A “post-squaring” detection algorithm, and related devices, that may reduce the complexity of maximum likelihood detection (MLD) schemes while preserving their performance is provided. Rather than search for optimum metrics (such as minimum distance metrics) based on squared norm values, a search may be based on un-squared norm metrics, and the squaring may be postponed, for example, until subsequent log-likelihood ratio (LLR) computation. For certain embodiments, approximations of un-squared norm values may significantly reduce computation complexity.