Spatially Sparse Beamforming Training for Millimeter Wave MIMO Systems

To realize high beamforming gain and thus sufficient link budget, millimeter wave (MMW) multiple-input multiple-output (MIMO) systems employ large antenna arrays at both the transmitter and receiver. However, due to the power and cost limitation, only a limited number of radio frequency (RF) chains are available for MMW antenna arrays. In other words, the number of RF chains is far smaller than that of antenna elements. This MMW MIMO setup poses a formidable challenge for channel estimation, which is conventionally required to realize the optimal singular value decomposition (SVD) beamforming. To circumvent the formidable channel estimation, this paper proposes two iterative antenna training schemes for the SVD beamforming in MMW MIMO systems. Relying on the channel reciprocity in time division duplex, the proposed schemes employ power iteration and Lanczos iteration, respectively, to gradually approach the SVD beamforming. During iterations, we only need to estimate several channel-related vectors instead of the MIMO channel matrix. Thanks to the spatial sparsity in MMW channels, the training overheads required by the proposed schemes are moderate. Simulations demonstrate that the proposed schemes outperform the counterpart in terms of performance and training overhead, and can achieve the performance very close to that of the perfect SVD beamforming.