In this example we understand how the scheduling algorithm affects the UDP download throughput of a multi-user (UE) system where the UEs are at different distances from the gNB. Open NetSim, Select Examples ->5G NR ->Scheduling then click on the tile in the middle panel to load the example as shown in below screenshot
Figure 4‑31: List of scenarios for the example of Scheduling
The following network diagram illustrates, what the NetSim UI displays when you open the example configuration file.
Figure 4‑32: Network set up for studying the Scheduling
UEs at different distances and channel is not time varying#
Configuring the scheduling algorithm, and parameter settings in example config files
-
Set grid length as 5000m from Environment setting.
-
Set distance as follows.
-
gNB_7 to UE_8 = 500m
-
gNB_7 to UE_9 = 1000m, and
-
gNB_7 to UE_10 = 1500m
-
-
Go to Wired link properties and set the following properties as shown below Table 4‑21.
Wired Link Properties | |
---|---|
Uplink Speed | 5000 Mbps |
Downlink Speed | 5000 Mbps |
Uplink and downlink BER | 0.0000001 |
Table 4‑21: Wired Link Properties
- Go to gNB properties à Interface (5G_RAN), set the following properties as shown below Table 4‑22. In the first sample the scheduling type is set to Round Robin, in the second to Proportional fair, and in the third to Max throughput.
Properties | |
---|---|
Data Link Layer Properties | |
Scheduling Type | Varies: Proportional Fair, Max throughput, Round Robin |
Physical Layer Properties | |
CA Type | SINGLE_BAND |
CA Configuration | n78 |
CA1 | |
Numerology | 1 |
Channel Bandwidth | 100 MHz |
Outdoor_Scenario | URBAN_MACRO |
LOS NLOS Selection | USER_DEFINED |
LOS Probabillity | 1 |
Pathloss Model | 3GPPTR38.901-7.4.1 |
Shadow Fading Model | None |
Fading and Beamforming | NO_FADING_MIMO_UNIT_GAIN |
O2I Building Penetration Model | Low Loss Model |
Table 4‑22: gNB >Interface (5G_RAN) >Data Link layer properties
-
Set Tx_Antenna_Count as 2 and Rx_Antenna_Count as 1 in gNB properties.
-
Set Tx_Antenna_Count as 1 and Rx_Antenna_Count as 2 in all the UEs.
-
Go to Application properties and set the following properties as shown below Table 4‑23.
Application Properties | |||
---|---|---|---|
Application 1 | Application 2 | Application 3 | |
Application Type | CBR | CBR | CBR |
Source ID | 12 | 12 | 12 |
Destination ID | 8 | 9 | 10 |
QoS | UGS | UGS | UGS |
Transport Protocol | UDP | UDP | UDP |
Packet Size | 1460Bytes | 1460Bytes | 1460Bytes |
Inter-arrival time | 10μs | 10μs | 10μs |
Start Time | 1s | 1s | 1s |
Table 4‑23: Application properties
-
Plots are enabled in NetSim GUI.
-
Run Simulation for 1.5s and note down throughput value in the results window in each sample. Recall that each sample has a different scheduling algorithm configured.
Results and discussions
The results with all the three UEs simultaneously downloading data is as given below.
Throughput (Mbps) | ||||
---|---|---|---|---|
Scheduling | Application 1 | Application 2 | Application 3 | Aggregate |
Round Robin | 170.55 | 88.81 | 42.46 | 301.82 |
Proportional Fair | 170.55 | 88.81 | 42.46 | 301.82 |
Max Throughput | 510.36 | 0 | 0 | 510.36 |
Table 4‑24: UDP download throughputs for different scheduling algorithms when all three 3 UEs simultaneously downloading data
Next, consider a scenario with only one of the UEs seeing DL traffic (we don’t provide inbuilt configuration file for this, and since it is a simple exercise for a user) First, run for the UE at 1000m, then for UE at 1500m and finally for UE at 2000m. This gives the maximum achievable throughput per node since the gNB resources (bandwidth) is not shared between 3 UEs and is fully dedicated to just one UE. The results are below.
Distance from gNB (m) | Application ID | Throughput (Mbps) | Remarks |
---|---|---|---|
500 | 1 | 510.36 | UE 1 alone has full buffer DL traffic |
1000 | 2 | 266.77 | UE 2 alone has full buffer DL traffic |
1500 | 3 | 127.56 | UE 3 alone has full buffer DL traffic |
Table 4‑25: UE throughputs if they were run standalone (without the other UEs downloading data)
The PHY rate is decided per the received SNR. Therefore, a UE closer to the gNB will get a higher date rate than a UE further away. In this example the distances from the gNB are such that UE10_Distance > UE9_Distance > UE8_Distance.
In Round Robin PRBs are allocated equally among all three nodes. However, throughputs are in the order UE8 > UE9 > UE10 because of their distances from the gNB. The individual throughputs seen by each of the UEs is exactly $\frac{1}{3}$ of the throughput as shown in Table 4‑25.The PF scheduler results will match that of the RR scheduler since the channel is not time varying. In Max throughput scheduling the PRBs are allocated such that the system gets the maximum download throughput. The nearest UE will get all the resources and its throughput will be 3 times the throughput of the UE which got the max throughout in RR.
UEs equidistant with time varying channel. RR vs. PF scheduling#
A difference in the performance of the RR and PF schedulers can be seen when the channel is time varying (of the order of the coherence time which is 10ms). We consider the following case: all UEs are initially at a distance d= 2000 m from the gNB. Then the UEs move away from the gNB at the same speed of 0.1m every 10ms (or 0.01s ), which is a speed of 10m/s. The simulation is run for 10s, and the UEs end up at a distance of 2000 + 10 × 10 = 2100m from the gNB. Note that UEs are at all times equidistant from the gNB and hence pathloss is the same (at all times) for all UEs. To induce time varying randomness in the channel we enable log normal shadow fading. Thus, every time the UE moves, NetSim draws a normally distributed random variable from N ∼ (0, 4) dB, as the additional loss. Under these conditions, the RR scheduler would allot resources to the UEs in a round robin fashion, whereas the PF scheduler would give preference to the UE which sees the best channel. The channel quality is dependent on the draw from N ∼ (0, 4) since all UEs pathlosses are equal. The results are shown in Table 4‑27.
Configuring the scheduling algorithm, and parameter settings in example config files
-
Set grid length as 8000m from Environment setting
-
Distance between gNB and UEs is 2000m.
-
File based mobility is set in all UEs to move away from gNB at the > speed of 0.1m every 10ms(or 0.01s ).
-
Go to gNB properties à Interface (5G_RAN), set the following > properties as shown below table. In the first sample the > scheduling type is set to Proportional fair, in the second to > Round Robin.
Properties | |
---|---|
Data Link Layer Properties | |
Scheduling Type | Varies: Proportional Fair, Round Robin |
Physical Layer Properties | |
CA Type | SINGLE_BAND |
CA Configuration | n78 |
CA1 | |
Numerology | 0 |
Channel Bandwidth | 10 MHz |
Outdoor_Scenario | URBAN_MACRO |
LOS NLOS Selection | USER_DEFINED |
LOS Probabillity | 1 |
Pathloss Model | 3GPPTR38.901-7.4.1 |
Shadow Fading Model | LOG NORMAL |
Fading and Beamforming | NO_FADING_MIMO_UNIT_GAIN |
O2I Building Penetration Model | Low Loss Model |
Table 4‑26: Data Link Layer Properties
-
Go to Application properties and set the properties as shown Table 4‑23 with Inter arrival time of 389 μs.
-
Run Simulation for 11s and note down throughput value in the results window in each sample.
Results
UE ID | RR Throughput (Mbps) |
PF Throughput (Mbps) |
---|---|---|
UE1 | 12.33 | 13.99 |
UE2 | 12.32 | 14.13 |
UE3 | 12.15 | 13.89 |
Table 4‑27: Comparison of PF vs. RR throughput in a case involving time varying channels