BACKGROUND & RELATED WORK

LEO mega-constellations being a new problem area, we include relevant background to aid our readers.

What makes an LEO satellite network?

A large LEO constellation may comprise hundreds to thousands of satellites. These satellites are organized into a number of orbits. An orbit is described by: (a) its inclination, 𝑖, the angle its plane makes with the Equator traveling northward; and (b) its height above sea-level, ℎ < 2,000 km. Satellites within one orbit are uniformly spaced out. A set of orbits with the same 𝑖 and ℎ, and crossing the Equator at uniform spacing from each other, is called an orbital shell. (Typically orbits in a shell vary their elevation around ℎ to avoid collisions — these minor differences are largely immaterial to networking.) Large constellations may deploy one or more such shells. This description only captures a sub-class of LEO constellations, but all recently proposed constellations fit this sub-class.

一个大型低地球轨道（LEO）星座可能由数百到数千颗卫星组成。这些卫星被组织成多个轨道。轨道的描述包括：（a）其倾角 \(i\)，即轨道平面与赤道向北移动时所形成的角度；（b）其海拔高度 \(h < 2,000\) km。处于同一轨道中的卫星均匀分布。具有相同 \(i\) 和 \(h\) 的一组轨道，且彼此在赤道交叉时均匀间隔，称为轨道壳（orbital shell）。通常，轨道壳中的轨道在高度 \(h\) 周围会有所变化，以避免碰撞——这些微小的差异对于网络连接而言基本上是无关紧要的。大型星座可以部署一个或多个这样的轨道壳。该描述仅涵盖了LEO星座的一个子类，但所有最近提议的星座均符合这一子类。

Each satellite uses radio up/down links to communicate with ground stations (GSes), as shown in Fig. 1. A satellite can only connect to GSes from which it can be seen at sufficiently high elevation in the sky, as defined by the minimum angle of elevation, 𝑙. A satellite directly overhead a GS is at elevation 90°, while one at the horizon is at 0°. If the minimum elevation 𝑙 = 40°, only GSes that see a satellite at elevation 40°or higher in the sky can communicate with it. Thus, smaller 𝑙 values allow GSes to talk to satellites closer to the horizon, while larger 𝑙 values are more restrictive. However smaller 𝑙 values also have a downside: connections from lower elevation experience reduced antenna gain and signal quality due to beam contour widening and increased attenuation.

每颗卫星通过无线电上下行链路与地面站（GSes）进行通信，如图1所示。卫星只能与其在天空中以足够高的仰角可见的地面站建立连接，仰角的最小值定义为 \(l\)。卫星正好位于地面站正上方时，仰角为90°；而位于地平线上的卫星，其仰角为0°。如果最小仰角 \(l = 40°\)，则只有在天空中以40°或更高仰角可见的地面站才能与该卫星进行通信。因此，较小的 \(l\) 值允许地面站与更接近地平线的卫星进行通信，而较大的 \(l\) 值则限制性更强。然而，较小的 \(l\) 值也存在缺点：来自低仰角的连接由于波束轮廓扩展和衰减增加，导致天线增益和信号质量降低。

According to Kuiper’s FCC filings [46], each satellite will have multiple antennas, with each antenna supporting multiple steerable beams; the beam steering and frequency band allocation will be software-defined, with the goal of maximizing throughput. Whether each GS can also simultaneously connect to multiple satellites depends on the type of GS: a user terminal uses a single phased-array antenna, while an enterprise user or gateway terminal uses multiple parabolic antennas with more flexibility [46].

根据Kuiper在FCC的备案文件，每颗卫星将配备多个天线，每个天线支持多个可调波束；波束调整和频段分配将采用软件定义，以最大化吞吐量。每个地面站（GS）是否能够同时与多颗卫星连接取决于地面站的类型：用户终端使用单个相控阵天线，而企业用户或网关终端则使用多个抛物面天线，具备更大的灵活性。

Satellites also connect to each other, using laser inter-satellite links (ISLs). An end-end path between two GSes comprises a radio up-link from the source GS to the ingress satellite, followed by zero or more laser ISLs, and then the egress radio down-link to the destination GS.

卫星之间也通过激光星间链路（ISLs）进行连接。两个地面站（GS）之间的端到端路径包括从源地面站到入站卫星的无线电上行链路，随后是零个或多个激光ISL，最后是到达目标地面站的无线电下行链路。

The largest proposed constellations

To add concrete numbers to the above abstract description of LEO constellations, we describe the design parameters for the largest three proposed constellations.

为了为上述低地球轨道（LEO）星座的抽象描述添加具体数字，我们将描述三个最大提议星座的设计参数。

SpaceX Starlink: Table. 1 details the first phase of Starlink, with 4,409 satellites planned across 5 orbital shells [66–69]. SpaceX is currently deploying S1, with 1,584 satellites (72 orbits, each with 22 satellites), ℎ = 550 km, and 𝑖 = 53°. The minimum elevation, 𝑙 = 25°. S1 will cover most of the world’s population, but will not extend service to less populated regions at high latitudes. This coverage issue will be addressed by the higher inclination shells, S3S5. SpaceX’s stated plan is to deploy more than 42,000 satellites, but it is unclear how much of this is posturing to secure spectrum [34].

SpaceX Starlink

表1详细介绍了Starlink的第一阶段，计划在5个轨道壳中部署4,409颗卫星。SpaceX目前正在部署S1，包含1,584颗卫星（72个轨道，每个轨道22颗卫星），其高度为\(h = 550\) km，倾角为\(i = 53°\)。最小仰角为\(l = 25°\)。S1将覆盖世界大部分人口，但不会向高纬度的低人口地区提供服务。这个覆盖问题将通过倾角更高的轨道壳S3和S5来解决。SpaceX声明计划部署超过42,000颗卫星，但尚不清楚其中有多少是为了争取频谱而进行的夸张。

Amazon Kuiper: Kuiper plans three shells, with a total of 3,236 satellites at slightly different operating heights [47–49]. Kuiper entirely eschews connectivity near the poles, with all its shells having inclinations under 52°. The FCC filings mention a few possible values of 𝑙: “20(min)/30/35/45” [47].

Amazon Kuiper

Kuiper计划建立三个轨道壳，总共部署3,236颗卫星，工作高度略有不同。Kuiper完全避免在极地区域提供连接，其所有轨道壳的倾角均低于52°。FCC备案文件提到了一些可能的最小仰角值：“20（最小）/30/35/45”。

Telesat: Telesat plans two shells with a total of 1,671 satellites [73], roughly a fifth of which will cover the higher latitudes, using an inclination of 98.98°, with the rest focused on improving capacity at lower latitudes. Telesat plans 𝑙 = 10°, but the feasibility of this is unclear — unlike Starlink and Kuiper, whose filings detail how to address beam contour and antenna gain changes for 𝑙 ≥ 25°, Telesat’s filings thus far omit such information.

Telesat

Telesat计划建立两个轨道壳，总共部署1,671颗卫星，其中约五分之一将覆盖高纬度地区，倾角为\(98.98°\)，其余部分则专注于提高低纬度地区的容量。Telesat计划将最小仰角设置为\(l = 10°\)，但这一可行性尚不明确——与Starlink和Kuiper不同，它们的备案文件详细说明了如何解决\(l \geq 25°\)时波束轮廓和天线增益变化的问题，而Telesat迄今为止的备案文件则省略了此类信息。

Unique dynamics of LEO networks

A satellite’s height, ℎ, determines its orbital velocity, and thus, orbital period [56]. At ℎ = 550 km, the orbital velocity is more than 27,000 km/hr, and satellites complete an orbit around the Earth in ∼100 minutes [6]. As satellites travel fast across GSes, GS-satellite links can only be maintained for a few minutes, after which they require a handoff. ISLs also continuously change in length. The Earth’s shape and orbital geometry results in satellites coming closer at higher latitudes. This results in a continuous change in their relative positions and hence the ISL lengths and latencies.

The end-end path between two GSes thus changes both in terms of which satellites are involved, and in terms of the lengths of both the GS-satellite links and the ISLs.

卫星的高度 \(h\) 决定了其轨道速度，从而影响轨道周期。在 \(h = 550\) km 时，轨道速度超过 27,000 km/h，卫星绕地球一圈的时间约为 100 分钟。由于卫星在地面站（GSes）上空快速移动，GS-卫星链路只能维持几分钟，之后需要进行切换。激光星间链路（ISLs）的长度也会不断变化。地球的形状和轨道几何导致卫星在高纬度地区更为接近，这造成了它们相对位置的持续变化，从而影响 ISL 的长度和延迟。

因此，两个地面站之间的端到端路径在所涉及的卫星以及 GS-卫星链路和 ISL 的长度方面都会发生变化。

Mobility is, of course, well-studied in a variety of contexts, including cellular networks, high-speed trains, drones and airplanes, and swarms of mobile nodes. For many of these settings, there are also models of mobility, together with simulation and analysis infrastructure. However, LEO satellite mobility is unique for several reasons:

移动性在多种背景下都得到了充分研究，包括蜂窝网络、高速列车、无人机和飞机以及移动节点群体。在许多这些场景中，也有关于移动性的模型，以及相应的仿真和分析基础设施。然而，LEO 卫星的移动性因以下几个原因而独特：

• LEO mobility features much larger distances and velocities than terrestrial mobile networks.

• Unlike most other settings, LEO networks’ core infrastructure itself is mobile, rather than just the end-points.

• LEO mobility is predictable; this is not the case for the most well-studied setting, cellular networks.

• LEO networks feature thousands of network switches (satellites) capable of providing Tbps of connectivity. This scale is far beyond other networked swarms.

Each previously well-studied setting features one or two of the above characteristics, but not all of them. For instance, trains, and to a lesser extent, airplanes, also feature predictable motion, but none of the other characteristics.

• LEO的移动性特征具有比地面移动网络更大的距离和速度。
• 与大多数其他环境不同，LEO网络的核心基础设施本身是移动的，而不仅仅是终端。
• LEO的移动性是可预测的；而这一点在最为研究透彻的蜂窝网络中并不成立。
• LEO网络拥有数千个网络交换机（卫星），能够提供Tbps级别的连接。这一规模远超其他网络群体。

每个先前研究透彻的环境通常只具备上述特征中的一两个，但并不具备全部。例如，列车以及在较小程度上飞机也具有可预测的运动，但并不具备其他特征。

Large LEO networks need new research

Commercial satellite networks already provide varied network services. HughesNet [35] and Viasat [76] primarily serve areas poorly connected by terrestrial fiber, as well as aircrafts and ships. These are both GEO satellite constellations, and operating at 35,786 km, they incur hundreds of milliseconds of latency. Besides, their performance and service goals being different, their GEO satellites are, by definition, stationary with respect to the Earth, and thus do not feature LEO dynamics. Iridium [36, 37] operates in LEO, but primarily offers satellite telephony rather than broadband Internet. Iridium, with 82 satellites in operation, is the largest of the networks that pre-date the new LEO mega-constellations.

商业卫星网络已经提供了多样化的网络服务。HughesNet和Viasat主要服务于那些与地面光纤连接不良的地区，以及航空器和船舶。这两者都是静止轨道（GEO）卫星星座，运行在35,786 km的高度，因而会产生数百毫秒的延迟。此外，由于它们的性能和服务目标不同，这些GEO卫星在定义上相对于地球是静止的，因此不具备LEO动态特征。Iridium在低地球轨道（LEO）运行，但主要提供卫星电话服务，而非宽带互联网。Iridium拥有82颗在运卫星，是在新LEO超级星座出现之前最大的网络。

Thus, no prior networks have all the features of the new LEO networks, the largest of which are planned to operate thousands of satellites instead of tens, and provide mass market low-latency broadband Internet, rather than niche services. One of the upcoming constellations, Starlink, already has more than 400 satellites operational, and expects a public launch of their service as soon as 2020 [16, 23]. Over the long-term, such networks have the potential to fundamentally change the Internet, making it crucial for research to keep pace with the hectic pace of industry developments.

因此，之前的网络并未具备新LEO网络的所有特征，后者计划运营数千颗卫星，而不是几十颗，并提供大众市场的低延迟宽带互联网，而非小众服务。即将推出的星座之一Starlink已拥有超过400颗在运卫星，并预计将在2020年尽快公开发布其服务。从长远来看，这些网络有潜力从根本上改变互联网，因此研究必须跟上行业发展的快速步伐。

The networking community, recognizing this need, is indeed ramping up research in this direction. While there is a large body or earlier work from the 1990s on GEO and small LEO networks [2, 4, 14, 15, 18, 24, 43, 50, 53, 71, 78–80, 82], several position papers [5, 29, 44] have highlighted the new opportunities and challenges of megaconstellations, e.g., in intra-constellation routing [29] and interdomain routing [44], and end-end congestion control [5]. Followup work has since laid out novel proposals for topology design [6] and Internet inter-domain routing [26] in this context.

网络社区意识到这一需求，确实正在加大在这一方向上的研究力度。尽管1990年代关于GEO和小型LEO网络的研究成果丰富，但一些立场文件强调了超级星座的新机会和挑战，例如，在星座内部路由和跨域路由，以及端到端拥塞控制等方面。后续工作也提出了在此背景下拓扑设计和互联网跨域路由的新提案。

We are missing the right analysis tools

Unfortunately, the networking community lacks the right tools to attack many of the LEO networking challenges recent work has pointed out. We need software to simulate the behavior of such networks, so that we can deeply understand the problems, and new research ideas can be evaluated. Understanding the packet-level behavior of a network is obviously important for congestion control research, but ultimately, practitioners also want to evaluate routing and topology work in terms of how it impacts network packets, e.g., do some routing schemes cause more packet reordering, and thus, ultimately result in poor performance?

不幸的是，网络社区缺乏合适的工具来应对近期研究指出的许多LEO网络挑战。我们需要软件来模拟此类网络的行为，以便深入理解问题，并评估新的研究想法。理解网络的分组级行为显然对于拥塞控制研究至关重要，但最终，实践者也希望能够评估路由和拓扑工作对网络分组的影响，例如，某些路由方案是否会导致更多的分组重排序，从而最终导致性能下降。

Unfortunately, there is no simulator that fully addresses these needs. SNS3 [65] models GEO satellite communication channels, but does not support LEO satellites or inter-satellite connectivity. Another simulation effort [33] focused on the polar constellations of interest in the nineties, and the problems of interest therein, e.g., connectivity across “seams” that result from satellites traveling northward in one (longitudinal) hemisphere and southward in the other. While we could have extended this work for our study of modern LEO networks, we based Hypatia on the ns-3 platform to benefit from its more active development and support. Note that this prior work also did not analyze congestion control and traffic engineering, nor did it provide visualizations beyond the belowdiscussed SaVi tool [81]. A satellite mobility model is available for ns-3 [61], which can convert satellite trajectories in a specific format into a coordinate system compatible with ns-3. This capability is useful, and we build on it by adding models for inter-satellite and GS-satellite connections. Recent work on LEO inter-satellite topologies [6] evaluated topologies only in terms of path hop-counts and distances, not packet simulations. Likewise, work on interdomain routing [26] only modeled the network control messages and path distances. Another effort [21] estimates the throughput of new LEO networks using statistical methods, and minimizes the number of GSes needed to support the throughput. It does not account for network routing and transport dynamics.

不幸的是，目前网络社区缺乏能够全面满足这些需求的模拟器。SNS3模型主要针对静止轨道（GEO）卫星通信通道，但不支持低地球轨道（LEO）卫星或星间连接。另一个模拟工作则集中于90年代感兴趣的极地星座及其相关问题，例如，卫星在一个（经度）半球向北移动而在另一个半球向南移动所导致的“缝隙”连接问题。虽然我们可以扩展这一工作以研究现代LEO网络，但我们选择基于ns-3平台构建Hypatia，以利用其更活跃的开发和支持。值得注意的是，之前的工作也没有分析拥塞控制和流量工程，也没有提供超出下文讨论的SaVi工具的可视化。

ns-3中有一个卫星移动模型，可以将特定格式的卫星轨迹转换为与ns-3兼容的坐标系统。这一功能非常有用，我们在此基础上添加了星间连接和GS-卫星连接的模型。近期关于LEO星间拓扑的研究仅从路径跳数和距离的角度评估拓扑，而没有进行分组模拟。同样，关于跨域路由的工作也仅建模了网络控制消息和路径距离。另一个研究使用统计方法估计新LEO网络的吞吐量，并最小化支持该吞吐量所需的地面站数量，但并未考虑网络路由和传输动态。

We also need visualizations that help build sorely missing intuition for these new networks. While there are many beautiful visualizations, at least for Starlink [12, 22, 30, 32], most of these do not focus on networking concepts such as the evolution of paths, utilization, and congestion. The closest related work [29, 30] does not simulate packet-level behavior, and does not provide source code for its path-granularity computations or visualizations. NASA’s GMAT [57] can be used to visualize trajectories of objects in space; SaVi [81] can additionally render coverage of a satellite. However, neither provides the ability to define the topology, model network links, or run network-centric measurements.

我们还需要可视化工具，以帮助建立对这些新网络的迫切缺失的直觉。尽管有许多美观的可视化，尤其是针对Starlink的可视化，但大多数并未关注诸如路径演变、利用率和拥塞等网络概念。最相关的工作虽然提到了一些内容，但并未模拟分组级行为，也未提供其路径粒度计算或可视化的源代码。NASA的GMAT可以用于可视化太空中物体的轨迹；SaVi则可以额外呈现卫星的覆盖范围。然而，这两者都无法定义拓扑、建模网络链接或进行网络中心的测量。

While we expect that eventually the community will collect measurements from real clients on LEO networks, this will not alleviate the need for simulation and analysis tools. For a variety of network contexts, such tools continue to be valuable to understand existing phenomena, and to devise novel, hard-to-evaluate-in-thewild techniques.

因此，尽管我们预计最终社区将从LEO网络中的真实客户收集测量数据，但这并不能缓解对模拟和分析工具的需求。在各种网络环境中，这些工具仍然对理解现有现象以及设计新颖且难以在实际中评估的技术至关重要。