Multiuser detection in noise enhanced eigenvector subspace for large scale MIMO communications

This paper proposes a signal detection algorithm with good performance in the large scale uplink multiuser multiple-input multiple-output (MU-MIMO) systems. The proposed algorithm employs the minimum mean-square error (MMSE) detection result as the initial values, and project random noise to the orthonormal eigenvector subspace to amend the error of the noise enhancement of the MMSE detection, where the noise components become uncorrelated. To reduce the complexity, an approximated log likelihood function is employed, and only fixed number of candidates with small approximated log likelihood function values are used for further calculation. Then the detected signals are quantized and selected that minimize the log likelihood function. As the noise projected to each eigenvector is uncorrelated each other, the MU-MIMO detection algorithm is expected to achieve good performance. Computer simulations show that in a 128×64 uplink multiuser MIMO system, the BER performance of the proposed algorithm is superior to MMSE-SIC, while costing only a fraction of the complexity compared with MMSE-SIC.