Cognitive Radar Signal Processing-Detection And Tracking

Cognitive radar can adaptively select the configuration of the transmitter and the formof radar signal processing according to the characteristics of the targets and environment,and can improve the performance of detection and tracking by taking advantage of allavailable prior information. Cognitive radar is a new trend of modern radar technologydevelopment.This thesis focuses on the receiver of cognitive radar, and mainly discusses how totake advantage of available prior information to improve the performance of detectionand tracking. The track-before-detect algorithm based on Bayesian theory (Bayesiantrack-before-detect algorithm) is detailedly studied, and a detailed analysis of thecalculation steps of Bayesian track-before-detect algorithm is carried out. Further,modifications to the calculation steps of Bayesian track-before-detect algorithm aremade in this thesis in order to improve the performance of detection and trackingcompared with the traditional Bayesian track-before-detect algorithm. In addition, in thecase of the balance of the communication burden and the performance of detection andtracking, Bayesian track-before-detect algorithm is extended to multisensor fusionsystems in this thesis.The research works done in this thesis are summarized as follows:1. This thesis studies the radar detection method which takes advantage of sequentialdecision information, which includes how to modify the detection rule by takingadvantage of the available decision information in order to help improve theperformance of detection, the approach of determining the detection threshold and theevaluation of the proposed radar detection method.2. This thesis studies Bayesian track-before-detect algorithm, makes a detailedanalysis of Bayesian track-before-detect algorithm, and makes some modificationsabout the modeling of target existence state variable, the way of processing theunknown parameters of target motion model and system observation model, and themethod of estimating target motion state. This thesis detailedly introduces the basicprinciples of Bayesian track-before-detect algorithm, and gives the calculation flow ofBayesian track-before-detect algorithm. For the traditional Bayesian track-before-detectalgorithm, target existence state is often modeled as a homogeneous Markov chain;however, after some derivation, this thesis concludes that the homogeneous Markov chain does not provide any prior information on whether the target exists, which resultsin the degradation of the performance of detection. Therefore, this thesis models targetexistence state as an inhomogeneous Markov chain whose transition probability matrixis related with the posterior probability of target existence, which is line with thetransitions of target existence state and can further improve the performance ofdetection. When there are unknown parameters in target motion model or systemobservation model, the common trick is to treat these unknown parameters as extra statevariables and approximate these unknown parameters as some slowly varying Gaussprocesses for Bayesian track-before-detect algorithm. The trick can not integrate the realmodel of the unknown parameters dealing with the unknown parameters, and theestimates of the unknown parameters heavily depend on recent observations. This thesisapplies the Expectation Maximization algorithm to Bayesian track-before-detectalgorithm in order to obtain better estimates of the unknown parameters, which does nottreat the unknown parameters as extra state variables and can obtain the maximumlikelihood estimates of the unknown parameters by making full use of all observations.Therefore, better estimates of the unknown parameters are obtained, and theperformance of detection and tracking is improved. In the low signal-to-noise ratio case,there will be a lot of intensities of noise cells which are greater than the intensity of thetarget cell, which will lead to the multi-model characteristic of the posterior density oftarget motion state. In the particle-filter-based Bayesian track-before-detect algorithm,the estimate of target motion state will be highly affected by the noise if the particles aredirectly used to estimate target motion state. At different times, target moves accordingto target motion model; but the noise cells whose intensities are large are independent.Using the motion characteristics of target and noise, this thesis proposes a novel stateestimate approach which is based on the Weighted Rival Penalized CompetitiveLearning and the Dynamic Programming in order to alleviate the effect of the noise onthe estimate of target motion state.3. This thesis studies the approach of determining the detection threshold forBayesian track-before-detect algorithm. Because of the high nonlinearity andhigh-dimensional calculus involved in Bayesian track-before-detect algorithm, it isdifficult to quantitatively analyze the detection performance of Bayesiantrack-before-detect algorithm, and difficult to determine the detection threshold underNeyman-Pearson criterion. Under Neyman-Pearson criterion, this thesis derives theexpression of the test statistic starting from the likelihood ratio testing form, and obtainsthe relationship between the detection threshold and false alarm probability. Further, this thesis gives the approximate closed-form solution of the detection threshold under whiteGaussian noise background. Therefore, the detection threshold can be set in accordancewith the required false alarm probability in real time.4. This thesis studies Bayesian track-before-detect algorithm for multi-sensor systems.In the case of the balance of the communication burden and the performance ofdetection and tracking, this thesis proposes a distributed Bayesian track-before-detectalgorithm and a distributed Bayesian track-before-detect algorithm with feedback.Compared with the centralized Bayesian track-before-detect algorithm, the proposeddistributed Bayesian track-before-detect algorithm can greatly reduce the amount ofdata which is supposed to be communicated from the sensors to the fusion center, butthe performance of detection and tracking suffers a loss. Compared with the centralizedBayesian track-before-detect algorithm, the proposed distributed Bayesiantrack-before-detect algorithm with feedback can reduce the amount of data which issupposed to be communicated from the sensors to the fusion center, but the performanceof detection and tracking suffers a slight loss. Compared with the proposed distributedBayesian track-before-detect algorithm, the proposed distributed Bayesiantrack-before-detect algorithm with feedback can improve the performance of detectionand tracking significantly especially in the low signal-to-noise ratio case, but theamount of data which is supposed to be communicated from the sensors to the fusioncenter suffers a slight increase.