基于缓存的卫星传输性能优化策略

Performance optimization strategies for satellite transmission based on caching

摘要:

传统卫星透明转发模式在新型星地网络下会导致巨大的时延。为提高对地面用户的服务能力，本文在低轨卫星和地面内容服务器上设置缓存，并联合低轨卫星、地面内容服务器以及地面用户在同频干扰下，针对星地网络的传输吞吐问题进行建模。使用改进的粒子群(PSO)算法和支持超平面优化(SHOT)求解器求解。实验表明，对目标优化问题求解时，SHOT求解器获得的曲线较为平滑，优化的PSO算法相比SHOT求解器得到的结果，有2%上下的波动。在信干噪比阈值为8 dB和最大功率20 W，以及卫星或地面内容服务器干扰数量分别为2和3的约束下，可分别求得45 Mbps和37 Mbps的系统吞吐量。在分别设置低轨卫星、地面内容服务器以及地面用户数量的情况下，本文对两种算法的运行时间进行了分析，最坏情况下设置3类数量为4时，优化PSO解法比SHOT求解器快约10 s。结果表明，启发式算法更适用于未来复杂星地网络建模，更适合寻找非凸非线性全局解，并在高维问题下求解速度表现更优。

Abstract:

The traditional satellite transparent relay mode causes significant delays in the new satellite-ground network. In order to improve the service capability for ground users, this paper proposes a caching scheme on both the Low-Earth Orbit(LEO) satellite and the ground content server, and models the transmission throughput problem of the satellite-ground network considering co-frequency interference among LEO satellites, ground content servers, and ground users. The improved Particle Swarm Optimization(PSO) algorithm and the Supporting Hyperplane Optimization Toolkit(SHOT) solver are employed to solve the problem. The experimental results reveal that the SHOT solver can obtain a smoother result curve, and there is only about 2% fluctuation compared to the optimization PSO algorithm and the SHOT solver. Under the constraint of an 8 dB signal-to-noise ratio threshold and a maximum power of 20 W, the proposed approach can achieve a transmission rate of 45 Mbps and 37 Mbps, respectively, with interference from 2 satellites or 3 ground content servers. Furthermore, the study analyzes the running time of the two algorithms for various numbers of low-earth orbit satellites, ground content servers, and ground users. The optimization PSO algorithm is approximately 10 s faster than the SHOT solver, particularly when the number of the three categories is set to 4, in the worst-case scenario. The results indicate that heuristic algorithms are more suitable for modeling future complex satellite-ground networks and finding non-convex and nonlinear global solutions. They also perform better in terms of solving speed for high-dimensional problems.