Research On The Cognitive Tracking Radar Waveform Optimization Techniques Based On Machine Learning

In recent years,with the development of technology,the environment where radar used has become more and more complicated,so the demand for multifunction radar's performance is getting higher and higher.However,the working mode of traditional radar is usually fixed,and through this fixed mode,the system can only processing echo data and transmitting waveform passively.Even if the radar system is reasonable designed,it cannot be generalized to deal with the sophisticated environment.Then the concept of COGNITIVE RADAR was proposed by professor Simon Haykin,this kind of radar is changing the open-loop of traditional radar into closed-loop,and to a certain extent,it can solve the problems of facing complex environment.Compared with adaptive radar,the cognitive radar embedded with knowledgeaided module,which gives radar the ability of ‘Thinking' just like human do.And for different tasks of radar,the ability of ‘Thinking' reflected in different aspects.For common radar detection,based on the knowledge-aided module can improve the probability of detection,this can be done by take advantage of the prior knowledge of different environment;As for the tracking task,the accuracy of tracking can also be improved by adapting waveform based on different kinds of ways to process varying environments.And the technique of select waveform adaptively is one of the key points of the current cognitive radar systems,which is also the main topic of this thesis.The thesis is started with an brief introduction of the development of cognitive radar.Then there is a research on the basic structure of cognitive radar system,which is based on the basic framework proposed by Simon Haykin.And through this research,the paper conceptual analyze every single module of the entire cognitive radar system to illustrate the function of each module,which is on the basis of neural network and Bayesian filter theory.Based on all this effort,we refine the structure of cognitive radar system and analyze this the entire process in the last section of this chapter,to sums up the implementation of cognitive radar system in the signal transmission direction of tracking process.After that,the paper introduces the basic function of each piece of this new structure in details,and with the help of mathematical model of each module,we analyze how the waveform influences tracking task of cognitive radar,which gives the relationship between the measurement noise covariance and the waveform parameters.So that it can build up the bridge between waveform and tracking task.Then by using the cost-to-go function under the minimum mean square error criterion,this chapter compare the waveform-agile method with traditional closed-loop radar and radar with fixed waveform parameters through a linear motion simulation,based on this conclusion,the main signal type is determined and we can know the waveform-agile radar have a advantage in tracking tasks over other kinds of radar used in this chapter.By the same time,simulating the influence of interference on tracking task by changing the measurement value,the capacity of resisting disturbance of waveform agile radar is discussed.Secondly,the thesis gives a relatively detailed introduces about the artificial neural network and reinforcement learning.And based on the basic principle of neural networks,the paper explains how to deal with the training method of cognitive radar tracking task.Then using the trained neural network for tracking problem and compare the performance with waveform agile radar;besides,focusing on the Bellman optimal dynamic equation,we proposed two kinds of reinforcement learning algorithms and give the implementation process in a form of pseudo code.Then,the tracking performance of two reinforcement learning algorithms and waveform agility methods under linear model is analyzed in detail,and the superior performance of cognitive algorithm is confirmed.The method of machine learning proposed in this paper,which using the relationship between tracking process and waveform parameters,improved the tracking precision of the moving target.