Identification de canal et égalisation aveugles et semi-aveugles pour les communications mobiles

Most of the present mobile communication standards include a training sequence to estimate the channel. Blind techniques allow the estimation of the channel without requiring training symbols, thus increasing bandwidth efficiency, but lack from robustness. The purpose of semi-blind equalization is to exploit the blind information as well as the information coming from the known symbols. Semi{blind techniques robustify the blind problem and allow the estimation of longer impulse responses than possible with a certain training sequence length; for a desired estimation quality, they also allow the use of shorter training sequences.
Furthermore, they offer better performance than blind and training methods.
We present identifiability conditions for semi-blind FIR multichannel estimation: semi-blind methods are able to estimate any channel, even when the position of the known symbols in the burst is arbitrary. Performance bounds for semi{blind multichannel estimation are provided through the analysis of Cramer-Rao bounds and a comparison of semi{blind techniques with blind and training sequence based techniques is done. A study on performance under constraints is proposed to characterize blind performance.
The proposed semi{blind methods are mainly based on Maximum-Likelihood which can incorporate the knowledge of input symbols. For grouped known symbols, suboptimal criteria appear as a linear combination of a training sequence based criterion and the blind ML criterion. In order to build powerful semi{blind ML methods, we also focus on the study of blind ML methods. At last, we present methods that combine a blind criterion with a training sequence based criterion. Receiver structures are also presented. The structure of the burst mode equalizers are studied and especially the structure of the ISI canceller that we call Non-Causal Decision Feedback Equalizer (NCDFE): an implementation of the NCDFE is proposed based on soft decisions. At last, performance bounds on Maximum Likelihood Sequence Estimation (MLSE) are given when the channel order is underestimated.