Tomoya Takagi

An ultra-low-power VAD hardware implementation for intelligent ubiquitous sensor networks

We propose a power management method using a digital voice activity detection (VAD) module for intelligent ubiquitous sensor systems. When this VAD module detects a speech signal, a main signal processing circuit is connected to a power source. When no speech signal is detected, most circuits except VAD are blocked off, thereby reducing stand-by power for the specialized sensor nodes used for speech signal processing. We implemented the VAD algorithm, using zero crossing of input signals to an FPGA, thereby achieving 2.10 mW operation. We synthesized this VAD module using CMOS 0.18-µm process, achieving 3.49 µW power consumption for operation at 1.8 V and 100 kHz.

Microphone array network for ubiquitous sound acquisition

We propose a microphone array network that realizes ubiquitous sound acquisition. Nodes with 16 microphones are connected to form a huge sound acquisition system that carries out VAD, sound source localization and separation. The three operations are distributed among nodes. The VAD is implemented to manage power consumption. Consequently, the system consumes little power when speech is not active. The VAD module uses only 2.1 mW. The system can improve an SNR by 7.75 dB using 112 microphones.

Data Aggregation Protocol for Multiple Sound Sources Acquisition with Microphone Array Network

In this paper, we propose a microphone array network that realizes ubiquitous sound acquisition for multiple sound sources. Several nodes with 16 microphones are connected to form a huge sound acquisition system, which carries out voice activity detection (VAD), sound source localization, and enhancement. The three operations are distributed among nodes with multi-hop data communication. Using the distributed network, we produce a low-traffic data-intensive array network. In the sound source enhancement, we combine the delay-and-sum beam-forming algorithm with network data aggregation. The prototype of microphone array node is implemented in SUZAKU FPGA boards, which demonstrates a real-time sound-source enhancement operation.

0.45-V operating V t -variation tolerant 9T/18T dual-port SRAM

This paper proposes a dependable dual-port SRAM with 9T/18T bitcell structure. The proposed SRAM has two operating modes: a 9T normal mode and an 18T dependable mode. The 9T bitcell has an outside single-ended bitline as a dedicated read port along with a pair of conventional differential inside bitlines. Therefore, the 18T bitcell has two differential pairs of the outside bitlines and inside bitlines. For the dedicated read port, the 18T bitcell can exploit a differential sense amplifier operating at low voltage, but the 9T bitcell must have a single-ended readout inverter at high voltage. To achieve the 9T/18T SRAM architecture, an interleaved bitline scheme is incorporated for the dedicated read port. The 9T/18T dual-port SRAM can scale its speed, operating voltage, and power dynamically by combining two bitcells for one-bit information. We designed and fabricated the proposed SRAM using a 65-nm process. The measurement results show that the dependable read mode using the pair of the single-ended bitlines can reduce the operation voltage to 0.45 V at a frequency of 1 MHz because of the disturb-free read port, although the dependable read mode using the inside bitlines needs 0.54 V at the same frequency.

Live demonstration: Intelligent ubiquitous sensor network for sound acquisition

We propose a microphone array network that realizes ubiquitous sound acquisition. Nodes with 16 microphones are connected to form a large sound acquisition system that carries out voice activity detection (VAD), sound source localization and sound source separation. The three operations are distributed among nodes using network. Because the VAD is implemented to manage power consumption, the system consumes little power when speech is not active. The power of the VAD module is only 2.1 mW on an FPGA. The system can improve an SNR by 7.75 dB using 112 microphones.

Intelligent ubiquitous sensor network for sound acquisition

We propose a microphone array network that realizes ubiquitous sound acquisition. Nodes with 16 microphones are connected to form a large sound acquisition system that carries out voice activity detection (VAD), sound source localization and sound source separation. The three operations are distributed among nodes using network. Because the VAD is implemented to manage power consumption, the system consumes little power when speech is not active. The power of the VAD module is only 2.1 mW on an FPGA. The system can improve an SNR by 7.75 dB using 112 microphones.

Low-Traffic and Low-Power Data-Intensive Sound Acquisition with Perfect Aggregation Specialized for Microphone Array Networks

We propose a microphone array network that realizes ubiquitous sound acquisition. Several nodes with 16 microphones are connected to form a novel huge sound acquisition system, which carries out voice activity detection (VAD), sound source localization, and separation. The three operations are distributed among nodes. Using the distributed network, we achieve a low-traffic data-intensive array network. To manage nodes’ power consumption, VAD is implemented. Consequently, the system uses little power when speech is not active. For sound localization, a network-connected multiple signal classification (MUSIC) algorithm is used. The sound separation system can improve a signal-noise ratio (SNR) by 7.75 dB using 112 microphones. Network traffic is reduced by 99.11% when using 1024 microphones.

Data-Intensive Sound Acquisition System with Large-scale Microphone Array

We propose a microphone array network that realizes ubiquitous sound acquisition. Several nodes with 16 microphones are connected to form a novel huge sound acquisition system, which carries out voice activity detection (VAD), sound source localization, and sound enhancement. The three operations are distributed among nodes. Using the distributed network, we produce a low-traffic data-intensive array network. To manage node power consumption, VAD is implemented. The system uses little power when speech is not active. For sound localization, a network-connected multiple signal classification (MUSIC) algorithm is used. The experimental result of the sound-source enhancement shows a signal-noise ratio (SNR) improvement of 7.75dB using 112 microphones. Network traffic is reduced by 99.11% when using 1,024 microphones.

Positioning system for mobile terminals using a microphone array network as an intuitive interface

Physical positions are quite useful to realize an intuitive interface for communication among mobile terminals. We propose the placement of a microphone array on mobile terminals and the use of a network as an intuitive interface. The mobile terminals can obtain their relative positions by emitting a sound, which facilitates estimation of the directions of arrival (DOAs) among them. We produced a prototype using a tablet PC, 16 microphones, and an FPGA board. Then, we implemented two applications, exploiting the positioning system to elucidate its wider possibilities.