locating via GSM, revisited

[joerg at openmoko.org]

Am Mo  21. April 2008 schrieb Werner Almesberger:
> joerg at openmoko.org wrote:
> > From this data, we see it's quite possible to determine location to a 
> > precision of around 100 x 100m or even better.
> > Of course this depends on the density of BTS again.
> 
> Wait ... if I understand correctly, your experiments (great, by the
> way ! Finally real facts ;-) indicate that you can measure some
> differences if moving ~50m, correct ?

Yep


> 
> Now, how does this topology of relative signal strength translate to
> anything we can use ? 

Just gives fingerprints. You have to use a database to translate to geographic 
coordinates. Much the same way skyhook does for WLAN, only GSM-BTS are much 
more wellknown and steady. And our GSM-modem takes *much* less power to give 
us this data than WiFi does for AP-scan. :-)


> Can we translate it to a geographical location ? 
> If yes, how much error does the translation from signal path
> charracteristics to 3D location cause ? 

I didn't plan to do 3D location :-D.
*With* the cell reselection plus TA hack, you got 7 points of known (from 
database) position - let's call them BTS, and you know your distance to each 
of them (to a resolution of 550m). So it depends on luck whether you have 
500m error or (more probably) 20m.

> Also, how easily can we get 
> the real location of the base stations ?

For many countries there are ageold databases created by hobbyists doing 
antenna-spotting. In Germany, carrier O2 sends quite exact Gauss-Krueger 
coordinates on CBC 221 for each of his stations.


> 
> An alternative would be to use only a topology of signal strength,
> and to derive hints from it. E.g., one may not be able to say that
> location X is 117m at 32 deg from north, but one may be able to give
> an estimate that it may be at 10 times the distance from the last data
> point, if turning 45 deg counter-clockwise. (The latter part is
> tricky, because one would have to project an angle between vectors in
> an n-dimensional space to a losely correlated 2D space.)

Too sophisticated for me to understand. 
The first bet is "I'm registered at BTS 1234, so probably it's the closest BTS 
to my location". Sometimes fails, locating error (area of location) up to 
30km radius. 
Second: "I got TA of 3, so I know I'm between 1600m and 2150m from "my" BTS". 
Now the huge circle became a rather small ring.
Third: "I also can 'see' 6 BTS that are located North of 'my' BTS". Most 
likely we are *not* on the south half of the ring.
4.: " if we can get the TA distance to 2 other neighbour cells by doing this 
cell reselection trick, we should have something like a 200m x 200m area" by 
using very simple geometrics.

> 
> On the way to this suggested destination, updated and hopefully more
> accurate directions could be generated. Of course, there will be some
> area where the directions are only noise.
> 
> Anyway, sounds like an interesting research project. "Terrestrial
> navigation using GSM-based dynamic hypergeometries." :-)

SOUNDS interesting, for sure ;-)
In reality it's a little of WiFi positioning, a little GPS trigonometry with 
BTS instead of satelites, and a little of good o' "taking a bearing" like in 
(airo)nautical navigation.

/jOERG