Multiuser receiver structures for CDMA

CDMA (Code Division Multiple Access) with multiuser detection at the basestation is an effective way to increase the system capacity.; In the first part of this thesis, a delay-robust multistage successive interference cancellation (SIC) multiuser detector that is near-far resistant under delay mismatch is proposed. The detector is based on a linear SIC implementation of the decorrelating detector, which can be shown to be equivalent to the space alternating generalized expectation-maximization (SAGE) algorithm, which has guaranteed local convergence. Our proposed delay-robust SIC has a capacity close to 100% of the spreading factor and can be applied to general band-limited chip pulse shapes. The delay-robust SIC's asymptotic multiuser efficiency (AME) and bit error rate (BER) are both analyzed and confirmed by computer simulation.; In the second part of this thesis, we propose a new soft-decision function to be used in multistage SIC detectors. The soft-decision function combines the desirable convergence properties of the linear-soft decision function with the noise reduction of the hard-limiter decision function. The result is a SIC detector with both good noise performance and convergence. The multiuser detection and delay tracking for time-varying multi-path fading channel is considered and the tracking performance of the delay-robust SIC is evaluated by computer simulation.; The third part of the thesis considers the application of CDMA multiuser detection methods to multi-antenna systems known as MIMO (multiple input multiple output) systems, e.g., the BLAST (Bell Labs Layered Space-Time) system, by observing the similarity between a linear MIMO system and a synchronous CDMA multiuser system. An ordered SIC method was proposed by other researchers for bit detection in BLAST systems. However, its complexity is too high for high-rate applications. We apply a decorrelating decision-feedback CDMA multiuser detection method to BLAST systems. Since only one matrix decomposition is performed at the beginning of the algorithm, the computational complexity is greatly reduced. However, BLAST's decision-feedback ordering is according to decreasing signal power, not in the optimum decreasing signal-to-noise ratio (SNR) ordering. We further propose using a series of numerically stable unitary transformations to reorder the decomposed matrices. We show that complexity is an order lower than that of the ordered SIC while its stability is improved as well. (Abstract shortened by UMI.)...