Space-Time Audio processing based on Ray-space Transform (START) ;
For decades space-time processing has been an active area of research that focuses on a wide range of applications and involves different fields, ranging from radar sonar systems to acoustics. This branch of signal processing is currently facing an increased level of interest in both the academy and the industry thanks to the technological development. In fact, acoustic transducers are undergoing a sudden and dramatic transformation. Soon we will have access to seas of ultra-low-cost MEMS digital microphones and loudspeakers, deployed and seamlessly integrated with the environment, for applications of sound field capturing, processing and rendering that were unthinkable until now. This exciting innovation raises some problems on how to represent the signals coming from this huge number of sensors.
A particularly efficient representation paradigm is based on the concept of geometrical acoustics, and it assumes that acoustic radiance is constant along straight lines known as acoustic rays. Within this context, the Ray Space Transform, based on the use of Gabor frames, has revealed itself as a powerful and flexible tool for representing both near and far field data enabling invertible space-time decomposition of the array data. This tool is inherently based on the geometric paradigm and, in addition, it overcomes the limitations of geometrical acoustics as it extends the validity also to lower frequencies.
However, despite the fact that this representation solves some issues related to the problem of sound field representation, it has been studied, so far, to work with linear array of microphones. This poses some limitations in practical situations where the distribution of the sensors is unknown, thus opening new research challenges. In fact, we can think to have micro-arrays distributed in the space or even single sensor distributions. For these two cases, we have to rethink the ray space representation in order to gather as much information as we can. In the first case we can exploit the fact that a micro-array alone can acquire information related to the distribution of the energy along the rays incident to the array. In the second case, this is not true because the signal from a single sensor does not provide information related to the direction of propagation of the incoming signals. More specifically, the signal sensed by a microphone turns out to be the integral of the radiance of the rays crossing it.
In addition to representation problems, we can think of more practical problems when dealing with huge numbers of sensors. In particular, in both cases of distributions of micro-arrays and single sensors, there are problems such as self-calibration and synchronization. These are two extremely stimulating issues as well, on which we can investigate using the structure of the ray space representation. Moreover, many exciting applications can be envisaged from the development of the proposed frame-work ranging from ”classical” problems found in the array processing literature (e.g blind signal extraction) to exciting popular research topics such as the reconstruction of the acoustic scene (i.e. environment and sources) from the acquisition accomplished through a compact microphone array (ego-motion).
A particularly efficient representation paradigm is based on the concept of geometrical acoustics, and it assumes that acoustic radiance is constant along straight lines known as acoustic rays. Within this context, the Ray Space Transform, based on the use of Gabor frames, has revealed itself as a powerful and flexible tool for representing both near and far field data enabling invertible space-time decomposition of the array data. This tool is inherently based on the geometric paradigm and, in addition, it overcomes the limitations of geometrical acoustics as it extends the validity also to lower frequencies.
However, despite the fact that this representation solves some issues related to the problem of sound field representation, it has been studied, so far, to work with linear array of microphones. This poses some limitations in practical situations where the distribution of the sensors is unknown, thus opening new research challenges. In fact, we can think to have micro-arrays distributed in the space or even single sensor distributions. For these two cases, we have to rethink the ray space representation in order to gather as much information as we can. In the first case we can exploit the fact that a micro-array alone can acquire information related to the distribution of the energy along the rays incident to the array. In the second case, this is not true because the signal from a single sensor does not provide information related to the direction of propagation of the incoming signals. More specifically, the signal sensed by a microphone turns out to be the integral of the radiance of the rays crossing it.
In addition to representation problems, we can think of more practical problems when dealing with huge numbers of sensors. In particular, in both cases of distributions of micro-arrays and single sensors, there are problems such as self-calibration and synchronization. These are two extremely stimulating issues as well, on which we can investigate using the structure of the ray space representation. Moreover, many exciting applications can be envisaged from the development of the proposed frame-work ranging from ”classical” problems found in the array processing literature (e.g blind signal extraction) to exciting popular research topics such as the reconstruction of the acoustic scene (i.e. environment and sources) from the acquisition accomplished through a compact microphone array (ego-motion).
Back to Alumni