A while ago Gašper gave me this Energycount 3000 kit from Voltcraft for logging household electrical energy usage, with a wish that he would like to access its measurements from a computer. All this time it's been mostly gathering dust on my desk, but last week I've found some time to give it a closer look and made a few discoveries that are worth reporting.
The box contains two sensors that can be connected between a wall socket and a plug and a tiny battery-powered remote control for reading out the data. The short instruction leaflet explains that the sensors broadcast their measurements through radio every 5 seconds. With a push of the SCAN button you can set the remote control into a 6 second listening mode which catches the transmission and displays it on the small LCD screen.
The remote also has some calculation functions, like predicting the next electricity bill, but is otherwise nothing more than a remote display for the sensors, which apparently do all the logging. This makes sense: the remote control is limited by its batteries and as much functionality as possible is pushed to the wall plug where there is abundant power. The radio link is also obviously unidirectional. The sensors transmit their periodic reports and the remote receives them. All user interaction with the sensor is done through a push button on the sensor itself.
To get the data from the sensor it appears that all I would have to do is to eavesdrop on the transmission. Unfortunately, the box says this device operates on the 868 MHz ISM band, meaning that my 433 MHz receiver was useless. I could get a 868 MHz receiver module for it, but I suspected that these devices use something more complicated than on-off keying, so I looked for other possibilities.
Here is how sensors and the remote look from inside:
As you can see each device has two integrated circuits under a blob of epoxy. I'm guessing one is a microcontroller and the other obviously some kind of integrated ISM band transmitter or receiver. In both the sensor and the remote control they are connected with 5 copper traces. Having recently encountered and worked with chips like the CC1101, my first guess was something like that. I figured the five traces would carry a digital bus like SPI or I2C, so I soldered some wires to the tiny traces on the remote (destroying one of the termination capacitors in the process) and hooked it up to a logic analyzer.
Unfortunately, what I saw there didn't look at all like a synchronous transmission. Out of 5 lines, 2 seem not to carry any useful signals (all I saw there were some transients that looked like glitches). The remaining 3 might act as a digital bus immediately after the microcontroller wakes up the radio chip, but otherwise look like some pulse-width modulated signals for the majority of the 6 seconds when the radio is turned on.
This was a sort of a dead-end until I came across chips like the TH81112. These are ISM band receivers that can be used to receive ASK or FSK transmissions, but are much simpler that system-on-chip products like CC1101. They merely contain a tuner, intermediate frequency and a phase detector and therefore rely on the microcontroller or some other logic to do full FSK demodulation. In hindsight, it makes sense for Energycount to use something like this. It's a relatively cheap product and an expensive general-purpose transceiver like the CC1101 would probably be far too expensive for it.
Update: Gašper found a thread that has some interesting insights, particularly regarding the communication between the radio chip and the CPU. It's possible I was looking at the signals at a completely wrong time scale.
But at this point I didn't bother to mess further with the original receiver. I started up GNU Radio and Fun Cube Dongle Pro and tried to catch the transmissions with that. It turns out fishing out the correct channel in the 868 MHz ISM band is a challenge in itself if you're only limited to the 90 kHz bandwidth of the Fun Cube Dongle. But luckily having the transmitter and receiver close by meant that I was able to see the transmission burn through even when the receiver wasn't tuned exactly to the correct frequency. So after some bisection I found the transmission at 868.388 MHz.
The double peaks in the spectrogram gave me more confidence that this is indeed frequency-keying with 20 kHz deviation and with GNU Radio's FM demodulator block the bits in the packets became clearly visible.
Actually, it's impressive how easy software defined radio makes tasks like this. Throwing blocks around in the GNU Radio Companion in a few minutes is something that would otherwise take you weeks with a soldering iron. Next time I'm doing something similar I'll most likely skip the whole logic analyzer part and just skip right to sniffing the radio waves.
To conclude, this demodulated signal is now something that can be piped to the capture process from my AM433 project and it will hopefully produce binary data. But of course, getting useful information from a binary blob is a whole new matter and that will come in a follow-up post.