Handbook for Sound Engineers

Test and Measurement 1621

decayed prior to reaching the right side of the

screen. It should also be repeatable. Run the test

several times to check for consistency. Background

noise can dramatically affect the repeatability of the

measurement and the validity of the data.

Once the impulse response is acquired, it can be
further analyzed for spectral content, intelligibility infor-
mation, decay time, etc. These are referred to as metrics,
and some require some knowledge on the part of the
measurer in properly placing markers (called cursors) to
identify the parameters required to perform the calcula-
tions. Let us look at how the response of the loudspeaker
might be extracted from the data just gathered.
The time domain data displays what would have
resulted if an impulse were fed through the system.
Don’t try to correlate what you see on the analyzer with
what you heard during the test. Most measurement
systems display an impulse response that is calculated
from a knowledge of the input and output signal to the
system, and there is no resemblance between what you
hear when the test is run and what you are seeing on the
screen, Fig. 46-18.

We can usually assume that the first energy arrival is
from the loudspeaker itself, since any reflection would
have to arrive later than the first wave front since it had
to travel farther. Pre-arrivals can be caused by the
acoustic wave propagating through a solid object, such
as a ceiling or floor and reradiating near the microphone.
Such arrivals are very rare and usually quite low in level.
In some cases a reflection may actually be louder than
the direct arrival. This could be due to loudspeaker
design or its placement relative to the mic location. It’s
up to the measurer to determine if this is normal for a
given loudspeaker position/seating position. All loud-
speakers will have some internal and external reflections

that will arrive just after the first wave front. These are

actually a part of the loudspeaker’s response and can’t

be separated from the first wave front with a time

window due to their close proximity without extreme

compromises in frequency resolution. Such reflections

are at least partially responsible for the characteristic

sound of a loudspeaker. Studio monitor designers and

studio control room designers go to great lengths to

reduce the level of such reflections, yielding more accu-

rate sound reproduction. Good system design practice is

to place loudspeakers as far as possible from boundaries

(at least at mid- and high frequencies). This will produce

an initial time gap between the loudspeaker’s response

and the first reflections from the room. This gap is a

good initial dividing point between the loudspeaker’s

response and the room’s response, with the energy to the

left of the dividing cursor being the response of the loud-

speaker and the energy to the right the response of the

room. The placement of this divider can form a time

window by having the analyzer ignore everything later

in time than the cursor setting. The time window size

also determines the frequency resolution of the post-

processed data. In the frequency domain, improved reso-

lution means a smaller number. For instance, 10 Hz

resolution is better than 40 Hz resolution. Since time and

frequency have an inverse relationship, the time window

length required to observe 10 Hz will be much longer

than the time window length required to resolve 40 Hz.

The resolution can be estimated by f=1/T, where T is

the length of the time window in seconds. Since a

frequency magnitude plot is made up of a number of

data points connected by a line, another way to view the

frequency resolution is that it is the number of Hz

between the data points in a frequency domain display.

The method of determination of the time window

length varies with different analyzers. Some allow a

cursor to be placed anywhere on the data record, and the

placement determines the frequency resolution of the

spectrum determined by the window length. Others

require that the measurer select the number of samples to

be used to form the time window, which in turn deter-

mines the frequency resolution of the time window. The

window can then be positioned at different places on the

time domain plot to observe the spectral content of the

energy within the window, Figs. 46-19, 46-20, and 46-21.

For instance, a 1 second total time (44,100 samples)

could be divided into about twenty two time windows

of 2048 samples each (about 45 ms). Each window

would allow the observation of the spectral content

down to ( ) × 1000 or 22 Hz. The windows can be

overlapped and moved around to allow more precise

selection of the time span to be observed. Displaying a

Figure 46-18. Many analyzers acquire the room response
by digital sampling.

Amplitude

Time

# samples = Sample rate * time span

# samples determines frequency resolution

A/D resolution (# bits)

Impulse response

(^1) » 45