1
• A method to analyze the ability of ISF to obtain an arbitrary, user-defined, desired spatial response at the output of the filter was presented. • With perfect knowledge of the source DOAs the desired spatial response can be obtained at the output of the ISF. • Through the analysis it was shown that robustness against DOA estimation errors is important and essential for the ISF framework. 5. Simulation Experiment 6. Conclusions • ULA with M=5 • Source B placed at I = 37 discrete points • Anechoic environment • Source-microphone distance: 1.8 m Obtained vs Desired : Known DOAs Obtained vs Desired : DOA errors + Different passband widths SDI Obtained vs Desired Array Gain: Source B • DOA error variance : 15 O • Passband widths: 20 O and 30 O • DOA error variance: 5 O and 10 O • Passband widths: 20 O and 30 O ˆ A (n, k )= A +A (n, k ), ˆ B,i (n, k )= B,i +B,i (n, k ) DOA estimates with error σ 2 DOA =E{2 } Simulation Setup Error variance [1] O. Thiergart, M. Taseska, and E.A.P. Habets, “An informed parametric spatial filter based on instantaneous direction-of-arrival estimates,” IEEE Trans Acoust., Speech, Signal Process., vol. 22, no. 12, pp.2182–2196, Dec 2014. Source A Microphone Array Source B positions I =9 4. Proposed Analysis Method Average Directional Array Gain A d,i = 1 card(T i ) X (n,k)2T i G A (n, k ) G B,i (n, k ) G A (n, k )= | ˜ X A (n, k )| 2 |X A (n, k, d 1 )| 2 Proposed method can also be used for L =1 or L > 2. By varying the input signal-to-interference ratio (iSIR) the proposed method can be used to identify the critical iSIR where the single plane wave signal model is violated. G B,i (n, k )= | ˜ X B,i (n, k )| 2 |X B,i (n, k, d 1 )| 2 • Assumption: L = 2 • Sample complete DOA range at I discrete points. • Source A is kept static at broadside • Source B is moved through the whole DOA range, placing it at the Idiscrete points. Setup T i = {(n, k ): |X A (n, k, d 1 )|≥ A ^|X B,i (n, k, d 1 )|≥ B } 0 30 60 90 120 150 180 -40 -30 -20 -10 0 Magnitude [dB] DOA θ [ ] • The spatial response is an arbitrary, user-defined function that can be potentially complex valued and frequency dependent. • The design of the desired spatial response function is also dependent on the application. 3. Desired Spatial Response 2. Informed Spatial Filter: Review • Consider a uniform linear array (ULA) of M microphones located at . The microphone signals are given by • Assuming signal components are mutually uncorrelated • Desired signal • Estimate of the desired signal • Microphone signal of l-th plane wave • Sound pressure of l-th plane wave • m-th element of steering vector • Noise PSD matrix • Direction dependent gain • Filter weights • Propagation vectors of L sources Definitions • Formulation: s.t. • Solution: with • The direction dependent gain corresponds to the value of an arbitrary spatial response function g(✓,k ) , evaluated at the DOA of the l-th plane wave. Informed LCMV filter d 1...M y(n, k )= L X l=1 x l (n, k ) | {z } x(n,k) +x n (n, k ) y (n, k )=E{y(n, k )y H (n, k )} = x (n, k )+ n (n, k ) Z (n, k )= L X l=1 G(l ,k )X l (n, k, d 1 ) x l (n, k )=[X l (n, k, d 1 ) ...X l (n, k, d M )] T x l (n, k )= a(l ,k )X l (n, k, d 1 ) a m (l ,k ) = exp{jr m cos l (n, k )} n (n, k )= n (k ) I G(l ,k ) w (n, k ) A =[a(1 ,k ),..., a(L ,k )] w (n, k ) = arg min w w H n (n, k )w w H (n, k )a(l ,k )= G(l ,k ) w (n, k )= 1 n A A H 1 n A 1 g G(l ,k ) g =[G(1 ,k ),...,G(L ,k )] H b Z (n, k )= w H (n, k )y(n, k ) 1. Introduction • Propose a method to analyse the obtained spatial response and compare it to the desired response. • Gain insight into the obtained spatial response at the output of the spatial filter. • Analyse the effect of DOA estimation errors on the obtained spatial response, and provide an objective motivation for the need of robustness against DOA estimation errors. Aim A Method to Analyze the Spatial Response of Informed Spatial Filters Soumitro Chakrabarty, Oliver Thiergart, Emanuël. A. P. Habets International Audio Laboratories Erlangen E-Mail: [email protected] INTERNATIONAL AUDIO LABORATORIES ERLANGEN A joint institution of Fraunhofer IIS and Universität Erlangen-Nürnberg • Informed spatial filters (ISFs) [1] are designed to capture multiple sound sources with a desired, arbitrary spatial response at each TF instant while attenuating undesired signal components. • The desired spatial response function is different from the directivity pattern of the spatial filter. • A comparative analysis of the obtained and the desired spatial response has not been performed yet. Motivation

INTERNATIONAL AUDIO LABORATORIES ERLANGEN · soumitro. 5. Simulation Experiment 6. Conclusions • ULA with M=5 • Source B placed at I = 37 discrete points • Anechoic environment

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: INTERNATIONAL AUDIO LABORATORIES ERLANGEN · soumitro. 5. Simulation Experiment 6. Conclusions • ULA with M=5 • Source B placed at I = 37 discrete points • Anechoic environment

•AmethodtoanalyzetheabilityofISFtoobtainanarbitrary,user-defined,desiredspatialresponseattheoutputofthefilterwaspresented.•WithperfectknowledgeofthesourceDOAsthedesiredspatialresponsecanbeobtainedattheoutputoftheISF.•ThroughtheanalysisitwasshownthatrobustnessagainstDOAestimationerrorsisimportantandessentialfortheISFframework.

E-Mail:

soumitro.

E-Mail:

soumitro.

5. Simulation Experiment

6. Conclusions

•ULAwithM=5

•SourceBplacedatI=37discretepoints

•Anechoicenvironment

•Source-microphonedistance:1.8m

Obtained vs Desired : Known DOAs

Obtained vs Desired : DOA errors + Different passband widths

SDI Obtained vs Desired Array Gain: Source B

•DOAerrorvariance:15O

•Passbandwidths:20Oand30O

•DOAerrorvariance:5Oand10O

•Passbandwidths:20Oand30O

✓̂A(n, k) = ✓A +✓A(n, k),

✓̂B,i(n, k) = ✓B,i +✓B,i(n, k)

DOA estimates with error

σ2DOA = E{✓2}

Simulation Setup

Error variance

[1]O.Thiergart,M.Taseska,andE.A.P.Habets,“Aninformedparametricspatialfilterbasedoninstantaneousdirection-of-arrivalestimates,”IEEETransAcoust.,Speech,SignalProcess.,vol.22,no.12,pp.2182–2196,Dec2014.

Source A

Microphone Array

Source B

positionsI= 9

4. Proposed Analysis Method

Average Directional Array Gain

Ad,i =1

card(Ti)X

(n,k)2Ti

GA(n, k)

GB,i(n, k)

GA(n, k) =|X̃A(n, k)|2

|XA(n, k,d1)|2

•Proposed method can also be used for L =1 or L > 2.

•By varying the input signal-to-interference ratio (iSIR) the proposed method can be used to identify the critical iSIR where the single plane wave signal model is violated.

GB,i(n, k) =|X̃B,i(n, k)|2

|XB,i(n, k,d1)|2

•Assumption:L=2

•SamplecompleteDOArangeatIdiscretepoints.

•SourceAiskeptstaticatbroadside

•SourceBismovedthroughthewholeDOArange,placingitattheIdiscretepoints.

Setup

Ti = {(n, k) : |XA(n, k,d1)| ≥ ✏A ^ |XB,i(n, k,d1)| ≥ ✏B}

0 30 60 90 120 150 180

−40

−30

−20

−10

0

Magnitude[dB]

DOA θ[◦]

•Thespatialresponseisanarbitrary,user-definedfunctionthatcanbepotentiallycomplexvaluedandfrequencydependent.

•Thedesignofthedesiredspatialresponsefunctionisalsodependentontheapplication.

3. Desired Spatial Response

2. Informed Spatial Filter: Review

•Considerauniformlineararray(ULA)ofMmicrophoneslocatedat.Themicrophonesignalsaregivenby

•Assumingsignalcomponentsaremutuallyuncorrelated

•Desiredsignal

•Estimateofthedesiredsignal

•Microphonesignalofl-thplanewave• Soundpressureofl-thplanewave

•m-thelementofsteeringvector

•NoisePSDmatrix

•Directiondependentgain

•Filterweights

•PropagationvectorsofLsources

Definitions

•Formulation:s.t.

•Solution:with

•Thedirectiondependentgaincorrespondstothevalueofanarbitraryspatialresponsefunctiong(✓, k),evaluatedattheDOAofthel-thplanewave.

Informed LCMV filter

d1...M

y(n, k) =

LX

l=1

xl(n, k)

| {z }x(n,k)

+xn(n, k)

y(n, k) = E{y(n, k)yH(n, k)}= x(n, k) +n(n, k)

Z(n, k) =

LX

l=1

G(✓l, k)Xl(n, k,d1)

xl(n, k) = [Xl(n, k,d1) . . . Xl(n, k,dM )]T

xl(n, k) = a(✓l, k)Xl(n, k,d1)

am(✓l, k) = exp{jrm cos ✓l(n, k)}

n(n, k) = n(k) I

G(✓l, k)

w(n, k)

A = [a(✓1, k), . . . ,a(✓L, k)]

w(n, k) = arg minw

wHn(n, k)w wH(n, k)a(✓l, k) = G(✓l, k)

w(n, k) = 1n A

⇥AH1

n A⇤1

g

G(✓l, k)

g = [G(✓1, k), . . . , G(✓L, k)]H

bZ(n, k) = wH(n, k)y(n, k)

1. Introduction

•Proposeamethodtoanalysetheobtainedspatialresponseandcompareittothedesiredresponse.•Gaininsightintotheobtainedspatialresponseattheoutputofthespatialfilter.•AnalysetheeffectofDOAestimationerrorsontheobtainedspatialresponse,andprovideanobjectivemotivationfortheneedofrobustnessagainstDOAestimationerrors.

Aim

A Method to Analyze the Spatial Response of Informed Spatial Filters

Soumitro Chakrabarty, Oliver Thiergart, Emanuël. A. P. Habets

International Audio Laboratories Erlangen

E-Mail: [email protected]

INTERNATIONAL AUDIO LABORATORIES ERLANGENA joint institution of Fraunhofer IIS and Universität Erlangen-Nürnberg

•Informedspatialfilters(ISFs)[1]aredesignedtocapturemultiplesoundsourceswithadesired,arbitraryspatialresponseateachTFinstantwhileattenuatingundesiredsignalcomponents.•Thedesiredspatialresponsefunctionisdifferentfromthedirectivitypatternofthespatialfilter.

•Acomparativeanalysisoftheobtainedandthedesiredspatialresponsehasnotbeenperformedyet.

Motivation