Konstrukcija logaritamskih kvantizera za visoko kvalitetno adaptivno transformaciono kodovanje govornog signala

Abstract

In this dissertation, the quantizer design is being considered, devices which within telecommunication systems are used for the signal coding. The main goal of this dissertation is logarithmic quantizers design for speech signal coding that will provide high signal quality at the receiving end. The usage of transform coding and adaptation techniques increases coding quality comparing to classical telecommunicational systems suggested by International Telecommunication Union. Quantization, as one of the most important procedures in signal processing, presents discretization of the signal by amplitude as a basic step in analogue signal digitization. The characteristics of maintaining approximately constant coding quality in a wide range of input signal variances is called robustness, so it is suitable to use the logarithmic quantizers due their ability to satisfy this feature. The adaptation of the logarithmic quantizers provides higher reconstructed signal quality at the receiving end, in the manner that adaptive quantizers track the changes in signal strength, change the quantization levels to adapt to input signal, thereby ensuring consistently high quality in the wide range of the input signal variances. The usage of transform coding methods in coding scheme for speech processing additionaly increases quality of the reconstructed signal at the receiver. Transform coding involves association of adjancet signal samples, and the application of certain transformation over them, which indicated redistributing energy (information). In this dissertation, in the order of coding schemes design, will be used some of the best known transformations like Hadamard transform, discrete cosine transform and discrete wavelet transform. By leading the discrete signal to the input of a coding system, the compression is further enhanced, since the discrete signal is limited by amplitude and there is no overload distortion, which ensures high quality coding with lower bit rate. For this reason, dissertation proposes coding schemes for discrete input signal. The gain of the solutions that are proposed in this dissertaiton is clearly demonstrated by combining these quantization techinques and transform coding which results in high coding quality of speech signal at the receiving end even on the lower bit rates. The subject of this dissertation is a very popular in current scientific research directions, because with the development of communication-informational technologies and telecommunication systems, the application of the proposed solutions in the speech signal coding and transmission systems significantly increases. Results presented in this thesis may have many practical applications.