A fast recursive sphere decoder with linear storage

A novel fast recursive sphere decoder with competitive extension (CR-SD) is proposed for the exact maximum-likelihood (ML) detection in multiple-input multiple output (MIMO) system. CR-SD introduces a novel competition mechanism which enables the sub-trees to be searched in an optimized order and makes the search radius shrink to the minimum Euclidean distance as early as possible. Numerical results show that CR-SD achieves considerable computation complexity reduction even compared with the fastest ML Schnorr-Euchner enumeration (SEE) based sphere decoder (SE-SD) in literature. Unlike the ML stack [V. Sandrine et al., May 2002] [S. Baro et al., May 2003] or branch-and-bound (BBD) detectors [J. Luo et al., May 2003] in which the storage requirement is random and the exponential in worst case, CR-SD maintains the negligible storage upper-bounded by a deterministic constant linear with the search dimensions.