Extensions of the constant-modulus algorithm and the phase-locked loop for blind multiuser detection.

摘要

Multiuser detection is the process of mitigating interference among users in a multiuser communications system. This dissertation addresses the problem of blind multiuser detection, where the transmitters do not provide training sequences or other assistance to the receiver. We propose new low-complexity, adaptive, linear blind detection algorithms for square and tall channels by extending two well-established single-user algorithms, the constant-modulus algorithm (CMA) and the phase-locked loop (PLL), to multiuser channel framework.;We propose a multidimensional phase-locked loop (MPLL) for blindly resolving a unitary ambiguity. The MPLL, a multidimensional generalization of the PLL, is a decision-directed algorithm that exploits the discrete nature of digital communication signals. We investigate the convergence behavior of a first-order and second-order MPLL. Using computer simulations, we show that the MPLL offers fast convergence, low complexity, and excellent steady-state performance.;For memoryless channels, we propose the vector constant-modulus algorithm, which is a unique generalization of the CMA to vector-valued signals. In the absence of noise, we show that the vector CMA detector is a whitener for all non-CM input alphabets. As a result, this detector is compatible with shaped-input alphabets. In the presence of noise, this detector displays near MMSE-like performance. We also propose another whitener, the vector CMA with Gram-Schmidt constraint detector, that can be used for all input alphabets. Using numerical examples, we compare that the performance and complexity of these two detectors with other blind detectors.;We also extend both the vector CMA and vector CMA with GSC detectors to channels with memory. In the absence of noise, we show that the vector CMA detector converges to both unitary and non-unitary matrices and that the vector CMA with GSC detector is a whitener for a sub-Gaussian input alphabet. In the presence of noise, the performance of optimal vector with GSC detector is similar to that of the optimal MMSE detector. Finally, through simulations, we demonstrate that the vector CMA with GSC detector compares favorably with other blind detectors.