基于CLDNN的物联网设备个体识别

Individual identification of Internet of things devices based on CLDNN

摘要:

针对目前个体识别中特征工程存在信号序列长、特征鲁棒性差等问题，研究了基于深度神经网络的个体识别技术。借鉴语音识别中的卷积长短时全连接神经网络(CLDNN)，通过卷积神经网络提取信号的局部幅度特征，通过长短期记忆网络提取信号的全局时域特征，使用全连接网络实现特征图到设备标签的映射。在视距(LOS)信道下，采集8台LoRa调制的无线数传电台数据，加入高斯白噪声进行仿真测试。仿真表明，本文所提方法在信号序列长度为2 048点和低信噪比(0 dB)时，模型准确率达到95%；此外相较VGG16模型，本模型参数更少，在物联网设备部署方面具有一定的应用前景。

Abstract:

Aiming at the problems of long signal sequence and poor feature robustness in Feature Engineering in individual recognition, the technology based on deep neural network is studied. Drawing lessons from Convolutional, Long Short-Term Memory, Fully Connected Deep Neural Networks(CLDNN) in speech recognition, the local amplitude features of the signal are extracted through convolution neural network and the global time-domain features of the signal are extracted through long-term and short-term memory network. A fully connected network is utilized to map the feature to the device label. Under the line of sight channel, the data of eight Lora modulated wireless data transmission stations are collected, and the Gaussian white noise is added to the simulation test. The simulation shows that when the Signal-to-Noise Ratio is low(0 dB) , the accuracy of the model can reach nearly 95% under the signal sequence length of 2 048 points. In addition, this model needs fewer parameters compared with VGG16 model. The proposed model has a certain application prospect in the deployment of Internet of things devices.