In this paper, a novel low-complexity tree-search (TS) algorithm, which is referred to as a hierarchical tree search (HTS), is proposed for soft-output multiple-input-multiple-output (MIMO) detection to mitigate the performance loss caused by an early termination. The proposed HTS algorithm is realized by the following two components: the hierarchical set partitioning to find all hypotheses with reasonable quality as fast as possible and to fairly improve them and the new cost metric to determine the search order by considering the quality of the hypotheses found so far. Through simulation, it is shown that the performance-complexity tradeoff of the proposed HTS algorithm surpasses those of the existing algorithms in various MIMO configurations.
Index Terms-Multiple-input-multiple-output (MIMO) detection, softoutput MIMO detection, tree-search (TS) algorithm.
I. INTRODUCTIONRecently, due to the demands for both higher spectral efficiency and reliability, multiple-input-multiple-output (MIMO) systems incorporating with error-correcting codes (ECCs) have been adopted in wireless standards, such as IEEE 802.11n, IEEE 802.16e, and Third-Generation Partnership Project Long-Term Evolution, and it has been shown that a soft-output MIMO detector can dramatically improve the error performance compared with a hard-output MIMO detector [1]. Because the maximum a posteriori (MAP) detector for softoutput MIMO detection requires tremendous complexity, tree-search (TS) algorithms have been widely considered for practical soft-output MIMO receivers [2], which can fall into one of the following three categories: 1) a depth-first search (DFS); 2) a breadth-first search (BFS); and 3) a metric-first search (MFS). The K-best [3] and Malgorithm [4] belong to BFS, which expands the K best nodes in each level of the tree by forward-only directional search. Although BFS has fixed complexity and memory requirements so that it can be implemented by a parallel pipelined structure [3], it suffers from a poor performance-complexity tradeoff. Sphere decoding (SD) [5] and its variants, including list SD (LSD) [1] and single TS (STS) [6], belong to DFS, in which a bidirectional search is performed within a predetermined sphere. It is known that DFS can provide near-optimal performance with reasonable average complexity [6]. Dijkstra's algorithm [7] and its variants [8] belong to MFS, in which the search order is determined by the metric of nodes so that it can minimize the average complexity. However, although DFS or MFS requires relatively low average complexity, the worst-case complexity is the same as that of an exhaustive search. In a practical system, the maximum allowed