To exploit the multiplexing and the diversity gains in a multiple-input multiple-output (MIMO) wireless system, the singular value decomposition (SVD) is typically used to decompose the MIMO channel into parallel eigen-modes. However, the dynamics of the frequency selective channel fading variations often make the decomposed subchannels ill-conditioned and lead to unsatisfied performance in terms of bit error rate (BER). Recently, two improved SVD-based decomposition schemes, i.e. the geometric mean decomposition (GMD) and the uniform channel decomposition (UCD), are developed which try to achieve uniform subchannel gains. Unfortunately, both of them require even higher complexity than that of SVD with power allocation. In this paper, we propose a decomposition method, called LDL(H), which has moderate subchannel gains uniformity while exhibits the lowest complexity compared with that of aforementioned SVD-based schemes. Simulations are also given to show that the proposed LDL(H) scheme is more robust than all other schemes at higher SNR as channel estimation errors become severe.