|
Audio Measurement Basics
|
|
|
It is virtually impossible to get the most out of your
sound system without a way to objectively measure its perfomance.
It used to be that the knowledge and expensive hardware
required to make such measurements were out of the reach
of most home audio enthusiasts.
|
Now, with ETF, these sophisticated measurements can be
taken by anyone with a computer! Many audiophiles are finding
that a few hours of acoustic measurement result in greater
improvement sound quality than could be obtained by spending
thousands of dollars upgrading their equipment.
|
| Measurement
Fundamentals |
|
The first, most important measurement of a given speakers
performance in a given room is that of impulse response,
also referred to as transient response in some books and
magazines.
"Impulse" refers to a very short, very loud sound, such
as a gunshot. The ideal impulse requires an infinite amount
of energy delivered to a system over a time period approaching
0 seconds. In the real world, of course, this is not possible,
so an impulse is approximated by a very short time duration
high amplitude sound.
|
"Response" refers to the way the impulse sounds in the
room. The response will depend both on the speaker producing
the impulse, and the sound qualities of the room itself.
In the case of loudspeaker measurement, the impulse can
take the form of a very high voltage placed across the loudspeaker
terminals for a very short period of time, resulting in
a loud "click" from the loudspeaker. The speaker/room response
to this impulse is recorded on a chart recorder with a microphone.
An example of an impulse response chart is shown below
|
| 
Fig 1 An example of an Impulse Response curve
|
|
The vertical axis of the chart shows the amplitude of the
sound waves. You can think of this axis measuring sound
volume (technically, this is not 100% true, but it's a good
enough description for our purposes). On the horizontal
axis is time, measured in Milliseconds (1/1000 of a second).
The initial "spike" on the graph, at 0 ms, is caused by
the sound of the impulse itself reaching the recording microphone.
The oscillations and spikes that follow are echoes as this
sound reverberates through the room and, to a lesser extent,
within the speaker itself.
The problem is, impulse response measurements are very
difficult to interpret. In the case of loudspeakers and
rooms, it is difficult to differentiate "good sound" from
sound quality that may not be considered as good just by
looking at the graph. For this reason, other methods have
been developed to display this sound data in more informative
ways. It should be stressed that all the different types
of charts and graphs people have devised to look at sound
data are simply different ways of looking at the same information.
No one representation is inherently superior to another;
they simply tend to emphasise different aspects of the sound.
|
For example, the impulse response shown above shows us primarily
when sounds from the impulse arrive at the recording microphone.
It tells us nothing about the relative strengths of the
various frequencies in the sound.
For frequency information, we turn to what is far and away
the most popular way of presenting audio data: frequency
response, also known sometimes as the "steady state
response". The frequency response curve is generated by
applying a mathematical operation known as the Fourier Transform
to the impulse response. ETF does this for you automatically,
so you really have no need to understand Fourier transforms.
However, if you're interested in learning more, our list
of reference materials will point you in the right direction.
Frequency response is the undisputed king of audio measurement.
Without good frequency response, high fidelity sound is
not possible. Below you will see a frequency response chart
taken from the demonstration room project you will find
at the Acoustisoft support page. It shows the low frequency
response delivered by a particular subwoofer at 8 different
positions in a room.
|
Fig 2: Low frequency response for a subwoofer measured
in 8 positions around a room
|
| |
|
|
You will notice two obvious differences between this and
the impulse response graph. First, the horizontal axis now
represents frequency, not time. This representation tells
us nothing of when the sounds arrived. Instead, it adds
up the amount of sound energy of each frequency that arrived
at the microphone over a given period of time, known as
the gate time. Usually, this gate time is set long
enough to include the first big spike you see in the impulse
response (the sound coming directly from the speaker), but
short enough that all the subsequent spikes (the echoes
from around the room) are ignored. For this reason, the
frequency response as shown above is primarily the response
of the speaker, with room effects ignored. This is useful
in that it allows comparisons of different speakers independent
of the rooms they happen to be measured in. < /p>
So what do we look for in a frequency response curve?
Well, ideally, we want the curve that is the flattest (we
want the speaker deliver equal amounts of sound energy at
all frequencies), and the highest (delivering the most sound
energy).
|
In this case, the flattest curve seems to be number 5,
while the highest is number 1. We could decide to accept
speaker position 5, as it demonstrates the best natural
response, or start playing around with an equalizer, turning
various frequencies up or down to try to "flatten" curve
1. In either case, our end goal is to have the flattest
frequency response possible, as that will deliver the best
sound.
The other thing you will notice is that frequency response
doesn't oscillate back and forth from positive to negative
the way the impulse response does. It is in the nature of
waves to oscillate. As a more direct measurement of the
sound waves, this oscillation is seen in the impulse response.
As already mentioned, however, the frequency response sums
the amount of energy of each frequency that arrives at the
microphone over a given time interval, rather than displaying
the state of the sound wave at a given time. For this reason,
no oscillation is observed.
Many examples of impulse response and frequency response
measurements are included in various places in ETF literature.
|
|
|
Fig 3: The steps in converting an impulse response to
a frequency response
|
|
|
|
Earlier, we stated that placing a high voltage across the
speakers' terminals for a short period of time could approximate
an impulse. Unfortunately, this is easier said than done.
The practical problems in applying a short duration high
voltage signal to a loudspeaker are that it is inconvenient
and requires specialize (and possibly dangerous) hardware.
It is also necessary to take several such measurements and
average them to compensate for background noise. Fortunately,
the science of audio measurement has advanced to the point
that test signals such as this are no longer required.
Over the years, different types of test signal have come
into popularity, such as pink noise excitation, time delay
spectroscopy, and periodic impulse excitation among others.
Each in turn has made improvements measurement quality and
ease of use.
ETF employs he current "state of the art" measurement technique,
called Maximum Length Sequence, or MLS.
|
This is the test signal type of choice for many instrumentation
manufacturers because of its relative ease of implementation,
low cost of required hardware, and its high immunity to
measurement artefacts such as noise or distortion. It also
provides more actual information about system response than
any previous measurement technique.
An MLS measurement begins with several seconds of white
noise (that is, noise of equal energy at all audible frequencies).
The response to this signal is recorded, and converted it
to an impulse response by means of a bit of mathematical
magic known as the Hadamard transform. The MLS method is
very convenient in that the MLS signal can be accurately
applied with very low cost hardware such as in a low cost
computer "game" type sound card, or even single bit devices
in specialized hardware.
MLS makes a very accurate recording of speaker / room
impulse response. This allows the response to any form of
room excitation to be determined.
|
|
|
|
MLS signals are a form of "white noise". White noise contains
a signal level that is identical for all frequencies just
as the colour white contains all other colours in the visible
spectrum. The fact that the MLS test signal contains energy
at all of the possible frequencies of interest makes it
possible to calculate an equivalent impulse response of
a measured system.
A test signal such as this can damage loudspeaker drivers
if not used carefully. Excessive high frequency energy can
damage high frequency drivers. This is the reason for the
existence of pink noise. Pink noise has decreasing energy
levels with as frequency is increased, making it possible
to play the signal louder through the audio system before
damage to high frequency units occurs.
|
This disadvantage of MLS is easily compensated for when
using a two-channel measurement system such as ETF. ETF
measures a system response through one channel and measures
only the sound card response through the other channel.
This allows ETF to have knowledge of the sound card response
and therefore correct the measurement for sound card anomalies.
This also means that sound card tone controls may be adjusted
to boost low frequencies and attenuate high frequencies.
This technique is known as spectral shaping. The test signal
may be shaped with an equalizer or tone control to emphasise
or de-emphasise any frequency region. The advantages of
the other system types are therefore available in ETF through
this technique.
|
|
|
