Diversidade em modulação aplicada à transmissão de imagens em canais com desvanecimento.

Abstract

This thesis deals with techniques to improve the performance of communication systems that use modulation diversity (MD), which is based on the combination of a suitable choice of the reference angle of an MPSK constellation with independent interleaving of the symbol components. The MD technique is applied to the transmission of vector-quantized images over a fading channel. The thesis also covers robust vector quantization (VQ), in which a VQ codebook is made robust against the channel errors by means of an index assignment (IA) procedure. In this way, the quality of the reconstructed images is enhanced. Concerning the MD technique, many important aspects are evaluated, such as: optimum rotation angle for QPSK, 8PSK and 1GPSK constellations; study of the influence of the channel estimation errors on the system performance (a relevant problem, which in general is not covered in the literature on MD); the impact of the Doppler effect on system performance is studied and a trade-off between the interleaving depth and the error probability is stablished; explanation of the performance gain of the MD technique by using a novel geometric interpretation: The Ferris Wheel Effect. Regarding robust vector quantization, a new figure of merit for the index assigment problem is proposed in this thesis. This figure of merit is used j o i n t ly with the simulated annealing algorithm to assign binary indexes to the VQ codevectors. In this work, the M D and IA techniques are evaluated, as well as, the combination of both, in terms of the quality of the reconstructed imagens after transmission over a fading channel. Simulations results show that the combination of the MD and IA leads to the best system performance. This thesis also presents a new method for calculating the bit error probability (BEP) of modulation schemes subject to Rayleigh fading. In this method the communication channel is seen as an additive noise channel where the noise is modeled as the ratio between a Gaussian random variable (r.v.) and a Rayleigh r.v. The method consists of using the cumulative density function of that additive noise to obtain exact expressions for the bit error probability.