Study On Weak Moving Targets Detection And Localization For Space-based Early Warning Radar

The modern space-based early warning radars play an importantly military role in surveiling the air and space moving targets. However, it encounters with the challenge of detection, tracking, and positioning of maneuvering targets under low signal noise rate (SNR), which has attracted the attention of researchers all over the world. The SNR can be improved by increasing integration time. However, the signal energy could not be effectively accumulated by FFT-based traditional methods, because the high-speed and maneuvering motion of a target induces range migration in the long-time coherent integration period. Track-before-detect (TBD) technology directly makes correlation between multiple image data, which reserves most of information. Therefore, TBD is especially applicable for the target detection and tracking in the case of low SNR. In addition, in order to detect the space target, it is reqired to transmit wideband signal. In the meantime, the signal information can be sufficiently exploited using the wideband signal to position the moving target. This thesis investigates the challenging problem of detection and positioning for the weak targets in the terms of long-time coherent integration, TBD based on particle filter (PF) and wideband direction-of-arrival (DOA) estimation. The main content of this thesis can be summarized as follows:1. In this section, the model of a moving target and the long-time energy accumulation for the weak targets, which involve in the detection and tracking for the weak targets by the space-based early warning radar, is introduced. This provides reference for the subsequent research.2. Increasing integration time can improve the detection performance of weak targets. However, the signal energy could not be effectively accumulated, because the high-speed motion of a target induces range migration in long-time coherent integration period. In Chapter two, two different approaches are addressed for long-time coherent integration. Firstly, the signal is segmented based on the influence of echo's high order phase history on motion compensation. Secondly, the range migration is corrected with keystone transform and then the phase history is estimated with Fractal Fourier transform (FRFT) for each sub-segment. Finally, the motion parameters are refined by using least square algorithm among sub-segments and thus the long-term coherent accumulation can be achieved. For the moving target with constant acceleration and Doppler ambiguity, a new method is proposed by constructing the two-dimensional compression function in range-Doppler frequency domain, which can eliminate the coupling effect between range and azimuth directions. This method makes matching processing by one-dimensional search for the acceleration and does not require target velocity parameters. This method can correct the range migration caused by radial velocity and radial acceleration. The proposed algorithm can be efficiently implemented by using fast Fourier transform without interpolation and thus has low computational complexity. Simulation results show that the proposed algorithm improves the performance for detecting high-speed maneuvering targets.3. TBD based on particle filter algorithm can effectively handle the problem of the nonlinear, non-Gaussian and multi-modal state estimation, especially applicable for the detection for maneuvering and weak targets. Unfortunately, the traditional PF methods are apt to induce collapse and diversity loss of particles. Two improved TBD algorithms based on the particle filter are proposed to detect and track the weak target in low SNR. Firstly, an updating strategy is proposed by replacing the existing particles with low weights with new particles and then performing Markov chain Monte Carlo (MCMC) moving step after resampling particles. This strategy can improve the diversity among the particles, and simutanouesly guarantee that the particles are effective. Simulation analysis is given for the effect of the number of particles and the detection threshold on target detection. The other algorithm is presented by combining the particle filter with unscented Kalman filter (UKF). Because in the proposed method, the important probability density distribution is calculated based on the current measurement, the sampling particles are most likely to be in the region of high likelihood, which makes the particles distribution more approach to the posterior distribution of the state. Simulation results show that the proposed algorithm provides an improved performance of detecting and tracking weak targets compared with the conventional particle filter.4. An improved TBD algorithm based on the particle filter is applied for high-speed weak target detection. This method uses a unique measurement model for radar range-Doppler compression, which can effectively reduce the model error of the traditional sensor point spread function. An updating strategy is developed by the following step:newborn particles are uniformly distributed within the set with high-intensity bins and then that the existing particles with low weights are replaced by new particles with probability. This strategy can improve the diversity among the particles, and mitigate the effects of degeneracy. Simulation results show that the proposed algorithm has an improved performance of detecting and tracking dim target compared with the standard particle filter.5. It is generally to use the high resolution in range direction to indentify space target. However, the resource in the satellitic plateform is limited and thus it is necessary to exploit the signal resource. Therefore, it is significantly meaningfull to estimate DOA along with indentifying the target with wideband signals. It is known that the performance of the coherent signal-subspace method (CSM) algorithm degrades due to the estimation difference of the covariance matrix in the case of small data. Aiming to address this problem, chapter 5 proposes a new method for estimating the DOA of wideband signal, which employs particle filters to track array manifold at different frequency bands. Compared with the CSM, the proposed method utilizes the current observed data and does not require the estimated covariance matrix, thus it performs better in the case of a small sample set. In the meantime, since initial values of the parameters in this method can be selected arbitrarily, thus this method does not require the preliminary DOA estimates. Moreover, the proposed method can localize completely correlated sources because it is based on the idea of maximum likelihood. Simulation results show that the performance of the propose method is better than CSM when the sample set is small, the SNR is low, and the signal sources are correlated.