Soft-output sphere decoding: Single tree search vs. improved k-best

Multiple-Input Multiple-Output systems provide high multiplexing gain for digital transmissions. However, this is only achievable if an expedient detection method is used. A common method is Maximum Likelihood (ML) detection which enables soft decisions for each received bit along with good error performance. The drawback of this method is its demanding algorithm. In order to meet real-time constraints, the ML detection can be approximated. In this paper, we compare three different implementations of the soft sphere decoder: the single tree search which achieves true ML performance, a conventional k-best algorithm that delivers approximated ML detection, and a novel improved k-best algorithm with better ML approximation at cost of slightly increased execution time. We examine different performance aspects of these sphere decoder implementations and give a recommended complexity-border which indicates where the usage of an ML approximation becomes appropriate.