lnt2007 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/6224344
Active listening room compensation for massive multichannel sound

reproduction systems using wave-domain adaptive ﬁltering
Article in The Journal of the Acoustical Society of America · August 2007

DOI: 10.1121/1.2737669 · Source: PubMed
CITATIONS READS
102 1,064
4 authors, including:
Sascha Spors Rudolf Rabenstein

University of Rostock Friedrich-Alexander-University of Erlangen-Nürnberg
312 PUBLICATIONS 4,240 CITATIONS 272 PUBLICATIONS 2,763 CITATIONS
SEE PROFILE SEE PROFILE
All content following this page was uploaded by Sascha Spors on 27 May 2014.
The user has requested enhancement of the downloaded file.

Active listening room compensation for massive multichannel
sound reproduction systems using wave-domain adaptive
filtering
Sascha Sporsa兲 and Herbert Buchner
Deutsche Telekom Laboratories, Berlin University of Technology, Ernst-Reuter-Platz 7, 10587 Berlin,
Germany
Rudolf Rabenstein
Multimedia Communications and Signal Processing, University Erlangen-Nuremberg, Cauerstrasse 7,
91058 Erlangen, Germany
Wolfgang Herbordt
Center of Competence Signal Processing, Rohde & Schwarz GmbH & Co. KG, 81614 Munich, Germany
共Received 11 October 2006; revised 16 April 2007; accepted 16 April 2007兲
The acoustic theory for multichannel sound reproduction systems usually assumes free-field
conditions for the listening environment. However, their performance in real-world listening
environments may be impaired by reflections at the walls. This impairment can be reduced by
suitable compensation measures. For systems with many channels, active compensation is an option,
since the compensating waves can be created by the reproduction loudspeakers. Due to the
time-varying nature of room acoustics, the compensation signals have to be determined by an
adaptive system. The problems associated with the successful operation of multichannel adaptive
systems are addressed in this contribution. First, a method for decoupling the adaptation problem is
introduced. It is based on a generalized singular value decomposition and is called eigenspace
adaptive filtering. Unfortunately, it cannot be implemented in its pure form, since the continuous
adaptation of the generalized singular value decomposition matrices to the variable room acoustics
is numerically very demanding. However, a combination of this mathematical technique with the
physical description of wave propagation yields a realizable multichannel adaptation method with
good decoupling properties. It is called wave domain adaptive filtering and is discussed here in the
context of wave field synthesis. © 2007 Acoustical Society of America. 关DOI: 10.1121/1.2737669兴
PACS number共s兲: 43.60.Dh, 43.60.Pt, 43.60.Tj, 43.60.Ac 关EJS兴 Pages: 354–369
I. INTRODUCTION lowing will introduce such massive multichannel reproduc-

tion systems, will illustrate the influence of the listening
The ultimate goal of sound reproduction is to create the room on their performance, and will briefly review existing
perfect acoustic illusion. Several generations of engineers listening room compensation systems.
have invented various sound reproduction systems in the past
decades.1–5 The perfect acoustic illusion has never been re-
alized by either of them. Nevertheless, sound reproduction
A. Massive multichannel sound reproduction systems
has improved considerably in terms of quality and spatial
impression. However, there are still a number of open prob- Spatial audio reproduction systems using loudspeakers
lems. One of these, common to most reproduction systems, can be roughly classified into so-called stereophony based
is the influence of the room where the reproduction takes and sound field reconstruction based. Stereophony-based
place, the so-called listening room. This contribution intro- systems operate typically with five or seven spatial
duces a highly efficient approach to the active compensation channels.6 The loudspeaker driving signals for each channel
of the listening room for reproduction systems with a high are obtained by carefully processing the source material
number of reproduction channels. The problems that arise for 共usually in a sound engineering studio兲 and then stored or
systems with high channel numbers pose fundamental re- transmitted as separate tracks, typically one for each loud-
strictions on the realizability of active listening room com- speaker, in a standardized way. Consequently, the loud-
pensation using standard algorithms. In order to distinguish speaker setup at the user’s site has to conform to the corre-
these systems and their problems from conventional multi- sponding standard for optimal spatial reproduction. The
channel reproduction systems with low channel numbers techniques for processing the source material into tracks for
共⬍10兲 these are referred to as massive multichannel sound each loudspeaker channel are borrowed from conventional
reproduction systems in the context of this article. The fol- two-channel stereo recording and mixing. They rely mainly
on amplitude panning relative to the target position of the
listener. An optimal spatial reproduction is only achieved in
a兲
Electronic mail: [email protected] the vicinity of this target position, the so-called sweet spot.
354 J. Acoust. Soc. Am. 122 共1兲, July 2007 0001-4966/2007/122共1兲/354/16/$23.00 © 2007 Acoustical Society of America
FIG. 1. 共Color online兲 Simplified example that shows
the effect of the listening room on the auralized wave
field. The dashed lines from one virtual source to one
exemplary listening position show the acoustic rays for
the direct sound and one reflection off the sidewall of
the virtual recording room. The solid line from one
loudspeaker to the listening position shows a reflection
of a loudspeaker wave field off the wall of the listening
room.
In contrast, sound field reconstruction based systems addressing this issue. The following sections describe the
avoid this spatial constraint. They are characterized by a high influence of the listening room, discuss compensation meth-
number of reproduction channels, typically some ten to sev- ods, and give an outline of this article.
eral hundred. For these numbers it is neither possible nor
desirable to store or transmit all the loudspeaker signals for
playback. Since the loudspeaker setup often depends on the B. The influence of the listening room
listening room architecture, each system will potentially The influence of the listening room on the sound scene
have an individual speaker configuration. Therefore, the re- reproduced by a multichannel reproduction system will be
production system cannot be built according to a certain illustrated first in an intuitive fashion. For this purpose the
standard for the number and placement of loudspeakers. In simple reproduction scenario illustrated by Fig. 1 is consid-
this situation, the driving signals for the loudspeakers have to ered in the following. The projection of an acoustic scene in
be computed in real time by the reproduction system. The a church 共e.g., a singer performing in the choir兲 into the
input for this processing step are the signals of the sound listening room is shown as an example. For simplicity the
sources, their spatial position, and room acoustic information propagation of sound waves is illustrated by acoustic rays in
of the scene to be reproduced. This acoustic information is Fig. 1. The dashed lines in Fig. 1 from the virtual source to
either collected during the recording of the source signals one exemplary listening position show the acoustic rays for
共e.g., in a concert hall兲 or derived from models of synthetic the direct sound and several reflections off the sidewalls of
acoustic spaces. the church. The loudspeaker system in the listening room
Two well-known approaches for generating multichan- reproduces the direct sound and the reflections in order to
nel loudspeaker driving signals are ambisonics and wave create the desired spatial impression. The theory behind
field synthesis. Ambisonics systems represent the sound field nearly all sound reproduction methods assumes an anechoic
in an enclosure by an expansion into three-dimensional basis listening room which does not exhibit any reflections of its
functions. They typically use a speaker arrangement which own. The solid line in Fig. 1 from one loudspeaker in the
resembles a sphere around the listening area. The contribu- upper row to the listening position illustrates a possible re-
tion of the sound sources to each basis function may be reflection of the wave field produced by this loudspeaker off
the wall of the listening room. This additional reflection
corded with special microphone arrangements.7,8 Common
caused by the listening room may impair the desired spatial
realizations make use of only the contributions up to first
impression.
order.9 Wave field synthesis 共WFS兲 is based on a physical
The influence of the listening room on the performance
description of wave propagation illustrated by Huygen’s
of sound reproduction systems is a topic of active
principle. The mathematical foundation is given by the research.15–26 Since the acoustic properties of the listening
Kirchhoff-Helmholtz integral. Certain simplifications and ap- room and the reproduction system used may vary in a wide
proximations lead to the description of a realizable multi- range, no generic conclusion can be given for the perceptual
channel sound reproduction system.10–14 influence of the listening room. However, it is generally
Common to both these approaches for massive multi- agreed that the reflections imposed by the listening room will
channel sound reproduction is that they rely on free-field have influence on the perceived properties of the reproduced
acoustic wave propagation and do not consider the influence scene. These influences may be, e.g., degradation of direc-
of reflections within the listening room. Since these reflec- tional localization performance or sound coloration.
tions may impair the carefully designed spatial sound field, Summarizing, a reverberant listening room will superim-
their influence should be minimized by taking appropriate pose its characteristics on the desired impression of the re-
countermeasures. This contribution presents an approach to corded room. Listening room compensation aims at eliminat-
J. Acoust. Soc. Am., Vol. 122, No. 1, July 2007 Spors et al.: Active listening room compensation 355
ing or reducing the effect of the listening room. The quire a large number of reproduction channels. Additionally
following will classify and briefly introduce existing listen- an adequate analysis of the reproduced wave field will re-
ing room compensation approaches. quire a large number of analysis channels as well. Various
active listening room compensation systems for such mas-
sive multichannel reproduction systems have recently been
C. Listening room compensation systems
introduced.36–46 The proposed approaches can roughly be
Three basic classes of techniques can be identified: classified into three classes depending on the algorithms used
to compute the compensation filters: 共1兲 noniterative, 共2兲 it-
1. Passive listening room compensation,
erative, 共3兲 and adaptive room compensation techniques.
2. Consideration of the influence of the room in the render-
Most of the currently proposed listening room compen-
ing algorithm, and
sation schemes fall into the first two classes. They assume
3. Active listening room compensation.
that the acoustical environment is time-invariant and has
Passive listening room compensation applies acoustic insula- been characterized at some fixed time by impulse response
tion materials to the listening room as countermeasure measurements. The task of computing the compensation fil-
against its reflections. However, it is well known that acous- ters is typically formulated as a matrix inversion problem
tic insulation gets impractical and costly above an even which is solved with noniterative or iterative methods. In
rather modest level of sound absorption, especially for low general, the listening room characteristics may change over
frequencies. Thus, in practical setups passive room compen- time. For instance as a result of a temperature variation in the
sation alone cannot provide a sufficient suppression of listen- listening room the speed of sound and hence the acoustic
ing room reflections. properties will change.47,48 This calls for an adaptive compu-
The second class of approaches takes the influence of tation of the compensation filters on the basis of an analysis
the room on the auralized field into account in calculating the of the reproduced wave field. Therefore we will focus on the
loudspeaker driving signals. Almost every virtual scene will derivation of an adaptive algorithm within this article.
include at least some acoustic reflections, and the basic idea A wide variety of problems are related to the algorithms
in this class of approaches is to modify the rendering of the used for the adaptation of the compensation filters. The most
scene according to the additional reflections produced by the severe problem for a scenario with many playback and
listening room. However, no solutions or algorithms that suc- analysis channels is that the adaptation of the compensation
cessfully demonstrate the applicability of this idea are known filters is subject to fundamental problems.49 Thus it seems
to the authors at the time this article was written. A very that massive multichannel active room compensation may
rough sketch of some ideas can be found in Ref. 27. solve some of the problems of the multipoint room compen-
The third class of approaches uses concepts from active sation systems like position dependence and inversion prob-
control to perform the desired compensation. For sound re- lems but only at the expense of new problems related to the
production, synergies with the reproduction system are used multichannel adaptation.
in order to control the wave field. Ideally, the desired control This article will derive, discuss, and evaluate a novel
over the undesired reflections can be applied within the en- efficient approach to adaptive active listening room compen-
tire listening area by destructive interference. A concept sation for spatial audio systems that yields an extended com-
shared by most active room compensation approaches, due to pensated area compared to the multipoint approaches. It also
the causal nature of listening room reflections, is to prefilter summarizes and extends our earlier work in Refs. 40–43.
the loudspeaker driving signals using suitable compensation This approach is based on a technique that we denote as
filters. These filters are computed by analyzing the repro- wave domain adaptive filtering 共WDAF兲. In the following a
duced wave field. Most active compensation systems pub- short overview of this article is given.
lished in the past utilize only some few 共⬍10兲 loudspeakers
and analysis 共microphone兲 positions.28–31 As a consequence
D. Overview of this article
they are not able to provide sufficient control over the wave
field or are they able to sufficiently analyze the reproduced Section II discusses the concept of active listening room
wave field throughout the entire listening area. As a result, compensation in more detail and introduces the required no-
the influence of the listening room will be compensated tation. Section III presents a general concept for the adapta-
mainly at the analyzed positions with the potential occur- tion of the compensation filters and lists its known problems.
rence of severe artifacts at other positions.32–35 These ap- A framework for the solution to these problems is given in
proaches are therefore termed multipoint compensation ap- Sec. IV. It is based on a transformation into the so-called
proaches. eigenspace, which allows a decoupling into a number of in-
Advanced reproduction systems like wave field synthe- dependent adaption problems. Unfortunately, this transfor-
sis and higher-order ambisonics provide an improvement in mation turns out to be data dependent, which renders it un-
terms of control over the reproduced wave field up to a cer- suitable for practical computing. An escape is shown in Sec.
tain frequency 共spatial aliasing frequency兲. They allow one V by considering the physical meaning of this transforma-
to compensate for the reflections of the listening room tion. Based on acoustic intuition, a new inverse filtering ap-
throughout the entire listening area. Above the spatial alias- proach based on WDAF is presented. It is not optimal in the
ing frequency they may be supported easily by passive lis- sense of totally decoupling the problem, but it is suitable for
tening room compensation techniques. However, they re- practical application. Finally, its application to a widely used
356 J. Acoust. Soc. Am., Vol. 122, No. 1, July 2007 Spors et al.: Active listening room compensation
FIG. 2. 共Color online兲 Block diagram illustrating the
adaptive MIMO inverse filtering approach to room
compensation.
spatial rendering technique 共wave field synthesis兲 is shown ⫻共k兲 and the M ⫻ 1 vector l共M兲共k兲 are defined in an analo-
in Sec. VI. gous way as given by Eq. 共1兲. The output vector a共M兲共k兲
results from filtering the deriving signals d共N兲共k兲 with the
II. ACTIVE LISTENING ROOM COMPENSATION idealized M ⫻ N matrix F共k兲 of free-field impulse responses.
The error e共M兲共k兲 between free-field propagation a共M兲
The following presents the concept of active room com- ⫻共k兲 and the actual microphone signals l共M兲共k兲 describes the
pensation by prefiltering of the loudspeaker signals in more deviation of the rendered wave field from the reflection-free
detail. The need for adaptation of the prefilters arises in all case. It is used to adapt the matrix of compensation filters
practical listening room situations. Moderate variations of C共k兲. After convergence of C共k兲, the response of the prefil-
the room temperature or the movement of a single person tered reproduction system in the listening room produces the
may change the acoustic properties of a listening room to desired free-field sound field.
such an extent that a previously working compensation based The matrices of impulse responses R共k兲, F共k兲, and C共k兲
on fixed compensation filters is no longer valid.48 First, the describe discrete linear multiple-input/multiple-output
basic block diagram of an adaptive system for active room 共MIMO兲 systems. Assuming point source like propagation
compensation system is presented and then the required no- for the loudspeakers, the room R共k兲 and free-field F共k兲 ma-
tation is introduced. trices of impulse responses can be related to the so-called
Green’s function G共x 兩 x0 , ␻兲. This function characterizes the
A. Adaptive system for multichannel active room solutions of the wave equation with respect to boundary con-
compensation ditions. The Green’s function can be regarded as the acoustic
The block diagram of an adaptive system for multichan- transfer function from a spatial excitation point x0 to a mea-
nel active room compensation is shown in Fig. 2. The vector surement point x. Thus, its counterpart in the temporal do-
of input signals t 兵G共x 兩 x0 , ␻兲其 can be interpreted as the
main g共x 兩 x0 , t兲 = F−1
corresponding continuous-time room impulse response be-
d共N兲共k兲 = 关d1共k兲 d2共k兲 ¯ dN共k兲兴T 共1兲
tween the positions x and x0. This impulse response may be
represents the N loudspeaker driving signals dn共k兲, n of infinite length, but for practical purposes it is truncated at
= 1 , . . . , N provided by a multichannel spatial rendering sys- a reasonable time 共e.g., when its energy has decayed below a
tem which is based on the assumption of a reflection-free suitably chosen threshold兲 resulting in a finite impulse re-
listening room. Throughout this article the discrete time in- sponse 共FIR兲. The discretized impulse response between the
dex is denoted by k. The matrices of impulse responses C共k兲, nth loudspeaker position and the mth analysis 共microphone兲
R共k兲, and F共k兲 represent the compensation filters, the listen- position is defined as
ing room impulse responses, and the desired listening room rm,n共k兲 ª g共xm兩xn,kTs兲, 共2兲
impulses responses, respectively. In order to eliminate the
effect of the listening room, free-field conditions are desired with the temporal sampling interval Ts. The M ⫻ N matrix
for reproduction. Hence, the matrix F共k兲 is composed from R共k兲 captures all these impulse responses, and thus all sam-
the free-field impulse responses from each loudspeaker to pling points of the Green’s function G共x 兩 x0 , ␻兲 in the tem-
each microphone. This matrix is derived analytically from poral domain. If the respective impulse responses in R共k兲 are
the solution of the wave equation for the free-field case. Af- actually measured, then these will also contain the nonideal
ter the compensation filters there are N prefiltered driving loudspeaker and microphone characteristics 共e.g., directivity
signals for the loudspeakers, that are combined into a column and frequency response兲 as well as the influence of the em-
vector w共N兲共k兲. A total of M microphones are used for the ployed hardware.50
analysis of the resulting wave field. Their signals are com- Analogous definitions, as given earlier for R共k兲, apply
bined in the vector l共M兲共k兲, representing spatial samples of also to the M ⫻ N matrix F共k兲 of free-field impulse re-
the desired wave field on which the undesired listening room sponses. The matrix C共k兲 contains the sequences cn,n⬘共k兲 of
reflections are superimposed. The N ⫻ 1 column vector w共N兲 compensation filters which have to be determined by the
adaptation algorithm introduced in Sec. III. Since the respec- reviewed in the following section. A detailed discussion for
tive impulse responses are finite, they will be referred to as acoustic MIMO systems can be found, e.g., in Refs. 43 and
MIMO FIR systems in the following. 50. The solution of the normal equation with the filtered-x
recursive least-squares algorithm 共X-RLS兲 is additionally
B. Frequency-domain representation of signals and shown.
systems The adaptation of the compensation filters C共k兲 is driven
For the temporal frequency-domain description of the by the vector of error signals e共M兲共k兲. Each component em共k兲
signals and systems used in this article, the discrete time is determined by the components of the signal vector d共N兲
Fourier transform 共DTFT兲51 is used. The DTFT transform ⫻共k兲 and the respective impulse response matrices as
and its inverse for the spatially discrete loudspeaker driving em共k兲 = 兺 f m,n⬘共k兲dn⬘共k兲 − 兺兺 rm,n共k兲ĉn,n⬘共k兲dn⬘共k兲,
signal dn共k兲 are defined as follows: n⬘ n n⬘
⬁
共4兲
D n共 ␻ 兲 = 兺 dn共k兲e−j␻kT , s 共3a兲
k=−⬁ where the error em共k兲 depends on the estimates ĉn,n⬘共k兲 of the
ideal compensation filters cn,n⬘共k兲. With this error signal, the
dn共k兲 =
Ts
2␲
冕 0
2␲/Ts
Dn共␻兲e j␻kTsd␻ . 共3b兲
following cost function is defined:
k M
Analogous definitions apply to the other signals wn共k兲, lm共k兲, ␰共ĉ,k兲 = 兺 ␭

␬=0
k−␬
兺兩em共␬兲兩2 ,
m=1
共5兲
am共k兲, and em共k兲. The vector of loudspeaker driving signals
d共N兲共k兲 is transformed into the temporal frequency domain by where 0 ⬍ ␭ 艋 1 denotes an exponential weighting factor. The
transforming each element dn共k兲 separately using the DTFT cost function can be interpreted as the time-averaged energy
共3a兲. The resulting vector is denoted by dគ 共N兲共␻兲. Vectors and of the error em共k兲 between the desired wave field am共k兲 and
matrices of frequency-domain signals are underlined in the the actual reproduced wave field lm共k兲 averaged over all M
following. Analogous definitions as for d共N兲共k兲 apply to the analysis positions. The optimal filter coefficients in the
other signals used. mean-squared error 共MSE兲 sense are found by setting the
The matrix of room impulse responses R共k兲 is trans- gradient of the cost function ␰共ĉ , k兲 to zero with respect to
formed into the temporal frequency domain in the same way the estimated filter coefficients ĉ共k兲. The normal equation is
as d共N兲共k兲. The resulting matrix in the frequency domain is derived by expressing the error e共M兲共k兲 in terms of the filter
referred to as room transfer matrix and denoted by R គ 共␻兲. coefficients, introducing the result into the cost function 共5兲
The matrices F共k兲 and C共k兲 are transformed analogously. and calculating its gradient. To express the normal equation
The transfer matrix Fគ 共␻兲 is referred to as free-field transfer as a linear system of equations, the estimated compensation
matrix. filters are arranged in vector form,
III. ADAPTATION OF THE ROOM COMPENSATION ĉn,n⬘共k兲 = 关ĉn,n⬘共0兲,ĉn,n⬘共1兲, . . . ,ĉn,n⬘共Nc − 1兲兴T , 共6a兲
FILTERS
ĉ共N兲
n 共k兲 = 关ĉn,1共k兲ĉn,2共k兲 ¯ ĉn,N共k兲兴 ,
T T T T
共6b兲
Room compensation can be understood as an inverse
MIMO FIR filtering problem. For only few synthesis and 共N兲
analysis channels numerous solutions to this problem have ĉ共k兲 = 关ĉ共N兲 T 共N兲
1 共k兲 ĉ2 共k兲 ¯ ĉN 共k兲兴 ,
T T T
共6c兲
been developed in the past.32–35,50 However, algorithms for where Nc denotes the number of filter coefficients. Typically
massive multichannel systems still remain a challenge. The this number has to be chosen higher than the number of
following discusses the problems that arise when standard coefficients Nr of the room impulse responses rm,n共k兲 since
adaptive filtering algorithms are used for the computation of we are dealing with an inverse identification problem.52 The
the matrix of compensation filters C共k兲. To this end, it is resulting normal equation of the inverse filtering problem is
shown how to formulate the adaptation problem such that a then given as
powerful state-of-the-art least-squares algorithm can be ap-
plied. The presentation is just detailed enough to highlight ⌽̂dd共k兲ĉ共k兲 = ⌽̂da共k兲. 共7兲
the problems associated with the application of conventional
adaptation approaches to active listening room compensa- The N2Nc ⫻ NcN2 matrix ⌽̂dd共k兲 is the time and analysis
tion. It is not intended to provide all the details necessary for position-averaged autocorrelation matrix of the filtered driv-
an implementation. Instead the discussion of the problems ing signals
will lead to an alternative approach which is presented in k
Secs. IV and V. ⌽̂dd共k兲 = 兺 ␭k−␬DR共␬兲DRT共␬兲
␬=0
A. Least-squares error adaptation of the
compensation filter = ␭⌽̂dd共k − 1兲 + DR共k兲DRT共k兲, 共8兲
The derivation of the normal equation for the adaptive where DR共␬兲 denotes the matrix of filtered driving signals.
pre-equalization problem introduced in Sec. II A is briefly The matrix of filtered driving signals is given as follows:
application of the X-RLS algorithm to active listening
room compensation.
dm,n,n⬘共k兲 = dn⬘共k兲rm,n共k兲, 共9a兲 The X-RLS algorithm deviates from the standard RLS
algorithm by using a filtered version of the driving signal for
dm,n,n⬘共k兲 = 关dm,n,n⬘共k兲dm,n,n⬘共k − 1兲 ¯ dm,n,n⬘共k − Nc adaptation. The calculation of the filtered driving signals re-
quires knowledge of the room response R共k兲, which is in
+ 1兲兴T , 共9b兲 general not known a priori and will be time variant. Hence,
the room characteristics have to be identified additionally
共N兲
dm,n 共k兲 = 关dm,n,1
T
共k兲dm,n,2
T
共k兲 ¯ dm,n,N
T
共k兲兴T , 共9c兲 using a multichannel RLS algorithm. Its normal equation can
be derived in the same manner as shown earlier for the com-
d共N,N兲
M
共N兲
共k兲 = 关dm,1 共N兲
共k兲Tdm,2 共N兲
共k兲T ¯ dm,N 共k兲T兴T , 共9d兲 pensation filters. It is given by
⌽̂ww共k兲r̂共k兲 = ⌽̂wl共k兲, 共12兲

DR共k兲 = 关d共N,N兲
1 共k兲d共N,N兲
2 共k兲 ¯ d共N,N兲
M 共k兲兴. 共9e兲
Note that DR共k兲 is composed from all N2 combinations re- where ⌽̂ww共k兲 denotes the autocorrelation matrix of the fil-
sulting from filtering the driving signals dn⬘共k兲 with all inputs tered loudspeaker driving signals w共N兲共k兲, ⌽̂wl共k兲 the cross-
of the MIMO room responses rm,n共k兲 for all possible combi- correlation matrix between the filtered loudspeaker driving
nation of n and n⬘. This results from rearranging Eq. 共4兲 in signals w共N兲共k兲 and the analysis signals l共M兲共k兲, and r̂共k兲 the
order to isolate the filter coefficients ĉ共k兲. estimated coefficients of the room transfer matrix. Figure 3
The N2Nc ⫻ 1 vector ⌽̂da共k兲 can be interpreted as the illustrates the X-RLS algorithm applied to the active listen-
time and analysis position-averaged cross-correlation vector ing room compensation scenario.
between the filtered driving signals and the desired signals
which is defined as B. Fundamental problems of adaptive inverse filtering
k Three fundamental problems of massive multichannel
⌽̂da共k兲 = 兺 ␭k−␬DR共␬兲a共M兲共␬兲 adaptive pre-equalization can be identified from the normal
␬=0 equation 共7兲 and the definition 共8兲 of the autocorrelation ma-
= ␭⌽̂da共k − 1兲 + DR共k兲a共M兲共k兲. 共10兲 trix ⌽̂dd共k兲. These are:
1. Nonuniqueness of the solution,
The optimal pre-equalization filter with respect to the cost
function 共5兲 is given by solving the normal equation 共7兲 with 2. Ill-conditioning of the autocorrelation matrix ⌽̂dd共k兲, and
respect to the filter coefficients ĉ共k兲. The normal equation is 3. Computational complexity for massive MIMO systems.
typically not solved in a direct manner but by recursive up-
The first problem is related to minimization of the cost func-
dates of the filter coefficients. A recursive update equation
tion ␰共ĉ , k兲. Minimization of the cost function ␰共ĉ , k兲 may not
for the filter coefficients can be derived from the normal
provide the optimal solution in terms of identifying the in-
equation 共7兲, the recursive definitions of the correlation ma-
verse system to the room transfer matrix. Depending on the
trices given by Eqs. 共8兲 and 共10兲, and the definition of the
driving signals d共N兲共k兲 there may be multiple solutions for
error signal e共M兲共k兲 as follows:
ĉ共k兲 that minimize ␰共ĉ , k兲.54 This problem is often referred to
ĉ共k兲 = ĉ共k − 1兲 − ⌽̂dd
−1
共k兲DR共k兲e共M兲共k兲. 共11兲 as nonuniqueness problem.
The second and the third fundamental problems are re-
Equation 共11兲 together with the recursive definition of the lated to the solution of the normal equation 共7兲. The normal
correlation matrix ⌽̂dd共k兲 given by Eq. 共8兲 constitutes the equation has to be solved with respect to the coefficients ĉ共k兲
basis of the X-RLS algorithm. The inverse ⌽̂dd −1
共k兲 of the of the room compensation filter. However, the autocorrela-
autocorrelation matrix is typically computed in a recursive tion matrix ⌽̂dd共k兲 is typically ill-conditioned for the consid-
fashion by applying the matrix inversion lemma.53 ered multichannel reproduction scenarios.49 The filtered driv-
TABLE I. Complexity of adaptive listening room compensation using a eគ 共M兲共␻兲 = aគ 共M兲共␻兲 − I共M兲共␻兲 = Fគ 共␻兲dគ 共N兲共␻兲
multichannel filtered-x RLS 共X-RLS兲 algorithm in comparison to the pro-
posed approach of eigenspace adaptive filtering 共EAF兲 in conjunction with −R គ 共␻兲dគ 共N兲共␻兲.
គ 共␻兲C 共13兲
single-channel X-RLS algorithms. For the complexity of the transformations
Xគ H共␻兲 and V
គ 共␻兲 only their application has been considered but not their Optimal pre-equalization is obtained by minimizing this er-
computation by the GSVD.
ror: eគ 共M兲共␻兲 → 0. The least-squares solution to Eq. 共13兲 in this
X-RLS EAF sense, with respect to the compensation filters, is given
by53,56
Adaption of Dimension of ⌽̂dd共k兲 N N c ⫻ N cN
2 2
Nc ⫻ Nc
Ĉ共␻兲 Complexity O共N4Nc2兲 M · O共Nc2兲 គ 共␻兲 = R
C គ +共␻兲Fគ 共␻兲, 共14兲
Identification of Dimension of ⌽̂ww共k兲 NNc ⫻ NcN Nc ⫻ Nc where Rគ +共␻兲 denotes the Moore-Penrose pseudoinverse of
R̂共␻兲 Complexity O共N2Nc2兲 M · O共Nc2兲 គ 共␻兲.
R
Transformations Complexity O共NM兲 + O共M 2兲
គ H共␻兲 and V
X គ 共␻兲
B. Generalized singular value decomposition
ing signals DR共k兲 will contain cross-channel 共spatial兲 It will be assumed in the following that R គ 共␻兲 and Fគ 共␻兲
correlations due to the deterministic nature of most auraliza- have the dimensions M ⫻ N with N 艌 M. Hence, a reproduc-
tion algorithms. Also temporal correlations may be present tion system with equal or less analysis positions than loud-
for typical virtual source signals. Besides the ill-conditioning speakers is assumed. This restriction is meaningful with re-
spect to the solution of the pre-equalization problem, since in
also the dimensionality of the autocorrelation matrix ⌽̂dd共k兲
the general case an exact reproduction at M positions can
poses problems. Massive multichannel systems exhibit a
only be gained for such scenarios.57 However, the derived
high number of reproduction channels and additionally the
results can be generalized straightforwardly to arbitrary
length Nc of the inverse filter has to be chosen quite long for
channel numbers M and N.
a suitable suppression of reflections.52 As a consequence, the
The singular value decomposition 共SVD兲 states that any
derivation of the compensation filters will get computation-
matrix can be decomposed into two unitary matrices and a
ally very demanding. Table I illustrates the complexity of the
diagonal matrix.53,56 The concept of the SVD can be gener-
X-RLS algorithm. The complexity is given on the basis of a
alized to the diagonalization of a pair of matrices. This de-
straightforward solution of the normal equation without any
composition is known as generalized singular value decom-
optimizations. O共N兲 denotes “in the order of N” as measure
position 共GSVD兲.56 The GSVD of the matrices R គ 共␻兲 and
of complexity. A more detailed complexity analysis of the
Fគ 共␻兲 is given as follows:
X-RLS algorithm can be found, e.g., in Ref. 55.
The same problems as discussed earlier apply to the
identification of the room transfer matrix using a multichan- គ 共␻兲 = X
R គ 共␻兲R
គ̃ 共␻兲V
គ H共␻兲, 共15a兲
nel RLS algorithm.54
Fគ 共␻兲 = X
គ 共␻兲Fគ̃ 共␻兲U
គ H共␻兲. 共15b兲
IV. EIGENSPACE ADAPTIVE FILTERING
The matrices X គ 共␻兲, Vគ 共␻兲, and U គ 共␻兲 are unitary matrices
In this section we will derive a generic framework for with the dimensions M ⫻ M, N ⫻ M, and N ⫻ M, respectively.
pre-equalization which explicitly solves the problems of The matrix X គ 共␻兲 is the generalized singular matrix of R គ 共␻兲
complexity and cross-channel correlations by utilizing signal and Fគ 共␻兲, the matrices V គ 共␻兲 and U គ 共␻兲 the respective right
and system transformations. It will be shown additionally
singular matrices of R គ 共␻兲 and Fគ 共␻兲. The matrices R̃គ 共␻兲 and
that the other problems mentioned earlier are highly allevi-
ated by the proposed approach. The basic idea is to perform F̃គ 共␻兲 are diagonal matrices constructed from the singular
a joint decoupling of the MIMO systems R គ 共␻兲 and Fគ 共␻兲 in values of R គ 共␻兲 and Fគ 共␻兲. The diagonal matrix R̃ គ 共␻兲 is de-
the temporal frequency domain. This decoupling yields a defined as
coupling of the MIMO adaptation problem and the autocor-
relation matrix ⌽̂dd共k兲 as will also be shown. គ 共␻兲 = diag兵关R̃1共␻兲,R̃2共␻兲, . . . ,R̃ M 共␻兲兴其,
R̃ 共16兲
A. Nonadaptive computation of room compensation where R̃1共␻兲艌 R̃2共␻兲艌 ¯ 艌 R̃B共␻兲 ⬎ 0 denote the B nonzero
filters singular values R̃m共␻兲 of Rគ 共␻兲. Their total number B is given
by the rank of the matrix R គ 共␻兲 with 1 艋 B 艋 M. For B ⬍ M
In order to gain more insight into the solution of the
pre-equalization problem the nonadaptive case will be re- the remaining singular values R̃B+1共␻兲 , R̃B+2共␻兲 , . . . , R̃ M 共␻兲
garded first. For this purpose a frequency-domain description គ 共␻兲 apply
are zero. Similar definitions as given earlier for R̃
of the pre-equalization problem depicted in Fig. 2 is used. to the matrix F̃គ 共␻兲.
The error eគ 共M兲共␻兲 between the desired aគ 共M兲共␻兲 and the The relation given by Eq. 共15a兲 can be inverted by ex-
actual គl共M兲共␻兲 signal at the M analysis positions in the fre- ploiting the unitary property of the joint and right singular
quency domain can be derived by a DTFT of Eq. 共4兲 as matrices. This results in
eigenspace adaptive inverse filtering approach to room
compensation.
គ̃ 共␻兲 = X
R គ H共␻兲R
គ 共␻兲V共␻兲. 共17兲 Iគ̃ 共M兲共␻兲 = R គ̃ 共␻兲dគ̃ 共M兲共␻兲,
គ̃ 共␻兲C 共21兲
Hence each matrix R គ 共␻兲 can be transformed into a diagonal where គl̃共M兲共␻兲 = X
គ H共␻兲lគ共M兲共␻兲 and d̃គ 共M兲共␻兲 = U
គ H共␻兲dគ 共N兲共␻兲.
matrix R̃គ 共␻兲 using the joint and right singular matrix X គ 共␻兲 Decomposition of the desired system response according to
and V គ 共␻兲. A similar relation as given by Eq. 共17兲 can be Eq. 共15b兲 yields the desired signal in the transformed domain
derived straightforwardly for F̃គ 共␻兲. The GSVD transforms as
the matrices R共␻兲 and F共␻兲 into their joint eigenspace using
ãគ 共M兲共␻兲 = Fគ̃ 共␻兲dគ̃ 共M兲共␻兲, 共22兲
the singular matrices X គ 共␻兲, V
គ 共␻兲, and U គ 共␻兲. In general,
these singular matrices depend on the matrices R គ 共␻兲 and where ãគ 共M兲共␻兲 = X
គ H共␻兲aគ 共M兲共␻兲. Equation 共21兲 together with
Fគ 共␻兲. The GSVD is a data-dependent transformation. The Eq. 共22兲 allows one to express the error eគ 共M兲共␻兲 in the trans-
transformation of R គ 共␻兲 into its diagonal representation R̃
គ 共␻兲, formed domain
as given by Eq. 共17兲, can be interpreted as pre- and postfil-
tering the room transfer matrix R គ 共␻兲 by the MIMO systems eគ̃ 共M兲共␻兲 = aគ̃ 共M兲共␻兲 − Ĩគ 共M兲共␻兲 = Fគ̃ 共␻兲d̃គ 共M兲共␻兲
Vគ 共␻兲 and X គ H共␻兲. 共M兲
The SVD can be used to define the pseudoinverse R គ +共 ␻ 兲 គ̃ 共␻兲C̃
−R គ 共␻兲dគ̃ 共␻兲, 共23兲
of the matrix R គ 共␻兲,53
where ẽគ 共M兲共␻兲 denotes the error signal for all M components
គ +共 ␻ 兲 = V
R គ 共␻兲R
គ̃ −1共␻兲X
គ H共␻兲. 共18兲 in the transformed domain. Since R̃ គ 共␻兲, C̃
គ 共␻兲, and F̃គ 共␻兲 are
diagonal matrices, the mth component of the error signal
Equation 共15b兲 and Eq. 共18兲 can be combined to derive the
Ẽm共␻兲 in the transformed domain is given by the following
following result:
relation:
គ +共␻兲Fគ 共␻兲 = V
R គ 共␻兲R̃
គ 共␻兲Fគ̃ 共␻兲U 共␻兲,
−1 H
共19兲
Ẽm共␻兲 = F̃m共␻兲D̃m共␻兲 − R̃m共␻兲C̃m共␻兲D̃m共␻兲, 共24兲
គ 共␻兲 and Fគ 共␻兲 both have full rank.
where it is assumed that R
where R̃m共␻兲, C̃m共␻兲, and F̃m共␻兲 denote the mth component
Equation 共19兲 will be utilized in the following to derive a
decoupling of the room compensation filters. of the main diagonal of R̃ គ 共␻兲, C̃
គ 共␻兲, and F̃គ 共␻兲, respectively.
The error Ẽm共␻兲 depends only on the mth component of the
C. Decoupling of the MIMO adaptation problem respective signals and systems in the transformed domain.
Thus, Eq. 共24兲 states that the MIMO adaptive inverse filter-
The decompositions of the transfer matrices Fគ 共␻兲 and ing problem can be decomposed into M single-input/single-
Rគ 共␻兲 are given by Eq. 共15兲. It remains to choose a suitable output 共SISO兲 adaptive inverse filtering problems using the
decomposition of the compensation filter matrix C គ 共␻兲. The GSVD. The computation of the pre-equalization filters can
nonadaptive solution for the pre-equalization filter is given be performed independently for each of the M transformed
by Eq. 共14兲 in terms of the pseudoinverse Rគ +共␻兲 of the room components. The transformation of the systems and signals
transfer matrix. Hence, an eigenspace expansion of C គ 共␻兲 is is performed by transforming them into the joint eigenspace
given by Eq. 共19兲. However, the system transfer matrix R គ 共␻兲 of Rគ 共␻兲 and Fគ 共␻兲 using the GSVD. Therefore this approach
is not known in general and has to be identified additionally. will be referred to as eigenspace inverse adaptive filtering.
An expansion of the pre-equalization filter can be given by Please note that the transformation is not dependent on the
utilizing Eq. 共19兲 but with unknown expansion coefficients driving signals. Figure 4 illustrates the eigenspace inverse
គ 共␻兲,
C̃ adaptive filtering approach.
គ 共␻兲 = V
C គ 共␻兲C
គ̃ 共␻兲U
គ H共␻兲, 共20兲 D. Eigenspace adaptive filtering
គ 共␻兲 denotes a diagonal matrix, where some diagonal
where C̃ In the following, the normal equation of the multichan-
elements may be zero. Using Eq. 共20兲 together with Eq. nel adaptive pre-equalization problem presented in Sec. III A
共15a兲 yields the transformed signal គl̃共M兲共␻兲 at the analysis will be specialized to the decoupled MIMO system. Due to
points the decoupling, the cost function ␰共ĉ , k兲 given by Eq. 共5兲 can
be minimized independently for each component m of the MIMO adaptive inverse filtering problem into a series
= 1 , . . . , M. The normal equation in the transformed domain of single channel adaptive inverse filtering problems. This
is then given as way most of the fundamental problems 共see Sec. III B兲 of
adapting the room compensation filters for massive multi-
ˆ ˆ
⌽̃dd,m共k兲c̃ˆm共k兲 = ⌽̃da,m共k兲, 共25兲 channel systems were solved. However, the computation of
ˆ suitable transformations using the GSVD may become very
where ⌽̃dd,m共k兲 denotes the time-averaged autocorrelation complex. In order to overcome this problem the concept of
matrix of the mth component of the transformed filtered wave-domain adaptive filtering is introduced.
ˆ
loudspeaker driving signal, ⌽̃da,m共k兲 the corresponding
cross-correlation matrix between the filtered loudspeaker A. Generic concept of wave-domain adaptive filtering
driving signal and the desired signal, and c̃ˆm共k兲 the filter
Wave-domain adaptive filtering is based on two basic
ˆ
coefficients. The autocorrelation matrix ⌽̃dd,m共k兲 has the di- ideas:
mensions Nc ⫻ Nc. Due to this reduction in dimensionality,
1. Explicit consideration of the characteristics of the propa-
the solution of the M equations given by Eq. 共25兲 is much
gation medium for the derivation of suitable transforma-
more efficient than for the adaptation using the original 共not
tions, and
transformed兲 signals. Table I summarizes the complexity re-
2. Approximation of the concept of perfect decoupling of
duction of the eigenspace approach to MIMO inverse adap-
the MIMO adaptation problem.
tive filtering in contrast to the X-RLS algorithm.
Equation 共25兲 corresponds to the well-known single The listening room transfer matrix describes the sound trans-
channel normal equation.53 The cross-channel correlations mission from the loudspeakers to the analysis positions with
present in ⌽̂dd共k兲 have been removed in the transformed do- respect to the characteristics of the propagation medium and
main by the spatial decoupling of the MIMO systems. Thus, the boundary conditions imposed by the listening room.
the nonuniqueness and ill-conditioning problem discussed in Hence it has to fulfill the wave equation and the homoge-
Sec. III B are highly alleviated. There may still be time- neous boundary conditions imposed by the room. This
domain correlations present in the filtered input signals knowledge can be used to construct efficient transformations.
which cause problems when solving the normal equation Since these transformations inherently have to account for
共25兲. However, there are numerous approaches known in the the wave nature of sound in order to perform well, this ap-
literature on single-channel adaptive 共inverse兲 filtering to proach will be referred to as wave domain adaptive (inverse)
overcome these problems.53 filtering 共WDAF兲 and the transformed domain as wave do-
In practice, however, a number of problems emerge main in the following.
from the GSVD-based transformations used to decouple the The second idea is to approximate the perfect decou-
MIMO systems. These will be discussed in the following. pling of the MIMO adaptation problem in favor of generic
transformations which are to some degree independent of the
E. Fundamental problems of eigenspace adaptive listening room characteristics. In order to keep the complex-
filtering ity low, these generic transformations need not to be strictly
The major problems of eigenspace adaptive filtering are diagonalizing, but they should still represent the MIMO sys-
that tem with as few paths as possible.
The combination of both ideas allows one to derive fixed
គ 共␻兲, V
1. The singular matrices X គ 共␻兲, and U
គ 共␻兲 depend on transformations that provide nearly the same favorable prop-
គ 共␻兲 and free-field transfer matrix Fគ 共␻兲, and
the room R erties as the optimal GSVD-based transformations used for
2. Their computation using the GSVD is quite complex. eigenspace adaptive inverse filtering with the benefit of com-
putational efficiency.
The contents of the room transfer matrix will be time
Based on the approach of eigenspace adaptive filtering
variant in general due to changes in the acoustic conditions
and the above-mentioned considerations a generic block dia-
caused by, e.g., temperature changes or persons entering the
gram of the WDAF approach can be developed. Figure 5
listening room. This requires that the room transfer matrix
displays this generic block diagram. The signal and system
Rគ 共␻兲 has to be identified additionally and that the singular
transformations are performed by three generic transforma-
matrices have to be updated each time the room transfer
tions. Their structure is not limited to the MIMO FIR sys-
matrix changes. Both tasks are computationally very de-
tems derived from the GSVD. Transformation T1 transforms
manding for massive multichannel systems. In general no
the driving signals into the wave domain, T2 inversely trans-
optimizations can be performed in practice without placing
forms the filtered loudspeaker driving signals from the wave
restrictions on the structure of the transfer matrices Fគ 共␻兲 and
domain, and T3 transforms the signals at the analysis points
Rគ 共␻兲. Section V will introduce the concept of wave-domain
into the wave domain. As previously for the eigenspace do-
adaptive filtering in order to overcome these problems.
main, the signals and transfer functions in the wave domain
are denoted by a tilde over the respective variable, since
V. WAVE-DOMAIN ADAPTIVE FILTERING
suitable transforms will be based on the idea of a transfor-
In Sec. IV an eigenspace approach to adaptive inverse mation into the eigenspace of the respective systems. The
filtering was proposed. Its main feature was the decoupling adaptation is performed entirely in the wave domain.
FIG. 5. Block diagram illustrating the wave domain
adaptive inverse filtering approach to active room com-
pensation.
Note that the generic block diagram depicted by Fig. 5 have to be considered.61 A generalization of the proposed
also includes the eigenspace adaptive inverse filtering ap- decompositions to three-dimensional representations can be
proach. In this special case the transformations are given as derived straight-forwardly.58
the following MIMO FIR systems 共see Sec. IV兲: T1
=U គ H共␻兲, T2 = V
គ 共␻兲, and T3 = X
គ H共␻兲.
The following section introduces wave field expansions 1. Circular harmonics expansion
which are potential candidates as wave domain transforma-
tions. Since they are based upon the fundamental free-field The elementary solutions of the wave equation using
solutions of the wave equation they will only decouple the cylindrical coordinates are given in Refs. 58 and 62. The
free-field transfer matrix. elementary solutions using polar coordinates can be derived
from these by discarding the components which depend on
the z coordinate. The circular harmonics decomposition of an
B. Wave field representations
arbitrary wave field is then given by
冉冉冏冏冊
A first choice to characterize an acoustic wave field is ⬁
␻
given by the acoustic pressure P共x , ␻兲 captured at all posi-
tions x. For a wide variety of applications it is convenient to
P P共x p, ␻兲 = 兺
␯=−⬁
P̆共1兲共␯, ␻兲H␯共1兲
c
r e j ␯␣
冉冏冏冊冊
represent acoustic wave fields with respect to an orthogonal
␻
basis. This basis is typically constructed from the fundamen- + P̆共2兲共␯, ␻兲H␯共2兲 r e j ␯␣ , 共26兲
tal solutions of the wave equation considering the particular c
problem. The general solution of the wave equation is then where x p = 关␣ r兴T denotes the position vector in polar coordi-
given as a weighted superposition of all elementary solu- nates, H␯共1兲,共2兲共·兲 the ␯th order Hankel function of first/second
tions. Of special interest in the following are decompositions kind, and ␯ the angular frequency. The polar coordinates are
which are based on the fundamental free-field solutions of defined as x = r cos ␣ and y = r sin ␣ within this article. Quan-
the wave equation. These depend on the underlying coordi- tities and functions whose arguments are given in polar co-
nate system and its dimensionality. The basis functions con- ordinates are denoted by a P in their index. The infinite sum
nected to spherical, cylindrical, and Cartesian coordinates are over ␯ can be interpreted as Fourier series with respect to the
known as spherical harmonics, cylindrical harmonics, and
angle ␣. The coefficients P̆共1兲,共2兲共␯ , ␻兲 are referred to as cir-
plane waves. An arbitrary wave field can be represented by
cular harmonics expansion coefficients in the following and
the expansion coefficients with respect to these basis func-
will be denoted by a breve over the respective variable. The
tions. If two-dimensional wave fields are considered then the
Hankel function H␯共1兲共兩␻ / c兩r兲 belongs to an incoming 共con-
polar and Cartesian coordinate systems are common. Here
the basis functions are circular harmonics and 共two- verging兲 and H␯共2兲共兩␻ / c兩r兲 to an outgoing 共diverging兲 cylin-
dimensional兲 plane waves. Please note that according to the drical wave.58 Thus, the expansion coefficient P̆共1兲共␯ , ␻兲 de-
term “spherical harmonics” used for the elementary solutions scribes the incoming wave field, whereas P̆共2兲共␯ , ␻兲 describes
of the wave equation in spherical coordinates,58–60 the solu- the outgoing wave field. According to Eq. 共26兲 the total wave
tions in cylindrical coordinates have been termed as cylindri- field is given as a superposition of incoming and outgoing
cal harmonics and the solutions in polar coordinates as cir- contributions.
cular harmonics within this article. Typical sound In order to get more insight into the circular harmonics
reproduction systems aim at the reproduction in a plane only. expansion a closer look at the basis functions is taken. Each
The analysis of the reproduced wave field is then typically spatial variable has its own basis function. The angular co-
performed in the reproduction plane. We will therefore limit ordinate ␣ has an exponential function as basis. Figure 6
ourselves to two-dimensional wave field representations for illustrates the angular basis functions for different angular
the following discussion of the circular harmonics and plane frequencies. For the sake of illustration the plots only show
wave decomposition. However, since the reproduction will the absolute value of the real part. It can be seen clearly that
take place in a three-dimensional environment several arti- the angular basis functions exhibit a spatial selectivity in the
facts of two-dimensional sound reproduction and analysis angular coordinate. They can be interpreted as directivity
FIG. 6. Illustration of the angular basis functions e j␯␣ of the circular harmonics. The plots show the absolute value of the real part 共兩R兵e j␯␣其兩兲 for different
angular frequencies ␯.
patterns. Hankel functions are the basis functions of the ra- decomposition into plane waves and the decomposition into
dial coordinate r. See Ref. 63 for a detailed discussion of circular harmonics. Both are based upon orthogonal basis
their properties. functions derived from the free-field wave equation and will
The circular harmonics expansion coefficients for a par- therefore provide a decoupling of the free-field transfer ma-
ticular problem can be derived by introducing the expansion trix Fគ 共␻兲 but not necessarily of the room transfer matrix
共26兲 into the underlying wave equation. However, it is also Rគ 共␻兲. Either representation may be nevertheless a suitable
possible to measure an arbitrary wave field in a bounded choice for the wave domain representation as long as they
region and derive the expansion coefficients from these mea- provide an approximate decoupling of the room transfer ma-
surements. Due to the underlying geometry circular micro- trix. In order to qualitatively evaluate the two proposed de-
phone arrays are well suited for this task.64 compositions a closer look is taken at the room response
with respect to the basis functions. In the ideal case the re-
2. Plane wave expansion production of a basis function will result only in contribu-
Arbitrary solutions of the wave equation can be ex- tions belonging to the same basis function after analysis.
pressed alternatively as superposition of plane waves travel- Consider now typical listening rooms with a rectangular
ing into all possible directions. The plane wave expansion shape as a first approximation and their response to either
coefficients P̄共1兲共␪ , ␻兲 and P̄共2兲共␪ , ␻兲 describe the spectrum plane waves or circular harmonics.
of incoming/outgoing plane waves with incidence angle ␪. The response to a plane wave of a certain angle consists
They can be derived by a plane wave decomposition.43,64 in general of a mixture of plane waves with all kinds of
The plane wave expansion coefficients exhibit a direct link to different angles. This means that the desired effect of decou-
the expansion coefficients in circular harmonics pling is not achievable with plane waves in more or less
⬁ rectangular rooms.
4␲
P̄共1兲共␪, ␻兲 = 兺 j␯P̆共1兲共␯, ␻兲e j␯␪ ,
k ␯=−⬁
共27a兲 On the other hand, the response to a circular wave de-
pends on the wavelength. Since high frequencies can be
damped effectively by passive methods, active room com-
⬁
4␲ pensation can be restricted to low frequencies. For the cor-
共2兲
P̄ 共␪, ␻兲 = 兺 j␯P̆共2兲共␯, ␻兲e j␯␪ .
k ␯=−⬁
共27b兲
responding long wavelengths, scattering at the corners of the
listening room does not play a major role. Therefore each
Equation 共27兲 states that the plane wave decomposition of a circular harmonics component emitted by a well-designed
wave field is, up to the factor j␯, given by the Fourier series reproduction system will lead to reflections which can be
of the expansion coefficients in terms of circular harmonics. described mainly by the same component. Consequently the
desired decoupling effect will be more prominent than for
C. Circular harmonics as wave domain plane waves. This intuitive consideration suggests that circu-
transformation lar harmonics seem to be a more suitable choice for the wave
The optimal choice for the transformed signal represen- domain representation than plane waves. This assumption
tation depends on the acoustic characteristics of the listening holds also for reverberant rooms as long as the energy of
room and the desired free-field response. Two different wave other components than the emitted one is considerably at-
field representations have been introduced in Sec. IV: the tenuated.
Please note that the representation of a field in a trans- the transfer matrix F̆គ should also include an additional delay
formed domain requires adequate sampling of the measured to ensure the computation of causal room compensation fil-
field. Sampling theorems for circular microphone arrays can ters. If desired, certain artifacts of WFS systems like, e.g.,
be found, e.g., in Refs. 36, 43, 65, and 66. amplitude errors 共see Ref. 61兲 can be considered in the con-
struction of the matrix F̆គ .
VI. APPLICATION TO WAVE FIELD SYNTHESIS It was proposed by the authors in previous
Section V introduced wave domain adaptive filtering. publications40–42,74–76 to perform a Fourier transformation
Now its application to spatial reproduction techniques is with respect to the angular variable of the plane wave de-
shown using wave field synthesis as an example. First, this composition to derive the signals in the wave domain. This
reproduction technique is discussed shortly. Then the appli- Fourier transformation of the plane wave decomposed sig-
cation of wave domain adaptive filtering to wave field syn- nals is up to the factor j␯ 共and a frequency correction兲
thesis is shown by specifying the transformations T1 through equivalent to calculating the circular harmonics expansion
T3 in Fig. 5. Finally some selected results of performance coefficients, since the plane wave decomposition in terms of
evaluations are presented. circular harmonics is given by the Fourier series 共27兲.
A. Wave field synthesis

Wave field synthesis 共WFS兲 is a massive multichannel C. Performance evaluation
sound reproduction technique which overcomes the “sweet After describing the application of wave domain adap-
spot” limitation well known from stereophonic surround tive filtering for wave field synthesis, its performance for
sound methods. WFS techniques are based on Huygen’s decoupling the adaptation will now be evaluated. The evalu-
principle and are formulated in terms of the Kirchhoff- ation of the circular harmonics decomposition for listening
Helmholtz integral.62 These foundations have been initially room compensation with wave domain adaptive filtering has
developed by the Technical University of Delft10–12 and were been described in detail in Ref. 77. The following quote
later extended within the European project CARROUSO.67 some selected results.
Detailed descriptions of WFS can be found in Refs. 13, 14,
43, and 68–72.
1. Performance measures
B. Application of wave domain adaptive filtering to
wave field synthesis The performance of the proposed transformations and of
The generic concept of WDAF is now specialized to the the resulting active listening room compensation system is
reproduction utilizing a WFS system and the circular har- evaluated by the ability to compact the room characteristics
monics decomposition as wave domain transformation. The into less coefficients than using the microphone signals di-
transformations T1 through T3 are based on the decomposi- rectly. This ability is described by two measures: 共1兲 the
tion into circular harmonics of the respective wave fields. energy of the elements of the room transfer matrix and 共2兲
For active room compensation only the incoming parts 共see the energy compaction performance. The first measure is de-
Sec. V B 1兲 of the respective wave fields are of interest. For fined by calculating the energy of each spatial transmission
the particular scenario considered these transformations are path for the room transfer matrix in its different representa-
specialized as follows. tions. For its representation in the pressure domain 共pressure
microphones兲 the energy of the elements of the room transfer
共1兲 Transformation T1: This transformation transforms the matrix is defined as
virtual source signal S共␻兲 using a spatial model of the
virtual source into the loudspeaker driving signals d̆គ 共M兲
in the wave domain. Suitable models for the virtual
source characteristics are line/point sources or plane
E共m,n兲 =
1
2␲
冕
−⬁
⬁
兩Rm,n共␻兲兩2d␻ . 共28兲
waves.
共2兲 Transformation T2: This transformation generates the Similar definitions apply to the room transfer matrix in its
loudspeaker driving signals from the filtered driving sig- plane wave and circular harmonics representation yielding
គ 共M兲. Equation 共26兲 together with a suitable loud-
nals w̆ the energy representations Ē共␪ , ␪0兲 and Ĕ共␯ , ␯0兲. The second
speaker selection criterion73 can be used for this purpose. measure, the energy compaction, measures the ability of a
共3兲 Transformation T3: This transformation calculates the transform to compact the energy of the room transfer matrix
circular harmonics decomposition coefficients of the to as few coefficients as possible. For one particular room
wave field within the listening area from the microphone transfer matrix this measure is defined by calculating the
array measurements.64 ratio between the energy of the first i dominant elements and
the total energy of all elements. For this purpose the energies
The free-field transfer matrix F̆គ models the free-field E共m , n兲 are sorted in descending order yielding the sorted
propagation in terms of circular harmonics from the loud- elements Esort共␩兲. Then the ratio between the energy of the
speakers to the microphone array. In the ideal case this ma- first i sorted elements and the total energy of all elements is
trix would only model the propagation delay. Besides this, calculated as follows:
FIG. 7. Absolute value of the first eight right singular vectors 共f = 80 Hz兲 for the simulated circular WFS/wave field analysis system sorted by their descending
singular values 共top left to bottom right兲. The singular vectors for a plane wave reflection factor Rpw = 0.8 at the walls of the simulated room are shown.
i cessing Group at the University of Erlangen-Nuremberg.81

兺 Esort共␩兲
␩=0
Its geometrical and acoustical details as well as the loud-
EC共i兲 = , 共29兲 speaker and microphone setups are given in Refs. 43 and 77.
MN−1
The simulated room has the dimensions 5.90⫻ 5.80
兺
␩=0
Esort共␩兲
⫻ 3.10 m 共w ⫻ l ⫻ h兲.
A WFS system with 48 loudspeakers placed equidistant
where 0 艋 EC共i兲艋 1. Equivalent definitions apply for the on a circle with a radius of RLS = 1.50 m and a wave field
transformed representations. The larger portion of the total analysis system consisting of 48 angular sampling positions
energy that is captured by the first i elements, the better the at a radius of Rmic = 0.75 m was simulated. The simulations
performance in terms of energy compaction. The energy of were performed for a reverberant environment with a plane
the elements E共m , n兲, Ē共␪ , ␪0兲, and Ĕ共␯ , ␯0兲 illustrate the dis- wave reflection factor of Rpw = 0.8 at the walls. Due to the
tribution of the energy in the different transformed domains, spatial aliasing frequency of the WFS system and the micro-
while EC共i兲, EC共i兲, and EC ˘ 共i兲 illustrate the ability of a par- phone array all signals were low-pass filtered to a bandwidth
ticular transformation to compact the energy in few coeffi- of 650 Hz to calculate the results shown.
cients. An indication for the suitability of circular harmonics is
given by calculating the right singular vectors of the room
2. Evaluation results transfer matrix R共␻兲共for the pressure microphones兲 using
The following illustrates the performance of active lis- the GSVD. Figure 7 shows the absolute value of the first
tening room compensation for WFS using the proposed eight right singular vectors for the reverberant case Rpw
WDAF approach for simulated acoustical environments. The = 0.8. The singular vectors have been sorted by their de-
main benefits of using simulated versus real acoustical envi- scending singular values and thus by their energy. It can be
ronments are that the parameters of the simulated environ- seen that they look very similar to the circular harmonics
ment can be changed easily in order to simulate different illustrated in Fig. 6. This result and the considerations in Sec.
scenarios and that practical aspects such as noise, transducer V C give justification for using the circular harmonics repre-
mismatch, and misplacement can be excluded for a first sentation as wave domain transformation.
proof of the WDAF concept.
Figure 8 shows the energy of the elements of the room
For an accurate simulation of the acoustic environment,
transfer matrix for the pressure microphones and its circular
a numerical simulation of wave propagation based on the
harmonics representation. Figure 8共a兲 shows E共m , n兲 for the
functional transformation method was used.78 The particular
implementation used 共wave2d兲 simulates the wave equation pressure microphones. The direct path from the loudspeakers
in two dimensions.79 This method requires no spatial dis- to the microphones can be seen clearly along the main diag-
cretization and thus allows one to place the virtual micro- onal but also the reflections of the loudspeaker wave fields at
phones and speakers at their exact spatial positions. This is the walls of the listening room. Figure 8共b兲 shows the energy
not possible when using methods which perform a spatial of the room transfer matrix represented in circular harmon-
discretization, such as, e.g., the finite element method.80 ics. As desired, the main diagonal elements represent a major
The geometry of the simulated acoustical environment is portion of the energy. The off-diagonal elements are a result
a simplification of a real environment at the multimedia labo- of the reverberant enclosure. Please note the different scales
ratory of the Multimedia Communications and Signal Pro- used for Figs. 8共a兲 and 8共b兲.
FIG. 9. Energy compaction performance EC共i兲 for the different representa-
tions of the room transfer matrix. The index i denotes the number of sorted
elements which capture most of the total energy.
˘
EC共i兲. The energy compaction of the GSVD is shown for
reference. It can be seen clearly that the circular harmonics
representation of the room transfer matrix compacts the en-
ergy much better than its pressure and plane wave represen-
tation. It can also be seen that the GSVD provides the opti-
mal transformation in this sense.
More simulation results for other values of the reflection
factor and also for a rectangular loudspeaker array geometry
have been reported in Ref. 77. Furthermore, measurements
have been conducted for the circular wave field synthesis/
analysis system described earlier. These measurements con-
firm the simulation results.
The presented results indicate that the circular harmon-
ics decomposition is not the optimal transformation for the
reverberant case. However, the results also show that this
transformation provides a quite reasonable approximation to
the optimal GSVD transformation, especially when com-
pared to the plane wave decomposition and pressure micro-
phone representations.
Further results of the proposed algorithm with respect to
the adaptation performance and the resulting wave field in-
side the listening area after compensation can be found in
Refs. 41 and 48. They reveal that the proposed algorithm
provides fast and stable adaptation even for nonstationary
virtual scenes and results in a reproduced wave field without
listening room reflections throughout the entire listening
area.
FIG. 8. Energy of the room transfer matrix of the signals captured by the VII. CONCLUSIONS
pressure microphones E共m , n兲 and in the circular harmonics domain
This contribution has derived a method for active listen-
Ĕ共␯ , ␯0兲.
ing room compensation for massive multichannel sound re-
production systems. It has been shown that a major challenge
Not only the performance of the circular harmonics in is the adaptation of the matrix of compensation filters when
terms of energy compaction toward the main diagonal is of straightforwardly applying multichannel adaptation algo-
interest, but also its performance to represent the MIMO sys- rithms 共like the X-RLS algorithm兲 to massive multichannel
tem by as few coefficients as possible. This can be measured reproduction scenarios. In addition to the computation of the
by the energy compaction of the different representations. compensation filters also the matrix of room impulse re-
Results are shown in Fig. 9. Figure 9 illustrates the energy sponses has to be estimated. The numerical expense for such
compaction according to Eq. 共29兲 for the room transfer ma- a straightforward implementation would be immense: For
trix in the pressure domain EC共i兲, in the plane wave decom- reproduction systems with even only a few ten loudspeakers,
posed domain EC共i兲, and in the circular harmonics domain the computation of many thousands of single adaptation
5
problems would be required for each time step. Furthermore, M. Poletti, “A unified theory of horizontal holographic sound systems,” J.
Audio Eng. Soc. 48, 1155–1182 共2000兲.
the convergence of the adaptation will be slow due to a high 6
F. Rumsey, Spatial Audio 共Focal Press, 2001兲.
correlation between the channels. 7
J. Meyer, “Spherical microphone arrays for 3D sound reproduction,” in
Suitable mathematical techniques can help to reduce the Audio Signal Processing for Next-Generation Multimedia Communication
numerical expense and to improve convergence through de- Systems, edited by Y. Huang and J. Benesty 共Kluwer Academic, Dor-
drecht, 2004兲.
coupling the adaptation problem. More specifically, the ma- 8
S. Moreau and J. Daniel, “3D sound field recording with higher order
trix of compensation filters, the matrix of room impulse re- ambisonics—Objective measurements and validation of a 4th order spheri-
sponses, and the free field propagation matrix can be jointly cal microphone,” in 120th Audio Engineering Society Convention, Paris,
diagonalized through a GSVD. This method has been called France, 2006.
9
M. Gerzon, “Ambisonics in multichannel broadcasting and video,” J.
eigenspace adaptive filtering because the diagonalization can
Acoust. Soc. Am. 33, 859–871 共1985兲.
be seen as a transformation into the eigenspace of the prob- 10
A. Berkhout, “A holographic approach to acoustic control,” J. Audio Eng.
lem. Although eigenspace adaptive filtering provides optimal Soc. 36, 977–995 共1988兲.
11
decoupling, there is a major drawback for the practical ap- D. de Vries, E. Start, and V. Valstar, “The wave field synthesis concept
applied to sound reinforcement: Restrictions and solutions,” in 96th Audio
plication: The GSVD matrices for the transformation into the Engineering Society Convention, Amsterdam, The Netherlands, February
eigenspace are data dependent, since they depend on the 1994.
12
time-varying acoustic room response characteristics. Calcu- J. Sonke, D. de Vries, and J. Labeeuw, “Variable acoustics by wave field
lating these matrices anew in each time step would create synthesis: A closer look at amplitude effects,” in 104th Audio Engineering
Society Convention, Amsterdam, The Netherlands, May 1998.
another numerical problem. 13
S. Spors, H. Teutsch, A. Kuntz, and R. Rabenstein, “Sound field synthe-
At this point, the situation is similar as for coding of sis,” in Audio Signal Processing for Next-Generation Multimedia Commu-
audio or image data. The optimal energy compaction and nication Systems, edited by Y. Huang and J. Benesty 共Kluwer Academic,
decorrelation could be achieved by the Karhunen-Loéve Dordrecht, 2004兲.
14
R. Rabenstein and S. Spors, “Wave field synthesis techniques for spatial
transform, but the correct transformation matrix is data de- sound reproduction,” in Topics in Acoustic Echo and Noise Control, edited
pendent and hence not known. Thus practical coders are re- by E. Haensler and G. Schmidt 共Springer, New York, 2006兲, Chap. 13, pp.
alized with a different transformation which does not depend 517–545.
15
on the input data. For a wide variety of problems in audio D. Griesinger, “Multichannel sound systems and their interaction with the
room,” in 16th International Conference on Audio, Acoustics, and Small
and video coding, the discrete cosine transformation has Places, Audio Engineering Society, 1998, pp. 159–173.
proven to approximate the properties of the Karhunen-Loève 16
M. Dewhirst, S. Zielinski, P. Jackson, and F. Rumsey, “Objective assess-
transform reasonably well. A similar route has been taken ment of spatial localization attributes of surround-sound reproduction sys-
here. The search for an approximation of the transformation tems,” in 118th Audio Engineering Society Convention, Barcelona, Spain,
May 2005.
into the eigenspace by GSVD has been motivated by physi- 17
B. Klehs and T. Sporer, “Wave field synthesis in the real world. 1. In the
cal intuition. Inspection of the eigenvectors of the GSVD living room,” in 114th Audio Engineering Society Convention, Amster-
from simulations and measurements has shown that they re- dam, The Netherlands, March 2003.
18
semble the circular harmonics decomposition of free field E. Völker, “To nearfield monitoring of multichannel reproduction—Is the
acoustics of the living room sufficient?,” in Proceedings of the Tonmeis-
acoustics. This notion led us to the idea that an approach in tertagung, Hannover, Germany, 1998.
which the optimal transformations into the eigenspace is re- 19
E. Völker, W. Teuber, and A. Bob, “5.1 in the living room—On acoustics
placed by a transformation into the space of circular harmon- of multichannel reproduction,” in Proceedings of the Tonmeistertagung,
ics, the so-called wave domain, may give results that are only Hannover, Germany, 2002.
20
E.-J. Voelker, “Home cinema surround sound—Acoustics and neighbour-
slightly suboptimal. hood,” in 100th Audio Engineering Society Convention, Copenhagen,
The implementation of the transformation into the wave Denmark, May 1996.
21
domain depends on the specific spatial reproduction/analysis T. Sporer and B. Klehs, “Wave field synthesis in the real world. 2. In the
technique. Details have been given for reproduction by wave movie theatre,” in 116th Audio Engineering Society Convention, Berlin,
Germany, May 2005.
field synthesis. It turned out that the decoupling properties of 22
P. Gauthier and A. Berry, “Sound-field reproduction in-room using opti-
the transformation into the wave domain are still quite good mal control techniques: Simulations in the frequency domain,” J. Acoust.
and close to the theoretical optimum of the GSVD. In sum- Soc. Am. 117, 662–678 共2005兲.
23
mary, it has been shown that active listening room compen- S. Bech, “Timbral aspects of reproduced sound in small rooms. I,” J.
Acoust. Soc. Am. 97, 1717–1726 共1995兲.
sation for massive multichannel sound reproduction is a trac- 24
S. Bech, “Timbral aspects of reproduced sound in small rooms. II,” J.
table problem. For spatial reproduction with wave field Acoust. Soc. Am. 99, 3539–3549 共1996兲.
25
synthesis, wave domain adaptive filtering provides a solution S. Bech, “Spatial aspects of reproduced sound in small rooms,” J. Acoust.
which is both numerically feasible and almost optimal. It Soc. Am. 103, 434–445 共1998兲.
26
T. Caulkins and O. Warusfel, “Characterization of the reverberant sound
remains to extend these investigations to other spatial repro- field emitted by a wave field synthesis driven loudspeaker array,” in 120th
duction techniques. Audio Engineering Society Convention, Paris, France, May 2006.
27
E. Corteel and R. Nicol, “Listening room compensation for wave field
1
E. Torick, “Highlights in the history of multichannel sound,” J. Audio Eng. synthesis. What can be done?,” in 23rd Audio Engineering Society Con-
Soc. 46, 27–31 共1998兲. vention, Copenhagen, Denmark, May 2003.
2 28
G. Steinke, “Surround sound—The new phase. An overview,” in 100th L. Fielder, “Practical limits for room equalization,” in 111th Audio Engi-
Audio Engineering, Society Convention, Copenhagen, Denmark, May neering Society Convention, New York, September 2001.
29
1996. L. Fielder, “Analysis of traditional and reverberation-reducing methods of
3
D. Griesinger, “Surround: The current technological situation,” in 108th room equalization,” J. Audio Eng. Soc. 51, 3–26 共2003兲.
30
Audio Engineering Society Convention, Paris, France, February 2000. P. Hatziantoniou and J. Mourjopoulos, “Errors in real-time room acoustics
4
S. Kiriakakis, P. Tsakalides, and T. Holman, “Surrounded by sound,” IEEE dereverberation,” J. Audio Eng. Soc. 52, 883–899 共2004兲.
31
Signal Process. Mag. 16, 55–66 共1999兲. J. Mourjopoulos, “Digital equalization of room acoustics,” J. Audio Eng.
Soc. 42, 884–900 共1994兲. 56
G. Golub and C. Loan, Matrix Computations 共The Johns Hopkins Univer-
32
F. Talantzis and D. Ward, “Multi-channel equalization in an acoustic re- sity Press, Baltimore, MD, 89兲.
57
verberant environment: Established of robustness measures,” in Institute P. Flikkma, “An algebraic theory of 3D sound synthesis with loudspeak-
of Acoustics Spring Conference, Salford, UK, March 2002. ers,” in 22nd International Conference on Virtual, Synthetic and Entertain-
33
F. Talantzis and D. Ward, “Robustness of multichannel equalization in an ment Audio, Audio Engineering Society, 2002.
58
acoustic reverberant environment,” J. Acoust. Soc. Am. 114, 833–841 E. Williams, Fourier Acoustics: Sound Radiation and Nearfield Acoustical
共2003兲. Holography 共Academic, New York, 1999兲.
34 59
S. Bharitkar, P. Hilmes, and C. Kyriakakis, “Sensitivity of multichannel A. Pierce, Acoustics. An Introduction to its Physical Principles and Appli-
room equalization to listener position,” in IEEE International Conference cations 共Acoustical Society of America, New York, 1991兲.
on Multimedia and Expo 共ICME兲, Baltimore, MD, 2003. 60
D. Blackstock, Fundamentals of Physical Acoustics 共Wiley, New York,
35
P. Nelson, F. Orduna-Bustamante, and H. Hamada, “Inverse filter design 2000兲.
61
and equalization zones in multichannel sound reproduction,” IEEE Trans. S. Spors, M. Renk, and R. Rabenstein, “Limiting effects of active room
Speech Audio Process. 3, 185–192 共1995兲. compensation using wave field synthesis,” in 118th Audio Engineering
36
T. Betlehem and T. Abhayapala, “Theory and design of sound field repro- Society Convention, Barcelona, Spain, 2005.
62
duction in reverberant rooms,” J. Acoust. Soc. Am. 117, 2100–2111 P. Morse and H. Feshbach, Methods of Theoretical Physics. Part I
共2005兲. 共McGraw-Hill, New York, 1953兲.
37 63
T. Betlehem and T. Abhayapala, “A modal approach to soundfield repro- M. Abramowitz and I. Stegun, Handbook of Mathematical Functions 共Do-
duction in reverberant rooms,” in IEEE International Conference on ver, New York, 1972兲.
Acoustics, Speech, and Signal Processing 共ICASSP兲, Philadelphia, PA, 64
E. Hulsebos, D. de Vries, and E. Bourdillat, “Improved microphone array
2005, Vol. III, pp. 289–292. configurations for auralization of sound fields by Wave Field Synthesis,”
38
A. Sontacchi and R. Hoeldrich, “Enhanced 3D sound field synthesis and in 110th Audio Engineering Society Convention, Amsterdam, The Nether-
reproduction system by compensating interfering reflections,” in Confer- lands, May 2001.
ence on Digital Audio Effects 共DAFX-00兲, Verona, Italy, 2000. 65
H. Teutsch, “Wavefield decomposition using microphone arrays and its
39
M. Poletti, “Three-dimensional surround sound systems based on spherical application to acoustic scene analysis,” Ph.D. thesis, University of
harmonics,” J. Audio Eng. Soc. 53, 1004–1025 共2005兲. Erlangen-Nuremberg 共2005兲, http://www.lnt.de/lms/publications 共last
40
S. Spors, H. Buchner, and R. Rabenstein, “A novel approach to active viewed on 16 April 2007兲.
66
listening room compensation for wave field synthesis using wave-domain T. Ajdler, L. Sbaiz, and M. Vetterli, “Plenacoustic function on the circle
adaptive filtering,” in IEEE International Conference on Acoustics, with application to HRTF interpolation,” in IEEE International Conference
Speech, and Signal Processing 共ICASSP兲, 共2004兲. on Acoustics, Speech, and Signal Processing 共ICASSP兲, Philadephia, PA
41
S. Spors, H. Buchner, and R. Rabenstein, “Adaptive listening room com- 共2005兲.
67
pensation for spatial audio systems,” in European Signal Processing Con- S. Brix, T. Sporer, and J. Plogsties, “CARROUSO—An European ap-
ference 共EUSIPCO兲共2004兲. proach to 3D-audio,” in 110th Audio Engineering Society Convention,
42
S. Spors, H. Buchner, and R. Rabenstein, “Efficient active listening room 2001.
68
compensation for wave field synthesis,” in 116th Audio Engineering So- E. Start, “Direct sound enhancement by wave field synthesis,” Ph.D. the-
ciety Convention, Berlin, Germany, 2004. sis, Delft University of Technology, Delft, The Netherlands, 1997.
43 69
S. Spors, “Active listening room compensation for spatial sound reproduc- E. Verheijen, “Sound reproduction by wave field synthesis,” Ph.D. thesis,
tion systems,” Ph.D. thesis, University of Erlangen-Nuremberg, 2006. Delft University of Technology, Delft, The Netherlands, 1997.
44 70
J. Lopez, A. Gonzalez, and L. Fuster, “Room compensation in wave field P. Vogel, “Application of wave field synthesis in room acoustics,” Ph.D.
synthesis by means of multichannel inversion,” in IEEE Workshop on thesis, Delft University of Technology, Delft, The Netherlands, 1993.
71
Applications of Signal Processing to Audio and Acoustics, New Paltz, NY, W. de Bruijn, “Application of wave field synthesis in videoconferencing,”
2005. Ph.D. thesis, Delft University of Technology, Delft, The Netherlands,
45
R. von Zon, E. Corteel, D. de Vries, and O. Warusfel, “Multi-actuator 2004.
panel 共map兲 loudspeakers: How to compensate for their mutual reflec- 72
E. Hulsebos, “Auralization using wave field synthesis,” Ph.D. thesis, Delft
tions,” in 116th Audio Engineering Society Convention, Berlin, Germany, University of Technology, Delft, The Netherlands, 2004.
73
2004. S. Spors, “An analytic secondary source selection criteria for wave field
46
P. Gauthier and A. Berry, “Adaptive wave field synthesis with independent synthesis,” in 33rd German Annual Conference on Acoustics 共DAGA兲,
radiation mode control for active sound field reproduction: Theory,” J. Stuttgart, Germany, March 2007.
Acoust. Soc. Am. 119, 2721–2737 共2006兲. 74
H. Buchner, S. Spors, and W. Kellermann, “Wave-domain adaptive filter-
47
M. Omura, M. Yada, H. Saruwatari, S. Kajita, K. Takeda, and F. Itakura, ing: Acoustic echo cancellation for full-duplex systems based on wave-
“Compensating of room acoustic transfer functions affected by change of field synthesis,” in IEEE International Conference on Acoustics, Speech,
room temperature,” in IEEE International Conference on Acoustics, and Signal Processing 共ICASSP兲, 2004.
Speech, and Signal Processing 共ICASSP兲, Phoenix, AZ, 1999. 75
H. Buchner, S. Spors, and W. Kellermann, “Wave-domain adaptive filter-
48
S. Petrausch, S. Spors, and R. Rabenstein, “Simulation and visualization ing for acoustic human-machine interfaces based on wavefield analysis
of room compensation for wave field synthesis with the functional trans- and synthesis,” in European Signal Processing Conference 共EUSIPCO兲,
formation method,” in 119th Audio Engineering Society Convention, New 2004.
76
York, 2005. H. Buchner, S. Spors, and W. Kellermann, “Full-duplex systems for sound
49
M. Sondhi, D. Morgan, and J. Hall, “Stereophonic acoustic echo field recording and auralization based on wave field synthesis,” in 116th
cancellation—An overview of the fundamental problem,” IEEE Signal Audio Engineering Society Convention, Berlin, Germany, 2004.
Process. Lett. 2, 148–151 共1995兲. 77
S. Spors and R. Rabenstein, “Evaluation of the circular harmonics decom-
50
J. Garas, Adaptive 3D Sound Systems 共Kluwer Academic, Dordrecht, position for WDAF-based active listening room compensation,” in 28th
2000兲. AES Conference: The Future of Audio Technology—Surround and Be-
51
A. Oppenheim and R. Schafer, Discrete-Time Signal Processing 共Prentice- yond, Audio Engineering Society, Pitea, Sweden, June/July 2006, pp.
Hall, Englewood Cliffs, NJ, 1999兲. 134–149.
52 78
M. Miyoshi and Y. Kaneda, “Inverse filtering of room acoustics,” IEEE S. Petrausch and R. Rabenstein, “Highly efficient simulation and visual-
Trans. Acoust., Speech, Signal Process. 36, 145–152 共1988兲. ization of acoustic wave fields with the functional transformation method,”
53
S. Haykin, Adaptive Filter Theory 共Prentice-Hall, Englewood Cliffs, NJ, in Simulation and Visualization, 共Otto von Guericke Universität, Magde-
1996兲. burg, 2005兲, pp. 279–290.
54 79
J. Benesty, D. Morgan, and M. Sondhi, “A better understanding and an “Wave2D: Wave field simulation in 2D,” http://www.lnt.de/~stepe/demo/
improved solution to the specific problems of stereophonic acoustic echo Wave2D 共last viewed on 16 April 2007兲.
cancellation,” IEEE Trans. Speech Audio Process. 6, 156–165 共1998兲. 80
J. Reddy, Introduction to the Finite Element Method 共McGraw Hill, New
55
M. Bouchard and S. Quednau, “Multichannel recursive-least-squares algo- York, 2005兲.
81
rithms and fast-transversal-filter algorithms for active noise control and “Multimedia communications and signal processing at the University of
sound reproduction systems,” IEEE Trans. Speech Audio Process. 8, 606– Erlangen-Nuremberg,” http://www.lnt.de/LMS 共last viewed on 16 April
618 共2000兲. 2007兲.
View publication stats

lnt2007 16

Uploaded by

Copyright:

Available Formats

lnt2007 16

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

lnt2007 16

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Active listening room compensation for massive multichannel sound

Article in The Journal of the Acoustical Society of America · August 2007

Sascha Spors Rudolf Rabenstein

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

I. INTRODUCTION lowing will introduce such massive multichannel reproduc-

Analogous definitions apply to the other signals wn共k兲, lm共k兲, ␰共ĉ,k兲 = 兺 ␭

⌽̂ww共k兲r̂共k兲 = ⌽̂wl共k兲, 共12兲

A. Wave field synthesis

i cessing Group at the University of Erlangen-Nuremberg.81

View publication stats

You might also like