Design And Implementation Of Radiation Source Classification System Based On ZYNQ

At present,it is a research hotspot to extract electromagnetic fingerprint feature by convolution neural network.Application scenarios such as in-vehicle airborne put forward urgent requirements for low-power and high-performance hardware implementation;however,there are difficulties such as the flexibility of the radiation source signal and identifying unknown radiation sources.The hardware implementation of convolutional neural network not only needs to support the dynamic update of convolutional neural network model and parameters,but also needs to meet the requirements of high performance and low power consumption.In order to solve these problems,this thesis adopts an embedded platform of ARM + FPGA structure,such as ZYNQ,to design and implement a radiation source classification algorithm based on convolutional neural networks.Based on the characteristics of hardware resources in the hardware platform,the computational structure of the convolutional neural network is optimized.Moreover,in order to ensure that the update of the convolutional neural network will not affect the reasoning of the convolutional neural network,a dynamic update system is designed in the way of combination of hardware and software,so that the hardware platform can fully meet the application requirements and achieve the classification task of unknown radiation sources.The main work of this thesis is as follows:1.The design goals of radiation source classification and the radiation source classification algorithm based on convolutional neural network are introduced.The ZYNQ platform and its internal key hardware resources are analyzed,and the constraints of these hardware performances are discussed in combination with the characteristics of the algorithm.The system architecture of radiation source classification is proposed.2.An FPGA implementation scheme for large-scale convolutional neural networks is designed to meet the computational requirements of radiation source classification algorithms based on convolutional neural networks.The overall design of the algorithm and the design of each module are implemented in FPGA.Aiming at the constraints caused by algorithm complexity and scarce hardware resources,a series of improvement schemes are proposed.High performance and low power consumption can be achieved by the hardware implementation of the radiation source classification algorithm.3.A system design that uses the ZYNQ platform and a remote server to dynamically update the parameters of the convolutional neural network is proposed to meet the identification and classification of unknown radiation sources.Based on the FPGA design of the convolutional neural network,the overall design of the ARM core in ZYNQ and the remote server for convolutional neural network training is designed.Tasks are assigned based on the characteristics of each platform,and the design of each part of the software system is optimized.The classification algorithm can be smoothly and dynamically updated by the system,which meets the software and hardware system requirements for unknown radiation source classification,and the overall robustness of the system is improved.