Characterization and Suppression of Wind Noise Using a Large-Scale

Infrasound Sensor Array

Carrick L. TalmadgeDoug ShieldsKenneth E. GilbertThe University of Mississippi, Oxford, MS USA

Infrasound Detection

sourcepropagation

receiver

• Source characterization (wave form, signal level, directionality, etc.);.

• Effects of vertical variation in wind-velocity and temperature profiles; including diurnal effects

• Effects of turbulent scattering

• Terrain effects

Must consider effects of:• wind-generated sound

• atmospheric turbulence (intrinsic pressure fluctuations)

• pressure probe body on pressure measurements

Robust system design needs data-driven models of all three components!

Advantages of Sensor Arrays

Arrays provide

•directionality (source location);

•noise reduction using more sophisticated DSP algorithms;

•the ability to separately characterize the signal and the noise.

Old Array Technology

NSensors

D

A/D

Signal Processing

Total Cable Required N2D/4

Example: N = 40 D = 5m L = 200m Total Cable = 2000m

New Array Technology

NSensors

D

A/D

P

A/D

P

A/D

P

A/D

P

A/D

P

A/D

P

Signal Processing

Total Cable Required ND

Example: N = 40 D = 5m L = 200m Total Cable = 200m

Smart Sensor Arrays

Components of smart sensor element:

• Ruggedized sensor and sensor body

• Signal conditioning with programmable gain and anti-alias filter

• Analog-to-digital stage with programmable sampling rate

• Microcontroller-based networking using RS-485 standard (~ 2Mbps data rate).

• Remotely controlled self-calibration and diagnostic testing of individual sensors.

Approximate delivery dates for bimorph sensors:

8 prototypes January 1, 2002

100 manufactured units March 1, 2002.

Second-Generation Array

RS-485 networked sensor elements

DSP-based“collector” units with RS-485 to wireless modem

wireless link

central processor

sub

arra

ys

Bimorph Capsules

• Resonant Frequency - 3 kHz

• Sensitivity - 1 to 4 mV/Pa• Temperature Compensation

– Reverse bimorphs– Insulated enclosures,

small openings• Charge Generating

– Must operate into a high impendence

Potted Capsules & Housing

Acoustic Sensor Development

Gabrielson Piezo-Resistive Transducer

Features: • 0.005 - 1000 Hz• High Sensitivity (20 mV/Pa)• Low Power Usage• Low Cost (Less than $250 per

unit)

9V Batteries

Add anti-alias filter and 24-bit sigma-delta A/D converter. Replace acrylic back volume with insulated

brass back volume instrumented for internal temperature.

Move first-stage amplifier onto sensor board (small-outline IC).

Oxford Airport Array Experiment

Chris Clark, Ken Gilbert, Doug Shields, Carrick Talmadge, Ron Wagstaff Chad Williams, Jay Williams, Zak Williams

Experiment took place on April 6, 2001

40 minutes of data were collected over an ~ 1 hr interval

2 microphone arrays, sonic anemometer

Levels calibrated using a B&K sound meter

(not to scale)

Experimental Setup

southB&K array (6x)

anemometer

bimorph array (11x)

Single Sensor (B&K) ResultsS

ensor #

Successive sensor outputs have been shifted by –10Pa.Run #1 (50-Hz tone) — initial 10 seconds of recording

Amplitude and Phase Shown is the amplitude and phase for the 50-Hz tone from the sensor 1 read-out.

The propeller-driven plane which lands around 150 seconds had no noticeable influence on the 50-Hz tone…

“pops”

“Prop” plane landing (CPA)

Run 1: 50-Hz tone

Run 1: 50-Hz tone

Frequency [Hz]

Pressure [Pa]

Time [sec]

Si g

nal

Am

pl i

tud

e

Frequency [Hz]

Run 1: 50-Hz ToneTime Series Spectrum (Fourier Transform)

RMS Averaged Spectrum

Time [sec]

Lev

e l [

dB

SP

L]

Cou

nt

Level [dB SPL]

40-Hz bin 40-Hz bin

50-Hz bin

60-Hz bin 60-Hz bin

50-Hz bin

Run 1: 50-Hz Tone median RMS

Run 1: 50-Hz tone Note that “pops” in playback create erroneous spectra using RMS spectrogram.

Median spectrogram is more robust and will be used here throughout.

Calibration of B&K sensors is excellent between 10-250 Hz (error is on the order of 1 dB).

(RMS)

median

Run 6: 25-50 Hz Tones, 5-Hz Steps

wind-generated sound

Yocona River Array Experiment

1

16

15

30

6.46 m

0.8 mile

Sound SourceObservation Station

Observation Station350 meters

Yocona River

Hwy 7

NPrevailing Wind

Sewage Disposal Plant

Prevailing Wind

0 500 1000 1500 2000 2500 3000 3500 4000-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

LAG TIME IN SEC X 500

CO

RR

EL

AT

ION

CO

EF

FIC

IEN

T

CORRELATION VS LAG TIME

correlation between sensors 1 and 3

wind(1) time = 5-45 sec

0 2 4 6 8

116

15

30

6.46 m

N

3

-3 -2 -1 0 1 2 30.4

0.5

0.6

0.7

0.8

0.9

1

CORRELATION DISTANCE (m)

CO

RR

EL

AT

ION

CO

EF

FIC

IEN

T

CORRELATION COEFFICIENT VS CORRELATION DISTANCE

WIND(1) SENSORS 7-12, TIME 5-45 SECNOTE LOW CORRELATION

-3 -2 -1 0 1 2 30.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

CORRELATION DISTANCE (m)C

OR

RE

LA

TIO

N C

OE

FF

ICIE

NT

CORRELATION COEFFICIENT VS CORRELATION DISTANCE

wind(1) sensors 21-26, time 5-45 secnote high value of correlation

Wind Noise Correlation Strength vs. Distance

Correlation Parallel to Wind Direction Correlation Transverse to Wind Direction

Separation [m] Separation [m]

Max

imum

Cor

rela

tion

Conclusions

• Large-scale microphone arrays are a versatile, transportable alternative to pipe arrays.

• In general, they provide superior wind-noise reduction due to the more sophisticated DSP algorithms which are possible.

• The three components of wind noise (intrinsic pressure fluctuations; wind-generated sound; probe-body induced pressure fluctuations) can be decoupled and separately characterized.