In this thesis we shall focus on the applications of clustering techniques to simplify the implementation of several important modules in digital communication systems, where we are especially interested in systems with high data rate. Clustering technique is considered in this thesis because it is a computationally efficient signal processing tool. In particular, the thesis work will concentrate on the application of hyperplane and bipartitioning techniques for clustering. We have derived the optimal bipartitioning rules to achieve minimum sum of squares error clustering results. M-clustering problem can thus be solved by iterative application of the developed optimal hyperplane based bipartitoning. Simulation results showed that better clustering results can be obtained when compared...[ Read more ]

In this thesis we shall focus on the applications of clustering techniques to simplify the implementation of several important modules in digital communication systems, where we are especially interested in systems with high data rate. Clustering technique is considered in this thesis because it is a computationally efficient signal processing tool. In particular, the thesis work will concentrate on the application of hyperplane and bipartitioning techniques for clustering. We have derived the optimal bipartitioning rules to achieve minimum sum of squares error clustering results. M-clustering problem can thus be solved by iterative application of the developed optimal hyperplane based bipartitoning. Simulation results showed that better clustering results can be obtained when compared to those obtained by other well known algorithms in literature, yet the computational complexity of the proposed optimal hyperplane method is much lower than that of other algorithms in literature. The optimal hyperplane based clustering technique is then applied to solve various digital communication problems that include the nonlinear distortion problem induced by High Power Amplifier (HPA).

A novel joint precompensator and detector design method for QAM system with transmitter non-linearity is proposed, where the non-linearity can be induced by HPA. Simulation results showed that the directed-inverse precompensated system can achieve good performance without a prior information of the transfer function of the HPA. Further improvement of the system performance can be achieved by considering the design of joint precompensator and detector under various channel noise conditions. In order to the lower the computational complexity in the construction of detector and signal's detection, a simple tree-based decoder is employed. The bit error rate (BER) obtained from the simulations is shown to be compatible or even better than that of the Maximum Likelihood (ML) based detector.

Another application of the clustering algorithm is in the Automatic Modulation Recognition (AMR) problem. Simulation results showed that the proposed algorithm can achieve more than 90% recognition rate when the received signal to noise ratio (SNR) equals to 13 dB. It further demonstrated the robustness of the proposed algorithm towards channel noise where only a small amount of received symbols are required in the recognition problem.

Lastly, the thesis investigated the blind channel equalization problem using clustering algorithm. The Constant Modulus Algorithm (CMA) is recognized as a clustering problem, and new cost functions are thus proposed. A modified CMA (MCMA) is proposed under the clustering framework. Simulation results on blind equalization using MCMA shown a smaller mean square error (MSE) is obtained when compared to that obtained by the traditional CMA. Besides the equalization performance, the computational complexity and convergence are also important performance factors in blind equalization. A novel dual-mode algorithm that adapts the cost functions used in MCMA is proposed aiming to achieve a fast convergence rate and low computation complexity. Simulation results show 1.5dB and 2.5dB improvement in MSE for transmission over the frequency selective fading channel with 16-QAM and 64-QAM respectively.