Fundamentals in signal analysis of passive acoustic data
Dr Cédric Gervaise & Dr Lucia Di Iorio
Chaire CHORUS Remote sensing of Aquatic Environment By Passive Acoustics
Outline
• Introduction • Introduction to sound • Measurement chain • Spectral analysis of acoustic measurements • Time-frequency representation • Applied examples from our own research
gipsa lab Lucia Slide 1 Who are we ?
Cédric Lucia
Signal processing and Application of passive acoustics application to underwater Chair CHORUS to answer biological & acoustics Environmental ecological questions monitoring using Physical oceanography passive acoustics Animal behaviour Algorithm development Evolutionary biology/ecology Environmental monitoring Environmental monitoring
gipsa lab Lucia & Cedric Slide 2 Why sound in a time series conference?
hosts medium
Services deriving from the hydrosphere : 20 900 billions $ / year
The ocean: a vast still largely unknown environment Its observation : a priority of the 21st century gipsa lab Lucia Slide 3 Costanza, R.; d'Arge, R.; De Groot, R.; Farber, S.; Grasso, M.; Hannon, B.; Limburg, K.; Naeem, S.; O'Neill, R.; Paruelo, J. & others (1997), 'The value of the world's ecosystem services and natural capital', Nature 387(6630), 253--260.
gipsa lab Why sound in a time series conference?
Emitter = sources Receiver
Soundscape
Propoagation
AQUATIC ENVIRONMENT = ACOUSTIC SOUNDSCAPE
gipsa lab Cedric slide 4 • Pijanowski, B.; Villanueva Rivera, L.; Dumyahn, S.; Farina, A.; Krause, B.; Napoletano, B.; Gage, S. & Pieretti, N. (2011), 'Soundscape ecology: the science of sound in the landscape', BioScience 61(3), 203--216.
• Radford, C.; Stanley, J.; Tindle, C.; Montgomery, J. & Jeffs, A. (2010), 'Localised coastal habitats have distinct underwater sound signatures', Mar. Ecol. Prog. Ser. 401, 21-29
• Kennedy, E.; Holderied, M.; Mair, J.; Guzman, H. & Simpson, S. (2010), 'Spatial patterns in reef generated noise relate to habitats and communities: Evidence from a Panamanian case study', Journal of Experimental Marine Biology and Ecology 395 (1 2), 85 92.
• Gervaise, C.; Di Iorio, L.; Grall, J.; Chauvaud, L. ; Jolivet, A. ; Clavier, J. (2012),’La polyphonie côtière : des sons au fonctionnement des écosystèmes’, Chapitre HDR
gipsa lab Why sound in a time series conference?
- Recent technological advances in sound acquisition - Long-term monitoring, - Continuous recordings, - High resolution, - Deployment in areas/substrates difficult to access, - Relatively cost-effective, - Real-time acquisition possible….
gipsa lab • Lucia slide 5 Why sound in a time series conference?
gipsa lab Cedric slide 6 Why sound in a time series conference?
Soundscape
Propagation
Combination of ocean science and computational science
Environmental Characterisation of Sounds description & sound sources and knowledge propagation channel
Passive acoustic monitoring of the marine environment
Passive acoustic monitoring = an indirect measurement that needs processing! gipsa lab Cedric slide 7 Outline
• Introduction • Introduction to sound • Measurement chain • Spectral analysis of acoustic measurements • Time-frequency representation
• Applied examples from our own research
gipsa lab What is a wave ?
A mechanical WAVE
Initial information Propagation of information Oscillation of a particle Transmission of oscillation Different possible waves to explore the marine environment: - electromagnetic - optic - acoustic gipsa lab • Cedric slide 8 Why acoustic waves?
2 2 I0 (W/m ) I (W/m ) r
with α = coefficient of absorption (here in cm 1)
gipsa lab Cedric slide 9 Why acoustic waves?
Frequency 50Hz, range 20000km! Frequency 18Hz, range > 3000km
Frequency
Absorption
Range
Wavelength gipsa lab • Lucia slide 10 Munk, W.; Spindel, R.; Baggeroer, A. & Birdsall, T. (1994), 'The Heard Island Feasibility Test', The Journal of the Acoustical Society of America 96(4), 2330-2342.
Clark, C. W. & Gagnon, G. J. 2004. Low frequency vocal behaviors of baleen whales in the North Atlantic: Insights from integrated Undersea Surveillance System detections, locations, and tracking from 1992 to 1996. Journal of Underwater Acoustics (USN), 52 .
gipsa lab What is an acoustic wave ?
The equation of wave propagation is obtained from: - the law of conservation of mass - the law of conservation of momentum - the law of fluid dynamics (linking P to ρ) and from their 1st order linearization around the equilibrium pressure + v=0ms -1
1 ∂2p p − = 0 Wave equation c 2 ∂t 2
ρc 2 = p p = power/m 2 r ∂v r source Piezo electric ρ0 + ∇p = 0 receiver ∂t v
gipsa lab cédric slide 11 Jensen, F. (1994), Computational ocean acoustics, Amer Inst of Physics.
gipsa lab What is a sound ? Level (received µPa, emitted µPa@1m ?) in decibel as a function of:
1 0.8 0.6 0.4 Sound intensity level: 0.2 0
signal P 0.2 P =10log ( ) 0.4 dB re P0 10 Time domain Time 0.6 P0 0.8 1 0 200 400 600 800 1000 1200 1400 1600 1800 2000 temps T Time-frequency Frequency domain 1
0.9 T = 1/f
0.8
0.7
0.6
0.5
dsp V2/Hz 0.4
0.3
0.2
0.1
0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 gipsa lab f (Hz) Cedric slide 12 Sound speed
1 ∂2p air, c=330 ms 1 p − = 0 c 2 ∂t 2 water, c=1500 ms 1 ρc 2 = p rock, c =4000 ms 1
Sea water sound velocity c= G(T,p,S)=H(T,z,S)
Take home message (for T= 0 °C and S = 35 ° /°° ) • c increases with T (4.6m/s for T = 1 °C) • c increases with S (1.4m/s for S = 1 ° /°° ) • c increases with p and z (1.7m/s for S = 1000m) gipsa lab Cedric /Lucia slide 13 Sound speed
If sound velocity c changes with depth z, sound waves do not propagate on a straight line, they are refracted.
Sound waves always tend to converge towards the sound speed minimum.
gipsa lab Cedric slide 14 Sound speed
Polar profile c Shallow water / continental shelf profile c z
z
Deep water / temperate profile c
z
gipsa lab Cedric slide 15 Temperate deep water sound popagation
x10 4
gipsa lab Cedric slide 16 Temperate shallow water propagation
gipsa lab Cedric slide 17 The intensity of sound
p = force/m 2
v Power = strength x velocity
Intensity = amount of power transmitted through a unit area (m 2) = pressure x speed (W/m 2)
For planar waves : I
gipsa lab Cedric slide 18 The intensity of sound
Example = sound wave with an amplitude of 1 Pa
In water: ( ρ=1000kg/m 2; c=1500ms 1) => I =6.10 7W/m 2 eau I 2 1 3 2 In air: ( ρ=1.2kg/m ; c=330ms ) => Iair =2,5.10 W/m
Iair / Ieau = 4166 !!!!!!!!!!!
Air : compressible gaz Water : incompressible fluide
gipsa lab Lucia slide 19 Decibels – unit of Sound (pressure) Level
J represents a quantity and Jref a reference quantity
if J is an amplitude
if J is an intensity
Decibels are used to represent and compare amplitude quantities. The sound level in dB is a scale based on multiples of 10 (logarithmic scale):
10dB = 1* 10 -11 W/m 2 -> acoustic pressure increases 10 times 20dB = 1* 10 -10 W/m 2 -> acoustic pressure increases 100 times 40dB = 1* 10 -8 W/m 2 -> acoustic pressure increases 10000 times
gipsa lab Lucia slide 20 Can Decibels be a source of confusion ? Yes or No ! Two sounds with the following levels:
Are the sound of equal amplitudes ?
The quantity is an amplitude:
remember The quantity is an intensity:
gipsa lab Cedric slide 21 Can Decibels be a source of confusion ? Yes or No ! The sound level in dB of a sum of acoustic signals is not equal to the sum of the sound levels of each signa l!!!!!!!!!!!!!!
Two independent sounds => intensity of the sum = sum of the intensities in a linear scale:
(son = sound)
then then
gipsa lab Cedric slide 22 Source Level & Reveived Level
Source Level Level heared if @ 1m form an Received Level SL dB re isotropic source RL dB ref 1 Pa 2@1m ….. 1 Pa 2…..
Emitter = sources Receiver
Soundscap e
Propagation gipsa lab Cedric slide 23 Acoustic intensity : The energetic budget of a singing blue whale
Mean source level of blue whale song note: 160 dB re 1 Pa @ 1m (188 dB peak), signal duration = 18s with 960 signals/day, energetic value of krill = 96kcal/100g.
gipsa lab Biophony From invertebrates… Range
1m - > 200m
10m - > 1km
> 100m - > 1000km
… to marine mammals gipsa lab Lucia slide 24 Geophony Wind & rain Sounds of breaking ice
Underwater earthquake
gipsa lab Lucia slide 25 Anthropophony
gipsa lab Lucia slide 26 Sound Levels
Source (SL) Mesure (RL) Wideband effective source Wideband effective received level (SL), level (RL),
2 2 SL =10log10 rl( ); dBref1µPa @1m RL =10log10 rl( ); dBref1µPa 1 1 sl = m )t( 2 dt rl = m )t( 2 dt T ∫ T ∫ T T
Narrowband source level Narrowband received level Power spectral density Power spectral density
2 2 γ SL )f( = dBref1µPa /Hz@1m γRL )f( = dBref1µPa /Hz
Sound exposure level (SEL)
2 SEL =10log10 (sel); dBref1µPa s sel = ∫m )t( 2 dt T
gipsa lab Lucia slide 27 An inventory initatied by 2nd world war and cold war
Source whoi.edu, Wenz 1962, Wenz 1972, Urick 1984 gipsa lab Cedric slide 28 Natural sources Abiotic Biotic
Anthropogenic sources
gipsa lab Cedric slide 29 Ambient noise Background noise from many different sources excluding individually identifiable sounds
gipsa lab Cedric slide 30 Verydynamic amplitude large
Very large frequency dynamic gipsa lab Cedric slide 31 Hildebrand, J. (2009), 'Anthropogenic and natural sources of ambient noise in the ocean', Marine Ecology Progress Series 395, 5--20.
Wenz, G. (1972), 'Review of underwater acoustics research: Noise', The Journal of the Acoustical Society of America 51, 1010.
Wenz, G. M. (1962), 'Acoustic Ambient Noise in the Ocean: Spectra and Sources', The Journal of the Acoustical Society of America 34(12), 1936-1956.
gipsa lab Outline
• Introduction • Introduction to sound • Measurement chain • Spectral analysis of acoustic measurements • Time-frequency representation • Applied examples from our own research
gipsa lab Some measurement devices
10 miles
1826, lake of Geneva: Calladon & Strum gipsa lab Lucia slide 32 Some measurement devices
Portable device
Amplifier
IN
IN OUT
Recorder Hydrophone
gipsa lab Lucia slide 33 Some measurement devices
Autonomous recorders
RT SYS, France Aural, Multi Electronique, Qc.
gipsa lab Lucia slide 34 Some measurement devices
Cabled systems http://www.medon.info/
Yves Gladu
SO und SU rveillance System
gipsa lab Lucia slide 35 The elements of the acquisition chain
RS 20 s1 =10 p s p RS 1 dB = dB +
Amplifier
G 20 s 2 = 10 s1 Analogue digital s s G converter 2 dB = 1 dB +
Recorder/player
s s = (E 2 2nb−1 ) 3 D 2nb−1 s = s + 20log10( ) 3 dB 2 dB D
Sound analysis tools
gipsa lab Cedric slide 36 Receiver sensitivity
RSdB ref1V /1µPa
gipsa lab Cedric slide 37 A key element : the analogue to digital conversion
It allows the use of digital technology to process the acquired data.
gipsa lab Cedric slide 38 The analogue to digital conversion
s(t) t : independent variable s : dependent variable Discrete or continuous variable
Numérique/numérisation = digital/digitalisation Echantillonnage = sampling rate gipsa lab Cedric slide 39 Sampling rate
fmax is the maximal frequency of a signal s(t) only if
∀f > fma x ,S(f) = 0
The sampling of a signal s(t) is reversible only if s(t) is low-pass and
fe>2f max
gipsa lab Cedric slide 40 Sampling rate
gipsa lab Cedric slide 41 Sampling rate – the anti aliasing Physical filter phenomenon
fmax
measurement noise
f>fmax receiver f >2f fc≈fmax e c d filter PB d ⊗ fc ×××Te gipsa lab Cedric slide 42 Quantification The 3 key parameters: Vmin Vmax Number of quantification levels N D : dynamic range q : quantification step Quantification rule: b : number of bits gipsa lab Cedric slide 43 Quantification errors Saturation Discretization s(t) sq(t)=s(t)+ ε(t) ε(t) The quantification error limited Evenly distributed over white noise -> constant spetral component high frequencies gipsa lab Cedric slide 44 How to set a quantification chain? Case: need to measure two signals: an intense (amplitude A1) and a low one (amplitude A2) A1 Choose Vmin, Vmax, N in order to : A2 1) Avoid saturation of the intense signal q=2A /N 1 2) Record the faint signal with a minimal signal to noise ratio (so that it is audible)