Probabilistic Channel Feedback for Multiuser MIMO Broadcast Channels with Orthogonal Beamforming

In multiuser multi-input multi-output (MIMO) cellular systems operating in the frequency division duplexing (FDD) mode, the channel state information (CSI) feedback from users are required in order to schedule users effectively and achieve system capacity. When the number of transmitting antennas or the number of users increases, the tremendous amount of feedback load makes it unrealistic to have perfect CSI at the transmitter. Threshold based feedback reduction is a potential approach to reduce the feedback load. However, the previous works on threshold based feedback reduction only considered users with independent channels. In this paper, we prove that, while the threshold based feedback is optimal when the channels of the users are independent, it is not if there is correlation among users' channels. The main reason is that, when the users experience similarly poor channels, there is a higher probability that none of them feeds back, which results in insufficient information for the base station (BS) to perform scheduling, thus possible rate loss; while when the users' channels are similarly good, too many of them feedback and waste the feedback bandwidth. This observation motivates us to propose and investigate a more general probabilistic feedback strategy which can avoid the aforementioned problems and achieve higher system sum rate in correlated channels. In addition to the theoretical proof, numerical evaluation was conducted to validate the claim.