Channel Correlation-Based Approach for Feedback Overhead Reduction in Massive MIMO

For frequency-division duplex multiple-input-multiple-output (MIMO) systems, the channel state information at the transmitter is usually obtained by sending pilots or reference signals from all elements of the antenna array. The channel is then estimated by the receiver and communicated back to the transmitter. However, for massive MIMO, this periodical estimation of the full transfer matrix can lead to prohibitive overhead. To reduce the amount of data, we propose to estimate the updated channel matrix from the knowledge of the full correlation matrix at the transmitter made during some initialization time and the instantaneous measured channel matrix of smaller size, characterizing the link between the user and a limited number of reference array elements. The proposed algorithm is validated with measured massive MIMO channel transfer functions at 3.5GHz between a $9 \times 9$ uniform rectangular array and different user positions. Since measurements were made in static conditions, the criteria chosen for evaluating the performance of the algorithm are based on a comparison of the predicted channel capacity calculated from either the measured or estimated channel matrix.