Direction-Selective Filters for Sound Localization
Dean J. Schmidlin · InTech eBooks · 2011
An important problem in sound localization is the determination of the polar and azimuthal angles of far-field acoustic sources. Two fundamental approaches to the solution can be identified: spatial filtering (beamforming) and the parameter estimation approach. Van Veen and Buckley (1988) and Krim and Viberg (1996) give comprehensive reviews of the first and second approaches, respectively. Spatial filtering was carried out by an array of pressure sensors. A serious drawback to the filtering approach is that its performance depends directly on the physical size of the array (aperture), regardless of the data gathering time and signal-to-noise ratio. This aperture dependence together with more demanding applications motivated a good number of researchers to develop parametric estimation techniques. These methods can be separated into two main categories, namely, spectralbased and parametric approaches. The most famous example of the first is MUSIC (MUltiple Signal Classification) algorithm developed by Schmidt (1981) and Bienvenu and Kopp (1980), and of the second is the Maximum Likelihood (ML) method developed by Kumaresan and Shaw (1985) and Bresler and Macovski (1986). In contrast to beamforming techniques, a MUSIC estimate of arbitrary accuracy can be achieved if the data gathering time is sufficiently long, the SNR high enough, and the signal model sufficiently accurate. However, a significant limitation is the inability to resolve closely spaced signals with small sample sizes and low SNR. Further deterioration occurs for highly correlated signals and complete breakdown for coherent signals. The interested reader is referred to Krim and Viberg (1996) for discussions on how these limitations have been addressed. All of the methods for localizing acoustic sources had one thing in common. They used arrays composed of pressure sensors. This continued until Nehorai and Paldi (1994) introduced a new type of sensor called the vector sensor. An acoustic vector sensor measures the acoustic pressure and all three components of the acoustic particle velocity at a single point in space. The extra information provided by the vector sensor opened the door to improved source localization accuracy without increase in array aperture. Vector-sensor models and fundamental processing techniques were developed by Nehorai and Paldi (1994) and Hawkes and Nehorai (2000) for the case of sensors located away from and in the presence of a reflecting boundary, respectively. Parametric techniques that had been designed for arrays of pressure sensors were adapted to vector sensors. For example, Wong