2008 adaptive wave field synthesis for active sound field reproduction: experimental results

Upload: philippe-aubert-gauthier

Post on 03-Jun-2018

220 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    1/12

    Adaptive wave field synthesis for active sound field reproduction:Experimental results

    Philippe-Aubert Gauthiera and Alain BerryGroupe dAcoustique de lUniversit de Sherbrooke, Universit de Sherbrooke, 2500 boul. de lUniversit,

    Sherbrooke, Qubec J1K 2R1, Canada

    Received 31 May 2007; revised 28 January 2008; accepted 28 January 2008

    Sound field reproduction has applications in music reproduction, spatial audio, sound environment

    reproduction, and experimental acoustics. Sound field reproduction can be used to artificially

    reproduce the spatial character of natural hearing. The objective is then to reproduce a sound field

    in a real reproduction environment. Wave field synthesis WFS is a known open-loop technology

    which assumes that the reproduction environment is anechoic. The room response thus reduces the

    quality of the physical sound field reproduction by WFS. In recent research papers, adaptive wave

    field synthesis AWFS was defined as a potential solution to compensate for these quality

    reductions from which WFS objective performance suffers. In this paper, AWFS is experimentally

    investigated as an active sound field reproduction system with a limited number of reproduction

    error sensors to compensate for the response of the listening environment. Two digital signal

    processing algorithms for AWFS are used for comparison purposes, one of which is based on

    independent radiation mode control. AWFS performed propagating sound field reproduction better

    than WFS in three tested reproduction spaces hemianechoic chamber, standard laboratory space,

    and reverberation chamber. 2008 Acoustical Society of America. DOI: 10.1121/1.2875844

    PACS numbers: 43.38.Md, 43.60.Tj, 43.50.Ki AJZ Pages: 19912002

    I. INTRODUCTION

    With the constantly evolving digital signal processing

    and the relatively recent advent of multichannel audio, spa-

    tial audio has gained more attention in the past decades from

    researchers and practitioners for applications such as high-

    fidelity sound reproduction, music reproduction, virtual real-

    ity display, interactive multisensory environments, auraliza-

    tion, and sound installations Camurri and Ferrentino, 1999;

    Epain et al., 2004;AES Staff Writer, 2005;Woszczyket al.,2005;Keller et al., 2006; Blesser and Salter, 2007. The in-

    terest in immersion and convincing multisensory environ-

    ments is not new Grau, 2003 and various techniques for

    spatial audio have been introduced in the past Kendall,

    1995;Verheijen, 1997; Poletti, 2000; Rumsey, 2001; Davis,

    2003.

    Within spatial sound, sound field reproduction methods

    attempt to reproduce physical stimulus wave field, thereby

    avoiding any perceptual considerations in the implementa-

    tion. Sound field reproduction was investigated by research-

    ers in the past decades Berkhoutet al., 1993;Nelson et al.,

    1997; Verheijen, 1997; Poletti, 2000; Choi and Kim, 2004;

    Epain et al., 2004; Takane and Sone, 2004; Keller et al.,

    2006. One of the most active and recent related matters is

    room compensation Spors et al., 2003; Gauthier et al.,

    2005a; Betlehem and Abhayapala, 2005; Spors et al., 2005;

    Fuster et al., 2005; Gauthier and Berry, 2006, which is es-

    sential for sound field reproduction in a real reproduction

    space on the basis of objective, physically measurable, per-

    formances. This is especially true when acoustical treatment

    of the reproduction space is not possible, like for sound field

    reproduction in vehicle mock-ups where the visual reproduc-

    tion of the original space is important.

    This paper deals with the problem of sound pressure

    field reproduction using adaptive digital signal processing

    applied to adaptive wave field synthesis AWFS originally

    introduced by Gauthier et al. 2005b. More specifically it

    validates by experiments the AWFS concept.

    The concepts and results shown in this paper are not

    limited to audio applications. Indeed, sound field reproduc-tion may also be applied to experimental acoustics Veit and

    Sander, 1987; Bravo and Elliott, 2004, psychoacoustics

    Epain et al., 2004;Keller et al., 2006, and sound environ-

    ment reproduction or even used as a vibroacoustics design

    tool. These are promising applications of sound field repro-

    duction.

    This paper is divided in four parts. In Sec. I, sound field

    reproduction, WFS, and AWFS are described. The complete

    experimental procedures and setups are described in Sec. II.

    Results of experiments with AWFS are then reported in Sec.

    III for three different reproduction spaces. Section IV dis-

    cusses the results and exposes our conclusions.

    A. Sound field reproduction

    The main objective of sound field reproduction can be

    generally stated as the aim to recreate a given acoustical

    property of the sound field, such as sound pressure, sound

    intensityChoi and Kim, 2004; Merimaa and Pulkki, 2005,

    spatial diffusenessMerimaa and Pulkki, 2005, etc., over an

    extended region of space. This can be achieved using a re-

    production system including electroacoustical sources and

    receivers, signal processing, and the desired physical targetaElectronic mail: [email protected]

    J. Acoust. Soc. Am. 123 4, April 2008 2008 Acoustical Society of America 19910001-4966/2008/1234/1991/12/$23.00

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    2/12

    description. In several of these cases, use of adaptive filtering

    implies the minimization of a cost function which is repre-

    sentative of this reproduction objective Gauthier et al.,

    2005a; Gauthier and Berry, 2006. Adaptive signal process-

    ing for spatial sound reproduction has been considered invarious forms by researchers Asano and Swanson, 1995;

    Takane et al., 1999; Radlovi et al., 2000; Santilla, 2001;

    Epain et al., 2004; Choi and Kim, 2004; Gauthier et al.,

    2005b;Spors et al., 2005;Gauthier and Berry, 2006.

    B. Wave field synthesis

    WFS research started with the theoretical propositions

    by BerkhoutBerkhoutet al., 1993;Verheijen, 1997;Startet

    al., 1999. From the simple source formulation of the

    KirchhoffHelmholtz integral theorem Williams, 1999,

    WFS operators are designed to link a given simple virtual

    sourcetypically creating spherical or plane wave in an hori-zontal plane, the listening plane, fed by a monophonic sig-

    nal, to a loudspeaker array which reproduces the acoustic

    field of the virtual source that is the target, or virtual, sound

    field. A schematic and simplified representation of the prob-

    lem is shown in Fig. 1. The problem is usually limited to

    reproduction in the horizontal plane with a finite number of

    discrete reproduction sources using appropriate simplifica-

    tions of the integral formulation Verheijen, 1997. WFS

    studies have investigated: spatial aliasing de Vries et al.,

    1994;Startet al., 1995;Spors and Rabenstein, 2006;Corteel,

    2006a, objective performance, room effect, Klehs and

    Sporer, 2003;Sporer and Klehs, 2004, acoustic room com-

    pensationSporset al., 2003,2005;Fusteret al., 2005, WFS

    equalizationCorteel, 2006b, and more.

    On one hand, the benefit of current WFS prototypes is

    their effectiveness in transmitting a spatial impression in

    terms of sound localizationover a broad area surrounded by

    loudspeakers. On the other hand, WFS drawbacks are related

    to the definition of the synthesis operators: The reproduction

    room response or electroacoustical system limits Corteel,

    2006b are not considered in the definition of WFS. The

    typical WFS system is consequently based on an open-loop

    architecture assuming a free field as the reproduction space.

    Active room compensation or system equalization for WFS

    is an active research topic for objective sound field reproduc-

    tion in real room Elliott and Nelson, 1989; Asano and

    Swanson, 1995; Bouchard and Quednau, 2000; Santilla

    2001;Sporset al., 2003;Gauthier et al., 2005b;Spors et al.,

    2005;Fuster et al., 2005;Corteel, 2006b.

    To lighten the present paper, which focuses on experi-

    mental results, the readers are referred to Verheijen 1997

    for a more detailed review of WFS. A complete description

    of WFS adapted to the specific problem of AWFS was pub-

    lished by Gauthier and Berry 2006.

    C. Adaptive wave field synthesis and independentradiation mode control

    In a recent paper Gauthier and Berry, 2006, AWFS was

    suggested as a practical compromise between WFS and ac-

    tive room compensation that usually requires a large amount

    of sensors. AWFS is based on a cost function to be mini-

    mized. Although implemented here in a specific configura-

    tion, the AWFS concept described by this cost function can

    readily be applied to any configuration. The cost function is

    a quadratic function of: 1 the reproduction errors and 2

    the adaptive filters departure from the WFS solution ex-pressed as a set of finite impulse response FIR filter coef-

    ficients. The penalization of the departure from the WFS

    filters is what makes AWFS original in comparison with

    other research done on sound field reproduction using active

    noise control techniques. The interest of such an approach

    stems from a simple observation: The direct sound field re-

    produced by WFS approaches the virtual sound field

    Gauthier and Berry, 2007 and then allows for proper sound

    localization on the basis of precedence effect Blauert,

    1999. Accordingly, WFS can be taken as a starting point or

    an a priori solution for the adaptive algorithm which will

    minimize the reproduction errors caused by the room re-

    sponse. Moreover, the weighted penalization of any depar-ture from the WFS solution may prevent the degradation of

    sound localization since it limits the contribution of the sec-

    ondary sources which are normally not activated by the WFS

    solution, given the fact that the WFS solution already con-

    tains spatial information that cannot be completely measured

    using a limited number of error microphones. For example,

    when using more reproduction sources than error sensors, the

    WFS solution contributes to the proper reconstruction of the

    direct sound field outside the control region defined by the

    error sensor locations Gauthier and Berry, 2006. Finally,

    this penalization, which is controlled by a set of penalization

    parameters, can be used to control the balance between a

    purely WFS solution and a closed-loop realization of Am-

    bisonics sound field reproduction Gauthier and Berry,

    2006.

    This specific definition of a cost function for a multi-

    channel adaptive system leads to a simple modification of the

    leaky filtered-reference least-mean-square FXLMS algo-

    rithmElliott, 2001. The modification is the inclusion of the

    WFS solution in the adaptation rule. Here, we will refer to

    modified FXLMS for AWFS when this algorithm is used.

    Via the singular value decompositionSVDof the plant

    matrix frequency response functions FRFs between repro-

    FIG. 1. Term convention for WFS definition. The virtual source is located in

    x0. L is the reproduction source line, the virtual source is on the left of the

    source line and the reproduction space is on the right of the source line. All

    sources and sensors are located in the x1 x2 plane.

    1992 J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    3/12

    duction sources and error sensors involved in AWFS, it was

    shown that the underlying AWFS mechanism is the indepen-

    dent control of radiation modes Gauthier and Berry, 2006

    using plant decoupling. It is also related to the Principal-

    Components LMS PC-LMS algorithm Cabell and Fuller,

    1999. This suggested a practical implementation of AWFS

    signal processing which had already been textually described

    byGauthier and Berry2006.This reference on signal pro-

    cessing for AWFS should suffice for the purpose of this pa-

    per, see also Gauthier and Berry 2008. In the case of

    AWFS based on independent radiation mode control, a set of

    analysis filters is used to transform the sound pressure in the

    SVD basis and a set of synthesis filters is used to create the

    loudspeaker signals from the SVD basis. In this transformed

    domain a set of single-channel independent adaptive filters

    operate to control each radiation mode individually. This re-

    duces the computational burden and allows for a fine tuning

    of the convergence properties of the algorithm i.e., indepen-

    dent fine tuning of the radiation modes convergence proper-

    ties. Adaptive sound reproduction using plant decoupling

    via SVD was already proposed by Bai and Elliott 2004 via

    simulations for cross-talk cancellation. The main differences

    between Bais work and the present paper are 1 the experi-mental application to sound field reproduction, 2 further

    considerations for the proper construction of the synthesis

    and analysis filters from a signal processing perspective, and

    3 the inclusion of an a priori solution the WFS solution.

    The inclusion of the WFS solution in the cost function

    Gauthier and Berry, 2006 significantly changes the algo-

    rithm since the WFS solution must be projected on the SVD

    basis radiation mode synthesis filters and nullspace synthe-

    sis filters. The proper construction of the synthesis and

    analysis filters is also what make this paper on AWFS origi-

    nal. Moreover, this construction of the synthesis and analysis

    filters had proven to be of critical importance for the efficient

    projection of the WFS solution on the SVD basis. However,such detailed considerations for signal processing are beyond

    the scope of this paper.

    In this paper, the performance of AWFS based on modi-

    fied FXLMS and independent radiation mode control algo-

    rithms is derived from experiments with a real AWFS system

    in three different acoustical situations. This paper therefore

    validates the AWFS concepts, previously investigated in

    theory Gauthier and Berry, 2006. See Elliott 2001 for a

    general review on adaptive filtering for active noise control

    and sound field reproduction.

    II. EXPERIMENTAL PROTOCOL

    A. Experimental setups

    The tested system includes 24 reproduction sources and

    4 reproduction error sensors. The complete system is shown

    in Fig.2.The reproduction sources create a 2-m-diam circu-

    lar array in the horizontal plane. Sources are separated by

    26 cm, thus giving an approximate minimal spatial aliasing

    frequency of 634 Hz /2=26 cm, where is the acoustical

    wavelength Spors and Rabenstein, 2006. This minimal

    spatial aliasing frequency gives the frequency range of focus

    for the experiments. Above the spatial aliasing frequency, the

    sound field reconstruction is impossible over the reproduc-

    tion region. The error microphones form a cross, in the same

    horizontal plane as the reproduction sources, and their sepa-

    ration distance along x1 and x2 axes is 17.5 cm. Sources and

    sensors stand 1.22 m above the floor.

    The loudspeakers are studio monitors amplified two-

    way cabinets. The error sensors are 1 /4 in. electret micro-

    phones. For the off-line broadband AWFS implementation,

    the loudspeakers and the microphones are connected to a

    signal conditioner and a computer using a sound card 24

    analog inputs and 24 analog outputs. In this setup, theAWFS signal processing operates off-line. The setup is sche-

    matically shown in Fig. 3a.For the second setup involving

    harmonic target wave fields, the loudspeakers and micro-

    phones are connected to reconstruction and antialiasing fil-

    ters 440 Hz low-pass, eighth order, Butterworth, respec-

    tively, before being connected to a digital signal processing

    station used for on-line harmonic AWFS. The station is built

    around a Texas Instrument TMS320C40 floating point digital

    12

    11

    13

    10

    14

    9

    15

    8

    16

    ... 8

    7

    17

    x2

    6

    18

    5

    19

    1 ...

    4

    20

    3

    21

    2

    22

    1

    23

    24

    2

    m

    x1

    Sources

    Error sensors

    Monitoring sensors

    Sources: 24

    Sensors: 4

    FIG. 2. Schematic AWFS setup made of 24 reproduction sources, 4 repro-

    duction error sensors, and 8 monitoring sensors. Typical virtual source

    position.

    FIG. 3. Schematic representation of the AWFS instrumentations,a Broad-

    band and b harmonic.

    J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis 1993

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    4/12

    signal processor. The setup is shown in Fig. 3b.The experi-

    mental methods differ for these two setups. The system is

    pictured in Fig.4.

    B. Methods of experiments

    Two types of reproduction methods and experiments are

    reported. As for the first type, the AWFS algorithms operate

    off-line. In this case, the experimental procedure is:1iden-

    tification of the plant impulse responses from all reproduc-

    tion sources to all error sensors. Sweep-sine identification

    was used to cover from 0 to 600 Hz. The number of coeffi-

    cients varies according to the reproduction environment. 2

    Off-line running, by simulation, of the AWFS algorithms tex-

    tually described by Gauthier and Berry 2006. 3 Render-

    ing, with the real electroacoustical system, of the AWFS so-

    lution after convergence of the control filters. 4

    Measurement of the reproduced sound fields. The reproduced

    sound field measurements are based on reproduced impulseresponses from the reference signal to the eight monitor

    microphones see Fig. 2. Broadband AWFS results shown

    in Sec. III include measurements of the reproduced sound

    fields using swept sines.

    As for the second type of reproduction methods and ex-

    periments, the objective is to evaluate the performance of

    AWFS with an on-line adaptation system. To reach this goal,

    several practical trade-offs are included to reduce the com-

    putational burden so that on-line adaptation can be per-

    formed using the available hardware. In the case of harmonic

    sound field reproduction, all the AWFS filters including

    adaptive filters, target operators, synthesis, and analysis fil-

    ters are implemented using two-coefficient FIR filters. The

    algorithm then approaches the PC-LMS algorithm Cabell

    and Fuller, 1999 with a supplementary penalization term

    and an a priori solution. In this very specific situation, the

    computational load is drastically reduced and on-line adap-

    tation is possible using the algorithms as textually described

    by Gauthier and Berry 2006.

    For both types of AWFS realization, the convergence

    coefficients used in the adaptive algorithms are set near the

    maximum values, which guaranteed stability and conver-

    gence of the adaptation.

    III. AWFS EXPERIMENTS

    A. Acoustical characteristics of the reproductionrooms

    The three reproduction environments were selected to

    cover a large spectrum of reverberation properties. These en-

    vironments are: 1 a hemianechoic chamber, 2 a standardlaboratory space, and 3 a reverberation chamber.

    The hemianechoic chamber Fig. 4 has a volume of

    125 m3 6.556.253.05 m with a floor surface of 41 m2.

    A typical frequency response function FRF transfer func-

    tion between a reproduction source and an error microphone

    is shown in Fig.5.In this situation, the FRFs are smooth and

    the dip around 310 Hz is created by the destructive interfer-

    ence with the floor reflection at this frequency. In the hemi-

    anechoic chamber, the error sensor signals are mostly domi-

    nated by the direct sound field of the reproduction sources.

    The volume of the standard laboratory space is 469 m3

    8.2314.024.06 m with a floor surface of 115 m2. The

    typical FRF shown in Fig. 5 is more complex and showsvarious dips. The reverberation radius the distance from the

    source at which the sound pressure level of the direct sound

    field is equal to the sound pressure level of the diffuse sound

    field was estimated to be more than 1.4 m with broadband

    noise audio bandwidth. The approximation of the rever-

    beration radius is derived from the spatial sound pressure

    level decay curve from an omnidirectional loudspeaker array.

    The mean curve was estimated from four measurement lines

    in the horizontal plane 1.22 m above the floor randomly

    selected in the room. From the mean curve, the sound source

    power level is computed from the measured direct sound

    field using a curve fitting with a theoretical free-field decay

    curve. The homogeneous reverberation level is evaluated

    from the last part of the decay curve. The approximation of

    the reverberation radius is then derived from the crossing of

    this homogeneous level and the theoretical free-field decay

    curve for the approximated sound source power level. Given

    a reverberation radius of more than 1.4 m in this laboratory

    space, the error sensors at 1 m of the reproduction sources

    are exposed to the direct sound field of the reproduction

    source and the field reflected by the room walls with a well-

    balanced proportion in comparison with the hemianechoic

    space.

    FIG. 4. Experimental AWFS setup in the hemianechoic chamber.

    0 50 95 160 220 310 400 440 500

    -60

    -50

    -40

    -30

    -20

    -10

    0

    10

    20

    Freq. [Hz]

    Gain

    (dB

    ref.

    1)

    Hemi-anechoic chamberReverberant chamberArbitrary laboratory space

    FIG. 5. Typical identified FRFs from a reproduction source to an error

    microphone in the hemianechoic chamber, standard laboratory space, and

    reverberation chamber.

    1994 J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    5/12

    The volume of the reverberation chamber is 142 m3

    with a floor surface of 46.5 m2 7.56.23.05 m. Several

    sheets of damping material were placed in corners of the

    room to reduce the excessively long reverberation time. The

    typical FRF shown in Fig.5has gains that vary strongly with

    frequency and the transitions from frequency to frequency

    are very sharp. The reverberation radius was estimated to be

    between 0.45 and 0.51 m for broadband signal. Therefore, in

    this highly reverberant space, the error sensors are mostly

    exposed to the diffuse sound field of the reproduction

    sources. This is an hostile environment both for WFS and

    AWFS.

    B. Hemianechoic space

    1. Broadband AWFS

    The broadband off-line demonstration of AWFS is inter-

    esting because it validates the complete AWFS concept, as

    described in previous papers Gauthier and Berry, 2006. The

    broadband nature of the off-line system places a general

    point of view on the results.

    In a preliminary stage, the system FRFs are identified

    using swept sines with an average over 200 realizations. Theresulting impulse responses include 256 coefficients. The

    sampling rate is 1200 Hz for all the broadband experiments.

    Using the identified plant, the modified FXLMS algorithm

    can readily be applied Elliott, 2001; Gauthier and Berry,

    2006.

    The independent radiation mode control implementation

    requires an additional initialization step. Singular value de-

    composition of the system plant in the frequency domain is

    achieved as textually described by Gauthier and Berry

    2006 for each frequency. This gives the radiation modes

    source modes, singular values, and pressure modes at each

    frequency. Radiation mode reordering and phase optimiza-

    tion algorithms are then applied in the frequency domain tosmooth the source and pressure mode phase responses. This

    creates a novelty in comparison with Bais work with broad-

    band plant decoupling Bai and Elliott, 2004. Inverse

    discrete-time Fourier transform is then applied to obtain the

    synthesis filters and analysis filters to move to and from the

    SVD basisin the time domain for AWFS based on indepen-

    dent radiation mode control. Note that using such synthesis

    and analysis filters, the plant is uncoupled but not whitened.

    The synthesis filters Gauthier and Berry, 2006 for the first

    four source modes are shown in Fig. 6. Each group of syn-

    thesis filtersthere are 4 groups of 24 filtersproduces one of

    the source modes at the reproduction source array. The

    analysis filters for the four pressure modes are shown in Fig.

    7. Each group of analysis filters there are 4 groups of 4

    filters transforms the physical acoustical pressures in the

    pressure mode basis SVD basis. Interestingly, the synthesis

    and analysis filters show a sharp concentrated time response:

    Time leakage is reduced in comparison with SVD filters pre-

    sented by Bai and Elliott 2004, thanks to the radiation

    modes reordering and to the phase optimization algorithms.

    Figure8 shows the reproduced impulse responses from

    the virtual source to the monitoring sensor array shown in

    Fig.2 for WFS. The transfer function units are 1 /m sound

    pressure Pa divided by virtual monopole amplitude

    Pa m. The virtual source is a spherical source located at

    xo = 0 , 4 , 0 m. Clearly, the direct field of the reproduced

    impulse responses approaches the target impulse responses.

    After the direct wave front passage, the reflection from the

    floor appears and instants later the low frequency echo of the

    50 100 150 200 250

    -0.2

    -0.1

    0

    mp.

    Source mode #1

    50 100 150 200 250

    -0.1

    0

    0.1

    mp.

    Source mode #2

    50 100 150 200 250

    -0.1

    0

    0.1

    mp.

    Source mode #3

    1 64 128 192 256

    -0.2

    -0.1

    0

    0.1

    mp.

    Source mode #4

    Samples

    FIG. 6. First synthesis filtersGauthier and Berry, 2006in the time domain

    for source modes 14 in the hemianechoic chamber. Each plot includes 24synthesis filters to create the given source mode with 24 reproduction

    sources. Each filter includes 256 coefficients.

    50 100 150 200 250

    -0.2

    0

    0.2

    mp.

    Pressure mode #1

    50 100 150 200 250

    -0.4

    -0.2

    0

    0.2

    0.4

    mp.

    Pressure mode #2

    50 100 150 200 250

    -0.2

    0

    0.2

    mp.

    Pressure mode #3

    1 64 128 192 256

    -0.2

    0

    0.2

    mp.

    Pressure mode #4

    Samples

    FIG. 7. Analysis filtersGauthier and Berry, 2006 in the time domain for

    pressure modes 14 in the hemianechoic chamber. Each plot includes 4

    analysis filters to catch the pressure mode with 4 pressure sensors. Each

    filter includes 256 coefficients.

    J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis 1995

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    6/12

    hemianechoic chamber impinges the monitoring array. This

    echo is caused by the chamber which is hemianechoic only

    above 150 Hz. Moreover, another physical imperfection of

    the WFS reproduced sound field appears: The direct wavefront has an undesirable coloration the impulse is spread

    over two or three samples, passing from positive to negative

    values, possibly caused by the loudspeaker response or the

    WFS approximations.

    The convergence coefficient was set to 0.00001 for the

    modified FXLMS algorithm. The convergence coefficient

    m, for the independent radiation mode control algorithm,

    were set to 1 =0.0001, 2 =3 =0.0004, and 4 =0.002 for

    the four radiation modes, where the subscript indicates the

    radiation mode number. The penalization parameter for the

    FXLMS algorithm was set to = 20 Gauthier and Berry,

    2006. The penalization parameters for the AWFS based on

    independent radiation mode control were 1 = 2, 2 =3 =0.2and 4 =0.1.

    The impulse responses reproduced by AWFS are shown

    in Fig.9a by FXLMS and Fig. 9b by independent radia-

    tion mode control. Clearly, the imperfections of WFS shown

    in Fig.8are partly corrected by AWFS even outside the error

    sensor array note that the two central monitors see Fig.2

    correspond to two of the error sensors. Both the modified

    FXLMS and independent radiation mode control algorithms

    reduce these imperfections. The floor reflection is attenuated

    and the low frequency echo disappears from the sound field

    reproduced by AWFS. Moreover, the direct sound field re-

    produced by AWFS is closer to the target wave field than the

    sound field reproduced by WFS. Therefore, AWFS compen-

    sates for the room effects, for the loudspeaker colorations,

    and for some classical WFS approximations which introduce

    supplementary physical errors in the reproduced sound field.

    According to Fig. 9, the independent radiation mode

    control algorithm Fig. 9b achieves a better sound field

    reproduction than the FXLMS algorithm Fig. 9a at the

    farther monitoring sensors. This is visible for the direct

    sound field reproduced by independent radiation mode con-

    trol, in which case the negative sign excursion of the impulse

    responses one or two samples after the direct wave front

    passage is drastically reduced for nearly all monitoring sen-

    sors. This is due to the ability to fine tune each of the radia-

    tion modes with the independent radiation mode controller

    realization of AWFS. In the FXLMS case, the higher-orderradiation modes are often far too penalized and their possible

    beneficial contribution in the sound field reproduction pro-

    cess is greatly diminished. Further explanations are pre-

    sented in Sec. III D.

    A different representation of the results allows for a gen-

    eral comparison between WFS and the two AWFS algo-

    rithms in terms of the reproduction error reduction as a func-

    tion of space. Figure10shows the normalized energies of the

    reproduction errors at each of the monitoring microphones.

    They are computed from the differences between the virtual

    and reproduced impulse responses IRs shown in Figs. 8

    and 9 and from others IRs measured for different virtual

    source positions. The normalized energies are computed as

    the sums of the quadratic error signals differences between

    virtual IRs and reproduced IRs in the time domain over the

    length of the IRs normalized by the total quadratic sum of

    the virtual IR, divided by the number of monitoring micro-

    phones. The normalization is thus achieved through division

    by the mean virtual IR energy at the monitoring micro-

    phones. According to the results shown in Fig.10,the AWFS

    algorithms reduce on average the reproduction errors in com-

    parison with WFS by controlling the reproduction errors at

    the four error sensors two of which are monitors 4 and 5.

    12

    34

    56

    7

    8

    0.02

    0.04

    0.06

    0.08

    -0.5

    0

    0.5

    Im

    p.

    response

    [1/m]

    Time [s]

    Direct and targetwave fronts

    Floorreflection

    Low frequencyreflection

    Monitor no

    FIG. 8. Reproducedthick gray lines and virtual thin black lines impulse

    responses at the monitoring sensor array shown in Fig.2for WFS with the

    system in the hemianechoic chamber.

    12

    34

    56

    78

    0.02

    0.04

    0.06

    0.08

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    12

    34

    56

    78

    0.24

    0.26

    0.28

    0.3

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    (a)

    (b)

    FIG. 9. Reproduced thick gray lines and virtual thin black lines impulse

    responses at the monitoring sensor array shown in Fig. 2 for AWFS a

    FXLMS algorithm with a penalization parameter set to 20 and b indepen-

    dent radiation mode control with the system in the hemianechoic chamber.

    1996 J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    7/12

    AWFS based on independent radiation mode control effec-

    tively provides a larger reproduction region since the higher-order modes are included in the controller in that case. Note

    that the size of the effective control region is also blurred by

    this type of representation which includes all the frequencies.

    AWFS by modified FXLMS provides a significant reproduc-

    tion error reduction at the two central monitoring micro-

    phones, but the reproduction errors are not more important

    for the other monitoring microphones. AWFS performs better

    than WFS for all the reported virtual source positions.

    2. Harmonic AWFS and radiation modes at220 Hz

    Harmonic AWFS results in the hemianechoic space, not

    reported in this paper for the sake of brevity, were in agree-

    ment with the broadband results described in Sec. III B.2. To

    support previous propositions Gauthier and Berry, 2006

    concerning the shapes of the radiation modes, Fig. 11 pre-

    sents the four pressure modes at the error sensor array.

    Clearly, these pressure modes correspond to finite approxi-

    mations of pressure, pressure gradients, and crossed second-

    order spatial derivative. Accordingly, the interpretation of

    AWFS based on independent radiation mode control or in-

    dependent control of pressure, pressure gradients, and

    crossed second-order spatial derivative originally presented

    by Gauthier and Berry 2006 is supported by this experi-

    ment. This also explains why, in the broadband experiments,

    the size of the control region is larger for the AWFS algo-

    rithm based on independent radiation mode control which

    allows the higher-order radiation modes to converge. More

    detailed harmonic AWFS experiments are reported in Sec.

    III C.

    C. Laboratory space and reverberation chamber

    The following summarizes the results obtained for the

    laboratory space and the reverberation chamber. As these two

    reproduction environments enhance the room effect on WFS,

    only parts of the results are shown to support the effective-

    ness of AWFS to compensate for the room effect.

    1. Broadband AWFS

    Since the rooms IRs were longer than for the hemi-

    anechoic space, the identified IRs and control filters were

    selected to have 512 coefficients for the laboratory space and

    1024 for the reverberation chamber with an average over 200

    realizations. The resulting synthesis and analysis filters are

    shown in Figs. 12 and 13 for the laboratory space. Once

    again, a phase optimization algorithm is applied in the fre-

    quency domain to smooth the source and pressure modes

    phase responses before inverse discrete-time Fourier trans-

    form. Moreover, a bandpass filter fourth-order Butterworth,

    60540 Hz is applied to all synthesis and analysis filters to

    reduce DC components that tend to appear in long SVD

    filters. Clearly the responses of the synthesis and analysis

    filters are longer than for the hemianechoic room. These fil-

    ters are again concentrated impulses and show a reduced

    time leakage, thanks to the phase optimization and radiation

    modes reordering algorithms which avoid any abrupt phase

    or gain transitions in the frequency domain before inverse

    discrete-time Fourier transform.

    Measured impulse responses reproduced by classical

    WFS are shown in Fig. 14 for the laboratory space and in

    Fig. 15 for the reverberation chamber. The difference be-

    -0.6 -0.3 0 0.3 0.610

    -1

    100

    Normalized energies of the errors

    -0.6 -0.3 0 0.3 0.610

    -1

    100

    -0.6 -0.3 0 0.3 0.610

    -1

    100

    -0.6 -0.3 0 0.3 0.610

    -1

    100

    x1

    [m]

    Position #1

    Position #2

    Position #3

    Position #4

    FIG. 10. Normalized energies of the error signals at each monitoring micro-

    phone for four virtual source positions in the hemianechoic chamber. Posi-

    tion 1: xo =0 , 4 , 0 m, position 2: xo = 4,0,0 m, position 3: x0= 0,1.5,0 m, and position 4: x0 =1.19,0.91,0 m. WFS errors; errors of AWFS by FXLMS; and errors of AWFS by independentradiation mode control.

    Pressure mode 1 Pressure mode 2

    Pressure mode 3 Pressure mode 4

    FIG. 11. Measured pressure modes at 220 Hz in the hemianechoic chamber.

    Sensor position; positive real part; negative real part;

    positive imaginary part; and negative imaginary part. Symbol diameterillustrates the magnitude of the corresponding value. ---: Corresponding

    computed free-field directivity.

    J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis 1997

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    8/12

    tween the WFS reproduced sound field and the virtual sound

    field at the monitor sensor array is increased when compared

    with WFS in the hemianechoic room.

    AWFS was tested to reduce the reproduction errors at

    the four error microphones. The convergence coefficient was

    set to 0.000002, and the penalization parameter was set to 30

    for AWFS by FXLMS in laboratory space. The individual

    convergence coefficients were set to 1 =0.0000075, 2 =3=0.00003, and 4 =0.00015, and the penalization parameters

    were set to 1 = 2, 2 =3 =0.2, and 4 =0.1 for AWFS by

    independent radiation mode control in the laboratory. The

    convergence coefficient was set to 0.0000005, and the penal-ization parameter was set to 0 for AWFS by FXLMS in the

    reverberation chamber. The individual convergence coeffi-

    cients were 1 =0.0000125, 2 =3 =0.00005, and 4= 0.00025, and the penalization parameters were set to m=0 for AWFS by independent radiation mode control in the

    reverberation chamber. Although the penalization parameters

    are set to zero in the reverberation chamber, the WFS solu-

    tion still contributes to the AWFS solution because the adap-

    tive filters are initialized with the WFS solution. The conver-

    gence coefficients are smaller than for the hemianechoic case

    because the size of the control filters is increased.

    The reproduced impulse responses by AWFS are shown

    in Figs.16and17 Figs.16a and17a for AWFS by FX-LMS and Figs. 16b and 17b for AWFS by independent

    radiation mode control. The imperfections of WFS shown in

    Figs.14and15are partly corrected by AWFS even outside

    the error sensor array. Both the modified FXLMS and inde-

    pendent radiation mode control algorithms reduce these im-

    perfections for the two reproduction spaces. Remarkably, the

    100 200 300 400 500

    -0.1

    -0.05

    0

    0.05

    Amp.

    Source mode #1

    100 200 300 400 500

    -0.05

    0

    0.05

    Amp.

    Source mode #2

    100 200 300 400 500

    -0.04

    0

    0.04

    Amp.

    Source mode #3

    1 128 256 384 512

    -0.1

    0

    0.1

    Amp.

    Source mode #4

    Samples

    FIG. 12. First synthesis filters Gauthier and Berry, 2006 in the time do-

    main for source modes 14 in the laboratory space. Each plot includes 24synthesis filters to create the given source mode with 24 reproduction

    sources. Each filter includes 512 coefficients.

    100 200 300 400 500

    0

    0.2

    0.4

    Amp.

    Pressure mode #1

    100 200 300 400 500-0.2

    0

    0.2

    Amp.

    Pressure mode #2

    100 200 300 400 500

    -0.1

    0

    0.1

    Amp.

    Pressure mode #3

    1 128 256 384 512-0.4

    0

    0.4

    Amp.

    Pressure mode #4

    Samples

    FIG. 13. Analysis filters Gauthier and Berry, 2006 in the time domain for

    pressure modes 14 in the laboratory space. Each plot includes 4 analysis

    filters to catch the pressure mode with 4 pressure sensors. Each filter in-

    cludes 512 coefficients.

    12

    3 4

    56

    78

    0.02

    0.04

    0.06

    0.08

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    Direct wave front

    Floorreflection

    FIG. 14. Reproducedthick gray linesand virtualthin black linesimpulseresponses at the monitoring sensor array shown in Fig.2for WFS with the

    system in the laboratory space.

    12

    34

    56

    78

    0.02

    0.04

    0.06

    0.08

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    FIG. 15. Reproducedthick gray linesand virtualthin black linesimpulse

    responses at the monitoring sensor array shown in Fig.2for WFS with the

    system in the reverberation chamber.

    1998 J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    9/12

    room compensation is achieved over a wide time range.

    Again, more than simply reducing the undesirable room ef-

    fect in the reproduced sound field, AWFS more closely re-

    produces the direct sound field than WFS.The normalized energies of the errors at each of the

    monitor microphones are shown in Fig.18 for the laboratory

    space. The AWFS algorithms reduce on average the repro-

    duction errors in comparison with WFS by controlling the

    reproduction errors at the four error sensors. AWFS based on

    independent radiation mode control gives a larger reproduc-

    tion region as the higher-order radiation modes typically

    corresponding to higher-order spatial derivatives are in-

    cluded in the controller in that case. This again highlights the

    benefits of AWFS based on independent radiation mode con-

    trol. AWFS performs better than WFS for all the reported

    virtual source positions. Similar results were obtained for the

    reverberation chamber with different virtual source positions.

    2. Harmonic AWFS

    The harmonic AWFS results are only reported for the

    laboratory space. Consistent results were obtained for the

    reverberation chamber. According to the typical FRF shown

    in Fig.5,the following frequencies were selected for on-line

    AWFS: 133, 160, 220, 280, 340, and 400 Hz.

    Examples of pressure modes are shown for 220 Hz in

    Fig. 19. Again the radiation modes approach simple multi-

    pole directivity patterns: monopole, two orthogonal dipoles,

    and tesseral quadrupole at the sensor array.

    The convergence coefficient was set to 0.01 for all fre-quencies while the penalization parameter was fixed to 1

    except at 133 and 160 Hz where they were set to 0.0005 and

    0.005, respectively for AWFS by the modified FXLMS al-

    gorithm. The convergence coefficients and penalization pa-

    rameters for the AWFS algorithm based on independent ra-

    diation mode control were then adjusted to reach a roughly

    similar residual error level at the error sensors than for the

    FXLMS algorithm. However, each higher-order radiation

    modes were less penalized than for the FXLMS algorithm

    when possibleto increase the performance outside the error

    sensor location Gauthier and Berry, 2006. This was

    achieved using either m = or m+1m. The coefficients

    are shown in Table I.

    The results are summarized in Fig. 20, for a virtual

    source in xo = 0 , 4 , 0 m, where the color axis represents the

    ELS normalized criterion at each of the monitor positions

    along x1. The criterion ELS is the moving average of the

    quadratic sum of the reproduction errors normalized by the

    quadratic sum of the target signals at the monitor sensors. As

    one can note, the WFS performance is reduced in compari-

    son with the other algorithms. As for the hemianechoic re-

    sults, using the FXLMS algorithm, the error is effectively

    reduced near the error sensors. However, as frequency in-

    12

    34

    56

    78

    0.02

    0.04

    0.06

    0.08

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    12

    34

    56

    78

    0.46

    0.48

    0.5

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    (a)

    (b)

    FIG. 16. Reproducedthick gray linesand virtualthin black linesimpulse

    responses at the monitoring sensor array shown in Fig. 2 for AWFS a

    FXLMS algorithm with a penalization parameter set to 20 and b indepen-

    dent radiation mode control with the system in the laboratory space.

    12

    34

    56

    78

    0.02

    0.04

    0.06

    0.08

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    12

    34

    56

    78

    0.88

    0.9

    0.92

    0.94

    -0.5

    0

    0.5

    Monitor noTime [s]

    Imp.

    response

    [1/m]

    (a)

    (b)

    FIG. 17. Reproducedthick gray linesand virtualthin black linesimpulse

    responses at the monitoring sensor array shown in Fig. 2 for AWFS a

    FXLMS algorithm with =0 and b independent radiation mode control

    with m = 0 with the system in the reverberation chamber.

    J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis 1999

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    10/12

    creases, this control region is spatially reduced. This is inaccordance with typical active noise control results. AWFS

    based on independent radiation mode control advantageously

    produces, as shown in Fig. 20, a larger control region since

    the higher-order radiation modes which typically imply

    higher-order spatial derivatives at the error sensor array see

    Fig.19 are allowed to converge.

    D. Importance of the higher-order radiation modes

    The importance of the higher-order radiation modes is

    highlighted by a specific set of experiments. The size of the

    active sound field reproduction effective region is shown in

    Fig. 21 in relation to the acoustic wavelength. In Fig. 21,

    several harmonic AWFS results with the WFS solution

    forced to zero are presented for a harmonic wave field at

    400 Hz in the laboratory space. The WFS solution is forced

    to zero to illustrate only the effects of the individual radiation

    modes. Clearly, when AWFS based on independent radiation

    mode control includes only one radiation mode, the results

    correspond to AWFS by FXLMS. When the number of

    -0.6 0.3 0 0.3 0.610

    -1

    100

    Normalized energies of the errors

    -0.6 0.3 0 0.3 0.610

    -1

    100

    -0.6 0.3 0 0.3 0.610

    -1

    100

    0.3 0 0.3 0.6

    100

    -0.6 0.3 0 0.3 0.610

    -1

    100

    x1

    [m]

    Position #1

    Position #2

    Position #3

    Position #5

    FIG. 18. Normalized energies of the error signals at each monitoring micro-

    phone for four virtual source positions in the laboratory space. Position 1:

    xo = 0 , 4 , 0 m, position 2: xo = 4,0,0 m, position 3: xo = 0,1.5,0 m,

    and position 5: xo = 2.8289,2.8234,0 m. WFS errors; errors of

    AWFS by FXLMS; and errors of AWFS by independent radiation modecontrol.

    Pressure mode 1 Pressure mode 2

    Pressure mode 3 Pressure mode 4

    FIG. 19. Measured pressure modes at 220 Hz in the laboratory. Sensor

    position; positive real part; , negative real part; positive imagi-

    nary part; and negative imaginary part. Symbol diameter illustrates the

    magnitude of the corresponding value. --- Corresponding computed free-

    field directivity.

    TABLE I. Convergence coefficientsmand regularization parametersm

    for harmonic AWFS based on independent radiation mode control in labo-

    ratory space.

    Freq.Hz m m

    133 0.01, 0.1, 0.1, 0.5 0.1, 0.01, 0.01, 0.001

    160 0.02, 0.1, 0.1, 0.5 0.1, 0.01, 0.01, 0.001

    220 0.02, 0.2, 0.2, 0.5 0.1, 0.01, 0.01, 0.001

    280 0.02, 0.1, 0.1, 0.5 0.1, 0.01, 0.01, 0.0005

    340 0.02, 0.1, 0.1, 0.5 0.1, 0.01, 0.01, 0.001

    400 0.02, 0.1, 0.1, 0.5 0.1, 0.01, 0.01, 0.001

    -0.5

    0

    0.5

    133160 220 280 340 400

    WFS

    x1

    [m]

    -0.5

    0

    0.5

    133160 220 280 340 400

    AWFS by FXLMS

    -0.5

    0

    0.5

    133160 220 280 340 400

    AWFS by ind. radiation mode control

    Freq. [Hz]

    x1

    [m]

    x1

    [m]

    0 0.05 0.1 0.15 0.2 0.25

    ELS

    FIG. 20. Normalized ELS criterion at the monitoring sensors for various

    frequencies and harmonic algorithms after convergence in the laboratory.

    From top to bottom: WFS, AWFS based on FXLMS and AWFS based on

    independent radiation mode control. . Measurement points. The

    0.1 contour lines;--- the 0.25 contour lines; andthe 0.5 contour lines.

    2000 J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    11/12

    higher-order modes included in AWFS based on independent

    radiation mode control increases, the size of the effectivecontrol region increases from a quarter wavelength to half

    the wavelength. This supports the previous observation on

    the importance of the higher-order radiation modes to en-

    large the effective active sound field reproduction region.

    The adaptation coefficients and penalization parameters are

    shown in TableI.

    IV. CONCLUSION AND PERSPECTIVES

    This paper investigated the objective performances of

    AWFS to compensate for the room effects or any sound

    field reproduction errorson WFS in experimental situations.

    The basic idea of AWFS is a simple combination of activenoise control principles and classic WFS. Such a combina-

    tion is entirely contained in the AWFS general cost function

    Gauthier and Berry, 2006, which consists of minimization

    of the reproduction errors typically caused by the room re-

    sponseKlehs and Sporer, 2003;Sporer and Klehs, 2004 or

    the system limitation Corteel, 2006b at several points in

    space, along with a regularization that penalizes the adaptive

    solution departure from the classic WFS solution. This pe-

    nalizing is what makes AWFS original in comparison with

    other sound field reproduction techniques based on active

    sound control and adaptive filtering.

    The results presented in this paper show that the AWFS

    system successfully achieves active sound field reproduction

    in more or less reflective spaces: hemianechoic chamber,

    laboratory space, and reverberation chamber. These experi-

    ments validate the AWFS concept and demonstrate the physi-

    cal possibility of progressive sound field reproduction in re-

    flective rooms. For the three rooms, it was shown that AWFS

    reduces the reproduction errors with respect to WFS repro-

    duction errors in the reproduction region. AWFS based on

    independent control provides an extended effective area of

    sound field reproduction since each radiation mode conver-

    gence is independently adjusted so that higher-order modes

    converge in the allowed time. Since these higher-order

    modes are of great importance to enlarge the effective repro-

    duction region, the possibility to control them independently

    is a major advantage of AWFS based on independent radia-

    tion mode control in comparison with modified FXLMS,

    which does not allow such independent control of each ra-

    diation mode convergence.

    One of the original contributions of these results is that

    they establish, for the first time to the authors knowledge,

    the validity of the AWFS concept based on independent ra-diation mode control including SVD broadband filters for

    broadband impulse reproduction by the way of controlled

    experiments.Bai and Elliott 2004 already considered SVD

    plant decoupling for cross-talk cancellation, but their paper

    was limited to theoretical investigations. Moreover, at the

    heart of the AWFS concept, is the definition of an a priori

    solutionWFSwhich, to the authors knowledge, was never

    used or experimentally tested within an adaptive signal pro-

    cessing or active noise control architecture. This paper on

    experimental AWFS supports the practical interest of AWFS.

    The reported experiments were performed with a spe-

    cific reproduction source and error sensor configuration.However, AWFS is not limited to a specific configuration.

    These experiments with AWFS were performed to evaluate

    the method. AWFS could be tested with different configura-

    tions or for different practical problems: sound environment

    reproduction, mock-up with sound field simulation system,

    etc. A typical AWFS extension would, for example, include

    more loudspeakers and more error sensors. Indeed, we ex-

    pect that a larger effective reproduction region is achievable

    with a compact sensor array which would include more sen-

    sors, like a dense circular microphone array. Three-

    dimensional configurations of either loudspeaker or micro-

    phone arrays could also be implemented within the AWFSframework.

    The tested AWFS system and implementation should be

    regarded as proto-AWFS. Indeed, before AWFS can be used

    for a practical application, several modifications should be

    done. For example, to reduce the obstruction of the error

    sensors in the listening area, the control filters can be calcu-

    lated off-line for the virtual source positions and saved, after

    which the error sensors can be removed. An example of ef-

    fective bank of compensation filters for WFS direct-sound-

    field equalization has been reported by Corteel 2006b.

    Moreover, several adaptive algorithms could be applied to

    AWFS, such as frequency-domain adaptation, sparse adapta-tion, etc. Elliott, 2001 to improve the convergence proper-

    ties of the algorithm or to reduce the computational burden.

    Future research on AWFS should be conducted within a

    specific practical application such as sound environment re-

    production or sound reproduction in a dedicated listening

    room to evaluate the potential of the method in real situa-

    tions, possibly using more reproduction sources and more

    error sensors. AWFS and the corresponding algorithms

    should also be tested and evaluated on the basis of subjective

    performance.

    -0.5 0 0.510

    -2

    10-1

    100

    101

    x1

    [m]

    E

    LS

    (monitors)

    AWFS, FXLMS

    SVD with first mode only

    SVD with modes 1 to 3

    SVD with all modes

    /2

    0.8575 m

    /4

    FIG. 21. NormalizedELS criterion at the monitoring sensors for harmonic

    algorithms at 400 Hz after convergence in the laboratory space using AWFS

    with the WFS solution forced to zero. The wavelength and some corre-

    sponding fractions are included.

    J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008 P.-A. Gauthier and A. Berry: Experiments with adaptive wave field synthesis 2001

  • 8/11/2019 2008 Adaptive wave field synthesis for active sound field reproduction: Experimental results

    12/12

    ACKNOWLEDGMENTS

    This work was supported by NSERC Natural Sciences

    and Engineering Research Council of Canada, NATEQ

    Fond Qubecois de la Recherche sur la Nature et les Tech-

    nologies, VRQ Valorisation Recherche Qubec, and Uni-

    versit de Sherbrooke. This research was conducted in col-

    laboration with CIRMMT Center for Interdisciplinary

    Research in Music Media and Technology, McGill Univer-

    sity. The authors acknowledge the contribution of Emman-

    uel Corratg, who contributed to the construction of the har-monic AWFS system and harmonic experiments.

    AES Staff Writer 2005. Multichannel audio systems and techniques, J.

    Audio Eng. Soc. 53, 329335.

    Asano, F., and Swanson, D. C. 1995. Sound equalization in enclosures

    using modal reconstruction, J. Acoust. Soc. Am. 98, 20622069.

    Bai, M. R., and Elliott, S. J.2004. Preconditioning multichannel adaptive

    filtering algorithms using EVD- and SVD-based signal prewhitening and

    system and decoupling, J. Sound Vib. 270, 639655.

    Berkhout, A. J., de Vries, D., and Vogel, P. 1993. Acoustic control by

    wave field synthesis, J. Acoust. Soc. Am. 93, 27642778.

    Betlehem, T., and Abhayapala, T. D. 2005. Theory and design of sound

    field reproduction in reverberant rooms, J. Acoust. Soc. Am. 117, 2100

    2111.

    Blauert, J. 1999. Spatial Hearing, The Psychophysics of Human Sound

    Localization MIT, Cambridge.Blesser, B., and Salter, L.-R. 2007. Spaces Speak, Are You Listening?

    Experiencing Aural Architecture MIT, Cambridge.

    Bouchard, M., and Quednau, S. 2000. Multichannel recursive-least-

    squares algorithms and fast-transversal-filter algorithms for active noise

    control and sound reproduction systems, IEEE Trans. Speech Audio Pro-

    cess. 8, 606618.

    Bravo, T., and Elliott, S. J. 2004. Variability of low frequency soundtransmission measurements, J. Acoust. Soc. Am. 115, 29862997.

    Cabell, R. H., and Fuller, C. R. 1999. A principal component algorithm

    for feedforward active noise and vibration control, J. Sound Vib. 227,

    159181.

    Camurri, A., and Ferrentino, P.1999. Interactive environments for music

    and multimedia, Multimedia Syst. 7, 3247.

    Choi, J.-W., and Kim, Y.-H. 2004. Manipulation of sound intensity withina selected region using multiple sources, J. Acoust. Soc. Am. 116, 843

    852.Corteel, E. 2006a. On the use of irregurlarly spaced loudspeaker arrays

    for wave field synthesis, potential impact on spatial aliasing frequency,

    Proceedings of the Ninth International Conference on Digital Audio Ef-

    fectsDAFX, Montreal, Canada, Sept. 1820, 2006.

    Corteel, E. 2006b. Equalization in an extended area using multichannelinversion and wave field synthesis, J. Audio Eng. Soc. 54, 11401161.

    Davis, M. F. 2003. History of spatial coding, J. Audio Eng. Soc. 51,

    554569.

    de Vries, D., Start, E. W., and Valstar, V. G. 1994. The wave field syn-

    thesis concept applied to sound reinforcement: Restrictions and solutions,

    AES 96th Convention, Convention Paper 3812, Amsterdam, Feb. 1994.

    Elliott, S.2001.Signal Processing for Active ControlAcademic, London.

    Elliott, S. J., and Nelson, P. A.1989. Multiple-point equalization in room

    using adaptive digital filters, J. Audio Eng. Soc. 37, 899907.

    Epain, N., Friot, E., and Rabau, G. 2004. Indoor sonic boom reproduction

    using ANC, Proceedings of Active 2004, Williamsburg, USA, Sept. 2022, 2004.

    Fuster, L., Lpez, J. J., Gonzlez, A., and Zuccarello, P. D. 2005. Roomcompensation using multichannel inverse filters for wavefield synthesis

    systems, AES 118th Convention, Convention Paper 6401, Barcelona,

    Spain, May 28-31, 2005.

    Gauthier, P.-A., and Berry, A. 2006. Adaptive wave field synthesis with

    independent radiation mode control for active sound field reproduction:

    Theory, J. Acoust. Soc. Am. 119, 27212737.

    Gauthier, P.-A., and Berry, A. 2007. Objective evaluation of room effects

    on wave field synthesis, Acta. Acust. Acust. 93, 824836.

    Gaunthier, P.-A., and Berry, A. 2008. Adaptive wave field synthesis for

    broadband active sound field reproduction: Signal processing, J. Acoust.

    Soc. Am. 123, 20032016.

    Gauthier, P.-A., Berry, A., and Woszczyk, W. 2005a. Sound-field repro-

    duction in-room using optimal control techniques: Simulations in the fre-

    quency domain, J. Acoust. Soc. Am. 117, 662678.

    Gauthier, P.-A., Berry, A., and Woszczyk, W. 2005b. Wave field synthe-

    sis, adaptive wave field synthesis and ambisonics using decentralized

    transformed control: Potential applications to sound field reproduction and

    active noise control, J. Acoust. Soc. Am. 118, 1967.Grau, O. 2003. Virtual ArtFrom Illusion to Immersion MIT, Cam-

    bridge.

    Keller, M., Roure, A., and Marrot, F. 2006. Acoustic field reproduction

    for psychoacoustic experiments: Application to aircraft interior noise,

    Proceedings of Active 2006, Adelaide, Australia, Sept. 1820, 2006.

    Kendall, G. S. 1995. A 3-D sound primer: Directional hearing and stereo

    reproduction, Comput. Music J. 19, 2346.

    Klehs, B., and Sporer, T.2003. Wave field synthesis in the real world. 1.

    In the living room, AES 114th Convention, Convention Paper 5727, Am-

    sterdam, March 2225, 2005.

    Merimaa, J., and Pulkki, V. 2005. Spatial impulse response rendering. I.

    Analysis and synthesis, J. Audio Eng. Soc. 12, 11151127.

    Nelson, P. S., Kirkeby, O., and Takeuchi, T. 1997. Sound fields for the

    production of virtual acoustic images, J. Sound Vib. 204, 386396.

    Poletti, M. A. 2000. A unified theory of horizontal holographic sound

    systems, J. Audio Eng. Soc. 48, 11551182.Radlovi, B. D., Williamson, R. C., and Kennedy, R. A. 2000. Equaliza-

    tion in an acoustic reveberant environment: Robustness results, IEEE

    Trans. Speech Audio Process. 8, 311319.

    Rumsey, F. 2001. Spatial AudioFocal, Oxford.

    Santilla, A. O. 2001. Spatially extended sound equalization in rectangu-

    lar rooms, J. Acoust. Soc. Am. 110, 19891997.

    Sporer, T., and Klehs, B.2004. Wave field synthesis in the real world. 2.

    In the movie theatre, AES 116th Convention, Convention Paper 6055,

    Berlin, Germany, May 811, 2004.

    Spors, S., Kuntz, A., and Rabenstein, R. 2003. An approach to listening

    room compensation with wave field synthesis, Proceedings of the AES

    24th International Conference, Canada, June 2628, 2003, pp. 7082.

    Spors, S., Renk, M., and Rabenstein, R. 2005. Limiting effects of active

    room compensation using wave field synthesis, AES 118th Convention,

    Convention Paper 6400, Barcelona, Spain, May 2831, 2005.

    Spors, S., and Rabenstein, R.2006. Spatial aliasing artifacts produced bylinear and circular-loudspeaker arrays used for wave field synthesis, AES

    120th Convention, Convention Paper.

    Start, E. W., de Vries, D., and Berkhout, A. J. 1999. Wave field synthesis

    operators for bent line arrays in 3D space, Acust. Acta Acust. 85, 883

    892.

    Start, E. W., Valstar, V. G., and de Vries, D. 1995. Application of spatial

    bandwidth reduction in wave field synthesis, AES 98th Convention, Con-

    vention Paper 3972.

    Takane, S., and Sone, T. 2004. A new theory for active suppression of

    reflected sound waves from the walls based on Kirchhoff-Helmholtz

    boundary integral equation, Acoust. Sci. & Tech. 25, 3744.

    Takane, S., Suzuki, Y., and Sone, T. 1999. A new method for global

    sound field reproduction based on Kirchhoffs integral equation, Acust.

    Acta Acust. 85, 250257.

    Veit, I., and Sander, H. 1987. Production of spatially limited diffuse

    sound field in an anechoic room, J. Audio Eng. Soc. 35, 138143.Verheijen, E. N. G. 1997. Sound reproduction by wave field synthesis,

    Ph.D. thesis, Delft University of Technology, Delft, The Netherlands.

    Williams, E. G. 1999. Fourier AcousticsSound Radiation and Nearfield

    Acoustical Holography Academic, London.

    Woszczyk, W., Cooperstock, J., Roston, J., and Martens, W.2005. Shake,

    rattle, and roll: Getting immersed in multisensory, interactive music via

    broadband networks, J. Audio Eng. Soc. 53, 336344.

    2002 J A t S A V l 123 N 4 A il 2008 P A G thi d A B E i t ith d ti fi ld th i