Limited feedback and channel estimation for closed-loop MIMO systems

Multiple-input-multiple-output (MIMO) techniques have been one of the key features in the globally-endorsed 3GPP LTE cellular standard, and are also expected to be a key feature to compensate for the increased path-loss in the mm-wave spectrum. A MIMO transmitter requires knowledge of MIMO channel in order to harvest beamforming gain. Therefore, in frequency division duplex (FDD) system, or the time division duplex (TDD) system with uncalibrated uplink and downlink chains, the transmitter has to rely on limited feedback from a receiver. Feedback constraints are a consequence of the limited feedback link capacity.In this thesis, we firstly describe the MIMO precoding space as the Stiefel manifold, together with its MIMO-relevant quotient spaces; the Grassmannian and the Flag manifolds. Further, we introduce an alternative closed-form solution for geodesic construction on the Grassmannian manifold, and we introduce a framework for geodesic construction on manifolds, where a closed-form solution does not exist, i.e. the Stiefel manifold and the Flag manifold. Secondly, we discuss channel estimation on the spatially precoded user-specific reference symbols, with TDD-based Eigenbeamforming and FDD-based closed-loop LTE precoding. When the spatial precoder is varied across the allocated frequency band, the corresponding effective channel changes its frequency correlation properties and/or suffers from discontinuities. Herein, both issues are tackled by additional transmitter signal processing. The rest of the thesis is dedicated to feedback design for single-user (SU) MIMO, multi-user (MU) MIMO and Coordinated multi-point (CoMP). We develop a subspace packing algorithm called ECA which is analogous to the max-min algorithm placing points equidistantly within the wrap-around two-dimensional rectangle. Straight lines are substituted by geodesics. Further, we provide improvements to LTE double-codebook design for both uniform linear arrays and cross-polarized arrays. A generalized eigenvector MU-MIMO technique benefits from knowledge of the normalized covariance matrix at the transmitter. We separately quantize the normalized eigenvalues and corresponding eigenvectors, and study the optimal split of feedback bits between them. On the other hand, LTE does not support feedback of eigenvalues, and MU-MIMO performance may be improved only by improving the knowledge of eigenvectors at the transmitter. Set in an LTE context, we propose a low-complexity successive refinement technique using the additional m-th best feedback and geodesical interpolation at the transmitter. Finally, we discuss feedback for CoMP and introduce a concept of flexible layer arrangement for coherent and non-coherent SU-MIMO joint-transmission CoMP