I. Introduction
It is well known that the computational complexity of individually optimal detection for direct-sequence code-division multiple-access (DS-CDMA) grows exponentially with the number of users [1], as the computation of the marginal posterior-mode (MPM) distribution is required. Maximum a posteriori (MAP) probability detection for each user is, therefore, far too complex for practical CDMA systems with even a moderate number of users. The exponentially growing complexity has inspired a considerable effort in finding low-complexity suboptimal alternatives capable of resolving the detrimental effects of multiple-access interference (MAI).