Experimental Study of MIMO Channel Statistics and Capacity via Virtual Channel Representation

This work presents an experimental study of MIMO channel statistics and capacity via the recently proposed virtual channel representation, which describes the channel by a finite number of fixed virtual transmit/receive angles and delays. Our results confirm two important implications of virtual path partitioning: the virtual coefficients are approximately uncorrelated, and there exist fundamental angle-delay dependencies which limit the degrees of freedom in MIMO channels. The virtual channel power matrix reflects the distribution of channel power in angle-delay domain, quantifies the spatial-frequency correlation of actual channel coefficients, and also provides an intuitive explanation of the impact of scattering environments on capacity. The modeling accuracy of the popular Kronecker model, as well as the recently proposed eigenbeam model, is compared to the virtual channel model. The results indicate that both the virtual and eigenbeam models achieve good prediction accuracy, because they can model nonseparable 2D angular spectrum, while the Kronecker model results in larger prediction error due to the restrictive structure.