Schatting en controle van een akoestisch veld over een continue gebied

The goal of this thesis is to enable effective sound field control over continuous regions. Sound field control includes sound zone control, spatial active noise control and sound field reproduction. With a number of loudspeakers distributed in a room, the sound field control task consists of constructing loudspeaker signals such that the desired sound field is produced. In order to determine the loudspeaker signals, information from the acoustic environment is needed, such as room impulse responses from the loudspeakers to the continuous region of interest. Therefore, an essential part of sound field control is sound field estimation and spatial covariance estimation, both of which are methods for obtaining the required information from the acoustic environment. The thesis contains contributions within sound field control that improves flexibility, robustness to noise, and computational cost. It is complimented by contributions within sound field estimation and spatial covariance estimation that are more robust to noise, more effective in time-varying acoustic environments, and enables effective measurements with moving microphones. The thesis contains contributions to improve several aspects of sound field control, including improved flexibility, robustness, and computational cost. An audio reproduction system capable enough to produce sound zones is capable of other sound field control tasks. Therefore, one of the contributions of the thesis is a generalization of conventional sound zone control methods that easily integrates with other sound field control tasks. When a sound zone control system is used in noisy environments, the noise level has an effect on the audibility and the sound quality, implying that the sound zone control system should ideally adapt its operation to the noise level. Therefore, a sound zone control method is proposed that takes the noise level into account, leading to improved reproduction in noisy conditions. For active noise control and sound zone control, an adaptive filter approach is sometimes used, which can be computationally costly when there are many loudspeakers. Therefore, an adaptive filter for this purpose is proposed, which uses a low-rank representation of the filters which lowers computational cost and improves convergence speed. One of the primary challenges in sound field control is obtaining accurate information about the acoustic environment in a practical manner. Compared to traditional methods which require a stationary microphone, a more flexible and faster method of acquiring measurements is to use a moving microphone. Therefore, methods for sound field estimation with a moving microphone are developed in this thesis. Two approaches are taken, one based on kernel ridge regression, and one based on an infinite-dimensional spherical harmonic model. Both approaches ensure the estimates satisfy the homogenous Helmholtz equation, meaning that the sound fields are physically realizable. These approaches are shown to be more robust and provide more accurate estimates compared to traditional truncated or discretized basis expansion methods. The methods enable the use of directional microphones, as well as ways of incorporating prior knowledge of the sound field. The improvement in performance is validated on both simulated and real data. Many sound field control algorithms make use of spatial covariance matrices, a quantity related to the room impulse responses and the audio content that should be reproduced. The sound zone control methods proposed in this thesis make use of acoustic information in the form of the spatial covariance rather than room impulse responses. Therefore it is important to consider how to effectively estimate the spatial covariance. With full knowledge of the sound field, the spatial covariance can be estimated well with simple methods such as computing a sample covariance matrix. However, in practice only partial sound field knowledge is available, in which case estimating the spatial covariance is more difficult. In this thesis spatial covariance estimation methods have been proposed based on kernel ridge regression, ensuring that the physical properties of the underlying sound field are respected, and that the continuous region of interest is taken into account. In addition, because the acoustic environment is potentially time-varying, a method is provided that effectively combines new measurements with previous estimates, leading to shorter measurement times.