[Server-devel] [OLPC Networking] RSSI value questions
wad at laptop.org
Fri Apr 4 22:18:06 EDT 2008
A guiding design principle for any XO activity is that it be designed
to work without a school server. Learning doesn't stop at the school
The only thing special about an XS (or any access point) is that we can
know (absolutely) where it is. Whatever system is designed should
allow arbitrary peers to declare that they know where they are (and
should handle the fact that some of them either lie or have a very hazy
idea of where they are...) Perhaps an XS Active Antenna or Access Point
is simply an example of a certifiably trusted position beacon.
I still prefer the idea of using audio, a la Acoustic Measure by Ben or
a three-D, multiple device version:
(longer version at http://web.media.mit.edu/~vmb/papers/DaltonMS.pdf)
Research into less intrusive methods (using ambient noise, or sounds
generated by the systems while doing other tasks, as the
basis for obtaining the location information) is needed.
On Apr 4, 2008, at 4:39 AM, Oliver Mattos wrote:
> why exactly is an XS needed at all - what about just a mesh of
> laptops with no XS. I agree then there are NO refrence points at
> all, so orientation and world-position of the generated map can't
> be determined, but the rest of the info still remains useful. The
> XS is simply another node - there is no reason it should be required.
> In terms of an algorithm for calculating positions from a series of
> metrics with no known points, the best I can think of is successive
> approximation. Basicly, place all the nodes randomly on a map,
> attach "virtual springs" between nodes that have connectivity,
> where the springs ideal length is determined by the signal strength/
> other metric, and springgyness is determined by the metrics margin
> of error, and then do a physics simulation of where they all end up
> when released. Using that algorithm, multiple types of metric can
> be used to generate the same map.
> After generating the map once, future generations would require
> many fewer iterations of the physics simulation, therefore less
> processing time even for big meshes, so it would probably be
> possible to update the map in real time as new results come in.
> There are quite a few optimisations for the above, for example
> "replusion" springs with a negative force could be used for nodes
> that are currently close together on the map but have no
> connectivity. - that would provide much more accurate mapping in
> sparse meshes where some laptops have 2 or fewer neighbors.
> On Fri, Apr 4, 2008 at 7:06 AM, Ryan Crawford Comeaux
> <crawford.comeaux at gmail.com> wrote:
> Just to address a few other issues/questions raised...
> If there is only one antenna on a server, then as long as 3 other
> nodes are considered relatively stationary, I think their 2D
> locations can be deduced from each node's measurements of the other
> 4. An easy to use interface can allow the user to orient the
> generated map with respect to whatever reference point they like;
> ideally, the final program would allow for a floor plan of the
> building to be displayed underneath the topological mapping.
> With respect to granularity of different measurements, I think
> inaccurate measurements can be averaged over time, since some would
> necessarily be more accurate than others, allowing for a more
> accurate map as time passes.
> Ben stated that the dynamic gain isn't available in user space.
> I'm just wondering if there's a way to passively determine the gain
> and if this would even be helpful in determining location. Any
> ideas? I'm not so experienced in RF tech that I can come up with
> how knowing the gain would be useful, but if it is useful, then I
> think it'd be easy enough to figure out some sort of indicator
> that's relative to the fluctuations in whatever measurements the
> gain affects. Again, let me know if I'm that kid out in left field
> wearing his glove on his head and facing away from the bases...
> I feel pretty optimistic about the feasibility of this kind of
> project. There seem to be a few good measurement techniques to go
> by, as well different methods to compute the data. If the XOs
> pitch in and tell the server where they think other nodes and
> themselves are, relative to each other, that would provide another
> set of input to include when averaging out measurements.
> For those of you that would like some light reading on the topic of
> modeling this information and computing it, here are a couple of
> papers that attempt to do similar things with GSM signals and
> neural networks:
> - Crawford
> Networking mailing list
> Networking at lists.laptop.org
> Server-devel mailing list
> Server-devel at lists.laptop.org
More information about the Devel