Sequential Gsvd Based Prewhitening for Multidimensional Hosvd Based Subspace Estimation

Recently, R-dimensional subspace-based parameter estimation techniques have been improved by exploiting the tensor structure already in the subspace estimation step via a Higher Order Singular Value Decompostion (HOSVD) based low-rank approximation. Often this parameter estimation is performed in the presence of colored noise or interference, which can severely degrade the estimation accuracy. To avoid this degradation, prewhitening techniques are applied. In this contribution, we propose a Sequential Generalized Singular Value Decomposition (S-GSVD) based prewhitening scheme for multidimensional HOSVD based subspace estimation. By exploiting the Kronecker structure of the noise correlation matrix for the estimation of the correlation factors, we achieve an improved accuracy compared to matrix based prewhitening schemes. In addition, our S-GSVD approach is computationally more efficient than the classical matrix approach, since it has a lower complexity due to the n-mode GSVD operations.