There is a problem with OFDM weak signal detection in ath9k.  I'm
using a pair of AR9220-based cards as AP and STA, cabled together
through a series of attenuators.  Each card is putting out a clean
+5dBm.  In 802.11g mode, the cards switch abruptly from 54Mbps to
11Mbps at -72dBm actual receive signal strength, and rc_stats shows
that no more OFDM packets are being received (none of minstrel's MRR
attempts are successful at OFDM modulations).  The CCK rates are used
until link is lost entirely at -98dBm.  In HT20 mode at 2.4GHz the
cards stop being able to receive unicast traffic at -75dBm, and
finally lose association at -88dBm.

If I set ATH9K_ANI_USE_OFDM_WEAK_SIG to false, I can receive OFDM down
to -86dBm in 802.11g mode, and HT20 works down to -90dBm.  Has anyone
else seen this?  While I can receive OFDM down to -86dBm as opposed to
-72dBm, I now lose association at -92dBm as opposed to -98dBm (so I
seem to pick up 14dB of usable OFDM range while losing about 6dB of
overall receive sensitivity).

I'm running compat-wireless 2010-10-19, and will try 12-02, but I
really don't see anything in the diffs that would address this.  I
still need to enable ANI debug, so I can see what the noise immunity
level is doing, and what the various thresholds are set to.  I'll post
my findings, in case anyone is interested.

On a side-note, but somewhat related, has anyone tried calibrating the
reported signal strengths to reality?  The noise floor used in that
calculation is a static -95dBm, so I know right off the bat it won't
be that accurate.  In 802.11g mode it is usually between 4 and 8dB
lower than reality (iw will report -80, when in actuality it is
-74dBm).  In HT20 mode, it is usually between 8 and 12dB lower than
reality (iw will report -77 when in actuality it is -67dBm).  Maybe
getting those readings correct is as simple as using the noise floor
determined by the do_getnf priv op.


Brian Prodoehl
_______________________________________________
ath9k-devel mailing list
ath9k-devel@lists.ath9k.org
https://lists.ath9k.org/mailman/listinfo/ath9k-devel

Reply via email to