Abstract:

In wireless network communication,wireless channel fingerprint is commonly used to detect the location of wireless signals.This paper proposes a method of bisection location search,combined with Deep Belief Network(DBN) for the specific application,such as the radio channel fingerprint,in order to solve the phenomenon of higher error rate of the classification error when the data feature is related with the location.This method extracts the location feature information from the training data set,and predicts the relative location of the unknown data in one-dimensional space based on the extracted features.Experimental results show that compared with the traditional channel feature extraction and hierarchical classification method,DBN bisection location method can automatically extract the feature information,and reduce the scene recognition error tolerance to 10 m,the average error of channel scene recognition is reduced to 2.3 m.

Key words: Restricted Boltzmann Machine(RBM), Deep Belief Network(DBN), deep learning, bisection search, wireless channel fingerprint

摘要：

在无线网络通信中,无线信道的指纹常用来检测无线信号的位置。然而将无线信道指纹用于与位置相关的特定应用问题时,会产生分差错误率升高的现象。为此,结合深度信念网络(DBN)提出一种二分位置查找方法。从训练数据集中提取与位置相关的特征信息,并根据提取的 特征查找未知数据在一维时空中的相对位置。实验结果表明,与传统信道特征提取和层次分类方法相比,该方法可自动提取特征信息,且场景识别误差限降低到10 m,信道场景识别平均误差下降到2.3 m。

关键词: 受限玻尔兹曼机, 深度信念网络, 深度学习, 二分查找, 无线信道指纹

CLC Number: 

• TP393