Abstract
In CDMA systems, interference is caused by the non-orthogonality of the users' received signatures that comprise the users' information: the transmit spreading signature, transmit power, and channel impairments. By exploiting the cross-correlation between the users' received signatures, Multiuser Detection (MUD) techniques have been proposed to suppress the interference. Nevertheless, if a linear dependency between at least two signatures exists, as in an overloaded condition, most existing MUDs fail even to meet the basic requirement of a reliable communication, i.e. the Bit-Error-Rate (BER) exponentially reduces as the SNR increases. The optimum signatures for the overloaded condition can be derived by following the Welch-Bound-Equality. However, its optimality can be achieved only with the optimum, yet impractical, MUD. In this thesis, we propose a novel Low-Density Signature (LDS) structure that intelligently manages the interference by switching-off a large number of spreading signature chips. This new LDS structure also allows an efficient implementation of the optimum MUD via Message Passing Algorithm. Because it operates at the Chip-Level and is iterative in nature, the resultant new class MUD technique is termed the Chip-Level iterated (CLi) MUD. The performance of generic CLi MUD techniques that calculate the extrinsic information for each chip by using the MAP-based detections: Log-MAP and Max-Log-MAP with brute-force search can be characterized by its convergence behavior. Our results confirm that the brute-force CLi MUD can very closely approach single-user bound (SUB), in high SNR regimes, even when the system is 300% overloaded. Further to provide a greater flexibility in trading-off the performance and complexity, a Gaussian-Forcing (GF) soft Interference Cancelation (IC) is proposed and analyzed. However, since its performance depends too highly on the Gaussianity of the interference, a more sophisticated grouped detection is introduced and is termed grouped-GF (G-GF) CLi MUD. Combined with the proposed dynamic user grouping strategy, the dynamic G-GF CLi MUD is shown to approach closely the performance of a brute force CLi MUD, with far lower complexity. This thesis also extends the implementation of LDS structure to the Multi-Carrier (MC) CDMA systems for communication over multipath fading channels. Furthermore, the performance of a coded system employing an LDS structure is also investigated. We show that a partitioned turbo CLi Multiuser Detection/Decoding (MUDD) can approach the optimum MUDD with only a small number of turbo iterations. In addition, an Extrinsic Information Transfer (EXIT) chart analysis is used and is shown to have a good agreement with our simulation results.