ISBN 9783843914123

978-3-8439-1412-3, Reihe Kommunikationstechnik

Emna Eitel
Tracking of Time-Varying Multiple Input Multiple Output Channels

155 Seiten, Dissertation Universität Stuttgart (2013), Softcover, A5

Zusammenfassung / Abstract

One of the most crucial parts in the design of wireless telecommunications systems is the acquisition of precise channel state information (CSI). The accuracy of the channel estimate has notable impact on the achievable data throughput, which is especially true for multiple Input multiple output (MIMO) systems. In case of fast time-varying channels, tracking techniques must be applied to keep the CSI up to date. In this thesis, we deal with decision-directed (DD) channel estimation as a resource-efficient method for channel tracking. We show that pilot-assisted channel knowledge can be effectively exploited to enhance the performance of DD tracking algorithms. We consider the most widely applied adaptive recursive algorithms and demonstrate that periodic Training procedures can be improved to speed up the convergence, which is of paramount importance for packetized data transmission. A drawback of DD channel tracking relies in its sensitivity to successive detection errors leading to error propagation (EP). To tackle this problem, aperiodic approaches are suggested that apply training only when needed. This need arises as soon as an EP is detected and results in requesting pilots in order to stop it; a method which we denote as pilot on request Training (PRQT). EP detection is performed by metric thresholding. We distinguish between filterdependent and filter-Independent metrics. Filter-dependent metrics are tightly integrated in the tracking filter and can benefit from statistical information collected along the tracking process. In order to enable a wider application of PRQT regardless of the applied tracking algorithm, filter-independent metrics are analyzed which rely on the instantaneous signal-to- noise ratio. Methods to analytically determine the thresholds are proposed based on the probability density function of the metrics. Simulation results show that the suggested training methods lead to a significant Performance improvement in terms of bit error rate and spectral efficiency.