针对目前在扩频通信系统中,直接序列扩频信号用户数估计在各个用户载波频率未知以及低信噪比环境下性能较差的问题,提出了一种基于信号特征值直方图和 L-ResNet(Local Binary Patterns-Residual Network)的用户数估计方法.首先将收到的直接序列扩频信号构造为数据矩阵,利用信号矩阵的协方差矩阵的特征值构造特征值直方图.此方法利用特征值的大小以及梯度来放大不同用户数目的特征值直方图区分度,将用户数估计问题转化为分类问题.然后采用L-ResNet网络以区分不同用户数目的特征值直方图,引入 Local Binary Patterns(LBP)来约束网络模型的损失,增大不同类别的特征值直方图在梯度上的区分度,从而提高网络模型的直扩信号用户数估计性能.最后,实验结果表明,基于信号特征值直方图和 L-ResNet的直扩信号用户数估计性能优于特征值门限、相邻特征值之比的导数(DRAE)方法.
Estimation of the number of direct spread signal users based on signal eigenvalue histogram and L-ResNet
In order to solve the problem of poor performance in estimating the number of direct sequence spread spectrum signal user in the current spread spectrum communication system when the carrier frequency of each user is unknown and in a low signal-to-noise ratio environment,a method based on signal eigenvalue map and L-ResNet(Local Binary Patterns-Residual Network)is proposed.Firstly,the received direct sequence spread spectrum signal is constructed as a signal matrix,and the eigenvalue histogram is constructed by using the eigenvalues of the covariance matrix of the signal matrix.This method uses both the size and the gradient of the eigenvalues to amplify the discrimination of the eigenvalue histograms of different numbers of users,and transforms the problem of user number estimation into a classification problem.Then,build the L-ResNet network to distinguish the eigenvalue histograms of different numbers of users,introduce Local Binary Patterns(LBP)to constrain the loss of the network model,increase the degree of differentiation of different types of eigenvalue histograms on the gradient,and improve the performance of network model in estimating the number of direct sequence spread spectrum signal users.Finally,experimental results show that the estimation performance of DSSS signal user number based on signal eigenvalue map and L-ResNet is better than that of eigenvalue threshold and derivative of adjacent eigenvalue ratio(DRAE)method.
spread spectrum communicationuser number estimationneural networkhistogram
NETL
NSTL
万方数据
