In poking at some OM1Ps here I've found:
- Turning off diversity improves node-to-node ICMP response rate.
- Setting rxantenna to 2 improves rssi of associated client stations. (default is 0)
- Twiddling txantenna seems to have no effect. [?] (default is 1)
- Making the changes stick through a node's reboot cycle is needlessly complex.
On the last point, /proc/sys/dev/wifi0/(attribute) isn't supposed to stick,
/etc/sysctl.conf doesn't stick,
uci set (blah-di-blah) followed by a commit doesn't stick,
editing /etc/config/(blah-di-blah) is the same as running uci and doesn't stick,
edits to /lib/robin/postup.sh don't stick.
The magic spot, at least on r2693, is
Antonio documented that here:
|Only registered users can see links on this forum!|
Register or Login on forum!
(that's /viewtopic.php?t=3159&mforum=robin for those reading an archived copy without links)
From the madwifi documentation the choices for rxantenna and txantenna are 0, 1 and 2 where 0 means use diversity, 1 and 2 mean use just this one of the physical ports. That makes the flag flipped by the 'diversity' file in /proc/sys/dev look redundant, but who knows.
On the txantenna I was watching the node's rssi from another's shell while flipping the bit back and forth on the first. I tested it between two nodes about 15 feet apart at about the same elevation and at roughly the same orientation. I wound up telling it to use 2 for both of them just to give it fewer I/O pins to twiddle during normal use. Flipping txantenna on its own seemed to have no effect.
The improvement for choosing rxantenna=2 over rxantenna=1 held across every OM1P I tried. I don't think it's the flakey pigtail issue someone brought up earlier. The difference was 10 to 20 rssi units. Each was tested with just the soldered chip antenna and the stock 2dBi stick.
The feedback from athstats with receive diversity enabled suggested the opposite conclusion. Testing a node with a 19dBi dish on the antenna port and diversity enabled I found most of the traffic showing up as having been received on antenna 1 and that associating a client laptop physically close to the router led to occasional use of antenna 2 and a significant drop in performance to a distant node. Maybe there's a bug in madwifi that swaps the ports with the diversity bit. Again, who knows. Turning off diversity made it work better.
For diversity, rxantenna and txantenna on the OM1P my selections are 0, 2 and 2.
# documented as 0,1,2 for auto and two connectors