Cellular Networks in Non-Contiguous US Regions

In this section, we examine the cellular network deployments in AK and Maui, HI, and compare them against the deployments in the mainland US, focusing on two aspects: (1) network coverage, and (2) end-to-end performance.

3.1 Coverage

Figs. 3a, 3b, 3c show the cellular technology coverage of the three major US operators in AK, HI, and the mainland US route (ML), respectively, defined as the fraction of miles covered by a specific technology over the total distance covered during throughput and RTT measurements.

Fig. 3a shows that 5G coverage in AK is disappointingly low for both operators. Verizon offers no 5G service, while AT&T’s 5G coverage is 30%, but it is limited to 5G-low, which provides extended range with reduced data rates. Throughout our driving trip, we encountered no high-speed 5G (5G-mid or mmWave). Additionally, a non-negligible fraction of the coverage bars – 9% for AT&T and 23% for Verizon – is marked by "No Service", showing periods of complete service failures. In contrast, Fig. 3c shows that 5G coverage for all three operators is much higher along the mainland US route, even though that route also includes vast rural areas, similar to AK and HI. T-Mobile, which offers service only via roaming partners in AK, provides 92% 5G coverage along the mainland US route mostly in the mid bands (78%). The other two operators offer lower 5G coverage compared to T-Mobile (40%), but still significantly higher compared to AK. Additionally, both operators offer 5G midband service along the mainland route, in contrast to AK. Verizon’s 5G service is almost exclusively in the mid bands while AT&T’s 5G midband service is limited to 13%. Finally, all three operators offer continuous cellular coverage along the mainland driving route. The fraction of "No Service" periods in Fig. 3c is only 0.07% for AT&T and 0% for Verizon and T-Mobile.

Fig. 3b shows that the overall 5G coverage for all three operators is higher in HI compared to AK, but still worse than in the contiguous US, with the exception of AT&T. Surprisingly, AT&T’s 5G coverage is higher in HI than on the mainland route (55% vs. 40%), but, similar to AK, it is again limited to 5G-low. In contrast, in the mainland US, the operator offers limited 5G-mid coverage (13%). Verizon’s 5G coverage is only 16% in HI and also includes only 5G-low. T-Mobile has the largest 5G coverage in HI (63%) among the three operators, and it is the only operator offering 5G midband service (56% coverage). However, its coverage is still lower than on the mainland route (92% 5G coverage, with 78% contributed by 5G-mid). Additionally, all 3 operators face complete service failures, similar to AK, and the fraction of "No Service" is higher than in AK – 14%/18%/16% for AT&T/Verizon/T-Mobile.

Our coverage results, particularly the absence of 5G midband coverage in AK and the limited 5G midband coverage in HI, can be explained by the limited spectrum availability in these regions. In the contiguous US, AT&T and Verizon primarily rely on the C-Band (3.7–3.98 GHz) spectrum for their 5G midband deployments, while T-Mobile relies mainly on the 2.5 GHz band inherited from Sprint. However, these bands are not equally available in AK and HI. Specifically, the C-Band spectrum was not auctioned in AK and HI due to the continued operation of incumbent satellite services in the same frequency range, which the FCC deemed essential for regional connectivity [15]. Consequently, AT&T and Verizon provide only 5G-low or no 5G service in AK and HI, as we saw in Figs. 3a, 3b. Additionally, a substantial portion of the 2.5 GHz band in AK has been licensed to the Alaska Tribal Network through the FCC’s Rural Tribal Priority Window [16], resulting in limited commercial access to that spectrum; as a result T-Mobile relies exclusively on partnership networks in AK, as we mentioned in §2. In contrast, T-Mobile retains 2.5 GHz licenses in HI (subject to FCC divestiture conditions [17]), where it offers a substantial 5G-mid coverage (Fig 3b), although much lower compared to the mainland US.

We further break down the overall coverage based on the area type in Fig. 4. In AK, AT&T offers almost 90% 5G coverage in urban areas (Anchorage and Fairbanks), while LTE is the dominant technology in the rural areas, where 5G coverage is only 25% (Fig. 4a). Similarly, in HI, the urban areas exhibit higher 5G coverage, with AT&T and T-Mobile showing substantial (≥ 90%) 5G-low and 5G-mid coverage, respectively (Fig. 4b). In contrast, in rural areas, we observe again a sharp decline in 5G coverage for all 3 operators – from 94/27/93% down to 44/13/54% for AT&T/Verizon/T-Mobile, although T-Mobile still offers satisfactory 5G-mid coverage (45%). Additionally, in both regions, the "No service" periods observed in Fig. 3, are exclusively experienced in rural areas, further emphasizing the connectivity gap.

Fig. 4c shows a very different picture for the mainland route. In urban centers, all three operators offer high 5G-mid coverage, ranging from 52% for AT&T up to 98% for T-Mobile, with Verizon also offering a non-negligible 5G-mmWave coverage (9%). In rural areas, Verizon and T-Mobile offer much better 5G coverage than in the non-contiguous US regions, with a substantial amount of 5G-mid service (31% and 74%, respectively). Interestingly, AT&T’s 5G coverage in rural areas of the mainland route is worse than in rural Maui; only 32% (vs. 44% in rural HI) with negligible 5G-mid coverage. However, the operator offers almost full cellular coverage (0.07% "No Service") in rural areas of the mainland route, in contrast to rural HI.

UE-BS distance. To further understand the coverage results, we show in Figs. 5a-5c boxplots of the distances between the UE and the LTE/5G base stations (BSs) the UE was connected to during our drives in AK, HI, and the mainland route, respectively, broken down by area type. These distances provide an idea of the range of cellular BSs. We calculate the UE-BS distance using the Timing Advance (TA) field, 1 extracted from XCAL. We observe that distances to LTE BSs are typically longer than to 5G BSs and distances in rural areas are typically much longer than in urban areas (suggesting sparser deployments), although there are some exceptions. We also observe very large outliers regardless of region, area type, or technology; both LTE and 5G(-low) BSs can have a range longer than 30 km.

In urban areas, the distances are very short for both technologies in all three regions, with median and 75-th percentile values typically less than 2 km. The only exception is T-Mobile, with a 75-th percentile of 5.6 km for LTE in the mainland US and 21 km in HI. Recall from Figs. 4b, 4c that T-Mobile offers very limited LTE coverage in urban areas of HI and the mainland US, suggesting that the operator maintains very few LTE BSs in cities in those regions and BSs outside city limits are often used to fill 5G coverage holes inside cities. Another notable observation is that AT&T’s 5G distances are longer than the LTE distances in AK urban areas in spite of the near-full 5G coverage (Fig. 4a), suggesting a sparse 5G-low deployment.

In rural areas, the distances in AK are similar to the mainland for both technologies, with median values in the range of 4-5/2-3 km for LTE/5G and 75-th percentiles in the range of 7-10/5-7 km. In contrast, in rural HI, the LTE distances are much longer compared to the mainland US, with 50-th/75-th percentiles in the range of 6-9/14-21 km vs. 4-5/7-10 km. On the other hand, the 5G distances have lower median values compared to the mainland but much higher 75-th percentiles, suggesting heterogeneous deployments (with respect to the distance to the highways and/or Tx power). The only exception here is Verizon’s 5G-low service in rural HI with distances similar to those in urban HI (75-th percentile of 2 km), which explains the very limited 5G coverage offered by Verizon in these areas (Fig. 4b).

Signal strength. Figs. 5d-5f show boxplots of the Reference Signal Received Power (RSRP) in AK, HI, and the mainland route, broken down by area type and technology. In the case of LTE, the RSRP is typically higher in urban areas compared to rural areas; the median values range from -110 to -95 dBm in rural areas vs. -105 to -81 dBm in urban areas. The only exception is Verizon in HI, where the LTE RSRP has similar 50-th/75-th percentiles in both area types, despite the big discrepancy in UE-BS distances (Fig. 5b), suggesting that the operator uses power control to provide good LTE coverage in many rural areas. On the other hand, the differences in 5G RSRP between the two area types are much smaller. One notable discrepancy here is AT&T in HI, where the 5G RSRP is much higher in urban than in rural areas, which is consistent with our observations about the UE-BS distances in Fig. 5b. A second exception is T-Mobile in the mainland US. However, note that T-Mobile’s 5G RSRP in both urban and rural areas of the mainland US is the highest among the three operators.

When we compare the non-contiguous regions vs. the mainland US, we observe surprisingly that AT&T in urban AK areas offers better LTE signal than all three operators in urban areas of the mainland US. Similarly, Verizon’s LTE RSRP in urban AK is similar to the operator’s LTE RSRP in urban areas of the mainland US, where Verizon offers the highest LTE RSRP among the three operators. On the other hand, the LTE RSRP for all three operators in rural AK/HI is much lower than the LTE RSRP in rural areas of the mainland US. In the case of 5G, the RSRP for all three operators in the non-contiguous US regions is much lower than in the mainland US, regardless of the area type. The only exception here is AT&T, which surprisingly offers higher 5G RSRP in urban HI than in the urban centers of the mainland US.

Remarks. Overall, we observe a persistent digital divide between the mainland and non-contiguous US regions in terms of 5G coverage, which is further exacerbated in rural areas for two of the three major cellular operators. The disparity is evident not only in the overall 5G coverage but also in the lack of high-speed (5G-mid and mmWave) 5G deployments (with the exception of T-Mobile in HI). While urban areas benefit from better 5G coverage (although still worse than in the mainland US), rural areas continue to face significant challenges, with a heavier reliance on LTE along with sparse deployments, poor signal strength, and frequent "No Service" events.

We also observe different deployment strategies for the three operators in the non-contiguous vs. contiguous US. T-Mobile offers the best 5G coverage in the mainland US, but relies exclusively on roaming partners in AK. AT&T, on the other hand, offers the most extensive 5G coverage in AK and higher coverage than Verizon in HI (although limited to 5G-low), but its 5G coverage in the rural regions of the mainland route is lower than in rural HI.

5G 覆盖率低:

• 阿拉斯加 (AK): 5G 覆盖率极低.

  • Verizon 没有 5G 服务
  • AT&T 仅有 30% 的覆盖率且限于低频段 (5G-low), 无高速 5G(5G-mid 或 mmWave)

• 夏威夷 (HI): 5G 覆盖率高于 AK 但仍不及本土.

  • T-Mobile 拥有最广泛的 5G 覆盖 (63%) 且是唯一提供中频段 5G (5G-mid, 56% 覆盖率) 的运营商

• 无服务区域: 农村地区频繁出现完全无服务的情况, AK 的"无服务"比例高达 23% (Verizon), HI 高达 18% (Verizon)

  • 相比之下, 本土路线几乎实现了全覆盖(Verizon 和 T-Mobile 为 0%)

• 频谱受限:

  • 造成非本土地区 5G 中频段缺失的主要原因是 C-Band (3.7-3.98 GHz) 频谱因现有卫星服务干扰未在 AK 和 HI 拍卖
  • T-Mobile 在 AK 仅依赖漫游伙伴, 因为其 2.5 GHz 频段主要被许可给了阿拉斯加部落网络(Alaska Tribal Network)

• 城乡差异:

  • 城市地区的 5G 覆盖率通常较高, 但农村地区急剧下降, 且主要依赖 LTE
  • "无服务"情况几乎全部发生在农村地区

• 基站距离与信号:

  • 农村地区的基站距离通常远大于城市, 且信号强度 (RSRP) 更弱
  • 意外发现: AT&T 在 AK 城市地区的 LTE 信号强度甚至优于美国本土城市

3.2 Network Performance

DL Performance. Fig. 6a shows the TCP throughput with each operator in AK, HI, and the mainland route (ML). Our first observation is that there is a significantly high fraction of zerothroughput values for all three operators in the non-contiguous regions. In AK, the fraction of zero-throughput samples for AT&T/Verizon is about 24%/32%. The situation worsens in HI, where the fraction of zero-throughput samples for AT&T/T-Mobile/Verizon is 40%/47%/40%. Notably, the fraction of zero-throughput samples in Fig. 6a is much higher than the percentage of "No Service" coverage in Fig. 3, suggesting that, in certain cases, even if the UE was connected to cellular network, there was practically no data exchange with the network.

In AK, AT&T and Verizon exhibit comparable performance with 50-th/75-th percentiles of 15/67 Mbps vs. 11/49 Mbps, although AT&T’s 5G-low coverage yields a higher peak throughput (355 Mbps vs. 193 Mbps). In HI, T-Mobile outperforms the other two operators by a substantial margin, thanks to its 5G-mid service. The 50-th/75-th/100-th percentile with T-Mobile is 6/136/937 Mbps vs. 5/49/306 Mbps for Verizon and 3/26/272 Mbps for AT&T. Despite all operators demonstrating peak DL speeds of hundreds of Mbps, the overall throughput in both regions remains significantly lower compared to the mainland US. The 50-th/75-th/100-th percentile on the mainland route is 184/399/1469 Mbps with T-Mobile, 77/308/3154 Mbps with Verizon, and 57/133/2409 Mbps with AT&T. Additionally, the fraction of zero-throughput samples on the mainland route is 10-19% for the three operators, much lower than in the non-contiguous US.

In Fig. 8, we break down the DL throughput by cellular technology and area type. We make two observations: (1) The majority of the zero-throughput samples in AK/HI observed in Fig. 6a originate from LTE in rural areas, highlighting the challenges of cellular deployments in such areas. Figs. 5d, 5e show that the LTE RSRP is much lower in rural areas in both regions indicating very sparse deployments. (2) While urban areas exhibit in general higher performance than rural areas, as expected, we observe a notable exception for AT&T in AK. Its 5G-low service performs much better in rural areas, achieving a 50-th/75-th percentile of 75/129 Mbps compared to only 26/65 Mbps in urban areas, despite the similar RSRP in both areas (Fig. 5d). Fig. 7 shows that the number of 5G-low resource blocks for AT&T in rural areas is much higher than in urban areas. This suggests an insufficient deployment of 5G base stations in AK, which cannot meet the higher network demands from denser population in Anchorage and Fairbanks, leading to degraded performance due to contention for network resources among users.

UL Performance. Fig. 6b shows the UL throughput in AK, HI, and the mainland route. Similar to DL, we notice a significant portion of zero-throughput samples in both non-contiguous US regions (39-46% in AK, 52-53% in HI vs. only 14-22% in the mainland US). In AK, the UL throughput trends are similar to the DL ones, with both operators having roughly similar values – 3/12 Mbps vs. 2/11 Mbps at the 50-th/75-th percentile – and AT&T exhibiting higher peak values. In HI, unlike the DL case where T-Mobile outperformed the other two operators by a substantial margin, T-Mobile and Verizon achieve similar UL performance, while AT&T’s UL throughput is slightly lower. Despite these variations, UL TCP throughput across all operators remains significantly lower than in the mainland US. The median UL throughput is 0 Mbps in HI for all three operators and the gap with the mainland UL throughput at the 75-th percentile is 13-46 Mbps.

A breakdown of the UL TCP throughput based on area type and technology (Fig. 9) again shows that most zero-throughput samples in AK/HI are contributed by LTE in the rural areas. Additionally, for a given operator, regardless of the region, the UL LTE throughput is always lower compared to the 5G counterpart. Interestingly, T-Mobile’s 5G midband service does not offer significant advantages in the UL case, due to limited use of Carrier Aggregation [30, 69]. In particular, in rural HI, Verizon’s 5G-low service offer much better UL performance than T-Mobile’s 5G-mid service. The same observations hold true in the mainland with two notable exceptions for the rural areas: Verizon’s 5G-low performs very poorly and AT&T’s 5G-mid performs worse than 5G-low. However, 5G-low for Verizon and 5G-mid for AT&T contribute a negligible fraction to the overall cellular coverage of the corresponding operator in rural areas of the mainland US (Fig. 4c).

Network Latency. Fig. 6c shows the RTT of cellular networks in AK, HI, and the mainland route. We observe that the RTT values for Verizon in AK and all three operators in HI are much higher than in the mainland US (by 14 ms in AK and 26-42 ms in HI in the median case). The higher latency in non-contiguous US regions, especially for HI, can largely be attributed to the longer UE-server distance compared to the mainland US (recall that we used the same server in OR for all our measurements). Additionally, the Wavelength edge servers in three mainland cities further reduce the RTT for Verizon in the mainland US. Our results show a need for deployment of edge servers in non-contiguous US regions to reduce the end-to-end latency and improve the QoE of different latency-critical apps. Interestingly, AT&T exhibits very high RTTs along the mainland US route. AT&T’s RTT in the mainland is similar to Verizon’s RTT in HI in the median case (94 ms vs. 92 ms), and higher than both Verizon’s and T-Mobile’s RTT in HI at the 75-th percentile (129 ms vs. 100/123 ms), suggesting that the operator de-prioritizes ICMP traffic in the mainland US.

In AK, the RTT trends align with the throughput trends, with Verizon and AT&T exhibiting similar RTT values, despite Verizon offering only LTE service (Fig. 3a). Similarly, in HI, Verizon offers the lowest RTT despite its limited 5G coverage (Fig. 3b) compared to the other two operators. In particular, T-Mobile’s RTT is much higher than Verizon’s, despite the extensive 5G midband coverage, suggesting that T-Mobile de-prioritizes non-heavy ICMP ping traffic in HI.

We further break down the latency by technology and area type in Fig. 10. We make two observations. First, for a given operator, its 5G service does not always offer lower latency compared to its LTE service in the non-contiguous US (5G-low for AT&T in urban AK and T-Mobile in rural HI). In contrast, 5G offers always better RTT than LTE in the mainland US. In particular, Fig 10f shows that the LTE RTT with AT&T and T-Mobile is very high in mainland rural US areas. Recall that LTE contributes substantially to AT&T’s overall coverage in mainland rural areas but only a negligible fraction to T-Mobile’s overall coverage (Fig. 4). This explains the overall high RTT for AT&T in the mainland US in Fig. 6c. Second, the latency, in particular the peak values (upper quartile) typically increases substantially in rural areas compared to urban areas. The only exception is Verizon in HI, where its LTE service in rural areas offers lower latency than its 5G-low service in urban areas.

Remarks. In summary, we observe significant performance disparities between non-contiguous US regions and the mainland US in terms of both throughput and latency, which reinforce the digital divide we highlighted in §3.1. Frequent zero-throughput occurrences, particularly in rural areas, even when the UE is connected to the network, and sparse 5G deployments resulting in high contention for network resources and lower performance compared to LTE, highlight ongoing challenges. Network latency is a more complex issue. Greater UE-server distances increase the latency in non-contiguous US compared to the mainland US; however, operator policies related to de-prioritization of ICMP traffic for certain technologies in different regions further complicate the matter. Overall, these findings underscore the need for improved infrastructure and localized edge servers to bridge the performance gap with the mainland US.

下行性能 (DL Performance):

• "零吞吐量"高: 非本土地区即使连接上网络, 仍有极高比例的零吞吐量样本(例如 HI 的 T-Mobile 高达 47%), 远高于"无服务"比例
  • 这主要源于农村地区的 LTE 网络
• 吞吐量低: 整体吞吐量远低于本土
  • HI 的 T-Mobile 虽有中频段优势, 表现最佳, 但仍不及本土水平
• 拥塞问题: AT&T 在 AK 农村地区的 5G-low 性能反而优于城市, 原因是城市人口密集导致网络资源争用(资源块竞争)

上行性能 (UL Performance):

• 普遍较差: 所有运营商在上行方向均表现不佳, 零吞吐量样本比例极高(HI 高达 53%)
• 技术对比: LTE 的上行性能总是低于 5G.
  • 但在 HI 农村地区, Verizon 的 5G-low 上行性能优于 T-Mobile 的 5G-mid.

网络延迟 (Network Latency):

• 延迟更高: AK 和 HI 的 RTT 值显著高于本土(HI 中位数高出 26-42 ms), 主要原因是用户到服务器的物理距离更远
• ICMP 降级: T-Mobile 在 HI 的高 RTT 暗示其对非大流量 ICMP Ping 流量进行了降级处理
• 边缘计算缺失: 结果强调了在非本土地区部署边缘服务器以降低端到端延迟的必要性

5G-low vs. 5G-mid

1. 5G-low (低频段 5G)
   • Def: 使用低频段频谱部署的 5G 网络, 通常在 600 MHz 到 1 GHz 范围内
   • Traits:
     • 较广的覆盖范围: 类似于 LTE 的覆盖能力
     • 较低的数据速率: 无法提供 5G 承诺的超高网速
   • Where: 适合在农村地区、郊区等广域覆盖需求的场景中使用
   • 文中: 在阿拉斯加和夏威夷, 由于频谱限制, AT&T 和 Verizon 提供的 5G 服务主要是 5G-low
2. 5G-mid (中频段 5G)
   • Def: 使用中频段频谱部署的 5G 网络, 通常在 1 GHz 到 3 GHz 范围内
   • PS: 在文中被归类为“高速 5G”的一种
   • Traits: 显著更高的吞吐量和峰值速率
   • 频段:
     • AT&T 和 Verizon 主要依赖 C-Band (3.7–3.98 GHz) 频谱
     • T-Mobile 主要使用从 Sprint 继承的 2.5 GHz 频段
     • 文中:
       • T-Mobile 在夏威夷之所以性能大幅领先其他运营商，正是因为它提供了 5G-mid 服务
       • 由于 C-Band 频谱在阿拉斯加和夏威夷未进行拍卖（因卫星服务干扰）+ 部分 2.5 GHz 频谱被分配给部落网络，导致这些地区极度缺乏 5G-mid 部署