How to solve the voice delay problem after core network upgrade

picture

After an operator upgraded its core network, big data showed that the voice connection delay increased slightly.

picture

Performance statistics analysis

According to big data statistics, after the core network was upgraded, the voice service indicators fluctuated as follows.

1. The average latency of EPSFB increased by about 50ms and then stabilized, as shown in Figure 1.

Figure 1 Connection delay

2. The average duration of EPSFB in the core network is increased by about 40~50ms, a slight increase, as shown in Figure 2.

Figure 2 Average duration of EPSFB process

3. The process success rate indicator of the core network EPSFB increased by about 1% after the upgrade, which is a significant improvement, as shown in Figure 3.

Figure 3 Success rate indicators

4. The EPSFB indicators after the core network upgrade in other provinces are the same, as shown in Figures 4 and 5.

Figure 4 Process success rate

Figure 5 Average process duration

5. The average latency indicator of VoNR did not change. However, the latency indicator increased significantly and fluctuated after one week, as shown in Figure 6.

Figure 6 Latency indicators

6. Through the analysis of big data results, when the number of VoNR-to-VoNR calls decreases, the latency will increase and fluctuate greatly, as shown in Figure 7.

Figure 7 Connection delay

7. The indicators of VoNR calling VoNR in each city are broken down. When the number of calls decreases, the delay increases, and when the number of calls increases, the delay decreases. It can be seen that when the total number of calls is small, too little data sampling has a greater impact on the delay fluctuation, as shown in Figure 8.

Figure 8 Comparison of city indicators

8. Through statistical summary:

a. EPSFB success rate The core network has been greatly improved after the upgrade, while the average delay of the process has increased slightly.

b. The success rate and latency of VoNR2VoNR did not change significantly after the core network upgrade. The latency fluctuation of VoNR2VoNR is affected by the number of calls. When the number of statistical sampling times is large, the latency will tend to the real average latency value.

Business summary analysis

According to the comprehensive performance statistics and analysis results, the core network has optimized the voice service process, especially EPSFB, after this upgrade, which has improved the success rate, but also introduced a small increase in average latency. Based on this result, the core network side analysis results are as follows.

1. In the voice process, the SMF side has made many process optimizations. When establishing a dedicated bearer, the SMF receives event conflicts from other processes (for example, receiving a forward Xn switch). The SMF will handle the process conflict. The current SMF process conflict strategy is 3s×2 times, that is, each time the voice dedicated bearer establishment conflicts/timeouts, it will try to reestablish after 3 seconds, and try to establish it at most 2 times.

2. SMF pre-upgrade processing: Before the upgrade, all SMF event conflicts share the same policy configuration, which causes conflicting processes such as Xn switching, which occupy the total number of conflicting policies. The dedicated load cannot be triggered after the conflict because the maximum number of times is exhausted twice, and the voice call fails directly.

3. Post-SMF upgrade processing: After SMF upgrade, the conflict strategies of voice process, NGC process, and EPC process are configured differently, and no longer share the same strategy configuration. NGC process conflicts (such as Xn switching) no longer occupy the conflict count of voice process, as shown in Figure 9.

picture

Figure 9 Configuration diagram

4. Therefore, after the SMF version is upgraded, the voice establishment of the dedicated load conflicts with the business process. SMF can ensure that the dedicated load establishment process continues to pop up for retry to ensure that the voice process can be successful.

5. The impact of this optimization on the EPSFB process is shown in Figure 10.

Figure 10 EPSFB signaling process

a. EPSFB requires switching from 5G to 4G, and its process involves many interactive messages, resulting in a high probability of conflicts.

b. In the previous version, due to the failure to distinguish the conflict strategies under multiple conflicts, the subsequent voice could not try to rebuild the dedicated load (the number of times was exhausted).

c. This upgrade optimizes the conflict strategy implementation type distinction. The process that failed before the original upgrade can succeed after the upgrade through a reasonable dedicated load reconstruction method, but the total delay of this EPSFB establishment is not less than 3 seconds (a timeout reconstruction process must be performed), so this part of the process will simultaneously improve the success rate and delay of EPSFB.

6. The impact of this optimization on the VoNR process is shown in Figure 11.

Figure 11 VoNR signaling process

a. In the voice process of VoNR, there is no need to switch from 5G to 4G, the overall delay and process are reduced, and there are fewer intervals of multiple conflicts. Although this optimization is applicable to the VoNR process, the probability of triggering is low because the conflict scenario is small.

b. Therefore, the core network upgrade has no significant effect on the overall latency of VoNR.

7. The VoNR latency analysis is shown in Figure 12.

Figure 12 VoNR latency analysis

a. The VoNR call origination delay indicator (VoNR calling and called including EPSFB) has been showing a stable trend.

b. The VoNR2VoNR delay indicator (both the calling and the called are VoNR) fluctuates greatly as the number of users increases.

c. Therefore, the delay of VoNR mainly occurs in the called party side process.

d. Through big data, we can obtain the session signaling backtracking of high-latency VoNR calling VoNR. The calls with increased latency are all caused by the called party’s multiple paging calls, which increases the latency.

e. User 1 signaling is shown in Figures 13 and 14.

Figure 13 User 1 signaling process 1

Figure 14 User 1 signaling process 2

f. User 2 signaling is shown in Figures 15 and 16.

Figure 15 User 2 signaling process 1

Figure 16 User 2 signaling process 2

When the number of calls is small, individual multiple paging processes will increase the overall average latency and cause large fluctuations in the indicator. When the number of calls increases, the average session latency will be smoothed and present the overall latency indicator.

picture

1. After the SMF upgrade, the conflict strategy scenario was optimized, the defects of the voice process before the upgrade were solved, and the success rate of voice establishment was improved. However, the success of this part was achieved through the optimization of the conflict strategy, and the process delay would be extended during the conflict (no less than 3 seconds, maximum 6 seconds).

2. After the SMF upgrade, the EPSFB success rate increased and the latency also slightly improved. The latency of VoNR did not change before and after the upgrade.

picture