Comparative study of cone-shaped versus flat-panel speakers for
active noise control of multi-tonal signals in open windows
Woon Seng Gan
Digital Signal Processing Laboratory
School of Electrical and Electronic Engineering, Nanyang Technological University
S2-B4a-03, 50 Nanyang Avenue, 639798 Singapore
Openings on walls, such as windows or doors, facilitate the propagation of external noises to
interior spaces. Active noise control is a promising technique for the reduction of low
frequency noise propagating through such openings. To preserve the full functionality of
the opening, especially for natural ventilation, it is required to minimize the physical size of
the secondary sources. In this scenario, using compact speakers and limiting their distances
from the reference microphones are critical requirements. In this paper, we present a
comparative study on the noise reduction performance of an active noise control system
using compact cone-shaped and flat-panel speakers in identical configurations. The
performance was measured based on multi-tonal noise with frequency components below
1.2 kHz, using different setups of single reference sensor, single error sensor, and up to four
secondary sources. In evaluating the measurement data, including frequency responses,
harmonic distortion, radiation patterns, and noise reduction, we highlight key differences of
these two types of speakers and determine their respective application contexts.
With increasing knowledge and awareness of health risks associated with environmental
noise exposure, effective noise abatement is vital to safeguard public health. Based on a
WHO/European publication, noise pollution is ranked second only to air pollution as an
environmental stressor that adversely affects human health
. It has been estimated that almost a
quarter of the European Union population is daily exposed to potentially harmful noise levels
mostly due to traffic and concentrated in urban areas. The de facto noise mitigation methods are
passive structures that scatter and/or dampen acoustic waves (e.g., acoustic barriers). However,
passive techniques are ineffective at shielding high-rise urban dwellings and are not cost-effective
for land-scarce cities
Active means of noise control are primarily effective at low frequencies, typical of
, where passive methods are ineffective
. Active Noise Control (ANC)
relies on electro-mechanical components, such as loudspeakers, to generate an anti-noise wave
that reduces the “primary” noise, based on the superposition principle. ANC technology has seen
the most commercial success in ANC headphones, and it has also been applied in automotive,
home appliances, industrial plants and transportation
. However, in these examples, the
deployment was often limited to air ducts and exhaust pipes. Recent advances of ANC include
applications in motorcycle helmets, medical Magnetic Resonance Imaging (MRI) machines,
infant incubators, and noise control with high-directional loudspeakers
By leveraging on conditions that ANC has been known to be most effective at, the problem
of noise pollution control can be narrowed down to abating noise through wall openings. This has
been focused on windows that can be classified into two categories: closed-window ANC, and
open-window ANC. Closed-window ANC targets window fixtures and aims to reduce low
frequency sounds that double- or triple- glazed windows are ineffective at blocking
window ANC systems allow windows to retain their full functionality, an important feature for
homes. Recent advancements of open-window systems have yielded several variations with
varying degree of noise reduction capabilities up to 10 dB
in some cases. To enlarge the
noise reduction area, multichannel ANC systems with multiple reference microphones, secondary
sources and error microphones have been proposed
. To maintain the original functionality of
domestic windows, it is essential to minimize the obstruction caused by the ANC system.
Selecting small speakers as secondary (anti-noise) sources, and reducing their distance with
the reference microphones are critical design constraints. On this note, compact flat-panel
speakers have been proposed as effective secondary sources for narrowband noise
, or as
active structures limiting the noise wave propagation
. However, the characteristic differences
between flat-panel and cone-shaped speakers are not comprehensive when these are used as
secondary sources in ANC. In this paper, we compare the performance of two specific models of
cone-shaped and flat-panel speakers, both presenting small size and thus suitable for open-
window ANC systems. The rest of the paper is organized as follows. Section 2 includes the
technical features of the speakers and characterization measurement results. Section 3 details our
ANC test bench and the experimental settings. Results and conclusions are discussed respectively
in Section 4 and Section 5.
2 SPEAKERS CHARACTERISTICS
The two types of speakers we compare in this study are the “FRS 5”, a traditional cone-
shaped speaker, and the “Heyban”, a flat-panel speaker, manufactured respectively by Visaton
and Protro. A summary of the technical specifications is presented in Table 1. Both speakers
present compact dimensions relative to the typical size of domestic windows, and therefore result
in a minimal obstruction. The electromechanical transduction characteristics of the speakers, as
well as their geometries, present evident differences. The cone speaker can be modelled with a
cylinder shape, while its flat counterpart can be represented with a rectangular parallelepiped. The
first would determine a smaller obstruction in terms of surface area, while the second minimizes
the volume occupancy, featuring an extremely thin structure.
Both speakers have a power rating of 5 W, and the impedances are 8 Ohm and 6 Ohm,
respectively. The mean Sound Pressure Level (SPL) is about 6 dB higher for the cone speaker,
which is also evident in the respective frequency response plots of Fig. 1 (left). According to the
specifications from the manufacturer, the flat speaker has a wider frequency range, especially in
the low frequency range. To better characterize the response in the low- and mid-frequency
range, which is critical for ANC applications, we measured the SPL and the Total Harmonic
Distortion (THD) of tones in the band of 100 Hz to 2 kHz at a uniform frequency step of 100 Hz.
The THD measurement covers up to the 10
harmonic. The gains of the two independent
amplifiers were calibrated to drive a white noise signal with SPL of 100 dBA at 10 cm on the
frontal central axis. The results, illustrated in Fig. 1 (right), show a lower THD for the cone
speaker, especially when near the boundaries of the 100 Hz to 2 kHz band. The flat speaker
presents particularly high THD below 400 Hz with readings above -30 dBC. With concurrent low
SPL and high THD, at 100 Hz and 200 Hz we expect poor noise reduction when using the “FRS
5” or the “Heyban” as secondary sources. This was verified in preliminary tests, and thus we
considered only tonal noise starting from 300 Hz in the experiments detailed in Section 3.
Table 1 – Speakers technical specifications summary.
Mean SPL (1 W/1 m)
1% @ 1 kHz / 1 W
180° @ 4 kHz
5(d) x 3.1(h) cm
5.5(w) x 9.1(l) x 0.7(h) cm
Back propagation and radiation patterns are among the characteristics of the speakers that
need to be considered for ANC applications. These specifications are only partially provided by
manufacturers and hence we proceeded with an additional set of measurements. Excessive back
propagation may incur a feedback from the secondary source speaker to the reference
microphone, especially when noise and anti-noise SPL are high, which interferes with the noise
cancelling algorithm. In our context, feedback interference risk is high as the speaker-to-
reference-microphone distance is limited in a compact ANC setup. The back propagation to the
reference microphone was estimated at various distances on the central propagation axis of the
speaker, in the reverse direction as shown in Fig. 2 (right). The back propagation was measured
with the speakers calibrated for generating a SPL of 90 dBA at 6.5 cm in the direction of forward
propagation on the central axis. The results illustrated in Fig. 2 (left) show that the back
propagation of the cone speaker is on average 4.8 dBA lower than that of the flat speaker. This
gap is consistent across the measured ranges from 10 cm to 80 cm. In ANC implementations, the
drawbacks of the feedback interference captured by the reference microphone can be mitigated by
inserting sound absorbing material or using a digital cancelling filter
. However, selecting the
secondary sources with a small back lobe in the radiation pattern could improve the stability and
robustness of ANC systems.
Fig. 1 – Frequency responses (left), single tone SPL and THD (right) for cone and flat speakers.
Fig. 2 – Back propagation to reference microphone. SPL (left) and measurement setup (right).
The frontal radiation pattern was measured with the speakers placed in the same structure
used in ANC experiments, which is described and illustrated in Section 3.1. The amplifier was
calibrated to drive the speakers with 90 dBA white noise at 10 cm distance on the central axis.
We took 32 SPL readings at a distance of 11 cm from the speaker, with the measurement
microphones uniformly distributed in a 4×8 matrix with a length of 52.5 cm and width of 24 cm.
The same measurement was repeated by doubling the distance to 22 cm. Results are illustrated in
the heat-maps of Fig. 3. With respect to the heat-maps, speakers were located below the
coordinates CNTR-m3 (minor offset toward m4). The differences between the two directivity
patterns are evident. The cone speaker presents a wider emission angle than the flat speaker. On
average, the cone’s SPL readings are higher at every location of 3.5 dBA at 11cm and 2.6 dBA at
22 cm. The flat speaker propagates a slightly higher SPL than the cone-shaped only on the central
axis (0.4 dBA at 11cm and 0.1 dBA at 22 cm). The differences between the maximum and
minimum SPL for the low and high measurement plane are respectively 13.4 dBA and 6.8 dBA
for the “FRS 5”, while for the “Heyban” we measured 15.2 dBA and 9.5 dBA. In Fig. 3, some
reflections determined by the ANC experimental setup are visible at positions m1. For ANC, a
wider directivity pattern has the potential to reduce the noise in a larger region, but it can also
lead to a constructive anti-noise wave, resulting in SPL increase. This depends on the specific
ANC configuration, algorithm, number and positioning of secondary sources and error
microphones. Therefore, the selection of wide or narrow radiation pattern is strictly context
specific. Our observations on back propagation and directivity pattern for the “Heyban” speaker
confirm the “sharp directivity” and “sound on both sides” features claimed by the manufacturer.
Fig. 3 – SPL heat-maps approximating the frontal radiation pattern of the cone (left) and flat
(right) speakers at a distance of 11 cm (top) and 22 cm (bottom). The 32 measurement
microphones are uniformly distributed into a 4×8 matrix with size 24 cm by 52.5 cm.
The speakers are located below CNTR-m3.
Finally, we estimated the difference in response times between the two speakers. A fast
response time reduces the delay in generating the anti-noise from the moment the noise is
captured at the reference microphone. This in turn allows shorter microphone-to-speaker distance
given identical causality constraints, contributing to the compacting of the ANC setup. For this
characterization, we drove the cone and flat speakers simultaneously with an impulse while
recording the emitted sound with two independent microphones placed right in front of them. We
used a sampling rate equal to 96 kHz and linear interpolation was used to detect the inter-sample
peak position. The flat speaker was faster than the cone of about 27.5 μs, which is equivalent to 9
mm traveling distance for sound waves. The results were consistent across repeated
measurements, different microphone positions, impulse amplitudes, and test bench swap.
3 ANC EXPERIMENTAL SETTINGS
To roughly approximate a scenario in which the noise wave propagates into the interior
space through a window, we built a quasi-cubic chamber with the top face open, where the ANC
system was installed. The chamber is made of plywood and the inner surfaces are covered with
acoustic absorbent material, which mitigates reflections and resonant modes. The external noise
source is emulated with a 7 inch cone speaker installed in a wooden case and placed at the bottom
of the chamber. On the open face, we installed an array of cone speakers and an array of flat
speakers. These are mounted on plastic supports, symmetrical with respect to the center of the
chamber. The supports do not obstruct the back of the speakers so that we could evaluate the
effect determined by the secondary source back propagation.
In our ANC experiments, we used a configuration that includes 1 reference microphone, 4
secondary sources, and 1 error microphone (1×4×1). The speakers are distributed uniformly over
two arrays with a length of 40 cm. The reference and error microphones are respectively
positioned 20 cm below and 10 cm above the central axis of the speakers array. The primary
source speaker is also located along the same vertical axis. The geometry of the chamber, the
positions of speakers, and the positions microphones are illustrated in Fig. 4, in planar top and
side views, and a 3D model. Pictures of the described setup are shown in Fig. 5. In preliminary
ANC experiments, detailed in Section 3.3, we used a 1×1×1 configuration with the primary
source, error and reference microphones located on the vertical axis of the single active speaker.
Fig. 4 – From left to right: planar top view, planar side view, and 3D model of the chamber, with
details on the location of the primary and secondary speakers, reference and error
microphones, for the cone speaker setup (top) and flat speaker setup (bottom). Length is
expressed in millimeters.
Fig. 5 – Experimental setup with cone speakers, flat speakers, primary speaker, reference and
error microphones installed in the chamber for emulating an open window ANC setup.
The array of measuring microphones is visible in the right image.
Primary and secondary speakers were driven with a Dayton Audio MA1240a multichannel
class AB amplifier. The gain of each independent channel was manually calibrated for each
speaker. For the reference and error microphones, we used the G.R.A.S. 40PH high-sensitivity
condenser microphone. Experiments and measurements were conducted in a room with acoustic
absorbent material on walls, floor and the ceiling (partially). The ANC controller, as well as the
operator, was outside the measurement room.
3.2 Implementation platform and algorithm
The ANC processing was implemented on a National Instruments PXIe platform. The
microphone signals, including those used for SPL measurement, were acquired with the
multichannel data acquisition module PXIe-4499, equipped with high-precision Analog-to-
Digital Converters (ADC). The speakers were driven, through the MA1240a amplifier, with the
multichannel output module PXI-6733. The ANC algorithm was implemented in LabVIEW
running on the PXIe-8135 embedded controller. We worked with a sampling rate of 8 kHz,
sufficient for addressing noise in the low-mid frequency band considered in these experiments,
and a buffer size of 128 samples.
For this comparative study, we implemented the Filtered-X Least Mean Square (FxLMS)
with a feed-forward ANC structure. The length of the adaptive filters was set to 100
taps and the fixed step size was set to 5×10
for single tone noise cases, while we used
respectively 1000 taps and 1×10
step size for the multi-tonal noise cases. The length of the
secondary path filters is 100 taps and the coefficients were estimated offline using the least-mean-
square (LMS) algorithm.
The ANC causality constraints are determined by the distance between reference
microphone and secondary sources. In our compact setup, this distance is limited and the
resulting constraints are tight. These cannot be satisfied in our implementation based on a general
purpose operating system. The minimum input-output latency is high due to the high-quality
ADC and also the buffering that allows real-time computation. In this study, however, we are
addressing only periodic noises with invariant phase and amplitude sinusoidal components. In
this scenario, the feed-forward ANC structure can reduce the SPL without causality. The excess
of latency between reference microphone and secondary speaker can be modelled with a phase
shift of the periodic noise signal, which is compensated in the adaptive filter through the error
microphone feedback signal.
3.3 Test cases and measurement setup
To compare the noise reduction performance of cone and flat speakers, we measured the
SPL reduction over a large set of test cases. In the preliminary test, we used a 1×1×1 ANC
configuration with the following noise signals: single tone at 300 Hz, 600 Hz, 900 Hz, and 1200
Hz; dual tone at (300, 500) Hz, (800, 1100) Hz, and (500, 1200) Hz. For the 1×4×1
configurations, in addition to the test cases listed above, we also considered the following multi-
tone noise signals: (300, 600, 900) Hz, and (300, 600, 900, 1200) Hz.
For each test case, SPL readings with fast time weighting (equivalent to 125 ms intervals)
were taken with the ANC system enabled and disabled. In the results, these are averaged over 10
separate readings. We used an array composed of 8 40PH microphones, placed in 8 different
positions. A total of 64 measurement points were distributed on two parallel planes, at the vertical
distance of 11 cm (labelled “LOW”) and 22 cm (labelled “HIGH”) from the secondary source.
Each plane has 32 SPL reading points uniformly distributed in a 4×8 matrix with a length of 52.5
cm and width of 24 cm. The displacement of the measurement points is illustrated and labelled in
Fig. 5. For each measurement distance, the microphone array was positioned on the top and
parallel with the central axis of the secondary source array (labelled “CNTR”), shifted 8 cm
towards the outer of the chamber (labelled “OUT8”), shifted 8 cm and 16 cm towards the center
of the camber (labelled “IN8” and “IN16”). For the 1×1×1 experiments, the flat and cone
speakers enabled are located right below the measurement location CNTR-m3.
The ANC implementation described in this section, including the chamber emulating the
open window and the ANC algorithm, has room for improvement be improved. This work aims
to compare the performance of different compact speakers in a specific ANC application.
Therefore, we are focused on the noise reduction patterns, rather than on the absolute noise
reduction capability. Our setup can thus be considered as a solid test bench for evaluating cone
and flat speakers in identical operating conditions.
Fig. 6 – From left to right: planar top view, planar side view, 3D model of the 64 measurement
points for the cone speaker setup (top) and flat speaker setup (bottom). Length is
expressed in millimeters.
4 RESULTS AND OBSERVATIONS
The results of the test cases are presented in Table 2 separately for the cone-shaped and flat-
panel speakers, including average, maximum, and standard deviation of the SPL reduction for
each measurement plane. Results are grouped per ANC configuration and noise type. The heat-
maps of Fig. 7 and Fig. 8 illustrate the SPL reduction patterns over the 64 measurement points.
As expected, the highest SPL reduction values are always achieved in the region
surrounding the error microphone, which is located approximately at CNTR-m3 for 1×1×1 cases,
and between CNTR-m4 and CNTR-m5 for 1×4×1 cases. For the single secondary source with
single tone, we observed that the maximum SPL reduction along the vertical axis of the speaker is
3 to 4 dBA higher with the cone speaker. Moreover, the cone speaker yields an overall better
noise reduction on the higher measurement plane. A roughly similar pattern can be found with
dual tone noise. The SPL reduction is lower (about 1.5 dBA) and presents a more scattered
pattern with the single flat speaker, while a more uniform and better reduction of the noise SPL is
evident with the cone speaker. In this configuration, the wider directivity pattern of the cone
speaker has shown its advantage.
When using four speakers as secondary sources, the noise reduction obtained with cone and
flat speakers are more similar and uniform across the different noise cases used in our
experiments. At the lower measurement plane, the cone speakers provide a slightly better SPL
reduction but higher deviation, indicating a less regular spatial noise reduction. The scenario is
inverted at the higher measurement plane. The setup with flat speakers resulted in a slightly better
noise reduction as well as deviation. Increasing the number of the secondary speakers from 1 to a
line array of 4, higher SPL reduction (values above 8 dBA) is obtained in an extended spatial
region, along and across the central axis of the speaker array. Although the speakers are placed at
close distance in the ANC array, the wider directivity pattern of the cone speakers did not incur
any interference detrimental for the SPL reduction. In this configuration, the directivity pattern
differences between cone and flat speakers have negligible impact.
The noise reduction results obtained at 300 Hz show that the cone speaker outperforms the
flat speaker of an average 3 dBA and 2 dBA respectively in the 1×1×1 and 1×4×1 configuration.
This difference is compliant with the low SPL and high THD of the flat speakers at low
frequencies, as reported in Fig. 1. Therefore, the cone shaped setup is more suitable in ANC
systems that abate noises with lower frequency components.
In the various test cases and experimental settings, we used identical and fixed values for the
amplitude of tonal noise and for the gain of the amplifier driving the primary speaker. The noise
SPL was close to the limit after which the back propagation of the secondary speakers introduces
a fatal feedback to the reference microphone. This limit differs for the two speaker models. The
setup using the cone-shaped speakers, which present a lower back propagation, can handle tonal
noise up to 4 dB higher than the flat speakers. Finally, during the experiments, we noticed that the
setup with flat speakers usually converges faster. This observation was consistent across all test
cases, but not verified with any systematic measurements.
Table 2 – Summary of the SPL reduction results for all test cases, with average, maximum, and
standard deviation for each case on the LOW and HIGH measurement planes.
Single Dual Single Dual Multi
Cone Flat Cone Flat Cone Flat Cone Flat Cone Flat
average 3.78 3.95 3.85 2.66 3.99 3.98 4.26 4.37 2.56 2.03
maximum 11.23 8.21 8.25 9.81 10.66 10.31 11.63 10.50 12.65 10.26
deviation 2.53 2.41 2.06 2.08 2.54 2.83 2.64 2.55 3.07 3.34
average 4.72 3.20 5.06 3.53 4.08 4.65 5.10 4.83 3.48 3.24
maximum 10.27 6.99 7.70 6.33 8.36 10.12 9.03 10.06 7.48 8.43
deviation 2.06 1.45 1.59 1.59 1.93 2.45 2.04 2.39 2.31 2.62
Fig. 7 – SPL reduction heat-maps for cone speaker (left) and flat speaker (right) averaged over
the single tone cases (top), and dual tone cases (bottom) for the 1×1×1 ANC
Fig. 8 – SPL reduction heat-maps for cone speakers (left) and flat speakers (right) averaged over
the single tone cases (top), dual tone cases (middle), and multi-tone cases (bottom) for
the 1×4×1 ANC configuration.
We presented experimental settings and discussed the results of a comparative study on the
noise reduction performance of an ANC system based on compact cone-shaped and flat-panel
speakers. In identical configurations, we found that the cone speaker performs better in reducing
the noise at low frequencies, despite its slightly worse nominal low-frequency response as
compared to the flat speaker. The cone-shaped speaker outperforms the flat-panel type also in
single secondary source experiments, providing noise reduction in a considerable larger and
more uniform region. Measurements with ANC setup including the array of four closely spaced
secondary sources showed almost identical performance between the two types of speakers.
Therefore, the use of the cone speaker may be advantageous in ANC setup with larger spacing
between secondary sources, as for installations on larger windows and wall openings. The
shorter response time of the flat speakers allows realizing smaller ANC units, though a more
effective acoustic absorbent material is required to isolate the reference microphone from the
higher back propagation. In future works, we will use the same setup to investigate noise
reduction performance with multiple reference and error microphones, and also extend the study
to narrowband noise.
This material is based on research/work supported by the Singapore Ministry of National
Development and National Research Foundation under L2 NIC Award No. L2NICCFP1-2013-7.
1. Hellmuth, T., Classen, T., Kim, R. & Kephalopoulos, S. Methodological guidance for
estimating the burden of disease from environmental noise. (2012).
2. European Environment Agency (EEA). Noise in Europe 2014. (Publications Office of the
European Union, 2014).
3. Lam, B. & Gan, W. S. Active Acoustic Windows: Towards A Quieter Home. Potentials
IEEE (In Press).
4. Long, M. in Architectural Acoustics (Second Edition) (ed. Long, M.) 175–219 (2014).
5. Fiks, B. & Jagniatinskis, A. Assessment Of Environmental Noise From Long-Term Window
Microphone Measurements. Appl. Acoust. (2014).
6. Can, A., Leclercq, L., Lelong, J. & Botteldooren, D. Traffic noise spectrum analysis:
Dynamic modeling vs. experimental observations. Appl. Acoust. 71, 764–770 (2010).
7. Nelson, P. A. & Elliott, S. J. Active control of sound. (Academic, 1993).
8. Kuo, M. S. & Morgan, D. R. Active noise control systems : algorithms and DSP
implementations. (New York : Wiley, c1996., 1996).
9. Kuo, S. M. & Morgan, D. R. Active noise control: a tutorial review. Proc. IEEE 87, 943–973
10. Kajikawa, Y., Gan, W. S. & Kuo, S. M. Recent applications and challenges on active noise
control. APSIPA Trans. Signal Inf. Process. 1, 1–12 (2012).
11. Jakob, A. & Möser, M. Active control of double-glazed windowsPart I: Feedforward control.
Appl. Acoust. 2, 163–182 (2003).
12. Jakob, A. & Möser, M. Active control of double-glazed windows. Part II: Feedback control.
Appl. Acoust. 2, 183–196 (2003).
13. Hu, S., Rajamani, R. & Yu, X. Invisible speakers in home windows for simultaneous
auxiliary audio playback and active noise cancellation. Mechatronics 22, 1031–1042 (2012).
14. Hu, S., Rajamani, R. & Yu, X. Directional cancellation of acoustic noise for home window
applications. Appl. Acoust. 74, 467–477 (2013).
15. Murao, T. & Nishimura, M. Basic Study on Active Acoustic Shielding. J. Environ. Eng. 7,
16. Byoungho Kwon, Y. P. Interior noise control with an active window system. Appl. Acoust.
74, 647–652 (2013).
17. Murao, T., Nishimura, M., Sakurama, K. & Nishida, S. Basic study on active acoustic
shielding (Improving noise-reducing performance in low-frequency range). Mech. Eng. J. 1,
18. Chen, K. & Koopmann, G. H. Active Control of Low-Frequency Sound Radiation From
Vibrating Panel Using Planar Sound Sources. J. Vib. Acoust. 124, 2–9 (2001).
19. Nishimura, M., Ohnishi, K., Kanamori, N. & Ito, K. Basic Study on Active Acoustic
Shielding. in Inter.noise 2008 (2008).
20. Zhu, H., Rajamani, R., Dudney, J. & Stelson, K. A. Active noise control using a distributed
mode flat panel loudspeaker. ISA Trans. 42, 475–484 (2003).