Network Performance Measurements under Beam Switching

太长了懒得看. 这种 "毫无idea创新+纯测量类型" 的文章看的我真想吐... gemini总结一下:

In this section, we first present a detailed analysis and case studies on how obstructions lead to dynamic beam switching events, significantly impacting network performance in both stationary and mobile scenarios. Then, we illustrate how UT mobility influences the FOV and the visible satellites, contributing to more frequent beam switching events in mobile UTs. Finally, we assess the accuracy of the proposed mobility-aware satellite identification algorithm.

Table 1 summarizes the outage event statistics observed in both mobile and stationary obstructed scenarios. Both statistics are collected over a 5-hour measurement period. Note that the stationary obstructed UT (Fig. 1b) experienced more OBSTRUCTED outages due to the dense tree coverage. However, the mobile UT encounted more SKY_SEARCH outages, as transient obstructions such as highway overpasses frequently blocked the LoS during movement or FOV changes due to UT mobility, and it took longer for the UT to establish a new connection with a visible satellite. Additionally, long term performance monitoring indicates that, for the UT in Fig. 1b, the average obstructed time ratio has been reduced from around 10% in 2024 to less than 1% in December 2025, contributing to the proactive beam switching mechanism and the increased satellite density.

• 总体统计: 移动和固定场景的中断统计均基于 5 小时的测量数据.
• 中断类型差异:
  • 固定受阻 UT: 由于茂密的树木覆盖, 遭遇了更多的 OBSTRUCTED (受阻)中断.
  • 移动 UT: 遭遇了更多的 SKY_SEARCH (搜星)中断, 因为高速公路立交桥等瞬时障碍物在移动中频繁阻挡视距(LoS), 或者 UT 移动导致视场(FOV)变化, 且 UT 建立新连接所需时间更长.
• 长期性能改善: 对于固定受阻 UT, 平均受阻时间比率从 2024 年的约 10% 降至 2025 年 12 月的不到 1%, 这归功于主动波束切换机制和卫星密度的增加.

5.1 Stationary UT

Fig. 8a presents a snapshot of three timeslots of network performance for a stationary UT experiencing partial obstructions. Fig. 8b to Fig. 8d illustrate the corresponding cumulative obstruction maps and the connected satellite trajectories at the end of each timeslot.

At 58:57, after a regular handover to STARLINK-35165, the UT made a beam switching attempt to STARLINK-4709 to mitigate an OBSTRUCTED event resulting from low signal quality. However, as shown in Fig. 8b, for the remaining duration of this timeslot, the UT continued to experience lower signal quality without making further beam switching attempts. The downlink throughput performance dropped and continued to degrade at around 50 Mbps until the next regular handover at 59:12 happened.

In the second timeslot, following the regular handover at 59:12 to STARLINK-5271, the UT immediately attempted to conduct multiple beam switching to circumvent the OBSTRUCTED event. Eventually, the UT switched back to STARLINK-5271. The throughput performance started to recover until the connection to STARLINK-5271 was established, with a fluctuated throughput performance below 250 Mbps. As shown in Fig. 8c, it had a clear LoS without obstructions and strong signal quality for the rest of this timeslot.

In the third timeslot, after the regular handover to STARLINK-34280 at 59:27, the UT successfully kept to maintain the target bitrate at 250 Mbps without experiencing any outage events so no need to conduct any dynamic beam switching, as shown in Fig. 8d with the clear LoS.

• 切换失败导致性能下降:
  • 当 UT 遭遇因信号质量低而引发的 OBSTRUCTED 事件时, 会尝试进行动态波束切换(如切换至 STARLINK-4709)
  • 但如果切换未能改善信号质量且未进行进一步尝试, 下行吞吐量会下降并维持在约 50 Mbps 的低水平, 直到下一次常规切换发生
• 多次切换尝试:
  • 在某些情况下(如切换至 STARLINK-5271 后), UT 会立即进行多次波束切换尝试以规避遮挡, 最终可能切回原卫星
  • 吞吐量在连接建立前会波动, 但在无遮挡且信号强的情况下能维持在 250 Mbps 的目标比特率

5.2 Mobile UT

In Fig. 9, we present the impact of outage events and dynamic beam switching on network performance for a mobile Starlink UT, with both downlink and uplink throughput performance.

In the first timeslot of Fig. 9a, we observe a substantial downlink throughput drop (approximately 3 s) caused by a large highway sign. Although the obstruction map shows a noticeable SNR degradation, the UT maintains its serving beam and does not initiate a beam switch, indicating that the connection remains stable despite the short-term blockage. In the subsequent timeslot, a regular handover transitions the connection to STARLINK-35320. Shortly after, at 12:36, a bridge induces a brief outage (approximately 1 s), which triggers a beam switch to STARLINK-31884. Following this switch, the downlink throughput drops to roughly half of that observed with STARLINK-35320. At 12:42, when the next regular handover occurs, the UT switched back to STARLINK-35320, and the downlink throughput promptly recovers to its nominal level, on par with the previous timeslot before the outage and beam switch happened. This suggests that the UT, when forced to conduct a beam switch due to a hard obstruction event, falls back to a non-primary or backup beam in the serving cell with markedly lower performance.

In the first timeslot of Fig. 9b, we observe a significant uplink throughput drop caused by another large highway sign. One second before the regular handover at 06:57, a bridge-induced LoS obstruction triggers a beam switch, after which the uplink throughput requires (approximately 1 s) to recover during the next timeslot. A similar pattern appears later in the second timeslot. At 07:09, a bridge caused an outage that triggers a beam switch from STARLINK-30529 to STARLINK-5612. As in the downlink case, the uplink throughput does not fully recover until the next regular handover, when the UT returns to a more favorable serving satellite. At 07:12, a regular handover switches the serving satellite to STARLINK-1276. Immediately after a subsequent outage, the UT briefly switches to STARLINK-34550 before returning to STARLINK-1276. Once the vehicle exits the bridge, the UT re-establishes a stable connection with STARLINK-1276, followed by an approximately 2 s recovery period before throughput returns to nominal levels. Fig. 9c shows the GPS track of both events.

These events reveal a clear asymmetry between schedule-driven regular handovers and dynamic beam switches in LEO satellite networks. Schedule-driven handovers are pre-coordinated and UT-initiated based on cached satellite schedules, during which packets are briefly bicasted over both the old and new beams [4]. This intentional overlap allows the UT to transition to a highquality, pre-selected serving beam, often anchored to a planned and well-provisioned ground station, with only short-lived latency spikes and minimal throughput penalty. In contrast, dynamic beam switches triggered by sudden LoS obstructions are reactive and prioritize link continuity over link quality, without the benefit of packet bicasting or advance coordination. In these cases, the UT is often reassigned to a sub-optimal backup beam within the current serving cell, which may be anchored to a different or less favorable ground station, potentially introducing additional backhaul latency or congestion. Because no bicasting window exists, the UT must fully re-establish its link, resulting in prolonged throughput degradation. Notably, even after LoS conditions are restored, the UT does not immediately re-evaluate or return to the optimal beam. Throughput recovery is frequently delayed until the next scheduled handover, when bicasting and coordinated ground-station selection again become available.

(1) 瞬时遮挡的影响:

大型路牌或桥梁会导致短暂的吞吐量大幅下降(约 3 秒或 1 秒).

• 如果连接保持稳定(未切换波束), 吞吐量在阻塞后可恢复
• 若触发波束切换(如切至 STARLINK-31884), 吞吐量可能降至原先的一半

(2) 恢复滞后:

当发生因硬遮挡导致的动态切换时, UT 往往回落到性能较差的非主用或备用波束. 吞吐量通常无法立即恢复

直到下一次常规切换(regular handover)发生, UT 重新连接到更优的卫星(如 STARLINK-35320)后, 性能才恢复至正常水平

(3) 常规切换 vs. 动态切换:

• 常规切换:
  • 是计划驱动、预协调的, 利用数据包双播(bicasting)实现平滑过渡
  • 延迟峰值短, 吞吐量损失极小
• 动态切换:
  • 是反应式的, 由突发的 LoS 遮挡触发, 优先考虑连接连续性而非质量, 无双播机制
  • 通常被重新分配到次优备用波束(可能连接到不同或较差的地面站), 导致吞吐量长时间退化
  • 即使视距恢复, UT 也不会立即切回最优波束, 通常要等到下一次计划切换

5.3 FOV Change on Mobile UTs

Fig. 10 demonstrates how UT mobility affects the FOV and visible satellites. In this example, the red ellipse represents the UT’s FOV, which is calculated based on the specifications released by Starlink 5 , with the major and minor axes determined by the UT’s current tilt and azimuth. In Fig. 10a, the UT was connected to STARLINK-2518. As the UT rotates clockwise, STARLINK-2518 moves out of the FOV. This requires the UT to perform a beam switching event to select a new serving satellite. STARLINK-4525 was selected in Fig. 10b, which had a higher elevation and fit within the FOV. When the UT is in motion, uneven roads or vehicle turns are more likely to cause the loss of LoS between UTs and communicating satellites, requiring UTs to conduct frequent beam switching events.

• 旋转导致视距丢失: UT 的物理旋转(如车辆转弯)会导致当前连接的卫星(如 STARLINK-2518)移出视场(FOV)
• 强制切换:
  • 这迫使 UT 执行波束切换以选择视场内的新卫星(如仰角更高的 STARLINK-4525)
  • 移动中的不平坦道路或转弯增加了 LoS 丢失的概率, 导致频繁的波束切换

5.4 Satellite Identification Accuracy

To verify the accuracy of satellite identification results, particularly when the UT is in motion, we use the SGP4 [24] algorithm to calculate the expected satellite ground track for a specific Starlink satellite ID at a given time. We then reconstruct the anticipated obstruction map by incorporating the UT’s location, alignment, and orientation to determine the satellite’s azimuth and elevation relative to the UT, projecting this direction onto the obstruction map. We compare the reconstructed obstruction maps with the observed ones retrieved from the gRPC interface to validate the identification results, as shown in Fig. 11. The satellite trajectories of connected satellites in the reconstructed obstruction map, as shown in Fig. 11b, closely matches the observed obstruction map, with 1–2 pixel differences primarily due to rounding errors when projecting the satellite’s azimuth and elevation onto the obstruction map with a resolution of 123x123 pixels.

To evaluate the accuracy of the proposed mobility-aware satellite identification algorithm with quantitative metrics, we further calculate the angular separation between candidate satellite locations projected from the 2D obstruction map and the identified satellites using our method. In stationary scenarios, the average angular separation is 2.37 ◦ , with a standard deviation of 1.73 ◦ . In our mobility measurement, the average angular separation slightly increases to 5.44 ◦ , with a standard deviation of 3.64 ◦ .

• 验证方法: 使用 SGP4 算法计算特定时间卫星的地面轨迹, 结合 UT 的位置和姿态重构预期遮挡图, 并与实际观测的遮挡图对比. 重构轨迹与观测轨迹非常吻合, 仅有 1-2 像素的差异.
• 量化指标: 计算投影位置与识别卫星之间的角间距(Angular Separation).