Reduced Rank Filtering Techniques for Processing Multi-Aperture Radar

Using a non-uniformly distributed aperture radar system for forming a SAR image will result in data correlations between the SAR image resolution cells. Thus, this requires that a more robust filter than the Matched Filter, i.e. the MMSE or Wiener Filter to be used in the receiver processing. As the Wiener Filter involves a computationally expensive matrix inverse operation, it can be avoided by using the Kalman filter. But the error covariance matrix computation in the Kalman filter can become unstable from finite machine precision in conjunction with large variations of the covariance matrix Eigen values. However, this instability can be overcome by using the Square Root Covariance Filter (SRCF) that ensures that the resulting error covariance matrix will always remain positive definite after each measurement update. Besides the Kalman filter, a recent algorithm, namely the Multi-Stage Wiener Filter (MSWF) has been developed to overcome the matrix inverse problem in the Wiener Filter as well. Using orthogonal projections in each successive stage of decomposition, this filter is proven to achieve the same performance as the Wiener filter in a shorter computation period. In this thesis, the performance of the SRCF and the MSWF used to form a SAR image is evaluated as compared to the Wiener filter and the Kalman filter using data from an existing radar model simulator. In addition, the use of reduced rank techniques is applied to both algorithms so as to trade off between computation time