基于分布式深度学习的多星计算卸载策略

在边缘计算增强的低轨卫星网络场景下,低轨卫星集群协同处理地面任务能有效降低用户响应时延。对卫星集群的联合卸载决策和资源分配优化问题进行研究,将其描述为一个混合整数规划问题,并采用了一种基于分布式深度学习算法的卫星边缘计算卸载算法（deep learning based offloading algorithm,DLOA）。该算法使用多个并行DNN用于生成卸载决策并采用经验回放存储新生成的卸载决策,当采用隐藏层结构不同的DNN,收敛速度比同构DNN提升18%,收敛值与最优值的比值基本为1,可以认为已收敛至最优。此外,探讨了DNN的数量对所使用的算法的影响,仿真结果表明采用少量DNN就可以获得近优的收敛效果。通过对不同任务规模下采用不同算法的任务完成率进行研究,结果表明DLOA算法可通过采用异构DNN和优化资源分配方案显著提升完成率,其较单星运算方案任务完成率提升1倍,较二进制粒子群算法方案提升20%。     

一种改进的LEO卫星网络路由策略

文章结合LEO卫星网络的拓扑结构和运行规律,提出了一种时延受限的最小耗费（DCLC）路由算法。仿真结果显示,相对于最短路由算法,该算法具有较低的呼叫阻塞率和切换失败率,以及更好的服务质量。同时能够平衡各链路负载,使整个卫星网络更稳定。     

低树代价的低轨卫星网络组播算法

为了解决低轨卫星网络中现有典型源组播算法的信道资源浪费问题,提出了一套单核共享树组播算法,即核心群合并共享树(CCST)和加权核心群合并共享树(w-CCST)算法.CCST算法包括动态近似中心(DAC)选核方法和核心群合并组播路径构建方法.在核心群合并方法中,以核节点作为初始核心群,通过核心群和剩余组成员的最短路径方法逐步扩展直至整棵组播树构建完成,从而大大提高了网络带宽利用率.在w-CCST算法中,通过调整加权因子可以适度增大树代价、降低端到端传播时延,以支持某些时延性能要求严格的实时业务.仿真结果说明,CCST算法的树代价性能比其它算法有显著改善,w-CCST算法的端到端传播时延性能好于CCST算法.      

Single event upset investigations on the “Flying Laptop” satellite mission

The radiation effects in electronic parts are called single-event effects, which are deemed to be critical for space missions. This paper presents the Single Event Upsets that were observed in an onboard memory device of the Low Earth Orbit “Flying Laptop” satellite mission during its in-orbit operation. The Single Event Upsets were carefully mapped on the satellite orbital space itself and their root causes were investigated together with their rates of occurrence. Subsequently, the events were traced to show several root cause sources such as (i) trapped energetic protons leaking to low altitudes within the South Atlantic Anomaly, (ii) Solar Energetic Particles emitted by an impulsive event on 10 September 2017, and (iii) Galactic Cosmic Rays. A profound analysis was carried out on the observed flight data, and its corresponding results are actually in agreement with the standard energetic particle models. The presented results provide another important insight on the Single Event Upsets for future Low Earth Orbit satellite missions.     

UMR: A utility-maximizing routing algorithm for delay-sensitive service in LEO satellite networks

This paper develops a routing algorithm for delay-sensitive packet transmission in a low earth orbit multi-hop satellite network consists of micro-satellites. The micro-satellite low earth orbit(MS-LEO) network endures unstable link connection and frequent link congestion due to the uneven user distribution and the link capacity variations. The proposed routing algorithm,referred to as the utility maximizing routing(UMR) algorithm, improve the network utility of the MS-LEO network for carrying flows with strict end-to-end delay bound requirement. In UMR, first, a link state parameter is defined to capture the link reliability on continuing to keep the end-to-end delay into constraint; then, on the basis of this parameter, a routing metric is formulated and a routing scheme is designed for balancing the reliability in delay bound guarantee among paths and building a path maximizing the network utility expectation. While the UMR algorithm has many advantages, it may result in a higher blocking rate of new calls. This phenomenon is discussed and a weight factor is introduced into UMR to provide a flexible performance option for network operator. A set of simulations are conducted to verify the good performance of UMR, in terms of balancing the traffic distribution on inter-satellite links, reducing the flow interruption rate,and improving the network utility.     

Impact of covariance information of kinematic positions on orbit reconstruction and gravity field recovery

Gravity missions are equipped with onboard Global Positioning System (GPS) receivers for precise orbit determination (POD) and for the extraction of the long wavelength part of the Earth’s gravity field. As positions of low Earth orbiters (LEOs) may be determined from GPS measurements at each observation epoch by geometric means only, it is attractive to derive such kinematic positions in a first step and to use them in a second step as pseudo-observations for gravity field determination. The drawback of not directly using the original GPS measurements is, however, that kinematic positions are correlated due to the ambiguities in the GPS carrier phase observations, which in principle requires covariance information be taken into account. We use GRACE data to show that dynamic or reduced-dynamic orbit parameters are not optimally reconstructed from kinematic positions when only taking epoch-wise covariance information into account, but that essentially the same orbit quality can be achieved as when directly using the GPS measurements, if correlations in time are taken into account over sufficiently long intervals. For orbit reconstruction covariances have to be considered up to one revolution period to avoid ambiguity-induced variations of kinematic positions being erroneously interpreted as orbital variations. For gravity field recovery the advantage is, however, not very pronounced.     

原子氧和石墨在近地轨道环境中反应概率的数值估算

本文提出估算近地轨道环境中原子氧对高度规则材料的反应概率的分子动力学模型。该模型通过对所导出的随机微分方程组的数值求解估算出原子氧与材料表面相互作用的经典轨迹系统，并根据所提出的反应性与非反应性碰撞准则估算出相互作用的反应概率。作为算例，本文给出了原子戌石墨材料表面的反应概率数值计算并与航天飞机ＳＴＳ－４６所测出的实际反应概率进行了对比，表明所建立的模型在一定程度上中可靠的。     

低轨星载粒子探测器布局优化研究

基于卫星工具包(STK)、MATLAB软件和国际地磁参考场(IGRF)模型,对粒子探测器在低轨卫星任务中的投掷角测量范围进行了仿真,并给出了其探测器的星上布局优化方法。研究结果表明:粒子探测器的星上布局选择与任务的探测目标相关。如果探测目标是束缚在磁场中作弹跳运动的粒子,应将探测器在朝向卫星轨道坐标系±Y轴的基础上向卫星的前进方向偏转;如果探测目标是处于损失锥中的沉降粒子,应将探测器在朝向卫星轨道坐标系-Y轴的基础上向天顶方向偏转;为实现全投掷角探测范围的探测,要使用多个探测器通过构型布局优化进行联合探测。文章提出的投掷角测量范围分析与优化方法,可推广到多种轨道高度的粒子探测卫星总体设计中,为提升卫星的总体分析设计水平提供支持。     

一种GEO/LEO双层卫星网络路由算法及仿真研究

时间虚拟化策略的时间片过短将影响路由性能,提出一种基于GEO/LEO双层卫星网络的时间虚拟化和分层管理的路由算法（VLRA）,对时间虚拟化策略进行合理改进,通过新的分层管理思想,在每个时间片开始进行路由计算和更新。分析和仿真验证表明,改进的时间虚拟化策略得到的时间片更合理,利于网络的路由和通信。GEO/LEO双层卫星网络下VLRA算法时延稳定,并具有较好的丢包率和吞吐量性能。     

路由算法在LEO通信星座构型优化中的应用

低轨(LEO,Low Earth Orbit)通信星座的特殊功能决定了其构型优化需要综合考虑成本、覆盖和路由性能等因素.以往的研究往往忽略了路由性能,导致优化得到的构型不能很好的满足要求.首先给出了两个意义明确的星座路由性能的评价指标,然后通过将它们作为目标函数进行仿真,验证了其优化必要性和评价有效性.最后利用多目标遗传算法实现了星座的构型优化.仿真结果显示,将路由算法加入到星座的构型优化中,在仅对星座构型参数进行微调的前提下,不仅满足了成本、覆盖等基本要求,而且星座的路由性能大大提高.