Downlink Channel Feedback for FDD Massive MIMO Systems via Tensor Compression and Sampling

In frequency division duplex (FDD) massive MIMO systems, each user equipment (UE) device is required to frequently feed back downlink channel state information to its serving base station (BS), resulting in huge uplink overhead. This paper considers the problem of downlink channel feedback in practical wideband massive MIMO systems operating in the FDD mode. The starting point is to leverage the spatial structure of the BS antenna array to reshape the channel matrix in tensor form, and then bring in orthonormal Tucker (also known as multilinear / higher-order SVD) factorization to approximate the channel tensor on the UE side, and feed the obtained factors back to the BS. The resulting decomposition results in a significant reduction in the number of parameters that needs to be fed back from each UE to its serving BS, without sacrificing channel estimation accuracy. Furthermore, this paper exploits a recently-developed sampling/compression mechanism that efficiently handles the considered problem. Using this approach, each UE is only required to feed back a small number of samples/measurements from the channel tensor, thus shifting the computational burden of tensor decomposition/compression to the BS which naturally possesses higher computational capabilities compared to the UE. Experimental results using channels constructed from a practical ray-tracing simulator demonstrate the superiority of the proposed tensor-based approaches relative to the state-of-the-art.