Princeton University

School of Engineering & Applied Science

LLAMAS: Large-Area Microphone Arrays and Sensing Systems

Speaker: 
Josue Sanz-Robinson
Location: 
Engineering Quadrangle J323
Date/Time: 
Thursday, May 12, 2016 - 1:00pm to 2:30pm

Abstract
 
As electronics becomes ever more pervasive in our daily lives, it will no longer be confined to our phones and tablets, but rather will be seamlessly integrated into the physical environment in which we live, work, and play; thus, fostering collaborative spaces and enhancing interpersonal interactions. To support this vision not only do we require systems with a flexible form factor, for ease of deployment, but also systems that can readily support large numbers of spatially distributed sensors. This has led us to develop hybrid sensing systems, based on exploiting the complementary strengths of large-area electronics (LAE) and CMOS. LAE provides sensing functionality over physically expansive spaces, while CMOS has the computing capacity to transform the complex data from the sensors into actionable inferences about our environment. As a vehicle to explore the design of hybrid systems, we will focus on two aspects:
 
1) Thin-film Diodes: A key challenge for hybrid systems is how to robustly and scalably connect the LAE and CMOS domains. One approach is to use non-contact inductive interfaces to transmit data or power between these two domains, but this requires high performance, thin-film rectifiers. This led us to develop amorphous / nanocrystalline silicon diodes, which are deposited at low temperatures (less than 200°C) to enable processing on plastic.
 
2) Blind Source Separation: We develop a system for separating the voices of multiple simultaneous speakers, which can ultimately be fed to a voice-command recognition engine for controlling electronic systems. The cornerstone of this system is a large-area microphone array (2.25 m wide), attached to the wall, which uses a custom designed beamforming algorithm to isolate different sources in a real, reverberant room. A key feature of this algorithm is that it is “blind”, since it requires no knowledge about the location of the speakers or microphones, but instead it adapts to the unique acoustic environment of the room.