Identification And Interpretation Of RF Power Amplifier Based On Deep Neural Network

In the modern informat ion war,the electromagnet ic environment is becoming more and more complex,and the individual ident ification o f radar radiat ion sources is beco ming more and more pract ical significance.Individual ident ificat ion is the process of meticulously characterizing and ident ifying radiat ion sources with different subt le individual differences.The emph asis is no longer limited to parameter measurement and modulat ion mode o f the electro magnet ic signal,but the reflect ion o f addit ional modulat ion characteristics inherent in physical equipment.This characterist ic does not change with the change o f modulat ion parameters,and has the "fingerprint" characterist ic o f the individual equipment.Broadband RF power amplifiers are one o f the co mmo n basic components in radar and communication systems.They o ften have individual characterist ics.To analyze their memory characteristics,nonlinear distortion and other behavioral characterist ics and to identify individual power amplifiers have great significance.For the above problem,because the neural network can self-extracting featurs,this paper uses the deep neural network model for individual ident ificat ion of broadband RF power amplifiers,and at the same time uses the feature interpretation algorithm to explain the output result s of the neural network model.This art icle works as follows:First ly,introduced the topic background;review the development status o f RF power amplifier behavior model,neural network model and deep neural network model interpretabilit y algorithm at home and abroad;put forward the research content of this topic.Secondly,analyze the nonlinear distortion behavior o f the broadband RF power amplifier,the harmonic distort ion,intermodulation distortion and memor y effect are the main distort ion behaviors;Use the Matlab to establish the nonlinear behavior model of the broadband RF power amplifier and use the ADS to design the broadband RF power amplifier.It simulates the nonlinear distort ion behavio r of the broadband RF power amplifier fro m the software and hardware levels.Thirdly,by analyzing the network structure and characteristics o f different types of neural networks,it is found that convolut ional neural networks have great advantages in signal recognit ion.Design a convo lut ional neural network with five convo lutional layers,two pooling la yers,one fully connected layer and one output layer.The experiment o f measuring the output signal of the power amplifier proves that the network can individually identify the broadband RF power amplifier.Finally,research the network model interpretatio n method for the problem of the black box model of deep neural networks.And use Sobo l sensit ivit y analysis method,layered correlat ion propagation algorithm and deep lift ing algorithm to analyze the input layer and the middle hidden layer of the trained convo lutional neural network model,by calculat ing the correlation of each neuron to the final output result score,analyze the characterist ics o f the input signal that have a recognit ion effect,and complete the interpretation of the deep network model.