Study On Adaptive Target Detection For Multi-Channel Array In Complex Scenarios

The main task for radar is to detect a target in the background of noise. It plays an important role in military and civil areas, including the military applications of the surveillance of land and oceans, air defense early warning, navigation and weapon guidance, and the civil applications of traffic detection, weather prediction, seismic exploration and monitoring sea smuggling.The classic multi-channel array adaptive detection is often applied in the background of homogeneous Gaussian noise. In addition, it needs to satisfy these requirements, such as a number of adequate training data that is identical independent distribution, the exact steering vector of the target is known, and the target is regarded as a point–like target. A lot of investigations show that the background of adaptive detection becomes complex due to the space change of the geographical environments, tall buildings, man-made strong scatterer, the special structure of array, the increasing resolution of radar. The complex scenario includes the heterogeneous non-Gaussian characters of clutter, the limited training data, the range-spread target in high resolution radar systems, the mismatch between the presumed steering vector and the actual steering vector. This will cause severe performance degradation of adaptive detectors. Therefore, it is crucial to study robust effective detection algorithms in complex scenarios.This dissertation focuses on the problems for multi-channel array adaptive detection in the complex scenario, i.e., deficient training samples, non-Gaussian noise background, the mismatch of steering vector, the subspace interference. The structure information of the covariance matrix and the prior knowledge-aided information are exploited and partially homogeneous and compound-Gaussian processing are considered in this thesis. This dissertation studies the adaptive detection in the complex scenario, including non-Gaussian non-homogeneous target detection algorithms, range-spread target detection algorithms, robust adaptive beamforming, knowledge-aided space-time adaptive processing methods, and parametric detection method and so on. A brief summary is provided as follows:1. The problem of detecting a signal in partially homogeneous and homogeneous environments is addressed. In partially homogeneous environments, i.e., both the test data and training data share the same noise covariance matrix structure up to an unknown scaling factor, a persymmetric adaptive coherence estimator(Per-ACE) detector is proposed. By exploiting the persymmetric structure of the covariance matrix, the Per-ACE can reduce training data requirements. Furthermore, the expressions for the probabilities of false alarm and detection are derived along with the distribution of the loss factor. In homogeneous environments, a persymmetric adaptive matched filter(Per-AMF) detector has been presented. However, its probability of detection has not been obtained yet. Thus, we derive the expression for the probability of detection. For both the Per-ACE and Per-AMF, numerical results of these proposed expressions are confirmed with those of Monte Carlo trials. In addition, simulation results show that the proposed Per-ACE outperforms the conventional ACE in training-limited scenarios.2. The problem of detecting a signal in non-Gaussian clutter is dealt with in this part. Two methods to detect a target in compound-Gaussian clutter with random texture are proposed. First, the texture component in a compound-Gaussian process is modeled as a random variable, with Gamma, inverse Gamma or inverse Gaussian distribution. Then, the generalized likelihood ratio principle is used to derive the detectors in compound-Gaussian clutter with random texture. Moreover, the constant false alarm rate(CFAR) property is analyzed. Finally, the detectors are assessed by Monte Carlo simulations. Performance comparison of the proposed method with the traditional methods shows that the former is effective in compound-Gaussian clutter with random texture3. We focus on the problem of point-like and range-spread target detection in small sample support condition, and propose the knowledge-aided point-like target detection and range-spread target detection algorithms. For the point-like target, the covariance matrix in compound Gaussian noise is modeled as a random matrix, the prior distribution of which satisfies the complex inverse Wishart distribution. With prior distribution, the maximum a-posterior estimation of the covariance matrix is derived. Then, Rao detection is obtained based on the maximum a-posterior estimation. With the development of the high resolution radar, the size of the target is longer than the range resolution, and the target distributes over several range cells. Thus, the target is called the range-spread target. For the range-spread target, first, the covariance matrix in compound Gaussian noise is assumed to follow the complex inverse Wishart distribution. Then, the maximum a-posterior estimation of the covariance matrix is given. Next, the Bayesian Rao detection and Wald detection for distributed targets are derived. The proposed detectors combine the prior information of the covariance matrix and the training data. Simulation results show that the detection performance of the proposed detectors outperforms the traditional detectors when the amount of training samples is small.4. In order to improve the performance degradation in clutter suppression due to the steering vector mismatch and small sample support, we propose knowledge-aided adaptive beamforming and space-time adaptive processing approaches, including the following aspects:(a) a knowledge-aided adaptive beamforming approach against signal steering vector mismatch with small sample support is proposed. First, a combination covariance matrix of the prior covariance matrix, the sample covariance matrix and an identity matrix is obtained by the minimum mean-square error criterion. Then the mismatch between the actual and the presumed signal steering vector is modeled as a worst-case performance optimization constraint. Based on the combination covariance matrix and the constraint, the adaptive beamformer is formulated. And the problem is solved by second-order cone programming.(b) A novel robust structure-based beamforming algorithm is proposed. It includes three steps. First, the original samples are preprocessed by a window smoothing scheme. Given the processed data, the final covariance matrix can be constructed with the Toeplitz or the centro-hermitian structure information. Finally, the steering vector is estimated based on the angular sector of the mismatched steering vector. Simulation results show that the proposed method can not only mitigate the performance degradation caused by small sample support, but also is robust against the steering vector mismatch.(c) A knowledge-aided space-time adaptive processing approach is proposed, which mainly deal with the problem of solving the color loading factors. First, the knowledge-aided LCMV model is given. Then, the formula of the relationship between the color loading factors and the constraint constants is derived. Finally, the two color loading factors are solved. Simulation results show that the performance of the output signal-interference noise ratio loss is improved, which validate the effective of the proposed approach.5. A parametric detection method in partially homogeneous environments is proposed to improve the performance degradation due to the small sample support. In this method, the disturbance is modeled as a multichannel autoregressive process. To mitigate the effect of limited training samples, the subspatial aperture smoothing is performed in this method. Moreover, the persymmetric structure information is exploited to further reduce the sample requirements. The performance of the proposed method is assessed by numerical examples. Results show that this method outperforms other traditional detectors in sample-deficient scenarios.6. To detect range-spread targets for a high resolution radar, we propose a range-spread target detection algorithm in homogeneous and non-homogeneous environments plus subspace interference. Under the Bayesian frame, the subspace data model is presented. With the prior probability density of the covariance matrix, the generalized likelihood ratio principle is used to derive the one-step detection algorithm and two step procedure algorithm in homogeneous background. In addition, the adaptive detection algorithm in non-homogeneous is proposed. Simulation results indicate that the proposed Bayesian detection strategies outperform their competitors in training-limited scenarios.