IMPLICATIONS OF 5G HANDOVERS ON CARRIERS

5G 切换对运营商的影响

This section takes a network-side look at HOs in 5G. We: (1) present a 5G coverage landscape and highlight a coverage issue in NSA 5G, (2) discuss the impact of 5G HOs on network throughput, and (3) reveal challenges faced by NSA 5G HOs regarding the co-location of eNB and gNB.

6.1 Coverage Landscape in 5G

In cellular networks, the coverage of a cell determines when a HO will be performed. Since we did not have the tower (or cell) locations, we estimate the coverage of a cell by finding the continuous distance a UE travels while being connected to the same cell (i.e., the UE does not connect to a new PCI). Essentially, the estimation calculates the average diameter of a cell. Leveraging our extensive drive test, we first present the coverage landscape in 5G. Then we discuss how the effective coverage of a 5G cell can be affected by NSA.

We find that for NSA 5G, the coverage of a single 5G cell is 1.4 km, 0.73 km, and 0.15 km for low-band, mid-band, and mmWave, respectively. Notably, coverage reduces by 48% from low-band to mid-band. The mmWave coverage is 3.9× and 8.3× lower than midband and low-band coverage, respectively. The signal attenuation is frequency dependent in radio networks. This means higher frequency bands are more attenuated than lower ones, thus reducing cell coverage.

Reduction of effective coverage in NSA 5G. Our study collects data under both NSA and SA deployments of 5G. A key observation we make is that the coverage of the low-band NSA cell effectively reduces as compared to low-band SA. In our dataset, this reduction is found to be between 1.2 to 2×. Fig. 11(a) shows the effective coverage for low-band NSA (red shaded area) and SA (blue shaded area). The dashed lines correspond to the hypothetical (ideal) scenario of low-band NSA coverage, assuming the UE to be in the same coverage as long as the same PCI of 5G gNB is observed. We find that UE can travel over 2000m without a HO when using low-band (n71) SA 5G. Under NSA 5G using the same n71 band, the UE on average will experience a HO within 1000m only, thus effectively reducing the coverage by half. This nullifies NSA low-band’s advantage of extended coverage and infrequent HOs. To explain this, note that for low-band NSA 5G, although the data plane (5G-NR) operates on the low-band, its coupled control plane (NSA-4C) still uses the mid-band. As a result, a NSA-4C HO in NSA will always trigger 5G HO (SCGR), therefore reducing the effective coverage of a 5G cell. A similar case is found for mid-band (Fig. 11(b)) where NSA 5G’s effective coverage also slightly reduces when compared to the ideal scenario where NSA-4C’s impact is not considered. The above findings suggest that HOs in NSA 5G not only incur wild QoE fluctuations (§4) and long HO durations (§5.2), but also have implications on cell coverage.

6.2 Impact of 5G HOs on Bandwidth

Horizontal HOs are supposed to boost the network performance by associating a UE to a new tower with better signal strength.

However, we find that oftentimes this is not the case in NSA 5G. We next describe our findings and explain the root causes.

To get insights on the impact of HOs over the network bandwidth, while walking a 35+ minute loop, we perform bulk download using iPerf3 (§2) in areas with OpX’s 5G mmWave coverage. For each type of HO, we then measure the throughput in three phases: (i) Pre-HO (HO 𝑝 𝑟𝑒 ), which captures the throughput just 1-second before the HO procedure starts, (ii) During-HO (HO 𝑒𝑥𝑒𝑐 ), which captures the throughput during the execution of HO procedures, and (iii) Post-HO (HO 𝑝 𝑜𝑠𝑡 ), which denotes the perceived throughput 1-second after the HO procedures are complete. Fig. 12 compares the throughput in the three phases for inter-gNB (SCGC) handovers. We observe that the average post-HO throughput reduces by 14% compared to the average pre-HO throughput. This is counter-intuitive because inter-gNB HOs are supposed to improve the received signal strength of UE and hence its throughput. While prior literature identifies one reason to be suboptimal signal strength threshold settings [65], we identify a new reason in the 5G context, as detailed next.

As explained in §2, NSA 5G does not support direct HOs between gNBs. Instead, an SCGC HO (5𝐺 → 5𝐺) comprises of 5𝐺 → 4𝐺 and 4𝐺 → 5𝐺 HOs, and each of the latter two HOs is performed independently without accounting for the overall (5G→5G) signal strength improvement. As a result, an SCGC HO oftentimes shows no overall signal strength improvement. To mitigate this issue, NSA carriers may need to improve their inter-gNB HO logic by considering the overall HO sequence.

Besides SCGC HOs, using the same experimental methodology descried above, we find that other types of HOs also exhibit different throughput change patterns for the above three phases. The details can be found in Appendix A.3. Such patterns can be leveraged as features for HO prediction, as to be detailed in §7.4.

6.3 Impact of eNB and gNB Co-location

In NSA, the UE connects to both eNB and gNB, which may not be co-located at the same cell tower. To identify such co-location, we find that when the NSA-4C eNB and 5G-NR gNB are co-located at the same physical tower, their 4G and 5G PCIs are the same; on the other hand, their PCIs are typically different if they are not co-located 3 . Using this heuristic, we find that the NSA-4C eNB and 5G-NR gNB are co-located only in 5%–36% of the NSA low-band samples in our dataset across the three carriers.

We find that the non-co-located NSA-4C eNB and 5G-NR gNB incur a major side effect. Specifically, we find that for NSA HOs where the (origin or destination) gNB and eNB are co-located, their duration is significantly shorter than HOs whose gNB and eNB are not co-located. This can be clearly seen in Fig. 13 which shows that an NSA HO with same NSA-4C and 5G-NR PCI saves 13ms on average over a NSA HO with different PCIs. The additional latency is attributed to the cross-tower communication between NSA-4C and 5G-NR towers [60]. This finding suggests that NSA carriers can mitigate the impact of 5G HOs by facilitating NSA-4C and 5G-NR towers’ co-location, or at least take into account for 4G/5G antenna locations when making HO decisions.

TL;DR: 5G 切换对运营商的影响

5G 覆盖情况

在蜂窝网络中，单元格的覆盖范围决定了何时进行切换（HO）。研究发现，NSA 5G的单元格覆盖范围因频段不同而有显著差异：低频段为1.4公里，中频段为0.73公里，毫米波段仅为0.15公里。

毫米波段的覆盖范围比中频段和低频段分别减少了3.9倍和8.3倍。此外，NSA 5G的低频段有效覆盖范围与SA 5G相比减少了1.2到2倍，这是因为NSA 5G的数据平面使用低频段，但控制平面仍使用中频段，导致切换频率增加。

5G 切换对带宽的影响

理论上，水平切换应该通过将用户设备（UE）连接到信号更强的基站来提升网络性能。然而，在NSA 5G中，切换后平均带宽实际上减少了14%。这是因为NSA 5G不支持直接的gNB间切换，而是通过4G和5G之间的独立切换，这可能无法总体提升信号强度。

eNB和gNB的共址对切换的影响

在NSA部署中，eNB（4G基站）和gNB（5G基站）可能不在同一物理塔上。研究发现，当eNB和gNB共址时，切换延迟显著减少，平均节省13毫秒。共址可以减少跨塔通信的延迟，从而改善切换性能。因此，运营商可以通过促进eNB和gNB的共址或考虑4G/5G天线位置来优化切换策略。