Motivation and Related Work

A. Research Background: Observing the Earth via Satellites

Various earth observation missions. The earth observation (EO) ecosystem is continually evolving, toward seamless integration of new technologies, sensing modalities, and unconventional data sources. According to a recent report [45], nearly 45% of existing low earth orbit (LEO) satellites in space are launched for various EO missions, such as forest observation, weather forecasting, agriculture monitoring, crisis management and maritime surveillance, etc. Essential information on global areas collected by EO satellites enables us to monitor and protect our environment, manage our resources, respond to global disasters and enable sustainable development.

多样的对地观测任务 对地观测（EO）生态系统正持续演进，朝着无缝集成新技术、新传感模式和非常规数据源的方向发展。根据最近的一份报告[45]，太空中现有低地球轨道（LEO）卫星中近45%是为了执行各种对地观测任务而发射的，例如森林观测、天气预报、农业监测、危机管理和海事监控等。由EO卫星采集的关于全球各区域的关键信息，使我们能够监测和保护我们的环境、管理我们的资源、应对全球性灾害并实现可持续发展。

EO service model. Fig. 1 briefly plots the service model of existing EO ecosystems. At a high level, there are two major interactions. First, an EO service provider interacts with satellites via ground stations to assign EO tasks and collect data. In particular, an EO service provider owns and operates a number of EO satellites to perform EO tasks, e.g., acquiring information for a specific region of interest. To download data from space, existing commercial EO systems follow a “store first, download later” model, where space data is first acquired and stored in the satellite storage. All data will be downloaded to the ground for further processing and storage, via ground-satellite links (e.g., [47]), or satellite relays (e.g., [14]). Second, customers who require EO data for their own applications (e.g., remote monitoring) interact with the service provider via terrestrial Internet. Note that if the required contents have already been saved in the storage, the service provider sends them directly back to the customers. Otherwise, the service provider has to establish a new EO task to collect data, and then distribute content to the customers.

EO服务模型 图1简要描绘了现有EO生态系统的服务模型。宏观来看，存在两种主要交互。首先，EO服务提供商通过地面站与卫星交互，以分配观测任务和收集数据。具体而言，EO服务提供商拥有并运营若干EO卫星以执行任务，例如获取特定兴趣区域的信息。为从太空下载数据，现有的商业EO系统遵循“先存储，后下载”的模型，即空间数据首先被卫星获取并存储在星上存储器中。所有数据将通过地-星链路（例如[47]）或卫星中继（例如[14]）下载至地面，以供进一步处理和存储。其次，需要EO数据用于其自身应用（例如远程监控）的客户，通过地面互联网与服务提供商交互。

值得注意的是， 如果所需内容已保存在存储中，服务提供商会将其直接回传给客户。否则，服务提供官商必须建立一个新的EO任务来收集数据，然后再将内容分发给客户。

B. New Trends and Requirements Facing the EO Industry

With the rapid technical evolution in aerospace and remote sensing technologies, in recent years we have witnessed two critical trends in the EO industry.

随着航空航天和遥感技术的快速发展，近年来我们见证了EO产业的两个关键趋势。

T(1): from monolithic satellite to satellite constellations. Due to the high dynamics of LEO satellites and earth rotation, it is difficult for a single satellite to achieve high temporal resolution and maintain continuous observation in an EO mission. In particular, it may take hours to days for a satellite to revisit a specific region of interest upon the earth’s surface. Therefore, recent EO service providers leverage a constellation of EO satellites to cooperatively acquire data and reduce the revisiting time of a specific area of interest (AoI). For example, Planet [34] has launched and deployed 452 satellites consisting three constellations: PlanetScope [2], RapidEye [3], SkySat [4] to collaboratively perform EO missions. They can capture earth’s activities from multiple perspectives and dimensions with revisiting time less than one day.

趋势(1)：从单体卫星到卫星星座。 由于LEO卫星的高动态特性和地球自转，单颗卫星在EO任务中难以实现高时间分辨率和维持连续观测。具体来说，一颗卫星重访地球表面的特定兴趣区域可能需要数小时到数天。因此，近期的EO服务提供商利用EO卫星星座来协同采集数据，并缩短对特定兴趣区域（AoI）的重访时间。例如，Planet公司[34]已经发射并部署了由三个星座（PlanetScope [2], RapidEye [3] 和 SkySat [4]）组成的452颗卫星，以协同执行EO任务。它们能够从多个视角和维度捕捉地球的活动，重访时间小于一天。

TLDR

一颗卫星重访地球表面的特定兴趣区域可能需要数小时到数天

T(2): from low-quality to high-quality space sensors. In the early 1980s, the spatial resolution of EO satellites was around 30 meters as on LandSat-4 [46]. With the development of techniques, as low as 30 centimeters of spatial resolution is available in state-of-the-art EO satellites. Even CubeSats in PlanetScope can achieve a spatial resolution of around 3 meters [36]. The spectral resolution also improved dramatically over the past few decades, as sensors were refined and more bands became available for study. Some state-of-the-art satellite sensors can now capture pictures with more than 1000 spectral bands [39].

趋势(2)：从低质量到高质量的空间传感器。 在20世纪80年代初，如LandSat-4 [46]卫星的空间分辨率约为30米。随着技术的发展，最先进的EO卫星已能提供低至30厘米的空间分辨率。即便是PlanetScope中的立方星（CubeSats）也能达到约3米的空间分辨率[36]。在过去几十年里，随着传感器的不断改进和可供研究的波段增多，光谱分辨率也得到了显著提升。一些先进的卫星传感器现在能够捕获包含超过1000个光谱波段的图像[39]。

TLDR

分辨率 高高滴

In parallel with the new trends above, ideally emerging EO systems are expected to satisfy two critical performance requirements simultaneously, as described below.

R(1): fast information delivery. It is expected that the EO data can be downlinked to the ground as soon as possible. This is especially important for time-sensitive EO tasks, e.g., requiring fresh information from a wildfire or rescue scene.

R(2): scalability. As EO service providers leverage EO constellations to serve earth surveillance, it is expected that an EO system can maintain acceptable availability and performance as the number of EO sources grows up even in urgent cases.

与上述新趋势并行，理想情况下，新兴的EO系统需要同时满足以下两个关键性能需求。

需求(1)：快速信息交付。 EO数据被期望能够尽快下行链路至地面。这对于时间敏感的EO任务尤为重要，例如需要获取野火或救援现场的实时信息。

需求(2)：可扩展性。 随着EO服务提供商利用EO星座服务于地球监视，EO系统被期望在EO源卫星数量增长时，即使在紧急情况下也能保持可接受的可用性和性能。

C. Related Works

Many recent efforts have been proposed to optimize the data delivery for EO systems, which can be concluded in three aspects as described below.

近期已有许多工作被提出以优化EO系统的数据交付，可以归纳为以下三个方面。

Ground station networks. Many existing EO systems are using ground station networks directly to downlink the EO data, like [10]. But they mainly use a limited number of big ground stations which are high-cost to deploy and maintain, inducing high latency to deliver EO data for large EO constellations. Recent research [48] is proposed to use low-cost ground stations which are distributed all over the world to offer low latency downlink by downloading the data successively and cooperatively. This approach reduces the download time by increasing the time that a satellite can communicate with a ground station in one pass. However, since EO satellites move at high velocity, the visible window can only last for several minutes. Once an EO satellite moves out of the transmission range of a ground station, it has to interrupt the transmission process, waiting for another available ground station to continue the download process. Thus, the primary limitation of downloading data by distributed ground station networks is that: ground stations are difficult to be deployed on oceans which occupy nearly 70% of our earth surface, causing intermittent download and increasing the download completion time. Moreover, to take such an unprecedented amount of data to the ground is undoubtedly expensive and long duration.

地面站网络。许多现有EO系统，如[10]，直接使用地面站网络来下行传输EO数据。但它们主要使用数量有限的大型地面站，这些地面站部署和维护成本高昂，导致为大型EO星座传输数据时产生高延迟。近期的研究[48]提出使用分布于世界各地的低成本地面站，通过连续和协同下载的方式来提供低延迟下行链路。这种方法通过增加卫星单次过境时与地面站的通信时长来减少下载时间。

然而，由于EO卫星高速移动，可见通信窗口只能持续几分钟。一旦EO卫星移出某个地面站的传输范围，就必须中断传输过程，等待下一个可用的地面站来继续下载。因此， 通过分布式地面站网络下载数据的主要局限 在于：地面站难以部署在占地球表面近70%的海洋上，这导致了间歇性下载并增加了下载完成时间。此外，将如此空前数量的数据传输到地面无疑是昂贵且耗时的。

Satellite networks. Another prevalent approach for EO data transmission is leveraging GEO satellite relay networks, such as the European Data Relay Satellite (EDRS) system [18] owned by ESA, and Tracking and Data Relay Satellite (TDRS) system [40] operated by NASA. The key idea behind this method is to use satellite relay in geostationary orbit to establish long-duration and reliable LEO-to-GEO-to-ground communication path to transfer data acquired by EO satellites. However, downloading data by GEO satellite relays is scalability-limited, and is difficult to support a number of satellites in the EO constellation. Specifically, only 2 userspacecrafts can be connected to a TDRS relay [40] at the same time due to the limited on-board weight available for high-speed laser communication components. In addition, it’s economically difficult to launch many GEO relays to support more LEO satellites, since the cost to manufacture and launch a GEO satellite with the laser communication components is extremely high, e.g. $544 million for one EDRS relay [18].

Recent broadband LEO mega-constellations like Starlink [41] and Kuiper [5] are gaining popularity, which consist of thousands of inter-connected satellites with laser intersatellite links (ISLs). These mega-constellations promise to offer capacities up to 20Gbps [8], and provide broadband Internet service with lower latency [16]. A collection of recent works [16], [17], [25], [42] have proposed to leverage ISLs in mega-constellations to establish space routes consisting of ISLs and ground-satellite links (GSLs) for low-latency, high-speed data transmission. These prior efforts on LEO satellite routing suggest another viable path to download big EO data from space: exploiting multi-hop satellite routes from the EO satellite to transfer data to ground destinations. However, directly exploiting broadband constellations to download EO data can inevitably impose significant challenges for satellite systems. Continuously activating download links for highvolume EO data involves high energy consumption, which further requires to increase the size of battery or solar panel, involving big challenges on satisfying the stringent constraints on the mass, volume and cost of satellites.

卫星网络。另一种流行的EO数据传输方法是利用GEO卫星中继网络，例如欧洲航天局（ESA）拥有的欧洲数据中继卫星（EDRS）系统[18]，以及美国宇航局（NASA）运营的跟踪与数据中继卫星（TDRS）系统[40]。该方法的核心思想是利用地球静止轨道上的卫星中继，建立长时程、可靠的LEO-to-GEO-to-ground通信路径来传输EO卫星获取的数据。然而， 通过GEO卫星中继下载数据存在可扩展性限制，难以支持EO星座中的大量卫星 。具体而言，由于高速激光通信组件的星上重量有限，一个TDRS中继[40]同时只能连接2个用户航天器。此外，发射大量GEO中继来支持更多LEO卫星在经济上是困难的，因为制造和发射一颗带有激光通信组件的GEO卫星成本极高，例如一颗EDRS中继的成本高达5.44亿美元[18]。

近期的宽带LEO巨型星座，如Starlink [41]和Kuiper [5]，正日益受到关注，它们由数千颗通过激光星间链路（ISL）互连的卫星组成。这些巨型星座承诺提供高达20Gbps的容量[8]，并提供更低延迟的宽带互联网服务[16]。近期一系列工作[16], [17], [25], [42]已提出利用巨型星座中的ISL来建立由ISL和地-星链路（GSL）组成的空间路由，以实现低延迟、高速的数据传输。这些关于LEO卫星路由的先前工作为从太空下载海量EO数据提供了另一条可行路径：利用从EO卫星到地面目的地的多跳卫星路由来传输数据。然而，直接利用宽带星座下载EO数据不可避免地会给卫星系统带来重大挑战。为海量EO数据持续激活下载链路会涉及高昂的能耗，这又要求增加电池或太阳能帆板的尺寸，从而给满足卫星在质量、体积和成本上的严格约束带来了巨大挑战。

In-orbit data filtering. To decrease the amount of data that should be downloaded to the ground, orbital edge computing (OEC) [9] was proposed, which exploits improved onboard computing resources together with deep learning to filter out the valuable parts from the raw EO data. However, the “valuable” parts are hard to define, for different EO missions, the valuable items differ. For forest monitoring, the pictures of the sea may be useless but not in maritime surveillance cases. Other works like [15] also propose to transfer the processing process to the satellite edges to reduce the bandwidth consumption to accelerate the transmission process by leveraging emerging commercial off-the-shelf (COTS) system-on-chip (SoC) technologies. However, other than the limitations above, it can be also time-consuming to wait for visible ground stations since they don’t use the LEO satellite networks.

在轨数据过滤。为减少应下载到地面的数据量，轨道边缘计算（OEC）[9]被提出，该技术 利用改进的星上计算资源和深度学习来从原始EO数据中过滤出有价值的部分。然而，“有价值”的部分很难定义，对于不同的EO任务，有价值的内容也不同。

对于森林监测，海洋的图片可能无用，但在海事监控场景中则不然。其他工作如[15]也提出利用新兴的商用现成（COTS）片上系统（SoC）技术，将处理过程转移到卫星边缘，以减少带宽消耗，从而加速传输过程。但是，除了上述局限外，由于它们不使用LEO卫星网络，等待可见地面站的过程也可能非常耗时。

Summarily, prior solutions for EO delivery, are either efficiency-limited (like ground station networks) since they suffer from long delivery completion time due to the intermittent space-ground connectivity, or scalability-limited (like satellite networks), since they are unable to guarantee good delivery performance when the EO constellation size scales up. The state quo thus motivates us to explore a new solution to accomplish fast and scalable data delivery for EO constellations.

总结而言，先前的EO数据传输解决方案:

1. 要么是效率受限的（如地面站网络），因为它们因间歇性的空地连接而导致传输完成时间过长
2. 要么是可扩展性受限的（如卫星网络），因为它们在EO星座规模扩大时无法保证良好的交付性能

这一现状因此激励我们探索一种新的解决方案，以实现EO星座的快速和可扩展的数据交付