无线通信人工智能抗干扰关键技术研究

The revolutions in interference suppression theories and techniques have been driven by the mutual contradiction between the hostile transmission of military radio signals, the greedy use of the civilian radio spectrum, and the limitation of the electromagnetic wave in space-time-frequency dimensions. Artificial intelligence will become another milestone after the frequency-hopping spread spectrum, multi-antenna, transform domain, and cooperative cognition techniques in the development of interference suppression theories..In this project, the logic probability space representation of the interference information is first characterized, which will reveal the intrinsic relationship between the multi-source multi-type interference signals and the finite-dimensional probability representation. Then, the detection model of interference signals is established with the logic probability characterization. After that, the principles of the interference-pattern reasoning are explored by using the utility theory, from which the design procedures of sequential decision-making for fast matching the interference with its characterization is given. Lastly, the relationship between the proposed non-linear layered filtering network and the performance of the interference suppression is studied under the complex interference environment with multi-source multi-type nonlinear interference signals. In the end, the artificial-intelligence-based interference-suppression algorithmic framework is established, which is capable of the weights, connections and feedbacks among the network nodes..This project will efficiently expedite the application of the artificial intelligence in practical interference suppression systems. The outcomes of this project will bring important theoretical value and economic benefits to the communication devices/systems designs and applications, such as the mobile communication network, broadband wireless access network, military data link, and electromagnetic compatibility.

军事无线电信号的敌意发射、民用无线电频谱的贪婪使用与电磁波空时频三维度受限之间的矛盾，驱动着抗干扰理论与技术的变革；人工智能将是继扩跳频、多天线、变换域、认知协同之后，抗干扰思想发展过程中的重要里程碑。. 项目重点研究干扰信息知识的逻辑概率空间表征机理，分析多源多样式干扰信号与有限维度概率表征的内在关系，建立包含逻辑概率特征的干扰信号检测模型；探索干扰信号模式的效用推理规划准则，给出干扰信号快速匹配的序列决策设计方法；研究多源非线性复杂干扰环境下，非线性分层滤波网络与干扰信号抑制性能之间的关系，构造人工智能抗干扰算法的基本架构，实现滤波网络节点权值学习、连接、反馈的自主更新。. 人工智能抗干扰理论与技术的进步，将会揭示复杂电磁环境干扰与抗干扰手段互作用的深层机理。相关研究成果，对移动通信网、军用数据链、电磁兼容等无线电系统的设计和应用，具有理论与指导意义。

项目针对无线通信典型干扰场景，采用智能信号处理思想，开展了如下4个方面的研究工作，共发表论文25篇，其中SCI检索论文18篇，申请发明专利6项(4项已经授权)，完成了全部研究计划。. (1) 研究内容一：干扰知识的逻辑概率空间表征机理.针对多源复合干扰场景，提取和分析时变干扰的概率空间特征，研究了干扰信息知识的逻辑概率空间表征机理，分析了多源多样式干扰信号与有限维度概率表征的内在关系，建立包含逻辑概率特征的干扰信号检测模型。.侧重研究点包括：分析多源时变干扰信号的确定性和随机参数，研究干扰时变特性下知识关联方法，建立干扰信号的数学概率空间检测模型。. (2) 研究内容二：干扰模式匹配的快速推理策略.根据已经建立的干扰信号数学概率空间表征方法，研究了干扰模式匹配的快速推理策略。.侧重研究点包括：分析了历史干扰信号样本与现时干扰的匹配关系，探索干扰信号模式的效用推理规划准则，给出了干扰信号快速匹配的序列决策设计方法。. (3) 研究内容三：深度学习的多维关联抗干扰算法.根据干扰概率表征和模式匹配结果，研究了多维关联的非线性抗干扰算法。.侧重研究点包括：非线性同时同频自干扰环境下，研究了强化学习非线性分层滤波网络与干扰信号抑制性能之间的关系，实现滤波网络节点权值学习、连接、反馈的自主更新。. (4) 研究内容四：无线通信人工智能抗干扰实验验证.利用前期积累的可重构硬件实验平台，初步实验验证非线性干扰下的人工智能抗干扰算法。.侧重研究点包括：利用可重构的硬件实验平台，搭建非线性干扰环境，验证所设计的人工智能算法抗非线性干扰的能力，通过分析计算结果与工程实验测试结果之间的差异，从机理上给出了合理的解释。