This paper presents an iterative geometric mean decomposition (IGMD) algorithm for multiple-input-multiple-output (MIMO) wireless communications. In contrast to the existing GMD algorithms, the proposed IGMD does not require the explicit computation
of the geometric mean of positive singular values of the channel matrix and hence is more suitable for hardware implementation. The proposed IGMD has a regular structure and can be easily adapted to solve problems with different dimensions. We show that the proposed IGMD is guaranteed to converge to the perfect GMD under certain sufficient condition. Three different constructions of the proposed algorithm are proposed and compared through computer simulations. Numerical results show that the proposed algorithm quickly attains comparable performance to that of the true GMD within only a few iterations.
