The minimum mean-squared error (MMSE) linear detector [1], [2] has been pro
posed to successfully suppress the multiple access interference and mitigat
e the near-far problem in direct-sequence code-division multiple access com
munication systems. In the presence of unknown or time-varying channel para
meters, the MMSE linear detector can be implemented by the blind Griffiths'
algorithm [3], which uses the desired signal vector instead of a training
sequence of symbols for initial adaptation. In this paper, a variable step-
size (VSS) Griffiths' algorithm is proposed for accelerating the convergenc
e speed, especially in the presence of strong interference. Numerical resul
ts show that the convergence properties of the VSS Griffiths' algorithm are
robust against the wide eigenvalue-spread problem of the correlation matri
x associated with the received signal vector compared to the Griffiths' alg
orithm using a fixed step-size.