Aalborg Universitet

Low frequency sound field enhancement system for rectangular rooms, using multiple
loudspeakers

Celestinos, Adrian

Publication date:
2007

Document Version
Publisher's PDF, also known as Version of record

Link to publication from Aalborg University

Citation for published version (APA):

Celestinos, A. (2007). Low frequency sound field enhancement system for rectangular rooms, using multiple
loudspeakers.

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

- Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
- You may not further distribute the material or use it for any profit-making activity or commercial gain
- You may freely distribute the URL identifying the publication in the public portal -
Take down policy
If you believe that this document breaches copyright please contact us at [email protected] providing details, and we will remove access to
the work immediately and investigate your claim.

Downloaded from vbn.aau.dk on: October 02, 2022

Low frequency sound field enhancement
system for rectangular rooms,
using multiple loudspeakers

Ph.D. thesis by

Adrian Celestinos
Low frequency sound field enhancement
system for rectangular rooms,
using multiple loudspeakers

by

Adrian Celestinos

Ph.D. thesis

Acoustics, Department of Electronic Systems

Aalborg University
DK-9220 Aalborg East, Denmark
Aalborg, December 2006
Preface

This thesis is submitted to the Faculty of Engineering and Science, Aalborg University
in partial fulfillment of the requirements for the Ph.D. degree. This thesis is based on
the work conducted at Acoustics, Department of Electronic Systems, Aalborg University
during the period of September 2003 – December 2006, under the supervision of Assoc.
Prof. Sofus Birkedal Nielsen. The thesis is conformed as a plurality of three convention
preprints and a manuscript submitted to the Journal of the Audio Engineering Society.
The papers are referred in the text in bold letters as Paper A, Paper B, Paper C and
Paper D. In addition to the papers, the thesis consists of a general introduction, an
overview of the work performed, an extra Chapter E and two appendixes.

During this period I was appointed as a full-time Ph.D. fellowship holder financed by the
The Faculty of Engineering, Science and Medicine, at Aalborg University.

I wish to thank all the people who supported my work at Acoustics, in particular to my
supervisor Sofus Birkedal Nielsen for his endless patience and his fruitful guidance. I wish
to thank all my colleagues at Acoustics for the valuable comments and rich discussions. I
wish to thank all the staff for maintaining an extraordinary work place, to the secretariat
for their invaluable help and to the technical staff, Claus Vestergaard Skipper and Peter
Dissing for their professional assistance.

Finally I wish to thank my parents for their unyielding support no matter the distance.

Adrian Celestinos
Aalborg, December 2006

i
ii
Contents

Preface i

Contents iii

Abstract ix

Resumé (abstract in Danish) xi

Introduction 1

1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1.1 Room simulation model . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1.2 Low frequency analysis in rooms . . . . . . . . . . . . . . . . . . . . 4

1.1.3 Low frequency room equalization . . . . . . . . . . . . . . . . . . . . 6

1.1.4 Uniform sound field at low frequencies . . . . . . . . . . . . . . . . . 7

1.1.5 Controlled Acoustically Bass System (CABS) . . . . . . . . . . . . . 8

1.2 General Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.4 Summary of the Papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.5 List of Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

iii
Paper A: Multi-source low frequency room simulation using finite difference
time domain approximations 15

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 SIMULATION OF SOUND SOURCES IN A ROOM USING FDTD . . . . 18

2.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3 Sound Source Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 EVALUATION OF THE SOUND FIELD IN A ROOM . . . . . . . . . . . 20

3.1 Sound Pressure Level Distribution . . . . . . . . . . . . . . . . . . . 20

3.2 Optimization of Used Memory . . . . . . . . . . . . . . . . . . . . . 21

3.3 Visualization in Time Domain . . . . . . . . . . . . . . . . . . . . . 21

3.4 Acquisition of Impulse Response . . . . . . . . . . . . . . . . . . . . 21

4 RESULTS AND VALIDATION . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.2 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.3 Comparison of simulations and real measurements . . . . . . . . . . 26

5 DISCUSSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

6 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

7 ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

Paper B: Optimizing placement and equalization of multiple low frequency

loudspeakers in rooms 29

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2 SIMULATION PROGRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3 QUANTITATIVE PARAMETERS . . . . . . . . . . . . . . . . . . . . . . . 32

3.1 Magnitude Deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

iv
 3.2 Spatial Deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4 ANALYSIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 Room Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

4.2 Evaluation of Loudspeaker Configurations . . . . . . . . . . . . . . . 34

4.3 Positioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5 OPTIMIZATION STRATEGIES . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Multiple Point Equalization . . . . . . . . . . . . . . . . . . . . . . . 37

5.2 Equalization of Acoustic Radiation Power . . . . . . . . . . . . . . . 38

5.3 Optimization by Modifying Delay and Phase . . . . . . . . . . . . . 40

6 RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

7 MEASUREMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

8 DISCUSSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

9 CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Paper C: Low frequency sound field enhancement system for rectangular

rooms using multiple low frequency loudspeakers 49

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2 ANALYSIS OF THE SOUND FIELD ON THREE RECTANGULAR ROOMS 52

2.1 Room description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.2 Sound Field Room Simulations . . . . . . . . . . . . . . . . . . . . . 53

2.3 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3 THE EQUALIZATION SYSTEM . . . . . . . . . . . . . . . . . . . . . . . . 56

3.1 Creation of a Plane Wave . . . . . . . . . . . . . . . . . . . . . . . . 58

3.2 Removing the Reflection from the Back Wall . . . . . . . . . . . . . 59

3.3 Optimal Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . 63

v
 4 RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4.1 Simulation of the Equalization System . . . . . . . . . . . . . . . . . 63

4.2 Measurement of the Equalization System . . . . . . . . . . . . . . . 63

4.3 Evaluation of the Equalization System . . . . . . . . . . . . . . . . . 64

5 DISCUSSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

6 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

Paper D: Controlled Acoustically Bass System (CABS), A method

to achieve uniform sound field distribution at low frequencies in rectan-
gular rooms 69

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

2 LOW FREQUENCY SOUND IN RECTANGULAR ROOMS . . . . . . . . 72

2.1 The building up of a standing wave in the time domain . . . . . . . 72

2.2 Simulations on a three dimensional Virtual Room . . . . . . . . . . 74

2.3 Time and frequency analysis . . . . . . . . . . . . . . . . . . . . . . 75

2.4 Quantification Parameters . . . . . . . . . . . . . . . . . . . . . . . . 76

2.5 Traditional one point equalization . . . . . . . . . . . . . . . . . . . 80

3 UNIFORM SOUND PRESSURE DISTRIBUTION IN THE ROOM . . . . 81

3.1 Construction of a Plane Wave . . . . . . . . . . . . . . . . . . . . . . 82

3.2 Removing the Reflection from the Back Wall . . . . . . . . . . . . . 83

3.3 Controlled Acoustically Bass System (CABS) . . . . . . . . . . . . . 83

3.4 Simulation of CABS in the Virtual Room . . . . . . . . . . . . . . . 84

4 IMPLEMENTATION AND MEASUREMENT OF CABS IN REAL ROOMS 87

5 RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6 DISCUSSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

7 CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

vi
 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

Chapter E: CABS .2.2 In An Irregular Room 95

E.1 Low Frequency Sound Fields in An Irregular Room . . . . . . . . . . . . . . 97

E.1.1 Partitioning of the room . . . . . . . . . . . . . . . . . . . . . . . . . 97

E.1.2 Simulation of loudspeakers in an irregular room . . . . . . . . . . . . 99

E.1.3 Simulation of CABS .2.2 in an irregular room . . . . . . . . . . . . . 103

E.1.4 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . 105

Bibliography 107

Appendix I 109

I.1 Sound Field Room Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . 111

I.1.1 Discretization of the wave equation . . . . . . . . . . . . . . . . . . . 111

I.1.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

I.1.3 Sound source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

I.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

I.2.1 Wave dispersion errors . . . . . . . . . . . . . . . . . . . . . . . . . . 120

I.2.2 Transfer function measurement . . . . . . . . . . . . . . . . . . . . . 120

I.2.3 The walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

I.3 Graphical User Interface (GUI) of The Sound Field Room Simulator . . . . 122

Appendix II 129

II.1 Normal Modes of Vibration in Rooms . . . . . . . . . . . . . . . . . . . . . 131

vii
viii
Abstract

Loudspeakers are the last link in the sound reproduction chain, and they are typically
placed in small or medium size rooms. When low frequency sound is radiated by a loud-
speaker the sound distribution along the room presents large deviations. This is due to the
multiple reflection of sound at the rigid walls of the room. The reflected waves from the
rigid walls might meet the propagating waves from the loudspeaker itself at some places in
constructive phase and at some places in opposite phase. This may cause level differences
of up to 20 dB in the output level of the loudspeaker at the listener position. These devi-
ations change depending of the listener position and the loudspeaker placement. Some of
these deviations are associated with the standing waves, resonances or anti resonances of
the room.

The thesis contains an introduction and overview of the work done, four papers and one
additional chapter. The first paper is concentrated on a simulation model based on the
finite-difference time-domain method (FDTD) to predict the low frequency behavior of
loudspeakers in rectangular rooms. The second paper includes the simulation of different
configurations of subwoofers in a standard listening room and some equalization tech-
niques are revised. The third paper includes simulations and measurements of the typical
low frequency sound reproduction systems in rooms. Here a new method later named
Controlled Acoustically Bass System (CABS) is introduced. In the last two papers of the
thesis, the CABS system is simulated and implemented in two standard listening rooms.
The performance of the system is evaluated by measurements both in an IEC and a ITU
standard listening room. In the last paper the analysis in the time domain and frequency
domain of the low frequency sound in rooms is presented. The typical one subwoofer .1 vs
CABS are compared in simulations. The extra Chapter E concentrates on simulations
of an irregular room with the FDTD method and the performance of CABS in that room.
Two appendixes are added to the thesis; a more detailed description of the room simu-
lation program is presented in appendix I, and a brief description of the room modes of
vibration theory is outlined in appendix II.

The scope of the thesis and the research concerns itself with the performance of loud-
speakers in rooms at low frequencies. The research concentrates on the improvement of
the sound distribution in the room produced by loudspeakers at low frequencies. The

ix
work focuses on seeing the problem as an acoustic problem in the time domain. The result
of this work is the introduction of the CABS system as a novel and effective solution.
The thesis discusses the implementation and the performance of CABS in two standard
listening rooms and the simulations and implementation of working setups with CABS.

The methodology employed in this investigation was first to have a deep understanding of
the problem. This was accomplished by analyzing the behavior of low frequency sound in
rooms in the time domain. This could not be accomplished without investigating a room
simulation model based in the time domain. The outcome of this work is a intuitive and
effective system solution named Controlled Acoustically Bass System (CABS) that utilizes
loudspeakers equidistantly placed at the front wall of the room and extra loudspeakers in
anti phase at the rear wall. By using the acoustic cancellation principle, a digital pure
delay and the proper gain in the rear loudspeakers, the rear reflection is canceled, giving
a uniform sound distribution in the whole room approx. below 120 Hz. The working
range depends on the room size and the number of loudspeakers used. The smaller the
room the best CABS performs. The novelty of this solution is that differently from the
advanced room correction systems that only work at a restricted listening position, CABS
acquires even sound level distribution at low frequencies in the whole room with simple
signal processing. The CABS system works in the time domain therefore it performs as
well for transient signals as for steady signals.

By using CABS more even sound level distribution is obtained along the room. The effect
of the room resonances in the reproduced sound has been decreased remarkably. Compared
to the traditional one subwoofer setup that has typically standard deviations close to ±
6 dB and differences in spectral magnitude up to more than 20 dB. By using CABS in
an IEC standard listening room the spatial standard deviations are reduced to ± 1.6 dB
and the spectral standard deviations to ± 2.1 dB. In a ITU standard listening room the
spatial standard deviations were reduced to ±1.3 dB and the spectral standard deviations
to ± 2.1 dB. CABS can be integrated to stereo or multichannel systems. Preliminary
results of simulations of CABS in an irregular room have shown promising results however
measurements need to be carried out to arrive at objective conclusions.

x
Resumé (abstract in Danish)

Højttalere anvendes til reproduktion af lyd i rum, der er her ofte tale om små og mellem-
store rum. Især ved lave frekvenser vil der være meget store hørbare variationer i det
frembragte lydtryk. Disse variationer skyldes i det væsentlige refleksioner fra rummets
flader (vægge, gulv og loft). Reflekterede lydbølger vil blande sig med den oprindelig
udsendte lydbølge fra højttaleren og vil nogle steder i rummet være i fase med hinanden
og i andre steder i modfase, resulterende i variationer i lydtryk på op til 20 dB i forskel-
lige lyttepositioner. Disse store variationer i lydtryk afhænger af såvel lyttepositionen
som højttalerens placering i rummet. Nogle af variationerne skyldes stående bølger, res-
onansfrekvenser eller anti-resonansfrekvenser, men der vil være store variationer ved alle
frekvenser.

Denne PhD afhandling består af en introduktion der tjener til at give et overblik over
det udførte arbejde, der i det væsentlige er dokumenteret i 4 publikationer, bestående
af 3 afholdte konference indlæg og 1 tidsskriftartikel (indleveret, men endnu ikke pub-
liceret). Det første konferenceindlæg koncentrerer sig om konstruktionen af et simu-
leringsværktøj, et program baseret på finite difference time domain method (FDTD).
Dette simuleringsværktøj har været nødvendigt for at skaffe viden om, hvorledes lave
frekvenser fra højttalere opfører sig i rektangulære rum, der er de mest normale lytterum.
Det andet konferenceindlæg indeholder simuleringer af forskellige konfigurationer af sub-
woofere (lavfrekvente højttalere) i et standard lytterum og nogle equaliserings teknikker er
afprøvet. Det tredje konferenceindlæg inkluderer simuleringer og målinger af lavfrekvent
lydgengivelse i rum, og ny metode kaldet: Controlled Acoustically Bass System (CABS)
er introduceret. I det tredje konferenceindlæg og tidsskriftartiklen er CABS verificeret
ved simulering og målinger i 2 standard lytterum, henholdsvis i et IEC standard lytterum
og et IUT multikanals lytterum ved Akustik på Aalborg Universitet. Resultaterne ved
lave frekvenser er dokumenteret i såvel tidsdomæne som frekvensdomæne i de 2 rum. Den
typiske anvendelse at n subwoofer (.1) er sammenlignet med CABS, der anvender flere
højttalere. Afhandlingen indeholder 3 appendikser, hvor en mere detaljeret præsentation
af det konstruerede rumsimulerings program er beskrevet i appendiks I. Simuleringen af
CABS i ikke regulære rum er præsenteret i appendiks II, og en kort beskrivelse af rum-
modes og vibrationsteori er beskrevet i appendiks III.

xi
Formålet og omfanget af dette PhD projekt har været at udføre forskning, der kan tjene til
en bedre forståelse af hvorledes samspillet er mellem højttalere og rum ved lave frekvenser,
hvor der er fundamentale problemer. Udgangspunktet for analysen og løsningen er an-
derledes end mere traditionelle metoder idet der her er set på problemet som et akustisk
problem i tidsdomæne. Resultatet af denne analyse har ført til en effektiv løsning, der
introduceres som et nyt system, af nemheds grunde kaldet CABS. Afhandlingen belyser
implementeringen og anvendelsen af CABS i to lytterum, først ved simuleringer med det
til formålet udviklede værktøj, og efterfølgende ved målinger i de to rum.

Den metodik, der er anvendt i dette arbejde, har været først at få en dyb forståelse
for selve problemets art. Dette er sket ved at analysere hvorledes lavfrekvent lyd forde-
les i rum i tidsdomæne. For at kunne foretage disse komplekse analyser har det været
nødvendigt først at udvikle en rumsimulerings- model baseret på tidsdomæne analyse.
En intuitiv løsning (CABS) opstod, som anvender en ækvidistant placering af højttalere
ved frontvæggen og som det nye anvender ekstra højttalere i modfase ved bagvæggen.
Med denne metode sker der en akustisk minimering af refleksionen fra bagvæggen når
baghøjttalerne fødes med det oprindelige signal og med den rigtige forsinkelse og ampli-
tude. Denne metode giver en ensartet fordeling af lydtrykket i hele rummet under ca.
120 Hz, afhængig af rummets størrelse, antallet af højttaler og deres placering.

Ved at anvende CABS får man et mere ensartet lydtryk i rummet. Sammenlignet med den
traditionelle anvendelse af en subwoofer, der har en typisk spatial standardafvigelse tæt
på ±6 dB og en spektral standardafvigelse på ±7.5 dB, men med udsving på over 20 dB.
Ved at anvende CABS i samme IEC lytterum er den spatiale standardafvigelse reduceret
til ±1.6 dB og den spektrale standardafvigelse til ±2.1 dB. I et ITU standard lytterum
er den spatiale standardafvigelse reduceret til ±1.3 dB og den spektrale standardafvigelse
reduceret til ±2.1 dB.

Resultatet af dette PhD–arbejde er bl.a. introduktionen at en nyt og effektivt system

kaldet CABS, der i modsætning til traditionelle løsninger giver et ensartet lydtryk i et
rektangulært rum ved lave frekvenser, vel og mærke i hele rummet. Effekten af rummets
resonansfrekvenser i den reproducerede lyd er reduceret væsentligt. I modsætning til andre
avancerede rumkorrektions-systemer, anvender CABS simple former for signalbehandling.
CABS fungere i tidsdomæne og er derfor virksom for såvel transiente signaler som for
stationære signaler. Systemet kan integreres med traditionel Stereo systemer såvel som
multikanals systemer. Foreløbige simuleringer af CABS i ikke rektangulære rum har vist
lovende resultater, men bør udforskes nærmere.

xii
Introduction

Since the advent of stereophony the production of music signals in high fidelity has gained
the interest of researchers, professionals and a great amount of enthusiasts. With the
arrival of the digital signal processing technology the popularity of new reproduction
formats as multichannel surround sound has increased reasonably. From home theaters to
concert hall arenas it is possible to experience sound through full-range loudspeakers or
subwoofers dedicated to play back frequencies from 30 Hz to 100 Hz. The main quality of
these reproduction formats is that they give to the listener a sense of space. Formats such
as the traditional stereo or the multichannel surround sound are often called spatial sound
reproduction systems. They are based on more than one loudspeaker which are typically
placed in a living room. Other solutions that make use of Binaural technologies (also
known as 3D sound) are utilized to give the correct spatial sensation to the listener 16 .
The restriction of these solutions is the need of headphones and that the reproduction of
binaural signals by loudspeakers can be achieved only for a restricted listening position and
in very damped rooms. In music sound reproduction there are commonly two scenarios,
in the first of which the music material is produced by acoustical instruments and voice,
the second situation may be one in which the acoustic program is converted to electrical
signals, recorded or mixed, amplified and reproduced through loudspeakers. This work
will consider the second scenario where the program is already recorded and produced into
an audio reproduction format.

In an ideal situation, for example in stereo sound reproduction, only a person positioned at
the “sweet spot” will benefit by the qualities of this reproduction format, if the loudspeak-
ers are set correctly in the room. A more realistic situation is that the loudspeakers are
placed “more or less” symmetrically in the room and there will normally be more listeners
sitting in different positions. This situation is common for example in movie theaters
where a large listening area has to be covered. In these cases stereo reproduction may not
be sufficient and another kind of format as for example the surround sound 5.1 or wave
field synthesis 6 that make use of more channels and loudspeakers might be suitable. As
shown by Blauert 4 , Wightman and Kistler 27 and other authors the predominant cue for
human sound localization at low frequencies is the inter–aural time difference (ITD) there-
fore localization at those frequencies is quite poor. Human sound localization in listening
rooms worsens at frequencies where the wavelength is much longer than the distance be-

1
Introduction Low Frequency Enhancement System for Rooms

tween the ears. This has been the foundation to be able to mix the low frequency content
of all the channels in the surround sound formats or in the stereo format to the extra .1
subwoofer when the loudspeakers are not capable of reproducing very low frequencies 1,9 .

In the situation with the regular stereo setup with two full range loudspeakers in a room,
especially at low frequencies the perceived sound will be different depending on the position
of the listener in the room. These problems appear typically to loudspeakers placed in
small or medium size rooms and in some cases medium size halls. Modification of the
output level of the loudspeaker at the listeners ears occurs due to the multiple reflections
of the sound at the walls and different objects in the sound path, as for example furniture,
openings etc. These variations change depending on the position of the listener and the
loudspeaker in the room. This is often problematic to control by acoustic means due to the
long wavelengths involved. Mid and high frequencies involving relatively small wavelengths
can be attenuated by absorptive materials but when producing wavelengths in the range
of 10 to 3 meters (34 Hz – 114 Hz) the acoustic solutions become impractical. These
reflections will normally produce variations from 20 dB to 30 dB in the sound distribution
level along the room. Some of these variations at certain frequencies are caused by the
standing waves or resonance frequencies in the room. If these resonances are strongly
excited they will cause large differences in the sound pressure distribution in the room
and also structural vibration of the enclosure. What can be done acoustically is to build
the room with the best possible room-mode distribution, that can be achieved by choosing
the right dimensions ratio. Nevertheless the room, with its physical properties and the
placement of the loudspeaker, will highly influence the output level of the loudspeaker
measured at the different listening positions 11,2,3,20 .

In these cases there are two main identified problems. The first problem is for example
if a bass tone is played back through the loudspeakers it might be perceived very loud
at a determined listening position in the room, yet exactly the same sound would be
barely heard by another listener sitting in another position in the room. The second
problem is related to the variations in level at different frequencies also known as “Spectral
Coloration”. For example, in a fixed listening position, some notes of a scale or a chord
included in the recording, typically performed by instruments like an electric bass or an
pipe organ will not be heard as loud as some other notes that will be perceived louder, or
“booming”.

The strategy followed in this work was to get a deep understanding of the problem and
assimilate it as a physical phenomenon. To do that it was decided to investigate and
implement a room simulation model based on an element method. This approach led
to the use of the acoustic cancellation principle. Differently from other solutions, results
have shown that the sound pressure level distribution along the room can be improved
significantly from having deviations in the sound field below 100 Hz of typically ±12 dB
to ±3 dB.

2
Low Frequency Enhancement System for Rooms Introduction

1.1 Overview

In the following sections an overview of the project is outlined. Relevant aspects of the
work are presented, the room simulation model, the low frequency analysis in rooms, room
equalization systems, how to achieve uniform sound field distribution at low frequencies,
and finally a brief description of the outcome of this research.

1.1.1 Room simulation model

Generally the problem of low frequency sound in rooms has been widely analyzed by
solving the linear lossless wave equation for the propagation of sound in fluids (Kinsler
et al. 12 ). In this fashion a rectangular room with rigid walls is assumed where the normal
component of the particle velocity gets very close to zero. These well–known formulations
called modal decomposition techniques are mainly based on the complex sound pressure in
steady-state (Morse and Ingard 17 ).

Differently in this research work it was decided to inspect the problem by a model in
which it was possible to track down the sound pressure in the room as a function of time.
Other methods based on geometrical acoustics such as the Mirror Image model or Ray
Tracing, are no longer sufficient when the wavelength is comparable with the dimensions
of the room 5 . In this work a computer simulation program based on an element method
was developed and described in the Paper A of this thesis. The simulation program is
based on the finite–difference time–domain method (FDTD) 24,7,23,22,13 . This model also
begins by solving the linear lossless wave equation but in addition it applies the relation
between the particle velocity and the acoustic pressure. This second equation is known
as the linear inviscid force equation valid for acoustic processes of small amplitude 12 .
The main diference with other methods is that both equations (lossless wave equation
and force equation) calculate particle velocity and pressure as a function of time. In this
fashion these two equations are utilized to compute the acoustic pressure produced by a
number of sound sources in the entire enclosure (see Appendix I.1 for more details about
the discretization processes and solution of the mathematical expressions of the method).
With this program written in MATLAB the sound field produced by multiple loudspeakers
in a rectangular room can be simulated. Moreover irregular rooms can also be modeled
as presented in Chapter E of this thesis.

The advantage of the FDTD method is that it works in the time domain and therefore the
pressure amplitude and the particle velocity is always available for analysis or visualization
purposes. Besides that the impulse response of the transfer function at desired positions in
the room can be obtained. With other methods like the finite element method (FEM) or
the boundary element method (BEM) it is possible to obtain the complex sound pressure
level at the boundaries or within defined regions in the room but the time history not
always is available. Typically in these methods each discrete frequency has to be calculated

3
Introduction Low Frequency Enhancement System for Rooms

Mic

Figure 1.1: A room of 100 m3 volume with dimensions W=5.3 m, L=7 m and H=2.7 m
suggested by the standard BS.6840-13 8 , loudspeaker and microphone position.

separately. If the analysis is needed in a wide range of frequencies the simulation time
increases considerably.

The main disadvantage of the FDTD method is the limited frequency range where accurate
results can be achieved. It is well known that, at frequencies where the number of cells
per wavelength is lower than ten, wave dispersion errors occur. This limitation implies
the use of a huge amount of memory for simulating large spaces (see Fig.I.14 in Appendix
I.1). Another limitation is that in reality the wall reflections are frequency dependent
and this is difficult to implement on a FDTD scheme (Botteldooren 7 ). However as shown
by Olesen 18 assumptions can be made and good results can be achieved in simulations
of relatively small spaces at low frequencies. Although the FDTD method is a “brute
force” approach and has a limited frequency range for accurate simulations, it has been
decided to make use of it after the results obtained in Paper A. Simulations show good
agreement with measurements in an IEC standard listening room.

1.1.2 Low frequency analysis in rooms

In this work the understanding of the physical phenomena at low frequencies when a room
is excited by sound sources is presented. When acoustic energy is confined in an enclosed
space, as for example a room, a number of phenomena occur. If the wavelength of the
radiated sound is much longer than the largest dimension of the room, there is a frequency
below which the only propagating wave form that can exist is a plane wave 12 .

As explained in the introduction the multiple reflections with the walls of these front waves
produced by the loudspeaker will form large deviations in sound pressure level within the
room. Some of these patterns are very distinct and well–known as standing waves, reso-

4
Low Frequency Enhancement System for Rooms Introduction

Figure 1.2: Sequence of snap shots in time from left to right of the instantaneous pressure
produced by the loudspeaker in the room shown in Fig. 1.1.

nance frequencies or the normal modes of vibration of the enclosure when the dimensions
correspond to multiples of half of the wavelength, and when the dimensions correspond
to an odd multiple of one quarter of the wavelength the associated frequencies are called
anti–resonances or anti–modal frequencies. In Paper D, with the use of the simulation
program described in Paper A, the analysis in the time domain of the formation of these
patterns is given (see Fig. 1 and Fig. 2 in Paper D). In addition, the derivation of the
normal modes of vibration in an enclosure from the conventional solution to the wave
equation is presented in Appendix II.1.

The deviations in the sound pressure level distribution within the room will appear not
only at the modal or anti-modal frequencies but also at frequencies where the wavelengths
are long enough to be comparable to the dimensions of the room. Generally, for example,
there will always be a minimum in sound pressure level at a distance corresponding to one
quarter of a wavelength from a reflecting wall, since the reflected wave and the arriving
wave will always be in opposite phase.

To give an idea of the problem simulations of a loudspeaker placed in the room sketched
in Figure 1.1 are presented in Figure 1.2. In this sequence of pictures of the pressure
amplitude at discrete times, the interaction of the waves with the walls can be observed
forming the deviations in the sound field distribution. The building up of these deviations
can be originated even with quite short transients sounds. This can be inferred by ob-
serving the Cumulative Spectral Decay (CSD) introduced in Paper C and Paper D and
shown in Figure 1.3. The CSD performs a joint analysis in frequency and time. The CSD
is calculated over an impulse response. From this observation it is clear that in small and
medium size rooms the resonance frequencies keep ringing in time longer than the others.
In more damped rooms those frequencies do not necessarily keep ringing in time. However
there are still spectral deviations in the early part of the impulse response (see Fig. 4 and
Fig. 7 in Paper C).

5
Introduction Low Frequency Enhancement System for Rooms

dB dB

−20 −20

−25 −25

−30 −30

−35 −35

−40 −40

0 0
0.2 0.2
0.4 0.4
0.6 0.6
0.8 100 Hz 0.8 100 Hz
1s 1s
10 10

Figure 1.3: Cumulative spectral decay (CSD). Left, Loudspeaker measured in anechoic
conditions. Right, The same loudspeaker measured at the “Mic” position in the room
shown in Figure 1.1.

When low frequency sound is produced by a loudspeaker in a room there are two main
problems. The first is related to the deviations in sound pressure level as a function
of listening position and the second is related to the spectral deviations at a specific
listening position in the room. In Paper D a new parameter named the Mean Sound
Field Deviation (MSFD) is introduced. With the MSFD the two main problems are
quantified. The first problem associated with the deviations in sound level at one specific
frequency at different places of the room is represented by the Spatial Deviation SD. The
second problem, represented by the Magnitude Deviation MD, is identified at individual
listening positions and describes the deviations in sound level within a range of frequencies
between 20 Hz to 100 Hz. In this fashion the MSFD is calculated along a listening area
defined by microphone positions equally spaced in the room.

In Paper D another quantifier that operates in the time domain is presented. The pa-
rameter originally called in German “Deutlichkeit” D (translated to english as Definition)
mainly used in Room Acoustics is utilized to quantify the influence of the room at a num-
ber of microphone positions. Definition gives a criterion of the ratio of energy between
the early part of the impulse response (0 – 50 ms) and the remaining part 25 .

1.1.3 Low frequency room equalization

After having written a reliable simulation program a number of approaches have been
studied. As learned from the literature in order to deal with this problem several ap-
proaches have been investigated by a number of authors. Over the last three decades
Groh 11 , Allison 2 , Ballagh 3 (among others) have based their approaches on finding the op-
timum placement of the loudspeakers in the room. More recently Welti and Devantier 26

6
Low Frequency Enhancement System for Rooms Introduction

have based their investigation on the use of multiple subwoofers with different configu-
rations in the room. Another approach by Abildgaard 19 is based on the control of the
acoustic radiation power of the loudspeaker in a room. Large amount of research has been
carried out with the approach of designing the correct electrical filters commonly called
Room Correction systems to compensate for the negative effect of the room. Mäkivirta
et al. 15 has conducted research on the approach called Modal Equalization and Elliott
and Wilson 10 have worked on the technique named Multiple Point Equalization. These
filters are called “electrical filters” because they are included in a dedicated digital signal
processor (DSP) or apparatus connected before the loudspeaker and the amplifier. These
systems need a microphone to measure the output level of the loudspeaker in the room
or what is often called the “response” of the loudspeaker at a listening position. This
transfer function includes the effects of the loudspeaker and the room at this specific po-
sition. Then by means of adaptive digital techniques the effects are compensated for at
this specific listening position in the room.

As shown by Welti and Devantier 26 the addition of more loudspeakers carrying the same
signal and positioned at the mid points of each wall improves the problem related with the
spatial variations in sound pressure level. But on the other hand the deviations in level
along frequency increase considerably (see Fig. 4 in Paper B). This was verified in the
Paper B of this work where six configurations of subwoofers are simulated utilizing the
simulation program described in the Paper A. In fact in one of these configurations of
four subwoofers at the mid points of the walls a decrement of the overall power is observed
(see configuration LP4 in Fig. 4 in Paper B). After these results the configuration with less
spatial deviations was chosen in order to implement three different methods for optimizing
the low frequency sound field. The multiple point equalization and the equalization of the
acoustic radiation power near the loudspeakers are simulated. After implementing these
two types of equalization one can summarize that these methods partially alleviate the
problem. In a study conducted by Santillán et al. 21 an equalization system based on the
simulation of a plane wave traveling in a small room as in a free field is presented. In
this aproach 20 loudspeakers baffled in one of the walls and another 20 in the opposite
wall were utilized. In order to aquire the correct filter for each loudspeaker the transfer
function from each loudspeaker to each of the number of points at the listening planes are
needed (about 2880 impulse response measurements).

1.1.4 Uniform sound field at low frequencies

The main goal in this work is to improve the low frequency sound field produced by a
loudspeaker in a rectangular room. As explained before the cause of the large deviations in
sound pressure level is the reflections at the walls, and therefore to avoid the problem one
should cancel those reflections. But as shown in Figure 1.2 there are multiple reflections,
hence an elegant way of simplifying the problem is to cancel only the first reflection because
then only the direct sound would exist. This can be done by forming plane waves in one

7
Introduction Low Frequency Enhancement System for Rooms

end of the room traveling in only one direction towards the opposite wall. Where the
sound will be canceled by using extra loudspeakers with a delayed version of the signal
in anti–phase and with the correct gain. This is verified in Paper D by the simulation
of two loudspeakers in a rectangular room, positioned equidistantly in the front wall (see
Fig. 26. in Paper D). Then the rear wall is rendered as an opening instead of a rigid wall.
One can verify that the sound pressure level distribution is even over a large part of the
room and that this can be achieved up to frequencies where the distribution of the front
loudspeakers enables a plane wave front to be built. In the case where the loudspeakers
are not able to build a plane wave more loudspeakers would be needed. However at those
frequencies it is possible to attenuate these interferences by optimizing the placement of
the loudspeakers (see Sections 3.1 and 3.4 in Paper D).

1.1.5 Controlled Acoustically Bass System (CABS)

In contrast to the traditional equalization methods a different approach is presented in

this work. The name Controlled Acoustically Bass System (CABS) is first introduced in
Paper D. The system consists of the use of loudspeakers at the front wall of the room
and extra loudspeakers at the opposite wall in order to cancel the back wall reflection.
These extra loudspeakers are fed with the same signal as the front loudspeakers including
a delay according to the traveling distance in the direction of the plane wave to the back
wall, see the block diagram in Fig. 28 of Paper D. In addition to the delay the gain of
the extra loudspeakers has to be adjusted due to the traveling distance and the damping
characteristics of the room. Together with CABS the following notation is introduced:

.F.B

F stands for the number of low frequency loudspeakers (e.g. subwoofers) positioned at
the front wall.

B stands for the number of low frequency loudspeakers (e.g. subwoofers) positioned at
the back wall.

1.2 General Results

The system CABS .2.2 was implemented in a PC using a real–time signal–processing soft-
ware and an AD/DA multichannel converter. The parameters of the system were adjusted
empirically to achieve the best performance. The system was measured in two standard
listening rooms, the IEC Room and the ITU Room both at the Acoustics Laboratory at

8
Low Frequency Enhancement System for Rooms Introduction

Aalborg University. The general results are presented in Table 1.1 and Table 1.2 respec-
tively. In the IEC Room the spatial deviations of the sound field at 25 positions improved
by 6 dB, and the spectral deviations were enhanced by 8.6 dB. In the ITU Room the spa-
tial deviations improved 5.8 dB and the spectral deviations have shown an improvement
of 6.4 dB. For more details of the outcome refer to Section 4.3 in Paper C and Section 5
in Paper D.

Table 1.1: Results of improvement in measurements of CABS .2.2 at the listening area in the IEC
Room from 20 Hz to 100 Hz.
MSFD Definition
IEC Room SD (dB) MD (dB) D
.2.0 ± 4.6 ± 6.4 66.8 %
CABS .2.2 ± 1.6 ± 2.1 92.4 %
Improvement 6 dB 8.6 dB 25.6 %

Table 1.2: Results of improvement in measurements of CABS .2.2 at the listening area in the ITU
Room from 20 Hz to 90 Hz.
MSFD Definition
ITU Room SD (dB) MD (dB) D
.2.0 ± 4.2 ± 5.3 64.3 %
CABS .2.2 ± 1.3 ± 2.1 89.4 %
Improvement 5.8 dB 6.4 dB 25.1 %

1.3 Discussion

To summarize, in this work a deep understanding of the behavior of low frequency sound
in rooms has been acquired. First the investigation of a room simulation model based
on the wave equation in the time domain has been conducted. Some of the well known
equalization methods have been simulated. Finally, the outcome of this work is an effective
system solution named CABS that, independently from the traditional solutions, tackles
the problem in an effective manner.

The advantages of CABS are:

• More even sound level distribution below 100 Hz is achieved throughout the room.

• The effect of the room resonances in the reproduced sound is decreased considerably.

• Only simple signal processing is needed.

9
Introduction Low Frequency Enhancement System for Rooms

• The system works in the time domain and therefore is sufficient for transient signals
and for steady signals.

• Once adjusted, it works independently of the program material that is reproduced.

• The system can be integrated into stereo or multichannel systems. Informal listening
of music signals with CABS .2.2 integrated into full–range loudspeakers in a stereo
setup have shown remarkable results.

• Preliminary results of simulations of CABS .2.2 in irregular rooms have shown

promising results however measurements need to be carried out to get to objective
conclusions (see Chapter E).

• The working frequency range of the system extends as the room size decreases.
The smaller the room the fewer loudspeakers needed. CABS .2.2 could be applied
to sound reproduction systems in small enclosures (e.g. automobiles, small music
studios).

The main drawbacks are:

• More loudspeakers and power amplifiers are needed. All the loudspeakers have to
have the same phase and frequency response, or individual equalization is needed.

• The optimal placement of the loudspeakers might not be ideal for comercial aplica-
tions of the system.

• A detriment of 3 dB in the output power exists as a consequence of the acoustic

removal of one of the walls.

• Wide rooms might need more loudspeakers. Nevertheless the working frequency
range can be improved by optimizing the placement of the loudspeakers (see Fig. 39
in Section 3.4 of Paper D).

• In a stereo setup or a multichannel setup the low frequency content of all channels
has to be collected to only one channel, the CABS .2.2.

• If the temperature changes drastically in the room the delay must be re-adjusted.

• The system must include a low–pass filter to attenuate frequencies above the working
range.

• The system has been simulated and measured in empty rooms and no furniture has
been included. Furniture might decrease the effectiveness of the system.

10
Low Frequency Enhancement System for Rooms Introduction

Further investigations:

Further investigations may be conducted on the implementation of CABS on dedicated

signal processing hardware that could automatically adjust the parameters to give the best
performance. Measurements of CABS in existing living rooms with furniture, openings
and listeners could be conducted as further research on this project. In this work only
informal listening tests have been performed using CABS .2.2 integrated into a stereo
setup. Further investigations may be directed towards conducting listening experiments
in order to subjectively compare CABS with the standard formats. Another subject of
interest is the feasibility of integrating CABS into other spatial reproduction formats as
for example 3D sound.

11
Introduction Low Frequency Enhancement System for Rooms

1.4 Summary of the Papers

Paper A

Multi-source low frequency room simulation using finite difference time do-
main approximations

In this paper a simulation model written in MATLAB for the study of low frequencies in
audio reproduction such as ordinary stereo to multi–channel surround setups is described.
Simulations of multiple loudspeakers in a rectangular room are carried out to evaluate and
visualize their coupling with the room. Three kind of transfer functions are described.
First by using a Gaussian Pulse, second by using the MLS method and third by using a
near field impulse response of an existing loudspeaker. Two cases are simulated, first a
closed–box loudspeaker and second, two closed–box loudspeakers positioned in a stereo
setup. The simulations are compared to measurements in the existing room showing good
agreement.

Paper B

Optimizing placement and equalization of multiple low frequency loudspeakers

in rooms

A brief analysis of the characteristics of the room modes in rectangular rooms is given. Six
configurations of subwoofers are simulated utilizing the simulation program described in
Paper A. Three methods of equalization are simulated utilizing one of the configurations
with four subwoofers: The so–called multiple point equalization, the equalization of the
acoustic radiation power near the loudspeakers and the new method later named CABS.
The last method is implemented and measured in the IEC standard listening room at
Aalborg University.

Paper C

Low frequency sound field enhancement system for rectangular rooms using
multiple low frequency loudspeakers

Simulations of common sound–reproduction systems utilized in three different rooms are

presented. The chosen rooms are an IEC standard listening room, a ITU multichannel
listening room and a small concert hall. The description of the new method later called
CABS is given as well as results of simulations of the system on individual frequencies. The

12
Low Frequency Enhancement System for Rooms Introduction

new method is simulated in the three rooms. Results of the performance of CABS after
measurements in the actual rooms are presented. A discussion about the performance in
the three rooms is given.

Paper D

Controlled Acoustically Bass System (CABS), A Method to Achieve Uniform

Sound Field Distribution at Low Frequencies inside Rectangular Rooms

A detailed analysis in the time domain of low frequency sound in rooms is presented.
Simulations of a typical subwoofer in a rectangular room are given. A new parameter
called the mean sound field deviation (MSFD), utilized to quantify the deviations of the
sound field in a defined area, is introduced. Simulations of a room with an opening
instead of the back wall and two loudspeakers positioned at the front wall are presented.
The new method, Controlled Acoustically Bass System or CABS is formally introduced
and explained thoroughly. Results of the performance in simulations and measurements
of CABS both in an IEC and a ITU standard listening room are presented as well as
discussions and conclusion.

13
Introduction Low Frequency Enhancement System for Rooms

1.5 List of Publications

Paper A
“Multi-source low frequency room simulation using finite difference time
domain approximations,” A. Celestinos and S. B. Nielsen, presented at the 117th
Convention of the Audio Engineering Society, Journal of the Audio Engineering So-
ciety (Abstracts), vol. 53 pp. 105–106 (January/February 2005), convention preprint
6264.

Paper B
“Optimizing placement and equalization of multiple low frequency loud-
speakers in rooms,” A. Celestinos and S. B. Nielsen, presented at the 119th Con-
vention of the Audio Engineering Society, Journal of the Audio Engineering Society
(Abstracts), vol. 53 p. 1206 (December 2005), convention preprint 6545.

Paper C
“Low frequency sound field enhancement system for rectangular rooms
using multiple low frequency loudspeakers,” A. Celestinos and S. B. Nielsen,
presented at the 120th Convention of the Audio Engineering Society, Journal of
the Audio Engineering Society (Abstracts), vol. 54 p. 1206 (July/August 2006),
convention preprint 6688.

Paper D
“Controlled Acoustically Bass System (CABS), A Method to Achieve
Uniform Sound Field Distribution at Low Frequencies inside Rectangular
Rooms,” A. Celestinos and S. B. Nielsen, submitted to the Journal of the Audio
Engineering Society, (December 2006).

14
Paper A
Paper A Low Frequency Enhancement System for Rooms

16
 Multi-source low frequency room simulation using finite
difference time domain approximations
Adrian Celestinos a Sofus Birkedal Nielsen a
a Acoustics, Department of Electronic Systems, Aalborg University, DK-9220 Aalborg East, Denmark

Abstract

The sound level distribution generated by loudspeakers placed in a room can be simulated using numerical methods.
The purpose of this paper is to present an application based on finite-difference time-domain approximations (FDTD)
for the study of low frequencies in audio reproduction such as ordinary stereo to multi- channel surround setups. A
rectangular room is simulated by using a discrete model in time and space. This technique has been used extensively
and gives good performance at low frequencies. The impulse response can be obtained in addition to the sound level
distribution. Simulation of multiple loudspeakers in a room can be achieved to evaluate and visualize their coupling
with the room. A high frequency resolution can be obtained for auralization purpose.

1. INTRODUCTION cal acoustic approximations where the wavelength

is smaller than the room dimensions. The finite-
When a loudspeaker is placed in a rectangular room difference time-domain (FDTD) method has been
a number of problems arise. Modification of the re- used with great success to model electromagnetic
sponse of the loudspeaker at the listening position problems. In acoustics FDTD has shown good per-
occurs due to the strong influence of the room and formance to approximate the low frequency room
the position of the loudspeaker. The combination behavior. This method has been well described in
loudspeaker-room acts as a coupled system where early studies by Botteldooren in [6]. By using this
the room typically dominates by its distinct normal approximations the calculations are made directly
modes. When placing more than one loudspeaker in the time domain. The sound wave equation is dis-
in the room some of the attenuation produced by cretized both in time and space and the inclusion of
the room is less severe but still the room has a multiple sound sources in the room is possible since
strong influence at the different listening positions. the finite time and space are always available. Re-
In order to deal with this problem equalization cently other methods as the finite-element method
techniques have been investigated by several au- (FEM) and the boundary-element method (BEM)
thors as in [1],[2] or [3]. In connection to that a have been used extensively to simulate enclosures.
robust tool to simulate the low frequency behav- These approximations work only in the frequency
ior of multiple sound sources placed in rectangular domain, direct translation to a direct time domain
rooms is needed. During the last years two main formulation can be seen in [7] or [8].
approximations using numerical methods have been
developed which are the Ray tracing and the Image In this paper the FDTD method has been chosen
source method found in [4] and [5]. Such meth- to simulate a rectangular room exited by multiple
ods are no longer adequate to simulate frequencies sound sources. The estimation of the sound level dis-
below 100 Hz because they are based on geometri- tribution at low frequencies is calculated. The in-
clusion of loudspeakers assuming omni–directional
Paper A Low Frequency Enhancement System for Rooms

compact sources is implemented inside the room. point between two neighbour time/space points [12].
Since the particle velocity is always available, sound
power and intensity can be estimated for equaliza- After the derivation in time and space of Eq. (1)
tion purposes. Moreover a band limited impulse re- (force equation), the three components of the parti-
sponse of the room can be acquired. cle velocity are determined at positions:

ux(x± h ,yh,zh)
2. SIMULATION OF SOUND SOURCES 2

IN A ROOM USING FDTD uy(xh,y± h ,zh) (3)

2

uz(xh,yh,z± h )
In this section a description of the FDTD method is 2

presented as well as important aspects for the sim-

and at intermediate time t = (t + 21 )k by the follow-
ulation process which are, stability, the boundary
ing equations:
conditions and the modeling of the sound sources.
k k k
uxx+ h ,y,z (t + ) = uxx+ h ,y,z (t − ) −
2.1. Method
2 2 2 2 hρ0
h i
× px+h,y,z (t) − px,y,z (t) ,
Typically the FDTD method utilizes two coupled
first order differential equations. Since this method k k k
uyx,y+ h ,z (t + ) = uyx,y+ h ,z (t − ) −
works in the time domain it computes the derivative 2 2 2 2 hρ0
and linearized form of these two equations in the h i (4)
time domain. This is done by means of the central × px,y+h,z (t) − px,y,z (t) ,
finite difference [9]. k k k
uzx,y,z+ h (t + ) = uzx,y,z+ h (t − ) −
2 2 2 2 hρ0
2.1.1. Discretization of The Wave Equation h i
The first equation is the linear inviscid force equa- × px,y,z+h (t) − px,y,z (t) ,
tion valid for acoustic processes of small amplitude
where the acoustical pressure p and the particle ve- Similarly, from Eq. (2) (continuity equation), the
locity u are related as: acoustic pressure can be derived in time an space by:
∂~u
∇p = −ρ0 (1) px,y,z (t + k) = px,y,z (t)
∂t
where ρ0 is the density of the transmission media in c2 ρ0 k h x k k i
− ux+ h ,y,z (t + ) − uxx− h ,y,z (t + )
kg/m3 . The second equation is the linear continuity h 2 2 2 2
equation 2
c ρ0 k yh k k i (5)
− ux,y+ h ,z (t + ) − uyx,y− h ,z (t + )
h 2 2 2 2
1 ∂p
∇ · ~u = − (2) 2
c ρ0 k zh k k i
c2 ρ0 ∂t − ux,y,z+ h (t + ) − uzx,y,z− h (t + ) ,
h 2 2 2 2
where c is the wave propagation speed in the media
[10]. These are the set of equations that are used to cal-
culate particle velocity and acoustical pressure in
The typical formulation of the FDTD approxima- an alternate manner.
tion uses a Cartesian staggered grid [11], in which
pressure and particle velocity are the unknown In Fig. 1 an example of an enclosure can be seen
quantities. The acoustical pressure is determined where the layout of the grid for the calculation of
at the grid points (xδx, yδy, zδz), at time t = δt. the components of the particle velocity and acous-
In this paper δx = δy = δz = h that is the spatial tic pressure points in two dimensions is shown. The
discretization step and δt = k that is the time step. circular points represent acoustical pressure while
Both equations can be sampled in time and space squares are particle velocity component points in x
using the sampling rates k1 Hz and h1 m−1 . The dis- direction and stars are particle velocity components
cretization is done by means of finding the central in y direction. As it can be observed there are no

18
 # ∂p$ ∂ux ∂uy ∂uz
# c2 ρ0 k x $ = −c2 ρ0 + + . (8)
k) (t
=+ = p−x,y,z
c2 ρ0(t)
− u(t
+ ) (t + )
x k k
k − − ∂t ∂x ∂y ∂z
px,y,z (tp+
x,y,z p k) (t)
x,y,z h ux x+ 2 ,y,z hh 2(tu+ k
2),y,z
x+ h
x− h
x2 (t + x−
,y,z
u k h
2 ) 2 ,y,z 2
# c2and # from Eq. (8)2 the acoustical $ pressure $ is derived with
ρ0 k y y + k ) − uyk
k h (t (t + k
)
2
− c ρh0 k u−
y
x,y+hh
(tu+x,y+ ) − u
2 2 ,z x,y− h 2 (t + x,y−
2 ) 2 ,z (10)
h 2 (10)
2 ,z 2 ,z
2
# 2
# After the derivation
$ # k and$ linearization in the time domain both equations are
$1
z c ρ0 k uz k
− c ρh0 k u− hp h (t + ) −hu(tz + k ) − uzk
sampled2 h (t(t)
+ k(t +
) c2 ρ0hand
inx,y,z−
2time 2)
xspace using
+ k2 )the
− usampling rates ) k Hz and h m . From
1 −1
x,y,z+ x,y,z
2 (t2+
x,y,z+ 2 =
k) px,y,z
x,y,z− 2 − h2 ux+ h ,y,z
(t x
x− h
,y,z
(t + k
2
Eq.(7) the resulting# set of 2
equations for the 2
components $ of the particle velocity
Low Frequency Enhancement System arefor Roomsasc2 ρ0 k y
written y Paper A
− h ux,y+ h ,z (t + 2 ) − ux,y− h ,z (t + 2 )
k k
(10)
2 2
where the
whereacoustical pressure pressure
the acoustical is determined at the grid at
is determined points
the grid(xδx,points
yδy,
2
#(xδx, yδy, zδz)
zδz) $
at timeatt time
= δt and δt =
t = δx and = δz
δy δx == δyh=that
δz = is the spatial
h that discretization
is the spatial − cstep ρ0 k
discretizationand
u z
step
h
expressed
(t +and
k
) − u zin Eq. (6)
h (t
# + has
k
) to be held, more $ generally
h k x,y,z+ 2 k x,y,z− 2
δt = k δt
that is the time
= k that is the step.
1.2 mtime step.
Wall u x
h
x+ 2 ,y,z
(t + 2 ) = u x 2
h for
x+ 2 ,y,z
(t
a − )
three
2 − k 2
×
dimensional
hρ0 p x+h,y,z (t) − p
rectangular x,y,z (t)
grid , [9].
# $
In Eq. (9) The three 1.1components of the particle velocityuare y determined k at y
r
In Eq. (9) The three components of the particle velocity
h (t +are ) determined
= ux,y+ h ,z (tat− k2 ) − hρ 1k × p1x,y+h,z1(t) − px,y,z (t) , (9) (6)
times t = (t + 12 )k and1positions where the acoustical pressure
x,y+ 2 ,z is2determined 2 at≤
cδt the1/grid points
0 +
(xδx, +yδy,2zδz)
times t = (t + 2 )k and positions δx2 #δy 2 δz $
at time t = δt and δxu= z δy = δz =k h that z is the spatial
(t − k2 ) −discretization step and
h (t + 2 ) = u hρ0 × px,y,z+h (t) − px,y,z (t) ,
k
δt = k that is the time step.
x,y,z+ 2 x,y,z+ h 2
0.9 In this paper the sampling frequency f s was decided
to be 8 kHz, and the time step k = 1/f s. Never-
ux(x± h ,yh,zh) , uy(xh,y± hIn Eq.
,zh)
y
, u(9)
z The three
(xh,yh,z± h )
. components
(11)of the particle velocity are determined at
theless the program can be set to
find the minimum
2
ux(x± h ,yh,zh)
times
, 2 ut(xh,y± 1 , 2uz
= (t h+,zh)
2 )k and
andpositions
from Eq.
(xh,yh,z±
(11) pressure is derived with
h . (8) the acoustical
2 2 2) time step before it gets unstable.
0.7
Wall

Wall
2 # $
c2 ρ0 k
2 px,y,z (t + k) = px,y,z (t) −2.2. h uxx+ h ,y,z (t
Boundary + k2 ) − uxx− h ,y,z (t + k2 )
Conditions
0.5 2 2
ux(x± h ,yh,zh) , uy(xh,y± h ,zh) , 2uz(xh,yh,z±
# h . (11) $
2 2 c ρ0 k y 2) y
−Taking
h uthe
x,y+example
h
,z
(t + k
2 )
of− u
Fig.
x,y−1 h and
,z
(t +assuming
k
2 ) (10)
a right
2 2
ux ux hand
2
#rigid wall at the boundary of the $ room the
0.3
uy uy 2 − c ρh0 k uzx,y,z+ h (t + k2 ) − uzx,y,z− h (t + k2 )
component of2 the particle velocity2 in the x direc-
tion can not be calculated with Eq. (4) because the
ux p puxx
term px+h,y,z (t) is unknown. To solve this problem
0.1 y u
u uy
uy where the acoustical pressure is determinedfinite-difference
an asymmetric at the grid pointsapproximation
(xδx, yδy, zδz) for
0 at time t = δt and δx = δy =the
Wall = h that
δz space is the spatial
derivative is discretization
implemented as step and
in [11].
p 0.1 pp 0.3 0.5 0.7 0.9 1 m
δt = k that is the time step.
Fig. 1. Example of a 1m×1.20m enclosure. Circles are pres- Since the component of the particle velocity in the
sure points, stars are particle velocity inInthe
Eq. (9) The three
y direction, and components of the
x direction ux particle velocity
represents are determined part
the perpendicular at of
squares are particle velocity in the times t = (t + 12 )k and positions
x direction. the particle velocity to the wall, it is assumed that
the acoustic pressure p at the wall can be expressed
pressure points at the boundaries. In this manner by the product of the component of the particle ve-
the components of the particle velocity in for exam- locity in the x direction ux and the impedance Z of
ux(x± h ,yh,zh) , uy(xh,y± h ,zh) , uz(xh,yh,z± h ) .
ple the x direction are calculated at intermediate 2 that wall [12].
2 This manner2 an estimate of(11)
the ab-
pressure points as well as at intermediate time steps. sorption coefficient of the walls α can be introduced
The advantage of using this grid is that is easy to to calculate Z as
define the boundaries and it only requires two val- 2 √
ues of the acoustic pressure and particle velocity to 1+ 1−α
Z = ρ0 c √ . (7)
be stored in each grid cell. 1− 1−α
After these assumptions the new version of Eq. (4)
2.1.2. Cell Size for the component of the particle velocity at the
A fundamental constraint for the simulation method walls is introduced and for example for ux[0.9+ h ,y,z]
2
is the choice of the size cell. The frequency range of in Fig. 1 the boundary equation is defined as:
interest before aliasing and the accurate wave prop- ρ0 h
agation is given by the cell size. The cell size must be k −Z k
ux[0.9+ h ,y,z] (t + )= k
ρ0 h
ux[0.9+ h ,y,z] (t − )
much less than the smallest wave length for which 2 2 k +Z 2 2
accurate results are needed. Reasonable results can (8)
2
be achieved by using from five to ten cells per wave- + ρ0 h p [0.9,y,z] (t)
length [9]. In this paper a cell size of 10 cm has been k +Z
chosen since this cell size corresponds to 15 of 1a wave-
length therefore it is expected to have accurate
1 re- 2.3. Sound Source Model
sults below 600Hz.
1
1 1
The loudspeakers are modeled as typical closed-box
2.1.3. Stability loudspeakers with volume velocity function of time
After the cell size has been chosen, the time step has occupying a small volume (h × h × h) inside the
to be set. In order to have accurate wave propagation room. At low frequencies, where the wavelength of
and to minimized grid dispersion errors the relation sound in the air is much longer than the physical

19
Paper A Low Frequency Enhancement System for Rooms

Fig. 2. Listening test room, dimensions shape and measurement setup, adapted from [14].

dimensions of the loudspeaker, it propagates the sions, length 7.80 m; width 4.12 m; height 2.77 m;
sound in spherical waves radiating outwards uni- the mean reverberation time T60 is 0.47 s. The floor
formly in all directions [13]. is wooden and the walls are covered with special
panels that can be removed or moved to different
When the loudspeaker is driven by a sinusoidal sig- positions. The ceiling is curved in the corners cov-
nal it is modelled according to ered with special plaster panels (see Fig. 2). The
panels from the walls have been removed as well as
pxs ,ys ,zs (t) = A · sin(ω(t)) (9) the carpet that normally covers most of the floor.
A = SD ωu (10) The test room has been measured and simulated
early by Cherek and Langvad in [15] as well as by
where A is the volume acceleration, u is the parti- Krarup in [12].
cle velocity, SD is the effective area of the radiating
surface and ω is the angular frequency. In this man- An horizontal layer of the room at a height of 1.2 m
ner one or more than one compact sources can be has been chosen to calculate the sound pressure level
included in the model either in a pressure point or (SPL) distribution. The simulation time was set to
particle velocity using the volume velocity of the de- 1 second taking in to account that the reverberation
sired loudspeaker. In addition the loudspeaker can time of the test room simulated is less than 1 second.
be modelled as a membrane moving in a desire di- In case of a simulation of a more reverberant room
rection using some of the points of the components the simulation time should be grater than the mean
of the particle velocity. reverberation time.

3. EVALUATION OF THE SOUND FIELD 3.1. Sound Pressure Level Distribution

IN A ROOM
Since the root mean square (RMS) value of any
3.0.1. The Test Room signal is proportional to its energy content and
For the purpose of this paper the standard listening therefore is one of the most important and most of-
room at Acoustics, Aalborg University has been ten used measures of amplitude it has been decided
chosen to be simulated since this room has been to calculate this value over the area of interest in
well studied. The room has the following dimen- the room. As it is mention in section 2.1.1 two time

20
Low Frequency Enhancement System for Rooms Paper A

Fig. 4. Sequence of images at ascending times (from left to

right) of the computed pressure amplitude produced by a
loudspeaker close to the walls of the test room.

3.2. Optimization of Used Memory

A huge amount of memory is needed if one wants

to keep the pressure amplitude at all discrete times
all over the room. To optimize the use of memory
only two time steps of pressure amplitude and par-
ticle velocity all over the room are stored together
with one extra matrix for the average sound pressure
level. The averaged sound pressure level will only be
calculated in the horizontal plane of interest. Nev-
Fig. 3. Sound pressure level distribution resulting form the
simulation of four loudspeakers reproducing a sinusoidal fre- ertheless some other (virtual) microphones can be
quency of 65 Hz. set up all over the room in order to pick up the cal-
culated pressure amplitude at the discrete times at
steps of the pressure points and particle velocity of any desired position in the enclosure.
the whole room are needed to determine the acous-
tic pressure in the grid positions. Eq. (11) is used
to calculate the RMS value of the pressure over the 3.3. Visualization in Time Domain
area of interest where T is the relevant period over
which the averaging takes place and p is the instan- A very useful advantage of the FDTD method is the
taneous pressure [16]. visualization aspect. Since it runs in the time domain
the pressure amplitude in a desired area of the room
v can be observed at any discrete time. An animation
u ZT movie composed by indexed images of the pressure
u
u1
prms =t p2 (t) · dt (11) amplitude along the discrete simulation times can
T be obtained by the simulation program (see Fig. 4).
0

For this calculation an extra matrix is loaded to be

used as an accumulator of the result of the squared 3.4. Acquisition of Impulse Response
summation of the sound pressures for each time
step. From this pressure matrix the SPL distri- Two methods are considered to obtain a band lim-
bution over the chosen layer in the room can be ited transfer function impulse response from a num-
obtained (see Fig. 3). The SPL distribution is cal- ber of sound sources to specific locations in the room.
culated in the room according to: The first one is performed by reproducing a finite
length Gaussian pulse and picking up the impulse in
p the room at the desired position. The second method
LpSP L = 20Log10 (12) is implemented by reproducing a maximum length
po
sequence (MLS) and record the signal at the desired
where p is the sound pressure being computed and position in the room. In the next two sections both
po is the reference sound pressure being 20 µ Pa. methods will be described.

21
Paper A Low Frequency Enhancement System for Rooms

Fig. 5. Block diagram of the simulation program.

3.4.1. Gaussian Pulse to 2N − 1 being chosen according to how reverber-

The Gaussian pulse used in the simulation is defined ant is the room to simulate. The length of the input
by Eq. (13) and it is shown in Fig. 6. This kind of signal is two times the length of the MLS signal in
pulse has the characteristic of having a limited flat order to stabilized the filter. The input signal is low
frequency response. The cut off frequency is defined pass filtered to avoid aliasing from the simulation
by σ in Eq. (14) where ω = 2πf and f is the -3 dB itself. The cut off frequency is chosen according to
cut off frequency of the sound source. the frequency range of interest. The MLS method
has been extensively studied, details of the theory
1 −(t−t0 )2
pxs ,ys ,zs (t) = sin(t − t 0 )e σ2 (13) can be found in [17], [18] and other authors. An anti
σ2 aliasing filter at 2 kHz is implemented to filter the
2 recorded signal. The cross correlation between the
σ= (14) MLS input signal and the recorded sequence is cal-
ω
culated in order to obtain the impulse response. In
In Fig. 7 the recorded impulse response at position Fig. 8 the recorded impulse response and frequency
1 together with the frequency response is presented.
The test room was exited by one source at position
(1.2 m,2.0 m,1.2 m) refer to Fig. 2 to see the micro-
phone and loudspeaker position in the test room. As
it can be observed the influence of the room is severe
and some of the room modes are revealed.

3.4.2. MLS method

As well as the Gaussian pulse method an MLS se-
quence is reproduced by the sound source. An MLS
signal is very useful because it has high energy
content and it is very suitable for different impulse
lengths. It generates uniform probability density, its
spectrum is absolutely flat, and the most important
is that its periodic auto correlation is a unit sample
sequence. The motivation to use the MLS method
is that the MLS method unlike the Gaussian exci-
tation has high energy content in all frequencies.

The signal is implemented by generation of pseudo Fig. 6. Upper plot shows the frequency response of the Gaus-
random numbers. The sampling frequency is set to sian pulse, lower plot shows the Gaussian pulse in the time
8 kHz. The length of the impulse response was set domain. The cut off frequency has been set to 600 Hz.

22
Low Frequency Enhancement System for Rooms Paper A

Fig. 7. Upper curve is the transfer function of the room plus Fig. 8. Upper curve is the transfer function from the loud-
the loudspeaker calculated at microphone position 1, lower speaker to the the room calculated at virtual microphone
plot is the impulse response, method: Gaussian pulse. position 1. Lower curve is the impulse response, method:
MLS.
response of the simulated room is shown. The room lution of 30 degrees. The magnitude of the averaged
was exited by one sound source located at coordi- impulse response has been normalized with the near
nates (1.2 m, 2.0 m, 1.2 m) (see Fig. 2). field measurement in the frequency domain in order
to have the same gain as the near field measurement
3.4.3. Including a real loudspeaker impulse response (see Fig. 9).
In order to get a more accurate result in the simula-
tion the transfer function of a real loudspeaker can After obtaining the transfer functions of the loud-
be introduced in the model. This procedure is valid speaker they have been convolved with the MLS in-
just when the frequency range of interest is below put signal, then low pass filtered and reproduced
500 Hz since a loudspeaker behaves almost omni at the sound source position. The same procedure
directional within that range. The test loudspeaker explained in section 3.4.2 has been applied to ob-
(A) is a closed-box type with a volume of 12 litre tain the impulse response with an MLS signal (see
and 35 cm height 23 cm width and 23.5 cm depth, Fig. 5). The lower graph in Fig. 10 shows the im-
it has a woofer of 16.5 cm diameter, and a 1.9 cm pulse response recorded at position 1. The test room
dome tweeter. The test loudspeaker (A) has been is exited by one sound source located at position
measured in anechoic conditions in order to obtain (1.2 m,2.0 m,1.2 m) including the loudspeaker trans-
two impulse responses to be tested in the simulation fer function of the near field measurement, at the
program. upper graph in the same figure the frequency re-
sponse is presented, at the same graph the doted
For the first measurement it has been decided to line includes the averaged transfer function of the
measure the near field impulse response as close loudspeaker instead of the near field measurement.
as possible to the cone of the test loudspeaker (see
Fig. 9), more details about near field measurements 3.4.4. Implementation of a fine grid for auralization
can be found in [19]. By reducing the size of the cell the frequency range
of interest is increased. In order to obtain a fair fre-
The second impulse response is an average in the quency range for auralization purposes a fine grid of
frequency domain of measurements of the response 4 cm has been implemented. At the sound source po-
of the loudspeaker in the horizontal plane and ver- sition a finite length signal of music (speech) can be
tical plane at 1 m from the membrane with a reso- used as an input signal. The signal could be recorded

23
Paper A Low Frequency Enhancement System for Rooms

4. RESULTS AND VALIDATION

4.1. Measurements

In order to validate the results of the simulations

two set of measurements have been carried out in
the test room.

4.1.1. Sound Pressure Distribution Measurement

To validate the sound pressure distribution a rect-
angular area has been delimited in the test room. A
rectangular grid of 10x9 points separated each other
by 20 cm at a height of 1.20 m in the test room has
been set up (see Fig. 2). The panels that cover the
walls and the carpet from the floor were removed.
The test loudspeaker (A) was set in the room at
position (1.2 m,2.0 m,1.2 m). The sound pressure
level has been measured at each grid point. The
Fig. 9. Frequency response and time response of test loud- measurement was carried out driving the test loud-
speaker (A), thin line is the near field measurement, thick speaker with 65 Hz by a sine generator with 1.0 V
line averaged in horizontal and vertical planes.
RMS amplified by a reference stereo amplifier. As
mention before the test loudspeaker is a closed box
type with a volume of 12 litre and 35 cm height, 23
cm width and 23.5 cm depth it has a 16.5 cm diam-
eter bass driver unit and a 1.9 cm, polyamide dome
tweeter. The test loudspeaker (A) was pointing to
the grid area (as seen in Fig. 2).

At each point in the grid a pressure microphone con-

nected to a pre-amplifier was located with a preci-
sion of ± 1.5 cm. The output from the pre-amplifier
was connected to a measuring amplifier. The aver-
age time from the measuring amplifier was set to 1
second to obtain an RMS voltage value. The system
was calibrated by a piston-phone to 124 dB SPL at
250 Hz.

4.1.2. Impulse Response Measurement

To validate the acquisition of the impulse response
by the MLS method the impulse response at three
Fig. 10. Transfer function of the room and loudspeaker (A) microphone positions in the room were measured.
calculated at position 1, continuous line near field impulse re-
The microphones were located at 1.20 m height.
sponse is included, dashed line the averaged impulse response
of the real loudspeaker is included. In the time response only The room was exited first by the test loudspeaker
the averaged impulse response is included, method: MLS. (A) an afterwards by both test loudspeaker (A)
and test loudspeaker (B), the test loudspeaker (B)
at two microphone positions spaced by 16 cm. Since was also a closed box the same type, model and di-
the computation is done in the time domain it would mensions. In Fig. 2 the three microphone positions
take so much time to compute even one minute of and the loudspeaker positions are shown. At each
music. Instead the auralization has been performed microphone position a pressure microphone con-
by using direct convolution with the obtained im- nected to a pre-amplifier was placed. The output
pulse responses from the simulation program. from the pre-amplifier was connected to a measur-

24
Low Frequency Enhancement System for Rooms Paper A

Fig. 11. Comparison of sound pressure level distribution in the rectangular grid area and measurements, the grid has 9x11
points separated by 20 cm form each other.

limited impulse response. From the MLS measuring

system the output was connected to a analog band
pass filter set to 10 Hz and 600 Hz as cut off fre-
quencies. From the band pass filter the signal was
sent to a reference stereo amplifier and from there
to the test loudspeaker.The length of the impulse
response was set to 8191 samples with a sampling
frequency of 8 kHz. The bandwidth was set to 2 kHz
with a Butterworth 8th order low pass filter as a
anti aliasing filter.

The first measurement was done measuring at the

three microphone positions placing test loudspeaker
(A) and the second one was measuring again the
three microphone positions with test loudspeaker
(A) and test loudspeaker (B) included both. The ex-
citation signal was the same for both loudspeakers.

4.2. Simulation

Two main scenarios were simulated following the

measurements. The sound pressure level distribu-
Fig. 12. Sound pressure level distribution resulting from the tion over the selected grid area and the computation
simulation of one loudspeaker reproducing a sinusoidal fre- of the impulse response of the room by the excitation
quency of 65 Hz, the measuring grid can be observed. of first test loudspeaker (A) and secondly adding test
loudspeaker (B). The averaged impulse response of
ing amplifier and sent to a MLS measuring system the real loudspeaker was included in the model. In
board in a PC. The system was calibrated with a both cases the simulation time was 1 second at a
piston-phone producing a sound pressure level of sampling frequency of 8 kHz. The space grid was set
124 dB at 250 Hz. to 10 cm for both cases. For the impulse response
simulation the impulse response of the analog band
It was decided to band-pass filter the excitation pass filter and the anti aliasing filter from the MLS
signal since the result of the simulation is a band system were included in the model. The boundary

25
Paper A Low Frequency Enhancement System for Rooms

Fig. 13. Thick lines are frequency responses from measured impulse responses at microphone positions 1, 2 and 3; Thin lines
are the simulations. Left column only loudspeaker (A) is included while in right column test loudspeaker (B) is also included.

conditions for the walls were set as follows, for the surface layer of the room. In Fig. 13 the frequency
wooden floor the bulk characteristic impedance was response from the measured and simulated response
used as an approximation to the impedance of that are shown. In the left column the room was exited
surface being 1.575x106 mkg2 s , equivalent to an ab- by test loudspeaker A while in the right column both
sorption coefficient α = 0.0011. The absorption co- test loudspeakers were used. The excitation signal
efficient of the walls was set to α = 0.1000. Two ab- was the same for both. In Fig. 14 the measured im-
sorption coefficients were used for the ceiling, α = pulse responses derived by the simulation program
0.0797 and α = 0.1530 for the most absorptive sec- are shown. It can be observed that the impulse re-
tions. sponses calculated in the simulation have more en-
ergy content than the real ones. Nevertheless as it is
shown in Fig. 13 they do not differ so much in the
4.3. Comparison of simulations and real frequency domain.
measurements

In Fig. 12 the sound pressure distribution can be ob- 5. DISCUSSION

served after simulation at a height of 1.2 m. It is no-
ticeable the influence of the room forming the nodes As it is shown in Fig. 12 and in Fig. 11 the simula-
and antinodes by the stationary waves. In Fig. 11 tion program present good agreement with the mea-
two surface plots are shown, these graphs repre- surements. The main room resonances are revealed
sent the sound pressure level simulated and secondly by the simulation. It can be said that at very low
measured along the chosen surface area, in Fig. 12 frequency there is some divergence. Nevertheless the
the same simulation is shown along the complete simulation program can be used as a predictor tool

26
Low Frequency Enhancement System for Rooms Paper A

in order to know beforehand what would happen Acoustics,” Proceedings of AES 15th International
when a loudspeaker is placed in a rectangular room. Conference, Audio, Acoustics & Small Spaces, pp. 32-
47. (November 1998)
It has to be added that the ceiling of the room is
quite complex to model since it is not regular and [4] A. Krokstad, “Calculating the Acoustical Room
some parts are covered with very absorptive mate- Response by the Use of a Ray Tracing Technique,” J.
of Sound and Vibration, 8, pp. 118-125. (1968)
rial like rock wool but in some other areas the ceil-
ing is quite reflective. In connection to that it was [5] J. B. Allen and Berkley, “Image Method for Efficiently
Simulating small-room acoustics,” J. Acoust. Soc. Am.,
quite difficult to find absorption coefficients for the 65, pp. 943-950. (1979)
materials at very low frequencies. It was a good ap-
[6] D. Botteldooren, “Acoustical Finite-Difference Time-
proximation to model the boundary condition using
Domain Simulation in Quasi-Cartesian Grid,” J. Acoust.
the characteristic impedance of the materials. Soc. Am., 95, pp. 2313-2319. (May 1994)
[7] M. M. Boone and G. Janssen, “Modal superposition in
6. CONCLUSION the time domain: Theory and experimental results,” J.
Audio Eng. Soc., 97, pp. 92-97. (January 1995)
A simulation tool has been developed. The finite- [8] J. P. Coyette, “Transient Acoustics: Evaluation of Finite
difference time-domain FDTD method has been Element and Boundary Element Methods,” Proc. of
ISMA 19, pp. 223-234. (1994)
used to approximate the sound pressure and parti-
cle velocity produced by multiple loudspeakers in [9] K. S. Kunz, Finite Difference Time Domain Method for
Electromagnetics, CRC. (1993)
a rectangular room. The simulation program has
been tested with good results according to measure- [10] L. E. Kinsler, Fundamentals of Acoustics, 4th. Ed. John
Wiley & Sons, Inc. (2000)
ments. The developed application can be used as
a reliable tool for equalization purposes on multi- [11] D. Botteldooren, “Finite–Difference Time Domain
Simulation of Low–Frequency Room Acoustic
channel sound reproduction systems in conjunction
Problems,” J. Acoust. Soc. Am., 98, pp. 3302-3309.
with other approximations as Ray Tracing or Image (December 1994)
Source. The solution gives the possibility to evalu-
[12] S. K. Olesen, “Low Frequency Room Simulation
ate and visualize the interaction of multiple loud- using Finite Difference Equations,” Proc. AES 102nd
speakers in a room by an animation of pictures of Convention, Preprint 4422 (D2). (March 1997)
the pressure amplitude at discrete times. Moreover [13] J. Borwick, Loudspeaker and Headphone Handbook,
the possibility of direct auralization of multichannel Butterworth & Co. (1988)
signals is possible by convolution of the calculated [14] B. Langvad, H. Møller and G. Budzynski “Testing a
impulse response and anechoic recordings. New Listening-Room,” Archives of Acoustics, 14, 1-2,
pp. 45-60. (1989)
7. ACKNOWLEDGEMENTS [15] B. Chereck and B. Langvad, “Low Frequency Simulation
of a Listening Room,” Proc. of Nordic Acoustical
Meeting, pp. 265-270. (1990)
The authors thank Søren Krarup Olesen Associate
Professor at Acoustics, Aalborg University for his [16] J.R. Hassall and K. Zaveri, Acoustic Noise
Measurements, 5th Ed. 1st Print. B & K. (1988)
helpful advice and since part of this paper has been
motivated by his early work in low frequency room [17] D. D. Rife and J. Vanderkooy, “Transfer function
Measurement with Maximum-Length Sequences,” J.
simulations. Audio Eng. Soc., 37, pp. 419-444. (June 1989)
[18] J. Vanderkooy, “Aspects of MLS Measuring Systems,”
References J. Audio Eng. Soc., 42, pp. 219-231. (April 1994)
[19] D.B. Keele Jr., “Low-Frequency Loudspeaker Assess-
[1] S. J. Elliott and P. A. Nelson, “Multiple-Point ment by Nearfield Sound-Pressure Measurement,” J.
Equalization in a Room Using Adaptive Digital Filters,” Audio. Eng. Soc., Vol.22, pp.154-162. (April 1974)
J. Audio Eng. Soc., 37, pp. 899-907. (November 1989)
[2] P. A. Nelson, F. Orduña-Bustamante and H.
Hamada,“Inverse Filter Design and Equalization Zones
in Multichannel Sound Reproduction,” IEEE Trans.
Speech Audio Process, 3, pp. 185-192. (1995)
[3] R. Walker, “Equalization of Room Acoustics and
Adaptive Systems in the Equalization of Small Room

27
Paper A Low Frequency Enhancement System for Rooms

Fig. 14. Impulse responses measured and simulated by FDTD correspondent to the three microphone positions in the test room.

28
Paper B
Paper B Low Frequency Enhancement System for Rooms

30
 Optimizing placement and equalization of multiple low
frequency loudspeakers in rooms
Adrian Celestinos a , and Sofus Birkedal Nielsen a
a Acoustics, Department of Electronic Systems, Aalborg University, DK-9220 Aalborg East, Denmark

Abstract

Every room has strong influence on the low frequency performance of a loudspeaker. This is often problematic to
control and to predict. The modal resonances modify the response of the loudspeaker depending on placement and
listening position. In order to anticipate the behavior of low frequency loudspeakers in rooms a simulation tool
based on finite-difference time-domain approximations (FDTD) has been developed. Simulations have shown that by
increasing the number of loudspeakers and modifying their placement a significant improvement is achieved. A more
even sound pressure level distribution along a listening area is obtained. The placement of loudspeakers has been
optimized. Furthermore an equalization strategy can be implemented for optimization purpose. This solution can be
combined with multi channel sound systems.

1. INTRODUCTION tracing and Image source method [4],[5]. Such meth-

ods are no longer adequate to simulate frequencies
below 100 Hz because they are based on geometri-
When a loudspeaker is placed in a rectangular room cal acoustic approximations where the wavelength
a number of problems arise. Modification of the re- is smaller than the room dimensions.
sponse of the loudspeaker at the listening position
occurs due to the strong influence of the room and The finite-difference time-domain (FDTD) method
the position of the loudspeaker. The combination has been used with great success to model electro-
loudspeaker-room acts as a coupled system where magnetic problems. In acoustics FDTD has shown
the room typically dominate by its distinct normal good performance to approximate the low frequency
modes. When placing more than one loudspeaker in room behavior. This method has been well described
the room some of the attenuation produced by the in early studies by Botteldooren [6]. By using this
room is less sever but still the room has a strong in- approximations the calculations are made directly
fluence at the different listening positions. In order in the time domain. The sound wave equation is dis-
to deal with this problem equalization techniques cretized both in time and space and the inclusion of
have been investigated by several authors in [1],[2] multiple sound sources in the room is possible since
and [3]. In connection to that a robust tool to sim- the finite time and space is always available. In this
ulate the low frequency behavior of multiple sound paper the FDTD method has been chosen to simu-
sources placed in rectangular rooms is needed. Dur- late a rectangular room excited by multiple sound
ing the last years two main approximations using nu- sources. The estimation of the sound level distribu-
merical methods has been developed which are Ray tion at low frequencies is calculated. The inclusion
Paper B Low Frequency Enhancement System for Rooms

of loudspeakers assuming omnidirectional compact sional Cartesian staggered grid where particle veloc-
sources is implemented inside the room. Since the ity and pressure points are computed. By using this
particle velocity is always available, sound power method the impulse response with a number of vir-
and intensity can be estimated for equalization pur- tual microphones in the room can be obtained. The
poses. Moreover a band limited impulse response of impulse response is the instantaneous sound pres-
the room can be derived. sure at a point in the room including the transfer
function of one or more loudspeakers and eventually
Sound reproduction systems are typically placed signal processing and reflections of the room. These
in small or medium size rectangular rooms. Every virtual microphones can be set along a defined lis-
room has strong influence on the low frequency tening area or wherever is desired in the room. In
performance of a loudspeaker. The combination addition to that the sound level distribution on a
loudspeaker-room acts as a coupled system where specified section of the room can be obtained.
the room properties typically dominate due to the
parallel walls. This is often problematic to control The loudspeakers are implemented as point sources
and to predict since the modal resonances modify occupying one or more pressure or particle veloc-
the magnitude response of the sound source de- ity points in the room. The simulated loudspeakers
pending on the listening position and loudspeaker can be sealed boxes modeled as a 2nd order band
placement. In order to predict the behavior of low pass filtered version of a Gaussian asymmetric pulse.
frequency loudspeakers in small and medium size Moreover it is possible to include a real loudspeaker
rooms a robust simulation tool has been developed. impulse response of a near field measurement. The
absorption coefficient of the walls can be modified
By using the developed program different configu- as well as some of the sections of the room for exam-
rations of loudspeakers are analyzed from one to ple an open window or a door can be added in the
four loudspeakers at low frequencies on different lo- simulated room.
cations in the room. Comparison of these configu-
rations has been carried out by using quantitative A very useful advantage of the developed simula-
parameters. tion is the visualization aspect. Since the method is
based in the time domain an animation composed by
Three optimization strategies are proposed and sim- indexed images of the instantaneous sound pressure
ulated. First the equalization of the sound field at a in a desired area of the room can be rendered. These
limited listening area using Multiple point equaliza- images are saved as a video file for further analysis
tion is performed. Secondly the acoustic radiation and visualization. Moreover a graphical user inter-
power close to the loudspeaker is equalized and fi- face (GUI) (seen in Fig. 17) has been developed for
nally the modification of phase and delay of some of an easy use of the simulation program where some of
the loudspeakers is performed. The implementation the parameters can be modified as the number and
of a selected configuration as well as an optimiza- location of loudspeakers, dimensions of the room,
tion method is performed in a real setup. The setup virtual microphones, delays, and gains of each loud-
includes multiple loudspeakers placed in a standard speaker. The developed simulation program is well
listening room. Measurements have been carried out described and it was tested by the authors in [7] and
in one of the configuration in order to verify the per- the theory behind it can be found in [6], [8] and [9].
formance of the selected optimization strategy.

3. QUANTITATIVE PARAMETERS
2. SIMULATION PROGRAM

In order to asses the performance of the different

A numerical method based on finite-difference time- configurations of loudspeakers a quantitative metric
domain approximations (FDTD) has been created is needed. On a defined listening area in the room
in Matlab to simulate multiple loudspeakers in a the sound field will consist on the addition of the
room. By the developed application a rectangular contribution of the loudspeaker and the modal char-
room is simulated using a discrete model in time and acteristics of the room. This steady state sound field
space of the sound wave equation. In this fashion the will vary in amplitude according to position and fre-
room space can be represented by a three dimen- quency. In order to asses this variation two parame-

32
Low Frequency Enhancement System for Rooms Paper B

Fig. 1. Simulation of the sound pressure level (SPL) distribution averaged along 1.02 seconds in a rectangular room using the
loudspeaker shown in Fig.3. Left, driven frequency 22 Hz, room mode (1 0 0). Middle 44.6 Hz, room mode (2 0 0). Right
67 Hz, room mode (3 0 0).

ters were chosen, Magnitude Deviation and Spatial lated across positions. The lower the value the less
Deviation. variation exists between positions along the listen-
ing area. The standard deviation from the responses
at the microphone positions is calculated as
3.1. Magnitude Deviation
v
fhigh u np
1 X u 1 X
SVstd = t (xp,i − xi )2 (2)
If the main goal is to achieve an even sound pressure nf np − 1 p=1
i=flow
distribution along a listening area a flat frequency
response should be obtained on each microphone
position. To quantify how much the magnitude on where nf is the number of frequencies in the fre-
each microphone position deviates from an ideal flat quency range of interest from flow = 30Hz to
response the parameter Magnitude Deviation is used fhigh = 150Hz and np is the number of microphone
which is the standard deviation from this ideal flat positions, and xp,i is the ith frequency at position
response calculated across the given responses as p. A SVstd equal to 0 dB will indicate that all
v
u fhigh
magnitude responses are identical along the whole
u 1 X listening area.
M Dstd = t (xi − xi )2 (1)
nf − 1
i=flow This parameters are obtained from the N point dis-
crete Fourier transforms (DFT) of the impulse re-
where nf is the number of frequencies in the fre- sponses generated by the simulation program. The
quency range of interest from flow = 30Hz to Length of the impulse responses is N = 213 being
fhigh = 150Hz and xi is the ith frequency and xi 8192 samples with a sampling frequency f s = 8 kHz.
is the mean of xi , M Dstd is given in dB. A M Dstd No smoothing was applied.
equal to 0 dB represents an ideal flat magnitude
response. If the whole listening area is analyzed an
average of all individual M Dstd of positions is done
to give a single descriptor. 4. ANALYSIS

3.2. Spatial Deviation In the following section a briefly insight to the the-
ory behind the sound fields in rectangular rooms
is presented followed by an analysis of different
In order to quantify how much the magnitude varies loudspeaker configurations using up to four low
along the listening area the parameter Spatial Devia- frequency loudspeakers. Finally the effect of loud-
tion is used and it consists on the standard deviation speaker placement is illustrated on two loudspeaker
of every single frequency from the mean level calcu- configurations.

33
Paper B Low Frequency Enhancement System for Rooms

20

10

(dB re. Pa/V)

0

−10

−20

−30
10 100 1k
Frequency (Hz)

0.04

0.02

(Pa/V)
0
Fig. 2. Virtual room to be simulated seen from above, di-
mensions, shape, loudspeaker position and calculation of the −0.02
first 17th room modes are presented.
−0.04
0 0.05 0.1 0.15 0.2
4.1. Room Modes Time (s)

Fig. 3. Frequency response (upper) and time response (lower)

The purpose of this paper is not to analyze in deep of the real loudspeaker included in the simulations.
the modal theory. Nevertheless the sufficient back-
ground is presented here to support the further anal- nz are > 0. The zones where there will be minimum
ysis. When low frequency sound is confined in a sound pressure level are called nodes and the points
rectangular environment it will experience certain where exists a maximum of sound pressure are called
changes. Assuming a rectangular room with rigid anti nodes [10]. In Fig. 2 the first room modes of a
walls and if a loudspeaker is placed at the end wall rectangular room are calculated using Eq. (3).
of the longest dimension of the room, reproducing
continuously a pure tone of frequency where half of
the wavelength corresponds to that dimension, the 4.2. Evaluation of Loudspeaker Configurations
sound wave will reflect at the opposite wall and meet
at the middle of the room in opposite phase with the
wave traveling directly from the loudspeaker. This Extensive experimental investigation has been done
will cause destructive interference in the middle of by Welti in [11] where up to 16 subwoofers have
the room at this particular frequency and the travel- been used on different configurations in a rectangu-
ing wave will again hit the wall with the loudspeaker lar room. Results in this investigation have shown
in phase with the loudspeaker. It will also occur at that by increasing the number of loudspeakers a sig-
frequencies where an integer multiple of half of the nificant improvement is achieved on a centered lis-
wavelength corresponds to one or more of the di- tening area. Moreover those results shown that sym-
mensions of the room (see Fig. 1). This phenomena metrical configurations give better results than non-
it is often called standing wave, or Mode each mode symmetrical ones.
is related to a certain natural frequency given by On this paper up to six configurations are chosen to
c
r
nx
2 ny
2 nz
2 be simulated in a virtual room. The reason of choos-
fn = + + (3) ing this configurations is that they present special
2 lx ly lz
characteristics that could be used for sound repro-
duction systems. Since human sound localization is
Where c is the speed of sound in the air, nx , ny and quite poor at low frequencies then it is possible to
nz are integers starting with 0, 1, 2,... and ly , lx , lz add more loudspeakers with out destroying the per-
are the dimensions of the room. The room modes ceived sound image.
can be grouped in Axial Modes where only one of
the integers nx , ny , nz is > 0, Tangential Modes A rectangular room with an absorption coefficient of
which are two dimensional and two of the integers 0.10 in all walls is rendered with the created simula-
nx , ny , nz are > 0 and the Oblique Modes that are tion program in Matlab. The virtual room is slightly
three dimensional, where all three integers nx , ny , similar to a standard listening room at the Acous-

34
Low Frequency Enhancement System for Rooms Paper B

0 0
MDstd = 6.89 dB MDstd = 8.11 dB
SVstd = 5.31 dB SVstd = 3.12 dB
−10 −10

(dB re. Pa/V)

(dB re. Pa/V)

−20 −20
LP1 −30
LP2 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

0 0
MDstd = 9.34 dB MDstd = 7.97 dB
SVstd = 3.17 dB SVstd = 3.65 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20
LP3 −30
LP4 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

0 0
MDstd = 9.74 dB MDstd = 6.46 dB
SVstd = 3.37 dB SVstd = 4.27 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20
LP5 −30
LP6 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

Fig. 4. Frequency responses from the simulated configurations are shown together with the room and the position of the
loudspeaker seen from above. Green lines are the responses at each virtual microphone on the listening area. Blue lines are
the mean. The averaged M Dstd of all responses and the SVstd values are presented in every configuration.

tic department at Aalborg University. In Fig. 2 the walls along the longest dimension of the room has
simulated room is presented seen from above as well the lower Spatial Deviation with an SVstd value of
as the calculation of its room modes. A centered lis- 3.12 dB but in the other hand it has a high mag-
tening area in the room at a height of 1.25 m is de- nitude deviation value being M Dstd 8.11 dB, simi-
fined by 25 virtual microphone positions spaced by larly configuration LP3 has a low SVstd value being
40 cm from each other. A cell size grid of 10 cm has 3.17 dB and a high M Dstd of 9.34 dB. Although con-
been used to discretize the room. The magnitude figuration LP5 has the highest M Dstd it presents a
deviation M Dstd and the Spatial Deviation SVstd quite low variation across positions. Configuration
are calculated on each configuration of loudspeak- LP4 is the one that presents a fair compromise be-
ers. An impulse response of a real loudspeaker has tween variation across positions and magnitude de-
been used in the simulations which can be seen in viation. Even though configuration LP1 with one
Fig. 3 together with its frequency response. All the loudspeaker in the corner has the lower M Dstd value
loudspeakers are fed with the same signal and po- it has the highest variation across positions having a
sitioned 25 cm above the floor and 25 cm from the SVstd value of 5.31 dB nevertheless it is the one that
walls or at centered positions. seems to excite all room modes more evenly, a simi-
lar behavior shows configuration LP6 improving the
As it can be observed in Fig. 4 configuration LP2 M Dstd and SVstd compared to LP1.
with two loudspeakers at mid points on opposite

35
Paper B Low Frequency Enhancement System for Rooms

0 0
MDstd = 6.89 dB MDstd = 7.14 dB
SVstd = 5.31 dB SVstd = 5.62 dB
−10 −10

(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

0 0
MDstd = 9.74 dB MDstd = 9.47 dB
SVstd = 3.37 dB SVstd = 3.41 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

0 0
MDstd = 6.46 dB MDstd = 7.90 dB
SVstd = 4.27 dB SVstd = 4.60 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

Fig. 5. Green lines are the frequency responses at each virtual microphone in the listening area. Blue lines are the mean. Left
plots are the frequency responses of original configurations LP1, LP5 and LP6. Right plots are the relocated configurations.
In LP1 (relocated) the loudspeaker is 1.25 m off the lateral wall, 1.65 m from the front wall and at a height of 0.75 m. In LP5
and LP6 both (relocated) the loudspeakers are raised at a height of 1.25 m.

What it can be observed in Fig. 4 is that on config- sponding to 47.5 Hz it is canceled out as well.
urations LP2, LP4 and LP5 where the loudspeak-
ers are on opposite walls and particularly in LP2 After this analysis one can verify that by increas-
and LP4 where the loudspeakers are located at mid ing the number of loudspeakers the variation across
points they cancel out the room modes with odd nx , positions is improved at expenses of an increment
ny , nz integers and excite strongly the room modes on the magnitude deviation at every position. If the
corresponding to even integers (see Eq. (3)). The loudspeakers are located at opposite walls and at
axial room modes that have odd integers are those mid points they can cancel out some of the room
that present a node or minimum pressure at cen- modes but in the other hand increasing others. After
tered positions in the room. On the contrary the this observations one can propose that if the system
axial room modes that have even integers are those in question will be equalized, increasing the number
which present an anti node or maximum pressure of loudspeakers at mid positions will be favorable.
at centered positions in the room. For example seen This is because less microphone positions need be
Fig. 4 in configuration LP2 and LP5 the room mode used to have a knowledge of the sound field to im-
(3 0 0) corresponding to 67 Hz is heavily suppressed plement some kind of equalization. Besides that a
and from LP2 to LP4 the room mode (1 1 0) corre- single equalization filter can work for all loudspeak-
ers and more listening positions will benefited. If the

36
Low Frequency Enhancement System for Rooms Paper B

system will not be equalized the configuration LP6 reflection away from overlap a wall reflection.
with two loudspeakers on the front wall and LP2
with one loudspeaker in the corner can work up to The effect of relocating the loudspeaker in LP1 is
some extension. It should be pointed out that this shown in Fig. 5 at upper plots, where the loud-
two loudspeaker configurations can be improved by speaker has been moved from the corner to 1.25 m
for example optimizing its placement. In the next away from the lateral wall, 1.65 m beneath the front
section the effect of repositioning the loudspeakers wall and raised to a height of 0.75 m. From this plots
will be presented. it can be observed that an overall reduction in power
has been obtained. Besides that a reduction on the
excited resonances is experienced, specially on the
axial room modes (0 0 1) and (3 0 0) corresponding
4.3. Positioning to 63.7 Hz and 67 Hz respectively. Interestingly the
averaged magnitude deviation at all positions and
the spatial deviation have been slightly increased.
In case of having just one low frequency loudspeaker
it is well known that if it is placed within the room In Fig. 5, the effect of moving the loudspeakers in
at a anti node that resonance will be strongly ex- the symmetrical configuration LP5 at a height of
cited, and if the loudspeaker is located in a node 1.25 m can be seen. As it can be observed only a
that particularly mode will be weakly excited, this few modes are excited and since that height cor-
is often referred as if the sound source is well cou- responds to almost the mid point of that dimen-
pled to the room or not. The only position that en- sion the mode (1 1 1) corresponding to 79.4 Hz has
sure that all room modes are strongly excited it is been suppressed. Nearly only three room resonances
in a corner position, since all room modes have an are being strongly excited which are the ones with
anti node at the corners. Moreover a loudspeaker in even integers, (see Eq. (3)). It should be pointed out
the corner will experience an increment in power of that this configuration may not be pleasant to lis-
8 times that means approximately + 9 dB. If the ten with out some kind of control or equalization
room in question it is a middle size room with a re- since the resulting room modes excited will be exces-
verberation time T60 at 500 Hz of approx. 0.5 sec- sively boosted. Nevertheless this setup will simplify
onds it is recommended that the sound source will the equalization process.
be well coupled to the room in order to have a good
balance between mid frequencies and low frequen-
cies although some coloration at low frequencies will 5. OPTIMIZATION STRATEGIES
be inevitable [12]. A well coupled room loudspeaker
will mean an amplification on the power output of
the loudspeaker. This might be beneficial for a loud- In the next section three optimization strategies are
speaker with a poor frequency response at low fre- proposed and simulated, first a multiple point equal-
quencies. ization is presented followed by the equalization of
the radiated power of the loudspeakers and finally
From Fig. 5 in configuration LP1 it can be observed the modification of delay and phase of loudspeak-
that the loudspeaker excites evenly all room modes, ers. The three approaches are applied to the loud-
moreover if some improvement is required one could speaker configuration LP5 (relocated). Along this
relocate the loudspeaker moving it close to a node section the absorptions coefficients of the simulated
correspondingly to every axial dimension. That sit- virtual room have been changed to a more realistic
uation can be for example to move the loudspeaker room environment being, walls 0.12, floor 0.15 and
in configuration LP1 25 % off from the walls on each ceiling 0.2. The virtual room dimensions have been
dimension. Care should be taken to avoid that the kept the same as in section 4.
reflection coming from every corner wall cancel out
the loudspeaker itself. When ever a loudspeaker is
close to a corner it experiences seven reflections, 3 5.1. Multiple Point Equalization
coming from the walls, 3 coming from bi corners and
one from a tri corner. In order to avoid this situa-
tion one should keep the distances of the three walls Different approaches have been developed in the last
as different as possible but also keep the bi corner years to overcome a solution of equalizing a loud-

37
Paper B Low Frequency Enhancement System for Rooms

Fig. 6. Resulting filters for the multiple point equaliza- Fig. 7. Upper (thick black) curve is the pressure at 10 cm
tion technique. Thin (gray) curve is the normalized average. close to the loudspeaker located in the corner of the room.
Thick (black) curve is the equalization filter. Thin (black) Upper (thin black) curve is the pressure at 10 cm close the
curve is the equalized average. loudspeaker in an arbitrary position in the room. Upper
(thin gray) curve is the volume velocity. Lower curves are
the radiated sound power, in the corner (thick black) and in
speaker in an room [1], [2], [3]. This approaches an arbitrary position in the room (thin black). Lower (thick
called multiple point equalization can achieve per- gray) curve is the radiated power in anechoic conditions.
fect equalization at multiple points when the num-
ber of sources is more than the number of equal- LP5(relocated) configuration.
ization points. In the case of loudspeaker configura-
tion LP5 (relocated) perfect equalization would be
achieved for only four microphone positions if these 5.2. Equalization of Acoustic Radiation Power
approaches are applied. After seen Fig. 5 in LP5 (re-
located) one can observed that within the working
range of the loudspeakers the frequency response is In order to improve the coupled system loudspeaker-
more less similar at all positions in the listening area, room it is necessary to know how the loudspeaker
since this configuration shows a low value SVstd . Af- will interact with the room. This means in what
ter this observations one would suggest that an in- degree the loudspeaker will excite the room reso-
verse filter obtained from the averaged frequency re- nances or not excite them at all. This will depend
sponses can work for an extended listening area. on a number of factors such as the placement of the
loudspeaker, its own characteristics, the reverbera-
After obtained the impulse responses at the 25 lis- tion time of the room at different frequencies and
tening positions an average in the frequency domain so on. At low frequencies where the room dimen-
has been performed. Since the filtering is done off sions are comparable with those wavelengths loud-
line the sampling frequency has been kept the same speakers are not constant power generators. Below
as in the simulation program being f s = 8 kHz. Sub- the Schroeder frequency which is where three over-
sequently the average has been normalized to the lapping room modes occurs the statistical theory of
corresponding level of 30 Hz which is the cut off fre- sound fields in rooms can not be applied [14]. Since
quency of the loudspeakers. The frequency bins from at low frequencies a closed box loudspeaker acts as
0 to 30 Hz and from 150 Hz to the Nyquist frequency a point source radiating sound equally in all direc-
have been set to 0 gain. In order to be sure that the tions the radiated acoustic power can be calculated
system is causal and stable the minimum-phase has from its volume velocity and its radiation resistance.
been calculated using homomorphic filtering [13], The acoustic load of a loudspeaker placed in differ-
this assures that all poles and zeros are inside the ent environments will be reflected directly on the
unity circle so a stable inverse exists. Next the fil- sound power radiated by the source.
ter has been inverted and a direct finite impulse re-
sponse (FIR) is acquired, in Fig. 6 the resulting FIR The total power radiated by the source can be writ-
filter can be seen in the frequency domain. After this ten as:
process the resulting filter is loaded into the simu- 1
lation program and applied to the loudspeakers in W = U · U ∗ Re(Zr ) (4)
2
38
Low Frequency Enhancement System for Rooms Paper B

Fig. 8. Upper plot (thick) curve is the resulted FIR filter from Fig. 9. Upper (thin gray) curve is the volume velocity before
the radiated power, (thin gray) curve is the pressure before equalization of the radiated power, (thin dark) is the volume
equalization, (thin dark) is the pressure near the membrane velocity after the equalization. Lower curve (thick gray) is
after equalization. the radiation power before equalization and (thick dark) is
the radiation power after being equalized.
where W is the average radiated power, U is peak
volume velocity generated by the loudspeaker (∗ in- field conditions from
dicates the complex conjugate), Re(Zr ) is the ra- p
U= (8)
diation resistance (Re indicates the real part), and Zr
Zr is the radiation impedance. Considering the di-
aphragm of the loudspeaker of area S moving with having U then the radiated power by the loud-
a normal velocity component u then the radiation speaker can be obtained by substitution of Up in
impedance is expressed as 1 p
W = · U · U ∗ Re( ) (9)
Zr = Rr + jXr (5) 2 U

where p is the pressure close to the membrane and

where Xr is the radiation reactance and Rr is the
U is the peak volume velocity.
radiation resistance. Since the acoustical impedance
of the loudspeaker radiator is higher than the radia- In Fig. 7 the acoustic power radiated by a loud-
tion impedance, changes in the radiation impedance speaker in the corner and in an arbitrary position
have small effect on the volume velocity. Thus it can within the room can be seen. Although the sound
be said that a loudspeaker is a constant volume ve- pressure will variate according to the position of the
locity source [12], [14], [15]. listener it is clear that by this curves one can predict
which room modes this loudspeaker will excite.
If the response of the loudspeaker in free field is
known, the volume velocity can be estimated. As- After this theoretical background the proposed
suming the loudspeaker to be a baffled simple source method is to measure the acoustic radiation power
and for ka  1, being k the wave number and a the close to the loudspeaker and obtain a filter to at-
radius of the membrane, the radiation resistance can tenuate the room influence. A similar approach has
be calculated from been used in [16] where the radiation resistance
1 is equalized or replaced by one measured in a ref-
Rr = ρ̇c(kS)2 (6) erence room. Indeed the room modes will not be
2π
suppressed, they will be there any way but they will
and the radiation reactance become Xr = jρcSka not be excited as strong as the other frequencies.
therefore from the radiation impedance After being measured the acoustic pressure at 10 cm
p near the loudspeakers on configuration LP5 (relo-
Zr = = Rr + jXr (7)
U cated) the radiated power is calculated as explained
before. Since both the room and the setup LP5 are
where Rr is the real part of Up the volume velocity symmetric it is sufficient to used just one measure-
U can be obtained using the known pressure in free ment in front of one of the loudspeakers. From this

39
Paper B Low Frequency Enhancement System for Rooms

Fig. 11. Sound pressure level distribution in the Virtual Room at 1.25 m height. Frequency reproduced: 46 Hz. Left, all the
loudspeakers are in phase. Middle, rear loudspeakers are out of phase. Right, rear loudspeakers are delayed and out of phase.

gram. In Fig. 9 the radiation power before and after

the equalization is presented as well as the volume
velocity.

5.3. Optimization by Modifying Delay and Phase

A very intuitive approach is used to minimize the

effect of the excited room modes. Since by this loud-
speaker configuration LP5 (relocated) mainly the
axial modes corresponding to the longest dimension
of the room are excited one can used the principle of
Fig. 10. Upper (thin dark) curves are the acoustic pressure
both front and rear loudspeakers before the adjusting of am-
absorption by using the rear loudspeakers as acous-
plitude. Upper (thick dark) curve is the acoustic pressure tic absorbers.
of rear loudspeakers after the adjusting. Middle (thin dark)
curve is the radiated power at the rear loudspeakers be- A pure delay of 21.7 ms corresponding to the dis-
fore the adjusting. Middle (thick dark) curve is the radiated tance from the front loudspeakers and the back wall
power at the rear loudspeakers after the adjusting. Lower has been applied to the rear loudspeakers. Apart of
(thick dark) curves are the radiated power of front loud- the delay they have been inverted in phase so they
speakers before and after the adjusting. Lower (thin gray)
is the radiated power in anechoic conditions.
will cancel out the wave front coming from the front
loudspeakers. In Fig. 11 the effect of this procedure
can be seen, in the left plot all the loudspeakers are
measure a target filter is prepared in the frequency reproducing 46 Hz as a result the room mode (2 0
domain by normalizing the curve to 0 gain at the 0) is hardly excited and a very high sound pressure
level of 30 Hz. Then the target filter is squared and is measured in the center of the room. Differently in
inverted. The frequency bins from 0 to 30 Hz and the middle plot when the rear loudspeakers are out
from 150 Hz to the Nyquist frequency have been set of phase an attenuation of more than 50 dB occurs
to 0 gain. Having shaped the target filter a digital at the center of the room and in the right plot when
FIR filter is acquired as it is shown in Fig. 8. This the rear loudspeakers are delayed and inverted in
procedure is done by using the frequency sampling- phase the sound pressure level in the center of the
based design [17]. Afterwards the minimum-phase of room has been decreased by just 27 dB.
the FIR filter has been calculated the same manner
as in section 5.1 and loaded into the simulation pro- A similar approach has been proposed before by El-

40
Low Frequency Enhancement System for Rooms Paper B

liot and Johnson for noise control in [18] and [19]

where a method of adjusting a secondary source to
minimize the total power output of both primary
and secondary sound sources is achieved. In this
method the volume velocity of the secondary source
is adjusted to be inverted in phase and gradual in-
crement in amplitude to absorb power from the pri-
mary source is performed.
In this paper it has been found that if the rear loud-
speakers are adjusted to be out of phase an almost Fig. 12. Measurement setup used to calibrate the system and
complete cancellation of sound is achieved in the for acquisition of the impulse responses at the listening area.
center of the room. Since the listening area is at
centered position therefore the pure delay is applied section followed by discussions on the results and
obtaining a reasonable reduction in sound pressure measurements.
and not a complete cancellation. This effect is illus-
trated in Fig. 11 as it is noted the room mode is
effectively suppressed by inverting the phase of the 7. MEASUREMENTS
rear loudspeakers. By adjusting the amplitude of the
rear loudspeakers it is possible to achieved a more After simulating the three optimization strategies
even sound pressure level distribution. it has been decided to test the last optimization
In order to adjust the amplitude of the rear loud- method on configuration LP5 (relocated) in a real
speakers the radiated sound power has been mea- room. Four closed box active loudspeakers have been
sured as explained in section 5.2. The rear loud- used as in Fig. 5, LP5 (relocated) at a height of 1.20
speakers have been set to -6 dB and the radiated m and 1 m from the lateral walls. The frequency
power has been measured and stored. Next the rear response of the loudspeakers can be seen in Fig. 3.
loudspeakers have been set to -5 dB and so on up to The standard listening room of the Department of
+3 dB in increments of 1 dB. After this procedure Acoustics at Aalborg University has been used to
it has been found that the rear loudspeakers have to carry out the measurements, it has the following di-
have -2 dB gain compared to the front ones in order mensions, Length 7.80 m, Width 4.12 m and a height
to absorb enough power from the front loudspeak- of 2.77 m. The room has an averaged reverberation
ers and minimize the total radiated power. This can time T60 of 0.47 s. The floor has a carpet and it
be observed in Fig. 10 where the radiated power is wooden, the walls are covered with special pan-
from both front and rear loudspeakers is shown be- els that can be removed or moved to different po-
fore and after the adjusting in amplitude of the rear sitions. The ceiling is curved in the corners covered
loudspeakers. This adjustment is just good enough with special plaster panels [20]. The phase on the
for this room the situation may change in another rear loudspeakers has been inverted. A digital de-
room with for example different composite walls. lay and an attenuator have been connected before
the rear loudspeakers (see Fig. 12). The system has
been calibrated to have the same input level to all
6. RESULTS the active loudspeakers. The delay for the rear loud-
speakers has been adjusted so the sound pressure at
44.1 Hz (the room mode frequency) is minimum at
The results of the three optimization methods for a centered position within the listening area in the
this particular loudspeaker configuration LP5 (relo- room. Afterwards the rear loudspeakers have been
cated) are shown in Fig. 13 and Fig. 15 as surface adjusted increasing the gain in the attenuator from
plots arranged as rows of the listening area in the -4 dB to + 4 in steps of 1 dB. It has been found
room. As it is observed in Fig. 13 the three methods that a gain of + 2 dB was the optimum to attenuate
removed the peaks and particularly the method of the room mode at 44.1 Hz so it has the same ampli-
adding delay and inverting the phase showed a bet- tude as neighbor frequencies. After the calibration
ter performance than the others. After this results of the system the impulse response has been mea-
a validation measurement is presented in the next sured at the 25 microphone positions in the listen-

41
Paper B Low Frequency Enhancement System for Rooms

0 0
MDstd = 8.64 dB MDstd = 3.06 dB
SVstd = 3.42 dB SVstd = 3.42 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

0 0
MDstd = 7.57 dB MDstd = 4.69 dB
SVstd = 3.42 dB SVstd = 2.21 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

Fig. 13. Upper left are the frequency responses at the 25 virtual microphones positions in the listening area of configuration
LP5 (relocated). Upper right are the equalized responses by Multiple point equalization (blue), not equalized (green). Lower
left are the equalized responses by radiated power equalization (blue), not equalized (green). Lower right are the equalized
responses by Modification of delay and phase (blue), not equalized (green)

ing area. the results of the measurement can be seen than 6 dB in sound pressure level would be very
in Fig. 14 where the simulations are compared with noticeable. As for the spatial deviation the number
the measurements and in Fig. 16 they are presented SVstd shows a very small improvement. This may be
as surface plots. due to some asymmetries in the room.
As it can be seen in lower right plot in Fig. 14
Although some variations in amplitude exists the
system has removed the peak corresponding to the 8. DISCUSSIONS
room mode by 10 dB. The next room mode corre-
sponding to 67 Hz has also been attenuated by al-
After observed the simulations in Fig. 13 and Fig. 15
most 7 dB. It can also be observed that some of the
it can be said that the second method of equaliza-
notches are removed or minimize. Even though the
tion of radiated power removes only the peaks on
system did not performed as perfect as the simula-
the responses and interestingly the spatial variation
tion a significant improvement has been achieved.
did not change being SVstd 3.42 before and after the
The variations may be happen due to small varia-
equalization. In this method the magnitude devia-
tions on the adjustment of the system to its optimal
tion it is improved by only 1.05 dB. This method
performance. It has to be mentioned that the met-
can be used to remove part of the influence of the
ric M Dstd of the magnitude deviation do not really
room, and if the system is completely symmetric it
reflect the improvement, since a reduction of more
does not need many measurements.

42
Low Frequency Enhancement System for Rooms Paper B

0 0
MDstd = 6.88 dB MDstd = 6.98 dB
SVstd = 4.15 dB SVstd = 4.24 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

0 0
MDstd = 4.69 dB MDstd = 5.84 dB
SVstd = 2.21 dB SVstd = 4.13 dB
−10 −10
(dB re. Pa/V)

(dB re. Pa/V)

−20 −20

−30 −30

−40 −40

−50 −50
10 100 10 100
Frequency (Hz) Frequency (Hz)

Fig. 14. Left column are simulations of front loudspeakers with out the optimization (upper green). Left (lower blue) are
simulations of front and rear loudspeakers with the optimization system, (background green) the same as upper with out
optimization. Right (upper green) column are the measurements with only front loudspeakers and no optimization. Right
(lower blue) are the measurements with the rear loudspeakers and the delay and phase inversion, (background green) the same
as upper with out optimization.

The method of multiple point equalization can effectively removed and the spatial deviation has
achieved a quite good equalization from 30 Hz up been improved from a SVstd value of 3.42 dB to a
to 70 Hz and the average of the magnitude devia- SVstd value of 2.21 dB. A quite good equalization is
tion at the microphone positions has been improved achieved from 30 Hz to 70 Hz starting to deteriorate
from a M Dstd value of 8.64 dB to 3.06 dB. It can as the frequency is increased. This method can be
be seen that although this method produces quite used using the rear loudspeaker on a limited range
flat responses it does not improve the spatial de- from 30 Hz to 60 Hz by for example low pass filtering
viation. Besides that after seen the shape of the the rear signals. Contrasting to the other two meth-
filter applied to the loudspeaker in this method on ods one can say that this method has a better per-
Fig. 6 a high boost from 90 Hz to around 120 Hz it formance since no filtering is included, only a pure
is applied to the loudspeaker, that may cause non delay and phase inversion is used so no increment in
linear distortion. In this approach the room mode power is delivered to the loudspeaker at certain fre-
frequencies are heavily attenuated so it can be said quencies. Actually the rear loudspeakers are used as
that so much energy has been waisted. absorbers of energy to remove reflections from the
back wall and thereby reducing the room modes.
In comparison with the other two methods by modi-
fying the phase and applying a delay to the rear loud- In a real setup it has been confirmed that the sys-
speakers a very interesting improvement is achieved, tem with the delay and phase inversion works suf-
as seen in Fig. 13 the two room modes have been ficiently to removed the room modes and improve

43
Paper B Low Frequency Enhancement System for Rooms

also the notches situation. The other two methods in the room is achieved. This system removes effec-
will also take care of the peaks by applying heavy tively the room modes and it diminishes some of the
filtering. Nevertheless in the delay and phase inver- notches at a centered listening area in the room. The
sion method a more precise procedure to adjust the system of adding delay and phase inversion has been
rear loudspeakers should be found out in order to tested in a real room. Although the adjustment of
calibrate the system to its optimum performance. In the system has to be quite precise it shows an ac-
connection to the evaluation parameters we should ceptable performance. Further investigation should
find a more realistic parameter since the magnitude be carried out on finding an effective procedure on
deviation M Dstd does not really reflect what it is the adjustment of the system and on the compatibil-
perceived. Since it is well known that the notches are ity with standard sound reproduction systems from
less audible than the peaks therefore by smoothing stereo to 5.1 multichannel setups.
the frequency curves the notches will be some how
hidden and a better descriptor may be obtained. A
similar approach has been used in [11] where the
References
metric descriptors are obtained from smoothed fre-
quency responses.
[1] S. J. Elliott and P. A. Nelson, “Multiple-Point
Equalization in a Room Using Adaptive Digital Filters,”
J. Audio Eng. Soc., vol. 37, pp. 899-907. (November
9. CONCLUSIONS
1989)
[2] P. A. Nelson, F. Orduña-Bustamante and H.
In this paper a robust simulation program has been Hamada,“Inverse Filter Design and Equalization Zones
in Multichannel Sound Reproduction,” it IEEE Trans.
developed to simulate and predict the behavior of Speech Audio Process. vol. 3, pp. 185-192. (1995)
multiple low frequency loudspeakers in rectangular
[3] R. Walker, “Equalization of Room Acoustics and
rooms. A graphical user interface has been made for
Adaptive Systems in the Equalization of Small Room
an easy use of the simulation program. An anima- Acoustics,” it Proceedings of AES 15th International
tion composed by indexed images of the instanta- Conference, Audio, Acoustics & Small Spaces, pp. 32-
neous sound pressure in the room can be obtained for 47. (November 1998)
analysis purpose. Simulations have been performed [4] J. B. Allen and Berkley, “Image Method for Efficiently
placing from one to four loudspeakers in the room Simulating small-room acoustics,” J. Acoust. Soc. Am.,
at different locations. By using more than one loud- vol. 65, pp. 943-950. (1979)
speaker a significant improvement has been observed [5] A. Krokstad, “Calculating the Acoustical Room
on the sound pressure level distribution along a lis- Response by the Use of a Ray Tracing Technique,” J.
tening area. It is confirmed that symmetrical con- of Sound and Vibration, vol. 8, pp. 118-125. (1968)
figurations remove the room modes of odd integers [6] D. Botteldooren, “Acoustical Finite-Difference Time-
and improve the spatial variation at centered posi- Domain Simulation in Quasi-Cartesian Grid,” J. Acoust.
Soc. Am., vol. 95, pp. 2313-2319. (May 1994)
tions. After the evaluation of 6 configurations with
up to 4 loudspeakers, three methods of optimiza- [7] A. Celestinos, “Multi-source low frequency room simu-
tion have been proposed and simulated on one of lation using finite difference time domain approxi-
mations,” Proc. AES 117th Convention, Convention
the configurations. Four loudspeakers in a virtual Paper 6264. (October 2004)
room have been simulated, two on the front wall an
[8] D. Botteldooren, “Finite-Difference Time
two on the rear wall at a height of 1.25 m. The pro-
Domain Simulation of Low-Frequency Room Acoustic
posed optimization methods are, first the multiple Problems,” J. Acoust. Soc. Am., vol. 98, pp. 3302-3309.
point equalization secondly the equalization of the (December 1995)
radiated power closed to the membrane of the loud- [9] S. K. Olesen, “Low Frequency Room Simulation
speakers and finally the addition of a delay and in- using Finite Difference Equations,”Proc. AES 102nd
version of phase on the rear loudspeakers. Among Convention, Preprint 4422 (D2). (March 1997)
the three methods the method of adding delay and [10] L. E. Kinsler, Fundamentals of Acoustics, 4th. Ed. John
phase inversion shown a better performance in terms Wiley & Sons, Inc. (2000)
of simplicity and efficient use of energy, since the [11] T. Welti, “How Many Subwoofers are Enough,” Proc.
rear loudspeakers act as acoustical absorbers there- AES 112th Convention, Convention Paper 5602. (May
fore a significant improvement on the sound field 2002)

44
Low Frequency Enhancement System for Rooms Paper B

[12] J. Borwick, Loudspeaker and Headphone Handbook,

Butterworth & Co. (1988)
[13] A. V. Oppenheim, Discrete-Time Signal Processing,
Second Ed. Prentice Hall. (1998), pp. 788-792
[14] T. Salava, “Acoustic Load and Transfer Functions in
Rooms at Low Frequencies”, J. Audio Eng. Soc., vol.
36, pp. 763-775 (1988 October).
[15] K. O. Ballagh, “Optimum Loudspeaker Placement Near
Reflecting Planes”, J. Acoust. Soc. Am., vol. 31, pp.
931-935 (1983 December).
[16] J. Abildgaard Pedersen, “Adjusting a loudspeaker to its
acoustic environment”, Proc. AES 115th Convention,
New York, convention paper 5880, (2003 October).
[17] S.K. Mitra, Digital Signal Processing A Computer Based
Approach, First Ed. McGraw-Hill. (1998), pp. 462-468
[18] S. J. Elliot, P. Joseph, P.A. Nelson, and M. E. Johnson,
“Power output minimization and power absorption in
the active control of sound,” J. Acoust. Soc. Am., vol.
90, pp. 2501-2512. (November 1991)
[19] M. E. Johnson and S. J. Elliot, “Measurement of acoustic
power output in the active control of sound,” J. Acoust.
Soc. Am., vol. 93, pp. 1453-1459. (March 1993)
[20] B. Langvad, H. Møller and G. Budzynski “Testing a
New Listening-Room,” Archives of Acoustics, vol. 14,
1-2, pp. 45-60. (1989)

45
Paper B Low Frequency Enhancement System for Rooms

Fig. 15. Upper left are the frequency responses at the 25 virtual microphones positions in the listening area of configuration LP5
(relocated) plotted by rows and microphone position, the first row are the microphone positions closer to front loudspeakers.
Upper right are the equalized responses by multiple point equalization. Lower left are the equalized responses by radiated
power equalization. Lower right are the optimized responses by modification of delay and phase of rear loudspeakers.

46
Low Frequency Enhancement System for Rooms Paper B

Fig. 16. Frequency responses at the 25 virtual microphones positions in the listening area plotted by rows and microphone
position. Left (upper) column are simulations of front loudspeakers on with out optimization. Left column (lower) are simulations
with front loudspeakers and rear ones with delay and phase inversion. Right (upper) column measurements of only front
loudspeakers, with out optimization. Right (lower) are measurements with front loudspeakers and rear ones with delay and
phase inversion.

47
Paper B Low Frequency Enhancement System for Rooms

Fig. 17. Graphical user interface of the room simulator used in this paper.

48
Paper C
Paper C Low Frequency Enhancement System for Rooms

50
 Low frequency sound field enhancement system for
rectangular rooms using multiple low frequency
loudspeakers
Adrian Celestinos a , and Sofus Birkedal Nielsen a
a Acoustics, Department of Electronic Systems, Aalborg University, DK-9220 Aalborg East, Denmark

Abstract

Rectangular rooms have strong influence on the low frequency performance of loudspeakers. Simulations of three
different room sizes have been carried out using finite-difference time-domain method (FDTD) in order to predict the
behavior of the sound field at low frequencies. By using an enhancement system with extra loudspeakers the sound
pressure level distribution along the listening area presents a significant improvement in the subwoofer frequency
range. The system is simulated and implemented on the three different rooms and finally verified by measurements
on the real rooms.

1. INTRODUCTION strong influence on the low frequency sound field

and thereby also on the performance of the loud-
speaker. The combination loudspeaker-room acts as
In recent years and since the advent of the a coupled system where the room properties typi-
stereophony the reproduction in high fidelity of cally dominate. This is often problematic to control
music signals has drawn the attention of many re- since the magnitude response of the loudspeaker is
searchers, professionals of the audio industry and a modified depending on the listening position and
large amount of enthusiasts. More recently with the loudspeaker placement.
arrival of the digital signal processing technology
the popularity of sound reproduction formats like To deal with this problem several approaches have
the multichannel surround systems has increased been investigated by a number of authors, over the
reasonably. From home theaters to concert hall are- last three decades among others in [1] and [2] the so-
nas it is possible to experience sound through pow- lutions are based on finding the optimum placement
erful loudspeakers. When a loudspeaker is placed of the loudspeakers in the room and in [3] the ap-
in a room a number of problems arise. Modification proach is based in the use of multiple subwoofers on
of the response of the loudspeaker at the listening different configurations. Other approaches are based
position occurs due to the reflection of sound at on the control of the acoustic radiation power as in
the walls of the enclosure and the position of the [4], and a large amount of research has been done
loudspeaker. Sound reproduction systems are typ- on the approach of modeling the correct electrical
ically placed in small or medium size rectangular filters some times called modal equalization in [5],
rooms and in some cases large halls. Every room has or as in [6] that by means of adaptive filtering tech-
Paper C Low Frequency Enhancement System for Rooms

niques an specific listening position in the room or an wooden and the walls are quite reflective. The room
extended listening area is equalized (multiple point has a double metal door in one of the side walls. The
equalization). An interesting work done in [7] where room B is a multichannel listening room of approx.
an equalization system based on the simulation of 172 m3 that conforms to the recommendation ITU-
a plane wave in a small room seems to be a suit- R BS 775-1 for multichannel surround setups. The
able approach to come about to a solution to this walls of this room are quite damped except the back
complex problem even though this solution needs a wall that has large windows that cover most of the
large amount of loudspeakers and a large amount of wall. The ceiling is covered with special plaster pan-
measurements before the system is working. els. The floor is wooden and it has two metal doors
placed symmetrically on the side walls. The room C
In this paper the main goal is to improved the low is a hall of approx. 1200 m3 used as a concert hall for
frequency sound field in an extended listening area live performances of pop music. The floor is wooden
of three rectangular rooms by using multiple loud- and the ceiling has three section levels with the last
speakers. The idea is to excite only certain room section of the ceiling resting on four columns. All
modes by using constructive an destructive phase three rooms have a general characteristic of being
interference and to create traveling sound in one end rectangular and are used with sound reproduction
of the room and cancel the sound in the opposite systems. In Table 1 the room dimensions and the
wall by using extra loudspeakers delayed and in anti estimation of room parameters such as reverbera-
phase. This approach was described before in [9] by tion time T60 are shown. The reverberation time is
the authors. calculated from the measurements described in Sec-
In this paper the analysis at low frequencies of the tion 2.3 and using the loudspeaker setup 0.2.0. see
three rectangular rooms is performed. Simulations Section 2.2. The T60 is estimated as described in
are performed using a program based on finite dif- [13] from the 10 dB energy drop using the Schroeder
ference time domain approximations (FDTD) and backward integration method [14]. Also in Table 1
finally measurements in the real rooms are presented the Schroeder frequency (f g) is calculated accord-
testing the proposed equalization system. ing to
r
T 60
f g = 2000 (1)
2. ANALYSIS OF THE SOUND FIELD ON V
THREE RECTANGULAR ROOMS
where T60 is the reverberation time in seconds and
V is the volume of the room, this frequency can be
In this section the description of the three rooms
taken as the upper limit where the discrete standing
is given and the analysis of the sound field at low
waves predominate and the simplifications of the
frequencies produced by typical sound reproduction
statistical theory of sound field in enclosures can not
systems placed in the three different room sizes is
be applied [12], [13].
presented.
Following on the analysis of the rooms the first 25
room modes of the three enclosures are shown in
2.1. Room description Table 2 according to
s
c  n 2  n 2  n 2
x y z
The three different rooms have been simulated us- fn = + + (2)
ing a program based on the finite-difference time- 2 Lx Ly Lz
domain method (FDTD) presented by the authors
in [8] and [9]. The room A is a standard listening where c is the speed of sound in the air, nx , ny and
room of approx. 90 m3 that fulfill the IEC 268-13 nz are integers starting with 0, 1, 2,... and Ly , Lx ,
standard, which describes an average living room. Lz are the dimensions of the room [13].
This room has been well studied in [10] and [11].
The ceiling is a false ceiling tilted in the corners and The number of modal frequencies per 1 Hz and the
covered with special plaster panels with three dif- number of room modes both below the frequency f
ferent sections of absorptive materials. The floor is are computed according to equations

52
Low Frequency Enhancement System for Rooms Paper C

Table 1
Room dimensions, reverberation time (T60) in seconds, Schroeder frequency (f g), number of room modes (N ) below Schroeder
frequency and below 100 Hz, number of modal frequencies per 1 Hz (∆Nf ) below the Schroeder frequency and below 100 Hz.

Room L×W×H (m) V (m3 ) T60 (s) f g (Hz) N ∆Nf (Hz) N ∆Nf (Hz)
→ fg → fg →100 Hz →100 Hz
A 7.80×4.12×2.78 89.34 0.47 145 49.72 0.81 20.13 0.45
B 8.12×7.39×2.88 172.82 0.31 85 23.20 0.65 34.60 0.84
C 25.00×12.25×3.90 1194.40 0.89 55 42.16 1.83 190.26 4.95

f2 π f L 25 virtual microphones equally spaced by 48 cm at

∆Nf = 4πV 3
+ S 2+ (3)
c 2 c 8c a height of z=1.26 m. The configuration 2.1.0 was
tested in room B using the same full range loud-
and speakers as in room A and a subwoofer which has a
cut off frequency of 28 Hz. The subwoofer is located
4π  f 3 π  f 2 L f
N= V + S + (4)
3 c 4 c 8 c Table 2
The first 25 room modes of rooms A, B and C.
where S is the area of all walls 2(Lx Ly + Lx Lz + Room A Room B Room C
Ly Lz ), V is the volume of the enclosure and L =
4(Lx + Ly + Lz ) the sum of all edge lengths of the ny nx nz f n Hz ny nx nz f n Hz ny nx nz f n Hz
room [13]. 100 22 100 21 100 7
010 41 010 23 200 14
200 44 110 31 010 14
2.2. Sound Field Room Simulations
110 47 200 42 110 16
210 61 020 46 210 20
A typical setup of loudspeakers is simulated in each 001 63 210 48 300 21
of the rooms. Some of the details of each room like
300 66 120 51 310 25
the different ceilings, windows, columns and metal
doors are included in the simulation model. For sim- 101 66 001 60 400 28
plicity in the next sections the following notation is 011 75 220 63 020 28
introduced 201 77 101 63 120 29

Nr. of front . Nr. of front wall . Nr. of back wall 310 78 300 64 410 31
111 78 011 64 220 31
full range subwoofers subwoofers
020 82 111 67 500 34
to indicate for example a stereo setup of two full
120 85 310 68 320 35
range loudspeakers the notation 2.0.0 is used. For
a stereo setup of two full range loudspeakers plus a 211 87 030 70 510 37
subwoofer the notation 2.1.0 is used. For example 400 88 130 73 420 39
the notation 0.2.2 indicates a configuration with two 301 91 201 73 600 41
subwoofers in the front wall of the room and two
220 93 021 76 030 42
subwoofers on the back wall feed with a different
signal. 410 97 211 77 130 42
311 100 121 78 001 44
The configuration 2.0.0 has been simulated in
021 103 320 79 610 44
the room A, the two loudspeakers are located
at y=1.74 m from the front wall at a height of 320 105 230 82 520 44
z=1.26 m, they were simulated as two full range 121 105 400 85 230 44
type loudspeakers with a cut off frequency of 40 Hz. 401 108 301 87 101 45
The sound field is sampled in a listening area of
500 110 221 87 201 46
1.92 × 1.92 m centered in the room delimited by

53
Paper C Low Frequency Enhancement System for Rooms

Fig. 1. Simulation of sound pressure distribution of room A using setup 2.0.0. Left, SPL distribution produced by 44 Hz (modal
frequency). Middle, SPL distribution produced by 55 Hz (anti modal frequency). Right, SPL distribution produced by 66 Hz
(modal frequency).

0 −15
MDstd = 4.99 dB
SVstd = 4.90 dB −20
−10
−25
(dB)

−20 −30
(dB)

−35
−30
−40

−40 0

0.5
−50 100
10 100 1
Frequency (Hz) Time (s) 10 Frequency (Hz)

Fig. 2. Simulation of room A. Left, frequency response at the 25 virtual microphone positions produced by the setup 2.0.0.
Right, cumulative spectral decay (CSD) at one of the virtual microphone position.

on the floor at y=0 and x=Lx/2. The full range In Figs. 2, 4 and 6 the frequency and time analysis
loudspeakers were placed at y=1.89 m from the of the three rooms is presented. The frequency re-
front wall at a height of z=1.17 m. The sound field is sponse of the 25 virtual microphone positions along
sampled in a listening area of 2.88 × 2.88 m centered the listening area in rooms A, B and C is shown
in the room delimited by 25 virtual microphones respectively. The indicator Magnitude Deviation
equally spaced by 72cm at a height of z=1.17 m. In (M Dstd ) is an average of the 25 standard devia-
this case the crossover frequency was set to 85 Hz tions in the frequency range from 30 Hz to 150 Hz
using second order IIR Butterworth filters. As for from the ideal desired signal that in this paper is
the room C which is a concert hall the setup 0.2.0 the anechoic response of the loudspeaker, a value
was simulated using two subwoofers as this is the of M Dstd = 0 dB represents an ideal anechoic re-
normal setup to reproduce the low frequency con- sponse. The indicator Spatial Deviation (SVstd ) is
tent on live concerts at this venue. The sound field is the standard deviation per single frequency from the
sampled in a listening area of 4.8×4.8 m in the room mean value across all positions, a value of SVstd =
from y=8.85 m to y=13.65 m in the y direction and 0 dB indicates that all magnitude responses are
from x=3.75 m to x=8.55 m in the x direction de- identical along the whole listening area. The SVstd
limited by 25 virtual microphones equally spaced by is the mean of the deviations SVstd in the range
1.20 m. The location of the listening area is where from 30 Hz to 150 Hz. Additionally on the waterfall
most of the audience stay during the concerts. plots from one of the impulse responses in the lis-
tening area the cumulative spectral decay (CSD) is

54
Low Frequency Enhancement System for Rooms Paper C

Fig. 3. Simulation of sound pressure distribution of room B using setup 2.1.0. Left, SPL distribution produced by 42 Hz (modal
frequency). Middle, SPL distribution produced by 53 Hz (anti modal frequency). Right, SPL distribution produced by 64 Hz
(modal frequency).

−20
−10
−25

−20 MDstd = 4.06 dB

SVstd = 4.63 dB −30
(dB)

−35
−30
(dB)

−40

−40 −45

0
−50
0.5
−60 100
10 100 1
10
Frequency (Hz) Time (s) Frequency (Hz)

Fig. 4. Simulation of room B. Left, frequency response of the 25 virtual microphone positions produced by the setup 2.1.0.
Right, cumulative spectral decay (CSD) at one of the virtual microphone position.

calculated. This is done by applying a sliding rect- (MLS) of order N =14 with sampling frequency
angular window of 1s and calculating the discrete f s=8k Hz and analyzed by the discrete Fourier
Fourier transform (DFT) on the impulse response transform (DFT). The measurements of rooms A, B
to be analyzed [5], [15]. and C are presented in Fig. 7, where also the cumu-
lative spectral decay (CSD) on one of the impulse
The analysis of the sound pressure distribution can responses in each room is calculated.
be done by observing figures 1, 3 and 5 where the
sound pressure level (SPL) distribution along the As it can be seen from the simulations and measure-
room is plotted as surface plots. The frequencies cho- ments the sound field at low frequencies presents
sen corresponds to two modal frequencies and one high variations in magnitude, in some cases more
anti modal frequency. than ±20 dB. The response of the loudspeaker varies
from one position to another due to the standing
waves and parallel walls. These variations are de-
2.3. Measurements pendent of the size of the room and both the loud-
speaker and the listener position. It is also notice-
able that the room modes have stronger influence in
In order to verify the simulations measurements the smallest room. That can be explained because
have been carried out on the three real rooms in- the separation between modes become smaller as
cluding only the 25 microphone positions equally the room size increases. However when looking at
distributed in the listening area. The measurements the temporal responses one can observe in Figs. 4
were done by using maximum length sequences

55
Paper C Low Frequency Enhancement System for Rooms

Fig. 5. Simulation of the sound pressure level distribution of room C using setup 0.2.0. Left, produced by 34 Hz (modal
frequency). Middle, produced by 45 Hz (anti modal frequency). Right, produced by 55 Hz (modal frequency).

−10
−20

−20 MDstd = 3.81 dB

SVstd = 3.73 dB (dB) −30

−30
−40
(dB)

−40 −50

−50 0

0.5
−60
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

Fig. 6. Simulation of room C. Left, frequency response of the 25 virtual microphone positions produced by the setup 0.2.0.
Right, cumulative spectral decay (CSD) at one of the virtual microphone position.

and 6 that in room B and C the resonances are not tronic equalization would not be the best solution
as noticeable as in room A. Nevertheless the modal since the range of most equalizers would not be suf-
resonances are still there and that can be confirmed ficient to compensate for example a notch of - 20 dB
in Fig. 3 and Fig. 5. From the three rooms the one at 50 Hz. Even if it was possible to compensate a
that has more problems is room A since it presents notch of -20 dB the loudspeaker would not handle
the highest magnitude and spatial variations. In this the high boost and it will introduce large amount
case some of the low frequencies will sound boomy of distortion. In some cases electronic equalization
and others would be highly attenuated. In the three may work at one single position but it will make it
cases the spatial variations and magnitude devia- worse at some other positions. In [1] the optimum
tions become less problematic as the frequency in- loudspeaker placement relative to the listener posi-
creases. tion in the room has been investigated as well as in
[3] by using more than two subwoofers. Other solu-
tions are the so called multiple point equalization
3. THE EQUALIZATION SYSTEM in [6] where the sound field has to be sampled by
a distribution of microphones in order to find the
best suitable filters before the loudspeaker in the re-
During the last years several attempts have been production chain. Other approach in [4] attempts
carried out in order to tackle the problem of loud- to control the acoustic radiation power of the loud-
speakers in rooms at low frequencies. As learned in speakers and adjusting it to its environment, in this
section 2 the response of a loudspeaker in an en- approach the volume velocity of the loudspeaker has
closure would give peaks and notches of more than to be known in order to calculate the acoustic radi-
20 dB in magnitude difference, in these cases elec-

56
Low Frequency Enhancement System for Rooms Paper C

20 10

5
MDstd = 4.50 dB
10

(dB re. Pa/V)

SVstd = 4.47 dB
0
(dB re. Pa/V)

0 −5

−10
−10
−15

−20 0

0.5
−30 100
10 100 1
Frequency (Hz) Time (s) 10 Frequency (Hz)

20
MDstd = 4.47 dB 5
SVstd = 5.54 dB
10 0
(dB re. Pa/V)

−5
(dB re. Pa/V)

0
−10

−10 −15

−20
−20 0

0.5
−30 100
10 100 1
Frequency (Hz) Time (s) 10 Frequency (Hz)

20 10
MDstd = 4.37 dB
SVstd = 4.69 dB 5
10
(dB re. Pa/V)

0
(dB re. Pa/V)

0 −5

−10
−10
−15

−20 0

0.5
−30
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

Fig. 7. Upper, measurements on room A, setup 2.0.0. Middle, measurements on room B, setup 2.1.0. Lower, measurements
on room C, setup 0.2.0. Left column, frequency response of the 25 virtual microphone positions. Right column, cumulative
spectral decay (CSD) at one of the virtual microphone position.

57
Paper C Low Frequency Enhancement System for Rooms

75Hz 75Hz 75Hz

90 90 90

80 80 80
SPL (dB)

SPL (dB)

SPL (dB)
70 70 70

60 1 60 1 60 1
2 2 2
3 3 3
50 4 50 4 50 4
5 5 5
1 6 1 6 1 6
2 2 2
3 7 Length y(m) 3 7 Length y(m) 3 7
4 4 4 Length y(m)
Width x(m) Width x(m) Width y(m)

Fig. 8. Simulation of sound pressure distribution in a rectangular room measured at a height of z=1.38 m, driven frequency
75 Hz, room mode (0 1 1). Left plot, setup 0.1.0 the loudspeaker is located at z=0.06 m and x=Lx/2 and y=0. Middle plot,
setup 0.2.0 loudspeakers at z=0.06 m. Right plot, loudspeakers at z=1.38 m.

75Hz 75Hz 75Hz

2.26 2.26 2.26

Height z(m)

Height z(m)

Height z(m)
1.5 1.5 1.5

0.26 0.26 0.26

7 6 5 4 3 2 1 7 6 5 4 3 2 1 7 6 5 4 3 2 1
Length y(m) Length y(m) Length y(m)

Fig. 9. Simulation of sound pressure distribution in the same room as Fig. 8 measured at the vertical plane x=2.10 m, driven
frequency 75 Hz, room mode (0 1 1). Left plot, setup 0.1.0 the loudspeaker is located at z=0.06 m. Middle plot, setup 0.2.0
loudspeakers at z=0.06 m. Right plot, loudspeakers at z=1.38 m.

ation power. Lz =2.76 m similar to room A has been considered as

an example but assuming an absorption coefficient
of α=0.12 in all walls instead. On the left plot of Fig.
3.1. Creation of a Plane Wave 8 the sound pressure level distribution at z=1.38 m
is been simulated, one loudspeaker driven by 75 Hz
is located at y=0.06 m, x=Lx /2 and z=0.06. It can
To achieve optimum sound pressure level distribu- be observed that the reflection of the side walls and
tion within an extended listening area inside a rect- the ceiling produce destructive interference and it
angular enclosure of volume V = Lx Ly Lz and as- has not been able to create a plane wave traveling
suming a number of sound sources on the wall at y=0 in the y direction (notice that 75 Hz corresponds to
and a number of sound sources at the wall y = Ly , the room mode (0 1 1)). In the middle plot of Fig.
a traveling plane wave in the y direction has to be 8 two loudspeakers have been replaced instead at
simulated and only the axial modes corresponding y=0.06 m, x=Lx /4 and x=3Lx /4 respectively and
to this direction should be exited. at z=0.06 m driven by the same frequency (75 Hz).
One can observe that the interference caused by the
By placing the loudspeakers equidistantly in the x
side walls and both loudspeakers is been used to at-
an z directions mostly the axial modes in the y di-
tenuate the room mode corresponding to the x di-
rection will be exited and the amplitude of the other
rection and a traveling wave along the y direction
modes will be reduced significantly [7]. It has been
exists, still the attenuation caused by the standing
found that actually with a total of two sound sources
wave corresponding to the z direction is present see
placed at y=0, x = Lx /4 and x = 3Lx /4 respec-
middle plot in Fig. 9. That is alleviated by relocat-
tively and at a height z = Lz /2 a plane wave can be
ing the loudspeakers at z=Lz /2 which can be seen
created reducing the amplitude of the room modes
in the right plots in Figs. 8 and 9. This configura-
corresponding to (0 2 0) and (0 0 1) and their com-
tion should ideally create a traveling plane waves in
binations see Table 2 in Section 2.
the y direction at all frequencies below the modal
A room of dimensions Lx =4.20 m, Ly =7.8 m and frequency 103 Hz (0 2 1). In [7] it has been found

58
Low Frequency Enhancement System for Rooms Paper C

44Hz 55Hz 66Hz

90 90 90

80 80 80
SPL (dB)

SPL (dB)

SPL (dB)
70 70 70

60 1 60 1 60 1
2 2 2
3 3 3
50 4 50 4 50 4
5 5 5
1 6 1 6 1 6
2 Length (m) 2 Length (m) 2 Length (m)
3 7 3 7 3 7
4 4 4
Width (m) Width (m) Width (m)

Fig. 10. Simulation of sound pressure distribution measured at a height of z = 1.38m produced by setup 0.2.0 before equalization,
Left plot, driven frequency 44 Hz. Middle plot, driven frequency 55 Hz. Right plot driven frequency is 66 Hz.

44Hz 55Hz 66Hz

90 90 90

80 80 80
SPL (dB)

SPL (dB)

SPL (dB)
70 70 70

60 1 60 1 60 1
2 2 2
3 3 3
50 4 50 4 50 4
5 5 5
1 6 1 6 1 6
2 Length (m) 2 Length (m) 2 Length (m)
3 7 3 7 3 7
4 4 4
Width (m) Width (m) Width (m)

Fig. 11. Simulation of sound pressure distribution measured at a height of z = 1.38m after equalization produced by setup
0.2.2, Left plot, driven frequency 44 Hz. Middle plot, driven frequency 55 Hz. Right plot driven frequency is 66 Hz.

that an approximation of the maximum frequency N including a delay according to the traveling dis-
that can be equalized is given by fmax = c/d − ∆ε tance in the y direction of the plane wave. In ad-
where c is the speed of sound and d is the distance in dition the gain G of the extra loudspeakers has to
the x direction between two adjacent loudspeakers, be adjusted due to the attenuation of sound by the
and ∆ε is a constant that depends on the damping traveling distance and the damping characteristics
of the room. of the room. In Fig. 12 the block diagram of the
equalization system is shown.

3.2. Removing the Reflection from the Back Wall

As room modes or modal resonances are caused by

reflections and standing waves the obvious way to
reducing or removing these modes is to remove the
reflection which has to be made in the time domain
and it will ideally work for all frequencies. In order to
create a traveling plane wave in the y direction the
reflection of sound on the back wall has to be mini-
mized. This is achieved by placing the same number
of extra loudspeakers in antiphase with the sound Fig. 12. Block diagram of the equalization system to mini-
pressure at the back wall. mize the reflection of the back wall, G its a factor according
to the damping characteristics of the room and the attenu-
These loudspeakers are fed with the same signal as ation of sound by the air.

59
Paper C Low Frequency Enhancement System for Rooms

0
−10
MDstd = 5.23 dB
SVstd = 4.98 dB −15
−10
−20
−25

(dB)
−20 −30
(dB)

−35
−30 −40
−45
−40 0

0.5
−50
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

−10
−10
MDstd = 2.39 dB
SVstd = 1.83 dB −15
−20
−20
−25
(dB)

−30 −30
(dB)

−35
−40 −40
−45
−50 0

0.5
−60
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

Fig. 13. Simulation of the equalization system in room A setups 0.2.0 and 0.2.2. Left (upper, middle), frequency responses
at the 25 positions, (upper) before equalization (middle) after equalization. Right (upper, middle), cumulative spectral decay
(CSD) at one position, (upper) before equalization, (middle) after equalization. Lower, SPL distribution at z=1.26 m, driven
frequency 40 Hz, (left) before equalization, (right) after equalization.

60
Low Frequency Enhancement System for Rooms Paper C

−10 −20

MDstd = 4.33 dB −25

−20 SVstd = 4.44 dB −30
−35

(dB)
−30 −40
(dB)

−45
−40 −50
−55
−50 0

0.5
−60 100
10 100 1
Frequency (Hz) Time (s) 10 Frequency (Hz)

−10 −20
MD = 3.58 dB
−25
SV = 2.85 dB
−20 −30
−35
(dB)
(dB re. Pa/V)

−30 −40
−45
−40 −50
−55
−50 0

0.5
−60 100
10 100 1
Frequency (Hz) Time (s) 10 Frequency (Hz)

Fig. 14. Simulation of the equalization system in room B setups 0.2.0 and 0.2.2. Left (upper, middle), frequency responses
at the 25 positions, (upper) before equalization (middle) after equalization. Right (upper, middle), cumulative spectral decay
(CSD) at one position, (upper) before equalization, (middle) after equalization. Lower, SPL distribution at z=1.20 m, driven
frequency 40 Hz, (left) before equalization, (right) after equalization.

61
Paper C Low Frequency Enhancement System for Rooms

−10
−20

−20 MDstd = 3.81 dB

SVstd = 3.73 dB −30

(dB)
−30
−40
(dB)

−40 −50

−50 0

0.5
−60
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

−10
−20
MDstd = 4.89 dB
−20 SVstd = 4.84 dB
−30
(dB)

−30
−40
(dB)

−40 −50

−50 0

0.5
−60
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

Fig. 15. Simulation of the equalization system in room C setups 0.2.0 and 0.2.2. Left (upper, middle), frequency responses
at the 25 positions, (upper) before equalization (middle) after equalization. Right (upper, middle), cumulative spectral decay
(CSD) at one position, (upper) before equalization, (middle) after equalization. Lower, SPL distribution at z=1.65 m, driven
frequency 40 Hz, (left) before equalization, (right) after equalization.

62
Low Frequency Enhancement System for Rooms Paper C

3.3. Optimal Equalization 4.2. Measurement of the Equalization System

The same room used in section 3.1 is used here to First the equalization system is simulated and next
demonstrate the optimal equalization system on in- measured in the real rooms A, B and C. The mea-
dividual frequencies. The setup 0.2.0 is placed at surements are presented on Fig. 16, 17 and 18, where
x = Lx /4, x = 3Lx /4, y=0.06 m and at z = Lz/2. before equalization plots (upper) and after equaliza-
In Fig. 10 the sound pressure distribution at a height tion plots (lower) are shown, left plots are frequency
z = Lz/2 is measured using 44 Hz, 55 Hz and 66 Hz response curves and right plots are CSD waterfall
as driven frequencies before equalization. In Fig. 11 plots. The impulse responses were acquired by us-
the result of the equalization system is plotted with ing MLS sequences of order N =14 with a sampling
the extra loudspeakers. As it can be observed the frequency f s=8k Hz and processed by the discrete
back wall reflection has been minimized and a travel- Fourier transform (DFT) in Matlab. The loudspeak-
ing wave in the y direction has been created. Notice ers employed were four 35 cm × 29 cm × 35 cm close
that the sound level distribution is even in almost box type active loudspeakers with a 8 in driver unit
all the room. each.
In room A the sound field was measured at 1.26 m
4. RESULTS height with 25 microphone positions equally spaced
by 48 cm within an area of 1.92 × 1.92 m centered
in the room. The loudspeakers on setup 0.2.2 were
In this section first the equalization system is simu- placed at 1.50 m height and 6cm from the front
lated on rooms A, B and C and secondly the mea- and back wall respectively. As illustrated in Sec-
surements of the equalization system in the real tion 3.1 the loudspeakers should be placed at 1.38 m
rooms are presented. height but because of the complexity of the ceiling
this height (1.50 m) was assumed to be a better
approximation of Lz /2 since the concrete ceiling is
4.1. Simulation of the Equalization System at Lz =3.10 m in the room. The gain of the back
wall loudspeakers was G=-0.95 dB and the delay
∆t=22.44 ms
In room A the listening height has been chosen
z=1.26 m and the height of the loudspeaker setup In room B the sound field was measured at 1.20 m
0.2.2 was chosen to be z=1.50 m the same as in height on 25 microphone positions equally spaced
the measurements in the real room. The details of by 72cm within an area of 2.88 m × 2.88 m cen-
the ceiling, door and floor are included in the sim- tered in the room. The loudspeakers on setup 0.2.2
ulation program. In room B the listening height is were placed at 1.44 m height and 9cm from the
z=1.17 m and the loudspeakers 0.2.2 are located at front and back wall respectively. The gain of the
a height of z=1.53 m. In room C the listening height back wall loudspeakers was G=-3.7 dB and the de-
is z=1.65 m and the loudspeakers are placed on the lay ∆t=24.6 ms
floor at z=0.15 m. This room presents an special In room C the sound field was measured at 1.65 m
difficulty since in the real room the loudspeakers height on 25 microphone positions equally spaced by
can not be placed at the very end walls. The 0.2.0 1.20 m within an area of 4.8 × 4.8 m from y=8.85 m
loudspeakers are located as the typical subwoofer to y=13.65 m in the y direction and from x=3.75 m
placement in live concerts at y=5.80 m from the to x=8.55 m in the x direction. The loudspeak-
front wall, and the loudspeakers 0.0.2 where placed ers on setup 0.2.2 were placed on the floor and the
at y=19.20 m, 5.80 meters from the back wall re- 0.2.0 loudspeakers were placed at y=5.80 m from the
spectively. The delay was adjusted according to the front wall and the 0.0.2 loudspeakers were placed at
distance from the front loudspeakers to the rear y=19.20 m, 5.80 m from the back wall respectively.
loudspeakers. The gain of the back wall loudspeakers was G= -8.5
In Figs. 13, 14 and 15 simulation of setup 0.2.2 is dB and the delay ∆t=47.40 ms. The result of the
presented before and after equalization on rooms A, measurements is shown in Figs. 16, 17 and 18.
B and C respectively.

63
Paper C Low Frequency Enhancement System for Rooms

30
20

MDstd = 4.85 dB 15
20 SVstd = 4.63 dB

(dB re. Pa/V)

10
(dB re. Pa/V)

10
5

0
0
−5
−10 0

0.5
−20
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

30
20

MDstd= 2.92 dB 15
20 SVstd= 2.42 dB
(dB re. Pa/V)

10
(dB re. Pa/V)

10
5

0
0
−5
−10 0

0.5
−20
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

Fig. 16. Measurements of the equalization system in room A, setups 0.2.0 and 0.2.2. Left, frequency responses at the 25
positions, (upper) before equalization (lower) after equalization. Right, cumulative spectral decay (CSD) at one position,
(upper) before equalization, (lower) after equalization.

4.3. Evaluation of the Equalization System B the system performed better than room A, this
can be seen in Fig. 17, notice that the modal frequen-
cies are less noticeable. Concerning the spatial vari-
As it is clearly seen from the simulations and mea- ations the system improved from having variations
surements the equalization system performed very from one position to the nearest from around 6 dB
well in room A and B. The magnitude deviation im- to 3 dB in the worse cases. By observing the simu-
proved drastically from 20 Hz to 100 Hz being dif- lations on lower plots in Figs. 13 and 14 the system
ferences in magnitude from ±15 dB to ±6 dB see removed the standing wave not as perfect as in Fig.
Figs. 13 and 16, these deviations are fixed at the 11 in Section 3.1 but the sound pressure distribution
modal frequencies specially in room A. However by improved from ± 15 dB to ± 6 dB in the range from
observing Figs. 13 and 16 the system performed bet- 10 Hz to 100 Hz the distribution is more even not
ter than the simulations at higher frequencies, one only in the listening area but also along the room.
can notice that at the simulations from 100 Hz to Unfortunately a measurement of the whole listening
200 Hz the system did not correct for those peaks plane in the room was not performed nevertheless
but in the real room those peaks were attenuated see informal listening tests have been performed verify-
Fig. 16. One should notice that at that range of fre- ing the effectiveness of the system. In room B the
quencies the equalized system added more interfer- system did worse in frequencies from 90 to 100 Hz
ence resulting in more overlapped notches. In room

64
Low Frequency Enhancement System for Rooms Paper C

20 15
MDstd = 4.54 dB
SVstd = 4.79 dB 10
10

(dB re. Pa/V)

5
(dB re. Pa/V)

0 0

−5
−10
−10

−20 0

0.5
−30
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

15
20
10
MDstd = 3.59 dB
10 SVstd = 3.24 dB
(dB re. Pa/V)

5
(dB re. Pa/V)

0
0
−5

−10 −10

−20 0

0.5
−30 100
10 100 1
Frequency (Hz) Time (s) 10 Frequency (Hz)

Fig. 17. Measurements of the equalization system in room B, setups 0.2.0 and 0.2.2. Left, frequency responses at the 25
positions, (upper) before equalization (lower) after equalization. Right, cumulative spectral decay (CSD) at one position,
(upper) before equalization, (lower) after equalization.

see Fig. 17 since it increased a peak corresponding on each room has been plotted in Figs. 19, 21 and
to the room mode (4 2 0). In room C the system did 20.
not work neither in the simulations nor in the real
room. Very small improvement is seen from 15 Hz
to 27 Hz in the measurements but in the other hand 5. DISCUSSION
it makes it worse from 30 Hz to 50 Hz.
As it is clearly seen room A presents more prob- As seen from the analysis in Section 2 when loud-
lems than room B since the modal resonances are speakers are placed in an enclosure a number of
less overlapped than in room B. Nevertheless the problems appear, magnitude deviations from ±10
equalization system performed well up to 132 Hz. In dB to ± 20 dB occur on the worse cases depending
room B the equalization system performed well up on the size and damping of the room. The deviations
to 87 Hz and in room C very small improvement is in magnitude from one position to another varies at
shown. some frequencies from ± 6 dB to cases where there
In order to have an overview of the equalization sys- is almost not sound at all. By first creating a plane
tem in a general manner the mean of the 25 fre- wave in only one direction of the room which implies
quency responses before and after the equalization exiting only the axial modes of that direction and
secondly canceling that plane wave using loudspeak-

65
Paper C Low Frequency Enhancement System for Rooms

20 10
MDstd = 4.37 dB
SVstd = 4.69 dB 5
10

(dB re. Pa/V)

0
(dB re. Pa/V)

0 −5

−10
−10
−15

−20 0

0.5
−30
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

20 10

MDstd = 4.46 dB 5
10 SVstd = 4.79 dB
(dB re. Pa/V)

0
(dB re. Pa/V)

0 −5

−10
−10
−15

−20 0

0.5
−30
10 100 100
1 Frequency (Hz)
Frequency (Hz) Time (s) 10

Fig. 18. Measurements of the equalization system in room C, setups 0.2.0 and 0.2.2. Left, frequency responses at the 25
positions, (upper) before equalization (lower) after equalization. Right, cumulative spectral decay (CSD) at one position,
(upper) before equalization, (lower) after equalization.

ers delayed at the end wall in opposite phase with 16 and 17 the modal frequencies are much more no-
the traveling sound, optimal sound level distribution ticeable in room A than in room B so the improve-
can be obtained. First the equalization system was ment is worth in room A reducing the effect of the
tested on a simulation model and afterwards vali- modal resonances but not completely, in room B in-
dated by measurements in real rooms. After having stead the modal frequency do not ring as much as
been simulated and measured the equalization sys- in room A therefore the room modes in room B de-
tem for low frequencies it can be said that the sys- cay faster than in room A in this case the improve-
tem performed well in room A and B improving both ment is not as obvious than in room A. As it was
magnitude deviations and spatial variations. Gener- observed in the results the equalization system did
ally it worked not only in the listening area but also not performed well in room C actually the problems
in the whole room. in room C are not as bad as they are in room A or
B before the equalization. One could obviously see
The system presents some variations at the modal that the improvement is not needed in room C since
frequencies, this variations are due to asymmetries the room modes are very overlapped in the region
in the room and the complexity of the ceiling in room from 30 Hz to 100 Hz. Furthermore a slightly im-
A for example, and in room B because of the differ- provement is observed at very low frequencies from
ent impedance of the front wall and back wall. Inter- 10 Hz to 27 Hz where in order to perceive those fre-
estingly seen from the right waterfall plots in Figs.

66
Low Frequency Enhancement System for Rooms Paper C

30 30

20 20
(dB re. Pa/V)

(dB re. Pa/V)

10 10

0 0

−10 −10

−20 −20
10 100 10 100
Frequency (Hz) Frequency (Hz)

Fig. 19. Room A. Thin line mean of the frequency responses Fig. 21. Room B. Thin line mean of the frequency responses
of the measurements at the 25 microphone positions before of the measurements at the 25 microphone positions before
equalization. Thick line after equalization, setup 0.2.2. equalization. Thick line after equalization, setup 0.2.2.

20 ment depending on preference. Nevertheless as it is

observed on Figs. 19 and 21 when the equalization
10
system is on still there is a considerable effect of the
room, that can be seen as a boost at low frequencies
from 30 Hz to 50 Hz. One of the drawbacks of this
(dB re. Pa/V)

0
approach is that extra loudspeakers, power ampli-
fiers and simple signal processing equipment has to
−10 be added in order to cancel the sound at the back
wall. A further research can be addressed to inves-
−20 tigate the adequate amount of equalization that is
really needed in terms of human preference.
−30
10 100
Frequency (Hz) 6. CONCLUSION
Fig. 20. Room C. Thin line mean of the frequency responses
of the measurements at the 25 microphone positions before
equalization. Thick line after equalization, setup 0.2.2.
The analysis of the low frequency performance of
sound reproduction systems in three rectangular
quencies high acoustic power is needed, in fact most rooms of different size has been done. A simulation
of the subwoofers are able to reproduce efficiently program based in FDTD has been used to render
above 30 Hz. More loudspeakers at the front and at the sound field produced by typical sound repro-
the back walls might be needed to create a plane duction systems in rectangular rooms. The three
wave with a back wall cancellation. rooms are a standard listening room, a standard
multichannel listening room and a concert hall for
To summarize the system works depending on the live performances. An effective method to equalize
size of the room, the smaller the room the more con- low frequencies in rectangular rooms has been sim-
trollable the system will be. It can be said that if the ulated and implemented in these three rooms. The
equalization system is well implemented it should system uses two loudspeakers in the front wall of
work up to the frequencies where a plane wave is the room to create a traveling plane wave and an
formed at the front wall. A subject for discussion extra two low frequency loudspeakers in the back
is if a complete flat response is wanted. This may wall delayed and in opposite phase to remove the
depend on personal preference but the advantages reflection of that wall. After measurements of the
of this equalization system is that it could be ad- implemented system in the three rooms one can
justed parametrically to certain degree of enhance- conclude that the system can work effectively in

67
Paper C Low Frequency Enhancement System for Rooms

small and middle size rectangular rooms. The sys- Contents

tem can achieve fairly good responses not only in a
1 INTRODUCTION 51
single listening position but also within a listening
area and at very low frequencies in the whole room. 2 ANALYSIS OF THE SOUND FIELD ON
THREE RECTANGULAR ROOMS 52

2.1 Room description 52

References 2.2 Sound Field Room Simulations 53

2.3 Measurements 55
[1] A. R. Groh, “High-Fidelity Sound System Equalization
by Analysis of Standing Waves”, J. Audio Eng. Soc., 3 THE EQUALIZATION SYSTEM 56
vol. 22, pp. 795-799 (December 1974).
3.1 Creation of a Plane Wave 58
[2] K. O. Ballagh, “Optimum Loudspeaker Placement Near
Reflecting Planes”, J. Acoust. Soc. Am., vol. 31, pp. 3.2 Removing the Reflection from the Back Wall 59
931-935 (1983 December).
3.3 Optimal Equalization 63
[3] T. Welti, “How Many Subwoofers are Enough,” Proc.
AES 112th Convention, Convention Paper 5602. (May 4 RESULTS 63
2002)
4.1 Simulation of the Equalization System 63
[4] J. Abildgaard Pedersen, “Adjusting a loudspeaker to its
acoustic environment”, Proc. AES 115th Convention, 4.2 Measurement of the Equalization System 63
New York, convention paper 5880, (October 2003).
4.3 Evaluation of the Equalization System 64
[5] A. Mäkivirta and P. Antsalo, “Modal Equalization of
LoudspeakerRoom Responses at Low Frequencies” J. 5 DISCUSSION 65
Audio Eng. Soc., 51, pp. 324-353. (May 2003)
6 CONCLUSION 67
[6] S. J. Elliott and P. A. Nelson, “Multiple-Point
Equalization in a Room Using Adaptive Digital Filters,” References 68
J. Audio Eng. Soc., 37, pp. 899-907. (November 1989)
[7] A. O. Santillan, “Spatially extended sound equalization
in rectangular rooms,” J. Acoust. Soc. Am., 110, pp.
1989-1997. (October 2001)
[8] A. Celestinos and S. B. Nielsen, “Multi-source low
frequency room simulation using finite difference time
domain approximations,” Proc. AES 117th Convention,
Convention Paper 6264. (October 2004)
[9] A. Celestinos and S. B. Nielsen, “Optimizing placement
and equalization of multiple low frequency loudspeakers
in rooms,” Proc. AES 119th Convention, Convention
Paper 6545. (October 2005)
[10] B. Langvad, H. Møller and G. Budzynski “Testing a
New Listening-Room,” Archives of Acoustics, 14, 1-2,
pp. 45-60. (1989)
[11] B. Chereck and B. Langvad, “Low Frequency Simulation
of a Listening Room,” Proc. of Nordic Acoustical
Meeting, pp. 265-270. (1990)
[12] T. Salava, “Acoustic Load and Transfer Functions in
Rooms at Low Frequencies”, J. Audio Eng. Soc., vol.
36, pp. 763-775 (October 1988).
[13] H. Kuttruff, Room Acoustics, 3th. Ed. E & FN Spon.
(1999)
[14] M. R. Schroder, “New Method of Measuring
Reverberation Time,” J. Acoust. Soc. Am., 37, pp. 409-
412. (March 1965)
[15] J. D. Bunton and R. H Small, “Cumulative Spectra,
Tone Bursts and Applications” J. Audio Eng. Soc., 30,
pp. 386-395. (June 1982)

68
Paper D
Paper D Low Frequency Enhancement System for Rooms

70
Controlled Acoustically Bass System (CABS), A method
to achieve uniform sound field distribution at low
frequencies in rectangular rooms
Adrian Celestinos a , and Sofus Birkedal Nielsen a
a Acoustics, Department of Electronic Systems, Aalborg University, DK-9220 Aalborg East, Denmark

Abstract

Rectangular rooms have strong influence on the low frequency performance of loudspeakers. A simulation program
based on the finite-difference time domain method (FDTD) has been used to analyse the sound field produced by
loudspeakers in rectangular rooms at low frequencies. A new method called Controlled Acoustically Bass System
(CABS) is introduced. The system utilizes front loudspeakers and extra loudspeakers at the opposite wall of the room
processed to remove the back-wall reflection, which will give a more uniform sound field. The system works in the
time domain and presents good performance in the low frequency range. CABS is simulated and measured on two
different standard listening rooms.

1. INTRODUCTION duction systems are typically placed in small or

medium size rectangular rooms and in some cases
large halls. Every room has strong influence on the
In recent years and since the advent of the low frequency sound field and thereby also on the
stereophony the reproduction in high fidelity of performance of the loudspeaker. The response of a
music signals has drawn the attention of many re- loudspeaker will be highly influenced by its position
searchers, professionals of the audio industry and a in the room and the room properties. This is often
large amount of enthusiasts. More recently with the problematic to control since the modal resonances
arrival of the digital technology and the new sound modify the magnitude response of the sound source
reproduction formats like multichannel surround depending on the listening position and loudspeaker
sound the popularity of these systems has increased placement.
reasonably. From home theaters to concert hall are-
nas it is possible to experience low frequency sound To deal with this problem several approaches have
through full range loudspeakers or powerful sub- been investigated by a number of authors, over the
woofers dedicated to playback frequencies from 30 last three decades among others Groh in [1], Allison
to 100 Hz. When a loudspeaker is placed in a room in [2] and Ballagh in [3] have based their solutions
a number of problems arise. Modification of the re- on finding the optimum placement of the loudspeak-
sponse of the loudspeaker at the listening position ers in the room. More recently Welti in [4] has based
occurs due to the strong reflections in the enclosure his approach on the use of multiple subwoofers on
and the position of the loudspeaker. Sound repro- different configurations in the room. Another ap-
Paper D Low Frequency Enhancement System for Rooms

proach by Abildgaard in [5] is based on the control 2.1. The building up of a standing wave in the time
of the acoustic radiation power of the loudspeaker in domain
a room. Large amount of research has been carried
out on the approach of modeling the correct elec-
trical filters often called modal equalization in [6] Traditionally the problem of low frequency sound in
by Mäkivirta and Antsalo, or the so-called multiple rooms is analysed by the modal theory which parts
point equalization technique by Elliot in [7], that by from the solution of the wave equation in lossless,
means of adaptive filtering techniques compensate rigid-walled, rectangular enclosures as described in
a specific listening position in the room or an ex- [13] pp. 349. It assumes a steady state situation pro-
tended listening area. An interesting work done by duced by the sound source driven by pure frequency
Santillán et al. in [8] and [9] where the equalization tones. In order to clearly understand the physical
system is based on the simulation of a plane wave problem it is of great importance to perform the
traveling as in free field in a small room seems to analysis in the time domain. Assuming a room of
be a suitable approach to come about to a solution only rigid walls in both ends and a loudspeaker in
to this complex problem even though this solution one end of the room. If the dimension corresponds to
needs a large amount of loudspeakers and a large multiples of half of the wavelength of the produced
amount of measurements before the system is work- sound by the loudspeaker the reflection with the op-
ing properly. posite wall will meet the sound coming from the
loudspeaker in some places with constructive phase
The main goal of this paper is to improved the low and in some places with destructive phase. The re-
frequency sound field in an extended listening area flection will return to the wall at the loudspeaker
of a rectangular room by using multiple loudspeak- exactly in phase with the sound radiating from the
ers. The idea is to built a plane wave traveling to- loudspeaker. The resulting addition of these waves
wards the opposite wall where it will be canceled. coming from reflections of the walls and the loud-
This is done by using extra loudspeakers at the back speaker itself will form sections in the room where
wall with a delayed version of the signal but in anti there is almost no sound pressure and zones where
phase. This approach was described before in [10] there is a high sound pressure level. This frequen-
and [11] by the authors. In this paper the analysis cies are commonly known as resonance frequencies,
in the time and frequency domains at low frequen- modal frequencies or natural frequencies of the room
cies in rectangular rooms is presented. Simulations given by
are performed using a program based on the finite-
difference time domain method (FDTD) and finally c

frn = n (1)
measurements in two standard listening rooms are 2L
presented testing the performance of the enhance-
ment system. where L is the length of the room, c is the speed of
sound in the air and n is an integer starting with 1, 2,
2. LOW FREQUENCY SOUND IN 3,... In the cases where the dimension corresponds to
RECTANGULAR ROOMS an odd integer times one quarter of the wavelength
there will always be a minimum sound pressure level
at the proximity of the loudspeaker and a maximum
In this section the analysis of the sound field at low at the opposite wall, those frequencies are known
frequencies produced by typical sound reproduction as anti-resonances or anti-modal frequencies of the
systems placed in rooms is presented. The analysis room given by
is divided in three parts, first the physical problem is
c

observed in the time domain secondly the problem is fan = 2n − 1 . (2)
analysed in the joint time-frequency domains by the 4L
cumulative spectral decay (CSD) and finally in the
frequency domain by the digital Fourier transforma- This examples can be observed in the sequence of
tion (DFT). The impulse responses for the analysis plots in Fig. 1 for a resonance frequency and in Fig. 2
are produced by the program based on FDTD imple- for an anti-resonance frequency where sequences of
mented in MATLAB and presented by the authors snapshots in the time domain of the instantaneous
in [10] and [12]. pressure along the room are presented.

72
Low Frequency Enhancement System for Rooms Paper D

t = 2 ms t = 5 ms t = 8 ms
1 Pa 1 Pa 1 Pa

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 4m 0 1 2 3 4m 0 1 2 3 4m

t = 11 ms t = 14 ms t = 16 ms
1 Pa 1 Pa 1 Pa

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 4m 0 1 2 3 4m 0 1 2 3 4m

t = 20 ms t = 26 ms t = 32 ms
1 Pa 1 Pa 1 Pa

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 4m 0 1 2 3 4m 0 1 2 3 4m

Fig. 1. Analysis in the time domain of a resonance frequency.

t = 3 ms t = 7 ms t = 11 ms
1 Pa 1 Pa 1 Pa

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 4m 0 1 2 3 4m 0 1 2 3 4m

t = 15 ms t = 17 ms t = 21 ms
1 Pa 1 Pa 1 Pa

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 4m 0 1 2 3 4m 0 1 2 3 4m

t = 23 ms t = 31 ms t = 38 ms
1 Pa 1 Pa 1 Pa

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
0 1 2 3 4m 0 1 2 3 4m 0 1 2 3 4m

Fig. 2. Analysis in the time domain of an anti-resonance frequency.

73
Paper D Low Frequency Enhancement System for Rooms

In the case of a three-dimensional room the natural from a reflecting wall. As the reflected wave and the
frequencies are given by arriving wave will always be in opposite phase.
s
c  nx 2  ny 2  nz 2
fn = + + (3) 2.2. Simulations on a three dimensional Virtual
2 Lx Ly Lz
Room
where c is the speed of sound in the air, nx , ny and
nz are integers starting with 0, 1, 2,... and Lx , Ly , So far the analysis has been done in Fig. 1 and Fig. 2
Lz are the dimensions of the room [14]. carrying on simulations of a room of one dimen-
The zones where there will be minimum sound pres- sion now assuming a three-dimensional room from
sure level are called nodes and the points where there now named the “Virtual Room” with rigid walls and
exists a maximum of sound pressure are called anti width Lx = 4.20 m, length Ly = 7.8 m and height
nodes [13]. The number of modal frequencies per Lz = 2.76 m which is similar to the IEC standard lis-
1 Hz and the number of room modes both below f tening room at Aalborg University. The sound field
are computed according to equations produced by a typical subwoofer shown in Fig. 3 po-
sitioned in one of the corners on the floor as shown
f2 π f L in Fig. 5 is simulated. For simplicity the following
∆Nf = 4πV + S 2+ (4)
c3 2 c 8c notation is introduced

and
. F . B
4π  f 3 π  f 2 L f
N= V + S + (5)
3 c 4 c 8 c Nr. of front wall Nr. of front wall Nr. of back wall
full range subwoofers subwoofers
where S is the area of all walls 2(Lx Ly + Lx Lz +
Ly Lz ), V is the volume of the enclosure and L = to indicate for example a stereo setup of two full
4(Lx + Ly + Lz ) the sum of all edge lengths of range loudspeakers the notation 2.0.0 is used. For
the room [14]. This is often called modal density a stereo setup of two full range loudspeakers plus a
which are descriptors that can give an estimate of subwoofer the notation 2.1.0 is used. For example
the spread of the room modes below certain fre- the notation .2.2 indicates a configuration with two
quency knowing just the dimensions of the room. subwoofers in the front wall of the room and two
subwoofers on the back wall with a different signal.
Another descriptor called the Schroeder frequency
f g is calculated according to To have an overview on how the magnitude devia-
r tions are in more than one position within the room
T 60
f g = 2000 (6) dB
V

where T60 is the reverberation time in seconds and 0

V is the volume of the room, this frequency can be
taken as the upper limit where the discrete standing
waves predominate and the simplifications of the −10
statistical theory of sound field in enclosures can not
be applied [14], [15].
−20

The irregularities in the sound pressure level distri-

bution within the room will appear not only at the −30
modal or anti-modal frequencies but also on the rest
of the frequencies where the wavelengths are long
enough comparable to the dimensions of the room. It −40
10 100 Hz
is important to say that Generally for example there
will always be a node in sound pressure level at a dis- Fig. 3. Anechoic response of a typical subwoofer measured
tance corresponding to one quarter of a wavelength near the membrane.

74
Low Frequency Enhancement System for Rooms Paper D

1 10 11 20 21
Lz
Row 1

Row 2

Row 3

Row 4

Row 5
5 6 15 16 25

Ly

Lx

Fig. 5. The three dimensional Virtual Room model and loud-

Fig. 4. The 25 virtual microphone positions in the listening speaker setup .1.0 in the corner of the room.
area of the Virtual Room, loudspeaker setup .1.0 .
dB
the sound field is sampled in a listening area of MSFD = [±5.8 dB, ±15.7 dB]

(1.92 × 1.92 m) centered in the room delimited by

−20
25 virtual microphones equally spaced by 48 cm at
a height of z = 1.38 m as seen in Figs. 4 and 5. The
frequency response at the 25 positions are presented −30
all in one plot in Fig. 6 where it is clear how severe
the deviations are and how they change according
−40
to position. In some cases the differences in magni-
tude exceed more than 25 dB along the frequency
range from 20 to 200 Hz. In Fig. 8 the sound field −50
produced by the same subwoofer now positioned off
the corner as in Fig. 7 at x = 1.26 m, y = 1.62 m
and z = 0.18 m on the floor is presented. −60
10 100 Hz

Continuing with the analysis the setup .2.0 as shown

Fig. 6. Frequency responses at the 25 virtual microphone
in Fig. 13 at one end of the Virtual Room at a height positions in the listening area of the Virtual Room, setup
z = 1.38 m is simulated. Assuming that the loud- .1.0 the loudspeaker is in the corner of the room.
speakers at low frequencies behave as omnidirec-
tional sound sources and both producing the same quency, an anti modal frequency and 60 Hz this is
signal, in this case the low pass filtered impulse computed by the simulation program at the listen-
shown in Fig. 9 and Fig. 10 which output has a fre- ing height z = 1.26 m. It is clearly seen the sections
quency range from 1 Hz to 100 Hz is used as the where there is high sound pressure level and where
input for the sound source in the simulation. Then there is a minimum level. This differences can be
the instantaneous sound pressure has been obtained more than 20 dB in the extreme points depending
by the simulation model in an horizontal slice of the on the damping of the room.
room at a height of z = 1.26 m. This is shown in
Fig. 11 where a sequence of snap shots in the time
domain of the instantaneous pressure are presented. 2.3. Time and frequency analysis
As it can be observed the combination of both loud-
speakers produces a plane wave traveling along the
length of the room towards the opposite wall. Af- After analysing the problem in the time domain it
ter reflecting to the back wall it continues back and is of interest to know how severe is the problem in
forward until it dies out. both time and frequency domains. A way to do this
analysis is by calculating the Cumulative Spectral
In Fig. 12 three graphs are presented, the sound Decay (CSD) on one of the listening positions in the
pressure level distribution produced by a modal fre- Virtual Room. The same setup .2.0 shown in Fig. 13

75
Paper D Low Frequency Enhancement System for Rooms

0.03

Lz 0.02

0.01

Ly 0

Lx

−0.01
Fig. 7. The three dimensional Virtual Room model and loud- 0 10 20 30 ms
speaker setup .1.0 off the corner of the room.
Fig. 9. Impulse response of the loudspeaker used as the input
for the simulation in the time domain of Virtual Room on
dB Section 2.2.
MSFD = [±5.9 dB, ±12.6 dB]
dB
−20

0
−30

−10
−40

−20
−50

−30
−60
10 100 Hz
−40
Fig. 8. Frequency responses at the 25 virtual microphone 10 100 Hz
positions in the listening area of the Virtual Room, setup
.1.0 the loudspeaker is off the corner of the room. Fig. 10. Frequency response of the loudspeaker used as the
input for the simulation in the time domain of Virtual Room
on Section 2.2.
has been simulated and utilizing the two subwoofers
which anechoic response is presented in Fig. 3 . The 2.4. Quantification Parameters
CSD is performed by applying a sliding “apodized
rectangular window” and calculating the discrete In Fig. 14 the frequency response at the 25 positions
Fourier transform (DFT) to the impulse response as using the setup .2.0 with two subwoofers are pre-
in [6] and [16]. The first part of this window is built sented all in one plot. Although only 5 curves can
by the raising half of a gaussian window correspond- be seen there are 25 measurements but as the sound
ing to 32 ms long applied each 64 ms from t = 0 s distributes as a plane wave the five microphones in
to the end of the impulse response at the virtual mi- one row will give the same results. It is clear that
crophone position. The impulse response is 1024 ms in some cases the differences in magnitude exceed
long. The result of this is presented in Fig. 15 on a more than 20 dB along the frequency range from 20
waterfall plot where it is clearly seen how the modal to 200 Hz.
frequencies keep going in time longer than the others
and how severe the amplitude deviations are along In order to quantify the deviations of the sound
the frequency axes. field distribution along the listening area a new pa-

76
Low Frequency Enhancement System for Rooms Paper D

t = 12 ms t = 16 ms t = 21 ms
x10 −3 x10 −3 x10 −3
Pa Pa Pa
3 3 3
2 2 2
1 1 1
0 0 0
−1 m −1 m −1 m
3 3 3
1 2 2 1 2 2 1 2 2
3 4 5 1 3 4 5 1 3 4 5 1
6 m 6 m 6 m

t = 27 ms t = 35 ms t = 46 ms
x10 −3 x10 −3 x10 −3
Pa Pa Pa
3 3 3
2 2 2
1 1 1
0 0 0
−1 m −1 m −1 m
3 3 3
1 2 2 1 2 2 1 2 2
3 4 5 1 3 4 5 1 3 4 5 1
6 m 6 m 6 m

t = 59 ms t = 77 ms t = 100 ms
x10 −3 x10 −3 x10 −3
Pa Pa Pa
3 3 3
2 2 2
1 1 1
0 0 0
−1 m −1 m −1 m
3 3 3
1 2 2 1 2 2 1 2 2
3 4 5 1 3 4 5 1 3 4 5 1
6 m 6 m 6 m

Fig. 11. Sequence of snap shots in the time domain of the instantaneous sound pressure using setup of loudspeakers .2.0 in
the Virtual Room.

dB dB dB

80 80 80

70 70 70

60 60 60

50 50 50

1 1 1
2 2 2
3 3 3
4 4 4
Ly 5 m Ly 5 m Ly 5 m
6 3 6 3 6 3
m 2 m 2 m 2
1 Lx 1 Lx 1 Lx

Fig. 12. Sound pressure level distribution resulting from the simulation of the Virtual Room measured on a plane at a height
of z = 1.38 m using setup .2.0. Left produced by 44 Hz (modal frequency). Middle, produced by 55 Hz (anti modal frequency).
Right, produced by 60 Hz.

77
Paper D Low Frequency Enhancement System for Rooms

rameter is introduced the Mean Sound Field Devi-

ation (MSFD). This measure is calculated from the
frequency response of the 25 impulse responses in
the listening zone. The MSFD expressed in Eq. (9)
Lz is conformed by two numbers, the Spatial Devia-
tion (SD) which expresses the deviations within the
space in ± dB and the Magnitude Deviation (MD)
0 which reveals the magnitude spectral deviations also
in ± dB.

Ly To calculate this parameter the magnitude of the

frequency responses of all microphone positions are
Lx arranged in a table following the pad sketched in
Fig. 4, this is done for presentation purposes since
Fig. 13. The three dimensional Virtual Room model and the arrangement will not change the result of the
loudspeaker setup .2.0 . calculation. Then the whole listening area is repre-
sented in this table where the rows are the listening
dB positions and the columns are the frequencies from
MSFD = [±4.9 dB, ±14 dB] flow = 20 to fhigh = 100 Hz this can be seen in Table
1 where an example of the first five positions and five
−10
frequencies is presented. Next the standard devia-
tion on each frequency column is calculated so that
−20 the Spatial Deviation SD is the mean of all stan-
dard deviations of individual frequencies along posi-
tions as it is expressed in Eq. (7). The same manner
−30
the standard deviation is calculated on each row po-
sition so that the Magnitude Deviation MD is the
−40 mean of all standard deviations on individual posi-
tions along frequencies as it is expressed in Eq. (8).
v
−50 fhigh u np
10 100 Hz 1 X u t 1
X
SD = (xp,i − xi )2 (7)
nf np − 1 p=1
Fig. 14. Frequency responses at the 25 virtual microphone i=flow
positions in the listening area of the Virtual Room, setup v
.2.0 . np u fhigh
1 Xu 1 X
MD = t (xp,i − xp )2 (8)
np p=1 nf − 1
i=flow
dB
 
M SF D = SD ± dB, M D ± dB (9)
−20

To illustrate this parameters an example of the de-

−30 viations of sound pressure along all positions in the
listening area at 55 Hz is presented in Fig. 16. As
it can be observed the parameter SD reveals devia-
−40
tions within ±6.9 dB. In Fig. 17 an example of the
Magnitude Deviation M D is shown where there are
0
deviations of more than ±9 dB at one of the posi-
0.5 tions. In Fig. 18 the complete Table 1 is plotted as a
100 Hz surface plot to visualize the deviations of the sound
1s
10 field in the complete listening area.
Fig. 15. Cumulative spectral decay at the labeled 17 listening So far the MSFD has been calculated from a fre-
position in the Virtual Room, setup .2.0 . quency domain transformation but it is of interest

78
Low Frequency Enhancement System for Rooms Paper D

Table 1
Example of table for calculations of Mean Sound Field Deviation (MSFD).
M ic.P osition F requency (Hz) M D ± (dB)
20 21 22 23 24 25 . . . 100
1 -53.56 -47.27 -41.00 -34.81 -28.95 -26.00 . . . -29.77 14.55
2 -54.30 -52.55 -47.20 -40.20 -33.47 -29.85 . . . -43.52 14.68
3 -48.81 -47.72 -46.70 -45.00 -40.71 -36.63 . . . -28.91 13.87
4 -45.00 -42.89 -40.67 -38.32 -36.28 -37.27 . . . -27.23 13.90
5 -42.42 -39.80 -36.91 -33.71 -30.58 -30.23 . . . -33.12 13.08
. . . . . . . . . .
.. .. .. .. .. .. .. .. .. ..

25 -42.42 -39.80 -36.91 -33.71 -30.58 -30.23 . . . -33.12 13.08

SD ± (dB) 4.75 4.46 3.99 4.13 4.27 4.40 . . . 5.42

dB dB
SD = ±6.9 dB
−20 MD ±8.8 dB
−10

−25

−20
−30

−30
−35

−40 −40
5 10 15 20 25
Microphone positions
−50
10 100 Hz
Fig. 16. Example of the spatial deviations along the listening
area at 55 Hz. Dashed lines show the range of the standard
deviation, the horizontal line is the mean. The parameter Fig. 17. Example of the frequency response at position 11
SD reveals deviations up to ±6.9 dB, loudspeaker setup .2.0 in the listening area, the parameter MD reveals spectral
in the Virtual Room. deviations up to ± 8.8 dB, loudspeaker setup .2.0 in the
Virtual Room.
to have also a measure that can give information
from the time responses. An interesting parameter in
room acoustics named Definition used by Thiele in
[17] and originally called in German “Deutlichkeit” limit of integration 0 is the arrival of the direct
was chosen to give a criterion of the ratio of energy sound. An anechoic impulse response of a loud-
between the early part of the impulse response and speaker would give about 99% of D. On the other
the remaining part. The Definition (D) is obtained hand a loudspeaker measured in a normal living
by room would give lower percentages of D. The same
manner as the MSFD the frequency range of the
50ms analysis was from 20 Hz to 100 Hz therefore to ex-
Z
tract this number the impulse responses were low
[g(t)]2 dt
pass filtered before the calculation. In Fig. 19 the
D = Z0∞ 100% (10) first 500 ms of the 25 impulse responses align in time
are shown after the calculation of D. As expected
[g(t)]2 dt the parameter D = 56.4 % revealed a high influence
0 of the room on the time responses. It is also obvious
just by inspection of the impulse responses to see
where [g(t)] is the impulse response and the lower how the sound highly interacts with the room.

79
Paper D Low Frequency Enhancement System for Rooms

dB
dB
MSFD = [±4.9 dB, ±14 dB]
−10
0
−20

−30
−10
−40

−50 −20

10 −30
Mic.Pos.
15

20 100 Hz
−40
25
20

Fig. 18. Mean Sound Field Deviation (MSFD) table pre- −50
10 100 Hz
sented as a surface plot, setup .2.0 .

Fig. 20. Thin curve, frequency response at the microphone

−3 position before one point equalization. Thick curve same
x 10
1.5 microphone position after equalization, Virtual Room, loud-
speaker setup .2.0 .
1 D = 56.4%
typically by sampling the sound field with a micro-
phone at a listening position and designing a filter
0.5 with the different known adaptive techniques it may
work at one single position but it will make it worse
0 at some of the other positions.
This is illustrated in Fig. 20 where the frequency
−0.5
responses at the listening position before and af-
ter equalization are shown. The equalization filter
−1 design is performed in MATLAB by the method
of frequency sampling-based digital Finite Impulse
−1.5 Response (FIR) filters with arbitrarily shaped fre-
0 50 100 150 200 250 300 350 400 450 ms
quency response [18] [19].
Fig. 19. The 25 impulse responses of the listening area in the
Virtual Room, loudspeaker setup .2.0 after the calculation
The target filter response is the anechoic response of
of Definition D. the loudspeaker. It can be seen that the response has
been corrected in that particular position and some
2.5. Traditional one point equalization other positions but in contrast the remaining posi-
tions got worse now having boosted 20 dB a peak at
76 Hz, that can be seen in Fig. 21 and Fig. 22. Com-
As learned in this section the response of loudspeak- paring with Fig. 14 and Fig. 18 the MSFD does not
ers in small or middle size rooms would give peaks present a clear improvement in the complete listen-
and notches of more than 20 dB in magnitude dif- ing area. The only improvement is in the listening
ference, in these cases electronic equalization would position where the parameter D went from 57.0 %
not be the best solution since the range of most to 94.8 % shown in Fig. 23. However by looking at
equalizers would not be sufficient to compensate for the 25 impulse responses before and after the equal-
example a deep of -20 dB at 50 Hz. Even if it was ization in Fig. 24 it is clear that the responses did
possible to compensate a deep of -20 dB the loud- not get any improvement going from a D = 59.0%
speaker would not handle the high boost and it will to a D = 56.8%. Here the parameter D was calcu-
introduce large amount of distortion. In some cases lated up to 200 Hz in both situations (before and
the traditional electronic equalization implemented after equalization) since the equalization filter was

80
Low Frequency Enhancement System for Rooms Paper D

dB −3
x 10
MSFD = [±4.9 dB, ±16.6 dB] 5

4
0
3 D = 57.0%

2 D = 94.8%
−10
1

−20 0

−1

−30 −2

−3

−40 −4

−5
0 50 100 150 200 250 300 350 400 450 ms
−50
10 100 Hz Fig. 23. Gray curve, impulse response of the listening posi-
tion before equalization. Black curve the same position after
Fig. 21. Thin curves frequency responses of the 25 virtual mi- the equalization resulting of simulations of Virtual Room,
crophone positions in the listening area of the Virtual Room loudspeaker setup .2.0 .
after the traditional one point equalization. Thick curve is
the microphone position equalized, loudspeaker setup .2.0 .
−3
x 10
5
dB
MSFD = [±4.9 dB, ±16.6 dB] 4

−10 3 D = 59.0%

−20 2 D = 56.8%

−30 1

0
−40

−1
−50
−2
5
−3
10
Mic.Pos.
15 −4

20 100 Hz −5
0 50 100 150 200 250 300 350 400 450 ms
25
20

Fig. 24. Gray curves, the 25 impulse responses at the listening

Fig. 22. Mean Sound Field Deviation (MSFD) after the tra- area in the Virtual Room before equalization. Black curves,
ditional one point equalization, Virtual Room, loudspeaker after one point equalization. Loudspeaker setup .2.0
setup .2.0 .
equalization may work in a limited listening posi-
design to compensate to this frequency limit. tion while worsening the responses elsewhere in the
room.

3. UNIFORM SOUND PRESSURE A way to improve the sound pressure distribution

DISTRIBUTION IN THE ROOM in the whole room is to remove the reflection from
the back wall. This can be inferred after simulating
the setup .2.0 in the Virtual Room but in this case
As learned in Section 2 the response of a loudspeaker removing the back wall. This is done by setting the
placed in an enclosure will give irregular sound pres- impedance of that boundary to the impedance of
sure level distribution in the room due to the multi- the air. The results of this simulation are shown in
ple reflections of the sound to the walls. Electronic Fig. 25 and Fig. 26. It is clear that the sound field is

81
Paper D Low Frequency Enhancement System for Rooms

dB
MSFD = [±0.3 dB, ±13.1 dB]
dB

−20 80

70
−30
60

−40 50

1
2
−50 3
4
Ly 5 m
6 3
m 2
−60 1
10 100 Hz Lx

Fig. 25. Frequency response at the 25 positions resulting from Fig. 26. Sound pressure level distribution resulting from the
the simulation of the Virtual Room removing the back wall simulation of the Virtual Room removing the back wall by
by setting the impedance of that boundary to the impedance setting the impedance of that boundary to the impedance of
of the air, loudspeaker setup .2.0 . the air. Sound pressure measured on a plane at a height of
z=1.38m using loudspeaker setup .2.0 and driven frequency
of 44 Hz.
uniform in the whole room and very small spectral
deviations exists up to 110 Hz. However one must
be sure to create a plane wave traveling along the Considering the Virtual Room used in Section 2 as
room, in this manner the problem becomes unidi- an example simulations have been carried out using
mensional since the room will only be exited in one an absorption coefficient of α = 0.12 in all walls. On
direction. Instead of removing physically the back the upper left plot of Fig. 27 the sound pressure level
wall or having a full absorbing back wall which is distribution is measured at a height of z = 1.38 m,
unpractical and almost impossible to achieve, this one loudspeaker driven by 75 Hz is located on the
front wave can be canceled out exactly at this point floor at z = 0.06 m in one end of the room at y =
by producing the frontal sound delayed at the back 0.06 m and x = Lx /2. It can be observed that the
wall but in opposite phase and with proper ampli- reflection of the side walls and the ceiling produce
tude. destructive interference and it has been unable to
create a plane wave traveling in the y direction no-
tice that 75 Hz corresponds to the room mode (0 1 1)
3.1. Construction of a Plane Wave see Table 6. In the upper middle graph of Fig. 27 two
simulated loudspeakers have been replaced instead
at y = 0.06 m, x = Lx /4 and x = Lx 3/4 respec-
To achieve optimum sound pressure level distri- tively and at z = 0.06 m on the floor with the same
bution inside a rectangular room of volume V = driven frequency of 75 Hz. One can observe that the
Lx Ly Lz and assuming N number of sound sources interference caused by the side walls and both loud-
on the wall at y = 0 a traveling plane wave in the y speakers has attenuated the room mode correspond-
direction has to be constructed. ing to the x direction and a traveling wave along
the y direction exists, still the attenuation caused
By placing the loudspeakers equidistantly in the x by the standing wave corresponding to the z direc-
an z directions mostly the axial modes in the y di- tion is present see lower middle graph in Fig. 27.
rection will be exited and the amplitude of the other That is alleviated by relocating the loudspeakers at
modes will be reduced significantly [9]. It has been a height of z = Lz /2 which can be seen in the upper
found that actually with a total of two sound sources and lower right graphs on Fig. 27. This configura-
placed at y = 0, x = Lx /4 and x = Lx 3/4 respec- tion should ideally create a traveling plane wave in
tively and at a height z = Lz /2 a plane wave can the y direction at all frequencies below the modal
be constructed reducing the amplitude of the room frequency 103 Hz (0 2 1) see Table 6.
modes corresponding to (0 2 0) and (0 0 1) of Table
6 and their combinations. In [9] it has been found that an approximation of the

82
Low Frequency Enhancement System for Rooms Paper D

dB dB dB

80 80 80

70 70 70

60 60 60

50 50 50

1 1 1
2 2 2
3 3 3
4 4 4
Ly 5 m Ly 5 m Ly 5 m
6 3 6 3 6 3
m 2 m 2 m 2
1 Lx 1 Lx 1 Lx

m dB m dB m dB
2.26 2.26 2.26
80 80 80
Height Lz

Height Lz

Height Lz
1.38 70 1.38 70 1.38 70

60 60 60
0.26 0.26 0.26
50 50 50
1 2 3 4 5 6 7 m 1 2 3 4 5 6 7 m 1 2 3 4 5 6 7 m
Length Ly Length Ly Length Ly

Fig. 27. Sound pressure level distribution resulting from the simulation of the Virtual Room, driven frequency 75 Hz. Upper
left, setup .1.0 measured at a height of z = 1.38 m the loudspeaker is located at z = 0.06 m and x = Lx1/2 and y = 0.06 m.
Upper middle, setup .2.0 measured at a height of z = 1.38 m the loudspeakers are on the floor at z = 0.06 m and x = Lx 1/4
and x = Lx 3/4 respectively. Upper right, setup .2.0 measured at a height of z = 1.38 m the loudspeakers are at z = 1.38 m
and x = Lx 1/4 and x = Lx 3/4 respectively. Lower plots are the same setups as upper plots but measured at the vertical plane
x = 2.10 m in the Virtual Room, the driven frequency is the same as upper plots.

maximum frequency where it is possible to create a ers in opposite phase and the appropriate delay ac-
plane wave in a room is given by fmax = c/d − ∆ε cording to the traveling distance from the front wall
where c is the speed of sound and d is the distance in to the back wall.
the x direction between two adjacent loudspeakers,
and ∆ε is a constant that depends on the damping The loudspeakers used in this simulation are the
of the room. same as the example in Fig. 9 and Fig. 10 in Section
2. As it can be observed the front wave travels to-
wards the back wall until it is canceled at the very
end of the room by the back loudspeakers. There-
3.2. Removing the Reflection from the Back Wall fore the reflection of the back wall has been removed
and from the time t = 27 ms to time t = 100 ms the
sound pressure has been reduced significantly.
As room modes or modal resonances are caused by
reflections the obvious way to reducing or removing
these modes is to remove the reflection which has
to be made in the time domain and it will ideally 3.3. Controlled Acoustically Bass System (CABS)
work for all frequencies. In order to create a trav-
eling plane wave in the y direction the reflection of
sound on the back wall has to be minimized. This is In order to canceled out the reflection of the back
achieved by placing a number of extra loudspeakers wall a system called CABS .2.2 (Controlled Acous-
L = N in anti phase with the sound pressure at the tically Bass System) is introduced. This system con-
back wall including a delay according to the travel- sists on the addition of extra loudspeakers to the
ing distance in the y direction. .2.0 setup. This extra loudspeakers are fed with the
same signal as the front loudspeakers N including a
In order to minimize the reflection of sound on the delay ∆t ≈ Ly /c according to the traveling distance
wall at y = Ly , the same number of loudspeakers in the y direction of the plane wave, see Fig. 28. In
should be used in each of the walls. In Fig. 30 an addition to the delay the gain G of the extra loud-
example of the cancellation of the reflection of the speakers has to be adjusted due to the attenuation
back wall is shown using the setup .2.2 in the Virtual of sound by the traveling distance and the damping
Room as shown in Fig. 29 with the extra loudspeak- characteristics of the room.

83
Paper D Low Frequency Enhancement System for Rooms

Fig. 28. Block diagram of equalization system to minimize the reflection of the back wall, G its a factor according to the
damping characteristics of the room and the attenuation of sound by the air.

tion at a height z = Lz/2 is measured using 44 Hz,

55 Hz and 60 Hz as driven frequencies. In Fig. 31
the result of the CABS .2.2 system with the extra
loudspeakers is presented. As it can be observed the
Lz
back wall reflection has been removed and because a
traveling wave in the y direction has been physically
synthesized the sound field is even in almost all of
0 the room and just very close to the loudspeakers a
higher sound pressure level exists.

Ly In order to verify the performance of CABS .2.2 the

Cumulative Spectral Decay (CSD) is computed at
Lx one position and presented in Fig. 33 and the fre-
quency response at the 25 positions in the listen-
Fig. 29. Virtual Room model and loudspeaker setup .2.2 . ing area are presented in Fig. 34. By inspecting the
CSD and the impulse responses after using CABS it
3.4. Simulation of CABS in the Virtual Room is noticeable how the impulse responses have been
shortened and the parameter D has reached 88.7%,
this can be seen in Fig. 32. In addition the MSFD
The CABS .2.2 system (2 front and 2 back sub- is calculated and presented in Fig. 35. As shown a
woofers) has been simulated in the Virtual Room clear improvement has been achieved going from a
to demonstrate the optimal performance on indi- SD of ± 4.9 dB in Fig. 18 to a SD of ± 0.7 dB. Nev-
vidual frequencies. The front loudspeakers .2.B are ertheless the MD still kept high, it improved from ±
placed at x = Lx /4, x = Lx 3/4, y = 0.06 m and at 14 dB to ± 11.8 dB. This can be attributed first to
z = Lz/2 and the back loudspeakers .F.2 are placed the frequency response of the loudspeaker and sec-
at x = Lx /4, x = Lx 3/4 respectively and at y = ondly to the amplification caused by the room it-
7.74 m and z = Lz/2. The sound pressure distribu- self at very low frequencies. As it can be observed in

84
Low Frequency Enhancement System for Rooms Paper D

t = 12 ms t = 16 ms t = 21 ms
x10 −3 x10 −3 x10 −3
Pa Pa Pa
3 3 3
2 2 2
1 1 1
0 0 0
−1 m −1 m −1 m
3 3 3
1 2 2 1 2 2 1 2 2
3 4 5 1 3 4 5 1 3 4 5 1
6 m 6 m 6 m

t = 27 ms t = 35 ms t = 46 ms
x10 −3 x10 −3 x10 −3
Pa Pa Pa
3 3 3
2 2 2
1 1 1
0 0 0
−1 m −1 m −1 m
3 3 3
1 2 2 1 2 2 1 2 2
3 4 5 1 3 4 5 1 3 4 5 1
6 m 6 m 6 m

t = 59 ms t = 77 ms t = 100 ms
x10 −3 x10 −3 x10 −3
Pa Pa Pa
3 3 3
2 2 2
1 1 1
0 0 0
−1 m −1 m −1 m
3 3 3
1 2 2 1 2 2 1 2 2
3 4 5 1 3 4 5 1 3 4 5 1
6 m 6 m 6 m

Fig. 30. Sequence of snap shots in the time domain of the instantaneous sound pressure in the Virtual Room removing the
reflection from the back wall by adding the extra F.2 loudspeakers in anti phase with the sound at the wall and with the
appropriate delay.

dB dB dB

80 80 80

70 70 70

60 60 60

50 50 50

1 1 1
2 2 2
3 3 3
Ly 4 Ly 4 4
5 m 5 m
Ly 5 m
6 3 3 6 3
2 6 2
m m 2 m 1
1 Lx 1 Lx Lx

Fig. 31. Sound pressure level distribution resulting from the simulation of the Virtual Room measured on a plane at a height
of z = 1.38 m after using CABS .2.2. Left produced by 44 Hz (modal frequency). Middle, produced by 55 Hz (anti modal
frequency). Right, produced by 60 Hz.

85
Paper D Low Frequency Enhancement System for Rooms

−3
x 10
1.5
dB

1 D = 88.7%
−20

0.5
−30

−40
−0.5

0
−1
0.5
100 Hz
−1.5 1s
10
0 50 100 150 200 250 300 350 400 450 ms

Fig. 33. Cumulative spectral decay (CSD) at the labeled 17

Fig. 32. The 25 impulse responses of the listening area in the listening position in the Virtual Room after applying CABS
Virtual Room after applying CABS .2.2. and the Definition .2.2 .
D calculated.

Fig. 36 the MSFD has been computed using just the dB

transfer functions from both sound sources to the MSFD = [±0.7 dB, ±11.8 dB]
listening positions by deconvolving the loudspeaker −10
response, then the MD improved by 4.2 dB. A slope
of -20 dB/decade is observed which corresponds to
a 1st order low pass response. This amplification is −20
also shown in Fig. 14 and Fig. 18 with the loud-
speaker setup .2.0 before applying CABS and it can
−30
be explained because of the reflections from floor
and ceiling and the side walls at very low frequen-
cies where the wavelength is much longer than the −40
dimensions. In this situation the differences in phase
are more or less always constructive so they only add
−50
positively to the direct sound. In Fig. 37 the MSFD 10 100 Hz
has been computed after been high pass filtered the
transfer functions by a 1st order high pass filter with Fig. 34. Frequency responses of the 25 virtual microphone
positions in the listening area of the Virtual Room after
cut off frequency at 200 Hz. The MD is then ± 2 dB
applying CABS .2.2 .
and the SD is kept the same in ± 0.7 dB.
In Figs. 40, 41, 38 and 39 results of the simulation
Table 2
of CABS .2.2 now positioned at a height z = 0.66 m Comparison of the results of simulation of CABS .2.2 at the
from the floor are presented. As it can be observed listening area in the Virtual Room from 20 to 100 Hz.
the peaks from 124 Hz to 150 Hz have been attenu- MSFD Definition
ated corresponding to the reflections from the floor
SD (dB) MD (dB) D
and ceiling, room modes (ny = 0 nx = 0 nz =
2), (ny = 0 nx = 2 nz = 2) and the combination corner .1.0 ± 5.8 ± 7.5 68.1 %
with the reflections corresponding to the width of off corner .1.0 ± 5.9 ± 7.2 58.9 %
the room. An improvement is detected also in the
.2.0 ± 4.9 ± 6.6 56.4 %
parameter D been now of 95.9 % and in the MD
moved just to ± 1.9 dB. To summarize the results h = Lz/2 CABS .2.2 ± 0.7 ± 2.0 88.7 %
of the simulated setups compared with CABS .2.2. h ≈ Lz/4 CABS .2.2 ± 0.7 ± 1.9 95.9 %
are shown in Table 2.

86
Low Frequency Enhancement System for Rooms Paper D

dB dB
MSFD = [±0.7 dB, ±11.8 dB] MSFD = [±0.7 dB, ±2 dB]
−10 −10

−20 −20

−30 −30

−40 −40

−50 −50

5 5

10 10
Mic.Pos. Mic.Pos.
15 15

20 100 Hz 20 100 Hz

25 25
20 20

Fig. 35. Mean Sound Field Deviation (MSFD) at the listen- Fig. 37. Mean Sound Field Deviation (MSFD) of high pass
ing area in the Virtual room after applying CABS .2.2 . filtered transfer functions at the listening area in the Virtual
room after applying CABS .2.2. The loudspeaker response
has been deconvolved.
dB
MSFD = [±0.7 dB, ±7.6 dB]
dB
−10
MSFD = [±0.7 dB, ±11.8 dB]
−20
−10
−30

−40
−20

−50

5 −30
10
Mic.Pos.
15
−40
20 100 Hz

25
20
−50
10 100 Hz
Fig. 36. Mean Sound Field Deviation (MSFD) of transfer
functions at the listening area in the Virtual room after Fig. 38. Frequency responses of the 25 virtual microphone
applying CABS .2.2. The loudspeaker has been deconvolved. positions in the listening area of the Virtual Room after
applying CABS .2.2 the loudspeakers are positioned at a
4. IMPLEMENTATION AND height z = 0.66 m.
MEASUREMENT OF CABS IN REAL
ROOMS ied in [21] by Chereck and Langvad. The ceiling is
a false ceiling tilted in the corners and covered with
special plaster panels with three different sections
After simulating the CABS .2.2 system in a Vir- of absorptive materials. The floor is wooden and the
tual Room the system has been implemented in a walls are brick made covered with plaster. The room
PC using a real time signal processing software and has a double metal door in one of the side walls.
an AD/DA multichannel converter. The parameters Secondly the system has been measured in the ITU
of the system were adjusted empirically to achieve Room at Aalborg University which is a multichan-
best performance. First the system has been mea- nel listening room of approx. 172 m3 that conforms
sured in the IEC Room at Aalborg University which to the recommendation ITU-R BS 775-1 for multi-
is a standard listening room of approx. 90 m3 that channel surround setups [22]. The walls of this room
fulfills the IEC 268-13 standard, which describes an are quite damped except the back wall that has large
average living room [20]. This room has been stud- windows that cover most of the wall. The ceiling

87
Paper D Low Frequency Enhancement System for Rooms

dB
MSFD = [±0.7 dB, ±1.9 dB] dB
−10

−20 −20

−30
−30
−40

−50
−40

10 0
Mic.Pos.
15
0.5
20 100 Hz 100 Hz
1s
25
20
10

Fig. 41. Cumulative spectral decay (CSD) at the labeled 17

Fig. 39. Mean Sound Field Deviation (MSFD) of high pass
listening position in the Virtual Room after applying CABS
filtered transfer functions at the listening area in the Vir-
.2.2 the loudspeakers are positioned at a height z = 0.66 m.
tual room after applying CABS .2.2 the loudspeakers are
positioned at a height z = 0.66 m.
box type active loudspeakers with a 8 inch driver
x 10
−3 unit each.
1.5
In the IEC Room the sound field was measured at
1.26 m height with 25 microphone positions equally
1
D = 95.9% spaced by 48 cm within an area of (1.92 × 1.92 m)
centered in the room. The loudspeakers of the setup
0.5
.2.2 were placed at 1.50 m height and 6 cm from
the front and back wall respectively. As illustrated
0 in Section 3.1 the loudspeakers should be placed at
1.38 m height or at 0.69 m from the floor but because
−0.5 of the complexity of the ceiling this height (1.50 m)
was assumed to be a better approximation of Lz /2
−1 since the concrete ceiling is at Lz = 3.10 m in the
room.
−1.5
0 50 100 150 200 250 300 350 400 450 ms In the ITU Room the sound field was measured at
1.20 m height on 25 microphone positions equally
Fig. 40. The 25 impulse responses of the listening area in the spaced by 72 cm within an area of (2.88 × 2.88 m)
Virtual Room after applying CABS .2.2. the loudspeakers centered in the room. The loudspeakers of setup .2.2
are positioned at a height z = 0.66 m and the Definition D
were placed at 1.44 m height and 9 cm from the front
calculated.
and back wall respectively.
is covered with special plaster panels. The floor is
wooden and it has two metal doors placed symmet-
rically on the side walls. In Table 6 the first 25 room 5. RESULTS
modes of both roms are presented and in Table 5 the
room dimensions and some room parameters such
as reverberation time T60 and Schroeder frequency Results of the measurements of CABS .2.2 in the
f g are shown. The impulse responses were acquired two real rooms, the IEC Room and the ITU Room
by measurements using maximum length sequences respectively are presented from Figs. 42 to 47. First
(MLS) [23] of order N = 14 with a sampling fre- in Fig. 42 and 43 the 25 impulse responses aligned
quency f s = 8 kHz and processed by the discrete in time before and after CABS are presented. In the
Fourier transform (DFT) in MATLAB. The loud- IEC Room it is clear how the resonances are min-
speakers employed were four (35 × 29 × 35 cm) close imized by canceling the back reflection quite effec-

88
Low Frequency Enhancement System for Rooms Paper D

0.04 0.04

0.03 0.03
2.0 D = 66.8% 2.0 D = 64.3%
0.02 0.02
2.2 D = 92.4% 2.2 D = 89.4%
0.01 0.01

0 0

−0.01 −0.01

−0.02 −0.02

−0.03 −0.03

−0.04 −0.04
0 50 100 150 200 250 300 350 400 450 ms 0 50 100 150 200 250 ms

Fig. 42. The 25 impulse responses at the listening area re- Fig. 43. The 25 impulse responses at the listening area re-
sulting from measurements in the IEC Room and calcula- sulting from measurements in the ITU Room and calcula-
tion of Definition D. Gray lines setup .2.0. Black lines after tion of Definition D. Gray lines setup .2.0. Black lines after
applying CABS .2.2 . applying CABS .2.2 .

tively. This is expressed on the parameter Definition

D that went from 66.8% to 92.4%. In the ITU Room spectral deviations the MD went from ± 7.7 dB to
is not readily seen as in the IEC Room because the ± 4.1 dB in the IEC Room and from ± 6.5 dB to
reflections are not as strong as in the IEC Room but ± 4.5 dB in the ITU Room from 20 to 100 Hz in
still there is an effective improvement going from a both rooms.
D = 64.3% to a D=89.4%.
As explained in Section 3.4 the MSFD on Figs. 46
By observing Figs. 44 and 45 one can verify that and 47 has been calculated for both rooms by using
the magnitude deviation improved drastically from the high pass filtered transfer functions at the listen-
20 to 100 Hz being differences in magnitude from ing area. This is done to observe just the improve-
± 15 dB to ± 6 dB, these deviations are fixed at ment of the system and not the effect of the boost at
the modal frequencies specially in the IEC Room. very low frequencies neither the effect of the loud-
However the system performed better than the sim- speakers response. The result of this is shown in the
ulations at higher frequencies, one can notice that spectral deviations that went from a MD=± 6.4 dB
in the simulations on Fig. 34 from 100 Hz to 200 Hz to a MD=± 2.1 dB in the MSFD parameter for the
the system did not correct for those peaks but in IEC Room calculated from 20 to 100 Hz. As for the
the real room those peaks were attenuated. This is ITU Room the MD improved from ± 5.3 dB to ±
attributed to wave dispersion errors inherent in the 2.1 dB in the range from 20 Hz to 90 Hz.
simulation method as the frequency increases, these
small errors make the pure delay of the back loud- The results of the MSFD, Definition and the im-
speakers inaccurate for those frequencies. Although provements in dB in rooms IEC and ITU are pre-
in the ITU Room the CABS .2.2 did worse on fre- sented on Tables 3 and 4 respectively. As it is clearly
quencies from 90 Hz to 100 Hz the system performed seen the IEC Room presents more problems than
generally better than in the IEC Room. This is be- the ITU Room since the modal resonances are less
cause the ITU Room is a bigger room and the walls overlapped than in the ITU Room. Nevertheless the
are quite damped already. CABS performed well up to 132 Hz in the IEC Room
and in the ITU room the system performed well up
Concerning the spatial deviations the system im- to 87 Hz. It is remarkable that in both rooms the
proved from having a SD= ± 4.6 dB to a SD= parameter Definition is close to a anechoic response
± 1.6 dB in the IEC Room and from a SD= ± 4.4 dB being 92.4 % in the IEC Room and 89.4 % in the
to SD= ± 2 dB in the ITU Room. Concerning the ITU Room by using CABS.

89
Paper D Low Frequency Enhancement System for Rooms

dB
MSFD = [±4.6 dB, ±7.7 dB] dB

20

10
10

0
0

−10 0

0.5
100 Hz
−20
10 100 Hz 1s
10

dB
MSFD = [±1.6 dB, ±4.1 dB] dB

20

10
10

0
0

−10 0

0.5
100 Hz
−20
10 100 Hz 1s
10

Fig. 44. Measurements in the IEC Room. Left column, frequency responses at the 25 positions (upper) setup .2.0 (lower) with
CABS setup .2.2 . Right column, cumulative spectral decay (CSD) at position 17, (upper) setup .2.0 (lower) the same position
with CABS .2.2 .

Table 3
pending on the size and damping of the room and
Results of measurements and improvement of CABS .2.2 at
the listening area in the IEC Room from 20 to 100 Hz. the loudspeaker placement or listening position.
The deviations in magnitude from one position to
MSFD Definition
another varies at some frequencies from ± 6 dB to
IEC Room SD (dB) MD (dB) D cases where there is almost not sound at all. By first
.2.0 ± 4.6 ± 6.4 66.8 % creating a plane wave in only one direction of the
CABS .2.2 ± 1.6 ± 2.1 92.4 %
room and secondly canceling that plane wave using
loudspeakers delayed at the end wall in opposite
Improvement 6 dB 8.6 dB 25.6 %
phase with the traveling sound, optimal sound pres-
sure level distribution in the room can be obtained.
6. DISCUSSIONS
The CABS is a system that works in the time do-
main and once it is adjusted it works independently
As seen from the analysis in Section 2 when loud- of the program material that is reproduced. If the
speakers are placed in an enclosure a number temperature changes drastically in the room the de-
of problems appear, magnitude deviations from lay must be re-adjusted. As seen in the results CABS
± 10 dB to ± 20 dB occur on the worse cases de- .2.2 worked fine in the IEC Room up to 100 Hz and

90
Low Frequency Enhancement System for Rooms Paper D

dB
dB
MSFD = [±4.4 dB, ±6.5 dB]
10
10

0
0

−10
−10

−20 0

0.5
100 Hz
−30
10 100 Hz 1s
10

dB
dB
MSFD = [±2 dB, ±4.5 dB]
10
10

0
0

−10
−10

−20 0

0.5
100 Hz
−30
10 100 Hz 1s
10

Fig. 45. Measurements in the ITU Room. Left column, frequency responses at the 25 positions (upper) setup .2.0 (lower) with
CABS setup .2.2 . Right column, cumulative spectral decay (CSD) at position 17, (upper) setup .2.0 (lower) the same position
with CABS .2.2 .

Table 4
able. The working range of the subwoofer used was
Results of measurements and improvement of CABS .2.2 at
the listening area in the ITU Room from 20 to 90 Hz. from 30 Hz to 150 Hz.
MSFD Definition After having been simulated and measured the
ITU Room SD (dB) MD (dB) D CABS .2.2 system for low frequencies can be said to
perform well in both rooms improving both spec-
.2.0 ± 4.2 ± 5.3 64.3 %
tral deviations and spatial deviations. Generally it
CABS .2.2 ± 1.3 ± 2.1 89.4 % worked not only in the listening area but also in the
Improvement 5.8 dB 6.4 dB 25.1 % whole room. The system presents some variations
at the modal frequencies, this variations are due
to asymmetries in the room and the complexity of
in the ITU Room up to 90 Hz. Indeed if the system
the ceiling in the IEC Room for example, and in
is integrated to a full range reproduction system it
the ITU Room because of the different impedance
must include a low pass filter to attenuate frequen-
of the front wall and back wall. Interestingly seen
cies above these limits. Any how most subwoofers
from the right waterfall plots in Figs. 44 and 45 the
work within this range. In this paper the low pass
modal frequencies are much more noticeable in the
filter was not included in the setup in order to know
IEC Room than in the ITU Room so the improve-
up to which frequency the system can work accept-

91
Paper D Low Frequency Enhancement System for Rooms

dB dB
MSFD = [±4.6 dB, ±6.4 dB] MSFD = [±4.2 dB, ±5.3 dB]
20 20

10 10

0 0

−10 −10

−20 −20

5 5

10 10
Mic.Pos. Mic.Pos.
15 15

20 100 Hz 20 90 Hz
25 25
20 20

dB dB
MSFD = [±1.6 dB, ±2.1 dB] MSFD = [±1.3 dB, ±2.1 dB]
20 20

10 10

0 0

−10 −10

−20 −20

5 5

10 10
Mic.Pos. Mic.Pos.
15 15

20 100 Hz 20 90 Hz
25 25
20 20

Fig. 46. Mean Sound Field Deviation (MSFD) of high pass Fig. 47. Mean Sound Field Deviation (MSFD) of high pass
filtered transfer functions at the listening area measured in filtered transfer functions at the listening area measured in
the IEC Room. Upper setup .2.0. Lower after applying CABS the ITU Room. Upper setup .2.0. Lower after applying CABS
.2.2. The loudspeaker response has been deconvolved. .2.2. The loudspeaker response has been deconvolved.

ment is best in the first room. In the ITU room more overlapped and the front waves are less plane.
instead the modal frequencies do not keep going Moreover in bigger rooms this limit (ny = 0 nx = 2
in time as much as in the first room therefore the nz = 1) can be already over the Schroeder frequency.
room modes in the ITU Room decay faster than in This frequency limit could be used as the crossover
the IEC room in this case the improvement is not frequency if one wants to integrate the system with
as obvious as in the IEC Room. the full range loudspeakers. Although CABS .2.2
works fine with four loudspeakers within the typi-
To summarize the system works depending on the cal subwoofer frequency range in small and middle
size of the room, the smaller the room the more con- size rooms. For wider rooms there may be needed
trollable the system will be. It can be said that if more loudspeakers in the front wall and back wall
CABS .2.2 is well implemented it should work up to for example on positions corresponding to the nodes
frequencies corresponding to the room modes (ny = of room mode (ny = 0 nx = 4 nz = 1). To extend
0 nx = 2 nz = 1) or (ny = 0 nx = 2 nz = 2) the frequency range in the IEC Room there may be
depending of the height where the loudspeakers are needed more loudspeakers on positions correspond-
positioned and when this room modes are below the ing to the nodes of mode (ny = 0 nx = 2 nz = 2)
Schroeder frequency. Above this frequency as seen being two loudspeakers at a height Lz = Lz(1/4)
in Section 2 in small rooms the room modes become and two at Lz = Lz(3/4) or to the mode (ny = 0

92
Low Frequency Enhancement System for Rooms Paper D

nx = 3 nz = 2) being three loudspeakers equally of the room to create a traveling plane wave and
spaced along the width Lx at a height Lz = Lz(1/4) two extra low frequency loudspeakers in the back
and three more at a height Lz = Lz(3/4). wall delayed and in opposite phase to remove the
reflection of that wall giving a uniform sound field.
A subject to discussion is if a complete flat response After measurements of the implemented system in
is wanted where the room is completely removed. As the two rooms one can conclude that the system
it was observed the fact that the front loudspeak- works effectively in small and middle size rectangu-
ers are at the very wall and the side reflections from lar rooms. The system can achieve good responses
floor and ceiling are used to built a plane wave by not only in a single listening position but also in
utilizing a restricted number of loudspeakers it im- the whole room from 20 Hz to 100 Hz having spa-
plies that there is an amplification at very low fre- tial deviations in a large listening area of only ±
quencies that falls as the frequency increases. This 1.3 dB in the ITU Room and ± 1.6 dB in the IEC
slope was observed to be approx. 20 dB/decade in Room, contrary to the advanced room correction
the case of the two rooms examined. The correction systems that typically optimize to a single listening
of this boost may depend on personal preference if position. Informal listening with music signals inte-
it is necessary to correct for this amplification a 1st grating CABS with full range stereo loudspeakers
order high pass filter can be connected before CABS has shown evident improvement by removing the
to compensate that boost. On the other hand this booming sound which is always present in small or
amplification may be an advantage for loudspeakers middle size rooms. It presents a clear front sound
with poor power output at the low end frequency image to the listeners and the back loudspeakers
limit and also because we as humans are less sensi- were not heard at all. Since the system works in
tive at low frequencies. the time domain it works effectively with transient
The advantages of this system is that it works in sounds as well as with long durations tones.
the time domain and it could be adjusted paramet-
rically to certain degree of enhancement depending
on personal taste. One of the drawbacks of this ap-
References
proach is that extra loudspeakers and power ampli-
fiers are needed although simple signal processing
equipment has to be added in order to cancel the [1] A. R. Groh, “High-Fidelity Sound System Equalization
sound at the back wall. A further research can be ad- by Analysis of Standing Waves,” J. Audio Eng. Soc.,
dressed to investigate the amount of spectral devia- vol. 22, pp. 795–799 (December 1974).
tions at low frequencies that are tolerable in terms [2] R. F. Allison, “The Sound Field in Home Listening
of human preferences. Rooms II,” J. Audio Eng. Soc., vol. 24, pp. 14-19
(January/February 1976).
[3] K. O. Ballagh, “Optimum Loudspeaker Placement Near
Reflecting Planes,” J. Audio Eng. Soc., vol. 31, pp. 931–
7. CONCLUSIONS 935 (1983 December).
[4] T. Welti, “How Many Subwoofers are Enough,”
presented at the 112th Convention of the Audio
The analysis in time and frequency domains of
Engineering Society, J. Audio Eng. Soc. (Abstracts), vol.
sound fields at low frequencies produced by typical 50 p. 523 (June 2002), convention preprint 5602.
sound reproduction systems placed in rooms was
[5] J. A. Pedersen, “Adjusting a loudspeaker to its acoustic
presented. A simulation program based in FDTD environment,” presented at the 115th Convention of
has been utilized to render the sound field produced the Audio Engineering Society, J. Audio Eng. Soc.
by low frequency loudspeakers in rectangular rooms. (Abstracts), vol. 51 p. 1223 (December 2003), convention
preprint 5880.
A novel and effective method named Controlled
[6] A. Mäkivirta and P. Antsalo, “Modal Equalization of
Acoustically Bass System (CABS) to achieve op- LoudspeakerRoom Responses at Low Frequencies,” J.
timum sound pressure level distribution inside a Audio Eng. Soc., vol. 51, pp. 324–353. (May 2003)
rectangular room at low frequencies has been in-
[7] S. J. Elliott and P. A. Nelson, “Multiple-Point
troduced. The CABS .2.2 has been simulated and Equalization in a Room Using Adaptive Digital Filters,”
implemented in two standard listening rooms. The J. Audio Eng. Soc., vol. 37, pp. 899–907. (November
system utilizes two loudspeakers in the front wall 1989)

93
Paper D Low Frequency Enhancement System for Rooms

[8] A. O. Santillán, C. S. Pedersen and M. Lydolf,

“Experimental implementation of a low-frequency
global sound equalization method based on free field Table 5
propagation,” Applied Acoustics, in Press, accepted in Room dimensions, T60, Schroeder frequency (f g) and the
(May 2006). room mode density below f g and below 100Hz.
[9] A. O. Santillán, “Spatially extended sound equalization Room LxWxH (m) V (m3 ) T60 (s) f g (Hz) N ∆Nf (Hz)
in rectangular rooms,” J. Acoust. Soc. Am., vol. 110, −
→ −→ −
→ −→
pp. 1989–1997. (October 2001) f g 100Hz f g 100Hz

[10] A. Celestinos and S. B. Nielsen, “Optimizing placement IEC 7.80x4.12x2.78 89.34 0.47 145 49.72 20.13 0.81 0.45
and equalization of multiple low frequency loudspeakers ITU 8.12x7.39x2.88 172.82 0.31 85 23.20 34.60 0.65 0.84
in rooms,” presented at the 119th Convention of
the Audio Engineering Society, J. Audio Eng. Soc.
(Abstracts), vol. 53 p. 1206 (December 2005), convention
preprint 6545.
[11] A. Celestinos and S. B. Nielsen, “Low frequency
sound field enhancement system for rectangular rooms
using multiple low frequency loudspeakers,” presented Table 6
at the 120th Convention of the Audio Engineering The first 25 room modes of IEC and ITU Rooms.
Society, J. Audio Eng. Soc. (Abstracts), vol. 54 p. 1206 IEC Room ITU Room
(July/August 2006), convention preprint 6688.
ny nx nz f n (Hz) ny nx nz f n (Hz)
[12] A. Celestinos and S. B. Nielsen, “Multi-source low
frequency room simulation using finite difference 100 22 100 21
time domain approximations,” presented at the 117th 010 41 010 23
Convention of the Audio Engineering Society, J.
Audio Eng. Soc. (Abstracts), vol. 53 pp. 105–106 200 44 110 31
(January/February 2005), convention preprint 6264. 110 47 200 42
[13] L. E. Kinsler, Fundamentals of Acoustics, 4th. Ed. John
210 61 020 46
Wiley & Sons, Inc. (2000)
[14] H. Kuttruff, Room Acoustics, 3th. Ed. E & FN Spon. 001 63 210 48
(1999) 300 66 120 51
[15] T. Salava, “Acoustic Load and Transfer Functions in 101 66 001 60
Rooms at Low Frequencies,” J. Audio Eng. Soc., vol.
36, pp. 763–775 (October 1988) 011 75 220 63
[16] J. D. Bunton and R. H Small, “Cumulative Spectra, 201 77 101 63
Tone Bursts and Applications,” J. Audio Eng. Soc., vol.
310 78 300 64
30, pp. 386–395. (June 1982)
[17] R. Thiele, “Richtungsverteilung und Zeitfolge der 111 78 011 64
Schallrückwurfe in Räumen,” Acustica, vol. 3, pp. 291- 020 82 111 67
302 (1953).
120 85 310 68
[18] S.K. Mitra, Digital Signal Processing A Computer Based
Approach, 1st Ed. McGraw-Hill, pp. 462–468 (1998). 211 87 030 70
[19] L.B. Jackson, Digital Filters and Signal Processing, 3rd 400 88 130 73
Ed. Kluwer Academic Publishers, pp. 301–307 (1996).
301 91 201 73
[20] IEC 60268-13 BS.6840 13, “Sound System Equipment–
220 93 021 76
Part13:Listening Tests on Loudspeakers,” International
Electrotechnical Commission, Geneva, Switzerland, 410 97 211 77
(1988).
311 100 121 78
[21] B. Chereck and B. Langvad, “Low Frequency Simulation
of a Listening Room,” Proc. of Nordic Acoustical 021 103 320 79
Meeting, pp. 265–270. (1990) 320 105 230 82
[22] ITU-R BS.775-1, “Multichannel stereophonic sound 121 105 400 85
system with and without accompanying picture,”
International Telecommunications Union, Geneva, 401 108 301 87
Switzerland (1992–1994). 500 110 221 87
[23] J. Vanderkooy, “Aspects of MLS Measuring Systems,”
J. Audio Eng. Soc., vol. 42, pp. 219–231. (April 1994)

94
Chapter E
Chapter E Low Frequency Enhancement System for Rooms

96
CABS .2.2 In An Irregular Room

E.1 Low Frequency Sound Fields in An Irregular Room

Rooms with irregular shapes are more close to the typical listening environments where
sound reproduction systems are utilized therefore it is of interest to simulate them. By
using the Finite Differences in the Time Domain (FDTD) method basic room shapes can
be simulated. The boundaries of the enclosure can be defined by the normal component
of the particle velocity at the wall (see Appendix I). This is done by the calculation of
the impedance at the boundary either using an estimate of the absorption coefficient or
by the characteristic impedance of the wall. By using a staggered grid the room can be
divided in two sections by setting the components of the particle velocity at the desired
walls of the room.

E.1.1 Partitioning of the room

For example in the small enclosure seen from above in Figure E.4. The original room is
partitioned by the walls A and B forming an “L” shape room. The wall A corresponds
to the particle velocity points ux[0.6,0.7:1.1,z] represented by small squares. The wall B
corresponds to the particle velocity points uy[0.7:0.9,0,z] . Care should be taken in order to
completely close the room hence the smaller section is isolated. All boundaries are treated
the same manner as the regular rooms, first the original walls C, D, E and F are defined,
and finally the boundaries A an B that close the room. The boundary equation for wall
C is:

ρ0 h
k −Z k 2
ux[Lx,y,z] (t + )= k
ρ0 h
ux[Lx,y,z] (t − ) + ρ0 h
p[Lx− h ,y,z] (t), (E.1)
2 +Z 2 +Z 2
k k

for wall D:

97
Chapter E Low Frequency Enhancement System for Rooms

Figure E.4: Example of a calculation grid in a 1 m x 1.20 m irregular enclosure seen from
above. The circles are the pressure p points, stars are particle velocity points uy in the y
direction and the squares are the particle velocity points ux in the x direction.

ρ0 h
k −Z k 2
ux[x,0,z] (t + ) = k
ρ0 h
ux[x,0,z] (t − ) + ρ0 h
p[x,0+ h ,z] (t), (E.2)
2 +Z 2 +Z 2
k k

for wall E:

ρ0 h
k −Z k 2
ux[0,y,z] (t + )= k
ρ0 h
ux[0,y,z] (t − ) − ρ0 h
p[0+ h ,y,z] (t), (E.3)
2 +Z 2 +Z 2
k k

for wall F:

ρ0 h
k −Z k 2
ux[x,Ly,z] (t + )= k
ρ0 h
ux[x,Ly,z] (t − ) + ρ0 h
p[x,Ly− h ,z] (t). (E.4)
2 +Z 2 +Z 2
k k

The boundary equation for wall A is as follows:

98
Low Frequency Enhancement System for Rooms Chapter E

ρ0 h
−Z
ux[Lxc ,Lyc :Ly,z] (t + k2 ) = k
ρ0 h ux[Lxc ,Lyc :Ly,z] (t − k2 )
k
+Z (E.5)
+ ρ0 h2 p[Lxc − h ,Lyc :Ly,z] (t),
k
+Z 2

and for wall B is written as

ρ0 h
−Z
ux[Lxc :Lx,Lyc ,z] (t + k2 ) = k
ρ0 h ux[Lxc :Lx,Lyc ,z] (t − k2 )
k
+Z (E.6)
+ ρ0 h2 p[Lxc :Lx,Lyc − h ,z] (t).
k
+Z 2

E.1.2 Simulation of loudspeakers in an irregular room

The irregular room shown in Figure E.5 of dimensions Lx=7.08m,Ly=7.8m,

Lz=2.76m and Lxc =2.88m,Lyc =4.5m has been simulated. The estimated absorption co-
efficient used for the walls an floor was 0.12 and 0.13 and 0.15 for the ceiling. The
loudspeakers are modeled as (12×12×12) cm cubic sound sources.

Figure E.5: Irregular room model

On this room shape one can expect that the predominant resonance frequencies would be
the ones related to the length Ly, width Lx − Lxc and Lx but also the frequency which
wavelength relates to the pad Ly + Lx. This would depend on where the loudspeakers are
placed in the room. By having the loudspeakers equidistantly spaced at the front wall F
they will construct a plane wave traveling along the room towards wall D but at the abrupt

99
Chapter E Low Frequency Enhancement System for Rooms

90 90

80 80
(dB)

(dB)
70 70

60 60

50 50
7 7
6 6
5 5
4 1 4 1
3 2 2
3 3 3
2 4 2 4
1 5 1 5
6 6
7 Width (m) 7 Width (m)
Length (m) Length (m)

90 90

80 80
(dB)

(dB)
70 70

60 60

50 50

7 7
6 6
5 5
4 1 4 1
3 2 3 2
3 3
2 4 2 4
1 5 1 5
6 6
7 Width (m) 7 Width (m)
Length (m) Length (m)

Figure E.6: Sound pressure level distribution calculated at a height of 1.38 m in the
irregular room. Upper left driven frequency 44. Upper right 55 Hz. Lower left 60 Hz.
Lower right 80Hz.

corner formed by wall A and B the front wave would diffract itself forming a curved edge
towards the wall C. The main front wave will reflect to wall D with less amplitude on one
side due to the diffraction caused by the corner and then it will come back towards wall F.
The diffracted edge of the front wave will hit wall C and reflect towards wall E and so on.
These will be the main pads that will construct the patterns of the sound pressure level
distribution along the room. On Figure E.6 the results of the calculated sound pressure
level distribution in the irregular room are presented with 44Hz, 55Hz, 60Hz and 80Hz as
driven frequencies to both loudspeakers. As it can be observed the structures of the sound
level distribution are some how bended and not so regular along the width of the room
due to the diffracted wave reflected to wall C.

Three setups (A,B and C) are simulated in the irregular room with two common scenarios
each. First a typical subwoofer (.1.0) on the floor near one corner and secondly two
loudspeakers (2.0.0 or .2.0) as in a stereo setup both producing the same signal. In setups
A and B these loudspeakers are separated from the wall about 1.4 m and 1 m respectively,
and in the setup C the loudspeakers are placed at the wall. In all cases the sound field
has been sampled on a listening area of 1.92 m × 1.92 m situated at a listening height of
z = 1.38m, delimited by 25 virtual microphones equally spaced by 48 cm. The results of

100
Low Frequency Enhancement System for Rooms Chapter E

the simulations are presented in the following sections. The MSFD has been calculated
as explained in Paper D Section 2.4.

Setup A

Lz Lz
Ly Ly

Lx Lx

dB dB

−20 −10

−30 −20

−40 −30

−50 −40

−60 −50
10 100 Hz 10 100 Hz

dB
MSFD = [±4.4 dB, ±8.9 dB]
dB
MSFD = [±5.6 dB, ±7.2 dB]
−10

−20

−20

−30
−30

−40
−40

−50
−50

−60
5

5
10
10
15
15
100 Hz
20
20 100 Hz

25 25
20 20

Figure E.7: Left column, Setup A .1.0 . Right column, Setup A 2.0.0 . Upper row, room
model and loudspeakers. Middle row, 25 frequency responses at the listening area. Lower
row, Mean Sound Field Deviation.

101
Chapter E Low Frequency Enhancement System for Rooms

Setup B

Lz Lz
Ly Ly

Lx Lx

dB dB

−10 −10

−20 −20

−30 −30

−40 −40

−50 −50
10 100 Hz 10 100 Hz

dB dB
MSFD = [±4.3 dB, ±6.1 dB] MSFD = [±4.3 dB, ±8.7 dB]

−10 −10

−20 −20

−30 −30

−40 −40

−50 −50

5 5

10 10

15 15

20 20
100 Hz 100 Hz

25 25
20 20

Figure E.8: Left column, Setup B .1.0 . Right column, Setup B 2.0.0 . Upper row, room
model and loudspeakers. Middle row, 25 frequency responses at the listening area. Lower
row, Mean Sound Field Deviation.

102
Low Frequency Enhancement System for Rooms Chapter E

Setup C

Lz Lz
Ly Ly

Lx Lx

dB dB

−10 −10

−20 −20

−30 −30

−40 −40

−50 −50
10 100 Hz 10 100 Hz

dB dB
MSFD = [±4.8 dB, ±6.8 dB] MSFD = [±5.8 dB, ±7.2 dB]

−10 −10

−20 −20

−30 −30

−40 −40

−50 −50

5 5

10 10

15 15

20 20
100 Hz 100 Hz

25 25
20 20

Figure E.9: Left column, Setup C .1.0 . Right column, Setup C .2.0 . Upper row, room
model and loudspeakers. Middle row, 25 frequency responses at the listening area. Lower
row, Mean Sound Field Deviation.

E.2 Simulation of CABS .2.2 in the Irregular Room

As mention early in this work the optimal placement of the loudspeakers in the room is of
great importance for the performance of CABS. This to suppress as much as possible the
room modes caused by the side walls, floor and ceiling and with the action of CABS have
only propagating plane waves traveling in one direction. Now it is interesting to know
how CABS would perform in an irregular room where the conditions are not as optimal
as in a perfect rectangular room. In Figure E.10 the result of simulations of CABS .2.2 is
presented using the .2.0 loudspeakers at wall F and the rear loudspeakers .0.2 at wall D. As
it can be observed the reflection from the rear wall has not been completely removed. At
44Hz the reflection of the diffracted wave from walls C and part of wall D had disturbed

103
Chapter E Low Frequency Enhancement System for Rooms

90 90

80 80
(dB)

(dB)
70 70

60 60

50 50

7 7
6 6
5 5
4 1 4 1
3 2 3 2
3 3
2 4 2 4
1 5 1 5
6 6
7 Width (m) 7 Width (m)
Length (m) Length (m)

90 90

80 80
(dB)

(dB)
70 70

60 60

50 50

7 7
6 6
5 5
4 1 4 1
3 2 3 2
3 3
2 4 2 4
1 5 1 5
6 6
7 Width (m) 7 Width (m)
Length (m) Length (m)

Figure E.10: Sound pressure level distribution calculated at a height of 1.38 m using CABS
.2.2 in the irregular room. Upper left driven frequency 44Hz. Upper right 55 Hz. Lower
left 60 Hz. Lower right 80 Hz.

the pressure distribution at the listening area. Nevertheless the sound pressure level
distribution is more even than by using just the 2.0 loudspeakers. One would suggest that
the position of the rear loudspeakers might be optimized as well as the individual gain
and delay of the rear loudspeakers to cancel as much as possible the reflection of the back
wall.

To know how the performance of CABS .2.2 differs from the optimal condition the system
has been simulated as it was originally implemented with the same gain and delay in both
rear loudspeakers on the three scenarios A B C presented in Section E.1.2. Results of the
simulations are shown in Figures E.11 and E.12.

104
Low Frequency Enhancement System for Rooms Chapter E

Lz Lz Lz
Ly Ly Ly

Lx Lx Lx

dB dB dB

−10 −10 −10

−20 −20 −20

−30 −30 −30

−40 −40 −40

−50 −50 −50

10 100 Hz 10 100 Hz 10 100 Hz

dB dB dB
MSFD = [±2 dB, ±2.6 dB] MSFD = [±1.9 dB, ±3.5 dB] MSFD = [±3.2 dB, ±5.3 dB]

−10 −10 −10

−20 −20 −20

−30 −30 −30

−40 −40 −40

−50 −50 −50

5 5 5

10 10 10

15 15 15

20 20 20
100 Hz 100 Hz 100 Hz

25 25 25
20 20 20

Figure E.11: Left column, CABS .2.2 in Setup A. Middle column, CABS .2.2 in Setup B.
Right column, CABS .2.2 in Setup B. Upper row, room model and loudspeakers. Middle
row, 25 frequency responses at the listening area. Lower row, Mean Sound Field Deviation.

E.3 Summary and Conclusions

In this chapter the simulation model using the FDTD method of an irregular room has
been presented. Simulations of different setups of low frequency loudspeakers in the virtual
irregular room have been computed. The setup that had the worse spatial deviation was
Setup C .2.0 having a SD = ± 5.8 dB and the setup with the worse magnitude deviation
was Setup A 2.0.0 with M D = ± 8.4 dB. The setup that had slightly better performance
was Setup B .1.0 with one subwoofer having spatial deviations of SD = ± 4.3 dB and
magnitude deviations of M D = ± 6.1 dB.

The performance of CABS .2.2 was affected by the irregular shape of the room mainly
because of the abrupt corner that breaks the front wave diffracting it towards not only
the back wall but also to other walls of the room. Nevertheless CABS .2.2 improved the
spatial and magnitude deviations having less magnitude deviations in Setup A and less
spatial deviations in Setup B. In general it was observed that the symmetrical place of the
rear loudspeakers was not necessary the best placement. The deviations at frequencies in

105
Chapter E Low Frequency Enhancement System for Rooms

Figure E.12: Comparison between configurations 0.1.0 , 2.0.0 , 0.2.0 (Blue curves) and
CABS .2.2 (Red curves). The figure shows the impulse responses curves at the listening
area and the computed “Deutlichkeit” number. Left column Setup A. Middle column
Setup B. Right column Setup C.

the range of 20 to 50 Hz suggests that the amplitude and the delay to each of the rear
loudspeakers need to be slightly different. In setups A and B it seems that CABS .2.2 was
able to suppress the back wall reflection but not the diffracted waves by the corner and the
reflection of the side wall C. In Setup C the reflection of the rear wall is suppressed but the
curved edge of the front wave produced by the front loudspeakers would travel towards
wall D and C and come back to wall F forming a standing wave around 44 Hz. The precise
adjustment of CABS .2.2 (delay and gain) and optimal placement of the loudspeakers on
this kind of rooms may be a subject for future investigations. It would be also interesting
to test CABS .2.2 in real irregular rooms including furniture since only simulations have
been carried out.

106
Bibliography

[1] AESTD1001.0.01-05. Multichannel surround sound systems and operations. Audio

Engineering Society, Technical Council, New York, NY, 2001.

[2] R. F. Allison. The sound field in home listening rooms ii. J. Audio Eng. Soc., 24:
14–19, 1976.

[3] K. O. Ballagh. Optimum loudspeaker placement near reflecting planes. J. Audio Eng.
Soc., 31:931–935, 1983.

[4] J. Blauert. Spatial Hearing: The Psychophysics of Human Sound Localization. MIT
Press, 1997.

[5] R. H. Bolt. Normal modes of vibration in room acoustics: Angular distribution theory.
J. Acoust. Soc. Am., 11:74–79, 1939.

[6] M. M. Boone, E. Verheijen, and P. V. Tol. Spatial sound-field reproduction by wave-

field synthesis. J. Audio Eng. Soc., 43:1003–1012, 1995.

[7] D. Botteldooren. Finite-difference time-domain simulation of low-frequency room

acoustic problems. J. Acoust. Soc. Am., 98(6):3302–3308, 1995.

[8] IEC 60268–13 BS.6840 13. Sound System Equipment–Part13:Listening Tests on Loud-
speakers. International Electrotechnical Commission, Geneva, Switzerland, 1988.

[9] ITU–R BS.775–1. Multichannel stereophonic sound system with and without accompa-
nying picture. Rec., International Telecommunications Union, Geneva, Switzerland,
1994.

[10] S. J. Elliott and D. Wilson. Practical implementation of low–frequency equalization

using adaptive digital filters. J. Audio Eng. Soc., 42:988–998, 1994.

[11] A. R. Groh. High-fidelity sound system equalization by analysis of standing waves.

J. Audio Eng. Soc., 22:795–799, 1974.

[12] L. Kinsler, A. Frey, A. Coppens, and J. Sanders. Fundamentals of Acoustics. John

Wiley & Sons, Inc., NY, 2000.

107
Chapter E Low Frequency Enhancement System for Rooms

[13] K. S. Kunz and R. J. Luebbers. Finite Difference Time Domain Method for Electro-
magnetics. CRC Press LLC, N W, FL, 1993.

[14] H. Kuttruff. Room Acoustics. 3th. Ed. E & FN Spon., London, NY, 1999.

[15] A. Mäkivirta, P. Antsalo, M. Karjalainen, and V. Välimäki. Modal equalization of

loudspeakerroom responses at low frequencies. J. Audio Eng. Soc., 51:324–353, 2003.

[16] H. Møller. Reproduction of artificial-head recordings through loudspeakers. J. Audio

Eng. Soc., 43:30–33, 1989.

[17] P. Morse and K. Ingard. Theoretical Acoustics. McGraw–Hill, NY, 1968.

[18] S. K. Olesen. Low frequency room simulation using finite difference equations. In
102nd Audio Eng. Soc. Conv., 1997. preprint 4422.

[19] J. A. Pedersen. Adjusting a loudspeaker to its acoustic environment. In 115th Audio

Eng. Soc. Conv., 2003. preprint 5880.

[20] T. Salava. Acoustic load and transfer functions in rooms at low frequencies. J. Audio
Eng. Soc., 36:763–775, 1988.

[21] A. O. Santillán, C. Pedersen, and M. Lydolf. Experimental implementation of a low-

frequency global sound equalization method based on free field propagation. Applied
Acoustics, 2006. to be published.

[22] L. Savioja and V. Valimaki. Reducing the dispersion error in the digital waveguide
mesh using interpolation and frequency-warping techniques. IEEE Trans. Speech
Audio Processing, 8:184–194, 2000.

[23] L. Savioja, J. Backman, A. Jarvinen, and T. Takala. Waveguide mesh method for
low-frequency simulation of room acoustics, 1995.

[24] U. P. Svensson and U. R. Kristiansen. Computational modelling and simulation of

acoustic spaces. In Proc. of AES 22nd International Conference on Virtual, Synthetic
and Entertainment Audio, 2002. paper 266.

[25] R. Thiele. Richtungsverteilung und zeitfolge der schallrückwurfe in räumen. Acustica,

3:291–302, 1953.

[26] T. Welti and A. Devantier. Low-frequency optimization using multiple subwoofers.

J. Audio Eng. Soc., 54:347–364, 2006.

[27] F. L. Wightman and D. J. Kistler. The dominant role of low–frequency interaural

time differences in sound localization. J. Acoust. Soc. Am., 91:1649–1661, 1992.

108
I
Appendix I Low Frequency Enhancement System for Rooms

110
Appendix I

I.1 Sound Field Room Simulator

In this Appendix a detailed description of the room simulation program is presented. First
the analytical description of the finite-difference time-domain (FDTD) method is derived,
second the technical implementation is outlined and finally the description of the graphical
user interface (GUI) is presented.

I.1.1 Discretization of the wave equation

Typically the FDTD method utilizes two coupled first order differential equations, in
acoustics this equations are the simple force linear Euler’s equation where the pressure p
and the particle velocity u are related as

∂~u
∇p = −ρ0 (I.7)
∂t

where ρ0 is the density of the transmission media in kg/m3 . The second equation is the
linear continuity equation

1 ∂p
∇ · ~u = − (I.8)
c2 ρ0 ∂t

where c is the wave propagation speed in the media 12 . Since the acoustic pressure ∇p and
the particle velocity ∇ · ~u can be expressed as

∂p ∂p ∂p
∇p = x̂ + ŷ + ẑ (I.9)
∂x ∂y ∂z

111
Appendix I Low Frequency Enhancement System for Rooms

and

∂ux ∂uy ∂uz

∇ · ~u = + + (I.10)
∂x ∂y ∂z

the equations (I.7) and (I.8) can be rewritten as

∂p ∂p ∂p ∂~u
x̂ + ŷ + ẑ = −ρ0 (I.11)
∂x ∂y ∂z ∂t

∂ux ∂uy ∂uz 1 ∂p

+ + =− 2 (I.12)
∂x ∂y ∂z c ρ0 ∂t

then from (5) Eq. (I.7) yields to

" #
∂ux ∂uy ∂uz 1 ∂p ∂p ∂p
+ + =− x̂ + ŷ + ẑ (I.13)
∂t ∂t ∂t ρ0 ∂x ∂y ∂z

and from (6) Eq. (I.8) yields to

" #
∂p 2 ∂ux ∂uy ∂uz
= −c ρ0 + + . (I.14)
∂t ∂x ∂y ∂z

After the derivation and linearization in the time domain both equations are sampled in
time and space using the sampling rates k1 Hz and h1 m−1 . From Eq.(7) the resulting set
of equations for the components of the particle velocity are written as

h i
uxx+ h ,y,z (t + k2 ) = uxx+ h ,y,z (t − k2 ) − k
hρ0 × px+h,y,z (t) − px,y,z (t) ,
2 2
h i
uyx,y+ h ,z (t + k
2) = uyx,y+ h ,z (t − k2 ) − k
hρ0 × px,y+h,z (t) − px,y,z (t) , (I.15)
2 2
h i
uzx,y,z+ h (t + k2 ) = uzx,y,z+ h (t − k2 ) − k
hρ0 × px,y,z+h (t) − px,y,z (t) ,
2 2

112
Low Frequency Enhancement System for Rooms Appendix I

and from Eq. (8) the acoustical pressure is derived with

h i
c 2 ρ0 k
px,y,z (t + k) = px,y,z (t) − h uxx+ h ,y,z (t + k2 ) − uxx− h ,y,z (t + k2 )
2 2
h i
c 2 ρ0 k y y
− h ux,y+ h ,z (t + k2 ) − ux,y− h ,z (t + k2 ) (I.16)
2 2
h i
c 2 ρ0 k
− h ux,y,z+ h (t + 2 ) − ux,y,z− h (t + k2 )
z k z
2 2

where the acoustical pressure is determined at the grid points (xδx, yδy, zδz) at time t = δt
and δx = δy = δz = h that is the spatial discretization step and δt = k that is the time
step.

In Eq. (I.15) The three components of the particle velocity are determined at times t =
(t + 12 )k and positions

ux(x± h ,yh,zh) , uy(xh,y± h ,zh) , uz(xh,yh,z± h ) . (I.17)

2 2 2

113
Appendix I Low Frequency Enhancement System for Rooms

I.1.2 Boundary conditions

Assuming the following enclosure of volume V = Lx Ly Lz

Lz

Lx x

Ly
y

the boundaries of the enclosure are defined by the components of the particle velocity at
positions

ux[0,y,z] ux[Lx,y,z]
uy[x,0,z] uy[x,Ly,z] (I.18)
uz[x,y,0] uz[x,y,Lz]

and from the set of equations (I.15) in the x direction the equation

k k k h i
uxx+ h ,y,z (t + ) = uxx+ h ,y,z (t − ) − × px+h,y,z (t) − px,y,z (t) , (I.19)
2 2 2 2 hρ0

114
Low Frequency Enhancement System for Rooms Appendix I

for the wall at [Lx , y, z] can be rewritten as

k k k h i
ux[Lx ,y,z] (t + ) = ux[Lx ,y,z] (t − ) − × p[Lx + h ,y,z] (t) − p[Lx − h ,y,z] (t) , (I.20)
2 2 hρ0 2 2

but since the term p[Lx + h ,y,z] (t) is unknown the asymmetric finite-difference approximation
2
for the space derivative can be introduced 7

h i
p[Lx + h ,y,z] (t) − p[Lx − h ,y,z] (t) ≈ 2 p[Lx ,y,z] (t) − p[Lx − h ,y,z] (t) (I.21)
2 2 2

and substituted in (I.20) yields to

k k k h i
ux[Lx ,y,z] (t + ) = ux[Lx ,y,z] (t − ) − 2 × p[Lx ,y,z] (t) − p[Lx − h ,y,z] (t) . (I.22)
2 2 hρ0 2

To simplify the simulation model the complex part of the impedance of the wall has been
neglected. The acoustic pressure at the wall can be expressed by the product of the
component of the particle velocity ux at the wall and the characteristic impedance of the
wall

p[Lx ,y,z] (t) = Zux[Lx ,y,z] (t) (I.23)

where the impedance is expressed as

√
1+ 1−α
Z = ρ0 c √ (I.24)
1− 1−α

and α is the absorption coefficient of the wall. Replacing Zux[Lx ,y,z] (t) from (I.23) in (I.22)
yields to

k k k h i
ux[Lx ,y,z] (t + ) = ux[Lx ,y,z] (t − ) − 2 × Zux[Lx ,y,z] (t) − p[Lx − h ,y,z] (t) . (I.25)
2 2 hρ0 2

115
Appendix I Low Frequency Enhancement System for Rooms

The term ux[Lx ,y,z] (t) can be approximated by the first order (symmetric) estimation of the
first derivative of a function 18

ux[Lx ,y,z] (t + k2 ) + ux[Lx ,y,z] (t − k2 )

ux[Lx ,y,z] (t) ≈ . (I.26)
2

Inserting the right term of (I.26) in (I.25) yields to

ux[Lx ,y,z] (t + k2 ) = ux[Lx ,y,z] (t − k2 )

  x  (I.27)
u[L ,y,z] (t+ k2 )+ux (t− k2 )

−2 hρk 0 × Z x
2
[Lx ,y,z]
− p[Lx − h ,y,z] (t) .
2

After re–arranging terms and simplification, the component of the particle velocity in the
x direction at the boundary becomes

ρ0 h
k −Z k 2
ux[Lx ,y,z] (t + ) = k
ρ0 h
ux[Lx ,y,z] (t − ) − ρ0 h
p[Lx − h ,y,z] (t). (I.28)
2 +Z 2 +Z 2
k k

Since Eq.(I.20) assumes that the particle velocity has a positive sign going outwards from
a lower index to a higher index thus the appropriate change in sign in Eq.(I.28) has to be
done. Then the equations for the boundaries at x = 0 and x = Lx are

ρ0 h
k −Z k 2
ux[0,y,z] (t + ) = k
ρ0 h
ux[0,y,z] (t − ) − ρ0 h
p[0+ h ,y,z] (t) (I.29)
2 +Z 2 +Z 2
k k

and
ρ0 h
k −Z k 2
ux[Lx,y,z] (t + ) = k
ρ0 h
ux[Lx,y,z] (t − ) + ρ0 h
p[Lx− h ,y,z] (t). (I.30)
2 +Z 2 +Z 2
k k

The equations for the boundaries at y = 0 and y = Ly are

116
Low Frequency Enhancement System for Rooms Appendix I

ρ0 h
k −Z k 2
uy[x,0,z] (t + )= k
ρ0 h
uy[x,0,z] (t − ) − ρ0 h
p[x,0+ h ,z] (t) (I.31)
2 +Z 2 +Z 2
k k

and
ρ0 h
k −Z k 2
uy[x,Ly,z] (t + ) = k
ρ0 h
uy[x,Ly,z] (t − ) + ρ0 h
p[x,Ly− h ,z] (t). (I.32)
2 +Z 2 +Z 2
k k

The equations for the boundaries at z = 0 and z = Lz are

ρ0 h
k −Z k 2
uz[x,y,0] (t + ) = k
ρ0 h
uz[x,y,0] (t − ) − ρ0 h
p[x,y,0+ h ] (t) (I.33)
2 +Z 2 +Z 2
k k

and
ρ0 h
k −Z k 2
uz[x,y,Lz] (t + ) = k
ρ0 h
uz[x,y,Lz] (t − ) + ρ0 h
p[x,y,Lz− h ] (t). (I.34)
2 +Z 2 +Z 2
k k

I.1.3 Sound source

At frequencies where the wavelength is to large compared to the dimensions of the loud-
speaker they behave as omnidirectional compact sources. Therefore the loudspeakers are
defined in the model as small volumes occupying points in the discretized space represent-
ing compact sound sources. For example if a cell size of 12 cm is used the sound source
is defined by a cube of volume V =12cm3 . Additionally the sound sources can be modeled
as membranes moving in different directions. This can be done by using the components
of the particle velocity ux , uy or uz Two type of gaussian functions were used to describe
the sound source for visualization purposes. The first presented also in 7 is an asymmetric
gaussian function defined by

1 −(t−t0 )2
pxs ,ys ,zs (t) = sin(t − t 0 )e σ2 (I.35)
σ2
and

where the -3 dB cut off frequency is given by

117
Appendix I Low Frequency Enhancement System for Rooms

dB

−20

−40

−60

1 10 100 1 kHz

0.05

−0.05
0 6.3 12.5 18.7 25.0 ms

Figure I.13: Impulse response of the different sound sources. Upper, Frequency responses.
Lower impulse responses. Magenta, asymmetric gaussian pulse. Black, gaussian pulse
1 1
with N = 200Hz and alpha = 3.5. Blue, gaussian pulse with N = 200Hz and alpha = 5.0.
Red, 4th order Butterworth impulse response. Green, real subwoofer impulse response
measured at a few millimeters from the membrane.

2
σ= (I.36)
ω

The second function is defined by the gaussian pulse

h i2
(t−t0 )− σ
− 12 α σ
2
pxs ,ys ,zs (t) = e 2 (I.37)

where the -3 dB cut off frequency is given by

2π
σ= , (I.38)
ω

and α >= 2 is the reciprocal of the standard deviation of the function. The width of the
pulse is inversely related to the value of α, a larger value of α produces a more narrow
pulse.

118
Low Frequency Enhancement System for Rooms Appendix I

To obtain plain transfer functions of the room a different function was used. A dirac
function low pass filtered by a 4th order Butterworth digital filter was used. The cut off
frequency is given by the frequency limit of the simulation method which is given by the
cell size. It was decided to let 10 cells per wavelength therefore with a cell size of 10 cm
the simulation model is valid up to 343 Hz. In Figure I.13 the time responses and the
frequency response of the sound sources used are presented.

I.2 Implementation

The simulation program was written in MATLAB. The cell size was chosen between 10
or 12 cm. The sampling frequency f s was chosen to be 8000 Hz. The simulation time
was set to approx. 1 s being t = 2N and N = 13. The initial state of the media at t=0
is assumed to be homogeneous therefore the acoustic pressure has to be set to zero. Only
two time steps of the acoustic pressure and the particle velocity of the entire room are
needed for the simulation. Therefore to simulate a room of dimensions Lx=4.2 m, Ly=7.8
m and Lz=2.76 m in principle only 4 matrixes should be initialized:

1. the acoustic pressure px,y,z (t) of size 35 x 65 x 23 x 2

2. the component of the particle velocity ux of size 36 x 65 x 23 x 2

3. the component of the particle velocity uy of size 35 x 66 x 23 x 2

4. the component of the particle velocity uz of size 35 x 65 x 24 x 2

In addition the following matrixes can be initialized:

1. the particle velocity ux,y,z (t) of size 35 x 65 x 23 x 2

2. the differences Dx in the x direction of either pressure p or particle velocities ux of

size 35 x 65 x 23

3. the differences Dy in the y direction of either pressure p or particle velocities uy of

size 35 x 65 x 23

4. the differences Dz in the z direction of either pressure p or particle velocities uz of

size 35 x 65 x 23

5. the sound pressure level of size 35 x 65 x 1 x 3

119
Appendix I Low Frequency Enhancement System for Rooms

This is done if the particle velocity u(t) is needed for special calculations as for example
the acoustic radiation power. The extra matrixes Dx,Dy and Dx are initialized to store
the result of the built-in function of MATLAB diff which differentiates between adjacent
elements. For the calculation of the sound pressure level an extra matrix is needed in
order to store the result of the squared summation of the acoustic pressures for each time
step (see Paper A Section 3.1).

Two more matrixes are initialized including a Listening Area defined by 25 virtual micro-
phones and four extra virtual microphones to record the instantaneous acoustic pressure
px,y,z (t) at the desired positions in the room;

1. Listening Area of virtual microphones of size 5 x 5 x 8192

2. Extra virtual microphones 4 x 8192

Only in these matrixes the entire simulation time is stored therefore the memory in use is
optimized. With these numbers and this room size no more than 200 MB of memory in
RAM are used by MATLAB.

In FigureI.14 the relation between the room volume, number of cells and size cell for the
simulation program is presented. For example if a room of 100 m3 of volume has to be
simulated, and the limit of frequency interest is 500 Hz, about 463000 cells are needed.
This is by assuming acceptable results up to 10 cells per wavelength.

I.2.1 Wave dispersion errors

At frequencies where the number of cells per wavelength is less than 10 dispersion error
exists. In most of this work the limit of the frequency interest was below 200 Hz therefore
by choosing a cell size from 10-12 cm the wave dispersion errors were assumed negligible.
In the literature advanced methods to correct those errors can be found for example in 22 .
The correction of these errors was not in the scope of this work.

I.2.2 Transfer function measurement

To obtain the transfer function at low frequencies of the room from a loudspeaker to the
complete listening area or a virtual microphone an impulse response of a existing closed
box loudspeaker was utilized along most of this work. This impulse response was measured
at 5 mm from the membrane in anechoic conditions. Alternatively to obtain the transfer
function from a omnidirectional compact source the Gaussian pulse or the low passed dirac
impulse was utilized.

120
Low Frequency Enhancement System for Rooms Appendix I

1 cm − 3430 Hz
9
10

2 cm − 1715 Hz

8
10

3 cm − 1143 Hz

4 cm − 858 Hz

7 5 cm − 686 Hz
10
6 cm − 572 Hz

7 cm − 490 Hz

8 cm − 429 Hz

9 cm − 381 Hz

10 cm − 343 Hz
6 11 cm − 312 Hz
10
12 cm − 286 Hz
Number of Cells

13 cm − 264 Hz
14 cm − 245 Hz
15 cm − 229 Hz
16 cm − 214 Hz
17 cm − 202 Hz
18 cm − 191 Hz
19 cm − 181 Hz
20 cm − 172 Hz

5
10

4
10

3
10

1 10 100 1000
Room Volume (m3)

Figure I.14: Relation between room volume and number of cells necessary for the simu-
lation. The inclined lines indicate the frequency limit of the simulation program per cell
size. Ten cells per wavelength are assumed for the calculation.

121
Appendix I Low Frequency Enhancement System for Rooms

I.2.3 The walls

The absorption coefficient of each wall can be defined by setting α in Eq. (I.24). Alterna-
tively since it is assumed that the absorbing characteristics of the wall are determined by
the normal specific acoustic impedance.

I.3 Graphical user interface (GUI) of The Sound Field Room

Simulator

A graphical user interface shown in FigureI.15 has been developed in order to introduce the
different parameters for the room simulation. This user interface works under MATLAB
Ver 7.0.4 and can be lunched from the command window by typing SIMRoomV02 for regular
rooms or SIMRoomVi02 for irregular rooms.

1. Room

Dimensions
The dimensions of the enclosure are defined in meters. The origin of the sim-
ulation model is assumed to be in the upper left corner of the room seen from
above as shown in FigureI.15.
Absorption Coefficients
The absorption coefficient of the walls floor and ceiling can be defined here.
A number from 0.0001 to 1 can be inserted which indicates the absorption
coefficient of the wall.
Observation Layer
The observation height is defined here, this is for the calculation of the sound
pressure level (SPL) in a horizontal plane along the room.
Corner Coordinate
In the software version for irregular rooms a text box is included where the
coordinates of the partition corner can be entered (see Figure E.4).
Odd Mesh
Here an odd number of cells is forced for the discretization of the room.
Cell Size
The size of the cell in cm is defined here.

2. Sound Sources

122
Low Frequency Enhancement System for Rooms Appendix I

Figure I.15: Graphical user interface (GUI) of the Sound Field Room Simulator

Type of loudspeaker
In this menu different sound sources can be chosen, a pre-defined impulse re-
sponse of a real loudspeaker “Real Loudspeaker”, a modeled sealed box type
loudspeaker “Sealed box” and a “Pulse” which is a Gaussian pulse with flat fre-
quency response from 0 Hz to 100 Hz approx. Additionally by chosen “Transfer

123
Appendix I Low Frequency Enhancement System for Rooms

function” the low passed dirac impulse described in Section I.1.3 is utilized to
obtain the transfer function measurement of the room.
Number of loudspeakers
Up to six loudspeakers can be switched on in each simulation by activating each
box marked as L, C, R, SR, SL, and SW.
Coordinates
The coordinates of each loudspeaker in the room are defined in the text boxes.
Gain
A gain of ± 12 dB can be adjusted for each loudspeaker.
Delay
A pure delay from 0.125 ms (1 sample) up to 500 ms (4000 samples) can be
added to each loudspeaker.
EQ
By activating this button the loudspeaker is pre filtered by an FIR filter or an
IIR filter. The coefficients of this filter should be saved before in matrix file as
“eqfilters.mat”.
Speed of Sound in Air
The speed of sound in the media can be set here.

3. Simulation Mode
Two simulation modes can be utilized, “Impulse Response” and “Single Frequency”.
The first mode is to acquire impulse responses at the listening area and the four
virtual microphones. The second mode is to obtain an approximation of the SPL in
a horizontal section of the room. In this mode the loudspeakers are driven by a single
frequency. This frequency is pre-filtered with the loudspeaker impulse response.

Frequency
If the “Single Frequency” simulation mode has been chosen then the driving
frequency can be introduced by the slider or by entering the number into the
text box.
Simulation Time
The simulation time can be adjusted by modifying the “N” order button having
N=13, N=14 and N=15 as possible options. Time responses of 2N samples can
be acquired. The length in seconds is presented in the box below the “N” order
button.
Animation
The animation for visualization purposes can be activated by marking the box
“Step by Step” under the Animation frame. By doing this the horizontal layer
at the selected height under the box “Observation Layer” is displayed on the

124
Low Frequency Enhancement System for Rooms Appendix I

Sound Pressure Level at 1.25m Height Freq: 58Hz Sound Pressure Level at 1.25m Height Freq: 52Hz
90 90

1 85 1 85

80 80
2 2

75 75

3 3
Length (m)

Length (m)
70 70

4 4

65 65

5 5
60 60

6 55 6 55

dB dB
1 2 3 4 5 1 2 3 4 5
Width (m) Width (m)

Figure I.16: Example of results of room simulations using “Single Frequency” mode and
selecting “SPL 2D”. Sound pressure level distribution in the room, Left the loudspeaker
is driven by 52 Hz. Right, the loudspeaker is driven by 58 Hz.

frame Room Results. If the “Figure” button is activated the animation is

detached from the GUI and an individual figure is created.
Movie file
By marking the box “Create Movie” an .AVI file can be created. This is done
by indexed frames taken from the animated figure. In order to properly close
the movie file before completing the total simulation time, the button “Stop”
has to be pressed until the simulation stops.
Virtual Microphones
Four “Virtual microphones can be located at different positions within the
room. The microphone coordinates can be introduced on each text box.
Listening Area
By default the program fixes a Listening Area of 25 virtual microphones cen-
tered in the room. The location of the center can be adjusted by introducing
the new coordinates in the text box “Grid”. By marking the box “Listening
Area” the location of the microphones can be seen if the animation is displayed.

4. Plot Results After a simulation is completed the results can be displayed in different
forms by selecting the menus under the Plot Results frame. If the simulation mode
“Impulse Response” was chosen the impulse response at the four microphones are
displayed in the Room Results frame. For example, by selecting “FFT” into the first
or second menu from left to right the frequency response of the time responses can be

125
Appendix I Low Frequency Enhancement System for Rooms

Sound Pressure Level at 1.25m Height Freq: 58Hz Sound Pressure Level at 1.25m Height Freq: 52Hz

90 90

80 80

70 70

60 60
1 1

2 2
dB dB
3 3

1 4 1 4
2 Length (m) 2
5 5
3 3 Length (m)
4 6 4 6
Width (m) Width (m)
5 5

Figure I.17: Example of results of room simulations using “Single Frequency” mode and
selecting “SPL Surf”. Sound pressure level distribution in the room, Left the loudspeaker
is driven by 52 Hz. Right, the loudspeaker is driven by 58 Hz.

displayed (see Figure I.18). To display the time and frequency responses in the same
frame “IR FFT” must be selected (see Figure I.19). To display frequency response
and phase “FFT Phase” can be selected. To display the 25 frequency responses at
the listening area in one plot “FFT LArea all” must be selected (see Figure I.18). To
display the frequency responses by rows from the listening area “FFT LArea rows”
can be selected. If the selected simulation mode is “Single Frequency” different
options of displaying the results can be chosen under the menus. By selecting “SPL
2D” a two dimensional color plot is displayed (see Figure I.16). By selecting “SPL
Surf” a surface plot is generated (see Figure I.17). If “SPL Surf Grid” is selected a
surface plot delimited by the listening area is displayed.

Make Figure
By pressing the “Make Figure” button a separate figure can be created with
the above selected option.

5. Room Results
The results are displayed in this panel as well as the animation frames.

Percentage bar
During the simulation a percentage bar is displayed showing the completed
simulation in percentage. This is displayed in the Room Results panel too.

126
Low Frequency Enhancement System for Rooms Appendix I

dB dB

−10 −10

−20 −20

−30 −30

−40 −40

−50 −50

−60 −60

−70 −70

−80 −80
10 100 Hz 10 100 Hz
Frequency Frequency

Figure I.18: Frequency response of the time responses. Example of results of room sim-
ulations using “Impulse response” mode. Left, selecting “FFT LArea all” all positions in
the listening area are included. Right, selecting “FFT” only the four microphone positions
are included.

Fg
Under the Room Results frame the estimated Schroeder frequency f g is calcu-
lated with the given dimensions.
Simulation Time
When the “Step by Step” function is activated the simulation time t in mil-
liseconds is displayed here.

6. Start simulation After all parameters are set the button “Calculate” has to be
pressed.
7. Stop button In order to stop the simulation the “Stop” button has to be pressed.
8. Reset button
In order to reset the simulation program to the default settings the “Reset” button
has to be pressed.
9. File menu
Under the “File” menu a set of functions can be done as for example, to save just
the results on a small file, to save the simulation settings or to save the complete
simulation on a desired location in the hard disk of the PC.

Load Settings from File

A previously saved file containing the settings of the simulation can be loaded.

127
Appendix I Low Frequency Enhancement System for Rooms

dB
Mic1
−10 Mic2
Mic3
−20 Mic4

−30

−40

−50
10 100 1k
Frequency (Hz)
−3
x 10

1.5

0.5

−0.5

−1

−1.5

0.125 0.250 0.375 0.500 0.625 0.750 0.875 1

Time (s)

Figure I.19: Frequency responses and impulse responses. Example of results of room sim-
ulations using “Impulse response” mode selecting “IR FFT”. Upper, frequency response
at the four microphone positions. Lower, time responses at the four microphone positions.

Save Settings as
All settings of the simulation are saved on a desired location in the PC.
Save Results as
The impulse responses of the listening area and the four microphone responses
can be saved on disk as a texttt.mat file.
Save all as
The complete set of variables, settings and results can be saved into a file
texttt.mat. This file can be recalled to repeat the simulation or just to display
the results.
Load Results from File
A previously saved file by “Save all as” containing all set of variables, settings
and results of the simulation can be loaded. The simulation can be repeated or
the results can be displayed again.
Close Program
This option closes the simulation program and GUI interface after displaying a
warning window.

128
II
Appendix II Low Frequency Enhancement System for Rooms

130
Appendix II

II.1 Normal Modes of Vibration in Rooms

In this Appendix the derivation of the normal modes of vibration in an enclosure from the
conventional solution of the wave equation is presented.

The propagation of sound in fluids contained in regions of space can be derived by the
linear, lossless wave equation valid for acoustic processes of small amplitude 12 which is
given by

1 ∂2p
∇2 p = (II.39)
c2 ∂t2

where c is the speed of sound in the propagation media and ∇2 is the three dimensional
Laplacian in Cartesian coordinates therefore Eq.II.39 can be rewritten in terms of the
Cartesian coordinates

∂2p ∂2p ∂2p 1 ∂2p

+ + = (II.40)
∂x2 ∂y 2 ∂z 2 c2 ∂t2

In order to calculate the normal modes of vibration in the enclosure a conventional solution
to the wave equation (II.39) can be given in the form of

p(x, y, z, t) = Ψejwt (II.41)

where Ψ is a function of position and substitution of k = w/c yields to a

∇2 Ψ + k 2 Ψ = 0 (II.42)

131
Appendix II Low Frequency Enhancement System for Rooms

which is well known as the Helmholtz equation, after separation of variables and since Ψ
is a product of three functions each dependent on only one the dimensions Ψ(x, y, z) =
X(x)Y(y)Z(z) therefore Eq. (II.41) becomes

p(x, y, z, t) = X(x)Y(y)Z(z)ejwt (II.43)

and Eq. (II.42) can be rewritten as

∂2Ψ ∂2Ψ ∂2Ψ

+ + + k2 Ψ = 0 (II.44)
∂x2 ∂y 2 ∂z 2

2d
( dx 2
2 + kx )X = 0
d2
( dy 2
2 + ky )Y = 0
2
d
( dz 2
2 + kz )Z = 0
(II.45)

where the angular frequency must be given by

 w 2
= k 2 = kx2 + ky2 + kz2 (II.46)
c

Considering a rectangular room of dimensions Lx , Ly and Lz and assuming that the normal
component of the particle velocity is zero at all walls therefore the boundary conditions
become

   
∂p ∂p
= =0
 dx x=0  dx x=Lx
∂p ∂p
dy y=0 = dy y=L =0
y

∂p
  
∂p (II.47)
dz z=0 = dz z=L =0
z

Since the energy can not escape from the enclosure a solution to the wave equation can
be the cosines thus Eq.(II.43) becomes

132
Low Frequency Enhancement System for Rooms Appendix II

plmn = Almn cos kxl x cos kym y cos kzn zejwlmn t (II.48)

where the components of k are

kxl = lπ/Lx l = 0, 1, 2, · · ·
kym = mπ/Ly m = 0, 1, 2, · · ·
kzn = nπ/Lz n = 0, 1, 2, · · · (II.49)

by using this solution the only allowed angular frequencies where a standing wave will
occur are given by
s
 lπ 2  mπ 2  nπ 2
wlmn = c + + (II.50)
Lx Ly Lz

or in 14 as
s
c  n 2  n 2  n 2
x y z
fn = + + (II.51)
2 Lx Ly Lz

where c is the speed of sound in the air, nx , ny and nz are integers starting with 0, 1, 2,...
and Ly , Lx , Lz are the dimensions of the room. Each standing wave has its own modal
frequency and these are specified by the integers l, m, n or nx , ny , nz . The zones where
there will be minimum pressure are called nodes and the zones where there is a maximum
pressure are called antinodes. If a sound source is located at a node of a modal frequency
this mode will not be exited, on the contrary if a sound source is placed close to or at a
antinode of a modal frequency this will be greatly exited. Similarly if a receiver is located
at a antinode its output will be great and if the receiver is located at a node of a mode its
output will be minimum.

133