Distributed and Collaborative Processing of Audio Signals: Algorithms, Tools and Applications

Abstract

This thesis fits into the field of Information and Communications Technology (ICT), especially in the area of digital signal processing. Nowadays and due to the rise of the Internet of Things (IoT), there is a growing interest in wireless sensor networks (WSN), that is, networks composed of different types of devices specifically distributed in some area to perform different signal processsing tasks. These devices, also referred to as nodes, are usually equipped with electroacoustic transducers as well as powerful and efficient processors with communication capability. In the particular case of acoustic sensor networks (ASN), nodes are dedicated to solving different acoustic signal processing tasks. These audio signal processing applications have been undergone a major development in recent years due in part to the advances made in computer hardware and software. The development of powerful centralized processing systems has allowed the number of audio channels to be increased, the control area to be extended or more complex algorithmms to be implemented. In most cases, a distributed ASN topology can be desirable due to several factors such as the limited number of channels used by the sound acquisition and reproduction devices, the convenience of a scalable system or the high computational demands of a centralized fashion. All these aspects may lead to the use of novel distributed signal processing techniques with the aim to be applied over ASNs. To this end, one of the main contributions of this dissertation is the development of adaptive filtering algorithms for multichannel sound systems over distributed networks. Note that, for sound field control (SFC) applications, such as active noise control (ANC) or active noise equalization (ANE), acoustic nodes must be not only equipped with sensors but also with actuators in order to control and modify the sound field. However, most of the adaptive distributed networks approaches used to solve soundfield control problems do not take into account that the nodes may interfere or modify the behaviour of the rest. Therefore, other important contribution of this thesis is focused on analyzing how the acoustic system affects the behavior of the nodes within an ASN. In cases where the acoustic environment adversely affects the system stability, several distributed strategies have been proposed for solving the acoustic interference problem with the aim to stabilize ANC control systems. These strategies are based on both collaborative and non-collaborative approaches. Implementation aspects such as hardware constraints, sensor locations, convergenge rate or computational and communication burden, have been also considered on the design of the distributed algorithms. Moreover and with the aim to create independent-zone equalization profiles in the presence of multi-tonal noises, distributed narrowband and broadband ANE algorithms over an ASN with a collaborative learning and composed of acoustic nodes have been presented. Experimental results are presented to validate the use of the distributed algorithms proposed in the work for practical applications. For this purpose, an acoustic simulation software has been specifically designed to analyze the performance of the developed algorithms. Finally, the performance of the proposed distributed algorithms for multichannel SFC applications has been evaluated by means of a real practical implementation. To this end, a real-time prototype that controls both ANC and ANE applications by using collaborative acoustic nodes has been developed. The prototype consists of two personal audio control (PAC) systems composed of a car seat and an acoustic node, which is equipped with two loudspeakers, two microphones and a processor with communications capability. In this way, it is possible to create two independent noise control zones improving the acoustic comfort of the user without the use of headphones.