A subspace training method is presented for multiuser detection in dispersi
ve DS-CDMA channels with the presence of multiple access interference (MAI)
. By using a small number of training symbols of the desired user, the prop
osed method is computationally efficient and near-far resistant.