Optimization Of Ambisonics Sound Image Reproduction

Two-dimensional(2D) or Three-dimensional(3D) sound image reproductions of various loudspeaker arrays can be accomplished by 3D Ambisonics audio technique. However, the performance of irregular loudspeaker reproduction achieved by the solution of inverse matrix is poor. In order to improve the localization performance of sound image, the thesis concentrates on the optimization of Ambisonics sound image reproduction based on irregular configure of loudspeakers for ITU-4, ITU-5 and Auro-10. The main research contributions are summarized as follows:Firstly, the localization performance of 2D sound images reproduced with the first order Ambisonic is inferior at high frequencies since the multi-objective optimization is dominated by the low-frequency objectives. A technique called sum of weighted genetic algorithm is applied to automatically and logically remove this sort of domination, so as to make high-frequency objectives optimized fully and improve the localization performance of sound images at high-frequencies. A first order reproduction for ITU-4 certifies that superiority. And it is proved to have better auditory perception than references in objective test of Interaural Level Difference(ILD).Secondly, the convergence performance of genetic algorithm will be declined with the increasing of optimization parameters when the 2D sound images are reproduced by Higher-order Ambisonics(HOA). The thesis presents an optimization approach for sound images reproduction based on HOA derived from optimal symmetrical virtual microphone response. The number of optimization parameters is reduced to 50% with the proposed method, and the optimal symmetrical virtual microphone response which is well perceived can be easily acquired by genetic algorithm. A forth order reproduction for ITU-5 demonstrates that it is superior to any other traditional approaches. And it is proved to have better auditory perception than references in objective test of Interaural Time Difference(ITD) and ILD.Finally, the 3D Ambisonics sound image reproduction of Auro-10 is discussed. With regard to the problem that the sound image of each direction couldn‘t be realized primely by the fixed 3D loudspeaker gain parameters, this thesis puts forward to a multi-gain fusion approach based on Spatial Localization Quantization Point(SLQP). The directions of original sound source from Ambisonics signals are firstly estimated and then quantized using SLQP. The sound image of each quantized direction is reproduced perfectly by multiple sets of gain parameters. Localization performance of 3D reproduced sound image is better than the traditional method with the fixed gain parameters.