Local maximum likelihood detection method for MIMO systems

The paper proposes a local maximum likelihood signal detection method which has very low complexity and good performance. The method is composed of two stages. First, linear detection (e.g., zero forcing detection) is used to obtain the initial estimated signal vector, then, local ML detection is performed over restricted search vectors which are chosen around the initial estimated symbol based on two rules (rule A and rule B). Because the selection of these search vectors is operated within each transmit antenna signal constellation instead of all hyperplanes as sphere decoding, and only neighboring points are considered, both the selection and search complexities can be reduced significantly. The performances with rules A and B are comparable to that of the ordering MMSE and the lattice-reduction-aided MMSE successive-interference-cancellation (LR MMSE SIC) detection, respectively, while the complexity is only a small fraction of theirs.