Re: ath9k: Problems with diversity on AR9330

[Sven Eckelmann <sven@narfation.org>]

[-- Attachment #1: Type: text/plain, Size: 2188 bytes --]

On Friday 06 July 2012 19:32:41 Mohammed Shafi wrote:
> Hi Simon,
> 
> On Fri, Jul 6, 2012 at 6:20 PM, Simon Wunderlich
> 
> <simon.wunderlich@s2003.tu-chemnitz.de> wrote:
> > Hey,
> > 
> > we have trouble with an AR9330 (Hornet) based AP. This device has 2
> > antennas, and is supposed to support diversity (2 RX, 1 TX). However,
> > diversity is not really enabled because the "alternative" antenna is not
> > considered as "good", because not enough frames are received. It seems
> > that at 99% of the time, alt_rssi (read from rs->rs_rssi_ctl1) is invalid
> > (set to ATH9K_RSSI_BAD, -128), while main_rssi (read from
> > rs->rs_rssi_ctl0) usually has sane values.
> its been some good amount of time i had taken/tested take a look at
> this. actually Gabor Juhos added the support for AR9330.
> seems AR9003 family chipsets can exhibit this behaviour. i quickly
> checked with AR9285(AR9002) which seems to have a positive RSSI
> regulary in ctl0/ctl1.
> i also have a AR9485(AR9003 family) which might be similar to AR9330.
> just checked with removing the antenna in chain0 seems to bring
> positive values in rssi_ctl1.
> I just also observer the chain0 is "preferred" unless the RSSI is so
> poor in it so chain1 is taken.
> Please check if this work out for your case too.

Removing the antenna on chain0 doesn't change the alt_rssi value for us. We 
only noticed that frames with alt_rssi != -128 have  the value rx_ant_conf == 
1 (LNA2) when only chain1 is attached and rx_ant_conf == 2 (LNA1) when 
both/chain0-only are attached. Still most of the packets had alt_rssi == -128 
and we saw changes in curr_main_set and curr_alt_set (after lot of data was 
sent).

> also please see the throughput difference with/without antenna
> diversity with one antenna broken.

We (Simon and me) tried to play with the antennas and came to following 
results:

* Both attached: 45 Mibit/s
* chain0 attached: 45 Mibit/s
* chain1 attached: 15 seconds nearly nothing, after that 25 Mibit/s

> I will check out the proper technical reason for such a
> preference/compare with AR9285 (or) if its a bug in the code itself.
> thanks for looking into this feature!

Thanks

Kind regards,
	Sven

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]