I. Introduction

IN the uplink CDMA wireless systems, it is assumed that the base station has access to all users' channel state information (CSI) and spreading signatures; and multiuser detection (MUD) has been shown to be an effective way to combat interference and increase data throughput [12]. For the downlink, on the other hand, one can transfer the signal processing for interference suppression from the mobile receiver to the transmit base station by using precoding techniques. This is feasible if the base station has access to the CSI of all active mobile units, e.g., in time-division duplex (TDD) systems where the base station can exploit the channel reciprocity if the time difference between uplink and downlink transmission is shorter than the channel coherence time, or by using channel prediction techniques [1]. The simplest precoding method is pre-RAKE [2], which mitigates the multipath interference without considering the multiuser interference (MUI). Linear precoding techniques to remove the MUI and multipath interference were proposed in [13]. Nonlinear precoding techniques have been shown to offer superior performance although they complicate the receiver and the transmitter, since a modulo operation (which in general depends on the CSI) has to be implemented at both sides of the communication link [3], [9], [14]. Note that most work on linear precoding assumes that each user implements a RAKE receiver and hence assumes the knowledge of CSI at the receiver [13], [14].