John McDermon wrote:
Jim,
Our *real world* experience definitely supports your math (as if you
needed more confirmation).
The thing about this stuff is it sort of works when you start and even
though you're probably already suffering 50% or greater throughput
degredation due to interference and excessively long links, it doesn't
show up because your only sharing a DSL connection or a single T1.
So no, here at LCWA, we've got ourselves in a bit of a pickle -- most
of our APs have clients in at nearly all points of the compass. If we
want to minimize the high-gain omni issue you describe would we go to
point-to-point directional links to connect AP to AP (the back bone)
and then lower gain omnis for the clients? This will cause of problems
with number of available channels in 802.11b since some of our APs
have multiple APs binding to their omni.
Are there reasons why you can't make the backbone run in 5.8GHz? That
would probably be the best solution. If you have to stick in 2.4GHz, I
certainly hope that you have very large 'channel filters' between each
radio and the common antenna.
What you'll find is that even though the "3 non-overlapping channels" of
802.11b (and g) don't "overlap" in terms of the transmit mask, the
'selectivity' (adjacent/alternate channel rejection) of your typical
WiFi receiver is actually quite poor. The best I've seen is the
super-het designs (such as the Prism2, Orinoco and CIsco cards) which
manage about 41dB of adjacent channel rejection. The IEEE minimum is
35dB for 802.11b, and all of the "direct converstion" receivers I've
tested just manage to make it.
The ACR figure gets worse with increasing bitrate. The IEEE minimum ACR
at 54Mbps, for example, is "-1 dB".
What this means is that if you put two radios directly coupled to each
other (in your example, on the same antenna) and operate them on
adjacent channels (1 on 6, 6 on 1, 11 on 6, 6 on 11) then when radio 'a'
transmits (say, on ch1) then radio b (say on ch6) will see a degradation
to its noise floor.
If both radios are 20dBm (100mW) it looks like this:
20 dBm - loss of both cables - ACR figure.
I'll assume that you've spent the money and purchased good quality
cables, so you've got no more than 2dBm of insertion loss due to each
cable. This means that the 20dBm signal from radio 'a' hits the front
end of radio 'b' at 16dBm, where, if radio 'b' is on an adjacent channel
it is suppressed by another 41dB, presenting itself as a very strong,
but mostly non-coherent signal in 'b's receiver at -25dBm.
Meanwhile, you're probably trying to 'see' signals at 'b' from far-off
clients at -60dBm or even -80dBm.
The biggest issue occurs when 'b' (or 'a') is receiving from a far-off
client, and 'a' (or 'b') has *NO IDEA* that this is happening (because
the signal level, once suppressed by another 41dB is far too low to
trigger CCA) and therefore transmits *ANY* packet (even an ACK or a
beacon). This will destroy the incoming frame in the other radio.
None of the radio manufacturers published a spec for alternate (11 on 1,
1 on 11) channel rejection for 11b, but experimental data says you can
count on approximately 20-25dB of additional "suppression".
Even here, with a very good (read: expensive) channel filter (perhaps
another 30dB of rejection of off-channel signals) you've got a total of
96dB of rejection, while the noise floor is typically held to be at
-105dB. This is getting into the "hard to see in the field" arena, but
you've still got a 10dB degradation.
If you google around, you'll see me tell various people that the
quickest way to restore performance to their (gen1) Vivato switch is to
run all the "pointing directions" (radios) on the same channel. The (1st
gen) Vivato switch had circulators (good for about 20dB) and Orinoco
radios.
Multi-channel (in the same band) is very very hard. Its a 100dB problem.
If you're not backing away quickly when I make that statement, you might
want to remember that 100dB is "1:10,000,000,000".
The other thing you might try on the AP-AP links is running something
like 'frottle', (assuming your AP is build on linux or freebsd and that
you can change the code in it.) such that only one AP in the cell is
uplinking to the base at any one time. (Turning on RTS/CTS with a very
low threshold will achieve similar effect, with the potential benefit
that RTS works cross-channel (if it can be decoded).
Running frottle with all the uplinks in 5.8GHz would likely improve your
situation a lot.
The attached .png graphic is a spatially correct depiction of our
network rendered using open source SW and GPS coordinates.
One of our pairs of T1s is located at the Condesa AP, the other T1s
are located on the West side of the map at the Main AP.
Ideas would be appreciated.
Thx,
--John
http://www.lcwireless.org
------------------------------------------------------------------------
On Jan 20, 2006, at 3:47, Jim Thompson wrote:
sigh.
the last thing you want to do is plant a high gain omni on high ground.
here's the thing with antennas, they have the same gain in both
directions.
here's the thing with wifi, there are more of them then there are of
you.
Thus, if you employ a high gain omni with a clear LOS to a lot of
area, that area will have other WiFi devices in it.
If you can reach out far enough to communicate with one, it (or one
at the same or shorter range) will be able to, at the very least,
wreck havoc.
APs are like insects, where you find one, there are more, lurking,
waiting to cause interference with packets you're trying to receive.
They can't 'hear' each other (because the path loss at ground level
is too high), but they can all hear *you*, and, as a consequence, you
can hear *all* of them.
This is what killed Vivato.
I've tried explaining it to Matt and co up in Seattle (just like I
told them early on that their efforts with OSPF as a wireless routing
protocol were doomed, and that mesh was going to be the only way
forward.) In both cases, my mistake has been attempting to explain
'why' with mathematics to a crowd who are suspicious of mathematics
(potentially because they don't understand them) and always wants to
"see it in real life".
Matt eventually asked me to leave the Seattle Wireless lists because
he didn't want me to "spout off" about why he was wrong, when he was
wrong. I've stayed on the other community wireless lists (Austin, PTP
(Portland), here, and NYCwireless).
Community wireless (or rather Community Networking) has a chance (*),
but its not going to happen with high EIRP setups and people who
aren't willing to deal with the actual physics of RF.
Jim
(*) See, for example:
http://www.smallworks.com/archives/00000365.htm
http://www.smallworks.com/archives/00000372.htm
Begin forwarded message:
From: "Frank Keeney" <[EMAIL PROTECTED]>
Date: January 19, 2006 11:46:46 PM HST
To: <[email protected]>
Subject: [SOCALWUG] Wi-Fi Underground Revolutionaries
Reply-To: [email protected]
A very good summary of the grass roots community wireless Internet
movement:
"The idea of building citywide wireless networks from the community
level was suspiciously simple back in 2000, though the plans sounded
like the work of underground revolutionaries.
"All of us were very idealistic, and all quite strongly
opinionated," said Adam Shand, founder of The Personal Telco
Project, which had visions of such a network in Portland, Ore.
Seattle Wireless probably represents one of the highest achievements
in the area of neutrality, because they’ve built a network of what
looks like now over a dozen nodes that use an open- source mesh
routing protocol to create a neutral medium....."
Read the New York Times Article also at Cnet
Glenn Fleishman's Blog
Frank Keeney
Antennas, Cables and Equipment:
http://www.wlanparts.com
Blog: http://www.unwiredadventures.com
