MEB Rooftop mMIMO Experiments

[Brent Kenney]

Dear POWDER Community,

Has anyone had success with the MEB Rooftop mMIMO system using the MATLAB scripts?  I just tried to run the basic rl_ofdm_simo.m and rl_ofdm_mimo.m scripts, but I was unable to find the preamble (i.e., bad correlations on almost all antennas each time).  I'm guessing that I had something configured incorrectly.  It also looked like BS antennas 10, 26, and 30 were not receiving data, but that should not have affected the other paths.  I am only using one radio chain per IRIS SDR at this point (32 BS receivers).  I believe that these same MATLAB scripts were running fine on the anechoic chamber mMIMO system a few months ago.  Any tips would be appreciated.  I am going to try it again next week.

Best regards,

Brent Kenney

[Oscar Bejarano]

Hi Brent,

I have used both scripts in the MEB rooftop successfully. A few things to consider:

1) Unfortunately there's some 3rd party interference in the bands we can use so that makes the scripts fail from time to time.
You can check if there are any active interferers in the band you are using, by running SISO_RX.py with at least one board.

2) The defaults (gains/serials) are set for the chamber BS. But, given that we no longer have the chamber, I will go ahead and change those defaults.
You may want to play with the gains a bit though. That's usually the main culprit.

I'll try to run some tests this weekend and will let you know as soon as I have any updates.

Best,

-Oscar

--
You received this message because you are subscribed to the Google Groups "Powder Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to powder-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/powder-users/CAOPZb19P5bn5kxjvUtRPjLUz0r0cFCudO-m3fvJ7vMnk8ggY0g%40mail.gmail.com.

[Oscar Bejarano]

Hi Brent,

I just ran a few tests in the MEB rooftop deployment. Here's some info about the tests and some results:

1) I believe in one of our meetings I mentioned we were working on overhauling the Matlab framework to make it simpler for you to modify and test. For example: pushing radio calls to a driver instead of having them exposed.
Well, we have made good progress and those are the scripts I ended up testing. We have not merged this new branch with our main codebase so you can just do a `git checkout matlab_new_driver` to use them. Make sure my latest changes from today are in ('git log'). There's some bugs we need to take care of before releasing the final version (I'm sure you'll see them).
The scripts are:
rl_siso_NEW.m
rl_ul_mimo_NEW.m
rl_dl_mimo_NEW.m

I've also changed the defaults to the MEB hardware (including gains, etc).

2) You'll see that you can finally use both channels A and B (both antennas) in the base station boards. Enabling this was a bit tricky for several reasons so we don't really support that in the UEs, yet. Let us know if that's something you need urgently to see what we can do about it.

3) I'm attaching some (good) results from running all three scripts. SISO seems to be the most reliable one. I found a couple of bugs in the other two, however, these bugs mainly consist of bad error-catching when the radios fail to detect signals etc. In the UL mimo script, you'll notice that performance is strongly related to the condition number of the channel matrix (expected), and you'll also notice that having a single user is more reliable than having both users enabled (also expected).

3) In my experience, these scripts are a lot more reliable than what we had before but they do fail, especially outdoors (they are much more stable in our indoor testbed for sure). You can even see it from the results I'm sharing. For instance, you can trigger several consecutive transmissions (increase N_FRM) and in some cases you get zero errors but in others your correlator might fail. We will continue to work on robustness but there's so much we can do when channels are "bad". The other thing we need to work on is 'gain setting as a function of number of antennas'. We don't yet have a smart mechanism that adapts tx gain as you increase antennas (so the array gain leads to saturation at the receiver).

4) Max gain should be 81, but the current defaults seemed to work in most instances so you might be fine using those values.

I hope this helps. Please let me know if you have a specific scenario you cannot get to work so we can help tune things up.
We'll continue to try to facilitate the process of running tests so we appreciate any feedback you may have.

Best,

-Oscar

To view this discussion on the web visit https://groups.google.com/d/msgid/powder-users/cea4365a-4071-2379-8f5d-2a120af06dcf%40rice.edu.

[Brent Kenney]

Oscar,

Thank you for providing this information.  It was very helpful.  I found that the SISO script was very reliable, but the MIMO script was fully successful less than 20% of the time.  I had to modify some of the plots in that script to get a decent picture of what was going on.  The most useful plot was looking at the correlation data.  If the correlation was found, then things worked fine.  I saw a number of cases where there was a correlation/synchronization error on each BS antenna, which points to a Tx problem in the UE.  I also saw cases where the timing of the UE signals was considerably off (i.e., 2nd UE transmitted its pilot during part of the first UE's payload).  By looking at the received signals, I saw cases where only part of the Tx payload was sent.  When I get on the system again, I'll try to characterize the behavior that I'm seeing.  Despite these issues, I was happy to see things working part of the time.  Our group can start developing tests around the current limitations.

Thanks again,

Brent Kenney

To view this discussion on the web visit https://groups.google.com/d/msgid/powder-users/77e85d27-98b4-4635-a412-d095612ff2e8%40rice.edu.