Wednesday, August 15, 2012

My DIY Smartmeter

This one started with a power bill. One exceptionally high power bill, that is.

I had been reading and thinking about smart metering and smart grids for some time already, and when the bill hit me I decided it was time to act. But where to start? Those kill-a-watt-kind-of-things weren't an option because I needed to measure the power right at the point where the lines enter my apartment, because I didn't want to miss anything like the stove or the water heater, which don't plug into sockets, but are connected directly.

The guts of my Breakerbox:
In the lower row you can see the main breaker on the left
and the breakers for the stove and the water heater.
But I also didn't really want to mess with my breakerbox if there was any way I could avoid it. So the first things I had a look at were inductive sensors. Only two problems with that:

First, the coil or the ferrite core of the sensor has to go around the conductor you want to measure. So either the coil, the core or the conductor has to be split open to mount it. There are sensors made especially for this purpose, like these: https://www.flukso.net/content/50amp-current-clamp, so this issue could be addressed by simply throwing money at it, because those specialized sensors anren't too cheap. Actually there are even some DIY projects for these sensors out there.

The second problem is, that these sensors only measure current. We want to measure power consumption and power is the product of current and voltage. So we'd need to measure the voltage, too, if we want accurate measurements. Now this definitely isn't achievable without a physical connection to the conductor.

So there you have it: No accurate measurement without physical contact to hot wires. Just no way around it. Shit.

This meant, I couldn't do his by myself. Luckily that's not that much of a problem, because a friend of mine, who is a certified electrician, had offered to help me out.

The inevitable disclaimer:
Working with AC mains power can kill you! Don't do it unless you really know what you're doing!

This also meant, that I didn't have to awkwardly hack about with current clamps and later figure out how to calibrate and measure the whole shebang, but I now could look at an entirely different concept.

The meter(s):

There are lots of electronic powermeters out there, calibrated and uncalibrated, the main difference being that the calibrated ones are only required if you want to use them for billing purposes.

In my case an uncalibrated one would do fine. These meters have an interface that generates pulses when power is consumed. The pulses are 90ms long and come in different resolutions, 1000 or 2000 pulses per consumed kWh are the most common. Basically they all look like these.

So I picked up three, because where I live it's common to have three wires carrying 230V (and one neutral wire) going into your home, and all three of these have to be measured.

With all the basic supplies ready it was time to move on to the interesting part of this build:

The controller:

It looks like smart metering is about to become a big thing in Europe, but people are worried about the power companies being able to track their power consumption in such high resolution, and they are absolutely right about it! Just imagine someone at the power company who uses the data to keep track of who is at home and who is not, and sending a team of burglars over to the places where the owners are out for their weekly movie night, or whatever. Or imagine getting a call from the power company with the person on the other end of the line going: "We noticed that you didn't take a shower for two days in a row now, you really should take more care about your personal hygiene..."

Not wanting to give up this data to the power company is one thing, but there are still many positive aspects about smart metering, like, for starters, I want to know what it actually costs to take a shower.

This is why some folks in Germany started the volkszaehler.org (which roughly translates to "peoples power meter") project. This project is a really great starting point for everything about DIY smart metering. They've got open source software to track the consumption (among other things like temperatures) and several controller designs to connect your meters to the internet (or LAN).


This is the volkszaehler.org frontend.


Actually I only used the power supply and the input driver boards of the project, because the controller board would be connected by Ethernet and I don't have Ethernet anywhere near my breaker box.

There is another controller design by the same guy that uses Wifi and is compatible to the one I was building, so I tried to use that. That might have even worked had I not tried to make the PCB myself (The guy who designed these controllers also sells kits, but he had no pcb for the wifi-controller):

I have access to a little CNC-Mill capable of making PCBs via isolation routing. This is good for single sided boards (even with SMT parts like TQFPs), but it really sucks to make a double sided board on it. Especially if the board is designed to be made in a real fabrication house where they can make chemical vias, apply soldermask, and so on. The fact that I hadn't really worked with PCB design software like eagle before, didn't help.

So I just imported the design into eagle and didn't care about re-positioning the vias, so that they aren't below any parts, and milled away... That left me with a semi-functional board where everything would work fine for a while, and then the Wifi-module would perform an unprovoked factory reset.
This was the first design. It was based on a RN171-Wifi-module and an ATTiny2313, and didn't work. Notice how crammed the area above and on the right of the RN171 is. That's not only capacitors and resistors, there's a good deal of vias in there, too.
The bottom side of the beast.
I spent literally months of debugging before I finally decided to can it and start from scratch.

This is what I came up with:

Not much to it. The crucial parts are an ATMega, the RFM12, and the headers to connect to the other board.
Actually this design is derived from the JeeNode, which is (kind of) an Arduino clone with a RFM12B 868MHz FSK transceiver.

Since I was eager to keep things simple this time, and also because this was my first attempt to design a pcb in eagle (or any pcb-layout-software, for that matter), I went with a single sided board. For this all SMT-parts have to be mirrored and placed on the back. But after that epic failure with the first board I also decided to use as few SMT-parts as possible. Luckily the RFM12B is the only one and that's also comfortably big.
The layout in eagle. Notice the RFM12B on the back of the board
 Before you can feed the layout to the router, you have to convert the eagle file to gcode, and while you're at it, you could just widen the traces a bit, so that you'd have a bit less milling. If you don't know what I'm talking about, the next few pictures are pretty self-explanatory.

This is what the layout looks like before being "blown up"

This is what the layout looks like after the traces have been "blown up". The colors are added by the software to ease identifying traces and nets.

The milled and drilled board. I only remembered to take a picture after I already had started soldering stuff to it. Note that it's no problem to route traces between IC-pins. Also note that some holes (e.g. the ones for the pin headers) are wider than others.

This was done with pcb2gcode which is available at http://sourceforge.net/apps/mediawiki/pcb2gcode/index.php?title=Main_Page.

Unfortunately, there aren't any more photos of the construction because I broke my phone's camera function.


The receiver:

- Tell me, Captain Obvious, what good is a single RFM12B radio transceiver if it has no one to talk to?
- Well, not too much, obviously!

Anything with a RFM12B module (and probably others) can act as a receiver for the power meter.
I went and built a receiver into my Asus WL500gP Wifi-router, but you don't have to:

There are (Arduino compatible) kits and assembled USB-dongles that contain a RFM12B and an ATMega328 available at http://jeelabs.com/, so you could just get a JeeLink or a JeeNode USB and use that.


My breakerbox with my smartmeter in it: Upper row: The controller, middle row on the right: The three meters.




Resources:

I've only just started learning C, so the code is probably shit.

Credits:

All PCB milling was done at the Attraktor hackerspace in Hamburg (website in German).

Sunday, August 5, 2012

Garage door remote for real

(featuring my first real factory made PCB)

It's been a while since I built the garage-door-proof-of-concept-remote. As stated in the according post, the "industrial design" part of the PoC managed to be flimsy and clunky at the same time. So there was some room for improvement.

Back when I still thought the original remote was using 434 MHz, I bought two little "cloning remotes" or at least so I thought. I was eager to save a few Euros, so I got mine from some shady Chinese eBay-shop, and sure enough the ones I bought were far from capable of cloning anything. Instead they just sent out fixed codes, no matter what I tried. Well, at least they did something.

The enclosure.


But even if I had gotten the "real" ones, it wouldn't have helped a bit, because those still would only work with 434 MHz and I needed 868 MHz.

So there I was with my homemade (but working) remote and my two Chinese knockoffs. Even though those were useless, I still liked the enclosure and started wondering if I could just use that and design my own PCB to put inside. I could even harvest a few parts of the original PCBs, like the buttons and the blue LED.

The disassembled remote and the original PCB. They use a 12V battery for the 434 MHz transmitter.

The first thing I did was to take measurements of the original PCB and transfer that to a simple Eagle layout. After quintuple checking the positions of every hole, notch, button and the LED I locked their positions so that I wouldn't accidentally move them out of place later in the process.

The back of the board on a scanner.
And the front. At first I tried to take measurements from the scan, but I soon realized I would have to get a caliper.


The problem with my layout was that it would not work with the kind of PCBs I could make on the CNC router at the Attraktor because the board would have to be double sided and I've tried to make a double sided board before and failed miserably. So I had to have the board for this project manufactured in a real fab house. The problem with that is that it's really expensive for low volumes. If I had the board made on my own in Germany it would have cost me about $50. So that wasn't an option.

By chance a friend told me about the dorkbot PCB-service (now called OSHPark). This is an extremely cool service run by a guy who started out collecting PCB layouts for his local hackerspace and put them all together on a big panel to reduce manufacturing cost. To fill up the panels quicker he opened his service to the public, so that is where I went.

My Idea of a garage door remote in Eagle. Except for the antenna on the right I was really lazy and let Eagle autoroute most of the traces, which wasn't too brilliant.
Having the boards made and shipping them across half the globe takes about 4-6 weeks, so after a long wait I finally had these in the mail:

I had round edges and a little notch on the "Dimension" layer. The notch is there but the round edges aren't. I'll have to remember to sort that out before I order the next batch.

And when I finally held them in my hands, it wasn't long until I discovered the first fundamental design fault: There are a bunch of vias (gold plated, even!) right where the negative terminal of the battery would go. These I would have to insulate somehow.

Later that night one of the boards looked like this:

The ATMega is facing the wrong way. Ignore that.

I soldered the TQFP part using the solder wick technique shown at about 6:25 in this video (and probably many others).

I left out the LED and the accompanying resistor on this board. Also note that I left out the notches on the sides of the board. I figured that it would be easier to just cut the enclosure to fit instead. Which worked.
I insulated the vias under the battery clip with this stuff and I also put a lot of solder on the sqaure pad in the middle to raise its level above that of the vias.

The ISP connector is done with the pads on the left. Now I had to find a connector to fit. A PCI connector should fit, or maybe an old floppy drive connector. Too bad I didn't have any of those around. All I could find was an old connector for a C64 datasette port, if anyone still remembers those.

My barely working ISP Plug.


The pitch was totally off but it was the only thing I could find, so I used it. It was a royal pain in the ass but at least sometimes it worked.

Not that night, though... The programmer wouldn't even find a controller to talk to. After checking all of the connections were good, I finally found it: The ATMega was facing the wrong way. Nice work.

De-soldering a TQFP part is no fun but it can be done. You have to add lots of solder, so that all pins are covered, and heat all four sides in a swift procession. The trick is to have the fourth side heated while the first is still hot. You'll have to go around the chip for a few times but eventually it'll come loose and you can just wipe it off the board. The major rule is: Patience > Violence.

After having this fixed I could flash the controller but the damn thing still wouldn't work. I had to get out the logic analyzer. Luckily the connection between the ATMega and the RFM12B is an SPI connection, which is basically the same as the ISP connection for the programmer, and it also uses the same pins (mostly). So I could connect all but one of the analyzer probes to the programming connector. With help of the analyzer I could see that the controller was talking to the RFM12B on the MOSI line but wasn't getting an answer on the MISO line. This meant that the RFM12B wasn't working properly for whatever reason, most probably a short circuit somewhere on the board.

After poking around with the multimeter for another while I found it. Or, more precisely, I found out which nets were shorted out, but I had no idea where.

By now I had a little practice in de-soldering larger SMD parts, but even after removing the ATMega and the RFM12B (the de-soldering trick described for the ATMega works with that, too), the short was still there. It was between the SEL signal of the RFM12B and VCC, i.e. the positive terminal of the battery. But since these two nets virtually never cross paths or even come close to one another it had to be somewhere else. And here it was:

The SEL trace of the RFM12B crossing the drawing of the battery clip on the "Dimension" layer. Looks perfectly innocent in Eagle.
This is what it looks like in real life:
Not so much in real life. The metal of the battery clip cut through the soldermask somehow and shorted out VCC with SEL.
After gently bending the battery clip everything worked like a charm. Well, at least the hardware did.

The software:
The PoC didn't have a real program logic at all. As soon as you connected the power, it would transmit away. For this project I wanted something slightly more elegant. In addition to just sending the garage opening signal, the buttons should send something I could use around the house, like a unique ID, that I could process in projects to come. Also I had to do something about the current consumption. Both, the ATMega328p and the RFM12B have ultra low power sleep modes and I needed to utilize those to make the coin cell last more than a few hours. Again I had lots of help from the internets, particularly here and here somewhere near the bottom. After some tweaking I had the standby power consumption down to under 1 µA.

At this point I'd like to praise the Internet. I couldn't do shit without it.
The only problem with the code was that it wasn't very stable, and I couldn't really fix that.

(The PCB design actually reflects how little trust I had in my own coding skills: I put in a little solder jumper to be able to use one of the keys as a reset button for the AVR. But that was planned as a last resort in case I really couldn't get the code to run properly.)

Luckily, for cases just like this, the ATMega (and probably other microcontrollers as well) has a watchdog timer, that will reset the whole controller if you don't reset the timer before it runs out. After implementing that, everything is running smoothly for several days now.

Conclusion:
I've learned quite a few things about PCB design doing this project:

  • Don't trust the autorouter of your PCB design software. Better yet, on a project as simple as this, don't even use it.
  • Think 3D! It's hard to imagine a real life PCB just from an Eagle layout. There are 3D rendering plugins for Eagle and other software with this functionality built in, but I found it the easiest to just print out the PCB layout on a piece of paper and place the parts on it.
  • Only use parts in your design that are actually available to you. I have  yet to find a correctly pitched edge connector for the ISP connection.
Total cost of the project:

  • PCB: 3.30 €
  • ATMega328p: 1.80 €
  • RFM12B: 4.00 €
  • Chinese fake-cloning-remote: 2.50 €

Makes a total of 11.60 € or US $ 14.35 at the time of writing. All prices include tax and shipping and all items were purchased in low volumes.

Files:
All design files as well as the source code for the controller are available at my github repository:



Update:
As it turned out, the watchdog timer (or rather the way I implemented it) didn't really help, actually it made matters worse! After a long and tedious debugging process I finally think I've got it right. Read about it here.