A Study Of Multi-feature Based Method For LoRa Signal Source Identification

Long-distance and large-scale IoT technology has emerged as a viable means for supporting applications like smart city,smart farm,wild animal and plant monitoring,etc.However,the traditional wireless communication technologies based on RFID,ZigBee and WiFi still have limitations.Recently,with the development of wireless communication technology,the emergence of low power wide area network(LPWAN)has become a powerful boost for large-scale IoT(Internet of Things)applications.As one of the specific implementations of LPWAN,the long range(LoRa)network has attract wide attention from industry and academia,cause its mature protocol architecture and its advantages in unlicensed frequency bands.At present,LoRa deployment has covered more and more fields,such as environmental monitoring,target location,trajectory tracking and passive sensing.With the expanding deployment range of LoRa network,the number of LoRa signal sources increases,which dramatically increases the probability of collision between signal sources In order to ensure the accurate decoding of collision LoRa signals,timely identification of information for localization/tracking,and prevention of malicious signal source attack,we aim to study the method of recognizing LoRa signal source in this work.Our research emphasis can be summarized as followings :(1)LoRa signal preprocessing: Since the LoRa signal utilizes the chirp spread spectrum(CSS)modulation technology,it can not be extracted by received signal strength indicator(RSSI)based methods.In this work,we propose TripleC,a LoRa packet detection method based on the signal cross-correlation,which has the ability to extract correct data packets from collision signals interfered by other LoRa sources.(2)Signal feature extraction: This thesis analyzes the characteristics of LoRa signal in time domain,frequency domain and time-frequency domain,and combines the fractal dimension in fractal theory to extract the amplitude,IMF amplitude,marginal spectrum,box dimensions and information dimensions of LoRa signal.These indicators form the feature set for describing different signal sources.(3)Signal source recognition: This work presents a multi-feature based LoRa signal source recognition scheme Mu-FA.The basic idea of Mu-FA is to associating the extracted multi-feature sets with machine learning classifiers for identifying different LoRa signal sources and verify the validity of the method in real scenario.Comparing with the existing methods,the experimental results of Mu-FA shows that it can achieve 70% accuracy when identifying 10 signal sources in the outdoor environment.