The problem of constructing an adaptive multiuser detector (MUD) is conside
red for direct sequence code division multiple access (DS-CDMA) signals tra
nsmitted through multipath channels. The emerging learning technique, calle
d support vector machines (SVMs), is proposed as a method of obtaining a no
nlinear MUD from a relatively small training data block. Computer simulatio
n is used to study this SVM MUD, and the results show that it can closely m
atch the performance of the optimal Bayesian one-shot detector. Comparisons
with an adaptive radial basis function (RBF) MUD trained by an unsupervise
d clustering algorithm are discussed.