Optimising LEO Gateway Placement for the People

Low Earth Orbit (LEO) satellite networks are rapidly becoming the default broadband solution in developing countries, especially in areas lacking reliable terrestrial infrastructure. This paper investigates how gateway (GW) placement affects user experience in such networks, focusing on key performance metrics such as hop count and latency. Using real-world Starlink deployment data and global population distributions, we compare three gateway placement strategies: Starlink’s current deployment (382 GWs), a country-centroid model (241 GWs), and Internet Exchange Point (IXP) co-location (382 and 787 locations). Our analysis shows that IXP co-location delivers superior performance, connecting (96.8%) of global users via direct bent-pipe links compared to countrycentroid placement (94%) and Starlink’s current layout (83.9%). While country-centroid placement appears optimal theoretically, it fails practically in landlocked regions lacking Point of Presence (POP) backhaul infrastructure. Rwanda exemplifies this challenge, forcing Starlink to deploy community gateways instead. Conversely, Starlink’s strategy of leveraging existing IXP and POP locations proves effective, demonstrating that even revolutionary satellite technology must align with terrestrial infrastructure realities. Since bent-pipe connectivity offers significantly lower latency than multihop inter-satellite paths, these results demonstrate that optimal gateway placement requires alignment with both population density and existing internet infrastructure. This is especially crucial in the Global South, where LEO networks are increasingly relied upon for everyday internet access.

低地球轨道（LEO）卫星网络正迅速成为发展中国家主要的宽带解决方案，尤其是在缺乏可靠地面基础设施的地区。本文旨在研究网关（GW）布局如何影响此类网络中的用户体验，重点关注跳数和延迟等关键性能指标。我们利用真实的星链部署数据和全球人口分布数据，比较了三种网关布局策略：星链当前部署（382个网关）、国家中心点模型（241个网关），以及互联网交换点（IXP）共址模型（382个和787个位置）。

我们的分析表明，IXP共址策略性能最优，通过直接的“弯管”链路连接了全球96.8%的用户，而国家中心点模型为94%，星链当前布局为83.9%。

虽然国家中心点布局在理论上看似最优，但在缺乏接入点（POP）回传基础设施的内陆地区实际上并不可行。卢旺达便是这一挑战的典型例子，迫使星链不得不部署社区网关作为替代方案。相反，星链利用现有IXP和POP位置的策略被证明是有效的，这表明即使是革命性的卫星技术也必须与地面基础设施的现实情况相结合。

由于“弯管”连接提供的延迟远低于多跳星间路径 ，这些结果表明，最优的网关布局需要同时考虑人口密度和现有的互联网基础设施。这一点在全球南方国家尤为关键，因为这些地区日益依赖LEO网络来满足日常互联网接入需求。

Introduction

Low Earth Orbit (LEO) satellite networks promise connectivity in regions where traditional terrestrial infrastructure is sparse or unreliable. Despite this advantage, LEO technology is not without its limitations, particularly regarding latency. The time it takes for data to travel from a user’s device to a server and back, can significantly impact the user experience, making real-time applications such as video conferencing, online gaming, and remote work frustrating or even impractical [11, 18].

A central challenge in LEO satellite network design is balancing the use of Inter-Satellite Links (ISLs) with the deployment of ground gateways close to users. ISLs allow satellites to communicate directly, forming a mesh network in space that can route data without touching the ground. While this is beneficial in areas lacking nearby ground stations, it often leads to higher latency compared to a bent-pipe connection to a local ground station, as data must traverse longer paths through space [2]. By placing these gateways closer to users, data can follow a shorter, more direct route, significantly reducing latency and enhancing overall service quality.

Starlink has been actively investing in expanding its ground infrastructure in Africa, with two new PoP locations established in Nairobi and Johannesburg at the beginning of 2025, along with several new ground stations in the region [9]. Prior to this deployment, Kenyan users routinely experienced latencies above 100 milliseconds, often around 120 milliseconds despite Starlink’s dense ISL mesh. This high latency was primarily because all traffic was routed through distant European ground stations, thousands of kilometers away [14]. Within days of the Nairobi PoP coming online, latencies for Kenyan users dropped dramatically to as low as 26 milliseconds, as reported by Starlink resellers [16]. Independent measurement by Ookla for the first quarter of 2025 showed median latencies in Kenya at 53 milliseconds [15], still a substantial improvement over previous levels. This reduction not only halved round-trip time, but also significantly improved jitter and overall service quality, making Starlink a more viable option for latency-sensitive applications. This highlights a broader and increasingly relevant question: Given a satellite constellation’s ISL capabilities, how many (or which) gateways should operators build to meet latency and jitter targets for their users? As satellite internet services such as Starlink expand globally, understanding and optimizing this trade-off becomes crucial. Too few gateways may force reliance on ISLs, leading to higher latency and poorer performance. Too many gateways could be prohibitively expensive and logistically challenging to maintain. Striking the right balance is essential to ensure that satellite internet can truly bridge the digital divide, providing not just connectivity, but also a high-quality user experience.

This study addresses these trade-offs by analysing Starlink’s gateway deployment strategy and evaluating alternative placement models using real satellite and population data. We compare three configurations: Starlink’s current layout, a country-centroid model, and an IXP-based strategy assessing their performance in terms of latency, hop count, and user coverage. Our results show that the IXP-based model, which locates Gateways at IXPs that are already distributed in locations of high population density, offers the best latency profile and bent-pipe coverage, while countrycentroid placement ensures more equitable geographical access in remoter areas.These findings point to the need for hybrid gateway strategies that balance performance and deployment feasibility as LEO networks continue to scale.

低地球轨道（LEO）卫星网络有望为传统地面基础设施稀疏或不可靠的地区提供连接。尽管有此优势，LEO技术并非没有局限性，尤其是在延迟方面。数据从用户设备传输到服务器再返回所需的时间，会显著影响用户体验，使得视频会议、在线游戏和远程办公等实时应用变得令人沮丧甚至不切实际 [11, 18]。

LEO卫星网络设计的一个核心挑战是在使用星间链路（ISL）与在用户附近部署地面网关之间取得平衡。ISL允许卫星之间直接通信，在太空中形成一个网状网络，无需经过地面即可路由数据。虽然这在缺少附近地面站的地区很有利，但与通过本地地面站的 “弯管”连接 相比，它通常会导致更高的延迟，因为数据必须在太空中经过更长的路径 [2]。通过将这些网关部署在离用户更近的地方，数据可以沿更短、更直接的路径传输，从而显著降低延迟并提升整体服务质量。

星链一直在积极投资扩建其在非洲的地面基础设施，于2025年初在内罗毕和约翰内斯堡建立了两个新的接入点（PoP），并在该地区新增了几个地面站 [9]。在此部署之前，尽管星链拥有密集的ISL网状网络，肯尼亚用户的延迟通常仍超过100毫秒，常在120毫秒左右。这种高延迟主要是因为所有流量都需通过数千公里外的欧洲地面站进行路由 [14]。内罗毕PoP上线几天内，据星链经销商报告，肯尼亚用户的延迟急剧下降至26毫秒 [16]。Ookla在2025年第一季度的独立测量显示，肯尼亚的中位延迟为53毫秒 [15]，这仍是相较于之前水平的巨大进步。这次优化不仅使往返时间减半，还显著改善了抖动和整体服务质量，使星链成为延迟敏感型应用更可行的选择。这引出了一个更广泛且日益重要的问题：考虑到卫星星座的ISL能力，运营商应建造多少（或哪些）网关才能满足用户的延迟和抖动目标？

随着星链等卫星互联网服务的全球扩张，理解和优化这种权衡变得至关重要：

• 网关过少可能迫使网络依赖ISL，导致延迟升高和性能下降
• 网关过多则可能成本高昂且维护困难

取得适当的平衡对于确保卫星互联网真正弥合数字鸿沟至关重要，它不仅要提供连接，还要提供高质量的用户体验。

本研究通过分析星链的网关部署策略，并利用真实的卫星和人口数据评估替代布局模型，来探讨这些权衡。我们比较了三种配置：星链当前布局、国家中心点模型和基于IXP的策略，并评估它们在延迟、跳数和用户覆盖方面的性能。

我们的结果表明，将网关设在人口密度高的IXP位置的基于IXP的模型，提供了最佳的延迟表现和“弯管”覆盖，而国家中心点布局则确保了偏远地区更公平的地理接入。这些发现指出，随着LEO网络的持续扩展，需要采用兼顾性能和部署可行性的混合网关策略。

Related Work

Gateway placement is a cornerstone of satellite network design, directly influencing critical performance metrics such as latency, throughput, and reliability. In LEO satellite networks, the strategic placement of gateways is crucial for determining how efficiently satellites can transfer user data to the internet backbone. Research in this area has primarily focused on optimising gateway locations to balance traffic loads, minimise latency, and reduce the number of gateways [1, 2, 5].

Guo et al. [10] propose a gateway optimization (GPO) method for Low Earth Orbit (LEO) satellite networks, aiming to balance traffic loads while minimizing the number of gateways. Their approach models the problem as a combinatorial optimization task, using a gravity model to estimate traffic matrices between satellites and gateway. They employ a discrete particle swarm optimization (PSO) algorithm to identify optimal gateway locations, considering constraints such as link interference, satellite bandwidth, and the number of satellite antennas. Their simulations, based on constellations such as Starlink, demonstrate that these methods can reduce the number of gateways while maintaining balanced traffic loads. However, they do not consider the interplay between gateway placement and ISL routing, an aspect of the trade-off we investigate. Similarly, Cao et al. [4] address gateway placement in 5G-satellite hybrid networks, focusing on reliability optimization. They propose two algorithms: an Optimal Enumeration Algorithm (OEA) and a Cluster-based Approximation Placement Algorithm (CAPA). OEA exhaustively evaluates all possible gateway combinations to find the optimal placement, while CAPA uses clustering techniques to reduce computational complexity, achieving near-optimal results with significantly lower processing demands. Their focus is on terrestrial-satellite integration, and they do not explicitly address the role of ISLs in reducing latency or the impact of regulatory constraints on gateway placement.

In the context of the gateway-ISL trade-off, Zhang et al. [21] take a step closer to our research by addressing gateway placement in LEO mega-constellation networks with a focus on minimizing ISL usage. They introduce a novel metric for ISL usage and formulate a mixed-integer optimization model for the Gateway Site Optimization (GSO) problem. Their approach uses an Improved Binary Discrete PSO (IBD-PSO) algorithm to solve the optimization problem, considering global user traffic distribution. Their simulations, based on the Starlink constellation, show that optimal gateway placement can significantly reduce the number of ISL hops, thereby lowering latency. This work is highly relevant to our study, as it explicitly considers the trade-off between gateway placement and ISL usage.

The trade-off arises because ground gateways require physical infrastructure, which can be costly and logistically challenging to deploy, especially in remote areas. ISLs, on the other hand, leverage the satellite constellation to provide coverage without additional ground stations but introduce delays due to longer signal paths. In regions with dense terrestrial infrastructure (e.g., North America), users benefit from nearby PoPs, with latencies as low as 25–30 ms [15, 18]. In contrast, African region, with sparse infrastructure, often rely on ISLs to reach distant PoPs, resulting in higher latencies (e.g., 55–80 ms or more) [19].

网关布局是卫星网络设计的基石，直接影响延迟、吞吐量和可靠性等关键性能指标。在LEO卫星网络中，网关的战略性布局对于决定卫星将用户数据传输到互联网骨干网的效率至关重要。 该领域的研究主要集中在优化网关位置，以平衡流量负载、最小化延迟并减少网关数量 [1, 2, 5]

Guo等人[10]提出了一种用于LEO卫星网络的网关优化（GPO）方法，旨在平衡流量负载的同时最小化网关数量。他们的方法将该问题建模为组合优化任务，使用引力模型估算卫星与网关之间的流量矩阵，并采用离散粒子群优化（PSO）算法来确定最佳网关位置，同时考虑了链路干扰、卫星带宽和卫星天线数量等约束。他们基于星链等星座的仿真表明，这些方法可以在保持流量负载均衡的同时减少网关数量。然而，他们没有考虑网关布局与ISL路由之间的相互作用，而这正是我们研究的权衡点。类似地，Cao等人[4]研究了5G与卫星混合网络中的网关布局问题，重点关注可靠性优化。他们提出了两种算法：最优枚举算法（OEA）和基于聚类的近似布局算法（CAPA）。OEA通过穷举评估所有可能的网关组合来找到最优布局，而CAPA则使用聚类技术降低计算复杂性，以显著减少的处理需求实现了近优结果。他们的研究重点是地面与卫星的集成，并未明确探讨ISL在降低延迟中的作用或监管限制对网关布局的影响。

在网关与ISL权衡的背景下，Zhang等人[21]的研究与我们的方向更为接近，他们关注LEO巨型星座网络中的网关布局，旨在最小化ISL的使用。他们引入了一种新的ISL使用度量标准，并为网关选址优化（GSO）问题构建了一个混合整数优化模型。他们的方法使用改进的二进制离散粒子群优化（IBD-PSO）算法来解决该优化问题，并考虑了全球用户流量分布。他们基于星链星座的仿真表明，优化的网关布局可以显著减少ISL跳数，从而降低延迟。这项工作与我们的研究高度相关，因为它明确考虑了网关布局与ISL使用之间的权衡。

这种权衡的产生是因为地面网关需要实体基础设施，其部署成本高昂且在后勤上充满挑战，尤其是在偏远地区。另一方面，ISL利用卫星星座提供覆盖，无需额外的地面站，但由于信号路径更长而引入了延迟。在地面基础设施密集的地区（如北美），用户受益于附近的PoP，延迟可低至25-30毫秒 [15, 18]。相比之下，基础设施稀疏的非洲地区通常依赖ISL到达遥远的PoP，导致延迟较高（例如，55-80毫秒或更高）[19]。

Problem Statement

Starlink is rapidly becoming a primary broadband provider in many parts of the world, especially in developing countries with limited terrestrial infrastructure. While designed with extensive InterSatellite Link (ISL) capabilities, the lowest latency connections in Starlink’s architecture often rely on a bent-pipe model that routes traffic through a ground-based gateway (GW). This makes GW placement a critical factor in determining user experience, especially for latency sensitive applications.

As of 2025, Starlink has deployed or planned around 382 GWs globally, with most located near Internet Exchange Points (IXPs) [20]. However, it remains unclear how well these placements serve the global population, especially in regions with poor coverage or as yet unlicensed regions.

This raises several key questions: How many people can be served by the current GW layout using the bent-pipe architecture? How should GW placement evolve as LEO networks aim for lower latency? And what would global network performance look like if more GWs were deployed?

It is important to note the scope of our analysis. This work evaluates gateway placement strategies based on a specific objective minimizing latency for a user base that assumed to be proportional to global population density. We acknowledge that this represents a simplification. Commercial operators such as SpaceX are guided by a different set of objectives, including regulatory approval, market affordability, and the practical logistics of deploying infrastructure near reliable power and fiber backhaul. Their primary marker often consists of users in underserved regions where population density may be low, a factor our model does not explicitly capture. Therefore, this work aims not to prescribe a definite business strategy, but to provide a quantitative benchmark for how different placement philosophies impact network performance on a global scale.

Our goal is to assess the population reach, latency trade-offs, and deployment implications of Starlink’s current and future GW infrastructure, with the aim of informing better planning for LEO satellite networks.

星链正迅速成为世界许多地区主要的宽带提供商，尤其是在地面基础设施有限的发展中国家。尽管星链设计具备广泛的星间链路（ISL）能力，但其架构中延迟最低的连接通常依赖于通过地面网关（GW）路由流量的“弯管”模型。这使得网关布局成为决定用户体验的关键因素，特别是对于延迟敏感型应用。

截至2025年，星链已在全球部署或计划部署约382个网关，其中大多数位于互联网交换点（IXP）附近 [20]。然而，这些布局在多大程度上服务于全球人口，尤其是在覆盖较差或尚未获得许可的地区，目前尚不清楚。

这引出了几个关键问题：当前的网关布局使用“弯管”架构能服务多少人？随着LEO网络追求更低延迟，网关布局应如何演变？如果部署更多网关，全球网络性能将会如何？

需要注意的是我们分析的范围。本研究基于一个特定目标来评估网关布局策略：为与全球人口密度成正比的用户群最小化延迟。我们承认这是一种简化。像SpaceX这样的商业运营商受一套不同目标的指导，包括监管批准、市场可承受性，以及在可靠电力和光纤回传附近部署基础设施的实际后勤问题。他们的主要市场通常是服务欠缺地区的用户，而这些地区的人口密度可能较低，这是我们模型未明确捕捉的因素。因此，本研究的目的不是规定明确的商业策略，而是为不同布局理念如何在全球范围内影响网络性能提供一个定量基准。

我们的目标是评估星链当前及未来网关基础设施的人口覆盖范围、延迟权衡和部署影响，旨在为LEO卫星网络的更好规划提供参考。

Dataset

4.1 Gateway Location Dataset

To build our dataset of potential gateway locations, we turned to PeeringDB, a comprehensive database of Internet Exchange Points (IXPs) and co-location facilities where networks interconnect. PeeringDB’s API (accessed in June 2025) provided up-to-date information on 1,223 IXPs (Figure 2), which we geocoded to 787 distinct facilities [17]. While Starlink does not publish an official list of its gateways, a comprehensive list is maintained by the Starlink community and can be accessed through an interactive map at Starlink Gateway Locations [9] (accessed in June 2025). This information, corroborated by regulatory filings with agencies such as the FCC in the US and Ofcom in the UK, was analyzed to identify 382 gateways. Similarly, using the Geopandas library, we also extracted the centroid of 241 countries, islands, and overseas territories, which serve as gateway locations [7]. These gateways play a vital role in the network, enabling the transfer of signals between user terminals and satellites, as illustrated in Figure 2.

为了构建我们的潜在网关位置数据集，我们参考了PeeringDB，这是一个包含互联网交换点（IXP）和网络互联的共址设施的综合数据库。PeeringDB的API（于2025年6月访问）提供了1,223个IXP的最新信息（图2），我们将其地理编码到787个不同的设施 [17]。

虽然星链未公布其官方网关列表，但星链社区维护着一个全面的列表，可通过交互式地图在“Starlink Gateway Locations”[9]上访问（于2025年6月访问）。我们将这些信息与美国FCC和英国Ofcom等机构的监管文件进行核对，确定了382个网关。同样，我们使用Geopandas库提取了241个国家、岛屿和海外领土的中心点，作为网关位置 [7]。如图2所示，这些网关在网络中扮演着至关重要的角色，实现了用户终端和卫星之间的信号传输。

4.2 Starlink Satellite Locations

The evaluations rely on Starlink’s first shell (comprising 72 orbits with 22 satellites per orbit, totalling 1584 satellites), we use these positions to determine distances between ground segments and satellites. If a bent-pipe architecture is not viable, we use ISLs to route traffic, identifying the shortest path to the nearest ground station using Dijkstra’s algorithm. The Skyfield library is queried every 15 seconds to update satellite positions, recalculating groundto-satellite links (GSLs) and ISLs accordingly, as detailed in Table 1.

评估依赖于星链的第一壳层（包含72个轨道，每个轨道22颗卫星，共1584颗卫星），我们利用这些位置来确定地面部分与卫星之间的距离。如果“弯管”架构不可行，我们使用ISL来路由流量，通过迪杰斯特拉算法确定到最近地面站的最短路径。我们每隔15秒查询一次Skyfield库以更新卫星位置，并相应地重新计算地对星链路（GSL）和ISL，具体细节如表1所示。

4.3 World Population Density

To estimate where people might need Starlink, we used population data from the Socioeconomic Data and Applications Center (SEDAC) at a 55 km resolution [6]. This helps us see how many potential users are in different areas based on population size. Figure 3 shows a world map with population distribution, using colors to indicate how many people live in each area on a logarithmic scale.

为了估算人们可能需要星链的地区，我们使用了来自社会经济数据与应用中心（SEDAC）的55公里分辨率的人口数据 [6]。这有助于我们根据人口规模了解不同地区的潜在用户数量。图3展示了一张世界人口分布图，使用颜色以对数标度表示每个地区的人口数量。