Unsupervised Spectrum Anomaly Detection Based On Adversarial Autoencoder

With the development of wireless communication technology,electromagnetic spectrum resources have begun to become tight.Cognitive radio technology has been proposed as an effective solution,but there is a risk of malicious interference attack on cognitive radio.In modern warfare,electromagnetic warfare is inevitable,and timely identification of enemy interference and abnormality of our equipment is of strategic significance.Electromagnetic spectrum monitoring and spectrum anomaly detection are important parts of radio spectrum management,which can help detect radio anomalies such as malicious interference and illegal occupation,and is of great significance for ensuring radio management and electromagnetic environment assessment.In view of the large size,high power consumption,and inflexible deployment of current electromagnetic spectrum monitoring equipment,the micro spectrum sensor network protocol and micro node designed in this paper take flexible deployment,dynamic networking,small size,and low power consumption as the design goals,which can meet he electromagnetic environment monitoring needs in areas where people are difficult to reach in the field,border and coastal defense,ocean,islands,etc.In view of the huge amount of data collection of the micro sensor network system,and the high cost of artificial spectrum anomaly detection,the current spectrum anomaly detection algorithm cannot meet the requirements of spectrum anomaly detection of the system.This paper proposes an algorithm for detecting anomalies in the electromagnetic spectrum over a wider frequency band using adversarial autoencoders.The main work and innovations of this paper are as follows:(1)This article first designs the hardware architecture of the micro spectrum sensor node,and designs the overall architecture of the micro spectrum monitoring device node with an ARM processor as the main control module and FPGA as the data processing module.In view of the large amount of data generated by micro nodes and complex interactions,it is proposed to use UIO interrupts in conjunction with the internal AXI bus of the ZYNQ chip to complete the frequent and complex interactions between the main control ARM and the data processing module FPGA to provide high-speed internal data channels for both.At the same time,the work flow of the micro spectrum sensor network is analyzed,and the protocol stack of the spectrum sensor network is selected.In view of the complex structure of the micro spectrum sensor network,the high data communication stability requirements,and the complex and diverse functions,the design is composed of the transmission layer and the micro spectrum sensor network application layer protocol composed of two-layer protocols at the business layer.(2)The functional requirements and non-functional requirements of micro nodes in micro spectrum sensor networks are analyzed,and the functions of micro nodes are realized.Specifically,micro node network management functions are designed and implemented to ensure reliable and stable data communication.The design and realization of single frequency scanning,digital scanning,and frequency hopping signal analysis functions provide the management center with valuable electromagnetic spectrum monitoring data.The design realizes the working mode switching function,which ensures that the micro node can switch to the specified low power working mode according to the management center's instructions,and extends the running time of the micro node for a single charge.(3)This paper studies the current electromagnetic spectrum anomaly detection algorithm.Aiming at the problem that the current algorithm needs to train multiple copies of the same model for different frequency bands,an unsupervised anomaly detection algorithm based on adversarial autoencoder is proposed.Experiments on the model by using the data actually collected by the micro spectrum sensor node designed in this paper.The experiment shows that the performance of the spectrum anomaly detection algorithm based on the adversarial autoencoder is better than that of the variational autoencoder algorithm.By appropriately increasing the implicit encoding vector dimension can further improve detection performance.