Research On The Method Of LoRa Signal Recognition In Deep Learning

With the rapid development of Internet of Things(IoT)technology,the process of interconnecting everything has been further promoted.Traditional communication technologies such as RFID,Bluetooth and Zig Bee can no longer meet the growing needs of Internet of Things(IoT)application scenarios.Low Power Wide Area Network(LPWAN)is an emerging technology in the network layer of the Internet of Things(IoT),which provides a new technical idea for the large-scale application of the Internet of Things(IoT).LoRa,as a great potential communication technology in LPWAN,has become one of the mainstream technologies for the Internet of Things(IoT)due to its characteristics of long transmission distance,perfect network protocol and working in unauthorized frequency band,and has been widely concerned by industry and academia.With the popularity of LoRa applications,there are also some problems and challenges in practical deployment and use.The Adaptive Data Rate(ADR)mechanism is often called for LoRa terminal nodes to increase the spreading factor and ensure the reliability of data transmission at the cost of node power consumption and transmission delay,which makes the node life cycle far from the expected situation.In recent years,deep learning has been used to solve complex problems or replace traditional methods in many fields due to its excellent representation learning and data processing capabilities.Therefore,using deep learning to solve related problems in LoRa can be considered to have some research value.This paper carries out relevant research on the application of deep learning in LoRa signal recognition,mainly including:(1)For the unpublished LoRa physical layer modulation and demodulation technology patent,on the basis of analyzing the characteristics of the chirp signal,combined with the matched filtering method,the process of the LoRa signal modulation and demodulation are deduced,and the bit error rate simulation is carried out.The bit error rate results are compared and verified with the theoretical values.(2)For the denoising problem of LoRa signal containing additive white Gaussian noise,combined with the energy aggregation characteristics of chirp signal after fractional Fourier transform,the mathematical form of the LoRa signal fractional Fourier transform process is deduced,and this paper proposes a LoRa signal filtering method based on fractional Fourier transform that can effectively suppress noise.Aiming at the contradiction between computational complexity and precision of the optimal fractional Fourier transform domain search in the filtering method,an optimization method based on particle swarm optimization is also proposed.(3)In view of the problem that the survival cycle of LoRa terminal nodes is shortened due to the ADR mechanism that increases the spreading factor,and combined with the power supply characteristics of LoRa gateway.A method of deploying the deep learning model on LoRa gateway for LoRa signal recognition is proposed to improve the communication quality and reduce the power consumption of nodes.In order to carry out the research of deep learning in LoRa signal recognition,on the basis of comparing the performance of several common neural network models,a CNN model for LoRa signal recognition is proposed.The simulation results show that compared with the traditional matched filter demodulation,the model has a great improvement in bit error rate performance and signal transmission distance.On this basis,the filtering method based on fractional Fourier transform is also used to preprocess the LoRa signal sampling data,and a CNN model for LoRa signal recognition after preprocessing is proposed,which further improves its performance in LoRa signal recognition.