Closed-Loop Compressive CSIT Estimation in FDD Massive MIMO Systems With 1 Bit Feedback

Abstract

One major practical issue for the implementation of frequency division duplex (FDD) massive MIMO systems is that the acquisition of channel state information at the transmitter side (CSIT) requires overwhelming pilot training and feedback overhead. Recently, compressive sensing (CS) based CSIT estimation approaches have been proposed to reduce the pilot training overhead for massive MIMO systems. However, it is very difficult to compute the minimum required pilot training overhead at the base station because of loose restricted isometry property (RIP) bounds for successful CS recovery and unknown sparsity levels. In this paper, we consider a framework of closed-loop compressive CSIT estimation with 1 bit feedback to learn the minimum required pilot overhead to achieve a certain target CSIT MSE without explicit knowledge of channel sparsity. We analyze the convergence behaviors of the multi-loop pilot overhead adaptation based on the Lyapunov approach. Simulations show that the proposed closed-loop compressive CSIT estimation framework has substantial performance gain over conventional open-loop algorithms and is very robust to dynamic sparsity as well as model mismatch.