Rate loss analysis of transmitter cooperation with distributed CSIT

We consider in this work the problem of determining the number of feedback bits which should be used to quantize the channel state information (CSI) in a broadcast channel (BC)
with K transmit antennas (or equivalently K single-antenna transmitters (TXs)) and K single-antenna receivers (RXs). We focus on an extension of the conventional centralized CSI at
the TX (CSIT) model, where instead of having a single channel estimate, or quantized version, perfectly shared by all the TX antennas, each TX receives its own estimate of the global multiuser channel. This CSIT configuration, denoted as distributed CSIT, is particularly suited to model the joint transmission from TXs which are not colocated. With centralized CSIT, a very important design guideline for the feedback link was provided by Jindal [Trans. Inf. Theory 2006] by providing a sufficient feedback rate to ensure that the rate loss stays below a maximum value. In the distributed CSIT setting, additional errors occur and the design guidelines for the centralized case are no longer valid. Consequently, we obtain a new relation between the rate loss and the number of feedback bits. Interestingly, the feedback rate derived in the distributed CSIT setting is roughly K log2(K) bits larger than its counterpart in the centralized case. This highlights the critical impact of the CSIT distributedness over the performance.