临近空间飞行器通信中频谱的快速预测与精准感知技术研究

There is a rapid growth in the number of near space vehicles. It is reported that in the US, the number of high-altitude unmanned aircrafts grew by nearly 50% in the past 5 years. Therefore the research of near space is a hot topic. However, the near space communication has not yet been assigned licensed frequency bands. How the users access to unlicensed bands is a primary problem of communication. Because the service of licensed users includes satellite/air navigation and rescue, the communication of licensed users can not be jammed. Moreover, the occupancy rate of spectrum is very high. Therefore the near space users must first sense (find spectrum holes), and then access. For supersonic users in near space, the spectrum sensing methods need to meet the requirements of high dynamic communication effectively, while the traditional sensing approaches scan frequency bands based on ergodic/random methods and cost too much time. This project aims to improve the speed and accuracy of the sensing procedure by predicting the possible hole before sensing, based on a new and fast machine learning method. Multi-dimensional and idle duration based hidden Markov models are proposed to rapidly predict spectrum holes. In addition, existing fast sensing methods are difficult to detect signals for low SNR. The project will optimize the system estimation based on convex optimization theory to improve the accuracy of fast sensing. The project has important significance for the research of spectrum sharing and unauthorized access, in the application of spatial high dynamic communication.

临近空间飞行器数量具有快速增长趋势，据报道美国近5年仅高空无人机数量就增长近50%，是空间开发的热点之一。但该空域通信尚未被分配授权频段，用户如何接入非授权频段是通信的首要问题。由于授权用户业务的特殊性（如卫星/航空的导航与救援），其通信不可被干扰，且频谱占用率高，临近空间用户需先感知（找到频谱空穴）、再进行非授权接入。针对超音速用户，频谱感知方法需有效应对高动态的通信环境，而传统感知需遍历/随机扫描频段，耗时过长。本项目拟在感知前先通过新型的、快速的机器学习方法，预测可能的空穴，提高感知速度及准确度。特点是提出基于时空多维的、空闲时长的隐马尔科夫模型，快速预测频谱空穴。此外，针对低信噪比的授权用户信号，现有快速感知方法难以检测。本项目将基于凸优化理论优化系统估计量，保证感知快速的同时提高检测准确性。本项目对于空间高动态通信背景下的频谱共享研究与非授权接入应用具有重要意义。

无人飞行器无论是临近空间还是低空飞行器，近年来均快速增长。其通信业务量增多，对频谱资源需求持续上升。已分配的授权频段大多与ISM、Wi-Fi、蓝牙等通信频段共享，远远不足无人机要求，无人机需求以次用户身份机会接入非授权频段进行通信。由于授权用户业务的高优先级（如卫星/航空的导航与救援），其通信不可被干扰，无人机需准确预测、感知频谱空穴。针对无人机的高动态性，本项目研究了在运动中进行无历史学习数据的频谱状态预测技术与频谱状态持续时长预测技术。前者运用同伦理论，研究了对下一位置历史数据的线性同伦估计，并基于隐马尔可夫模型进行频谱状态预测。后者进一步分析了频谱状态分布，采用非线性同伦法估计下一位置的时长参数，基于非平稳隐马尔可夫模型预测频谱状态持续时长。这两项研究内容主要是运用同伦理论估计无人机通信非独立、非平稳信道的历史数据，解决了无人机在飞行中因缺乏下一位置的历史数据而无法预测的问题。针对授权用户信号在低信噪比时仍需准确感知的要求，本项目提出了基于优化噪声估计量的频谱感知技术与基于恒漏检率的信号检测技术。前者考虑不同频段之间并非独立同分布的问题，运用凸优化的线性规划对噪声估计量进行优化、校准，大大改善了噪声不确定度带来的影响，提高了感知性能。后者为本项目进行中发现的新问题，通过重新定义信号检测基准方法，提出基于恒漏检率的信号检测法，替代传统的基于恒虚警率检测法，可达到不同信噪比下恒定信号检测概率的目的，提高了在低信噪比下对主用户信号的感知性能。这两项研究内容基于简单快速的能量检测法，解决了无人机在高速运动中需求既快速又准确的感知要求。本项目四个研究内容均采集了实测数据进行仿真实验，验证了四项技术的有效性。本项目产生成果共14项，其中期刊论文7篇（SCI检索6篇，EI检索1篇），国际会议论文5篇（EI已检索4篇），申请专利2项，毕业硕士生及本科生5名。