Research On Behavior Recognition And Prediction Techniques Against Multi-Function Radar

As a key part of the radar countermeasure(RCM),radar electronic reconnaissance plays a crucial role in modern electronic warfare(EW).However,with the development and application of the active electronically scanned array(AESA),multi-function radar(MFR)has become a sensing system with multiple operating modes and multiple tasks that are of great intelligence,flexibility,and adaptability.The extensive applications of MFRs have posed a great challenge to the technologies of RCMs and electronic intelligence(ELINT).Therefore,in the game relationship with the powerful and complete MFRs,only if ELINT systems sense the dynamical changes of the MFR inner states timely and accurately,can the RCM equipment be guided to make the optimal jamming decisions agilely.For this reason,the cognition to MFRs has become an urgent problem in the field of RCM.In order to describe the MFR signals more accurately,the concept of “radar behavior” is defined as “anything that an MFR does involving inner action and outer response to warfare stimulation and electromagnetic environment”.With the background of electronic countermeasure against the air-defense radar network,the characteristics of MFR behaviors is studied by analyzing the intercepted signals based on the theoretical tools such as syntactic pattern recognition,dynamical systems and artificial neural networks(ANN).In this dissertation,the cognition to MFR behaviors is divided into four progressive branches that are the characterization,modeling,recognition and prediction of radar behaviors,and presented as follows:Firstly,the method is investigated for describing the MFR behavior characteristics which is suitable for the complex and flexible MFR signals.The radar word after signal sorting and word extracting for a single radar is used as the carrier of MFR behaviors,and the Markov property in the MFR word sequence is analyzed.The hierarchical structure of radar signal is extended from specific pulse-Doppler(PD)radar to all the pulse radars,by improving some of its conceptions.With the extended model which is able to utilize the information more efficiently,two methods for radar word extraction are proposed,making the radar word easier to be recognized and analyzed.Secondly,the MFR radar word sequence is viewed as a time-discrete dynamical system,and the reverse modeling method for MFR behaviors is studied based on the Predictive State Representation(PSR).On the base of the normal algorithm for PSR discovering and learning,an algorithm for PSR modeling and training is proposed and presented in detail,which is improved and optimized according to the flexibility and diversity of MFR signals.Analysis shows that the improved algorithm is able to effectively reduce the complexity,and simulations demonstrate that the novel method performs well in MFR signals modeling.Thirdly,the methods for MFR operating mode recognition are studied.In order to constructing the automaton of MFR and achieving the operating mode recognition,the knowledge-assisted and data-driven modeling methods are respectively proposed based on the MFR synthetic model.A grid-filter algorithm is proposed based on the PSR framework of MFR,which has better performance in recognition for the advantage of PSR in dynamical system description compared with hidden Markov model(HMM).For the problem of unavailable transition probabilities,an algorithm is proposed by accumulating the predictive states,which is helpful to reduce the dependence of MFR prior information.Finally,the methods are studied for MFR signal sequence prediction.The framework is developed for MFR signal prediction,and a normal algorithm based on linear PSR predictor is proposed,which is able to reduce the computational complexity.Another prediction algorithm is proposed by viewing the predicted observation of each step as a known condition,thus avoiding the large computation brought by the uncertainty of future observations.Combining PSR with the ANN by analyzing the settings of network layers and choices of functions,the trained neural network is used for MFR signal prediction,performing well in prediction accuracy.The work of this thesis completes the theory of MFR behavior description,and the reverse model of MFR is constructed by utilizing the intercepted radar signals.Several methods are proposed for MFR behavior recognition and prediction,which are of great significance in the intelligent jamming decision and adaptive RCMs,supporting the practical application of cognitive electronic warfare.