【作者】 吴建新；

【导师】 保铮；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2009， 博士

【摘要】 空时二维自适应处理能够有效提高机载相控阵雷达的地杂波抑制性能和动目标检测性能。理论上,全空时自适应处理可以实现最优处理,但是其计算量和实现的复杂度令人们难以接受,并且难以获得估计协方差矩阵所需的足够多的样本数。后来的研究重点就转移到降维处理算法上。研究人员在降维算法的研究上做了大量的工作,提出了很多种方法,发表了大量的文章。随着空时二维处理研究的不断深入,研究人员对强孤立杂波和强运动目标等非均匀环境的杂波抑制问题、杂波二维分布随距离变化的非平稳杂波抑制问题以及运动目标检测后的运动目标参数估计问题进行了广泛的研究,得到了很多有效的方法。降维处理、非均匀杂波抑制、非平稳杂波抑制和运动目标参数估计是空时自适应处理工程化必须解决的四个主要问题。本文主要围绕这四个方面做了一些工作,概括如下:利用Kronecker积的性质,提出了采用一维FFT实现自适应方向图和一维输出信杂噪比的快速计算,采用二维FFT实现空时二维频率响应和二维输出信杂噪比的快速计算。在滑窗距离样本选取方法的基础上,提出了滑窗递推QR分解方法。它采用样本矩阵QR分解得到空时处理的自适应权,这样矩阵条件数比较小,不容易受到有限字长等因素的影响,具有更好的数值稳定性。它通过双曲线Householder变换实现滑窗距离样本选取方法的权值递推,可以明显减少空时自适应处理权值计算所需的计算量。利用机载非正侧阵雷达的近程杂波和目标在俯仰角度域可以区分的特点,提出了两种近程杂波抑制方法:第一种方法是基于DOA估计的俯仰投影矩阵方法,它首先采用DOA估计方法估计出近程杂波的俯仰角,然后通过投影方法抑制该近程杂波,相比于直接计算近程杂波俯仰角度的俯仰非自适应处理方法,该方法不需要先验信息,可以避免地形起伏对近程杂波俯仰角度计算的影响;第二种方法是稳健的俯仰自适应波束形成方法。由于目标俯仰角度的任意性,我们不可能对某一个具体的俯仰角度保持固定增益,只能是对目标可能出现的一个俯仰角度范围保持一个稳定的增益,然后尽可能地抑制其它俯仰角度的杂波,此时在俯仰角度上离目标比较远的近程杂波就会被抑制。在没有阵元幅相误差的情况下第二种方法的性能不如第一种方法,但是在有阵元幅相误差的情况下该方法的性能要明显好于第一种方法。提出了一种适合于相邻距离单元统计特性完全不同的极端非同态环境的运动目标检测方法。不同于一般的空时自适应处理方法,该方法直接在功率谱进行运动目标检测。该方法首先采用空间滑动和时间滑动来获得样本用于协方差矩阵估计,然后用估计得到的协方差矩阵进行功率谱估计。为了减少功率谱估计的计算量,我们提出了采用二维FFT实现功率谱估计。估计得到数据功率谱后,利用杂波功率谱能量比较大的特点,提取出功率谱中的大能量点用于杂波谱线拟合。拟合出杂波谱线后,利用目标到杂波谱线的距离不为零的特点进行目标检测,同时实现目标参数估计。在这个过程中需要对原始的数据功率谱进行两次门限检测,第一次门限检测是提取出大能量的杂波谱用于杂波谱线拟合,第二次门限检测是提取出所有可能的目标用于目标检测。提出了两种针对均匀线阵的目标角度估计方法。第一种方法是基于实多项式求根的最小二乘方法,它首先采用空时自适应处理方法得到多个杂波抑制后的空间通道,然后用这些空间通道的数据进行最小二乘目标角度估计。为了避免复杂的角度搜索,该方法采用实多项式求根将角度搜索转化为代价函数的极值搜索,大大减少了角度估计的计算量。第二种方法是基于实多项式求根的最大似然方法,它直接利用数据的最大似然函数来估计目标角度,为了避免复杂的角度搜索,它同样采用了实多项式求根方法。提出了两种针对平面阵的目标方位角度和目标俯仰角度估计方法。第一种方法是基于交替最大化和实多项式求根的最小二乘方法,它首先采用空时自适应处理方法得到多个杂波抑制后的空间方位通道和空间俯仰通道,然后用这些空间通道的数据进行最小二乘目标角度估计。为了避免复杂的二维角度搜索,该方法采用交替最大化方法将方位角和俯仰角联合估计转化为方位角和俯仰角分别迭代估计,然后采用实多项式求根将角度搜索转化为代价函数的极值搜索,大大减少了角度估计的计算量。第二种方法是基于交替最大化和实多项式求根的最大似然方法,它直接利用数据的最大似然函数来估计目标方位角度和俯仰角度,为了避免复杂的二维角度搜索,它同样采用了交替最大化方法和实多项式求根方法。更多还原

【Abstract】 Space-time adaptive processing (STAP) provides great potential over improving the performance of the planar array radar on clutter suppression and moving target detection. Fully STAP can obtain the optimal performance in theory. However, the computation load and the implementation complexity of Fully STAP are too expensive to be accepted and it is really very difficult to obtain enough secondary data to estimate the covariance matrix. The focus of successive research is concentrated on the reduced-dimensional algorithms and many reduced-dimensional methods have been proposed. With the development of STAP, clutter suppression in nonhomogeneous environment, such as strong moving targets and strong isolated clutter, nonstationary clutter suppression and target parameters estimation are widely investigated and various methods are developed. The aforementioned aspects are mainly concerned in this dissertation, and they are summarized as follows:Utilizing the characteristic of the kronecker product, fast evaluation of the adaptive direction diagram and the output signal-to-clutter-plus-noise ratio (SCNR) using 1D-FFT are presented, and fast implementation of the two dimensional frequency response and the two dimensional output SCNR using 2D-FFT are also developed.Based on the sliding window range secondary selection method, the sliding window recursive QR decomposition method is presented. The presented method computes adaptive weight by QR decomposition of the secondary data matrix. Since the condition number of a covariance matrix is the square of that of a secondary data matrix, and thus the numerical stability of STAP based on QR decomposition of the secondary data matrix is much better than that of STAP based on covariance matrix inversion. Moreover, the presented method realizes weight recursion by the hyperbolic Householder transformation and greatly reduces the computational load of the computation of the weight vector.Considering the fact that the near-range clutter and moving targets are distinguishable in elevation space, two methods for range-dependent clutter suppression are developed, where the first method is based on the projection matrix method, the second method is based on the robust beamforming method. The former method estimates the elevation angle of the near-range clutter by spectral Capon rooting method, and the projection matrix is then constructed for suppressing the near-range clutter according to the estimated elevation angle. Compared to the nonadaptive method for range dependent clutter suppression, the method is free of priori knowledge. Due to the arbitrary of the moving target, the elevation angle of the moving target can not be accurately determined. The target signal will be treated as an unwanted interference signal by the adaptive processor and therefore will tend to be suppressed. It dramatically degrades the output signal-to-interference-plus-noise ratio (SINR). In order to avoid the target signal cancellation, the latter method is presented for improving the robustness of the elevation adaptive beamformer. Thus, the near-range clutter will be suppressed, while the far-range clutter and the moving target will be preserved. Under the ideal case that no array errors are occurred, the performance of the first method is slightly better than that of the second method. However, under the non-ideal case that array errors are occurred, the performance of the second method is much better than that of the first method.A moving target detection method based on power spectrum suitable to the extremely nonhomogeneous environment is proposed. Different to the general STAP methods, the presented method realizes moving target detection in the spatial-temporal spectrum domain. The secondary data obtained by spatial smoothing and temporal smoothing in single range unit can be considered identically distributed. Therefore, these secondary data can be used for covariance matrix estimation and then the power spectrum is estimated from the estimated covariance matrix. To alleviate the computation load of the power spectrum computation, 2D FFT is proposed to estimate the power spectrum. Utilizing the fact that the mainlobe clutter energy is much stronger than the noise power and the target power, the points with strong energy in the power spectrum can be extracted to fit the clutter ridge. Next, considering the fact that the distance between the moving target and the clutter spectrum is larger than zero, moving target detection can be implemented in the power spectrum. In this method, twice threshold detection are implemented, where one is implemented to extract the clutter with strong energy for clutter ridge fitting, the other is implemented to extract the targets for moving target detection.Two methods of target parameters estimation for uniform linear array (ULA) are developed. The first method estimates the target angle from the clutter-suppressed data using the least squares method. To avoid the grid search, the real polynomial rooting method is employed to convert the grid search to the extreme value search. The second method estimates the target angle from the likelihood function. Similarly, to avoid the grid search, the real polynomial rooting method is also used.Two methods of target parameters estimation for planar array are developed. The first method is the least squares method employing the alternating maximization method and the real polynomial rooting method. The second method is the maximum likelihood method employing the alternating maximization method and the real polynomial rooting method. The first method estimates the target angle from the clutter-suppressed data using the least squares method. To avoid the two-dimensional grid search, the alternating maximization method is used here to convert the joint estimation of the azimuth angle and the elevation angle to the iterating estimation. Furthermore, the real polynomial rooting method is employed to estimate the target angle instead of the grid search. The second method estimates the target angle from the likelihood function. Similarly, to avoid the two-dimensional grid search, the alternating maximization method and the real polynomial rooting method are also utilized.更多还原