LAUSR

Acquisition of accurate channel state information (CSI) at the transmitter (CSIT) is a major challenge of deploying frequency-division duplexing massive MIMO systems. Although compressive sensing (CS)-based CSIT estimation approaches have been proposed to reduce the pilot training overhead for massive MIMO systems, the existing schemes cannot properly dimension the minimum required pilot symbols to estimate the CSIT of all users at the required CSIT quality, because of the loose bounds on the required number of measurements for successful CS recovery and the unknown sparsity levels of user channels. In this paper, we propose a robust closed-loop pilot and CSIT feedback resource adaptation framework which not only exploits the joint sparsity of the multiuser massive MIMO channels to improve the CSIT estimation performance, but also has the built-in learning capability to adapt to the minimum pilot and feedback resources needed for successful CSIT recovery under unknown and time-varying channel sparsity levels. We establish the convergence of the proposed closed-loop adaptation algorithm under certain conditions. Simulations show that the proposed framework has substantial performance gain over conventional schemes with a fixed number of pilots and is very robust to dynamic sparsity as well as model mismatch.