A Constant Throughput Geometric Mean Decomposition Scheme Design for Wireless MIMO Precoding

Abstract

The geometric mean decomposition (GMD) algorithm is a popular approach in developing a precoding scheme for joint multiple-input-multiple-output (MIMO) transceiver designs. In this paper, the adverse effects of conventional GMD algorithms on hardware implementations are first reviewed. Then, a constant throughput modified GMD algorithm is presented. The proposed GMD scheme is constructed on a QR decomposition framework and requires no singular value decomposition (SVD) preprocessing. The new scheme is exempt from the convergence problem, which may seriously degrade throughput performance. It also features lower computational complexity and permutation-free operations and supports hardware sharing between precoding and signal detection modules. Quantitative analysis shows that, under similar symbol-error-rate (SER) performance, the proposed scheme possesses a computational complexity edge over conventional schemes by a margin of 30%. The complexity breakdown indicates that the SVD nullification sweep is the dominant factor of the SVD-based GMD schemes. Even when the sweep number is set to twice the matrix size (2N), the implementation loss is still 0.5 dB inferior to the proposed scheme. Finally, an architecture design of the proposed scheme is given to demonstrate a constant throughput implementation and the feasibility of hardware sharing between precoding and signal detection modules.