Radar Transmitting Resource Management Method For Target Parameter Estimation

Being the main warning and detection equipment in defense system,radar plays a very important role in national defense system.With the increase of complexity of battlefield environment and the rapid development of penetration technology,modern radar has to face the challenge of multi-target saturation attack and multi-task.However,the transmitting resources of radar system are always limited.Therefore,how to schedule the limited radar resources efficiently and reasonably to maximize the radar performance is of great significance.This dissertation focuses on signature extraction of space cone targets and cognitive tracking method,and studies radar transmitting resource management method for target parameter estimation based on digital array radar technology and cognitive radar technology.The main work is described as follows:1.With respect to the problem that long-dwell observation time is required by micro-motion feature extraction of space precession cone target,an optimization-based multiple shortdwell observation(MSDO)strategy is proposed.Its aim is to achieve the accuracy requirement of micro-motion signature extraction with minimum time resource consumption.Due to the non-stationarity of precession cone target echo signal,an estimation bias will inevitably be caused by using short-time Fourier transform to estimate the instantaneous frequency,which will degrade the micro motion feature extraction accuracy.In this dissertation,this instantaneous frequency estimation bias is introduced into the derivation of Cram é r-Rao lower bound of micro-doppler parameter estimation,and an accurate mathematical relationship between signature extraction accuracy and MSDO parameters is theoretically established,which provides a more reliable and accurate performance evaluation for MSDO parameter optimization.2.To handle the mismatch problem of state transition model in maneuvering target tracking,a maneuvering target resource allocation method is proposed based on recurrent neural network(RNN)prediction.Considering that the state transition model of maneuvering target is generally not available,using the alternative model to predict the target information will inevitably cause performance loss,and then lead to inefficient resource allocation results.This dissertation studies the problem based on the idea of data-driven,and trains an RNN to learn the state transition model of maneuvering target from historical data,to realize the information prediction of maneuvering target.Compared with the model-based method,this method can learn more accurate state transition model of maneuvering target from massive data,and get more accurate target prediction information while getting rid of the shackles of the model.3.With respect to the problem of radar power allocation that takes long-term and short-term multi-target tracking performance into consideration,a deep reinforcement learning(DRL)-based power allocation method is proposed,and a cognitive resource allocation framework is established based on deep learning.The aim is to maximize the performance of long-term and short-term multi-target tracking by optimizing power allocation with limited power constraint.If optimization method is used to solve the problem,it is necessary to obtain accurate multi-step prediction and tracking performance evaluation of the target state.However,in practice,the accuracy of multi-step prediction cannot always be guaranteed.Especially in the maneuvering target scenario,the model mismatch will further degrade the prediction accuracy.Therefore,a resource controller is designed based on DRL.Compared with the optimization-based method,this controller can avoid the multi-step prediction problem and complicated solving process,and ensure the real-time resource allocation.Furthermore,by introducing an RNN-based memory,a deep learning-based cognitive resource allocation framework is established.The framework further is able to improve the cognitive ability of radar system by taking advantage of the decision-making ability of DRL and the ability of memory and prediction of RNN.4.Based on the established cognitive resource allocation framework,an in-depth study is carried out on the problem of joint allocation of multiple resources.The main studies are described as follows:(1)For multi-target tracking,a joint power and beam allocation method is proposed based on DRL,which aims to improve the performance of long-term and shortterm multi-target tracking as much as possible with limited resource constraints.To address this problem,the resource controller is redesigned to generate beam and power allocation parameters.At the same time,the target RCS prediction is added to the memory so that the influence of target RCS can be taken into resource allocation.(2)For multiple radar system(MRS),a joint radar selection and power allocation method is proposed based on constrained DRL.The purpose is to minimize the power consumption of MRS meeting the tracking accuracy requirements,in which the tracking accuracy requirement is introduced into the policy gradient of DRL using Lagrange relaxation technique,so that the resource allocation strategy learned by DRL can achieve the preset tracking accuracy requirement.