The Accuracy of Tensor and Directional Methods for MIMO Channel Modeling

The goal of this work is to identify MIMO modeling strategies with high accuracy and relatively few parameters. Candidates are existing correlation tensor models (Kronecker, Weichselberger, and maximum entropy), a new generalized tensor model based on the higher-order singular value decomposition (HOSVD), and an unstructured diffuse directional model. The performance of the models is investigated by applying them to 8 x 8 MIMO reference channels generated with a realistic double- directional cluster channel model. It is shown that all of the existing models exhibit high error in the reconstructed spatial spectra (10-30%) and even higher error in the reconstructed full covariance (20-35%). Analysis of a new sparse core tensor model indicates that this error decays slowly (logarithmically) as the number of parameters is increased. For the same number of parameters as the simple Kronecker model, the directional model is able to reconstruct the spectrum with improved accuracy (about 3% error). More surprisingly, the error in the reconstructed covariance matrices (about 16%) is also substantially lower than all of the reduced-order correlation tensor methods.