Satellite Networking Primer

这一部分是常识普及

1. 星座平均每天采集的数据量高达 120 TB
2. LEO EO卫星: 高度500km附近
3. TLE数据存在: 卫星与地面站的接触时间点是可预测的
4. GS-Cloud: 骨干网, bw在数百 Mbps 到数 Gbps

There are nearly 5000 satellites in orbit today, up by 5× compared to a decade ago [42]. The increase has been driven by reduced cost of designing and launching hardware for small satellites (e.g., “shoebox-sized” cube-sats). A single rocket can launch multiple such satellites using ride-share agreements that amortize cost.

如今，在轨运行的卫星数量已接近5000颗，相较于十年前增长了5倍 [42]。这一增长主要得益于小型卫星（例如“鞋盒大小”的立方体卫星 CubeSat）硬件设计与发射成本的降低。通过分摊成本的“搭乘共享”协议，单枚火箭便可将多颗此类卫星送入轨道。

Satellite Orbits: Emerging LEO constellations for earth observation typically operate in polar orbits around 500 km above the Earth. A given satellite may return to the same location above Earth only every six to twelve days. Satellite operators—e.g., Planet Inc. [14], and Spire [1])—deploy large-scale constellations comprising hundreds of satellites to increase imaging frequency to multiple images per day. This is in contrast to traditional earth observation constellations that have only a few satellites, e.g., 2 satellites in the European Space Agency’s Sentinel 2 [15].

A satellite’s location with respect to Earth is reasonably predictable using Two Line Element (TLE) orbit descriptors published at regular intervals by multiple agencies such as Celestrak [29]. This means that the satellite-ground station contact time points are predictable and we use these as an input for scheduling. Past work [44] has also demonstrated the ability to predict the radio link quality across time.

卫星轨道

新兴的地球观测 LEO 星座通常运行在距离地球表面约 500 公里的极地轨道上。特定的一颗卫星可能需要六至十二天才会返回至地球上空的同一位置。为了将成像频率提高到每天多次，Planet Inc. [14] 和 Spire [1] 等卫星运营商部署了由数百颗卫星组成的大规模星座。这与传统的地球观测星座（通常仅有少数几颗卫星，如欧洲空间局的哨兵2号星座仅有2颗卫星 [15]）形成了鲜明对比。

借助 Celestrak [29] 等多家机构定期发布的TLE数据，卫星相对于地球的位置可以被相当精确地预测。这意味着 卫星与地面站的接触时间点是可预测的，我们将其作为调度的输入信息 。过去的研究 [44] 也已证明，可以预测无线电链路质量随时间的变化情况。

Imaging Equipment and Data Volume: Earth observation satellites capture images of Earth in different parts of the frequency spectrum, e.g., RGB, Radio Waves, Infrared, etc. The multi-spectral imagery as well as the high resolution are responsible for high volumes of data transfer from satellite to Earth. The Dove constellation captures 120 TB of data per day on average, in our evaluation period.

成像设备与数据量

地球观测卫星在频谱的不同部分（如可见光RGB、无线电波、红外线等）捕捉地球影像。多光谱成像与高分辨率是导致从卫星到地球产生海量数据传输的主要原因。 在我们的评估周期内，Dove 星座平均每天采集的数据量高达 120 TB

Ground Station Design: The satellite to ground station link is a high frequency link (e.g., X-band 8-10 GHz) with downlink bandwidths of up to 2 Gbps and uplinks of a few Kbps [13]. Bandwidth varies as a function of distance between satellites and ground stations during a contact period, and across contact periods. A single ground antenna can only talk to one satellite at a time. However, a satellite operator may deploy multiple antennas at the ground station, with each antenna talking to an independent satellite. We assume that satellites cannot communicate amongst each other, i.e., there are no inter-satellite links (this is true for all major LEO constellations today).

The ground station locations are selected using several constraints such as land availability, spectrum licensing requirements, lack of interference, orbital calculations, etc. The ground stations transfer data to the cloud using a backhaul link. The quality of the backhaul connection depends on the location of the ground station and can vary from 100s of Mbps to a few Gbps. This bandwidth is relatively stable across time (as opposed to the satellite-ground station bandwidth which varies due to orbital motion). While there is scant public information about the nature of these links, the range of 100s of Mbps to a few Gbps is consistent with anecdotal evidence based on both our conversations with satellite operators and public statements by ground station operators [38].

地面站设计

卫星到地面站的链路是高频链路（例如 X 波段，8-10 GHz），其下行链路带宽最高可达 2 Gbps，而上行链路仅为数 Kbps [13]。在一个接触周期内，带宽会随着卫星与地面站之间距离的变化而变化，并且不同接触周期之间的带宽也会有所不同。单座地面天线一次只能与一颗卫星通信。但是，卫星运营商可以在一个地面站部署多座天线，每座天线与一颗独立的卫星通信。我们 假设卫星之间无法相互通信，即不存在星间链路（目前所有主流 LEO 星座皆是如此）

地面站的选址受多种因素制约，如土地可用性、频谱许可要求、无信号干扰以及轨道计算等。地面站通过回程链路将数据传输至云端。回程链路的质量取决于地面站的地理位置，其带宽可在数百 Mbps 到数 Gbps 之间变化。 与因轨道运动而变化的星地链路带宽不同，回程链路的带宽随时间相对稳定。尽管关于这些链路性质的公开信息甚少，但根据我们与卫星运营商的交流以及地面站运营商的公开声明所获得的经验性证据 [38]，数百 Mbps 到数 Gbps 这一带宽范围是相符的

Data Download Process: The images collected by a satellite arrive at the cloud endpoints via two stages: satellite to ground station first, and then ground station to cloud. Each stage incurs hour-level latencies today. For the first step, the access to a ground station is the key bottleneck, i.e. a satellite must wait till its orbit brings it near a ground station, before it can transfer data. For the second step, the backhaul connectivity (to the cloud) is a bottleneck, especially for ground stations that are remote and/or get disproportionately high amount of data from satellites.

卫星采集的图像通过两个阶段抵达云端端点：首先是从卫星到地面站，然后是从地面站到云端。目前，每个阶段都会产生小时级别的延迟。对于第一阶段，与地面站的接触机会是关键瓶颈，即卫星必须等到其轨道使其接近地面站时才能传输数据。对于第二阶段，（到云端的）回程连接是瓶颈，特别是对于那些位置偏远和/或从卫星接收了不成比例的大量数据的地面站而言

Assumptions: In this paper, we assume that (a) satellites cannot send data to each other directly. This is true for all major LEO earth observation satellite constellations today; (b) ground stations cannot send data to each other either, which is because ground station-ground station communication would consume the same bandwidth (to the Internet) that the ground station could use to communicate to the cloud and (c) ground stations are not shared across multiple applications Nevertheless, we believe our withhold scheduling algorithms generalize to any topology that relaxes these assumptions. We also discuss these relaxations at the end of the paper.

假设

在本文中，我们做出如下假设：

(a) 卫星之间不能直接相互发送数据。目前所有主流的 LEO 地球观测星座均遵循此模式

(b) 地面站之间也不能相互发送数据，因为地面站之间的通信会占用其本可用于与云端通信的互联网带宽

(c) 地面站不由多个应用共享

尽管如此，我们相信我们的预扣调度算法可以推广到任何放宽了这些假设的拓扑结构。我们也会在论文末尾讨论这些假设放宽后的情况。