Some Attributes of Linear-Phase Loudspeakers

Compiled by Bohdan Raczynski, March 2013

Introduction

Source: http://www.kfs.oeaw.ac.at/content/blogcategory/0/378/

“…Early psychoacoustic research suggested that the human auditory system is

insensitive to differences in the relative phases of spectral components of a multi-
component sound. However, research from the last two decennia provides evidence
that listeners can detect phase differences between the stimulus components that
interact within a single auditory filter. The most impressive demonstration of phase
sensitivity is given by the masker-phase effect, i.e. the more than 20-dB variation in
masking effect caused by a harmonic complex when varying the phase relations
between its components. This masking paradigm is widely used to obtain a
psychoacoustical measure of the phase response of the cochlea….”

I must admit, I did not know about the above research and it’s results. I have
been researching the internet for about a year before I came across the above, simple
information. There is a lot more too. Now, I realised, there is a volume of research
results that clearly indicates, that rather than asking “is phase distortion audible?” we
should now be asking question “how does the phase distortion manifest itself?”.

Naively, and without any prior experience in how should I actually do it, I
conducted my own listening tests by comparing the sound from traditional, minimum-
phase loudspeakers to the sound of linear-phase loudspeakers. I am talking here about
acoustically linear-phase loudspeaker. During my short, initial listening tests on
linear-phase loudspeakers, I was surprised by how indifferent the linear-phase mode
was to my ear.

Was I doing the right thing then?. This result definitely required further
investigation on much more diverse listening material.

Listening Habits

Traditionally, when I listened to the quality of the sound reproduced by my

audio playback equipment, I focus on tonal balance (frequency response), dynamics
of the sound (SNR), residual noise floor ( inaudible ), distortion ( inaudible ).

Interestingly, all of the above characteristics can be assessed and visualized in

frequency domain. It was simply the easiest way to listen to the sound and evaluate
what I was hearing, but I now realize, that I was only considering the steady-state
analysis in the frequency domain – see pictures below.
Frequency response, distortion, dynamics and noise floor – all in frequency domain.

I was doing the same type of analysis over, and over again for years, and grew
accustomed to this ritual. It was easy to compare with measured results, so it felt
comfortable, that I can correlate my measurements with what I can easily hear (or can
not hear).

Recently, things have changed for me. I came across a simple paper,
http://www.audiophilerecordingstrust.org.uk/articles/speaker_science.pdf which
inspired me to take a more comprehensive look at my listening tests. Having read the
paper, I re-examined information from other internet re-sources, and as a result I came
to the conclusion, that my listening tests were only a starting point of what I should
have listened to when examining linear-phase loudspeakers.

To put it simply – I needed to significantly extended the evaluation of time-

domain characteristics of the loudspeaker in my listening habits.

In the brief conclusions of my short, initial listening tests presented in

http://www.bodziosoftware.com.au/Home_Theatre_Conclusions.pdf I have pointed
out one perceptible difference – I felt closer to the stage/musicians. This was more of
an accidental and unexpected impression, to which I did not pay much attention. But
this indeed relates to time-domain characteristics of a loudspeaker, rather than
frequency domain.

Yes, it appears, that I have been covering only half of what I should have been
paying attention to. And the paper mentioned above made it startlingly clear to me.

New Listening Habits

The remaining part of this paper is my crude attempt to summarise audible

attributes of linear-phase loudspeakers. This is what you need to listen for when
evaluating linear-phase loudspeakers. I do not pretend, that the list is complete, but
it’s a start. It clearly points to the time-domain characteristics of the loudspeaker, and
this is something, which may of us (till recently, including myself) are not
accustomed to. I simply did not know what to listen for.

Below, I present the “nominated attribute(s)”, showing the source, followed by

a short description from the source.
1. Tighter bass

2. Wider and deeper sound stage (quite dramatic)

Source: http://redspade-audio.blogspot.com.au/2012/03/bathurst-2011-audio-event-of-year.html

DEQX demo

“…A highlight this year was a demo of the capabilities of DEQX. This came about
from discussions of my active crossover listening comparisons, in which a small
group could not hear any improvement with DEQX. Terry argued that we had
dumbed down the DEQX and prevented it from showing what it can do. This is
certainly true, we wanted to test sound quality only and in that regard found no reason
to spend the extra compared to cheaper options. However, Terry set up a demo in
which two profiles were created on DEQX. One was limited to the processing power
of MiniDSP and DCX. The other allowed DEQX to strut its stuff. In particular, it was
allowed to correct for phase and group delay. We then blind tested this with instant
switching, not knowing what was being heard. I was the first to sit in the chair and do
the demo and quite soon I didn't need to be told which was which, because the
difference was obvious.

Changes noticed with DEQX:

much tighter bass

wider and deeper sound stage (quite dramatic)

Both had a basic level of time alignment with digital delays. Both were matched in
level and in response closely. These differences were related to the group delay
correction. Without it, the sound was flat and almost lifeless in comparison.

I then watched as others sat through the demo, each person noticing the same
differences, differing only in the amount of time taken before declaring what they
heard…..”

Personally, I can testify to the tighter bass audible during linear-phase mode. I
operate large, 18”/vented enclosure subwoofers, tuned to 20Hz. Playing impulsive
sounds, in minimum-phase mode, the subs overshot and then add and prolong the
ringing - past steep, impulse-like signals. This unwanted flabbiness is unfortunately
audible in minimum-phase mode on low-frequency impulsive signals.
http://www.bodziosoftware.com.au/LP_MP_Subwoofer_Tests.pdf

However, in linear-phase mode, the punch is still deep, but tight, without the
“aftersounds”.
3. Realism

Source: http://www.audiophilerecordingstrust.org.uk/articles/speaker_science.pdf
(This is a must-read article in it’s entirety)

"…..Another area in which loudspeakers are disreputable is in the neglect of the

time domain. The traditional view is that all that matters is to be able to reproduce
continuous sine waves over the range of human hearing.

A very small amount of research and thought will reveal that this is a misguided
view. Frequency response is important, but not so important that the attainment
of an ideal response should be to the detriment of realism. One tires of hearing
that "phase doesn't matter" in audio or "the ear is phase deaf". These are outmoded
views which were reached long ago in flawed experiments and which are at variance
with the results of recent psychoacoustic research.

The ear works in two distinct ways, which it moves between in order to obtain the
best outcome from the fundamental limits due to the Heisenberg inequality. The
Heisenberg inequality states that as frequency resolution goes up, time resolution goes
down and vice versa. Real sounds are not continuous, but contain starting transients.
During such transients, the ear works in the time domain. Before the listener is
conscious of a sound, the time domain analysis has compared the time of arrival
of the transient at the two ears and established the direction. Following the
production of a transient pressure step by a real sound source, the sound pressure must
equalise back to ambient.

The rate at which this happens is a function of the physical size of the source. The ear,
again acting in the time domain, can measure the relaxation time and assess the size of
the source. Thus before any sound is perceived, the mental model has been told of the
location and size of a sound source.

In fact this was the first use of hearing, as a means of perceiving a threat in order to
survive. Frequency analysis in hearing, consistent with the evolution of speech and
music came much later. After the analysis of the initial transient, the ear switches over
to working in the frequency domain in order to analyses timbre. In this mode, the
mode that will be used on steady state signals, phase is not very important. However,
the recognition of the initial transient and the relaxation time are critical for
realism. Anything in a sound reproduction system which corrupts the initial transient
is detrimental.

Whilst audio electronics can accurately handle transients, the traditional loudspeaker
destroys both the transient and the relaxation time measurement. Lack of attention to
the time domain in crossover networks leads to loudspeakers which reproduce a single
input step as a series of steps, one for each drive unit at different times..."
4. Depth

5. Resolution

6. Separation of ambience

Source: http://www.bostonaudiosociety.org/bas_speaker.htm
http://www.bostonaudiosociety.org/pdf/bass/

Boston Audio Society has an interesting view on time-corrected loudspeakers.

“….If the stereo loudspeakers differ in their time-shift behaviour by more than about
thirty millionths of a second (or a finer tolerance, perhaps, for critical listeners), the
stereo image will be perceptibly smeared. The two speakers must "speak" together
at all frequencies if the subtlest details in the stereo field are to be preserved.

This, quite simply, may be the principal advantage to be gained from "linear-phase"
or "time-corrected" loudspeakers. The manufacturers who are striving to reduce the
time dispersion of loudspeakers to zero may also be ensuring that there will be no
significant differences in signal propagation timing between the two speakers in a
stereo pair. The delicate timing information in a stereo recording is thus accurately
retained and is transmitted to the listener unaltered…”

They also point to some of the advantages of such loudspeakers:

1. Depth.
This may surprise some listeners when they first hear it, since many speakers (and
records) elicit only a general left-to-right spread. But "stereo", as originally
conceived, implied a three-dimensional sound in which voices or instruments could be
localized at different apparent distances from the listener as well as at various lateral
positions. Listeners to time-aligned speakers consistently report hearing a stereo
image with unusual depth.

2. Resolution.
The stereo image is reproduced precisely, each voice or instrument having its proper
place and width. In complex sound sources such as symphony orchestra, individual
instruments can be resolved with unexpected clarity. In the old cliche, "I hear details I
never knew were in the recording. " Some listeners have incorrectly attributed the
improved resolution of detail to more accurate transient response, but the better
definition of details is simply the result of the reduction of blending in the stereo
image.

3.Separation of ambience.
With loudspeakers whose stereo image is slightly blended because of time-smear, any
hall ambience or reverberation in the recording tends to become slightly mixed with
the instrumental sounds, causing coloration of those sounds. Consequently, with such
speakers closely-microphoned recordings tend to sound better because of their
distinctly defined sound. But with time-corrected loudspeakers, the ambience is
resolved as a separate sound, and larger amounts of hall ambience in recordings can
be enjoyed…….”
7. Inter-channel accuracy of sound reproduction.
Source: http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf

“…….5. Audibility of Phase Distortion

One of the confusing issues regarding the audibility of phase is that the discussion is
generally considered to be a single topic when in reality should be discussed as two
distinct situations. The audibility of phase distortion must be evaluated as follows:

1) Inter-channel phase distortion. Characterized as differences in phase response

between two or more channels.

2) Intra-channel phase distortion. Characterized by non-linear phase response within a

channel with the stipulation that the phase response is matched between all channels
within the system (i.e. inter-channel phase distortion is equal to 0 msec)

6. Inter-Channel Phase Distortion

We use the amplitude and phase relationship between the sounds received by our ears
to localize the source of the sound. Modern audio systems use this attribute to create
what is known as imaging, or the perception that an instrument or vocal is coming
from a location that is different than the actual speaker location. The audible effects of
inter-channel phase distortion can be easily demonstrated by simply reversing the
speaker connections on one channel of an otherwise properly configured stereo
system. The loss of imaging is immediately noticeable even to those without a trained
ear. Granted this test is rather dramatic and 180 degrees of inter-channel phase
distortion is not indicative of standard operation but it does demonstrate the potential
effects. As a result of this test, you would be hard pressed to find someone that would
argue that 180 degrees of inter-channel phase distortion is acceptable, but where
between the two extremes is the threshold of audibility? Tom Holman reports [10]
that in his laboratory environment at the University of Southern California that
is dominated by direct sound, a channel-to-channel time offset equal to one
sample period at 48 kHz is audible. This equates to 20 µsec of inter-channel
phase distortion across the entire audio band. Holman [10] also mentions, “one
just noticeable difference in image shift between left and right ear inputs is 10
µsec”.

7. Intra-Channel Phase Distortion

Recall that we use the differences in signal amplitude and phase to localize or
determine the source of sound and relatively small amounts of inter-channel phase
distortion can be audible. But how does our hearing react when each channel in a
multi-channel system is subjected to non-linear phase response but the phase response
is matched between all channels? Douglas Preis [11] did an extensive survey of
existing literature and Tom Holman's [10] experiences and research through his work
at USC gives us an interesting insight into this phenomenon. Both report that the
threshold of audibility is frequency dependent, which correlates with all other
audibility thresholds. In laboratory environments when using test tones and
headphones, research has shown that the human ear is sensitive to intra-channel phase
differences of 0.25 msec [8] or +/-0.5 msec [9] in the mid-range with the threshold
increasing at higher and lower frequencies. Preis states “the tolerances shown.... are
not directly applicable to speech or music signals irradiated by loudspeakers in a
reverberant environment. Most likely, the perceptual thresholds for these conditions
would be at more than twice those shown”. Essentially, the data suggests that for high
quality music or speech reproduction in a reverberant environment intra-channel
phase distortion of 1 msec is inaudible to a trained listener. Notice that this threshold
is a relatively conservative statement and is still two orders of magnitude greater than
that for inter-channel phase distortion!.....”

8. Precedence effect or “law of the first wavefront”

Source: http://en.wikipedia.org/wiki/Precedence_effect

“…..The precedence effect or law of the first wavefront is a binaural

psychoacoustic effect. When a sound is followed by another sound separated by a
sufficiently short time delay (below the listener's echo threshold), listeners perceive a
single fused auditory image; its perceived spatial location is dominated by the
location of the first-arriving sound (the first wave front). The lagging sound also
affects the perceived location. However, its effect is suppressed by the first-arriving
sound…..

The precedence effect appears, if the subsequent wave fronts arrive between 2 ms and
about 50 ms later than the first wave front.

The precedence effect is important for the hearing in enclosed rooms. With the
help of this effect it remains possible to determine the direction of a sound source
(e.g. the direction of a speaker) even in the presence of wall reflections….”

9. Importance of Phase in Transients

Source: http://sound.media.mit.edu/Papers/kdm-phdthesis.pdf

Page 44

“….Since Helmholtz, there has been a figurative tug-of-war between proponents of

his “spectral theory” of musical sound and researchers who recognized the importance
of sound’s temporal properties. Analysis-by-synthesis research, by trying to discover
methods for synthesizing realistic sounds, has revealed several critical limitations of
purely spectral theories. Clark demonstrated that recordings played in reverse—which
have the same magnitude spectra as their normal counterparts—make sound-source
identification very difficult. Synthesis based on Fourier spectra, with no account of
phase, does not produce realistic sounds, in part because the onset properties of the
sound are not captured (Clark et al., 1963). Although most musical instruments
produce spectra that are nearly harmonic—that is, the frequencies of their components
(measured in small time windows) are accurately modeled by integer multiples of a
fundamental—deviations from strict harmonicity are critical to the sounds produced
by some instruments. For example, components of piano tones below middle-C (261
Hz) must be inharmonic to sound piano-like (Fletcher et al., 1962). In fact, all freely
vibrating strings (e.g., plucked, struck, or released from bowing) and bells produce
inharmonic spectra, and inharmonicity is important to the attack of many instrument
sounds (Freedman, 1967; Grey & Moorer, 1977). Without erratic frequency behavior
during a note’s attack, synthesized pianos sound as if they have hammers made of
putty (Moorer & Grey, 1977).

So Helmholtz’s theory is correct as far as it goes: the relative phases of the

components of a purely periodic sound matter little to perception. However, as
soon as musical tone varies over time — for example, by turning on or off —
temporal properties become relevant. In the real world, there are no purely
periodic sounds, and an instrument’s magnitude spectrum is but one of its
facets…..”

10. Pitch, Timbre, and Source Separation

Source: David Greisner http://www.davidgriesinger.com

http://www.davidgriesinger.com/Acoustics_Today/Pitch,%20Timbre,%20Source%20
Separation_talk_web_sound_3.pptx
11. Confirmation of two-stage processing by the ear, as discussed in (3).

Source: http://www.hauptmikrofon.de/theile/ON_THE_LOCALISATION_english.pdf

4.3.1 The “law of the first localisation stimulus”

“….For a conventional stereo-up, a phantom source shifts from = 0° to = 30°

if the time difference between two broadband loudspeaker signals is increased
from zero to about 600 µs. The association model could explain this phenomenon
(time- as well as level-based stereophony) by means of psychoacoustic principles of
the gestalt association stage. The localisation stimulus arriving at the gestalt
association stage first has a greater weight compared to the second stimulus (the
equivalent for level based stereophony would be the localisation stimulus with the
higher level). Despite their identity and relative time delay, the localisation stimuli
can be discriminated, since each of them is present in the binaural correlation pattern
in a complete and discriminable form (see Section 4.1).

Yet, a further increase in the inter-channel time difference leads to an exceedance of

the maximal time delay τmax. For stationary broadband signals (continuous noise),
this causes a disruption of the localisation stimulus selection, which manifests itself in
the form of a reduced suppression of the comb filter effect, for example. In this
particular sound field constellation, the law of the first wavefront cannot be observed
in accordance with the association model. Analysable wavefronts that would allow for
a localisation stimulus selection of the impinging sound components do not exist.

In contrast, for non-stationary impulsive signals (clicks, speech, impulsive tones) an

increase in the inter-channel time difference has a different effect. In the association
model, evaluation of the amplitude envelope ensures that the primary and the delayed
sound (reflection) can be discriminated as localisation stimuli. According to a
hypothetical function of the gestalt association stage, the primary localisation stimulus
determines the auditory event. It does this even more so the larger the time difference
between the arriving localisation stimuli gets. Only when a time difference of about
10 … 30 ms is exceeded will the subsequent localisation stimulus gain in perceptual
weight. Beyond the echo threshold (for a definition see BLAUERT 1974), it will be
perceived as a separate auditory event.

It appears that the “law of the first wavefront” can be interpreted as the “law of the
first localisation stimulus”…..”

“…..6. Summary
According to the association model presented in the preceding chapters, the
functioning of the auditory system with respect to spatial hearing is due to two
different processing mechanisms. Each of these two processing mechanisms manifests
itself in the form of an associatively guided pattern selection.

A current stimulus stemming from a sufficiently broadband sound source gives

rise to a location association in the first and to a gestalt association in the second,
higher-level processing stage because of auditory experience. Although the two
stages work independently of each other, they always determine the properties of
one or multiple simultaneous auditory events in a conjoint manner. The rigorous
differentiation of these two stimulus evaluation stages corresponds entirely to the
two elementary areas of auditory experience. The received ear signals can be
attributed to the two sound source characteristics of “location” and “signal”,
which are independent of each other but always occur in a pair-wise fashion.
Therefore, the presented association model is in agreement with many phenomena
related to localisation in the superimposed sound field……”

12. Confirmation of a need to process timing information:

Source: http://arxiv.org/pdf/1208.4611v2.pdf

Gave the following summary:

"..The time-frequency uncertainty principle states that the product of the temporal and
frequency extents of a signal cannot be smaller than 1/(4PI). We study human ability
to simultaneously judge the frequency and the timing of a sound. Our subjects often
exceeded the uncertainty limit, sometimes by more than tenfold, mostly through
remarkable timing acuity. Our results establish a lower bound for the nonlinearity
and complexity of the algorithms employed by our brains in parsing transient
sounds, rule out simple "linear filter" models of early auditory processing, and
highlight timing acuity as a central feature in auditory object processing…."

And further:

"…In many applications such as speech recognition or audio compression (e.g. MP3
[18]), the first computational stage consists of generating from the source sound
sonogram snippets, which become the input to latter stages. Our data suggest this is
not a faithful description of early steps in auditory transduction and processing,
which appear to preserve much more accurate information about the timing and
phase of sound components [12, 19, 20] than about their intensity…."

And finally:

"…Early last century a number of auditory phenomena, such as residue pitch and
missing fundamentals, started to indicate that the traditional view of the hearing
process as a form of spectral analysis had to be revised. In 1951, Licklider [25] set the
foundation for the temporal theories of pitch perception, in which the detailed pattern
of action potentials in the auditory nerve is used [26, 28], as opposed to spectral or
place theories, in which the overall amplitude of the activity pattern is evaluated
without detailed access to phase information. The groundbreaking work of Ronken
[22] and Moore [23] found violations of uncertainty-like products and argued for
them to be evidence in favour of temporal models. However this line of work was
hampered fourfold, by lack of the formal foundation in time-frequency distributions
we have today, by concentrating on frequency discrimination alone, by technical
difficulties in the generation of the stimuli, and not the least by lack of understanding
of cochlear dynamics, since the active cochlear processes had not yet been discovered.
Perhaps because of these reasons this groundbreaking work did not percolate
into the community at large, and as a result most sound analysis and processing
tools today continue to use models based on spectral theories. We believe it is
time to revisit this issue….."

13. Transient and localization

Some very interesting information on transients and localization comes from the
development work of Joseph Manger. The whole paper is recommended for reading.

Source: http://www.manger-audio.co.uk/PDFs/acoustical_reality.pdf
How distorted transients can be – Manger illustrates it on the following pictures:
14. AES Technical Document on phase accuracy and transient fidelity.

In 2002, the AESTD1001.1.01-10 drove the stake in the ground, and pegged the 10usec
as the maximum allowed timing difference between stereo loudspeakers across the entire
audio band.

Source: http://www.aes.org/technical/documents/AESTD1001.pdf

Some comments are presented in

http://www.bodziosoftware.com.au/AES_Document_Comments.pdf
15. Audibility of transients

I have come across an interesting paper in JAES ,Vol.38, No.11,1990 November,

"On the Correlation between the Subjective Evaluation of Sound and the Objective
Evaluation of Acoustic Parameters for a Selected Source".

The authors performed subjective and objective analysis of several woofers using
impulsive tones, and concluded:

"…A detailed analysis of the results of the subjective evaluation of

loudspeakers showed that the subjective evaluation of the obtained sounds was
decisively influenced by the work of the loudspeaker in a transient state. It
appeared that the longer the duration of final transient and the smaller the
value of coefficient D, the greater the sharpness of the sounds emitted by the
loudspeaker…."

16. Transients and Localization

The following paper, clearly indicates, that transients are critical in localization
process.

Source: http://www.pa.msu.edu/acoustics/rooms1.pdf

I have located an interesting piece of information in the paper "Localization

of sound in rooms", from JASocAm. 74 (5) Nov 1983. The paper is by WM Hartman
from Michigan State University, Dept. of Physics, and provides the following
summary:

"…This paper is concerned with the localization of sources of sounds by human

listeners in rooms. It presents the results of source-identification experiments
designed to determine whether the ability to localize sound in a room depends
upon the room acoustics, and how it depends upon the nature of the source
signal.

The experiments indicate that the localization of impulsive sounds, with

strong attack transients, is independent of the room reverberation time, though
it may depend upon the room geometry.

For sounds without attack transients, localization improves monotonically with

the spectral density of the source.

Localization of continuous broadband noise does depend upon room

reverberation
time….."
More papers by Hartmann and Rakerd.

“Localization of sound in rooms, II: The effects of a single reflecting surface”

http://www.pa.msu.edu/acoustics/rooms2.pdf

“….Our results indicate the following: (1) A sound must include transients if the
precedence effect is to operate as an aid to its localization in rooms. (2) Even if
transients are present the precedence effect does not eliminate all influences of room
reflections. (3) Due to the interference of reflections large interaural intensity
differences may occur in a room and these have a considerable influence on
localization; this is true even at low frequencies for which IID cues do not exist in a
free field. (4) Listeners appear to have certain expectations about the reliability and
plausibility of various directional cues and perceptually weight the cues accordingly;
we suggest that this may explain, in part, the large variation in time-intensity trading
ratios reported in the literature and also the differing reports regarding the importance
of onsets for localization. (5) In this study we find that onset cues are of some
importance to localization even in free field.

“Localization of sound in rooms: III: Onset and duration effects”

http://www.pa.msu.edu/acoustics/rooms3.pdf

Conclusions

“…(1) A rapid onset facilitates localization in a free field by a measurable but

small amount, about 0.5deg. It facilitates localization in rooms by substantially
larger amount because the onset allows the precedence effect to operate and
without the precedence effect localization is poor due to misdirection cues in
steady-state sound field.

(2) The precedence effect is maximally effective when the signal onset is
instantaneous. Its effectiveness begins to diminish as the onset duration is
increased…..”

17. More on localization and transients

Source: http://www.pa.msu.edu/acoustics/rakhar2.pdf

In a paper by Brad Rakerd and William M. Hartmann “Localization of noise in a

reverberant environment” (Michigan State University), they conclude:

“…(1) Localization of noise is enhanced by an attack transient. An attack transient

appears to be particularly helpful when the direct-reverberant ratio is low. Attack
transients give an advantage over slow onsets when the reflections are not much
delayed re the direct sound. By contrast, attack transients are of only marginal value
when noise is presented by headphones or tones are presented in an anechoic room
(Tobias and Schubert, 1959; Rakerd and Hartmann, 1986).

(2) Onsets are a great leveler among individuals. Whereas the ability to localize
steady steady-state sounds varies greatly among listeners, the ability to localize
sounds with an onset transient shows best to worst differences less than 1.5 degrees
among our seven listeners….”

18. Even more on localization and transients

Source: AES library. Preprint 2745. 86th Convention.

“Localization of sound in a room with reflecting walls” W.M. Wagenaars

“….3. CONCLUSIONS

In this study localization of sound in a room with reflecting walls was tested. Eleven
stimuli were used, differing in spectral and temporal information. For such a room the
following can be concluded:

- Signal bandwidth is an important cue for localization. The broader the frequency
spectrum of a sound, the better localization performance.

- Offsets seem to be an equally important cue for localization as onsets.

Localization performance are similar for signals with an abrupt onset, offset, or
both.

- Localization performance for steady state sinusoids is frequency-dependent. For

simply gated sinusoids performance is not dependent of frequency.

Although many of the errors made were distance errors, subjects are able to localize
distance quite well. Furthermore subjects usually select the correct side, even for the
hard to localize steady state sinusoids…..”

19. Sound Quality and Transient Response.

In the next paper: “Correlation of Transient Measurements on Loudspeakers with

Listening Tests” by M. Corrington, published in JAES, JANUARY 1955, VOLUME 3,
NUMBER 1, we find an interesting measurement method, allowing for separation of the
“overhang transient” – see below
The paper reads well, and has the following interesting conclusion:

“….This information supplements the steady-state sound pressure measurements. We

have never found any system with low transient distortion that did not also have a smooth
sound-pressure curve; on the other hand, we have measured systems with fairly sharp and
small peaks in the sound-pressure response that produced objectionable transient
distortion.

There is very good correlation between transient distortion and subjective listening
tests. Whenever there are peaks in the transient distortion, one can be sure that the
listening tests will reveal unpleasant distortion, even though the sound-pressure
curve is quite smooth….

Extensive measurements show that for a high-quality audio system the sound-pressure
curve must be smooth and properly shaped, and that the transient distortion should be
down at least 18dB throughout the range. One can then be fairly certain that the system
will pass very careful listening tests….”
20. Confirmation of two-stage processing by the ear, as discussed in (3) and (11).

Yet another interesting paper. It puts the early reflections in somewhat different
perspective.

“The Significance of Early High-Frequency Reflections from Loudspeakers in

Listening Rooms”, Preprint 4094, David Moulton, David Moulton Professional
Services, Groton, MA

“…Any reverberant space yields comb-filtering effects, and virtually all listening to
music via loudspeakers is done in such spaces. Therefore, logically speaking, all
listening is done under compromised conditions, where a primary attribute of accurate
sound reproduction (fiat amplitude response) is negated. Yet we must acknowledge
that music playback systems seem to work well: listeners enjoy listening, they readily
and accurately identify sounds (and will testify to their realism), and some listeners
are able to detect truly microscopic differences between alternate components in the
playback system.

This anomaly raises the question: how can individuals listen effectively to
loudspeakers in reverberant spaces and why don't the ubiquitous comb-filtering
interference effects always pose problems for the listener?

I suggest that the answer lies in the nature of our auditory localization
capability, which makes use of interference effects such as comb-filtering as a
function of performing the sound source localization task.

That task is performed at a pre-conscious neurological stage and most early

reflections are localization information that is not presented to the conscious
mind. Further, we do not consciously perceive the amplitude response
characteristics of comb-filtering effects that occur in reverberant spaces as a
result of early reflections, even though such effects are clearly measurable….”

The above statement confirms earlier findings of Gunter Theile, Watkinson and
Manger about the ear processing the incoming audio stimulus in two stages: The
received ear signals can be attributed to the two sound source characteristics of
“location” and “signal”, which are independent of each other but always occur in a
pair-wise fashion.
20. General Conclusions From Papers Presented Above

First of all – the room itself.

Accordingly to Bernd Theiss, Malcolm O. J. Hawksford in AES Preprint 4462:

“…Early reflections < 2.5 ms.

Early reflections occurring less than 2.5 ms after the original sound sensation are
known to shift the image towards their direction and to blur the image.

Early reflections < 5 ms.

Early reflections occurring more than 2.5 ms but less than 5 ms after the original
sound sensation are known to blur the image, although they keep the direction of the
image constant….”.

So if your goal is to deliver the sharpest image, or most accurate localization, you
would be well advised to take care of transient origination (loudspeakers) and also
provide some acoustical treatment to the walls/room.

There are basically three areas where linear-phase loudspeakers differ from minimum-
phase loudspeakers.

1. Linear-phase speakers provide more accurate spatial information, rather than

timbral. Tonal balance is the same for both loudspeaker types. This is where
the tests are falling apart, because listeners are looking for tonal differences,
rather than subtle spatial clues – sharper image, better located soloists, stage
depth. It’s subtle, but it is there.

2. Identical phase response for all loudspeaker in the system. The phase response
in correctly equalized multi-channel linear-phase system is 0deg in every
loudspeaker. Therefore it immediately satisfies AESTD1001.1.01-10 for phase
accuracy and transient fidelity to perfection. The measurements of linear-
phase loudspeaker are presented on my website, and comments on
AESTD1001.1.01-10 are presented in -
http://www.bodziosoftware.com.au/AES_Document_Comments.pdf

3. Tighter bass. Even Dr. Floyd Toole quoted other researchers (Craven and
Gerzon) on this subject on page 420. The most obvious difference is the
tighter bass. I have conducted extensive tests on this subject -
http://www.bodziosoftware.com.au/LP_MP_Subwoofer_Tests.pdf
21. Square Wave Loudspeaker Testing

Another interesting paper from 94'th AES Convention. I would recommend reading
the entire paper.

Source: "Directions for Qualified Loudspeaker Evaluations", AES Preprint 3603,

Peter M. Pfleiderer, 1993.

The paper concludes with the following summary:

”…Summary

An almost unbelievable state of perfection has been reached for electronic

components within the electroacoustical reproduction chain due to competent
applications of measurement technology. With loudspeakers, on the other hand,
competent measurement methods are currently not even in practical use.
Obviously, test methods are required which are capable of uncovering major
changes to signal waveforms originating from linear and acoustical errors.

Measurements with square wave signals should be included as standard testing

procedures in order to be able to detect errors with sound quality and spatial
imaging in all HiFi components, but especially in loudspeaker systems. Many
technical and acoustical faults can namely not be registered with SPL or
frequency measurements, although they have induced significant irregularities
into the relevant audio signal waveform.

This is the reason why loudspeakers of proven square wave response capability
are an important prerequisite for the natural reproduction of sound. Moreover,
it is only possible to detect acoustic faults with this type of technically
faultless reference loudspeaker. It should be clearly noted that all other
current components in the electroacoustical reproduction chain already transmit
square wave signals correctly.

Correct square wave reproduction with loudspeakers has the same importance as
was the case for correct square wave reproduction with amplifiers in the 1960's.
Both constitute fundamental advances and establish important conditions for high
quality reproduction of music. Nothing can propagate the concept of high
fidelity more than these types of advances….”
 Time Domain Instrument Testing

Real-life loudspeaker example

The system under test discussed here consists of a filter and a loudspeaker in an
enclosure. These two components that will introduce time delay are the filter and the
combination of driver and the enclosure itself. To illustrate the above, a 12” guitar
loudspeaker in a vented box was measured and it’s minimum-phase responses were
obtained with a help of an MLS measurement technique – see below. It is immediately
observable, that the loudspeaker has rather irregular frequency response. Since the
loudspeaker is essentially a minimum-phase device, the corresponding phase response is
also highly irregular, and definitely not flat.

Let’s establish the frequency response of interest, which is the frequency range where the
SPL will be equalized to flat response. In my example it will be: 90Hz – 5500Hz.

A 300Hz square wave reproduced by this loudspeaker is highly distorted. Strong ringing
is due to 10dB sharp SPL peak located at 3.5kHz. You can see, that there are about
11periods of ringing waveform in one period of 300Hz square wave.
 Instrument test results obtained from linear-phase loudspeakers reveal their
true superiority in time domain. The following test results were obtained by John
Kreskovsky of Music and Design ( http://www.musicanddesign.com )

As John points out: “….The measurements were not taken in an anechoic

environment and are of the continuous time type, recorded over numerous cycles,
windowing over a reflection free period can not be performed. Thus, there is some
contamination by room reflections resulting is some degradation in the observed
response.

The first figure shows the 300 Hz response. This is close to the low frequency cut off
of the system where the phase rotation and group delay due to the 200 Hz high pass
cut off would normally result in loss of flat top behaviour and the 2k Hz crossover
would cause distortion of the initial rise. This is shown in the insert at the upper right
of the plot for the linearized system and confirmed by the lower plot which if for the
standard LR4 system. The white trace is the input, orange the acoustic output from the
speaker system.

300 Hz square response of linearized system, left, and standard LR4 crossover, right.

500 Hz square response of linearized system, left, and standard LR4 crossover, right.
1kHz square response of linearized system, left, and standard LR4 crossover, right.

2kHz square response of linearized system, left, and standard LR4 crossover, right.

….” End of quote.

My own measurements on 18” McCauley subwoofers further confirm time

domain superiority of linear-phase loudspeakers.

20Hz square wave: Linear-Phase Mode and Minimum-Phase Mode

Shown above, the time-domain comparison measurement results speak for

themselves. It needs to be remembered, that we are dealing here with a very heavy-
coned, 18” driver, low-pass filtered, in a vented (resonating) enclosure, and yet, the
time domain performance is near-perfect accurate. It’s pretty amazing to see a vented
loudspeaker, holding the acoustic pressure nearly constant for 25ms.

Next, I used 2ms-wide pulses separated by 350ms space as the source signal.
On the 2ms pulse, the minimum-phase version delivered a more of a “thump” instead
of a pop or a click. This is perhaps not surprising, as the post-ringing of the pulse
extended to130ms and far exceeded the 30ms “memory effect” of the auditory
system. Here, the driver, filter and vented enclosure added it’s own, combined
signature. It is also observable, that the minimum-phase version of the subwoofer has
converted the clearly asymmetrical pulse into a much more symmetrical bi-polar
pulse with post-ringing. This is clearly visible on the screen shots below.

5ms Impulse in Linear-Phase Mode and Minimum-Phase Mode

When a 2ms bi-polar pulse was used for excitation, the minimum-phase
version has done the opposite, and converted the symmetrical bi-polar pulse into a
pulse with clear asymmetrical tendency. The ringing past the pulse is due to a more
distant microphone placement, so now, the mike picks some of the room reflections.

2ms Bi-polar pulse in Linear-Phase Mode and Minimum-Phase Mode

When a 10ms bi-polar pulse was used for excitation, the minimum-phase
version has even more asymmetrical tendency.
 10ms Bi-polar pulse in Linear-Phase Mode and Minimum-Phase Mode

Finally some more square wave measurements from UE User’s Manual.

The linear phase result is on the left and the nonlinear phase result on the right.
It should be noted that there is some distortion in the wave forms that that must be
attributed to room reflections. Square wave testing is a steady state test and without a
true anechoic chamber the effects of room reflections can not be eliminated.

Never the less, for the 300 Hz case shown in the first figure, the linear phase
system shows the sharp rise and fairly flat top expected. The nonlinear phase case
shows early tweeter response followed by the woofer response and the sloped top is
an artefact of the nonlinear phase. The response also significantly overshoots the
correct level. This latter effect is seldom discussed when comparisons of linear and
nonlinear phase systems are made. Even though the amplitude of reach frequency
component is correctly reproduced in the nonlinear phase system, the lack of linear
phase means that the different frequency components do not sum correctly since that
are delayed by different amounts. The overshoot is a result of time distortion.

300 Hz square wave response, Linear phase, left; Nonlinear phase, right

The next figure shows the same comparison for a 1kHz square wave. Again,
some distortion is observed due to room reflections. However, the linear phase case
again shows the expected sharp rise and relatively flat top. The nonlinear phase
system more clearly shows the time lag between the woofer and tweeter response.

1kHz square wave response, Linear phase, left; Nonlinear phase, right.

The next figure shows the result for a 3k Hz square wave. The differences
between linear and nonlinear phase, while clearly evident, are less significant because
the fundamental is above the crossover point and there is little contribution from the
woofer due to the 4th order low pass response. With the system designed another
interesting feature of the linear phase system can be examined, the effect of crossover
slope.

3kHz square wave response, Linear phase, left; Nonlinear phase, right.

The next figure shows the 1kHz response of the linear phase and nonlinear
phase system when the slope of the woofer to tweeter crossover is increased to 8th
order, 48dB/octave. With the Ultimate Equalizer this is easily accomplished by
selecting the new 48dB/octave slopes and clicking Show complete system to calculate
and load the new filters.
 1kHz response of linear and nonlinear phase system with 8th order crossover.

This result should be compared to that of figure where the crossover was 4th
order. Changing the order has no effect on the linear phase system at the design point.
The nonlinear phase system response is significantly different solely due to the
change in crossover order.

Finally, the last figure shows the effect of reducing the crossover to 2nd order.
The response of the nonlinear phase system looks somewhat better now. However, for
flat response the tweeter must be connected with inverted polarity in the nonlinear
phase system and the initial tweeter pulse is therefore in the wrong direction. It should
be noted that many audio enthusiasts feel the 2nd order crossover sound better than
those of higher order. This may be a result of the improved wave form observed here
and could be an indication of the potential of linear crossover and speakers of any
order since they will all preserve wave form relative to the design point.

1kHz response of linear and nonlinear phase system with 2nd order crossover.
Conclusions

At the time of this writing, linear-phase loudspeakers are still a new “kid on
the block”. Past attempts in creating them resulted in offerings that were simply too
expensive for wide-spread use. The most accurate implementation of linear-phase
loudspeaker requires a full set of individual driver measurements, coupled with a DSP
approach, in addition to an active amplification system. This really makes the linear-
phase system highly customized device – a world of difference in comparison to the
current approach of loudspeaker industry.

However, this particular feature makes the linear-phase system an ideal DIY
device. In our world, everything is custom-built, with an aim to typically outperform
comparable commercial designs. Linear-phase loudspeakers offer everything that
minimum-phase loudspeakers can offer, and then reward you with often vastly
superior performance in time domain, as explained in the pages above.

It appears, that my poor and outdated listening/evaluating habits, coupled with

lack of standard listening methodology for time/space-domain assessment of
loudspeakers conspired to cloud my ability to really critically listen to the full set of
my loudspeakers during some of my evaluation tests. Secondly, not every musical
material will reveal all time-domain characteristics to the same degree. For instance,
tight, well-defined bass, will manifest itself on gunshots and explosions in DVD
movies, but will not stand out during low-frequency, seismic earthquake effects on
LFE channel. In more critical tests, I did pick the “tighter bass” characteristic, as it
was too obvious to miss on the large, 18” subs. Also, I pointed out earlier the effect of
feeling closer to the orchestra, as if I could better discriminate their sitting
arrangement. Both of these effects have really nothing to do with frequency domain –
they are both more of the time/space domain phenomena.

It is clear, that designing loudspeakers using frequency-domain characteristics

as the main (or only) criteria leads to stagnated, oversimplified, and ultimately
inaccurate system. If I continued to design loudspeakers that never reveal time-
domain or spatial-domain subtleties, I would never even know of the existence of
such subtleties, therefore, I would never be motivated to change – thus allowing the
vicious cycle to continue. It is evident, that the ear examines the incoming audio
stimulus in two-stage process: (1) location – here the transient of the stimulus is
examined, and (2) signal – here the spectral properties of the stimulus are examined.
The two processes always work in-tandem. It is therefore essential, that the
loudspeaker provides undistorted waveforms to the auditory system to enable correct
processing of both stages.

So, here I am. Struggling to come out of the “frequency-domain box” and into
the new world of time/frequency/space-domain characteristics of contemporary
loudspeakers. But even at these early stages of adopting a new technology, I find it
already very rewarding. This is because it’s evident that a new, accurate and
realistic acoustic transduction technology is being achieved in much more
accessible commercial way.

Thank you for reading.

Bohdan
Appendix A

Source: http://en.wikipedia.org/wiki/Sound_localization

Lateral information (left, ahead, right)

For determining the lateral input direction (left, front, right) the auditory system
analyzes the following ear signal information:

• Interaural time differences

Sound from the right side reaches the right ear earlier than the left ear. The
auditory system evaluates interaural time differences from
o Phase delays at low frequencies
o group delays at high frequencies
• Interaural level differences
Sound from the right side has a higher level at the right ear than at the left ear,
because the head shadows the left ear. These level differences are highly
frequency dependent and they increase with increasing frequency.

For frequencies below 800 Hz, mainly interaural time differences are evaluated (phase
delays), for frequencies above 1600 Hz mainly interaural level differences are
evaluated. Between 800 Hz and 1600 Hz there is a transition zone, where both
mechanisms play a role.

Localization accuracy is 1 degree for sources in front of the listener and 15

degrees for sources to the sides. Humans can discern interaural time differences
of 10 microseconds or less.[5][6]