Algoritmi adaptivi cu aplicații în comunicațiile audio și de voce / Adaptive algorithms with applications in audio and Voice communications

The purpose of the thesis is to develop new algorithms or improved variants of the existing ones for the identification of unknown systems through adaptive filters which overcome the existing solutions and aim realtime applications. The third chapter introduces four adaptive LMS-based algorithms by considering a new optimization criteria, which is much closer to a real-world perspective. Their performance and parameters are studied so that to provide a proper view about using them in certain configurations and scenarios. Moreover, the thesis develops optimization methods that improve the overall performance of the adaptive algorithms. In the fourth, we aim to improve one of the algorithms proposed in the third chapter, to make it even more practical and performant, by taking the advantage of the data-reuse technique. Also, we extend the data-reuse method to analyze the convergence modes of the normalized LMS (NLMS) algorithm and APA, to develop two variable step-size (VSS) algorithms, and open a gate to new methods of optimizing the use of constant parameters. In the end, solutions for multi-path signals (e.g., the reverberation effect) that can occur in practical applications are developed, aiming to reduce the computational complexity of the rudimentary algorithms. The fundamental research of the thesis is analyzed in depth through various experiments, so that to outline the advantages and disadvantages compared to other existing solutions, and to provide a proper vision of performance, parameters in use, computational complexity, and use cases.