Emitter Identification And HLS Implementation Based On Complex Neural Network

The identification of special emmiter(SEI)palys an important role in military and civilian fields.Aiming at the problem of poor anti-noise performance of real convolutional neural network(CNN),based on the mechanism of individual fingerprint of radiation emmiter and deep complex CNN,the main work of this thesis is as follows:(1)We study and realized SEI based on deep complex CNN.In this method,the part of effective signal is first extracted by energy detection,peak detection method,signal length detection method,and then the one-dimensional complex signal is transformed into a two-dimensional time-frequency by frame slicing and STFT transformation as the training and test samples of the network.Then we design a complex CNN to identify different emmiters.Finally,this thesis designs a real convolutional neural network and a complex convolutional neural network with similar network structures,and tests them on nine emiiters data actually collected.The results prove that the complex CNN has stronger anti-noise performance than the real CNN.(2)We proposed the theory of complex activation function based on probability amplitude method,and we designed two probability normalization methods,c-softmax and c-norm,according to this theory.Afterwards,based on the end-to-end complex CNN model,this thesis conducts identification on the two probability normalization methods on the public wifi datasets of 16 emiiters.The results proves that the performance of csoftmax is close to the traditional cat method and compared with which,c-norm has stronger robustness to noise.(3)Based on HLS technology,this thesis designed a configurable complex CNN hardware deployment scheme.The scheme consists of direct memory access(DMA),configurable complex convolution IP core,configurable complex pooling IP core and ARM.In the computing engine,the channel dimension is expended in the convolution process,and 32-bit parallel pipelined processing technology design methods are used to improve computing performance;in the control logic,parameter configuration and hardware deployment system are realized through PYNQ framework.At last the hardware system is tested on the public wifi datasets of 16 emiiters in this thesis.The results show that the hardware deployment system in this thesis can achieve 17 FPS at the operating frequency of 100 MHz,which can be deployed on the terminal.