Advice on long distance wireless links?

[Jim Carter]

On Sun, 8 Sep 2002, Bruno Lopes F. Cabral wrote:
> I did a long distance point to point link here, 45km,
> 30dBs 1.5meters antennaes on both sides (105m high on
> one side, 15m high on the other, no higher obstacles
> between the points)
>
> the Link Quality shows 19/92, Signal level: -87 dBm
> Noise level:-99 dBm. auto bitrate states 1Mb, but
> fixing it at 11Mb also seems to "work" (i.e. pinging
> the other side works -- most of the time)
>
> do I need to increase/decrease one or more of the ...
> - Fragmentation threeshold
> - Listen interval (or beacon interval)
> - RTS threeshold
>
> would be better to use ad-hoc mode instead of managed

45 Km link with big antennas.  Reported SNR = 19 dB, level -87 dB, noise
-99 dB (calculated SNR: 12 dB).  Link quality ratings are generally given
as:
    Below 10 dB:	Bad
    10-20		Marginal
    20-30		Good
    Above 30		Excellent

A marginal link should be useable with occasional trashed packets. These
should not be noticed on TCP transmissions, but missing UDP packets are
either lost forever, or retransmitted after a user-visible timeout handled
by the application. NFS is the most notorious example of the latter.

Don't tempt fate -- use managed mode.  There seems to be fewer bugs in
firmware for managed mode, compared to Ad-Hoc mode.

Roundtrip delay (90 Km) = 3e-4 sec (300 usec). All delays in the wireless
card must be greater than this. Unfortunately my book is at work so I can't
quote exactly which delays you need to fiddle with. The wlan_ng driver
(Prism-II / Intersil firmware only, not Orinoco) can set just about any
card parameter, but I only tried it once and I can't give you any
recommendations about how stable the code is.

Frag threshold: On extremely poor links with lots of lost packets, set a
small frag threshold and only the trashed fragment has to be retransmitted.
But there's overhead for each packet, so on a "normal" net you leave the
frag threshold at the maximum value, which is the default. Probably won't
help you; your link isn't that bad.

RTS threshold: On a heavily loaded net with lots of collisions, you can use
a RTS/CTS extension, which avoids collisions at the cost of an extra
administrative packet for each data packet. So you only want to do this for
big packets. Many cards are not capable of doing RTS/CTS anyway. This is
irrelevant for your situation, with only two stations (one at each end).

Beacon interval: If your net has lots of PDA's with tiny batteries, for
which it's important to keep the radio powered off except when actually in
use, you can lengthen the beacon interval, so they don't have to power up
their radio so often. But then, the access points have to save queued
packets destined for the PDA's for a longer time, which could fill up their
memory and cause the queued packets to be tossed. Irrelevant for you.

Recommended book:
Matthew Gast, "802.11 Wireless Networks: The Definitive Guide",
O'Reilly (http://www.oreilly.com/) (USA), 2002, ISBN 0-596-00183-5,
about US$ 45.

James F. Carter          Voice 310 825 2897    FAX 310 206 6673
UCLA-Mathnet;  6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA  90095-1555
Email: jimc at math.ucla.edu    http://www.math.ucla.edu/~jimc (q.v. for PGP key)