The Gram-Schmidt (GS) orthogonalisation procedure has been used to improve
the convergence speed of least mean square (LMS) adaptive code-division mul
tiple-access (CDMA) detectors. However, this algorithm updates two sets of
parameters, namely the GS transform coefficients and the tap weights, simul
taneously. Because of the additional adaptation noise introduced by the for
mer, it is impossible to achieve the same performance as the ideal orthogon
alised LMS filter, unlike the result implied in an earlier paper. The autho
rs provide a lower bound on the minimum achievable mean squared error (MSE)
as a function of the forgetting factor lambda used in finding the GS trans
form coefficients, and propose a variable-lambda algorithm to balance the c
onflicting requirements of good tracking and low misadjustment.