Tech Topic: What You See Is Not What You Get
Why measured response doesn’t always match what’s
heard
By
Pat Brown
|
|
 |
Many audio field technicians are now in possession of measurement
systems that can be used to assist the listening process in equalizing
sound reinforcement systems. But, they’re often surprised to find
that the measured system response correlates poorly with subjective
impression of how the system sounds. |
lLet's look at why the eye and ear do not always agree on what is best
regarding the response of the sound system.
First, consider the most popular methods of measuring the response of
the sound system. By “response,” I am referring to the magnitude of the
frequency response as displayed on a dB (vertical) vs. logarithmic (horizontal)
scale. The goal of technical system equalization is to produce a “flat”
horizontal line on this display.
|
|
|
WORKING IN REAL TIME The original RTAs used analog meters, but current versions use
a vertical row of LEDs for each 1/n-octave band. One-third octave
resolution is the most popular, and correlates well with the response
of the human auditory system. RTAs can also be software-based, utilizing the sound card on a personal computer to provide the A/D conversion of the microphone output voltage. |
A mathematical algorithm (the FFT) is used to produce the previously described dB vs. frequency display. These “digital” analyzers emulate their analog counterparts in how the information is displayed, but differ in that the filters and display is the product of a computer algorithm rather than analog filters. This type of RTA is more versatile, as the octave-fractions, colors, etc. are under software control.
|
|
|
Regardless of which type is used, the standard method-of-use is to drive the sound system with pink noise (equal energy per 1/n octave) and adjust the system equalizer for a “flat” magnitude response on the analyzer display. RTAs are powerful tools when certain guidelines are followed, but indoors they can indicate a system response with poor correlation to what the listener is hearing. The major consideration is the placement of the measurement microphone. |
If the mic is placed in the near field of the loudspeaker (typically
less than 10 feet), the correlation with human hearing is pretty good.
At this position, the direct energy from the loudspeaker dominates what
is being observed on the analyzer and very little of the reflected energy
from the room is included in the displayed response. Adjustment of the
equalizer for a flat direct sound field on the analyzer produces a desirable
result.
The down side to the near-field placement is that the measured response
is very sensitive to small vertical movements of the microphone when the
loudspeaker has offset vertical components (as most do). This sensitivity
can be reduced if the microphone is moved to a greater distance from the
loudspeaker (into the far field) since the path-length difference back
to the individual components becomes more equal. But, as the microphone
is moved further away, the reflected energy from the room begins to dominate
the displayed response.
GIVING EQUAL WEIGHT
Microphones have no “perceptual” abilities. They do not localize sound
or discriminate early sound energy from late energy like humans do. A
listener at a distance remote from the loudspeaker will pay more attention
to the direct field of the loudspeaker than sound that is building-up
in the room. A microphone gives equal weight to all energy without regard
to where it is coming from.
A simple experiment to verify this is to stand at the microphone position
and listen to the loudspeaker and then route the mic through a headphone
amplifier and listen to it through headphones - not the same thing at
all.
|
|
|
Low frequency sounds tend to linger in rooms longer than high frequency sounds, because most rooms have more high frequency absorption than low frequency absorption. As such, the room becomes “bass heavy” when the total sound field is considered. This extra low frequency information will dominate what is observed on the RTA, and the knee-jerk reaction is to attempt to “flatten” the response by boosting the high frequency bands on the equalizer. The result is a system with excessive high frequency output and a resultant “harsh” sound quality. When RTAs are used in this manner, it is important to equalize to a “target curve” rather than for a flat frequency response. |
The popular “X” curve for theaters is flat to 2 kHz, where it starts rolling
off the high frequency response at about 3 dB per octave. It is -10 dB
at 10 kHz relative to 2 kHz. This represents 1/10th power at 10 kHz relative
to flat response. The one-third-octave analyzer and the target curve have
served sound practitioners well for years, and remains a viable approach
to system calibration.
RECENT METHODS
Technology has yielded some new methods for acquiring the system response
at a listener position. A complex comparison (both time and frequency
information) of the input and output of a system is called the transfer
function. It includes both the magnitude and phase response of the loudspeaker/room
at the microphone position. This has become a popular method of analysis,
as it allows any input stimulus to be used to test the system, since the
displayed response is just the difference between “what you put in” and
“what you got out.”
|
|
|
Transfer function analysis has the added advantage of the ability
to use a “time window” to exclude late arriving energy from consideration
in the response. This can prevent the low-frequency build-up problem
that plagues traditional real-time analysis. With proper implementation
of a time window, the system response can be adjusted without the
need for frequency weighting via a target curve. The stimulus (the reference signal) always has a much shorter path back to analyzer input than the output of the measurement microphone. Sources of delay include the travel time through the air and the latency of digital processors. |
Failure to properly synchronize the reference signal and the microphone’s signal will result in an erroneous display of the system’s response. The length of the time window must also be selected - in other words, “how much of the room decay do I want to include in the response?”
|
|
|
Unfortunately, there is not an optimum size for the entire spectrum. A short time window excludes much of the room decay at the expense of low-frequency resolution. A long time window improves frequency resolution at the expense of gathering too much of the room’s decay. A compromise is required. The human auditory system perceives pitch on a proportional (logarithmic) frequency scale. This is one reason that we use constant-percentage bandwidth filters for tuning audio systems - the bandwidth grows with increasing frequency. |
Frequency-dependent bandwidth suggests that the length of the windowing
function used in transfer function analysis should be varied in the same
manner - a decreasing length with increasing frequency. This produces
a somewhat “anechoic” response at high frequencies with increasing frequency
resolution as frequency decreases. The time window length is a function
of frequency, with even the longest window (highest frequency resolution)
excluding much of the late energy from the room.
Another caveat of this type of analysis is that much greater frequency
detail is possible than with the typical 1/3-octave banded display. Phase
interference effects from reflections or multiple drivers are clearly
visible on the analyzer. Such anomalies are almost always position-dependent,
so careful “corrections” at one seating position will be inappropriate
for another. Both the loudspeaker and the measurement microphone should
be carefully positioned to avoid the creation of very early high-level
reflections.
SPECIAL EFFECTS?
The “floor bounce” effect is a common example of a very early reflection
(typically within a few milliseconds of the first sound arrival) that
produces a unique acoustic response for each listener seat for all but
the lowest octaves of the spectrum. This is an example of “less is better”
when measuring the response, as a 1/3-octave display lacks the resolution
to observe the effect in detail and produces less of a temptation to “fix”
it.
|
|
|
The floor bounce effect can be minimized by use of an appropriate
frequency-dependent time window or by simply laying the measurement
microphone on the floor, or on a board placed across the listener
seats. The effect usually disappears with the presence of an audience,
so we do not wish to consider it when tuning the sound system. |
Near-field techniques do not include air absorption for the simple fact
that the sound has not traveled very far before it is picked up by the
microphone, so it hasn’t passed though enough air to be significantly
attenuated.
By far, the biggest problem with tuning sound systems is failure on the
part of the technician to recognize anomalies that cannot be corrected
with equalization. The equalizer is a “global” device, meaning that its
response curve will be impressed on all of the sound radiated from the
loudspeaker, regardless of the direction in which it is radiated.
Many, if not most, of the anomalies observed on the analyzer are unique
to each listener position. The technician must learn to recognize and
ignore such events. They include:
• Floor-bounce effect;
• Interference between multiple drivers;
• Reflections from objects near the mic or loudspeaker.
Events that produce a more global effect, and can therefore be addressed with equalization include:
• Boundary-loading of loudspeakers;
• Coupling between multiple low-frequency drivers;
• The direct-field loudspeaker response.
With training and experience, the system technician can implement methods
that reveal system imperfections that are correctable, and hide those
that are not - regardless of the analysis method used. Better yet, system
designers can design systems with fewer “un-equalizable” problems.
BAD IS ALWAYS BAD
The old adage “an ounce of prevention...” could never be MORE true. System
equalization then becomes meaningful and fast, providing the “icing on
the cake” of the performance of a sound system. It makes a good loudspeaker
sound better, and brings the system to its fullest potential given the
acoustic environment into which it is placed. A bad room is a bad room,
regardless of how we process the electrical signal that drives the loudspeakers.
|
|
|
When used properly, the traditional 1/n-octave real-time analyzer
is a useful tool outdoors at any distance. Indoors, the effects
of reflected sound and non-frequency-uniform room absorption produce
some problems for this method at measurement distances remote from
the loudspeaker. |
Transfer function analysis addresses some of the shortcomings of the 1/n-octave
RTA, but it requires greater expertise on the part of the user. Failure
to properly compensate for the time differential between the reference
and measured signal can produce wildly erroneous results. The time window
length must also be selected by the user, and different lengths will produce
different displayed responses. A frequency-dependent time window produces
a display that correlates well with human perception.
The most important feature of either measurement method is a knowledgeable
operator - one who understands the caveats of each approach along with
the basic characteristics of the human auditory system. None of the questions
raised here have a single, correct answer. This means that experience,
good judgment, and common sense rooted in Newtonian physics are still
a part of the measurement process.
Sound is a relatively easy quantity to measure, but measurements that
correlate with human perception are much more difficult. Analyzers driven
by omni directional microphones do a poor job of emulating the human listener.
At this point one could ask, “So why measure at all? Why not just listen?”
Next issue, we’ll have a look at this provocative question.
Pat and Brenda Brown own and operate Syn-Aud-Con, conducting audio training sessions around the world. For more info, go to www.synaudcon.com
October 2003 Live Sound International