We present a new multiple-input-multiple-output (MIMO) precoding scheme using LDLH channel decomposition. The proposed method is verified to have lower complexity and better bit-error-rate (BER) performance than that of conventional singular-value decomposition (SVD). SVD is commonly used to decompose the MIMO channel into parallel eigenmodes. However, the large variation of the singular values often leads to ill-conditioned channel and degraded error performance. Recently, two SVD-based decomposition schemes that aim to improve such subchannel gain uniformity have been developed. They are geometric mean decomposition (GMD) and uniform channel decomposition (UCD). Albeit the error performance of GMD and UCD is optimized, both of them require relatively higher complexity than that of SVD with power allocation. In this paper, we present a scheme with moderate subchannel gain uniformity while exhibiting the lowest complexity compared with that of SVD, GMD, and UCD. Simulations also show that LDLH can outperform all aforementioned schemes as the channel-estimation errors in the receiver become severe.