Subspace Beamforming for Near-Capacity MIMO Performance

A subspace beamforming method is presented that decomposes a multiple-input multiple-output (MIMO) channel into multiple pairs of subchannels. The pairing is done based on singular values such that similar channel capacity is obtained between different subchannel pairs. This new capacity balancing concept is key to achieving high performance with low complexity. We apply the subspace idea to geometric mean decomposition (GMD) and maximum-likelihood (ML) detection. The proposed subspace GMD scheme requires only two layers of detection/decoding, regardless of the total number of subchannels, thus alleviating the latency issue associated with conventional GMD. We also show how the subspace concept makes the optimization of ML beamforming and ML detection itself feasible for any K timesK MIMO system. Simulation results show that subspace beamforming performs nearly as well as optimum GMD performance, and to within only a few decibels of the Shannon bound.