David Lowe and Jason McDermott

This update is in three parts: ‘our approach’; ‘the hardware’; and ’sensing’. This entry is continued from ‘our approach’.

CRIN have explored some simple hardware platforms, particularly the Gumstix and Beagleboard ’sawn-off’ PCs (strictly, ‘computers-on-module’). Below, a few shots from a recent session discussing the hardware. Be warned, this will get a little geeky.

Note the core board here is a Gumstix Overo Earth – the smaller component – sitting on top of a larger expansion board, which gives us further connectivity, including an outlet for multiple USB connections amongst others (spec here). This is needed as CRIN are taking the ‘array of dongles’ approach to sensing via four Bluetooth dongles all scanning simultaneously. (You can see a couple of those behind the box in the shot below.)

Note also a 3G dongle is needed to deliver data from the board to the internet (and to enable updates to the software to be received on the board too).

There is also a wi-fi dongle, ready for the next stage of scanning (notes on that to follow).

The power adaptor, or battery pack, as seen here, is by far the biggest component. If the boxes are installed where reliable, secure power is available, then the box can be rationalised heavily, down to not much larger than the augmented Gumstix itself. In many cases – most urban bus-stops, public transport vehicles themselves, offices etc. – this power will be available. In far-flung suburban bus-stops, it may not be. So the battery approach isn’t ideal, but may be necessary in some environments. (Also, CRIN (David Lowe, Alex Gibson) reckon the program code is running at about 7% CPU at the moment. By rationalising other aspects of the software, the power load can come down even more. However, Alex is exploring running a background task to keep the 3G connection open (those service plans are not particularly designed for this, of course.)

CRIN can also wirelessly update the software on the boards, when required, by SSH-ing onto the card via their IP address (using Dynamic DNS, which avoids the issue of IP addresses changing regularly due to DHCP, usually used when connecting through an ISP like Dodo. Dodo have been chosen as the 3G connectivity as they’re one of the few to offer a long term pre-paid plans. Again, it’s interesting how these ancillary services can inadvertently shape the research. Though bandwidth not an issue here, due to the tiny packets of data being transferred.)

The data is being updated to Pachube. The problem here is that Pachube only currently supports 15 minute updates, and we also require a broader longitudinal history than it enables. Both of these aspects are being addressed by Pachube though, so we see great value in continuing to patronise that platform with our data (and here it is, for what it’s worth at this pre-installation point). The memory card on the board – a basic micro-SD card – has more than enough room for years worth of data stored locally, but getting the data into ‘the cloud’ via Pachube makes it far more malleable.

CRIN are going to test how scalable this approach is – say, by putting seven Bluetooth dongles on a Gumstix board, which are then scanning at slightly different start times. Results across all seven could be aggregated, and with duplicates removed, give a significantly better return in terms of phone detection.

The total cost of the box (all internals, connectivity etc.) is looking like it’s around $400 at this point. Note, at this stage, the boxes are being chosen for their robustness and unobtrusiveness. At a later date we will consider the visual design and affordances of such boxes, as the need to convey the existance of this system might figure for ethical and adaptive design reasons. Besides, why not make a nice box?

We’ll consider next steps, and the other sensing approaches in the ’stack’ in the next post.