by Akiba

Sunday, 29 November 2009

I didn’t get a chance (or actually I forgot) to wish everyone a Happy Thanksgiving. I’ll be doing my end of year reflection soon and I already know that I have a lot of things to be thankful for.

This Thanksgiving holiday wasn’t much of a holiday for me though. Since I spent so much time helping Tokyo Hackerspace prepare for the MAKE event, the holiday was spent trying to catch up with my designs and get things ready to roll out for the store. I’ve already decided to put off opening the store for this year since the holidays are quickly approaching and the last thing you’ll want to be doing over the holidays is prototyping wireless sensor circuits and writing code. However I do need to get over my mental blocks and insecurities and just get the damn thing open. You know things are bad when your wife starts pressuring you to open the store.

I already have the first wave of products ready to release. It’s going to consist of the 128 kB AVR board, the AT86RF230 radio board, and the prototyping breadboard. I’ll also have some other items like hi-gain directional antennas and such. It’s a pretty sparse initial offering, but there’s a reason for it.

I was initially going to release the Chibi board as well, but the time at MAKE and working with the scrolling LED board I designed taught me a couple of lessons. One of the things that I wasn’t too happy with is that Chibi is running a naked battery. By naked, I mean that it can run off a battery, but it’s running directly off of the battery with no power supply conditioning. The initial version of the scrolling LED boards I designed for Tokyo Hackerspace were doing the same thing, but I didn’t like the fact that the batteries dropped in voltage. Those nice flat discharge curves you see from the manufacturers aren’t exactly true. Those curves are flat so it looks like batteries retain their voltage until they’re almost fully discharged. Unfortunately, as those of you with cars might understand, when batteries are under load, the voltage does indeed drop as it discharges and it can be up to 1V or more. That’s why I wasn’t so happy to see the LED displays dimming out when the battery manufacturer had such a nice flat discharge curve for their lithium coin cells.

For LED displays, its not very critical because the display just starts getting dim, but for sensor circuits, its horrible because it may corrupt the sensor data. Also, MCUs will usually shut off after a certain voltage (the brownout voltage). In the case of AVRs, the lowest you can go is 2.7V after which, the MCU will stop functioning. This is under load, however so a 3V lithium cell that goes down to 2.7V probably still retains more than 50% of its energy.

Because of this, I’ve been testing out a boost converter that the Chibi battery would feed into. I’m using a PFM (pulse frequency modulation) boost converter IC that can go down to 0.7V and keep the voltage output constant at 5V or 3.3V. The problem was that I couldn’t decide on which voltage I wanted. There’s a big tradeoff here with two factors that are important for wireless sensing. Choosing 5V would allow me to drop the voltage to a very stable 3.3V via the regulator. Although this wastes power, having a stable 3.3V supply would yield the most accurate sensor measurements. The other choice was to use a 3.3V boost converter and run the Chibi board off of that supply only. This uses much less power but there would be ripple in the power supply which would affect the accuracy of sensor measurements.

So for the past couple of days, I’ve been playing with the boost circuit and measuring how much of a hit to accuracy a 3.3V boost would be, and how much of hit to power consumption a 5V boost converter would be. I was getting current consumption of ~10 mA running the 3.3V boost converter and ~17 mA with the 5V boost. In other words, using the 5V boost and dropping the voltage would result in an increase in current consumption of around 70%+. I suspect that it’ll be pretty proportional so if the 3.3V circuit consumes 50 mA, then the 5V one would consume about 85 (not counting changes in the boost efficiency). Those kind of numbers are pretty expensive in terms of power so then I checked how much of a hit to accuracy I would get. You can see two screenshots from my oscilloscope in the pictures below. Both shots are of the 3.3V supply and the one with the ripple is from the 3.3V converter. This was with both converters being loaded down by 70 mA. I used LEDs on the breadboard to add a realistic current load. As you can see, there was about a 20mV difference in supply stability between the two. Actually, that was after I had optimized the 3.3V boost circuit. It was much worse but I used a larger inductor, changed the diode, and added big ol’ 100uF tantalum caps on the input and output to get down to 20mV difference.

Basically, I’ve decided to take the easy way out. I’m going to add a 3.3V boost circuit to the Chibi board since it’ll improve the sensor accuracy plus improve the battery life by allowing it to fully drain the battery. For those that don’t like losing the 20mV of accuracy to power supply ripple, I’m going to have an optional board with a 5V boost converter and battery holder that plugs into the DC jack (usually used for the wall wart). This would allow the 5V converter output with ripple to go to the regulator and produce a flatter 3.3V supply.

Incidentally, all of this experimentation with power supplies is also helping to prep my designs for when my boards go solar/supercap + battery-backed. Solar supplies are notoriously unstable which is why power conditioning circuitry is essential. Once I get those going, it'll be possible for me to put together some real environmental sensors where the power supply might actually outlast the enclosure.

Also, I should mention that I designed most of my second wave of boards and they’re ready to be sent out for the initial prototype fab. These boards will consist of: