search

xieyaxiongfly / Atheros-CSI-Tool

Tool for extracting CSI from off-the-shelf Commercial Atheros WiFi NIC
Other
128 stars 73 forks source link

Strange IQ Values using verified AR9382 and AR9344 SOC (running OpenWrt) #6

Open niklasser opened 7 years ago

niklasser commented 7 years ago

Hi @xieyaxiongfly , my name is Niklas. I'am currently writing my bachelor thesis at Karlsruhe Institut of Technology and trying to reproduce the results from your paper [Precise Power Delay Profiling with Commodity WiFi ]. Currently im trying to extract CSI Information with your tool on an Atheros 9344 SOC running OpenWrt. We can extract information but as of right now, they don't seem to make much sense. In comparision to the sample file, we cant see a steady increase to pi and then the jump to -pi for the phase angle.

We thought it might occur due to an unsupported chip (Ar9344). Thats why we also tried it with an AR9382 which is on the supported list on your homepage. The AR9382 is running in two notebooks using a Kernel 4.1. One notebooks uses the 64bit architecture, the other one is on 32bit. The results are similar to the NIC's running OpenWrt.

We also use the default hostapd.conf from your Repo.

This is the IQ-Data from the sample file. iq_sample And this is the IQ-Data from a packet we recorded. our_iq our_iq2 This is a phase we recorded. our_phase And this one is from the sample file. It looks as expected. phase_sample

UPDATE: This is a phase from the sample file and our recording plotted in one graph. The y-Axis is in Radians. The x-Axis ist in Subcarrier. plot

Since you discovered the presence of the CSI information by coincidence maybe you have encountered similar behavior before and you can suggest an approach to parse the buffer correctly? Or do you have a suggestion what else we could try to get similiar I/Q results?

ps: We will also experiment with the AR9582 which a 2 antenna version of the AR9580 - so we get a better understanding where the problem may lie.

niklasser commented 7 years ago

Quick update: We did try the AR9582 yesterday, the results were similar to my previous post.

The cards we used were the following mini-PCi Express cards

I will attach file that we recoreded with the AR9582.

test_garage_notebook_mid.dat.zip

xieyaxiongfly commented 7 years ago

Sorry bro, I was busy catching up a deadline.

Can you replot the figure of phase and use the same scale of y axis?

Or you can plot the phase from the sample file and your phase in the same figure. I feel it look different because of the scale.

Can you update that？

niklasser commented 7 years ago

Hey,

sure, i updated the first post. I think the y-Axis was already the same (in Radians) Here is also the plot: plot

xieyaxiongfly commented 7 years ago

Yes, it looks reasonable. How do you think?

niklasser commented 7 years ago

Well, we are currently not sure why those two phases are so different. The distance should have an influence on the phase, but we don't know why our phases look like a sine wave shifted by pi. This makes it hard for us to reproduce phases as suggested in your paper http://www.ntu.edu.sg/home/limo/papers/com292-xieA.pdf

The phases also behaves independent from the distance, frequency and even bandwidth we are using. We tried all bands available with ath9k with all bandwidth combinations (2,4GHz and 5GHz; HT40 and HT20) and always ended up with phases shaped like a sine wave.

distance_0_all_24_bands In the picture you can see the average phase of multiple packages (more than 1000) of all 2.4GHz channels (HT20) put next to each other for a fixed distance.

If our assumption is correct, they should "somewhat" overlap in order to create a continuous phase.

Best regards

EuclidGH commented 7 years ago

Hello Niklas,

it seems, we are almost on the same way, but with different goals. Could you send me more information about your set up? (k-ieso@hotmail.com) We can discuss the existing problems together. I use the Atheros-CSI for measurement set up for my master thesis at Ruhr Uni Bochum.

lqh929289158 commented 6 years ago

@niklasser @EuclidGH

  1. Why the phase information looks like a sine wave? We have solved it. Please see "Perceiving accurate CSI phases with commodity WiFi devices" in INFOCOM 17. The cause is non-linear error.
  2. You said "The phases also behaves independent from the distance." I can not agree with you more. I am interested in it too. Would mind providing your e-mail so that we can discuss it more?