Performance Analysis of the
Eigencanceler-Eigenanalysis Based Space-Time
|Alex M. Haimovich
New Jersey Institute of Technology
Department of Electrical and Computer Engineering
University Heights, Newark, NJ 07102
Abstract In this presentation we suggest the eigencanceler for space-time adaptive radar and analyze its performance. The term eigencanceler refers to a suboptimum processor derived from the principal eigenvectors of the space-time covariance matrix. Other authors refer to similar methods as low rank or principal components detection. Previous work has shown that when the noise covariance matrix is unknown, but approximately low rank, the eigencanceler achieves a significantly higher convergence speed than the sample matrix inversion (SMI) processor. In this work we use a Nyquist sampling argument to show that the signals received by an N x L space-time radar (N elements and L-tap delay line filter at each element) are essentially confined to an (N + L -1) dimensional subspace. This subspace is spanned by the principal eigenvectors of the space-time convariance matrix and is referred to as the interference subspace, while the other eigenvectors are said to span the noise subspace. The eigencanceler filter is then constructed as the minimum norm vector constrained to the noise subspace.
Using asymptotic distributions an analytical expression is obtained for the probability density function of the eigencanceler normalized SNR. This expression is used to compute the average detection and false alarm probabilities. The theoretical expressions for the normalized SNR and detection probability are shown to closely match simulation results. Advantages of the eigencanceler over the SMI method are further illustrated by analysis of IDPCA Mountaintop data. It is shown that when the target is included in the training and when the number of training samples is small, the SMI causes target cancellation while the eigencanceler provides robust performance.
Direct comments and questions to: firstname.lastname@example.org
© MIT Lincoln Laboratory. All rights reserved.