From: Benoit PAPILLAULT Date: Fri, 30 Apr 2010 12:19:12 +0200 Subject: [ath9k-devel] [PATCH] ath9k: Added get_survey callback in In-Reply-To: <0290AB2F0A8E124B9FFC144088A1A95609D69B6BB9@EX-MBS04.nbl.local> References: <0290AB2F0A8E124B9FFC144088A1A95609D69B6BB9@EX-MBS04.nbl.local> Message-ID: <4BDAAEA0.9070508@free.fr> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ath9k-devel@lists.ath9k.org Anton Puolakka a ?crit : > > Hi, > > recently we?ve been organising work that would happen in the area of > get_survey. Seems that Benoit is already working in the field pretty > much ? we waited for Holger?s mac80211 patch to be publicly available? > and the waiting is paying off now, it seems. > > Anyway, our interest is to have the noise (someone else?s signal in > terms of Shannon) on all channels. This shall be not so fast > operation. It is good to have faster only-for-one-channel option as > well. We are happy to help so if Benoit could show us a good area, we > could start the work in the very near future. I look forward to Benoit > to contact us. If You gonna do it solo, we just have to wait a bit more? > That was just a pretty straightforward patch since I was using debugfs last_noise file before. > > Jouni?s comment was the most welcomed as it is the first time I see > any pondering and discussion on design purposes and design decisions. > Am I following too low-level mailing list or am I just so out of the > clue what?s happening?? > Yup, I'm very open to design discussion as well. I don't find get_survey very convenient for my needs. Basically, I need signal & noise to compute SNR for peer nodes, this means I need to sample signal & noise at the same time. As such, having those value per packet was quite convenient. Digging into the details, the signal reported by the card is the the difference compared to the configured noise floor (and noise floor is not the same thing as noise to me). This means that noise floor calibration needs to be done properly and it takes some time (how much? I don't know). Then, we need "noise" as "someone's else" signal like you mentioned. I think we could use some value from the noise floor calibration to do that. It's just an idea, maybe it's better suited to use a fixed noise of -101 dBm anyway. At least, the definition chosen needs to be compatible with computing SNR for each peer node. > > So far we?ve bumped into this calibration thing, it is not too fast, > but it is the way we?ve done digging out the noise in the current > implementation. We do not exactly like this solution but we?ve made > some progress with it. > > ?and I can?t help it but I have to say that wherever signal strength > is available, noise should be available, too, in similar API or in the > same data struct. This is because SNR is vital information in any > possible case. > I agree. I don't know exactly how it could be done with mac80211 currently. > > Kind regards, > > Anton > > -- > > Anton Puolakka, VP Product Development > > http://www.7signal.com gsm:// +358 (45) 1234 020 > > *What 7signal is about in 100 seconds:* > http://www.youtube.com/watch?v=mlbN4JZTu7M > Regards, Benoit