Indoor Localization in Energy Constrained Wireless Acoustic Sensor Networks

Advances in energy efficient electronic components create new opportunities for battery powered Wireless Acoustic Sensor Networks (WASNs). Such sensors can be deployed to localize unwanted and unexpected sound events in surveillance applications, home assisted living, etc. Commercialization of these systems is nowadays limited due to their high energy consumption, which is mainly a result of the high data rates required to wirelessly stream the audio data. This research concentrates on the complete system design of a Small Form Factor (SFF) Ultra-Low-Power (ULP) WASN node consisting of a MEMS microphone array. The array consists of 4 microphone elements, each having their own switchable two-stage amplifier. To optimize power consumption, it is completely configurable using an on board low-power microcontroller, enabling the activation of the required elements when they are actually needed. Due to the selection of ULP components and smart switching, the power consumption of the array is much lower than other similar designs, consuming only 1.4 mW when the array is activated and 66 nW during sleep. The proposed sensor node design is able to process an acoustic event every 3.4 s during one year. The microphone array enables the node to compute the incident angle of sound events. The angular information is wirelessly transmitted to a central localization system, estimating the location of the sound event. Since only the angle and not the complete audio stream is transmitted, a lot of energy is saved in wireless transmission. To benefit from this reduction, a computationally efficient delay-based Angle of Arrival (AoA) calculation method is implemented and depends on the cross correlation between microphone elements. Common Angle of Arrival (AoA) localization approaches use triangulation relying on a single angle per sensor node. However this work presents a way of using angular Probability Density Functions (PDFs) combined with a matching algorithm to localize sound events. The inputs of the matching algorithm are either precalculated Line-Of-Sight (LOS)-PDFs or fingerprints obtained from a prerecording stage using a White gaussian Noise (WN) signal. The localization algorithm is experimentally evaluated in a 4.26 m by 9.24 m room, localizing WN and vocal sounds. The localization results demonstrate the superior accuracy of the proposed matching algorithm over the common triangulation approach. When the WN fingerprinting technique is applied, it is possible to localize sound sources using a single reference node. Finally, the localization system is evaluated in a real life setup, where sub-meter mean localization results are obtained.

Publication year:2019

Accessibility:Open