Limitations for Classic Sat-Ground Interconnection Methodology

A. Classic methodologies for satellite-ground interconnection

Existing methodologies for satellite-ground interconnection, i.e., determining how a GS should select and connect to an access satellite, can typically be classified into three categories, based on their criteria for access satellite selection: (i) connecting to the satellite with longest remaining service time until it moves out of the transmission range (denoted as LRST); (ii) connecting to the nearest satellite until it disappears out of the transmission range (denoted as NSD); and (iii) always connecting to the nearest satellite, i.e., immediately handovering to a closer satellite if any (denoted as NSH). The above approaches are widely used in the early design of SNs [14], [32], [40], [41], [49].

现有的星地互联方法，即确定一个地面站（GS）应如何选择并连接到一个接入卫星，通常可根据其接入卫星的选择标准分为三类：

（i）连接到剩余服务时间最长的卫星，直至其飞出传输范围（记为LRST）

（ii）连接到最近的卫星，直至其消失在传输范围之外（记为NSD）

（iii）始终连接到最近的卫星，即一旦有更近的卫星出现便立即切换（记为NSH）

上述方法在早期的卫星网络（SNs）设计中被广泛使用 [14], [32], [40], [41], [49]

B. Why the above methodologies are limited in emerging SNs?

Emerging LEO SNs involve three new characteristics as compared to their predecessors.

• Intermittent satellite-ground connectivity. LEO satellites have the inherent “high mobility” property that differs greatly from the terrestrial nodes and GEO satellites. The satellite-ground topology is no longer static and can change frequently, making the interconnection more complex.

• Many visible LEO satellites. The scale of LEO constellation has experienced explosive growth, from less than one hundred satellites to thousands of satellites. The number of visible satellites in the GS has risen to dozens in low and middle latitudes [12], [19]. GSes must choose which satellite to connect to and larger constellation scale has greatly expanded the satellite-ground topology design space.

• Multi-hop routing requirements in SNs. In ISTN, satellites can store and transmit data, participating in routing. Therefore, satellite-ground interconnection will not only affect one hop between satellites and GS, but also have an impact on the performance of the entire network.

与前辈相比，新兴的低轨卫星网络呈现出三个新特征:

• 间歇性的星地连接: 低轨卫星具有固有的“高移动性”属性，这与地面节点和地球静止轨道（GEO）卫星截然不同。星地拓扑不再是静态的，而是会频繁变化，这使得互联变得更为复杂

• 大量可见的低轨卫星: 低轨星座的规模经历了从不足百颗到数千颗的爆炸式增长。在中低纬度地区，地面站的可见卫星数量已上升至数十颗 [12], [19]。地面站必须选择连接哪颗卫星，而更大规模的星座极大地扩展了星地拓扑的设计空间

• 卫星网络中的多跳路由需求: 在天地一体化网络（ISTN）中，卫星能够存储和转发数据，并参与路由计算。因此，星地互联将不仅仅影响卫星与地面站之间的一跳，更会对整个网络的性能产生影响

C. Quantifying the impact of satellite-ground topology on the routing performance over emerging SNs

To further quantify and understand the impact of satelliteground topology on the routing performance over emerging SNs, we carry out experiments under different constellations and GSes, using Starperf [34] to simulate these constellations. We select Starlink’s first shell, composed of 1584 satellites and Kuiper’s first shell, composed of 1156 satellites. Further, we use +Grid [13] to interconnect LEO satellites, and use the shortest path (minimizing number of hops) routing algorithm to forward data packets. We choose the GSes from the existing GSes of AWS [4]. The experiment simulates an orbital period of about 6000 seconds for each constellation.

为了进一步量化和理解星地拓扑对新兴卫星网络路由性能的影响，我们利用Starperf [34] 模拟器，在不同星座和地面站配置下进行了实验。我们选择了由1584颗卫星组成的Starlink首个壳层和由1156颗卫星组成的Kuiper首个壳层。此外，我们使用+Grid [13]模型来构建低轨卫星间的互联，并采用最短路径（最小化跳数）路由算法来转发数据包。我们从亚马逊云服务（AWS）[4]的现有地面站中选择实验站点。实验模拟了每个星座约6000秒的一个轨道周期。

1) Performance metrics: We examine following performance metrics in the networking stack to quantify the impact of different underlying topology options.

• Network convergence/reachability. The change of topology can lead to routing recalculation, which can further lead to routing instability and increase the overhead of routing protocols. The routing instability can be reflected in network convergence time [10] and has a direct impact on network reachability. The overhead of routing protocol includes the routing calculation overhead and the communication overhead caused by the transmission of routing control information. The former puts a load on the processor, and the latter takes up valuable data bandwidth.

• Latency and jitter quantify how emerging SNs can facilitate long-haul, low-latency Internet services for terrestrial users. To concentrate on the impact of topology options, we refrain the effect of other factors like link congestion caused by overloaded traffic, and we denote the latency as the one-way propagation delay between two network nodes.

1. 性能指标： 我们考察了网络协议栈中的以下性能指标，以量化不同底层拓扑选项的影响

• 网络收敛性/可达性。 拓扑的改变会导致路由重算，这可能进一步引发路由不稳定并增加路由协议的开销。路由不稳定性可以体现在网络收敛时间[10]上，并直接影响网络可达性。路由协议的开销包括路由计算开销和由路由控制信息传输引起的通信开销。前者给处理器带来负载，后者则占用了宝贵的数据带宽。

• 时延与抖动。 这两个指标量化了新兴卫星网络如何为地面用户提供长距离、低时延的互联网服务。为专注于拓扑选项的影响，我们规避了由流量过载引起的链路拥塞等其他因素的影响，并将时延定义为两个网络节点间的单向传播延迟。

networking metrics

1. 网络收敛时间 / 可达性
2. 时延 (one-way) + jitter

2) Observations: Summarily we make three key observations as described below.

High satellite-ground fluctuation caused by satellite dynamicity. Figure 2 plots the satellite-ground fluctuation under three different interconnection schemes. Handover interval, used to describe the degree of satellite-ground fluctuation, is defined as the time interval between two adjacent handover of satellites for a GS or multiple GSes. More specifically, we calculate the handover interval for a single GS (i.e., Ireland) and double GSes (i.e., Oregon and Sydney). First, as LEO satellites orbit in high velocity, the handover interval for a GS ranges from few seconds to about 300 seconds.

由卫星动态性引起的高星地波动。图2描绘了三种不同互联方案下的星地波动情况。 “切换间隔”（Handover interval）被用来描述星地波动的程度，其定义为单个或多个地面站两次相邻卫星切换之间的时间间隔。

具体来说，我们计算了单个地面站（爱尔兰）和两个地面站（俄勒冈和悉尼）的切换间隔。首先，由于低轨卫星高速运行，单个地面站的切换间隔从几秒到约300秒不等。

Further, the handover interval increases rapidly for double GSes, which inevitably has a negative impact on their end-to-end communication. Second, the LRST scheme attains longer handover intervals compared with other schemes, as it keeps using the same satellite until it moves out of the transmission range to avoid frequent handovers. Third, we observe that the constellation pattern can also affect the handover interval, as the satellite-ground interconnection changes less frequently under Kuiper. This is because GSLs can persist longer under Kuiper, as the coverage of a satellite under Kuiper is wider for its higher height and lower minimum elevation angle.

此外，对于双地面站场景，切换间隔迅速增加，这不可避免地对其端到端通信产生负面影响。其次，LRST方案获得了比其他方案更长的切换间隔，因为它会持续使用同一颗卫星直至其飞出传输范围，从而避免了频繁切换。第三，我们观察到星座构型也会影响切换间隔，因为在Kuiper星座下星地互联的变化频率较低。这是由于Kuiper卫星的轨道更高、最小仰角更低，其单星覆盖范围更广，使得地星链路可以维持更长时间。

实验现象

(1) 双地面站场景，切换间隔迅速增加 ?为什么?

????

(2) LRST方案获得了比其他方案更长的切换间隔 => 最不容易发生切换

因为LRST的策略是选择“最大可服务时长者”, 它会持续使用同一颗卫星直至其飞出传输范围, 从而避免频繁切换

(3) 星座构型影响切换间隔

Kuiper 的切换频率更低, 因为它轨道更高, 单星可覆盖范围更广

Impact of satellite-ground interconnection on network reachability. The satellite-ground instability exposed above may inevitably lead to routing churn and trigger routing reconvergence which affects end-to-end network reachability. Figure 3 depicts the end-to-end packet delivery ratio between representative geo-distributed GSes under different handover intervals. We run the OSPF protocol which exploits the Dijkstra algorithm to construct the shortest path for packet forwarding, and we adjust the HELLO INTERVAL in OSPF to tune the frequency of probing link states. As shown in Figure 3, we observe that the satellite-ground instability has a significant impact on the reachability in SNs. Frequent connectivity changes cause poor network reachability, e.g., the packet delivery ratio is no more than 54% if the handover interval is about 25 seconds. Therefore, the network reachability is affected by the interconnection schemes, which lead to different handover intervals. Further, we observe that the reachability is also affected by the probing interval of the routing protocol. This is because shorter probing intervals can shrink route discovery time, thereby speeding up network convergence time and increasing network reachability.

上文揭示的星地不稳定性可能不可避免地导致路由抖动（routing churn）并触发路由重敛，从而影响端到端网络可达性。图3描述了在不同切换间隔下，代表性的地理分布式地面站之间的端到端分组投递率。我们运行OSPF协议，该协议利用Dijkstra算法构建最短路径以进行数据包转发，并通过调整OSPF中的HELLO间隔来调节探测链路状态的频率。如图3所示，我们观察到星地不稳定性对卫星网络的可达性有显著影响。频繁的连接变化导致了较差的网络可达性，例如，如果切换间隔约为25秒，分组投递率将不高于54%。因此，网络可达性受互联方案的影响，因为不同方案导致了不同的切换间隔。此外，我们观察到可达性也受路由协议探测间隔的影响。这是因为更短的探测间隔可以缩短路由发现时间，从而加速网络收敛，提高网络可达性。

Impact of satellite-ground interconnection on end-to-end latency and jitter. Figure 4 plots the latency between representative GSes under various satellite-ground interconnection schemes. First, we observe that for each scheme the latency suffers from high variations, from 40ms to about 120ms. Through our in-depth analysis, we identify that such high variation is caused by not only the inter-satellite fluctuation, but also the suboptimal selection of access satellite. Figure 5 illustrates an example explaining the high latency variation caused by the handover of access satellite. When the Sydney GS switches its access satellite, the shortest path to the Oregon GS changes dramatically and their end-to-end latency also changes from only 49.93 milliseconds to 113.51 milliseconds. This phenomenon is common in all three schemes. Second, we find that the achievable latency is jointly affected by both the interconnection scheme as well as the location of source/destination GSes. The change of the access satellite at either end has an impact on their path and latency.

图4描绘了在各种星地互联方案下，代表性地面站之间的时延。首先，我们观察到每种方案下的时延都经历了剧烈波动，从40毫秒到约120毫秒。通过深入分析，我们发现如此剧烈的波动不仅由星间链路的波动引起，也源于接入卫星的次优选择。

图5用一个例子解释了由接入卫星切换引起的高时延波动。当悉尼地面站切换其接入卫星时，其到俄勒冈地面站的最短路径发生巨大变化，端到端时延也从仅49.93毫秒变为113.51毫秒。这一现象在所有三种方案中都很普遍。其次，我们发现可达时延受到互联方案以及源/目的地面站位置的共同影响。任意一端的接入卫星变化都会对其路径和时延产生影响。

3) Takeaways: Collectively, we identify that:

(i) although LRST brings longer handover interval and higher network reachability, compared with NSH and NSD, it only reduces the interval from the perspective of a single GS;

(ii) access satellite of different GSes will jointly affect end-to-end latency and jitter;

(iii) none of the existing satellite-ground interconnection schemes take stable routing and low-latency communication into consideration.

---

(i) 尽管与NSH和NSD相比， LRST带来了更长的切换间隔和更高的网络可达性，但它仅仅是从单个地面站的角度减少了切换频率

(ii) 不同地面站的接入卫星将共同影响端到端的时延和抖动

(iii) 现有的星地互联方案均未将路由稳定和低时延通信纳入考量