摘要：

基于指纹的定位方法是定位技术中的重点研究内容。现有指纹定位方法大多数是为接收端设备的自定位而设计的，无法直接应用到对信号发射端的定位中。为此，设计一种适用于信号发射端的辐射源指纹定位模型，并提出相应的定位算法。利用频谱地图进行指纹定位，基于信号指纹生成算法和指纹匹配定位算法实现指纹匹配定位。信号指纹生成算法将空间插值转化为机器学习多变量回归问题，基于测量点的数据建立数据集并训练随机森林回归模型，结合蛇优化算法改进模型的参数设置，利用改进的随机森林回归模型补全频谱地图中的缺失数据，进而得到信号指纹。指纹匹配定位算法通过深度学习框架完成，利用卷积神经网络从频谱地图中估计辐射源的位置坐标。仿真实验结果表明，该定位模型在使用10%的测量数据时，平均定位误差为12.91 m，且定位误差在20 m以内的置信概率达到了84.5%，能够实现对信号发射端的指纹定位，与其他方法相比, 具有较优的定位精度和定位稳定性。

关键词: 指纹定位, 频谱地图, 随机森林, 蛇优化算法, 卷积神经网络

Abstract:

Fingerprint-based positioning method is the key content in positioning technology. Most of the current fingerprint positioning methods are designed for self-positioning of receivers, and cannot be directly applied to the positioning of signal transmitter. Considering the abovementioned problem, this study proposes an radiation source fingerprint positioning model, which can be used for signal transmitters, and designs the corresponding positioning algorithms.The model utilizes a spectrum map for fingerprint positioning, which is implemented by the signal fingerprint generation algorithm and fingerprint matching algorithm.The signal fingerprint generation algorithm transforms spatial interpolation into a machine learning multivariable regression problem.The dataset is established based on measurement points and then used to train the Random Forest(RF)regression model.Parameters of the model are optimized with the help of the Snake Optimizer(SO). The improved RF regression model is used to complete the missing data in the spectrum map to obtain the signal fingerprint. The fingerprint matching algorithm is implemented through the deep learning framework, and the position coordinates of radiation sources are estimated from the spectrum map by the Convolutional Neural Network(CNN). Simulation experiment results show that the proposed method has an average positioning error of 12.91 m when the proportion of measurement data is 10%, and the fiducial probability of the positioning error within 20 m is 84.5%.The proposed method can realize the fingerprint positioning of signal transmitters and has better performance in positioning accuracy and stability compared with other methods.

Key words: fingerprint positioning, spectrum map, Random Forest(RF), Snake Optimizer(SO)algorithm, Convolutional Neural Network(CNN)