An Efficient K-Best MIMO Detector for Large Modulation Constellations

Abstract

For K-best multiple-input multiple-output (MIMO) detection using real-valued decomposition (RVD), we need to obtain the $K$ surviving candidates from $K \sqrt {M}$ candidates, where $M$ is the modulation order. This paper presents a sorter-free detection algorithm, where the $K$ surviving nodes can be obtained in ${\mathrm {log_{2}}} {K}$ iterations, which is independent of modulation size. The $K \sqrt {M}$ candidates are arranged into a multiple-layer table using the proposed path metric discretization. A bisection-based search algorithm is used to obtain the locations of the $K$ surviving candidates. A low-complexity fully-pipelined architecture is devised in order to implement the proposed MIMO detection without the need to use any dividers. In addition, an efficient method for storing information from child nodes is proposed, which requires significantly less storage space compared to the conventional Schnorr Euchner (SE) enumeration approach. Implementation results show that the proposed K-best MIMO detector supports a 6.4Gb/s throughput that has a $0.32~\boldsymbol{\mu }\text{s}$ latency in a 90 nm process for a 256-quadrature amplitude modulation (QAM) 4 $\times $ 4 MIMO system. In addition, compared to the sorter-based baseline detector, the proposed detector improves the hardware efficiency by 77%.

Keywords

• K-best detection
• multiple-input multiple-output (MIMO)
• very large-scale integration