BLIND CHANNEL EQUALIZATION USING NEW STATISTICAL SIGNAL PROCESSING TECHNIQUES
In this dissertation we have given some new algorithms for blind equalization. In case of the equalization based on second order statistics (SOS), a sufficient condition for blind equalization is given and proved when the source symbols are coloured. Based on the given condition a new algorithm for coloured sources has been derived. The performance of the new algorithm as given by simulations is comparatively better than the previous works, especially at low signal to noise ratios. Then a new energy matching technique in SOS domain is introduced. It is very simple and computationally light scheme, which provides faster convergence. This scheme is valid only for equal energy source symbols. It outperforms the existing techniques of same category in simulations.
The work is then carried out for the domain of higher order statistics (HOS). In order to take advantage of both the SOS and HOS based techniques the hybrid approach of HOS-SOS is developed. The cost function of HOS based algorithms is modified by imposing the conditions based on SOS, which resulted into another new algorithm for binary source symbols. The algorithm gives excellent results and converges in three to five epochs only. This work is then further extended to cut down the computational complexity. Some over–stressing conditions with little contribution to the performance but possessing higher level of complexity have been removed. Energy constraint as given in SOS domain is also applied in the hybrid domain. Further modification is done to deal with complex source symbols e.g. 4–QAM. All this resulted in a much faster algorithm, which converges in one epoch only. Simulation results of each algorithm demonstrate the validity and superior performance of the new algorithms over the previous works. Some concluding remarks and future directions in this field have been proposed at the end.