High Latency 5G NSA

5G NR
, , ,
1

Hi, we have noticed a very strange behaviour in our network where 5G latency has high variation. Percentile 10 and 50 are fine but Percentile 90 is very high.
If i lock user to LTE only, latency is uniform and better.
From Stationary DT , it is clear that when UE TX power is high>22dBm, the latency increases. Checked radio condition and cant see any change in RSRP/SINR/Pathloss.
Cant figure out why UE TX power is increasing without any change in radio condition

Next - Disabled closed loop power control but issue still persist. P0nominal PUCCH/PUSCH/F1F3 all set to -90
I feel there is some issue with power control. Ideas will be appreciated

2

Check UL retransmission rate. If high maybe there is UL interference, too many retransmissions cause high latency.

4

According to your description, maybe for 5G you have a more aggressive ACM (adaptive coding and modulation) algorithm in your system. So even when the channel condition is good, the Tx tried to get the maximal throughput and push up Tx power within its power budget. Thus receiver modem’s latency increases or even as @upf mentioned reTx rate increase, which definitely results in high latency.

5

After some more analysis, i figured out that latency increases when UL goes on LTE leg only
Case 1 - UL on NR+LTE - Latency good
Case 2 - UL on LTE leg only( due to high pathloss, SINR still good) - high latency variation and poor samples
Case 3 - UL on LTE (No 5G ENDC available) - Latency good

So the problem is when due to poor UL condition ( mainly high pathloss, SINR still good ) , UL leg switches to LTE only.
I have seen in few networks that they only use SINR thresholds for UL data split and Pathloss threshold is kept very relaxed so that UL split is mainly decided by UL SINR

Resolution- I think we need to keep data on NR leg more if SINR is still good so thinking of pushing pathloss threshold. Also we can try increase ulDataSplitThreshold . Currently set as b3200What do u think?

6

Let me recap your case 2,

In 5G NSA ENDC mode, the LTE is operating in lower frequency band e.g., (600MHz~800MHz), and NR is on higher frequency band, e.g. mid-band (2.5GHz or above), and the pathloss is higher for NR, but SINR is comparable for LTE and NR legs since there is less interference in mid-band (?).

In this scenario, if the traffic split is based on pathloss, it will be switched to LTE. But if the traffic is split based on SINR, it will be equally split. (But your description on SINR and pathloss threshold setting is somehow conflict here.

I think if you put more strict pathloss threshold here, more traffic may go to LTE.

So I think the solution should be, you should split the traffic based on the channel capacity (there is a terminology for it, but I can’t remember it now) threshold, which is derived from both the SINR and channel bandwidth. If you configure the traffic splitting like this, you can get what you want.

Thanks for sharing your thoughts,

Wenzhen Li

7

For case 3, is UE using anchor cell or not?

8

Hi Wndy, In Nokia, UL split can only be done through pathloss or SINR. Currently for UL data to switch to LTE completely, SINR threshold is -.4.5 and pathloss is 118. Once the pathloss threshold is reached, UL switch to LTE only and DL still manage to go on 5G leg. 5G slot utilization go down as UL traffic is not moving on LTE leg. Due to this i can degradation in latency and every 1 out of 4 pings is bad approx 100ms
I am wandering whats the best way to tackle this. Relaxing thresholds too much means that UL might get drag on NR leg which again will cause poor customer experiece

9

Case 3 - UE is in LTE only mode, NR coverage goes bad and ENDC is released so both DL and UL is on LTE

10

Is the LTE cell anchor cell or not? Is it the cell the one for EN-DC?

11

yes the cell is same on LTE

12

If in case 2 LTE is not too bad, maybe you can check X2 bandwidth between LTE and NR cell.

13

According to your description, I think you are saying that, pathloss threshold is used to switch UL traffic from NR to LTE, and SINR threshold is used to split DL and UL traffic between NR and LTE. This means LTE has a larger coverage because its lower carrier frequency or narrower channel bandwidth.

You also mentioned that NR’s slot utilization went down due to UL traffic occupied in the LTE leg, here LTE and NR are sharing the same channel? If not, they can concurrently operate, right?

So I want to know,

is NR in TDD or FDD mode?

Is LTE in TDD or FDD mode?

NR and LTE are sharing the same channel in UL and DL? Which channels are they using?

You also mentioned in the previous message it is ENDC, so I think they are using different channels. So please clarify the above information first.

You also mentioned before your system is NSA, so I assume that all RRC (control, configuration, and scheduling information) traffic go through DL LTE, including UL and DL slot allocation.

After all these information is clarified, we can brainstorm the possible optimization.

Thanks for sharing!