Real-Time Audio Processing on FPGA for Microphone Arrays

Abstract

The widespread adoption of drones has presented new requirements for reliable
localization and tracking systems. For this reason acoustic sensing has become an
area of growing interest for a passive and cost-effective solution. However, process-
ing large amounts of microphone data in real time presents significant hardware
challenges.
This thesis studies two different types of signal channelization techniques de-
signed to reduce the data rate while preserving acoustic features for beamforming
and drone tracking. A filter bank consisting of parallel band-pass filters is com-
pared to a Fast Fourier Transform (FFT)-based approach. The channelizers are
paired with a dynamic subband selection algorithm that isolates active frequency
components and lowers the total bandwidth. Both designs are implemented on an
Xilinx Zynq-7000 Field-Programmable Gate Array (FPGA) and integrated into a
custom microphone array sampling system.
Both of the systems are able to reduce the data rate to less than 30% compared
to the raw audio data. The filter-bank channelizer achieved slightly lower data
rates while the FFT implementation provides greater spectral isolation and has
more resources available for future development.

Popular Abstract

Drones have become an increasingly common sight in our skies. With their num-
bers increasing so do the risks and incidents. To ensure safe operations and protect
flight restricted areas drone detection systems are quickly becoming essential for
vital infrastructure.
One of the simplest and most cost-effective ways to detect drones is to listen
for them. Drones make a distinctive sound and by arranging microphones in an
array, a system can both detect, locate and track them. This method is called
acoustic detection and it works much like how our ears help us figure out where a
sound is coming from.
A big challenge with real-time acoustic detection systems is the large amount
of data produced by the many microphones. This high data rate constrains the
system, often forcing a compromise between resolution and refresh rate. This
thesis explores ways to reduce the amount of data without losing important in-
formation. To achieve this two signal processing methods are designed, simulated
and implemented into an existing sound sampling system running on an FPGA.
The first method performs filter operations on each incoming sample to divide
the frequency spectrum. This is followed by a dynamic selection algorithm to
extract only the active parts of the audio spectrum. The second method works in
a similar way but instead of filters it utilizes a Fast Fourier Transform to divide
the frequency spectrum. The limited resources on the FPGA, the large number
of microphones and the real-time requirement force both designs to be highly
optimized and efficient for their application.
The result is two systems that both significantly reduce the data rate. When
recording a moving drone, both channelizer implementations achieve average data
rates of less than 30% compared to the original system. This is made possible
by effectively isolating and transmitting only the parts of the sound spectrum
produced by the drone. The significant reduction in data boosts the performance
of the acoustic detection system by reducing the processing load.
To summarize, the system listens and extracts only the valuable parts of the
audio, significantly reducing the data rate. This makes acoustic detection more
efficient which helps keep our skies safer.

• author: Nilsson, Ivar
• supervisor: Liang Liu
• organization: Department of Electrical and Information Technology
• course: EITM01 20251
• year: 2025
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• keywords: FPGA Signal Processing, Microphone Array, Real-Time, Beamforming, Drone Detection, FFT-based Channelizer, Filter-bank Channelizer, Audio Signal Processing
• report number: LU/LTH-EIT 2025-1078
• language: English
• id: 9205511
• date added to LUP: 2025-06-26 13:17:13
• date last changed: 2025-06-26 13:17:13