Related Work

The problem of handover in satellite network has been extensively discussed. Considering the satellite network’s topology is a natural approach [35], [36]. One mainstream method is the virtual node method [35], in which partition mappings satellites and ground stations into geographic region. However, this method only works in polar orbit constellation networks and is not applicable to mainstream constellation configurations, such as the Walker constellation primarily used by Starlink.

Another mainstream method is based on calculating all possible network topology states to predict the network topology at any given moment [36]. However, this method’s computational complexity and storage requirements increase exponentially with the number of satellites and users, making it impractical for large-scale satellite networks.Another approach is modifying the network structure on the base of standard 5G network [8], [34], [37], [38]. This includes introducing new NFs and leveraging advanced 5G handover procedures. However, most of these methods do not specifically prioritize handover latency optimization or provide sufficient improvements in reducing latency. A recent work [8] introduced stateless network elements on the satellite side, allowing users to store the information required by core network, thus avoiding interactions with core network during handover switches. However, this work primarily focused on the core network and neglected RAN, which could lead to an underestimation of the overall latency in handover procedure.

Our proposed scheme involves predicting the network topology (i.e., the user’s access satellite). However, there is no need to compute the global topology; instead, each UPF calculate the users it is responsible for respectively, significantly reducing the computational load. On the other hand, we modifying the handover procedure. By completely eliminating the interaction with core network during handover, we achieve a substantial reduction in latency, surpassing existing methods. Our solution takes both core network and RAN into consideration, and we conducted comprehensive systemlevel experiments on a latency platform. This ensures that experimental results are close to real-world scenarios.

卫星网络中的切换问题已被广泛讨论。 一种思路是考虑卫星网络的拓扑结构 [35], [36]。其中一个主流方法是虚拟节点法[35]，该方法通过划分将卫星和地面站映射到地理区域中。然而，此方法仅适用于极地轨道星座网络，而无法应用于主流的星座配置，例如星链（Starlink）主要使用的沃尔克（Walker）星座。另一个主流方法则基于计算所有可能的网络拓扑状态，来预测任意时刻的网络拓扑[36]。然而，该方法的计算复杂度和存储需求随卫星与用户数量呈指数级增长，使其在大型卫星网络中不具备实用性。

另一种思路是在标准5G网络的基础上修改网络架构 [8], [34], [37], [38]。这包括引入新的网络功能（NFs）以及利用更先进的5G切换流程。然而，这些方法中的大多数并未专门针对切换延迟进行优化，或在降低延迟方面未能提供显著的改进。近期一项工作[8]在卫星侧引入了无状态网络元素，允许用户存储核心网所需的信息，从而在切换时避免与核心网的交互。但是，该工作 主要聚焦于核心网而忽略了无线接入网（RAN） ，这可能导致对整体切换流程延迟的低估。

我们提出的方案同样涉及到预测网络拓扑（即用户所接入的卫星）。然而，我们无需计算全局拓扑；取而代之的是，每个用户平面功能（UPF）实体仅分别计算其所负责的用户，从而显著降低了计算负荷。另一方面，我们修改了切换流程。通过在切换过程中完全消除与核心网的交互，我们实现了延迟的大幅降低，其效果超越了现有方法。我们的解决方案同时兼顾了核心网与无线接入网，并在一个延迟仿真平台上进行了全面的系统级实验。这确保了实验结果能够贴近真实世界的场景。