Treffer: Advanced array signal processing algorithms for multi-dimensional parameter estimation

Multi-dimensional high-resolution parameter estimation is a fundamental problem in a variety of array signal processing applications, including radar, mobile communications, multiple-input multiple-output (MIMO) channel estimation, and biomedical imaging. The objective is to estimate the frequency parameters of noise-corrupted multi-dimensional harmonics that are sampled on a multi-dimensional grid. Among the proposed parameter estimation algorithms to solve this problem, multi-dimensional (R-D) ESPRIT-type algorithms have been widely used due to their computational efficiency and their simplicity. Their performance in various scenarios has been objectively evaluated by means of an analytical performance assessment framework. Recently, a relatively new class of parameter estimators based on sparse signal reconstruction has gained popularity due to their robustness under challenging conditions such as a small sample size or strong signal correlation. A common approach towards further improving the performance of parameter estimation algorithms is to exploit prior knowledge on the structure of the signals. In this thesis, we develop enhanced versions of R-D ESPRIT-type algorithms and the relatively new class of sparsity-based parameter estimation algorithms by exploiting the multi-dimensional structure of the signals and the statistical properties of strictly non-circular (NC) signals. First, we derive analytical expressions for the gain from forward-backward averaging and tensor-based processing in R-D ESPRIT-type and R-D Tensor-ESPRIT-type algorithms for the special case of two sources. This is accomplished by simplifying the generic analytical MSE expressions from the performance analysis of R-D ESPRIT-type algorithms. The derived expressions allow us to identify the parameter settings, e.g., the number of sensors, the signal correlation, and the source separation, for which both gains are most pronounced or no gain is achieved. Second, we propose the generalized least squares (GLS) algorithm to solve the ...