Waveform Selection Method Based On Adaptive Dynamic Programming

Adaptive waveform selection has been widely applied in radar, sonar and cognitive radio and so on. During these years, especially after the concept of cognitive radar was proposed by Haykin professor, the research on adaptive waveform selection has turned out to be a major issue in radar system.Because of the widely use of the electromagnetic spectrum, diversity and high moving of targets and the complexity of radar scene environment, the traditional waveform design methods can't satisfy the real need. The waveform selection method based on adaptive dynamic programming can fully use the information feedbacked from the radar receiver, makes the transmitted waveforms to optimally match to the radar scene environment and then gives the best transmitted waveforms sequence.In this paper, we describe the development of radar waveform design methods, analyze the traditional waveform design methods and waveform design methods based on the information theory systematically, study the waveform selection method based on adaptive dynamic programming fully. On this basis, we analyze the solving method of the adaptive waveform selection problem under the partially observed Markov decision problem model, propose the Bellman's optimal equation of waveform selection problem. Under the detection radar application, we give two methods to get the optimal waveforms sequence:1. In this paper, we use the backforward dynamic programming algorithm to solve the finite horizon adaptive dynamic programming waveform selection problem, get an optimal transmitted waveforms sequence.2. The traditional value iteration methods of solving the Bellman's optimal equation all need the state observed transition probabilities. However, these probabilities are hard to obtain in real world, so we propose a Q-learning method to solve the optimal equation. In this method, we don't need the knowledge of the state observable transition probabilities and can low down the compute duty and the storage space, and then give a suboptimal policy compared with the backforward dymanic programming algorithm.The theory analysis and computer matlab simulation experiment all show that the above two methods are practical, effective and easy to realize.