Research On Deep Learning Method For Individual Recognition Of Wi-Fi Signal Radiation Source

With the popularity and application of wireless network,Wi-Fi signal has been ubiquitous in our daily life.Although Wi-Fi networks bring us convenience and efficiency,the identification and authentication technologies based on digital forms commonly used in Wi-Fi networks are easy to be compromised,leading to certain security risks.The Wi-Fi signal radiation source individual identification technology can be used to find unauthorized access points or malicious Wi-Fi jamming devices,to avoid hackers using these devices to carry out network attacks or steal sensitive information,and has a positive role in improving the security of the Wi-Fi network.Based on the in-depth study of Wi-Fi signal characteristics,communication radiation source individual recognition technology,and deep learning algorithm,this paper designs three kinds of deep learning methods for Wi-Fi signal radiation source individual recognition to solve the problems of low recognition rate,poor interpretation of network model and low recognition rate under the condition of small samples.The main research content of this paper is as follows:1.The influence of RF fingerprints produced by DA converters,filters,orthogonal modulators,and local oscillators on the modulation domain constellation and error vector amplitude of the transmitted signal is analyzed,and the causes of the two stray characteristic components in the transmitted signal,DC bias,and third-order intermodulation interference,are studied.2.An expert feature-driven Wi-Fi signal radiation source individual identification method is designed.In this method,the leading code of the received Wi-Fi signal is selected as the target signal segment for feature extraction.Expert features are extracted by short-time Fourier transform,Hilbert-Yellow transform,and cyclic spectrum analysis,and input into the deep residual network through channel feature fusion to complete classification and recognition.The experimental results show that under the condition of 500 training samples,the recognition accuracy of 5 types of Wi-Fi signal radiation sources can be achieved at 91.6%,and when the training samples are reduced to 100,the recognition rate remains above 80%,indicating that this method effectively solves the problem of network recognition rate reduction under the condition of small samples.3.To improve the ability of the deep learning network to extract individual characteristics of radiation sources from original I/Q signals,a data-driven Wi-Fi signal radiation source individual recognition method is designed.The proposed method is based on a convolutional neural network that integrates the multi-stage double-attention mechanism and combines two data enhancement schemes of adding random noise and random moving slices to realize highaccuracy identification of Wi-Fi signal radiation source individuals in a data-driven way.Experimental results show that the identification accuracy of the proposed method can reach 96.8%for five Wi-Fi signal radiation sources,and the comparison experiment proves that compared with several traditional data-driven recognition methods,the proposed method has more advantages in recognition rate and network reference number.4.To solve the problems of complex expert feature extraction in expert feature-driven method and poor interpretation of network model in data-driven method,a Wi-Fi signal radiation source individual recognition method based on depth adaptive wavelet network is designed.The deep adaptive wavelet network can implement the process of lifting wavelet transform algorithm in its network structure to avoid extra expert feature extraction,and thus make its recognition process interpretable to a certain extent.The experimental results show that the identification accuracy of this method can reach 95.5%when identifying 5 different Wi-Fi radiation sources,and it has good anti-noise performance.