| In today’s complicated military environment,multifunctional phased-array radars play a critical role.Because electronic reconnaissance,as a pioneer of electronic warfare,can provide operational intelligence and influence the deployment of operational strategies through intercepted enemy radar radiation sources,the problem of identifying radar working modes as an important component of electronic warfare has attracted research efforts.This thesis is first carried out at the phased array radar system level,starting from a block diagram of the phased array radar system structure and focusing on several modules including antenna radiation patterns,waveform design,beam position arrangement and resource scheduling.In the waveform design section,the LFM signal is used,the generation of the echo signal is simulated,and the role of the resource scheduling module is demonstrated to establish a system model in preparation for further analysis of the characteristics and modelling of different working modes in subsequent chapters.Secondly,the characteristics of search and tracking signals under four classical multifunctional phased array radar working modes are analyzed,reconnaissance interception models are established,and the four radar working modes are modelled on top of the modelling of the phased array radar system level in combination with the set target motion scenarios,and the airspace pulse signal schematics of the four working modes are simulated,in order to provide data for the subsequent study of intelligent identification methods of the working modes.The simulation of the airspace pulse signal schematics of the four working modes provides data guarantee for the subsequent research on the intelligent identification of the working modes.Finally,three recognition methods are investigated.Firstly,the BP neural network is combined with the Ada Boost algorithm,which belongs to the Boosting type algorithm in integrated learning,to propose a radar working mode recognition model based on BP-Ada Boost,and it is compared with the improved fuzzy clustering method in the unsupervised method to prove the superiority of the BP-Ada Boost model compared with the unsupervised clustering,but it is found that the BP-Ada Boost model has poor recognition stability and cannot solve the working mode misclassification problem well.To address the working mode misclassification problem,the one-dimensional CNN is proposed for the recognition of radar working modes by taking advantage of the one-dimensional CNN in processing sequential data,and the effectiveness of the one-dimensional CNN in the radar working mode recognition problem is clarified through network structure optimisation and other comparison experiments.For the small sample working mode recognition problem,the long short-term memory network is used to store the state information,and the radar working mode recognition model based on the LSTM network is proposed.The deep LSTM network model is established by superimposing the LSTM network,and the feasibility of the network for the small sample radar working mode recognition problem has been demonstrated by simulation experiments.. |
