A Binaural Neuromorphic Auditory Sensor for FPGA: A Spike Signal Processing Approach

Area of Science:

• Neuroscience
• Electrical Engineering
• Computer Science

Background:

• Traditional digital cochleae use classical digital signal processing.
• Neuromorphic computing offers event-driven, spike-based information processing.
• Spike-based auditory sensors can mimic biological auditory systems.

Purpose of the Study:

• To present a novel neuromorphic binaural auditory sensor architecture and design flow.
• To implement and analyze the system on a field-programmable gate array (FPGA).
• To enable researchers to study brain-inspired audio processing and learning.

Main Methods:

• Designed a neuromorphic binaural auditory sensor entirely in the spike domain using pulse frequency modulation.
• Developed a systematic design process for creating tunable, feature-rich audio frequency decomposers.
• Implemented a 64-channel binaural system on a Virtex-5 FPGA and tested with diverse audio signals.

Main Results:

• Achieved a frequency range of 9.6 Hz to 14.6 kHz (adjustable).
• Demonstrated a maximum output event rate of 2.19 Mevents/s with low power consumption (29.7 mW).
• The system utilized 11141 slices on the FPGA operating at a 27 MHz clock frequency.

Conclusions:

• The proposed neuromorphic auditory sensor provides frequency-decomposed audio information via an address-event representation interface.
• The generic design process facilitates custom neuromorphic auditory system synthesis on low-cost FPGAs.
• The implemented system offers a valuable tool for investigating auditory processing and learning mechanisms in the brain.