Direct beamformer estimation for dynamic TDD networks with forward-backward training

This paper investigates direct beamformer estimation in dynamic time division duplexing (TDD) system with the objective of weighted sum rate maximization. For a given TDD frame, base stations (BS) are allocated to either uplink or downlink based on the instantaneous traffic state. The weighted sum mean-squared error minimization framework is used to obtain the decentralized iterative solution for the beamformer optimization. The received precoded pilot training matrices are directly used in the decoupled optimization problem, and two different over the air bi-directional signaling strategies are used for iterative forward-backward training of both transmit and receive beamformers. Detailed flow of signaling exchange and the beamformers estimation procedures are described for each bi-directional signaling strategy. Moreover, the proposed beamformer signaling schemes allow non-orthogonal and overlapping pilots, which greatly reduces the resource allocation effort. The numerical examples illustrate the system performance of the proposed algorithms against the training sequence length.