Subspace versus eigenmode quantization for limited feedback block-diagonalization

In this paper, we compare the performance of block-diagonalization and regularized block-diagonalization precoding, assuming quantized channel state information (CSI) at the transmitter as obtained from limited receiver feedback. We identify the necessary CSI for the calculation of the precoders and propose quantization strategies to efficiently represent the CSI in the corresponding Grassmannian and Stiefel manifolds. We evaluate the trade-off between quantizing only subspace information to facilitate block-diagonalization versus quantizing the magnitude and direction of the individual channel eigenmodes to enable precoder regularization, improving the performance at low SNR.