Introduction

Abstract

5G cellular networks and Low-Earth-Orbit (LEO) satellite networks, such as Starlink, promise enhanced performance and coverage capabilities. While a large number of research works have evaluated these technologies in the mainland US, their performance in non-contiguous US regions remains under-explored, despite their unique challenges and significant visitor demands. Through extensive drive tests covering over 4200+ km across Alaska, Maui (Hawaii), and the mainland US (as a baseline), while simultaneously measuring the performance of the three major US mobile operators and Starlink, this work presents the first detailed evaluation of cellular and Starlink network coverage and performance in the non-contiguous US regions. Our study shows a persistent digital divide between mainland and non-contiguous US for cellular networks in terms of coverage and performance, highlighting the challenges of cellular deployments in non-contiguous US regions. Starlink provides substantially higher performance than cellular networks most of the time, but area-specific challenges, including unique terrains in Hawaii and sparse satellite deployment in Alaska, significantly degrade performance compared to the mainland US. Additionally, we explore the spatiotemporal diversity between cellular and Starlink performance and study the potential of multipath transport to bridge the connectivity gap in non-contiguous US regions.

5G 蜂窝网络与低轨卫星网络(如 Starlink)有望提供增强的性能与覆盖能力. 尽管已有大量研究评估了这些技术在美国本土的应用, 但在美国非本土地区, 尽管面临独特的挑战且存在巨大的访客需求, 其性能表现仍鲜有研究涉足. 本文通过在阿拉斯加, 夏威夷茂宜岛以及美国本土(作为基准)进行的超过 4200 公里的广泛路测, 同时测量了美国三大移动运营商与 Starlink 的性能, 首次对美国非本土地区的蜂窝网络与 Starlink 网络的覆盖范围及性能进行了详尽的评估. 研究结果表明, 在蜂窝网络的覆盖范围与性能方面, 美国本土与非本土地区之间存在持续的"数字鸿沟", 这凸显了在非本土地区部署蜂窝网络所面临的挑战. 尽管 Starlink 在大多数情况下提供的性能显著优于蜂窝网络, 但特定区域的挑战(包括夏威夷独特的地形以及阿拉斯加稀疏的卫星部署)导致其在这些地区的性能相较于美国本土有显著下降. 此外, 本文还探讨了蜂窝网络与 Starlink 性能之间的时空多样性, 并研究了利用多路径传输技术弥合美国非本土地区连接差距的潜力.

Introduction

The wireless landscape is currently undergoing a significant transformation, with 5G networks playing a pivotal role in this process. Via a combination of innovations such as higher-order modulation schemes, (massive) MIMO, wider channels, flexible slot scheduling, network slicing, etc., 5G promises ultra-high bandwidth and ultra-low latency, far surpassing the performance of its precursor, 4G LTE. Such high data rates and low latency promise to unlock bandwidth-intensive, latency-critical applications such as Augment Reality, Mixed Reality, Connected Autonomous Vehicles, and industrial automation.

Despite these ambitious promises, the real-world deployment of 5G has revealed several limitations. While existing studies and articles [2, 3, 20, 29, 36, 56] demonstrate that 5G can achieve peak data rates of 3+ Gbps and sub-1 ms PHY latency under ideal conditions, other works [6, 10, 28, 30, 56] have also highlighted that its coverage and performance often remain fragmented and inconsistent, especially when traveling across diverse regions in the US. Notably, prior evaluations have largely focused on the contiguous United States, leaving a gap in our understanding of how 5G performs in non-contiguous regions such as Alaska (AK) and Hawaii (HI). These regions face distinct challenges, including diverse terrains, limited access to fiber backhaul in remote rural environments, which necessitating the use of alternatives such as microwave or satellite links [26, 66], a heavy reliance on long-distance terrestrial or submarine cables, and limited spectrum availability [15], all of which can significantly impact network performance and availability. As such, the first goal of this work is to assess 5G’s reach and effectiveness in these regions, which is crucial to determine whether a digital divide exists between the mainland and the non-contiguous US. Such a divide would carry important implications, not only for local residents who rely on cellular connectivity for daily communication and services, but also for the millions of tourists traveling through these regions, where network availability and performance could significantly affect navigation, safety, and overall experience.

Assuming such a digital divide between the mainland and non-contiguous US indeed exists, our second research goal is to understand whether and how the emerging Low Earth Orbit (LEO) satellite networks, such as Starlink, can reliably bridge connectivity gaps left by terrestrial cellular networks and mitigate the digital divide. LEO satellite networks have gained traction in recent years as a promising alternative Internet access technology in remote and inaccessible regions where traditional wired or wireless networks fall short. Recent studies [27, 28, 34, 50] suggest that LEO satellite networks can offer competitive performance compared to cellular networks in the mainland US, raising the possibility that they could extend meaningful connectivity to harder-to-reach regions such as AK and HI.

However, whether LEO satellite networks can reliably supplement cellular performance in the non-contiguous US remains an open question. While prior work suggests promising results in the mainland US, the unique geographic and infrastructure challenges of AK and HI complicate the picture. These challenges include: (1) The long distance of Hawaii and Alaska from mainland may introduce higher latency, as many ground stations are concentrated within the mainland US. (2) The diverse terrains and challenging topography of these regions can significantly impact communication. For example, obstructions caused by dense rain forests in HI or rugged mountains in AK can severely block Line of Sight (LoS) communication leading to performance degradation. (3) Varying satellite orbits and inclinations in different regions may lead to large differences in coverage and performance. As such, answering our second question requires the first in-depth study of LEO satellite network coverage and performance in these non-contiguous regions.

To address both of our research questions, we conduct, to our knowledge, the first measurement study of cellular and LEO satellite networks in the non-contiguous US. Via three extensive drive trips covering over 4200+ km in Alaska, Maui, HI, and the mainland US (Los Angeles to Omaha), as a baseline, we simultaneously measure the coverage and performance of the three major US cellular operators (AT&T, Verizon, and T-Mobile) and Starlink’s LEO satellite network. Our measurements span diverse area types and comprehensively evaluate key network metrics such as downlink/uplink (DL/UL) throughput, network latency, and outage rates. Additionally, we explore the spatiotemporal diversity between cellular and Starlink performance and study the potential of multipath transport protocols (e.g., Multipath TCP) to bridge the connectivity gap in the non-contiguous US. The key findings of our work are summarized as follows:

• Our results show significantly lower 5G coverage in the non-contiguous US compared to the mainland US, which is further exacerbated in rural areas with frequent periods of no connectivity. The disparity is evident not only in terms of overall 5G coverage, but also in the lack of high-speed 5G deployments (with the exception of T-Mobile in HI). As a result, cellular networks consistently underperform in the non-contiguous US in comparison to mainland US, in terms of both throughput and latency. Zero-throughput occurrences are frequent even in areas with cellular connectivity, highlighting ongoing coverage challenges. Additionally, greater UE-server distances and limited network resources due to sparse deployments pose as a bottleneck in achieving higher performance in these regions.

• Similar to cellular networks, Starlink’s performance in the mainland US significantly surpasses its performance in Alaska and Hawaii due to area-specific challenges. Sparse satellite coverage in Alaska leads to coverage gaps and increased latency, whereas unique terrains in rural Hawaii lead to high outage rates despite dense satellite coverage. Additionally, the longer UE-server distances in the non-contiguous US compared to the mainland US significantly inflate the RTT in those regions.

• Starlink outperforms cellular networks in most scenarios across the non-contiguous US, providing superior DL throughput and enhanced reliability. However, Starlink provides much lower DL peak data rates compared to 5G-mid, and it often exhibits lower UL TCP throughput and higher RTT (especially in Alaska) compared to cellular networks, highlighting inherent challenges such as packet loss and congestion control inefficiencies in LEO satellite networks. Our study further reveals substantial diversity in the performance of cellular networks and Starlink at a given location and time, suggesting that multipath transport protocols have great potential to improve overall performance and reliability.

• To assess this potential, we evaluate the performance of Multipath TCP (MPTCP) via tracedriven emulation. Our results confirm that multi-link access techniques can significantly improve performance and reduce outage rates, especially in rural areas. While combining Starlink with a cellular network can simultaneously boost throughput and reliability in the DL direction, Starlink’s lower UL TCP throughput compared to cellular networks introduces an interesting tradeoff between performance and reliability in the UL direction, emphasizing the need for careful network pairing strategies to maximize benefits.

Contributions. In summary, this work makes the following contributions: (1) We collect a large dataset covering both cellular and Starlink network performance in the non-contiguous and contiguous US, obtained via extensive drive-test measurement campaigns spanning over 4200+ km.

(2) Leveraging this unique dataset, we conduct the first large-scale evaluation of cellular and LEO satellite networks in the non-contiguous US, examining their coverage, throughput, latency, and reliability, in comparison to the mainland US. (3) We evaluate the potential of MPTCP to improve performance and reliability in the non-contiguous US, highlighting opportunities for a synergistic integration of cellular and LEO satellite networks to address connectivity gaps. (4) Our dataset and scripts are publicly available [1].

The rest of the paper is organized as follows: §2 describes the data collection methodology; §3 and §4 discuss the coverage and performance of cellular networks and Starlink, respectively, in non-contiguous US regions; §5 compares the performance of the two technologies in the noncontiguous US regions, and explores their spatiotemporal diversity; §6 evaluates the benefits of combining the two technologies via multipath transport (MPTCP); §7 discusses the related work and §8 concludes the paper.

研究背景与缺口:

• 尽管 5G 承诺提供超高带宽和超低延迟, 但现实部署中其覆盖和性能往往是不一致的, 且现有研究主要集中在美国本土, 缺乏对阿拉斯加(AK)和夏威夷(HI)等非本土地区的了解.
• 这些非本土地区面临独特挑战, 如多样的地形, 偏远地区光纤回程受限, 依赖长距离电缆以及频谱可用性有限.

研究目标:

1. 评估 5G 在这些地区的覆盖范围和有效性, 以确定美国本土与非本土地区之间 是否存在"数字鸿沟"
2. 探究低地球轨道(LEO)卫星网络(如 Starlink)是否能可靠地填补蜂窝网络的"数字鸿沟"
   • 尽管其面临地面站距离远, 地形遮挡(如雨林, 山脉)和卫星轨道差异等挑战

研究方法:

• 进行了首个针对美国非本土地区蜂窝和 LEO 卫星网络的测量研究
• 通过在阿拉斯加, 夏威夷茂宜岛以及作为基准的美国本土(洛杉矶至奥马哈)进行的超过 4200 公里的路测, 同时测量了三大移动运营商和 Starlink 的覆盖与性能(吞吐量, 延迟, 中断率)

主要发现:

• 蜂窝网络: 本土性能 > 偏远地性能

  • 美国非本土地区的 5G 覆盖率显著低于本土, 且性能(吞吐量和延迟)表现不佳, 特别是在农村地区经常出现无连接的情况, 证实了数字鸿沟的存在

• Starlink 网络: 本土性能 > 偏远地性能

  • 由于阿拉斯加稀疏的卫星覆盖和夏威夷独特的地形挑战, Starlink 在这些地区的性能也显著低于美国本土
  • 且由于用户与服务器距离较远, 导致往返时延(RTT)较高

• 对比分析: 虽然 Starlink 在非本土地区的大多数场景中提供了优于蜂窝网络的下行吞吐量和可靠性!

  • 但其下行峰值速率较低, 且上行 TCP 性能和延迟往往不及蜂窝网络

对比维度	Starlink	Cellular Networks
总体场景表现	在美国非本土地区的大多数场景中表现优于蜂窝网络	在美国非本土地区的大多数场景中表现不及 Starlink
下行 (DL) 吞吐量	提供更优的下行吞吐量	总体下行吞吐量表现较弱
可靠性	具有增强的可靠性	可靠性相对较低
下行 (DL) 峰值速率	远低于 5G 中频段 (5G-mid)	5G 中频段 (5G-mid) 可提供更高的峰值数据速率
上行 (UL) TCP 性能	通常表现较低	通常表现优于 Starlink
往返时延 (RTT)	较高（特别是在阿拉斯加地区）	相对较低
面临的挑战	存在丢包和拥塞控制效率低下等固有挑战	相对较少面临此类 LEO 特有的挑战

解决方案与贡献:

• 研究表明多路径传输协议(如 MPTCP)具有弥合连接差距的巨大潜力, 通过结合两种技术可以显著提高性能并降低中断率, 尤其是在农村地区
• 本文的主要贡献包括收集了大规模路测数据集, 进行了首次大规模对比评估, 评估了 MPTCP 的潜力, 并公开了数据集和脚本