Grassmannian predictive coding for limited feedback multiuser MIMO systems

Grassmannian beamforming is an efficient way to quantize channel state information in multiple-input multiple-output wireless systems. Unfortunately, multiuser systems require larger codebooks since the quantization error creates residual interference that limits the sum rate performance. To reduce the feedback requirements in multiuser systems, we propose Grassmannian predictive coding to exploit temporal channel correlation. The proposed algorithm exploits the differential geometric structure of the Grassmann manifold. The difference between points, prediction, and quantization are defined using the tangent space of the Grassmann manifold. We show that with practical feedback rates, a significant sum rate improvement can be obtained as a function of the channel correlation.