Semidefinite programming relaxation based virtually antipodal detection for MIMO systems using Gray-coded high-order QAM

Download

Preview

PDF - Post print Download (908Kb) | Preview

Description/Abstract

An efficient generalized semidefinite programming relaxation (SDPR) based virtually antipodal (VA) detection approach is proposed for Gray-coded high-order rectangular quadrature amplitude modulation (QAM) signalling over multiple-input--multiple-output (MIMO) channels. Albeit the decomposition of symbol-based detection to a bit-based one is desirable owing to its reduced complexity and increased flexibility, Gray-mapping is nonlinear, and hence the direct bit-based detection of Gray-coded-QAM MIMO systems constitutes a challenging problem. In this paper, we find a way of exploiting the structural regularity of Gray-coded high-order rectangular QAM, and transforms the classic symbol-based MIMO detection model to a low-complexity bit-based detection model. As an appealing benefit, the conventional three-step ``signal-to-symbols-to-bits'' decision process can be substituted by a simpler ``signal-to-bits'' decision process for the classic Gray-mapping aided high-order rectangular QAM, and hence any bit-based detection method becomes potentially applicable. As an application example, we propose a direct-bit-based VA-SDPR (DVA-SDPR) MIMO detector, which is capable of directly making binary decisions concerning the individual information bits of the ubiquitous Gray-mapping aided high-order rectangular QAM, while dispensing with symbol-based detection. Furthermore, the proposed model transformation method facilitates the exploitation of the unequal error protection (UEP) property of high-order QAM with the aid of the low-complexity bit-flipping based ``hill climbing'' method. As a result, the proposed DVA-SDPR detector achieves the best bit error ratio (BER) performance among the known SDPR-based MIMO detectors in the context considered, while still maintaining the lowest-possible worst-case complexity order of O (NT log2M + 1)3.5.