[Server-devel] Technical questions

James Cameron quozl at laptop.org
Wed Jan 20 16:47:42 EST 2010


On Wed, Jan 20, 2010 at 05:43:31AM -0500, Reuben K. Caron wrote:
> And by non-overlapping ranges he means you can ONLY USE channels 1, 6,
> and 11, spaced out so that two APs with channel 1 do not overlap.

For collective interest ... the air is a commons, divided by
frequencies.

The radios in the access points and the laptops switch rapidly between
transmitting and receiving, and they do this using a timing beacon
provided by the access point.  They share in time.

Access points on separate channels (frequencies) do not hear each other
directly, and so they are unable to collaborate well on sharing the
commons.  Often one receiver may be trying to receive from a distant
transmitter while a nearby transmitter on a separate channel transmits.

A radio receiver is less sensitive if there is a nearby transmitter.
This is called desensitisation, and it manifests as higher noise or
lower signal level in the receiver.  If the noise is high enough, or the
signal low enough, then the packet is not decoded and a retry must
happen.  Retries pollute time commons.

The amount of desensitisation depends on the difference in frequency;
and falls off more sharply the greater the difference.

For typical access point hardware, the desensitisation caused by five
channel spaces is sufficiently low to provide reasonable range; hence 1
vs 6 vs 11.

Slightly more range can be had by reducing the number of access points
and improving the antenna systems.

To cover a larger open area you may gain greater overall performance by
restricting the number of access points to 3, using a larger
omnidirectional antenna and increasing their height above ground.

To cover a larger closed area you may gain performance by using
directional panel antennas that face all in the same direction ...
creating a series of cells.

Above all you should test performance before and after making a change;
with all radios fully active and operating.  Theory is helpful, but
reflections from metal surfaces in a built environment can mock theory
badly.

-- 
James Cameron
http://quozl.linux.org.au/


More information about the Server-devel mailing list