I have detected that the main antena in the WRT54GS V1.1 (wl0_antdiv=0) Is the one close to the reset button. Not the one close to the power jack, as the documentation sais.

Can some body confirm than?

Thank you,

yip, I can confirm it, I posted last week with the same concerns, see http://www.openwrt.org/forum/viewtopic.php?id=1626

This is on my v2.2 hardware, running experimental, but it shouldn't really make a difference which firmware is being run.

on the same idea, the other thing I found quite weird, is that although 0 is the one next to the reset button, and is the primary antenna (If you check the tracks on the board), I get the best signa out of the one next to the power connector (1)

bUT my testing isn't really scientific, it was done with the default card on a centrino system, running netstumbler, and watching which antenna gave highest results on the graph.

I'm glad someone else also noticed this, its been gnawing at me for a while!

I have to agree with Dorris. The ant furthest away from the reset button has in all my tests also delivered the best signal. I think the little cable that runs to the board from the furthermost ant has the right attenuation, whilst the connector on the tracks might not be exactly have the right run of copper.

The difference in antenna port performance is bothersome.  I think we can see the antenna switch right there outside the radio so the difference must be from there out.  I think roDent meant so suggest that the copper traces might not have the proper impedence (attenuation would be bad).  So we might be getting a higher VSWR on the closer antenna port.  Another (or additional) possibility is that the exposed copper traces are picking up system noise whereas the shielded cable is helping keep that out. 

> Check the reported noise number with "wl" with each antenna setting
> Check the reported signal numbers as well
> Check the bulkhead connectors - right angle connectors exhibit more loss than straight thru

Higher noise may mean the exposed copper trace is picking up CPU clock noise.  At these frequencies a 2.5 inch long trace could look like an antenna.  Even an 1/8 wave 1.2 inch trace could pick up a noticeable amount of noise.  One experiment would be to replace the closer-port's bulkhead connector with straight-thru to cable, cut the PCB trace and then attach the cable directly to the RF switch. 

Clearly these are not the best laid out PCBs; but then again, a whole lot of work might only yield just a little bit more SNR improvement.  It might be more productive to work on shedding more heat from the PCB - it seems a lot of the heat from the network switch, CPU, and sdram, is all flowing into the PCB and this isn't helping the analog stuff in the radio.  So as one last (and simple experiment) try chilling your WRT with power off and then making the signal and noise power measurements right after power on.  If you get substantially better SNR, then we need to start a cooling project.  Be careful cooling the WRT in a freezer - condensation isn't all that great for electronics.  (note: an A/B AP project at a well known company used the pre-cooling trick during product development to get thru stupid program"gates")

Final note - other high-frequency equipment nearby can have a huge affect on system noise and reflections (both due to electrical noise and metal construction).  One lab had to take all the video monitors OFF the table tops and put then on the floor.  The RF shields in the monitors were causing too many crazy multi-path reflections that the radio had to struggle to get signal thru.   

VSWR info: http://www.antennex.com/preview/vswr.htm