E LEKTROTEHNIŠKI VESTNIK 78(3): 91-95, 2011
E NGLISH EDITION
MLS and Sine-Sweep
Sweep technique comparison in room-acoustic
room
measurements
Franco Policardi
Università di Bologna, viale Risorgimento 2, 40136 Bologna, Italia
franco.policardi@studio.unibo.it
Abstract.Modern room acoustic measurements are still in their evolution phase,, as stated in ISO 3382. MLS test
signals coexist with Sine-Sweep test signals because both show pros and cons. While MLS technique provides
good rough data results, Sine-Sweep
Sweep technique is more accurate butalso more time-consuming consuming and
computationally demanding. During 2009Trieste Politeama Rossetti room-acoustic
room acoustic measurement campaign, a
comparison was made between the two techniques.
techniques After a theoretical discussion,the
the paper presents some testing
results.(77)
Key words:room acoustics, ISO 3382, RT60, impulse response, MLS, Sine-Sweep technique
by pistol shots or balloon explosions) presents various
1 INTRODUCTION problems as for example low S/N, possible non-linear
phenomena due to too fast signal slew-rate
slew and last but
Every physical space is definedfined by specific acoustics not least extremely low and high frequencies,manacing
frequencies
and evenven in contemporary virtual-reality,
virtual acoustics to damage electroacoustic transducers.
plays a fundamental role in creating reality illusion. Early acousticians used hand-claps
hand evaluation to
Reliable room-acoustic
acoustic measurements have been understand room acoustic behaviour;
behaviour but this was
developed between the end of XIXth and beginning of happening during a less technological era.
era It was a non-
XXth centuries leading to a multitude ltitude of acoustic-
acoustic technological approachin a lot less measured room
parameter definitions through specific algorithms,
alg first background noise.Different non-direct
non IR measurement
of all reverberation time decay for the first 60dB (RT60). techniques have then been developed, starting with
Reality still shows differences between calculated and organ pipes in 1895 by W. C. C Sabine [2], thenthe
measured parameters and ISO 3382 standard [1] gives artificially
ificially generated white noise and now pseudo-
some main guidelines in room acoustic measurements.
measurement random maximum length sequence
seque (MLS) and Sine-
Besides different measurement
easurement setups, this standard Sweep signals coexisting.
allows different room excitation
itation signalsand
signal a lot of
freedom in different operator choiceswhichcan
choices interfere
with final results.
2 MEASUREMENT METHODS
The fundamental hypothesis in analysingan
analysing acoustic 2.1 Theory
or electroacoustic system behaviour is its linear time
invariant (LTI) behaviour during measurement.
measurement This is A simple method to measure any system consists of
necessary to determine its impulse response (IR) and to applying a unitary impulse in input
in and to observe its
calculate from it its frequency response and other output. The more the input
put signal is similar to the ideal
parameters. Room
oom impulse response g(t) is the pressure-
pressure one, the more accurate will be the system impulse
time response function at a specific receiver position as response as shown in Figure 1.
a result of the excitation by an impulsive source. This
enables research ofaa specific algorithm, defining the
univoque relation between sound source and receiver
through the unknown behaviour of the acoustic space.
energy function is therefore E(t) = [g(t)]2.
The time-energy
By definition,IR is the system output signal obtained
by the input signal,mathematically defined as Dirac δ,
which is digitally PCM defined as a sequence of 0s
containing a 1 at a full-scale strength.
strength This sort of IR Figure 1: Dirac δ impulse, system under measurement and
measurement (many times analogically approximated system IR
Received 25 May 2011
Accepted 10 June 2011
92 POLICARDI
This method is accurate but,to
,to give good results, input
signal has to be as similar as possible to the ideal
impulse. Unfortunately, in real conditions,
conditions Dirac δ is too
short to contain enough energy and simple
s calculation
shows that at normal PCM 44,1kHz 44,1 sampling
has a 10-6s magnitude
frequency, a single pulse lengthhas
order (about 22,68µs) and is absolutely too short to be
correctly reproduced by any known power loudspeaker
for at least 60dB above background noise.
noise Even if this
were possible, any small non-linearity
non might
compromisee good measurement results. A real situation
is shown in Figure 2 [3],, where at least
transducerdistortions are surely added.
added
Figure 3: MLS self-correlation produces a Dirac δ
Any difference from Dirac δ shows acousticnon-
linearbehaviour. From this difference,the
difference measured item
frequency response and many other significant
Figure 2: Linear system measurement with noise
n parameters can be easily obtained through FFT. FFT That is
why this measurement method has been so popular
2.2 MLS method between 1975 and 2000 (known as TEF or MLSSA), ML
MLS signal is useful to measure any input-output when computational
utational power was still limited.
limited In the
system asmost equipment and room acoustics are. At particular case of room acoustics
ustics, it has to be noted that
the beginning (1965)MLS signal was generated by the MLS signal duration must be longer than room
primitive
tive polynomials and subsequently
subsequent by digital shift reverberation time RT60 to allowsound field to reach a
registers. MLS signal is white noise comparable and steady-state in the room and thus ensure correct
white noise is non-periodical and random, thus its measurement results.
measurement requires long measurement
rement time averaging Thorough and extensive studies devoted to analysing
(minutes) to be sure to correctly
tly estimate its spectrum. this specific measurement technique have revealed its
In practice, this kind of signal consists of a randomly intrinsic limits as different types of distortion can occur
distributed sequence of same amplitude, same positive in IRs because the entire system is not LTI,
LTI determining
and negative impulses,, so that the t sequence is uncertainty in different results. Instead of
symmetrical around 0.. This sort of signal contains at anhomogenous Schroeder function functio decay, different
least 1 x 104 pulses per sequence, thus having higher peaks occur in measurement,,mainly due to non-LTI
energy content than the theoretical Dirac δ signal, which electroacoustic system behaviour.
behaviour
contains just one pulse per sequence [4]. MLS 2.3 Linear Sine-Sweep method
numerical sequence length is 2n - 1, where n is the
number of digital shift registers; in acoustics 16 digital Little by little other methods have entered the acoustic
shift registers are normally used implying 65.535 measurement field:first the linear Sine-Sweep method
samples for a typical MLS room-acoustics
acoustics measurement (LSS) better known as a time-delay
time spectrometry or
signal. TDS using TEF analyser and afterwards exponential
A very important feature ofMLS signal use is that its sine sweeps (ESSs). Introduced
ntroduced by R. C. Heyser in 1967
self-correlation
correlation calculation generates a perfect Dirac δ, [6],the LLS method uses a linear time-growing time
as its self-correlation generates the single pulse shown frequency to »sweep« the measured item, in this
in Figure 3. remembering W. C. Sabine'ss five different-frequency
Advantages of the pseudo-random
random MLS signal are 1) its organ-pipes early experiments in room acoustics [2].
complete frequency spectrum, 2) given its determined TDS specifically consists of a broad-spectrum
broad technique
sequence, it is easy to mathematically find its time using a signal stimulus mathematically expressed as
reversal MLS-1, 3) because MLS is a binary sequence, exp[i θ (t)]; delay-tracking
tracking filters process the response
its convolution product is simplified as multiplications of the measured item extracting useful informations.
become sums, 4) because of signal and time-domain
time Sine-Sweep techniques show important
convolution computational simplicity, it only requires computational advantages as the test signal inverse filter
Fast Hadamard Transform (FHT) and last but not least x(t)-1 is just test signal x(t) time reversal. Unfortunately,
Unfortunately
5) given its flat sound spectrum, it is easy to calculate MLS FHT T computational simplicity is lost but the actual
the IR signal frequency response. computational power allows for a real-time and precise
solution through specific »select-save
»select FFT« fast
MLS AND SINE-SWEEP
SWEEP TECHNIQUE COMPARISON
COMP IN ROOM-ACOUSTIC MEASUREMENTS 93
convolution. The main advantage of Sine-Sweep
techniques is that it becomespossible
possible to discriminate
distortions from the signal even if the whole measuring
chain is not LTI or in any way non linear as for example
electroacoustic transducers,, as shown in Figure 4.
4
Figure 6: ESS and measurement with distortion products
(upper parallel)
In the specific case of ESS, test-signal
test deconvolution
inverse filter x(t)-1 has to be slightly modified to avoid
uncorrect frequency filtering,and and to compensate ESS
Figure 4: LSS and measurement with distortion test signal which is energetically not frequency-flat but
products(upper not parallel) decreases by 3dB per octave, in this similar to pink
noise. The solution is very simple consisting in a 6dB
Convolving y(t) output signal withx(t)-1 test-signal per octave positive-amplitude
amplitude filtering
filter directly during
inverse filter, it becomespossibleto separate high-order
high computational generation oftest signal deconvolution
distortion products to obtain vertical right desired IR inverse filter x(t)-1 [8].
containing the F[x(t)] measured item behaviour,
behaviour as As for LSS, ESS output y(t) x(t)-1 inverse filter =
shown in Figure 5. IR and different harmonic non-linear contributions
forestall IR. IR is on the right and other order harmonic
distortions are now perfectly distinguishable on its left,
left
in this being better than LSS,, as shown in Figure 7. 7
Some electroacoustic experiments showed up to 50
harmonic products, allowinglowing very deep measured-item
measured
understanding.
Figure 5: LSS x(t)-1 inverse filter and IR in time-frequency
time
domain
main contribution is
LSS techniquemain i its possibility to
discriminate between system IR and electroacoustic
non-linearities. LSS technique hardly permits to well
separate different harmonic-distortion
distortion products and
unfortunately still sufferss from time aliasing if output -1
Figure 7: ESS x(t) inverse filter and IR in time-frequency
time
signal is not significantly longer than RT60, in the last domain
remembering MLS technique.LSS method is actually
seldom in use in room acoustics because of the growing
interest in ESSs. 3 EXPERIMENTAL RESULTS
2.4 ExponentialSine-Sweep method
During Trieste Politeama Rossetti 2009 measurement
An interesting
nteresting development from LSS is ESS campaign, different measurement systems were used to
familyin which frequency-growthth law can be arbitrarily investigate their result comparability.
comparability
chosen. ESS measurement technique
nique is a frequency- The computer-generated
generated Dirac δ (Section 2.1), MLS
determined sine sweep exponentially growing with (Section 2.2), and ESS (Section 2.4), signals drovea
time,which in the specific case of logarithmic
logar lawis 01dB acoustic-measurement
measurement power amplifier linked to a
much more similar tohuman ear and music logarithmic 01dB Dodecaedronacoustic-measurement
measurement loudspeaker
behaviours. In this case, y(t) shows a parallel time- system. Receivers were two Neumann KU100 dummy- dummy
frequency logarithmic-plane slope of x(t) as shown in head microphone ne systems. Even if planned, LSS-TDS
Figure 6. (Section 2.3),and TEF equipment was unfortunately not
available during the measurement campaign.
Starting from the beginning as stated in 2.1, a
progressively more energetic Dirac δ was sent tothe
power-amplifier
amplifier and loudspeaker system, but rather than
power amplifier temperature slight growth because of
94 POLICARDI
the joule effect, almost no appreciable impulse came exempt from system non LTI, as shown in Figure 9 and
from loudspeaker system. As expected, apart from some because averaging was not performed. A comparison
clicks from loudspeaker system tweeters, no appreciable between Figures 8 and 9 clearly shows a greatly better
Dirac δ resulted, because mid-frequency and even more S/N, which can be correctly calculated for 60dB (now at
low-frequency power loudspeaker moving system lower energy levels, too) and RT60 is now surely
inertia was too big. Just to remember, a normal detected in late room-acoustic decay also. Its slope is
loudspeaker moving system is made of diaphragm, dust now evident and measurement results can be assumed
cap, voice coil, suspension system (spider + much more correct, compared to MLS method.
surround)voice coil connexion cables and voice coil
front + rear moved air mass [9].
MLS and ESS methods were then compared using
the same electroacoustic measuring chain in the same
Trieste Politeama Rossetti theatre at the same time.
Experimentswere made at various receiver positions and
most interesting comparison results are here shown for
the same microphone in the same position.
As expected, MLS technique showed to be sensitive
toacoustic noise,uncorrect low-frequencies
omnidirectional measurement loudspeaker system
reproduction, infrasound, amplifier insufficient high-
frequencies slew-rate, joule effect in the electroacoustic
chain, harmonic and/or intermodulation products and
many more as show different peaks in Figure 8. RT60
decay plot can be correctly calculated for 20dB only and Figure 9: distortion products free ESS measurement
so for RT20 only, not being sure if at lower energy levels
or on the other hand in the remaining part of the RT
(which is a time expression) the room behaviour is 4 DISCUSSION AND CONCLUSION
extensible.
As sound perception in humans is not homogenous [10],
some acoustic objective evaluation parameters had to be
developed to describe room acoustics as for example
RT60. Slow evolution in room acoustics seems nownot
to be mainly due to recently solved computational
requirements but to imperfect electroacoustic chain
components and to measurement techniques.
Electroacoustic-chain components drawbacks are
related to component high costs and mainly to their
physical limits, as linear power amplifiers do not exist
as long as linear loudspeakers on 3D propagation planes
[9]. Operators link the two main problems, as they are
mainly not enough trained and measurements are
consequently not comparable because ofISO 3382
uncertainties. Huge work in this particular applied
physics sector is still needed, where ESS solved at least
Figure 8: MLS measurement with distortion products previously impossible discrimination between
measurement chain distortion and room acoustic
Even if still in use, MLS signal implies an incorrect behaviour even in non LTI situations.
or at least aleatory results risk. In these conditions, it As theory first and experimental results afterwards
normally becomes necessary to average many demonstrate, ESS in room-acoustic measurements is a
measurements(losing in result accuracy).Here is then big step towards the comprehension of the room IRat a
discovered one of the main causes preventing further specific listener position and this is the main difference
correct calculations during room acoustics measurement between MLS and ESS measurement methods. ESS
or restoration. shows the specific algorithm which defines the
As expected, ESS techniqueallowedfor acoustic- univoque relation between the sound source and the
system linear IR extraction with no electroacoustic sound receiver towards the step by step less unknown
distortions and quantifiedthe whole system non-linear behaviour of acoustic spaces.
response weighting every harmonic order IR. Linear IR
measurement had better S/N and was completely
MLS AND SINE-SWEEP TECHNIQUE COMPARISON IN ROOM-ACOUSTIC MEASUREMENTS 95
ACKNOWLEDGEMENTS
The author acknowledges ing. Gianni Amadasi from
01dB Padova for the 01dB acoustic measurement power
amplifier and 01dBDodecaedron loudspeaker system
and ing. Stephan Peus from Neumann Gmbh Berlin for
KU100 microphone systems lending during
measurement sessions.
BIBLIOGRAFY
[1] ISO 3382 standard.
[2] Sabine Wallace Clement “Collected papers on acoustics”,
Harvard University Press 1922 (post).
[3] Cuomo Mario “Analisi qualitativo-comparativa di sistemi di
riverbero artificiali”, tesi magistrale, Adria 2009.
[4] Karjalainen Matti “Transfer-function measurements in audio”,
(Eds.) Äänenkäsittelyn seminaari 1996: Digitaaliaudion
signaalinkäsittelymenetelmiä (Seminar on audio signal
processing: “Signal processing methods in digital audio”),
Helsinki University of Technology , Laboratory of Acoustics and
Audio Signal Processing, Report 41, Espoo, Finland 1996,
pp.181-204.
[5] Schroeder Manfred Robert “Synthesis of low-peak-factor signals
and binary sequences with low autocorrelation”, IEEE Trans.
Information Theory, January 1970, pp.85-89.
[6] Heyser Richard Charles“Acoustical measurements by time delay
spectrometry”, J. Audio Eng. Soc. 1967, 15 pp. 370-82.
[7] AA. VV. “Acustica musicale ed architettonica”, chapt. 19, ed.
UTET, Torino 2009.
[8] Farina Angelo, Capra Andrea, Campanini Simone “La misura
dela risposta all'impulso per la caratterizzazione di sistemi
acustici e vibrazionali”, Seminario di Acustica “Strumenti e
metodi di misura per l’acustica e le vibrazioni”, Università
Politecnica delle Marche - Dipartimento di Energetica, Ancona
2008, pp. 75-90.
[9] Policardi Franz “Linearity and reliability in measurements”,
European Acoustical Association - EAA Euroregio, Ljubljana
2010, Acta Acustica vol 96 pp. 96.
[10] Burnik Urban, Tasič Jurij Franc “Človeško zaznavanje in
objektivno vrednotenje zakasnitev multimedijskih storitev”,
Elektrotehniški vestnik, 2006, 73 (1), pp. 7-12.
Franco Policardiwas born in Trieste, Italy in 1963 where he
graduated in 1983. He received his Master in Music
technologies degree in Adria and Ph.D. degree in applied
physics (Industrial engineering) from Bologna University in
2005 and 2010, respectively. From 1985 to 2002 he was
incharge of top quality recording and live music concert
amplification (Wiener Philharmoniker Orkester, Pavarotti and
friends...) and has afterwards extended his early interests in
room acoustics.Currently he works towards his second Ph. D.
in Mechanical engineering at the L'Aquila University.
