Cheap semidefinite relaxation MIMO detection using row-by-row block coordinate descent

This paper considers the problem of low complexity implementation of high-performance semidefinite relaxation (SDR) MIMO detection methods. Currently, most SDR MIMO detectors are implemented using interior-point methods. Although such implementations have worst-case polynomial complexity (approximately cubic in the problem size), they can be quite computationally costly in practice. Here we depart from the interior-point method framework and investigate the use of other low per-iteration-complexity techniques for SDR MIMO detection. Specifically, we employ the row by-row (RBR) method, which is a particular version of block coordinate descent, to solve the semidefinite programs that arise in the SDR MIMO context with an emphasis on the QPSK scenario. In each iteration of the RBR method, only matrix-vector multiplications are needed, and hence it can be implemented in a very efficient manner. Our simulation results show that the RBR method can indeed offer a significant speedup in runtime, while providing bit error rate performance on par with the interior-point methods.