伴随日益强烈的民用、军事和国家战略利益拓展的应用需求，我国急需具有自适应干扰抑制的智能化卫星通信系统。本项目研究基于可变多核卷积神经网络-深度学习(MKCNN-DL)的卫星通信物理层干扰认知、抑制的理论与技术方法，为自适应抗干扰的智能化卫星通信提供有力支撑。相关研究内容主要包括：(1)构建可变的多核卷积神经网络(MKCNN)模型。依据物理层干扰和信道特点设计针对性卷积核，增强模型对物理层复杂干扰信号的特征提取能力；(2)基于MKCNN的物理层干扰认知。对采集数据进行多维预处理，生成高效学习训练样本数据，通过变步长学习训练算法进行模型快速学习，利用训练完毕模型完成干扰准确认知；(3)结合认知结果的自适应干扰抑制。通过自适应对消方法对可重构干扰进行消除，同时利用通信参数学习优化方法实现其他干扰抑制。

With the increasing applications for civil, military and national strategic interests, we urgently need an intelligent satellite communication system with adaptive interference suppression. The theory and technical methods of interference recognition and suppression of satellite communication physical layer, based on variable multi-core convolutional neural network-deep learning (MKCNN-DL), will be studied and analyzed in this project, which can provide strong support for intelligent satellite communications with adaptive anti-interference. The research content mainly includes: (1) Construction of variable multi-core convolutional neural network (MKCNN) model. The targeted convolution kernels are designed based on physical layer interference and channel characteristics to enhance the model's feature extraction capability for complex interference signals in the physical layer. (2) MKCNN-based physical layer interference recognition. The multi-dimensional pre-processing of the collected data will be operated to generate efficient learning training sample data, and the model can achieve rapid learning through variable step learning training algorithms. Consequently, the interference can be accurately recognized by the trained model. (3) Adaptive interference suppression based on cognitive results. The reconfigurable interference will be eliminated by an adaptive cancellation method, and other interference suppression methods can be implemented using a communication parameter learning optimization method.