locating via GSM, revisited

Andy Green andy at openmoko.com
Sun Apr 20 22:58:17 CEST 2008

Hash: SHA1

Somebody in the thread at some point said:

|> What I think we need to do with this stuff is feed it into one daemon,
|> from GPS, WLAN AP and GSM masts alike.  These WLAN and GSM
|> "fingerprints" have something deep in common, they are a list of
|> transmitters with estimated distance (radius) and estimated error.  I
|> think if we collect this datatype we can leverage it in very cool ways,
|> with or without the advantage of GPS.
| Yep! For GTA04 I like to see parts of this daemon running on the MPU,
at least
| to the point where we can detect changes caused by movement so we know we
| have to wake the MCU. I guess this should be quite feasible for GSM
and GPS,

I dunno this is wise, it makes me think about U-Boot.  Kernel and
usermode daemon would be "good at" this kind of function, we should
leave it to the thing that is good at it.  Since we only do it
periodically it doesn't need to make a big power impact.

The only thing that makes sense is motion sensor management to be under
the MPU control I think.  It can do a nice job using threshold and then
dead reckoning (integration) then back to threshold when the motion is
stopped for a while to save power.  And we only need to check for new
"place" radio fingerprint when there has been movement seen by motion
sensors... otherwise he is definitely in the same "place" and we needn't

|> Although pure signal strength or reflections can make variations in what
|> we think about one transmitter, it is unlikely to affect the
|> constellation of transmitters randomly spread around in the same way,
|> this stuff could be more robust than one might expect.
| That's exactly what I think. And the best part of the story is: from
GSM we
| get a continuous position fix almost for free (unlike GPS or even
WiFi), as
| long as GSM is registered anyway :-)))

Since it is too much for MPU, it can be a reason to wake periodically
for the CPU, so we can still have this.  It also definitely is in a
class of wake cases where we don't want to bring up LCM like Werner
proposed might exist.

- -Andy
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org


More information about the openmoko-kernel mailing list