GOLDBERG AND MÄKIVIRTA
AUTOMATED IN-SITU EQUALISATION
AES 23RD CONFERENCE, May 23-25, 2003
10
In Figure 4 the results are pooled for all products and
for each product type, excluding the main monitors
where there were only three cases. The change in
quartile difference and RMS deviation for the broad-
band and the subbands is illustrated. For all models,
the broadband flatness is improved by 0.4 dB and the
mean reduction in the LF subband RMS deviation is
1.4 dB. The RMS deviation for all models pooled to-
gether has been reduced by equalisation and the larg-
est reduction is seen for the three-way systems. To
some extent this is similar for the quartile difference
but the small two-way and two-way systems do not
see such large improvements due to equalisation. This
indicates that the improvement is mainly a reduction
of extreme magnitude values (heights of peaks and
notches) in the low frequency response.
5. DISCUSSION
The objective of this paper is to present an automated
system for choosing appropriate room response con-
trol settings once an in-situ frequency response meas-
urement has been made and to show that it is effec-
tive.
The room response controls in active loudspeakers
implement discrete filter parameter values rather than
a continuous parameter value range. The number of
possible filter parameter value combinations can be
quite large and so even an experienced operator can
find it difficult to choose the optimal settings.
The task of the automated optimiser is to find the op-
timal combination from the possible combinations of
discrete filter parameter values. The cost of perform-
ing a brute force search of all value combinations and
then choosing the best among them is prohibitive in
terms of computer processing time. The approach cho-
sen is to exploit the heuristics of experienced calibra-
tion engineers and to reduce the number of alterna-
tives by dividing the task into subsections that can re-
liably be solved independently. A significant part of
the heuristics is the order in which these choices
should be taken. A considerable improvement in the
speed of optimisation was achieved relative to a full
exhaustive search.
The optimisation algorithm is relatively robust to a
wide variety of situations, such as varying room
acoustics, different sized loudspeakers with differing
anechoic responses and varying in-situ responses [41].
The optimisation is efficient and so the software is fast
enough to be used routinely at in-situ loudspeaker
calibrations.
A case study demonstrates the statistical changes due
to the optimisation algorithm’s recommended room
response control settings. The settings achieve im-
proved equalisation in the form of a smaller RMS de-
viation from the target response. The improvement is
not limited by the optimisation method but by the
room response controls which are not intended to cor-
rect for narrow-band deviations in the frequency re-
sponse. Examples of these are response variations re-
sulting from acoustic issues such as cancellations as-
sociated comb filtering due to reflections. These
should be solved acoustically rather than electroni-
cally.
The statistical analysis of 63 loudspeakers shows that
the automated equalisation is able to systematically
reduce the variability in the equalised responses and to
improve the frequency response flatness relative to the
target response. It achieves this by improving the
broadband frequency balance relative to the target re-
sponse and by reducing the variability in the response,
particularly in the low frequencies. Across all loud-
speaker groups the main improvement is in the reduc-
tion of extreme (outlier) values in the low frequency
band of the response.
It is interesting to note that the most commonly used
room response control was the midrange level, fol-
lowed closely by the treble level and bass tilt control.
This is explained by the fact that the algorithm in most
stages minimises the RMS deviation, and in so doing
affects most efficiently the extreme deviations from
the median level.
For all models pooled together, the broadband flatness
was improved by equalisation. This improvement is
mainly due to a reduction of the extreme magnitude
values (heights of peaks and notches) in the low fre-
quency response (LF subband).
The lack of improvement in the quartile values and
RMS deviation in the midrange and high frequencies
(MF and HF subbands) is because the room related
response variation becomes narrow band. Some im-
provement in the equalisation could be obtained with
room response controls offering a tilting or shaping of
the response within the mid-to-high frequency range.
The largest variability of the improvement in the low
frequency range can be explained by the acoustics
found in listening rooms [42]. At low frequencies the
radiation from the loudspeaker can be considered om-
nidirectional and the sound field in the room is usually
not very diffuse. This results in strong room effects
and hence large variations in the magnitude response
at these frequencies.
The largest improvement is seen for the three-way
systems and can be explained by two main factors.
Firstly, the rooms in which this type of loudspeaker is
typically installed are of a higher quality acoustical
design, so the sound field in them is well controlled.
Conversely, smaller loudspeakers are often installed in
rooms with little or no acoustical design, making cor-
