| 摘要: |
| 目前无线通信网络频谱环境时空分布复杂多变，现有多用户协同感知方法数据预处理繁琐，感知效率低下。为此，在由用户感知层和边缘融合层构成的系统架构下，提出了一种基于协同学习的频谱智能感知算法。用户感知层采用多分支卷积循环门控神经网络，利用原始归一化能量信号的底层结构信息，实现本地感知。边缘融合层基于自注意力机制进行消息传播，融合用户感知层中各个非授权用户的感知结果得出最终决策。实验表明，在信噪比为-20 dB以及5个用户协同感知的情况下，该方法能在虚警概率为1.91%时达到18.3%的检测概率，相比对比模型提升了6.1%，且不需要对原始数据额外预处理，降低了算法的复杂度。 |
| 关键词: 智能频谱感知 协同学习 卷积神经网络 门控循环单元 自注意力机制 |
| DOI：10.20079/j.issn.1001-893x.220721005 |
|
| 基金项目:国家自然科学基金资助项目(61931004，61801073)；江苏省自然科学基金项目(BK20210641) |
|
| An Intelligent Spectrum Sensing Method Based on Collaborative Learning |
| PAN Chengsheng,CAI Ren,SHI Huaifeng,SHI Jianfeng,WANG Yuyue |
| ((1.School of Electronics and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;2.School of Automation,Nanjing University of Science and Technology,Nanjing 210094,China;3.National Mobile Communications Research Laboratory,Southeast University,Nanjing 211189,China)) |
| Abstract: |
| The spatial and temporal distribution of current heterogeneous network spectrum environment is complex and variable,the data preprocessing of existing multi-user cooperative sensing methods is cumbersome,and the sensing efficiency is low.For above problems,a cooperative learning-based spectrum intelligent sensing algorithm is proposed under a system architecture consisting of user sensing layer and edge fusion layer.The user-aware layer uses a multi-branch convolutional recurrent gated neural network to realize local sensing by using the underlying structural information of the original normalized energy signal.The edge fusion layer performs message propagation based on a self-attention mechanism and fuses the sensing results of each unauthorized user in the user-aware layer to arrive at the final decision.Experiments show that when the signal-to-noise ratio is -20 dB and five users are sensing cooperatively,the proposed method is able to achieve a detection probability of 18.3% at a false alarm probability of 1.91%,an improvement of 6.1% compared with the comparison model,and does not require additional pre-processing of the raw data,thus reducing the complexity of the algorithm. |
| Key words: intelligent spectrum sensing collaborative learning convolutional neural network gated cycle unit self-attention mechanism |
