LLAMAS: Large-Area Microphone Arrays and Sensing Systems

Abstract

Large-area electronics (LAE) provides a platform to build sensing systems, based on distributing large numbers of densely spaced sensors over a physically-expansive space. Due to their flexible, "wallpaper-like" form factor, these systems can be seamlessly deployed in everyday spaces. They go beyond just supplying sensor readings, but rather they aim to transform the wealth of data from these sensors into actionable inferences about our physical environment. This requires vertically integrated systems that span the entirety of the signal processing chain, including transducers and devices, circuits, and signal processing algorithms. To this end we develop hybrid LAE / CMOS systems, which exploit the complementary strengths of LAE, enabling spatially distributed sensors, and CMOS ICs, providing computational capacity for signal processing.

To explore the development of hybrid sensing systems, based on vertical integration across the signal processing chain, we focus on two main drivers: (1) thin-film diodes, and (2) microphone arrays for blind source separation:

1) Thin-film diodes are a key building block for many applications, such as RFID tags or power transfer over non-contact inductive links, which require rectifiers for AC-to-DC conversion. We developed hybrid amorphous / nanocrystalline silicon diodes, which are fabricated at low temperatures (<200 deg C) to be compatible with processing on plastic, and have high current densities (5 A/cm^2 at 1 V) and high frequency operation (cutoff frequency of 110 MHz).

2) We designed 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. On a device level, we developed flexible PVDF microphones, which were used to create a large-area microphone array. On a circuit level we developed localized a-Si TFT amplifiers, and a custom CMOS IC, for system control, sensor readout and digitization. On a signal processing level we developed an algorithm for blind source separation in a real, reverberant room, based on beamforming and binary masking. It requires no knowledge about the location of the speakers or microphones. Instead, it uses cluster analysis techniques to determine the time delays for beamforming; thus, adapting to the unique acoustic environment of the room.