University of Parma

Angelo Farina Angelo Farina

Industrial Engineering Dept.

Industrial Engineering Dept.

University of Parma

University of Parma - - ITALY ITALY HTTP://

HTTP:// www.angelofarina.it www.angelofarina.it

University of Parma

Advancements in impulse Advancements in impulse

response measurements by response measurements by

sine sweeps sine sweeps

122

AES

Convention

2007 May 5–8

Vienna, Austria

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

Basic

Basic sound sound propagation propagation scheme scheme

Omnidirectional receiver Direct Sound

Reflected Sound

Point Source

Direct Sound

Reverberant tail

Measurement process

ƒ The desidered result is the linear impulse response of the acoustic propagation h(t). It can be recovered by knowing the test signal x(t) and the measured system output y(t).

ƒ It is necessary to exclude the effect of the not-linear part K and of the background noise n(t).

Not-linear, time variant

system K[x(t)]

Noise n(t)

input x(t)

+ output y(t) linear system

w(t)⊗h(t) distorted signal

w(t)

Exponential Sine Sweep method

ƒ x(t) is a sine signal, which frequency is varied

exponentially with time, starting at f₁ and ending at f₂.

⎥ ⎥

⎥ ⎥

⎦

⎤

⎢ ⎢

⎢ ⎢

⎣

⎡

⎟⎟

⎟ ⎟

⎠

⎞

⎜⎜

⎜ ⎜

⎝

⎛

−

⋅

⎟⎟ ⎠

⎜⎜ ⎞

⎝

⎛

⋅

⋅ π

= ⋅ ^{⎟⎟ ⎠}

⎜⎜ ⎞

⎝

⋅ ⎛

1 e

f ln f

T f

sin 2 )

t (

x ¹

2

f ln f T

t

1

2

1

Test Signal – x(t)

Measured signal - y(t)

ƒƒ Not-Not-linearlinear behaviourbehaviourofof the loudspeakerthe loudspeaker causescauses manymany harmonicsharmonics totoappearappear

Inverse Filter – z(t)

The The deconvolutiondeconvolution ofof the IR isthe IR is obtainedobtained convolvingconvolving the the measuredmeasured signalsignal y(t)

y(t) withwith the inverse filterthe inverse filter z(t) [equalizedz(t) [equalized, , time-time-reversedreversed x(t)]x(t)]

Deconvolution of Log Sine Sweep

The “time reversal mirror” technique is employed: the system’s impulse response is obtained by convolving the measured signal y(t) with the time-reversal of the test signal x(-t). As the log sine sweep does not have a “white” spectrum, proper equalization is required

Test Signal x(t) Inverse Filter z(t)

Result of the deconvolution

The last impulse response is the linear one, the preceding

2° 1°

3°

5°

IR IR Selection Selection

ƒƒ AfterAfter the sequencethe sequence ofofimpulseimpulse responsesresponseshashas beenbeen obtained, obtained, itit isis possiblepossible totoselectselect and and extractextractjust onejust one ofofthemthem::

Post processing of impulse responses

ƒƒ A specialA special pluginplugin hashas beenbeen developeddevelopedforfor the computationthe computation ofof STI accordingSTI according toto IEC-IEC-EN 60268EN 60268--16:200316:2003

Post processing of impulse responses

ƒƒ A specialA special pluginplugin hashas beenbeen developeddevelopedforfor performingperforminganalysisanalysis ofofacousticalacoustical parameters

parameters accordingaccording totoISO-ISO-33823382

The new AQT plugin for Audition

ƒƒ The newThe new modulemodule isis stillstillunder developmentunder development and willand will allowallow forfor veryvery fast fast computation

computation ofofthe AQT (Dynamicthe AQT (Dynamic FrequencyFrequency ResponseResponse) curve from) curve from withinwithin Adobe

Adobe AuditionAudition

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

Pre-ringing at high and low frequency

ƒƒ Pre-Pre-ringingringing at high frequencyat high frequency due todue to improperimproperfadefade-out-out

ThisThispicturepicture showsshows the preringingthe preringing obtainedobtained deconvolvingdeconvolvingdirectlydirectly the test the test signal

signal, , withoutwithout passingpassingthroughthrough the system under testthe system under test

Pre-ringing at high and low frequency

ƒƒ PerfectPerfect Dirac’Dirac’s delta s delta afterafter removingremovingthe fadethe fade--outout

ThisThis picturepictureshowsshows the resultthe result obtainedobtained deconvolvingdeconvolvingdirectlydirectly the test signalthe test signal, , without

without passingpassingthroughthrough the system under test, and employingthe system under test, and employing a sinea sine sweep

sweep goinggoing up toup to the Nyquistthe Nyquist frequencyfrequency

Pre-ringing at high and low frequency

ƒƒ Pre-Pre-ringingringing at low frequencyat low frequency due due totoa bad sound card featuringa bad sound card featuring frequency-frequency- dependent

dependent latencylatency

ThisThis artifactartifact can becan be correctedcorrected ifif the frequencythe frequency-dependent-dependent latencylatencyremainsremains the samethe same, , byby creatingcreatinga suitablea suitable inverse filterinverse filter withwiththe the KirkebyKirkebymethodmethod

Kirkeby

Kirkeby inverse inverse filter filter

ƒƒ The KirkebyThe Kirkeby inverse filterinverse filter isis computedcomputedinvertinginverting the measuredthe measured IRIR

( ) [ ( ) ]

[

( )

]

( ) ( )

Conj

f H f Conj

C = ⋅ +ε

1) The IR to be inverted is FFT transformed to frequency domain:

H(f) = FFT [h(f)]

2) The computation of the inverse filter is done in frequency domain:

Where ε(f) is a small, frequency-dependent regularization parameter

3) Finally, an IFFT brings back the inverse filter to time domain:

c(t) = IFFT [C(f)]

ε_est

ε_int

f_low f_high

Δf Δf

Inverse Inverse filterfilter Frequency

Frequency--dependentdependent regularizationregularizationparameterparameter

Pre-ringing at high and low frequency

ƒƒ ConvolvingConvolvingthe timethe time-smeared-smeared IR withIR with the Kirkebythe Kirkeby compactingcompacting filter, a filter, a veryvery sharp

sharp IR isIR is obtainedobtained

The sameThe same methodmethod can alsocan also bebeappliedapplied forfor correctingcorrectingthe responsethe response ofof the the loudspeaker

loudspeaker/microphone/microphone system, ifsystem, if anananechoicanechoic preliminarypreliminary test istest is donedone

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

Equalization

Equalization of of the the whole whole system system

ƒƒ An anechoicAn anechoic measurementmeasurement isis first performedfirst performed

Equalization

Equalization of of the the whole whole system system

ƒƒ A suitableA suitable inverse filterinverse filter isis generatedgenerated withwith the Kirkebythe Kirkeby methodmethodbyby invertinginverting the anechoicthe anechoic measurementmeasurement

Equalization

Equalization of of the the whole whole system system

ƒƒ The inverse filterThe inverse filter can becan be eithereither pre-pre-convolvedconvolved withwith the test signalthe test signal or postor post-- convolved

convolved withwith the resultthe result ofof the measurementthe measurement

ƒƒ Pre-Pre-convolutionconvolution usuallyusually reducesreduces the SPL the SPL beingbeing generatedgenerated bybythe the loudspeaker

loudspeaker, , resultingresultingin worstin worst S/N ratioS/N ratio

ƒƒ On the otherOn the other hand, hand, postpost--convolutionconvolution can makecan make the background noisethe background noise toto become

become “coloured“coloured”, and ”, and hencehence more more perciptibleperciptible

ƒƒ The resultingThe resulting anechoicanechoicIR becomesIR becomes almostalmost perfectlyperfectly a Diraca Dirac’’s Delta s Delta function

function::

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

Sensitivity to abrupt pulsive noises

ƒƒ OftenOften a pulsivea pulsive noisenoise occursoccursduringduringa sinea sine sweepsweepmeasurementmeasurement

Sensitivity to abrupt pulsive noises

ƒƒ AfterAfter deconvolution, the deconvolution, the pulsivepulsive sound causessound causes untolerableuntolerable artifactsartifacts in the in the impulse

impulse responseresponse

The artifactThe artifact appearsappears asas a downa down--slopingsloping sweepsweep on the impulseon the impulse response. response. At the 2 kHz

At the 2 kHz octaveoctave band the band the decaydecay isis distorteddistorted, and the reverb, and the reverb. . timetime isis artificiallyartificially increasedincreased fromfrom 2.13 to2.13 to 2.48 s2.48 s

Sensitivity to abrupt pulsive noises

ƒƒ SeveralSeveraldenoisingdenoisingtechniquestechniquescan becan be employed:employed:

► Brutely silencing the transient noise

► Employing the specific “click-pop eliminator” plugin of Adobe Audition

► Applying a narrow-passband filter around the frequency which was being generated in the moment in which the pulsive noise occurred

ƒƒ The thirdThe thirdapproachapproachprovidesprovidesthe betterthe betterresults:results:

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

Clock mismatch

ƒƒ WhenWhen the measurementthe measurement isisperformedperformed employingemploying devicesdeviceswhichwhich exhibitexhibit signifcant

signifcant clock mismatchclock mismatch betweenbetween playback and recordingplayback and recording, the , the resultingresulting impulse

impulse responseresponseisis “skewed“skewed”” (stretched(stretched in timein time):):

The picturesThe pictures show the show the resultsresults ofof anan electricalelectrical measuremntmeasuremnt performedperformed connecting

connecting directlydirectly a CD-a CD-player player withwith a DAT a DAT recorderrecorder

Clock

Clock mismatch mismatch

ƒƒ ItIt isispossiblepossible totore-re-pack the pack the impulseimpulse responseresponseemployingemploying the alreadythe already-- described

described approachapproach basedbased on the usageon the usage ofof a Kirkebya Kirkeby inverse filterinverse filter::

However

However, , thisthis isis possiblepossible onlyonly ifif a “a “referencereference”” electricalelectrical (or anechoic(or anechoic) ) measurement

measurement hashas beenbeen performedperformed. But. But, in , in manymany casescases, one, one onlyonly getsgets the rethe re--recordedrecorded signalssignals, and no reference, and no reference measurementmeasurement isis available, so available, so

the Kirkebythe Kirkeby inverse filterinverse filter cannotcannot bebe computed.computed.

Clock

Clock mismatch mismatch

ƒƒ However, However, ititisis alwaysalwayspossiblepossible totogenerate a pregenerate a pre--stretchedstretched inverse filterinverse filter, , which

which isis longerlonger or shorteror shorter thanthanthe “the “theoreticaltheoretical””oneone --byby properproper selectionselection ofof the lenghtthe lenght ofofthe inverse filterthe inverse filter, , ititisis possiblepossible toto deconvolvedeconvolveimpulseimpulse responses

responses whichwhich are almostare almost perfectlyperfectly “unskewed“unskewed””::

The picturesThe pictures show the resultshow the result ofof the deconvolvutionthe deconvolvution ofof a clocka clock-- mismatched

mismatched measurementmeasurement, in which, in which a prea pre--strecthedstrecthed inverse filterinverse filter isis employed

employed, 8.5 ms longer, 8.5 ms longer thanthan the theoreticalthe theoretical one.one.

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

High High - - frequency frequency cancellation cancellation due due to to averaging averaging

ƒƒ WhenWhen severalseveral impulseimpulseresponseresponsemeasurementsmeasurements are synchronouslyare synchronously-- averaged

averaged forfor improvingimproving the S/N ratiothe S/N ratio, the late part , the late part ofof the tailthe tail cancelscancelsout, out, particularly

particularly at high frequencyat high frequency, due , due toto slightslight timetimevariancevariance ofofthe systemthe system

Comparison

Comparison ofof a single sweepa single sweep 50 s long with50 s long with the synchronousthe synchronous averageaverage ofof 50 sweeps50 sweeps, 1 s long , 1 s long eacheach..

Spectrum of a single sweep of 50s (above) versus 50 sweeps of 1s (below)

short-FFT spectrum at 200 ms after direct sound

4 kHz 4 kHz

4 kHz 4 kHz

High High - - frequency frequency cancellation cancellation due due to to averaging averaging

ƒƒ However, However, ififaveragaingaveragaing isisperformedperformed properlyproperly in spectralin spectral domain, and a domain, and a single

single conversionconversion tototimetime domain domain isis performedperformed afterafter averaging, averaging, thisthisartifactartifact isis significantlysignificantly reducedreduced

ƒƒ The newThe new “cross “cross FunctionsFunctions”” pluginplugin can becan be usedused forfor computingcomputingH1:H1:

Result

Result ofoftransfer functiontransfer function H1, processing a H1, processing a sequence

sequence ofof 50 sine50 sine sweepssweeps(above(above)) 4 kHz 4 kHz

( )

LL 1 LR

G f G

H =

ƒ Basics of Exponential Sine Sweep (ESS) method

ƒ Pre-ringing at high and low frequency before the arrival of the direct sound pulse

ƒ Pre/post equalization of the test signal performed in a way which avoids time-smearing of the impulse

response

ƒ Sensitivity to abrupt pulsive noises during the measurement

ƒ Skewing of the measured impulse response when the playback and recording digital clocks are mismatched

ƒ Cancellation of high frequencies in the late part of the tail when performing synchronous averaging

ƒ Directivity of transducers (loudspeakers and microphones)

Topics

Directivity

Directivity of of transducers transducers

ƒ ƒ Analysis Analysis of of performances performances of of binaural binaural dummy

dummy heads heads

ƒ ƒ Analysis Analysis of of performances performances of of omni omni / / figure

figure - - of of - - 8 8 microphone microphone assemblies assemblies

ƒ ƒ Polar Polar patterns patterns of of dodechaedron dodechaedron loudspeakers

loudspeakers

Spatial

Spatial analysis analysis by by directive directive microphones microphones

ƒƒ The initialThe initial approachapproach waswas totouseuse directivedirective microphonesmicrophonesforfor gatheringgathering some some information

information aboutabout the spatialthe spatial propertiesproperties ofofthe sound the sound fieldfield“as“as perceivedperceived byby the listenerthe listener””

ƒƒ TwoTwo apparentlyapparentlydifferentdifferent approachesapproachesemerged: emerged: binauralbinaural dummydummy headsheads and and pressure

pressure--velocityvelocity microphones:microphones:

Binaural Binaural microphone

microphone (left(left))

and and

variable

variable--directivitydirectivity microphone

microphone (right)(right)

“ “ objective objective ” ” spatial spatial parameters parameters

ƒƒ ItIt waswas attemptedattemptedtoto “quantify“quantify”” the “the “spatialityspatiality”” ofofa rooma room bybymeansmeans ofof

“objective“objective”” parameters, parameters, basedbased on 2-on 2-channels channels impulseimpulse responsesresponses measuredmeasured withwith directivedirective microphonesmicrophones

ƒƒ The mostThe most famousfamous “spatial“spatial”” parameterparameter isis IACC (Inter AuralIACC (Inter Aural Cross Cross Correlation

Correlation), ), basedbased on binauralon binaural IR measurementsIR measurements

LeftLeft

Right Right

80 ms 80 ms

( )

( ) ( )

( ) ( )

pp_LL(τ(τ))

pp_RR((ττ))

[ ] ( )

[

]

Max IACC

d t p

d p

d t p

p

t _E

ms 80

0 2R ms

80

0 2L ms 80

0

R L

+

−

∈ ρ

= τ

⋅ + τ

⋅ τ

⋅ τ

τ

⋅ + τ

⋅ τ

= ρ

∫

∫

∫

“ “ objective objective ” ” spatial spatial parameters parameters

ƒƒ OtherOther “spatial“spatial”” parametersparameters are the Lateralare the Lateral Energy ratiosEnergy ratios: LE, LF, LFC: LE, LF, LFC

ƒƒ TheseThese are definedare defined fromfrom a 2-a 2-channels channels impulseimpulse response, the first response, the first channelchannelisis a a standard

standard omniomni microphonemicrophone, the second, the second channelchannel isisa a “figure“figure--ofof--eighteight”” microphone

microphone::

( ) ( )

∫ ( )

∫

τ

⋅ τ

τ

⋅ τ

⋅ τ

= 80ms

ms 0

o2 ms 80

ms 5

o 8

d h

d h

h LFC

Figure Figure

of of 88 OmniOmni

( )

∫ ( )

∫

τ

⋅ τ

τ

⋅ τ

= 80ms

ms 0

o2 ms 80

ms 5

82

d h

d h

LF

hh_oo(τ(τ))

hh₈₈((ττ))

( )

∫ ( )

∫

τ

⋅ τ

τ

⋅ τ

= 80ms

ms 0

o2 ms 80

ms 25

82

d h

d h

LE

Robustness

Robustness of of spatial spatial parameters parameters

ƒƒ BothBothIACC and LF dependIACC and LF depend stronglystronglyon the orientationon the orientation ofofthe microphonesthe microphones

ƒƒ BinauralBinaural and pressureand pressure--velocityvelocity measurementsmeasurements werewereperformedperformed in 2 in 2 theatres

theatres employingemploying a rotatinga rotating tabletable forfor turningturningthe microphonesthe microphones

(1-LF) Auditorium Parma – Sorgente a sx

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

0 10

20 30

40 50

60 70

80 90 100 110 120 130 140 150 170 160 190 180

200 210 220 230 240 250 260 270 280 290

300 310

320 330

340 350 Sorgente

(1-LF) Auditorium Roma (Sala 1200) – Sorgente a sx

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

0 10

20 30

40 50

60 70

80 90 100 110 120 130 140 150 170 160 190 180

200 210 220 230 240 250 260 270 280 290

300 310

320 330

340 350 Sorgente

Theatre 1-LF IACC Parma 0.725 0.266 Roma 0.676 0.344

IACC Auditorium Parma - Sorgente a sx

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0

10 20 30

40 50

60 70

80 90 100 110 120 130 140 150 170 160 190 180

200 210 220 230 240 250 260 270 280 290

300 310

320

330 340 350 Sorgente

IACC Auditorium Roma (Sala 1200) - Sorgente a sx

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0

10 20 30

40 50

60 70

80 90 100 110 120 130 140 150 170 160 190 180

200 210 220 230 240 250 260 270 280 290

300 310

320

330 340 350 Sorgente

Are Are binaural binaural measurents measurents reproducible reproducible ? ?

ƒƒ ExperimentExperiment performedperformedin anechoicin anechoic roomroom --samesame loudspeakerloudspeaker, same, same source source and receiverand receiver positions, 5 positions, 5 binauralbinaural dummydummy headsheads

Are Are binaural binaural measurents measurents reproducible reproducible ? ?

ƒƒ 90°90° incidenceincidence --at low frequencyat low frequency IACC isIACC is almostalmost 1, at high frequency1, at high frequency the the difference

difference betweenbetweenthe headsthe heads becomesbecomes evidentevident

IACCe - 90° incidence

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

31.5 63 125 250 500 1000 2000 4000 8000 16000

Frequency (Hz)

IACCe

B&K4100 Cortex Head Neumann

Are Are binaural binaural measurents measurents reproducible reproducible ? ?

ƒƒ Diffuse fieldDiffuse field --the differencethe difference betweenbetween the headsthe heads isis nownowdramaticdramatic

IACCe - random incidence

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

31.5 63 125 250 500 1000 2000 4000 8000 16000

Frequency (Hz)

IACCe

B&K4100 Cortex Head Neumann

Are LF

Are LF measurents measurents reproducible reproducible ? ?

ƒƒ ExperimentExperiment performedperformedin the Auditorium ofin the Auditorium of Parma -Parma - samesame loudspeaker, loudspeaker, samesame source and receiversource and receiver positions, 5 positions, 5 pressure-pressure-velocityvelocity microphonesmicrophones

Are LF

Are LF measurents measurents reproducible reproducible ? ?

ƒƒ At 7.5 m distanceAt 7.5 m distance, the results, the results alreadyalready exhibitexhibit significantsignificant scatterscatter

Comparison LF - measure 1 - 7.5m distance

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

31.5 63 125 250 500 1000 2000 4000 8000 16000

Frequency (Hz)

LF

Schoeps Neumann Soundfield B&K

Are LF

Are LF measurents measurents reproducible reproducible ? ?

ƒƒ At 25 m distanceAt 25 m distance, the , the scatterscatter isis eveneven larger....larger....

Comparison LF - measure 2 - 25m distance

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

31.5 63 125 250 500 1000 2000 4000 8000 16000

Frequency (Hz)

LF

Schoeps Neumann Soundfield B&K