Nr module (5G LENA) - desynchronised antennas: from ICI to CLI - Problem with SINR and ICI

[Stefano Biccari]

Hello everyone,

I'm new here so... nice to meet you. My idea would be to create a 5G simulation environment whit NR module (5G-LENA) where we can study the interference effects in the case of desynchronised antennas operating at the same frequency. The ns3 versione that I'm using is 3.45.

I tried to create a simplified basic structure for this, where there are two GNBs, each connected to a UE, which are then connected to the EPC and a remote server. In this simulation, I created both uplink and downlink traffic. Through the flow helper, I saw that the packets reach their destinations, so from a macroscopic point of view, it seems to work.
My goal at this stage is not yet to manage desynchronisation but to actually see that ICI is being managed. However, when I check the NR traces, particularly those relating to SINR, I see that it remains fixed at each transmission, which I would have expected to change based on the interfering moments of ICI between the two UEs and the two GNBs.

I realise that at the moment I am not yet managing a situation of desynchronised antennas, but my idea is to take it step by step. I don't know if a real situation of desynchronised antennas is actually feasible in ns3, but I have seen that there is the possibility among the settable parameters to set the TTD pattern, so my idea was then, once I understood how to fix the ICI and SINR issue, to set the two TTDs of the antennas as if they were out of phase by one slot and start to see what happens, such as DDDUD and DDDDU.

I have included my simulation code below, in case anyone with experience can spot any logical or implementation errors.

Since I'm here, I'll give you a preview of the tests I've done, in case you're interested. I added the option to enable beamforming, as I wanted to see if it would improve the number of lost packets, and spoiler alert: yes, it turns out that when I enable it, I no longer have any losses. However, I justified this because I have a large increase in gain, which increases the SINR, and since the packet drop is linked to this, they are probably no longer dropped. Another thing I am unsure about is the reason for the drop: whether they are lost in the air or whether they are received and then dropped.

Changing the distance of the UEs only changes the SINR of the individual UE, and not that of the other, as if the interference were not seen. Depending on how the distance is changed, the packet drop of the UE also changes, which makes sense if the loss on free space is working.

I hope I have been as clear as possible and that someone can help me in some way. But regardless, to you who are reading this, I thank you for taking the time to read it!

I look forward to any updates, and in the meantime, I will continue to work on it.

[Gabriel Ferreira]

I don't understand what you mean by desynchronized antennas.

Interference between different cells in same frequencies, we do support. 

For interference sources you can use PDSCH and/or CSI-IM. PDSCH alone only measures interference on used rgbs, which requires having data traffic to be able to measure anything significant. Full buffer if you want to measure entire bandwidth. CSI-IM on the other hand estimates wideband interference from other cells based on CSI-RS signals. 

[Stefano Biccari]

First of all, thank you for your reply.

By desynchronised antennas, I mean antennas that do not have their TTD aligned, with possible partial slot overlaps and, consequently, the generation of CLI. What I want to see in my final result is how even small desynchronisations lead to degradation of 5G transmission.

But before thinking about the CLI situation, I thought it would be better to understand how the ICI situation was being handled.

As for my simulation, I think I took the approach you suggested, which is to saturate the PDSCH and PUSCH, and I did this by inserting a host that sends and receives 1000 packets, using UDP apps. At that point, my idea was to insert the simplest possible scenario, i.e. one with a GNB and a user connected to it (GNB1 - UE1) and another GNB with its UE connected (GNB2 - UE2). This was done in a UMa scenario, without BF, with isotropic antennas and a 3GPP channel. These antennas are 200 m apart (gnb1: 0;0;0 - gnb2 0;200;0) and I first tried to put the two UEs in the same spot (0;100;0) and they have -12.558 and -12.3736 dB as SINR. If I move UE towards gnb1 and I have a situation where UE1: 0;50;0 and UE2: 0;100;0, I expect both to improve because one gets closer, reducing path loss, and the other suffers less ICI interference. The values they obtain are SINR -11.205 and -12.3721.

So there has been an improvement, but it is practically imperceptible in terms of ICI. And so I was wondering if it made sense that, even though we were now moving away from it, the improvement was so fleeting.

Another test I did was to place the UE 1 exactly where the gnb2 was, so that all its power would interfere with the UE2. What I got was that the SINR of the UE2 rightly dropped to -21 dB, while that of the UE1 obviously dropped even further because it moved away, reaching -40 dB.

So I was also wondering if these values made sense to you and if there was anything you would set to try to visualise these values better. Oh, right, I should point out that this data was from the UL.

I also tried inserting BF, and when I do so, the situation improves, bringing the SINR above 8 dB. Here too, I am not satisfied because the ideal BF works for me, but when I put in the real one, it gives me “died <signal.SIGSEGV: 11>”.

Thank you again for your time!

[Gabriel Ferreira]

I am still not quite sure what you mean by that. 

The TDD slots are aligned in time, since we do not model propagation delay nor local time innacuracies, which allowed us not to care about synchronization signalling/procedures. 

Regarding SINR, when using 3GPP channel, you cannot look at a single value, because it is stochastic. Collect sinr for UEs over time, then use a t-test to compare if distributions are statistically the same or are different. You can do it getting traces from received transport block traces.

Without the stacktrace of the real beamforming crash, I can't help much. 

[Stefano Biccari]

Hi! Thanks again for the reply.

So, regarding the SINR, in the end I enabled the traces through the PHY-layer helper, and through them I was able to observe the SINR I was referring to. From what I understand, the SINR is calculated at the slot level, not at the RB level. Also, from what I have seen in the calculation, what happens is that we have:

SINR = (useful received power) / (all received powers − useful received power + noise), all of this at the slot level.

The useful received power also takes the scenario into account, including its path loss.

However, once the position is set, the SINR remains fixed. So when you talk about stochastic behavior, are you telling me that depending on where the nodes are placed, with the same settings, the SINR changes in an unpredictable way (or rather, partially predictable based on how parameters and positions are set), but that at a fine-grained level you cannot know exactly how movement will affect the SINR itself?

However, once a fixed position is assigned to all nodes, the SINR should no longer vary at this point, because everything becomes deterministic. Is that correct? If so, it would make sense that my SINR values are constant, since my nodes are fixed.

As for desynchronization, I am aware that this is a simulator that, by default, assumes that this issue does not need to be handled. All antennas assume that the TDDs are aligned, and that is fine for me. I am not so much interested in simulating desynchronized antennas during the actual synchronization procedure, but rather in studying only the effects of poor synchronization. So, for example, when I have one antenna transmitting perfectly synchronized with all nearby antennas, but there is one that, instead of starting its slots at time 0, starts them with a delay of some number of nanoseconds, thereby generating ICI + CLI as interference.

Now, I do not know whether it is feasible to introduce some kind of offset that makes an antenna always start its slots slightly later; I have not yet really dug into this issue. For now, I was mainly interested in seeing whether both CLI and ICI were actually observed, and it seems that both are handled, because when I set two antennas with patterns such as: DDDUD and DDDDU. In the simulation where the patterns were aligned, I had −12 dB SINR, while in the one with offset patterns I obtained −18 dB. This is probably because the interfering signal is no longer the one from the UE that is transmitting, but from the gNB antenna which, by transmitting, naturally sends more power and thus overwhelms the signal.

So, at this point, I would mainly be interested in understanding whether, in your opinion, it is feasible as an idea to add a temporal offset to make an antenna start later in transmitting and receiving information (thus effectively adding a temporal offset to its TDD).

Additionally, do you have any idea why ideal beamforming works for me, whereas the realistic one generates the error mentioned above? I thought that, more or less, the way both of them operate would be roughly the same.

Thanks again in advance for your time!

Stefano Biccari

--
Posting to this group should follow these guidelines https://www.nsnam.org/wiki/Ns-3-users-guidelines-for-posting
---
You received this message because you are subscribed to the Google Groups "ns-3-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ns-3-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ns-3-users/765e2088-dcb1-409f-878a-6f436c2f656bn%40googlegroups.com.

[Kent Huns]

>>However, once a fixed position is assigned to all nodes, the SINR should no longer vary at this point, because everything becomes deterministic. Is that correct?

3GPP channel model:TR38.901 simulates multipath fading, and the angle of arrival of each path fluctuates because surrounding objects don't have to be fixed.

https://groups.google.com/g/ns-3-users/c/Igl9zAmuDY8/m/lxjEtMwWEgAJ

This kind of interference is often studied in dynamic TDD, full duplex or NTN, so it would be better to look for such researchers or examples in ns-3/5G-LENA forum.

To view this discussion visit https://groups.google.com/d/msgid/ns-3-users/CAOFwKy-5CM_oJXJTKjovd0CiRXOTjF3okJATdLN5jWj93nBe3A%40mail.gmail.com.

[Stefano Biccari]

Hello! Thank you for your reply!

But it's very strange. In the simulation I considered, I set the channel as a UMa. So there should already be integrated fading for this scenario. Yet when I set my UEs and GNBs to fixed, when I went to look at the traces of incoming packets on the UL side, I saw a fixed SINR, as if, given a position, the SINR value did not depend on time but only on who was actually transmitting at that moment, on the position of the nodes in the scenario and nothing else. It was as if a certain fading was set at that position, which, reading your reply, should not be the case. Compared to the code I ran, can you tell me if there is a part that I may have set incorrectly that causes me to get a fixed SINR over time?

Thank you very much!

[Gabriel Ferreira]

To have fast fading, you need some velocity. For the SINRs to change over time, you have both interference and channel updates to take into account.
If you channel update time is set to 0, it won't be updated at all, unless the nodes move, you change antenna, settings, etc.
If you have full buffer traffic causing interference, same nodes, same positions, same channel, yes, SINR will appear to be fixed.
If you don't want your nodes moving around, just the fast fading, use FastFadingConstantPositionMobilityModel from nr/utils/fast-fading-constant-position-mobility-model.cc.
That is what we use for testing.

[Stefano Biccari]

Hello! Thank you very much for your reply.

I understand what you said, so basically when I instantiate UE and gNB, if I don't introduce movement or fading that changes the SINR, it will appear fixed.

I just have one last question: I should create the scenario with hexagonal grid logic. I saw that there is a helper for doing this, but then it becomes inconvenient for me to manage some things I wanted to try, such as placing users at a specific point or moving the antenna (a scenario I need for the future creation of a WAB device). If I see the positions where the antennas are placed in the case of the hexagonal grid and I put them in the same position without actually using the grid helper, from a simulation point of view, do I get the same effects as if I had set up the grid, or does something change? Because depending on the answer, I need to study the grid helper and how to manage it in my simulation.

Thank you very much for your time and sorry for the endless questions (if I have any more doubts, I will take advantage of the advice given to me in the previous answers and ask in the group more relevant to 5G NR, now that I know it exists).

Thanks again!

Stefano Biccari

To view this discussion visit https://groups.google.com/d/msgid/ns-3-users/c3fa50e4-2297-419f-b914-f69b0bb635a4n%40googlegroups.com.