Research On Waveform Optimization And Design In Cognitive Radar

With the continuous exploration and development of space, more extensive use of the electromagnetic spectrum and diversity and high-speed of target all lead to electromagnetic environment becoming more complex. Traditional radar transmits single waveform, lacks flexibility and hardly adapt to complex working environment. As a kind of intelligent radar, cognitive radar can interact with environment, percept environment parameters in real time, and adaptively construct corresponding waveform library and transmit suitable waveform, so radar performance can be improved. The research on cognitive radar is receiving more and more attention.Waveform optimization and design is an important problem in the research of key technology in cognitive radar, and it contains two aspects of waveform selection and waveform design. In this thesis, we systematically analyze basic principles and composition of cognitive radar, and basic methods for traditional waveform design. We propose valid algorithms of waveform optimization and design, mainly involves optimal waveform selection algorithm based on dynamic programming, waveform selection algorithm based on approximate dynamic programming and optimal waveform design algorithm baesd on mutual information.The problem of waveform selection can be attributed as problem of maximization or minimization. For example, minimize error, minimize energy or maximize value, etc. Dynamic programming is a powful tool for solving these problems. We analyze radar measurement parameters, divide range-Doppler resolution cells, calculate measurement probability, select reward function, use dynamic programming theory and propose dynamic programming model of waveform selection. This model can reflect the reward difference when radar transmit different waform and detect different target, and it provides the theory basis for dynamic programming methods using in waveform selection. Based on this model, we propose optimal waveform selection algorithm based on dynamic programming. This algorithm fully ergodic the whole state space and waveform space, and optimal waveform sequence can be obtained. It can be seen in the simulation results that state estimation error of optimal waveform selection algorithm based on dynamic programming is lower than that of fixed waveform.Although optimal waveform selection algorithm can be gained using traditional dynamic programming algorithm, this calculation needs state observation transition probability, and each iteration need precise solution. When state space and waveform number become large, each iteration solution can hardly realize, and precise dynamic programming method is invalid. For this problem, we propose waveform selection algorithm based on temporal difference learning. This algorithm only need to calculate temporal difference value and do not need to fully ergodic the whole state space and waveform space and it overcomes the disadvantage of optimal waveform selection algorithm to some extent. The computational efficiency of waveform selection algorithm based on temporal difference learning is much higher than optimal waveform selection algorithm, and the performance of state estimation error is very close to optimal waveform selection algorithm. However, in the calculation of waveform selection algorithm based on temporal difference learning, state observation transition probability is needed. We propose waveform selection algorithm based on Q-learning to avoid computing state observation transition probability and further improve efficiency.In this algorithm, it only needs to calculate Q value and do not need to fully ergodic the whole state space and waveform space. Morever, it does not need state observation transition probability which is suitable for radar. The computational efficiency of waveform selection algorithm based on Q-learning is higher than waveform selection algorithm based on temporal difference learning, and the performance of state estimation error is also very close to optimal waveform selection algorithm based on dynamic programming.The excellent performance of cognitive radar not only depends on rational waveform selection, but also depends on the design of waveform itself. The design of radar waveform is task-dependent. For detection task, the optimal radar waveform should be able to put as much transmitted energy as possible into the largest mode of the target to maximize the SNR. For estimation task, the optimal radar waveform should allocate the energy between the received signal and the target signature. As intelligent cognitive radar, detection task and estimation task should be considered simultaneously. That means mutual information between the received signal and target signature should be maximized on the basis of certain SNR. For the problem of waveform design of cognitive radar, we propose the optimal waveform design algorithm baesd on mutual information and design criterion for transmitted waveform in cognitive radar. We consider two situations of no clutter and clutter, and propose the corresponding mutual information models. Considering both detection performance and estimation performance, and with the constraints of SNR and energy, the problem of waveform design of cognitive radar can be converted into the problem of convex optimization. Finally we use interior-point method with Newton method to solve this problem. It can be concluded from the simulation results that when power spectrum peak of cognitive radar waveform changes with that of target’s impulse simultaneously, the mutual information between the received signal and target signature can reach maximum, no matter in the situation no clutter and clutter, and no matter what the target signature is. That means the received signal contains the most information on target signature, and it can be as design criterion for transmitted waveform in cognitive radar.