5G NR Radio Network Analysis
Services
This is your Services section. This is a great place to give more information about the services you provide. You can write a general description of what your business offers then add more details below.This section can be adapted for your website.
You may choose to highlight other things like courses or programs, or to share special features about your business that you want to promote. Double click on the text box to edit the text and make it your own.
5G NR DL Throughput Analysis
One of the most common problem in the mobile radio network is to analyze the root-cause of poor DL/UL throughput. Issues might lies on different network elements (UE, gNB, UPF, Servers) i.e. analysis must carry on different interfaces deploying different tools/methods.This example case illustrates the comparison of different HTTP Download with various setting i.e. single HTTP server with 1...5 threads and several HTTP servers.
The above cdf plot of the total PDSCH throughput shows that:
-
The median value (50%tile) of 4 threads with different HTTP servers has DL throughput better than single HTTP server with 3, 4 and threads by approx. 500Mbps!
-
The median value of single server with 1 or 2 threads has better DL throughput comparted to 3, 4 and 5 threads by approx. 250...350Mbps
-
The peak throughput of 4 threads with different HTTP server has the best value, which is approx. 100Mbps better than single server with 1 and 2 threads and approx. 500 Mbps better than single server 3 and 5 threads.
In summary, the 4 thread with different server provide the best performance. It is questionable, with single server, why more threads provide less DL throughput. With ROMES, we can check the NR radio quality to see whether or not it affects the PDSCH throughput.
ROMES: 1T Single Server HTTP Download Sessions
The below 2D graphs show a single server with 1 thread:
-
Net PDSCH throughput (Mbps): the graph show a typical TCP slow ramping behavior. In the first download session, it took more than 25s to reach the saturate data rate (--> Suggest further check-up with the Flow Control configuration at the server side). The 2nd download session contain a throughput dip (highlight in black dot circle). The lower assignment of PDSCH Layer and MCS make the throughput smaller. However, if we compare the CQI and RI reported over the same period, we see that UE see the radio quality better than what the gNB scheduled. Particularly on PDSCH Layer compared to RI. This suggests a further check-up on the S1-U interface whether the poorer DL scheduling of gNB is due to radio condition or due to lack of data in the buffer
-
MCS and Avg. PDSCH Layer indicate radio quality seen by the gNB
-
CQI and RI indicate radio quality seen by the UE
-
Number of PDSCH RB and PDSCH Slot Occupation indicate DL scheduling function of gNB
Wireshark I/O Graph: S1-U
The Wireshark I/O graph of S1-U interface below shows the all packets (bps) of the single server with 1 thread. The TCP slow ramping up behavior is seen at S1-U interface. The 2nd download session shows a throughput dip (--> this indicate that the low throughput that seen earlier in ROMES does not cause by radio condition, but due to the low buffer in gNB).
The TCP Retransmission (tcp.analysis.retransmission) shown below is in the unit of bit as well.
Note that the retransmission is negligible on S1-U interface.
