一种改进的特征值-LSTM微弱信号盲检测方法  

An improved eigenvalue-LSTM based weak signal blind detection method

作　　者：任昕 REN Xin(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信学院,黑龙江哈尔滨150001

出　　处：《应用科技》2022年第5期67-73,共7页Applied Science and Technology

摘　　要：微弱信号盲检测方法中,最大最小特征值检测应用到样本协方差矩阵的信息较少,且其判决门限难以准确估计,导致信噪比较低时检测稳定性差、性能较低。针对以上问题,提出一种基于特征值-长短时记忆(eigenvalue-LSTM)神经网络的微弱信号盲检测方法,在最大最小特征值检测算法的统计量中引入平均特征值,并对检测概率、虚警概率进行了推导分析,最后将改进的统计量作为长短期记忆(LSTM)神经网络输入进行训练得到分类器。仿真实验结果表明,该方法无需估计检测门限,且在相同采样点的条件下,检测性能在-12 dB时较直接采样-LSTM方法提高了9%。因此基于特征值-长短时记忆网络神经网络方法具有更优的检测性能。Among various weak signal blind detection methods,the maximum and minimum eigenvalue detection algorithm detects little information when applied to the sample covariance matrix,and it is difficult to estimate corresponding judgement threshold accurately,resulting in poor detection stability and low performance under the low signal-to-noise ratio condition.Against above problems,based on the eigenvalue-LSTM neural network,a blind detection method for weak signals is proposed in this paper.An average eigenvalue is introduced to the statistics of the maximum and minimum eigenvalue detection algorithm.Then the detection probability and false alarm probability are deduced and analyzed.Finally,The improved statistics are used as the input of the long short-term memory(LSTM)neural network for training to obtain the classifier.The simulation experiment results show that this method requires no detection threshold estimation and that under the condition of the same sampling point,its detection performance is 9%higher than that of the direct sampling-LSTM method when the signal-to-noise ratio is-12dB.

关 键 词：微弱信号 盲检测 特征值 神经网络 存在性检测 最大最小特征值检测 长短时记忆 多天线接收 

分 类 号：TN92[电子电信—通信与信息系统]