Blind multiuser detection based on second-order statistics

Multiuser detection is the process of recovering information from mutually interfering users of a shared communication channel. Typical applications include wireless networks, bundled cables, and multitrack magnetic recording systems. A multiuser detector exploits the structure of the multiuser interference in order to improve system performance or capacity. Because users sharing the channel usually operate autonomously, it is often desirable for a multiuser detector to function blindly, with no a priori knowledge of the channel nor any transmitter cooperation. This thesis addresses the problem of blind multiuser detection.;The research focuses on tall channels, those having more outputs than inputs, because of their many special properties. Tall channels arise, for example, in multisensor receivers or in code-division multiple access systems. We develop adaptive detectors based on primarily on second-order statistics, a philosophy which offers fast, reliable convergence, low computational complexity, and inherent compatibility with shaped constellations having near-Gaussian distributions.;For memoryless channels, we propose a novel adaptive signal-noise subspace separator can be used to simplify subsequent processing in any detector. We extend the technique to perform singular-value decompositions (SVDs) adaptively, and propose blind implementations of the minimum-mean-square-error (MMSE) and zero-forcing (ZF) detectors based on the adaptive SVD. We propose a canonical whiten-rotate (WR) detector offering near-MMSE performance with an adaptive implementation based on spatial prediction. We also propose an adaptive channel-diagonalization algorithm.;For channels with memory, we propose detectors based on spatio-temporal prediction. These detectors use temporal prediction to virtually eliminate channel memory, and then apply spatial algorithms. We propose family of detectors that operate on a stacked observation such that the effective channel is memoryless. The detectors are robust to their estimate of the signal subspace dimension, and they need not know the number of interfering users or the size of their signal alphabets. We also propose information-lossless space-time precoding technique of finite complexity.