Abstract
Turbo codes discovered in 1993 by Berrou et al. [1] were the first coding technique
that approached the Shannon theoretical limit of information transmission to within
0.5 dB. Soon after this discovery, it was recognized that the same Turbo principle
could be applied to a variety of detection/decoding problems such as equalization,
multi-user detection, joint channel estimation and decoding etc. The computational
complexity of such schemes remains a challenging issue since many Turbo detectors
require computationally complex Maximum Likelihood (ML) detection in combination
with channel decoding. Consequently, a class of low complexity Turbo detectors
employing linear and decision feedback filtering instead of ML have been invented
recently in order to solve this problem.
This dissertation describes low complexity adaptive turbo detection methods for
wireless communications, namely Turbo equalization and Turbo Multi-user detection.
The adaptive turbo detectors are shown to outperform their conventional Minimum
Mean Squared Error (MMSE) counterparts regarding the SNR-BER performance.
For Turbo equalization the most remarkable improvement has been achieved
for highly frequency-selective channels. For Turbo Multi-user detection most of the
gain is obtained for overloaded DS-CDMA systems where the number of users exceeds
the processing gain. A theoretical analysis of Turbo equalization provides a
new set of MMSE coefficients. The proposed new detector is shown to outperform
all turbo detectors of similar computational complexity.
The second major contribution of this thesis is a proposed adaptive method for
user ordering for Successive Decision Feedback multi-user Detectors (S-DFD). The
method is shown to outperform all previously proposed ordering methods delivering
significant improvement in SNR-BER characteristic. The analysis of S-DFD has
revealed that a proper user ordering can significantly improve the performance of
the system. The proposed ordering has also been implemented in the adaptive
iterative S-DFD improving the BER performance especially after the first turbo
iteration since this is the most critical part, which determines the SNR floor at
which the Turbo effect starts.