Andy Isaacson wrote:
> [...]
> This is all very interesting, but a radio friend of mine made a good
> point when I was discussing the Moos Event with him.
> 
> In general, the AP is going to have a better transmit side than the NIC;
> it has more power to work with, a better antenna system, etc.
> Conversely, the NIC probably has lower transmit power and a lesser
> antenna.
> 
> Furthermore, an AP on Moos is going to be barraged with noise, since it
> has LOS to just about every radio source in Hennepin and Ramsey
> counties, while the NIC can hear the AP and any other radiators in its
> immediate vicinity, but not much else.

All this is true, very true.

The directional antenna at the remote sites helps keep the
spurious rx somewhat under control at that end of the link.

There are going to be two major sources of in-band rx interference
at Moos:

1 - general RF (including 802.11b transmissions in adjacent
    channels.) This is going to have the effect of increasing
    the baseline noise (aka noise floor) that the receiver(s)
    have to deal with. Note that packet error rates correlate
    with Signal to Noise ratios, meaning that we need more signal
    to acheive the same error rate in the presence of increase noise.

2 - 802.11b transmissions on the same channel. This is going to require
    a remote site to "capture" the receiver, meaning that a particular
    signal exceed competing signals by some measured amount (usually
    measured in dB.) In addition to classic hidden-node (which is somewhat
    solveable by configuring RTS/CTS) we may see truncated packets because
    some other node captures the receiver during receipt of a packet. The
    good news is that DSSS modulations generally boast pretty good capture
    ratios.

Obviously people are successfully doing similar things in the metro area
(e.g. Implex - http://www.implex.net/services/access/wireless-area.cfm,
[I'd take that map with a grain of salt] though they are using sectorised
antennas, and karlnet firmware rather than vanilla 802.11b.) Anyone from
Implex on the list ? Care to share your perspective ? Correct my description
of your solution ?

> These points suggest that we should do more than just running Kismet to
> test connectivity between Moos and a mobile station.  Kismet is a
> fabulous tool, but it explicitly does NOT test the host-to-AP side of
> the connection (please correct me if I'm wrong in this claim).  We
> should also find out how well mobile nodes can associate with the AP,
> and do some round-trip tests, to ensure that the radio round-trip is
> usable.

Ack.
Agreed 100%.

Thanks for doing the somewhat thankless task of collating the
info and making the maps, Matt. I have, however, noticed a slight
inconsistancy in the posted maps. The SE corner of the moos6 coverage
seems to move from Hwy5 & 35W (moos-survey1.png) to Hwy5 and Randolph
(moos-survey4.png.) I'm interested in what kind of connectivity was
achieved at this location (whichever it is.) The effect of that
corner of the coverage envelope adds several square miles to the
actual coverage area (moos-survey3.png.)

> Does anybody out there have any experience doing this kind of
> connectivity testing?  If so, are you willing to share your experience
> with the group?

I've never used kismet for a site survey. I've always associated, moved data
and measured throughput, but I've never tried general coverage mapping of
a moos-tower like substance using 802.11b. I suspect a script that continually
looped associating with a central site, capturing stats with iwconfig or /proc,
dhcp request, move some data, dhcp release, all the time snarfing gpsd data
would be useful.
-- 
andyw at pobox.com

Andy Warner		Voice: (612) 801-8549	Fax: (208) 575-5634