Académique Documents
Professionnel Documents
Culture Documents
net/publication/228361126
CITATIONS READS
30 1,161
1 author:
Colin H Hansen
University of Adelaide
290 PUBLICATIONS 4,561 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Establishing the physiological and sleep disruption characteristics of wind farm versus traffic noise disturbances in sleep View project
All content following this page was uploaded by Colin H Hansen on 31 May 2014.
ABSTRACT
The status of active noise control in terms of its application to industrial problems is discussed
and reasons for the apparent lack of enthusiasm for the technology by industry are postulated.
An industrial installation in which the author was involved is used as an example to illustrate the
complexities involved and the reasons why implementation costs are so high. The future of
active noise control in industry is dependent on a number of issues associated with hardware
configuration and cost, user friendly software, generalisation of system design, development of
low-cost, rugged actuators and sensors together with an acceptance of what is possible and what
is not. Novel approaches to achieving the control objective of reduced noise levels at the ears of
industrial employees, which sidestep limitations imposed by the physical properties of sound and
vibration fields, are also required to enable practical application of the technology in many cases.
One such novel approach, which involves virtual sensing combined with very local control and
1
1. INTRODUCTION
Ten years ago there was widespread belief that in ten years time, industrial applications of
active noise control (ANC) would be quite common. Five years even before that, extensive
advertising and promises were made by at least one company that active noise control would
soon be a feature in most consumer products. In a 1997 paper by the author1, it was pointed out
that the number and range of long term commercial installations was relatively small and that
perhaps the main reason for this was the excessive cost of system installation which could easily
exceed the cost of the basic control hardware by a large amount. The need for tuning of systems
by experts in the technology also adds a significant cost component as does the unique nature
It was pointed out in the 1997 paper that what was needed was an “inexpensive, clever,
commercial control system which includes a selection of source and sensor transducers to satisfy
most problems, and software to guide users in the correct choice and location of such
transducers”. Details of the system requirements are also discussed in the 1997 paper. Such a
applications of the technology are to increase. In the years since 1997, no such system has been
Although the lack of a suitable generic controller has hindered industrial application of ANC
technology, another contributor has been the bad publicity resulting from claims made by some
companies in the past, which were never realised. As in many emerging fields of technology,
unscrupulous patenting of ideas which have been published by other unrelated researchers in
journal papers or consulting reports has also hindered commercialisation efforts by small
2
companies who cannot afford to become embroiled in court battles proving the invalidity of
relevant patents.
This paper focusses on industrial applications of ANC technology and as such, military and
aerospace applications are not discussed, even though there are a number of successful examples
of ANC in these areas. However, these applications are not affected by the same cost constraints
that are a reality in industry and so the extension of the technology to general industry is rarely,
if ever, feasible.
In the remainder of the paper, some currently practical industrial applications of ANC will
author, which will be used to illustrate some of the issues involved and problems that need to be
overcome. Next, the types of commercially available control systems currently available will
be described and this will be followed by a description of current hardware and software
development being undertaken by the author and his colleagues. Finally, novel approaches to
reducing noise levels around the heads of people working in noise environments will be
discussed. This includes virtual sensing, head tracking, control source localisation and steering.
Current industrial applications of active noise control technology are limited mainly to the
control of plane wave sound propagation in air handling ducts, gas turbine exhausts or diesel
engine exhausts, and these are mainly feedforward systems. Active ear muffs represent a
successful application and these are usually feedback systems. Considerable development effort
has been spent on these and they have been successfully used in environments in aircraft and in
some industries. Because active ear muffs use feedback control principles (as this requires no
3
reference signal), they perform best when the error signal auto-correlation coefficient is
relatively high. They also suffer from problems in high level noise environments due to the
limited output capability of the small loudspeakers used. In impulsive noise environments use
of a feedback control system can result in ringing because of the limited frequency response of
the control system. Active ear muffs are relatively well known and will not be discussed further
here.
Another application that has been successfully commercialised is the reduction of tonal noise
in propeller driven aircraft using active engine mounts and vibration actuators mounted on the
fuselage rings2. This technology is more applicable to the aerospace industry rather than general
The most successful industrial applications of active noise control target single or multiple
tonal noise, but applications also exist that produce significant noise reductions (10 to 15 dB)
over a 1-1/2 to 2 octave frequency range. It is difficult to achieve a greater bandwidth of active
control with existing commercially available hardware and software. However, it is possible to
extend the control bandwidth by using a filtering system coupled with multi-rate sampling. This
has yet to be implemented in any commercial controller and will be discussed in Section 4A.
that must be considered, even though most of them can be completely ignored for a laboratory
a robust, reliable system in an industrial environment. The requirements that must be met for a
• a noise free reference signal that is well correlated with the error signals;
4
• robust control sources well protected from the environment;
• robust and reliable error sensors well protected from the environment;
• a reliable control system capable of self starting from a power shut down by throwing a
single master switch with a time delay between switching amplifiers on and switching on the
• a system that is either immune to transient events or can shut itself down on detection of such
• automatic readjustment of the controller should an error sensor or control source fail;
• automatic shutdown in the event of sufficient sensor or actuator failures to make the system
ineffective;
• alarms to indicate which error sensor channel or control source channel is defective; and
• fail-safe mechanisms to prevent noise levels ever exceeding the existing primary noise levels
prior to control for any reason such as controller instability caused by transients or transducer
failure.
It is important to remember that until now and probably for the foreseeable future, active control
systems are an extremely complex form of noise control technology compared to existing passive
alternatives. This complexity has in the past, translated into unreliability and excessive cost so
that now the situation exists where active noise control solutions in industry are only considered
where alternative passive solutions are either impractical to implement or exorbitantly expensive.
The only means by which this trend will be reversed is by gradual demonstration of relatively
low cost, easily maintained, reliable systems and this situation is still some way into the future.
5
3. AN EXAMPLE OF THE DEVELOPMENT AND INSTALLATION OF AN
Perhaps the requirements for successful ANC systems are best illustrated by an example of
an actual installation that includes many of the capabilities listed in the previous section. The
example is a system used to control sound propagation in the exhaust stack of a spray dryer unit
used in a dairy factory. Perhaps the most complicating aspects of this example were:
• the frequency at which control was desired was above the higher order mode cut on
frequency; and
• the sound field in the duct was never steady so that the error sensor outputs varied rapidly
The aim of the ANC system to be described here was to reduce the amplitude of tonal noise
at the fan blade pass frequency, emanating from the top of the spray dryer exhaust stack and
radiating into the surrounding community. The diameter of the exhaust stack is 1.6 m, the
temperature range of exhaust air is from 60 to 90 degrees C and the blade pass frequency ranges
from 170 to 190 Hz. Calculations show that for this stack, two higher order modes propagate,
thus greatly complicating the control system required. Not only do higher order modes exist, they
exhibit “spinning” characteristics which result in the nodal lines twisting with respect to a
reference, with the twist angle being a function of axial location in the duct. The twist angle at
a fixed axial location was very sensitive to temperature fluctuations and this resulted in quite
rapid and large sound pressure fluctuations at any given location in the duct. This is discussed
a little more in the controller section to follow. Of course, a conservative approach would have
6
been to weld axial splitter plates in the duct so that only plane waves could propagate between
the control source and error sensors. For these to be effective, the welds would have to have been
air tight. As the duct was stainless steel and vertical, it was considered too difficult to implement
One of the expensive and time consuming aspects of the installation was the signal cabling.
Extensive lengths of signal cabling were needed to connect the reference and error sensors to the
controller and to connect the controller to the loudspeakers, all of which were physically
separated by many tens of metres. Care had to be taken in joining wires to minimise the
possibility of corrosion of the joints and the introduction of extraneous electrical noise. In all
cases, water proof junction boxes were used. Due to the sensitive nature of the DSP chips and
other electronic components, the controller had to be located in a temperature controlled room
and space was found along side other process control instrumentation.
Interesting practical aspects associated with each element of the control system are described
in the following sections. It is assumed that readers are familiar with the general configuration
of a multi-channel feedforward active noise control system, and if not appropriate textbooks may
be consulted 3. The main components and their associated technical challenges, which will be
discussed here, are the reference sensor, the error sensors, the control sound sources and the
controller.
A. Reference Sensor
The reference sensor is used to derive a signal that is used by the controller to generate the
driving signal for the control loudspeakers. For the spray dryers, the reference signal was derived
using a Hall effect tachometer which produced an electronic pulse each time a gear tooth passed
7
its field of view (see Figure ?). The gear wheel was manufactured especially for the project and
was mounted on the fan drive shaft. The reference signal was generated digitally by passing the
tachometer signal into an Analog Devices Sharc EZ-kit 21061 DSP board. The time between
successive tachometer pulses was measured by passing the signal into the digital interrupt port
which interrupts the central processor and allows it to count clock cycles between successive
pulses. As the CPU runs at 40 MHz, the resolution is 25 nano-seconds. The DSP is programmed
to generate a sine wave with the same period (or any required multiple thereof) and output it with
a sample rate of 16 kHz through a low pass filter into the control system reference signal input.
The reference signal sine wave thus remains synchronised with the fan blade passing frequency
at all times.
B. Error Sensors
The large turbulent pressure fluctuations in the duct propagate at the speed of the air flow and
not the speed of sound and thus their presence complicates the error sensor design which ideally
should only respond to acoustic pressure fluctuations if undesirable control signals are to be
avoided. Initially, microphones were placed in the end of a 2m long porous tube with the
intention of amplifying the acoustic pressure fluctuations and attenuating the turbulent pressure
fluctuations. This approach had two main problems. The turbulent pressure fluctuation rejection
was insufficient and anchoring the tubes in the vertical duct as well as protecting them during
cleaning operations was problematical. They also represented a possible site for bacterial growth.
The final design was a flush wall-mounted sensor, consisting of an electret microphone inserted
in a tube with its own preamplifier, connected to an enclosure filled with a block of acoustic
foam protected by a metallic foil to filter out some of the turbulence-related pressure fluctuations
8
(see Figure 2). Small disks of Vyon, a porous plastic, were inserted in the tube in front of the
microphone to attenuate the sound level incident on the microphones so as to avoid mechanical
saturation of the microphone membrane, because of the high acoustic levels present inside the
duct. As shown in the figure, a continuous air flow is maintained over the microphones to keep
them cool. The aluminum foil facing on the acoustic foam insert prevents contamination of the
With this design, background pseudo sound was reduced to more than 13 dB below the
targeted acoustic error signal. These sensors have been operating successfully for over a year
Twelve error sensors were installed on the exhaust stack, flush with the duct wall. They
were organised in three rings of 4, 5 and 3 sensors respectively, as shown in Figure 3. The
three rings are respectively at a distance of 0.20 m (No. 6,7,8,9), 0.70 m (No. 1,2,3,4,5), and
C. Sound Sources
Due to the high temperature and sometimes wet environment, it was necessary to provide
protection for the loudspeaker cones and cooling for the coil. At first, a membrane of Viton, a
rubber-like material, was tried. Even though this was limp and relatively lightweight with a
transmission loss close to zero at 200 Hz, it had a dramatic effect on the loudspeaker output,
reducing the maximum achievable sound level by 20 dB, probably as a result of the viton
material acoustically loading the loudspeaker and changing its radiation impedance. As the
loudspeakers have to generate sound levels in the duct of up to 138 dB, this amount of loss was
unacceptable. The final design consists of a 600W paper cone, low-frequency loudspeaker,
9
coated with a silicon compound to protect it against humidity, and then a reflective coating to
protect it against dust particles and heat build-up. To further discourage the build up of milk
powder dust on the speaker diaphragm, it would be useful to introduce a small air jet close to
where build-up is possible. The enclosure at the back of the speaker is connected to the front face
of the speaker with a small tube to ensure pressure equalisation of the front and back surfaces
of the speaker cone. Without this equalisation, the speaker cone would suffer a DC shift and fail
to function correctly. The heat generated by the speaker coils in operation is dissipated using
pressurised air injected in the vented speaker backing enclosures. The speakers have survived
so far for over a year in their hot and humid environment, with failures that have occurred being
due to over pressurisation of the air cooling system, resulting in bursting of the paper cones.
Each speaker enclosure is mounted flush with the duct wall and one to two speakers are
mounted at each of four different axial locations, covering a total of seven different locations,
as shown in Figures 4 and 5. The top speakers are mounted approximately 3500 mm below the
duct exit plane. During operation, the 600W speakers are driven at about 5W, which minimises
D. Controller
A feedforward controller using a periodic block FXLMS algorithm4, (Eqs. 12-14), with control
filter updates carried out every 15th sample) was chosen for the task. The periodic block FXLMS
algorithm is less computationally intensive than the standard FXLMS algorithm as it allows
control filter updates to be done after a block of error sensor data has been collected. There are12
input channels for the error signals and 6 output channels for the control signals. The reference
signal is derived from the tachometer on the fan shaft so there is no acoustic feedback from the
10
control sources to the reference sensor. The cost function is the average sound pressure from all
12 microphones and even though the sound pressure recorded by individual microphones
changes quite quickly, the average changes much more slowly and this allows the control system
to converge to an optimum. Note that the cost function also includes a quantity proportional to
the control signal outputs, which is equivalent to including leakage3. Other cost functions such
as weighting the microphone signals differently so that more effort would be directed towards
controlling the larger signals, were tried but not found to be any more effective than the average
mentioned above. The optimum size of the convergence coefficient was found by trial and error
as a compromise between convergence speed and instability risk, and was adjusted on site during
system set up. Needless to say the acceptable stability risk was zero so a relatively small
It is well known3 that to ensure stability of the control system, it is necessary to have an
accurate estimate of the cancellation path impulse response or transfer function (CPTF). In many
laboratory experiments, this is done off-line prior to starting up the controller, using random or
pseudo random noise injected into each control source in turn. For any industrial application, this
is the preferred way to start up the system. However, once the system begins to operate, it is
generally not possible to shut it down to check the CPTF and an on-line procedure is needed.
Various schemes were tried on the spray dryer system. The use of the primary noise signal to
determine the CPTF as described by Sommerfeldt and Tichy5 is only accurate for single channel
systems. Another scheme that was considered for the spray dryer system involved the use of an
introduced tone, several decibels lower in level than the blade passing tone and at a range of
frequencies around but not too close to the fan blade passing frequency. When the blade passing
frequency varied, the frequencies missed previously would be tested, resulting in the gradual
updating of a table that could be stored in memory. Each time the fan frequency changed, the
11
table could be looked up for the most recent CPTF data. However, if the fan ran for any length
of time at the one speed and other environmental conditions changed, this method may not take
into account changes in the CPTF and the system could go unstable. Thus this scheme was
rejected as too risky and it was eventually decided to use low level random noise, 30 dB below
the level of the blade passing tone, introduced into each control speaker in sequence. Using such
low level noise, which was not detectable by ear, required very long averaging times to
E. Controller Hardware
The controller hardware used for the project is shown schematically in Figure 6. The
EMERALD modules are DSP boards based on the Analog Devices ADSP21062 processor and
they are suitable for executing both system-level software (such as the interface to the PC) and
signal processing software, such as system modeling and control filter adaptation. In the current
system, one EMERALD board is used for each purpose, with one acting as a slave to the other,
which is the master. The interface between the modules is implemented using the processor's
Not shown in the above layout is the Analog Devices EZ-Kit 21061 which was used to
process the reference signal input pulse train to produce a sine wave at the blade passing
frequency, which was used as the reference signal input to the system in the above figure. The
• excessive noise on the tachometer signal due to electricians laying signal cables next to
power cables;
12
• electromagnetic pickup in DSP board communication lines due to radiation from motor
• automatic identification and alarm set state for any failed speaker or microphone
channels;
• system shutdown and temporary rest if noise levels exceeded primary noise levels;
The performance of the system was measured in the community rather than the error sensors
in the duct (where the performance was much better). Noise reductions in the community ranged
from 9 to 14 dB at the blade passing frequency with the reduction being such that the tone was
To make active noise control more accepted in industry, considerable effort is needed to
conquer the last remaining barriers to this acceptance. It is true that the limitations of the
technology are better understood than they were ten years ago. Now the thrust is rightly directed
at using innovative means to overcome these apparent limitations, some of which are a result of
13
the laws of physics and some of which are a result of insufficient power and intelligence in
current commercial control systems, or the lack of sufficiently robust sound and vibration
generating transducers. Industrial applications of ANC are special in that any installed system
must be electrically as well as mechanically robust and should function for years with no
maintenance. Thus relevant research is or ideally should be directed towards resolving a number
• Development of low cost, yet rugged, high output inertial actuators for vibration isolation
and structural vibration (and noise radiation) suppression. This is important in cases where
• Development of low cost, yet rugged, high output sound sources for sound suppression.
Although this development is important if active noise control is to become more widely
accepted in industry, there is very little relevant work reported in the literature. The
generation of high sound levels in ducts and muffler systems may be achieved using horns
that are designed to couple with a duct rather that free space. Although no work has been
reported using this approach, some work has been reported recently on the use of a flapper
• Development of control source systems that are remote from the location where noise
reduction is required. These may take the form of non-planar speaker arrays to direct
cancelling energy into desired areas with minimal effects in other areas.
• Development of remote virtual sound pressure and energy density sensing systems so that
microphones do not need to be located near where the sound field is being cancelled or
suppressed.
• Development of tracking systems that can keep track of people’s head locations and direct
14
• Development of a third generation electronic controller with sufficient power to cope with
the new control source and sensing systems and sufficiently user friendly to be useful with
minimal training.
• Development of various control algorithms tailored for specific problems, some of which are
listed below. A number of appropriate control algorithms are discussed by Qiu and Hansen4.
• Development of system management software necessary for effective, automatic and user
sources and undesirable source outputs in the case of a transient event being recorded by an
error sensor.
Work at the University of Adelaide is currently focussed on three of the above aspects that
are driving the technology forward towards the goal of widespread industrial acceptance of the
technology. The first aspect is concerned with the development, currently underway, of a third
generation controller which, when complete, will be somewhere between existing controllers and
the required future intelligent controllers in terms of capability. The second aspect is concerned
with overcoming physical limitations associated with the small zone of silence that exists around
error sensors in enclosed sound fields. The third aspect is associated with developing a focussed
15
With continual advances in DSP memory capability and speed, and steady reductions in cost
per unit of capability, it is becoming more and more practical to develop multi-channel
controllers capable of having thousands of taps in the control filters and cancellation path
modelling filters. Note that the number of cancellation path modelling filters needed is equal to
One of the big problems with the development of an electronic active noise control system
is the difficulty in obtaining funding for it. It is too risky in terms of potential economic returns
to attract industry support and too applied to attract support from research granting bodies.
Perhaps the best hope lies with Defence Departments which can see a use for such technology
In deciding on the appropriate architecture for a third generation controller, one must choose
from a number of possible fundamental options. Briefly, there are three practical fundamental
1. Use currently available DSP boards and IO boards which can be inserted in a PC and
programmed in “C” and run via the Windows operating system. Thus if more powerful
boards become available, they can simply replace existing boards with negligible, if any,
suffers from a large overhead in terms of memory and a system programmed in this way
will invariably suffer due to processing speed limitations. This type of system is also too
16
2. Put a general-purpose DSP board in a custom enclosure with power supply and custom
I/O boards, programmed using DSP programming tools. Unfortunately the continued
supply of these boards is often problematic, their specifications are unreliable and they
have many components that are not needed for active noise control. There are also
considerable problems interfacing General purpose DSP boards with I/O boards from
another manufacturer.
3. Construct a multi-processor DSP board from scratch with sufficient power and memory
to meet the most demanding ANC application. Include it in a modular system which
allows the use of multiple DSP boards and multiple IO modules which can be tailored
to the number of channels needed and the processing power and memory required.
The third option offers more flexibility in terms of producing a system optimised for active
noise and vibration control. Such a system is currently being developed with the intention that
it will have all of the properties listed as dot points in Section 2. It will also have the capacity for
multi-rate filtering (as discussed in Section 4A) with the associated extension of good
performance from two to five octave bands. The intended future system configuration is
illustrated in Figure 9.
The multi-rate filtering will be achieved in the I/O modules (labelled as “Analog I/O” in the
figure), as each module will contain a high speed A/D converter, a D/A converter and a low cost
DSP, all sampling at 250 kHz. The DSP on each I/O board will manage the required down-
sampling and will allow multi-rate filtering as described in the next section. Note that there will
be one DSP for each channel, which will provide ample processing power for all the I/O
management tasks and multi-rate filtering as well as transducer failure and signal overload
management.
17
The central processor board (labelled “DSP” in the figure) will contain 4 Analog Devices
DSP Sharc (floating point) processors. Using more DSPs on the one board does not significantly
improve the overall speed of the system due to bottlenecks associated with using external
The controller will also be capable of being monitored, diagnosed, reset and set up remotely
The new controller will allow a choice of a number of different control algorithms, including
frequency domain algorithms optimised for both tonal and broadband sound field control.
Feedforward, feedback and hybrid algorithms will be available and advanced users will also be
able to program their own control algorithms. Cancellation path identification will be possible
The hardware will be configured to minimise digital delays and to provide a large capacity
for filter taps for both control filters and cancellation path estimates.
One of the limitations of current active control systems is the limited bandwidth over which
they operate (usually 1-1/2 to 2 octaves at most). However, it is possible to extend the control
bandwidth by using a filtering system coupled with multi-rate sampling, but as yet this has not
practice is that the input signals from the reference and error sensors are sampled at a very high
rate (250 kHz) and filtered into octave bands using a number of digital filters. The high sample
rate eliminates the need for anti-aliasing filters. Each octave band signal emerging from a digital
filter is then re-sampled at the optimum sampling rate for that particular frequency range (usually
18
about 15 times the centre frequency of the octave band). For each octave band filter, a control
filter with adjustable weights is needed to provide the control signal and the control filter outputs
for each octave band need to be re-combined prior to being converted to an analog signal and
sent to the amplifier driving the control actuator. For a multi-channel control system, one set of
A five-in, four-out multi-rate sampling system is illustrated in Figure 10, where the
arrangement is shown for a single octave band and the combination with the systems for other
octave bands is done at the output labelled as “sum over all octave bands”. Following the octave
band boxes on the inputs, a digital down-sampler adjusts the sample rate so that it is optimum
for the particular octave band. It can be seen that the processing power needed to implement such
a system is enormous and current technology is only just up to the task, even though the idea was
Another limitation with current control systems is the need for analog anti-aliasing filters
which must be duplicated for each sampling rate selected. One way around this is to use 1-bit
Codec A/D and D/A converters which sample at a very high rate (1 MHz or more) and through
some interesting hardware manage to produce a 16 bit result at 32 kHz, with the capability of a
final rate as low as 5 kHz. The most unfortunate property of codecs is that they have a 30 sample
delay at the final sample rate. At the 32 kHz rate this is 1 millli second and at the 5 kHz rate it
random noise. It is not a problem for control of periodic noise, nor for the control of random
noise in a duct where it is possible for the control source to be further than 2.1 metres from the
reference sensor. Delays introduced by loudspeaker control sources increase this required
distance in practice, of course. One way of achieving low sample rates for the control of very low
frequency disturbances is to use digital down sampling based on the highest sample rate possible.
19
In this way the delay is the 30 samples at the high rate plus whatever overheads are used in the
down sampling. Some results of doing this with one particular commercial controller are shown
in Figure ?, where it can be seen that the reduction in time delay is slightly less than the down-
sampling divider factor due to the computational overheads associated with the down-sampling.
It can also be seen that the time delay is slightly dependent on the frequency of the input signal
sufficiently large number of control sources are used in a feedforward active noise control
system, it is not too important where they are placed in terms of achieving a global reduction in
some cost function. That is, it is not necessary to go through an optimisation process to optimally
place the control sources and error sensors. In an attempt to further simplify the control process,
Fuller and Carneal9 suggested using hierarchical bio control in which a small number of signals
are sent from an advanced, centralized controller and are then distributed by local simple rules
to multiple control actuators. No commercial systems currently use this approach, even though
Some recent results comparing the effectiveness of multiple single channel control systems
versus a single multi-channel control system attached to the same sensors and actuators are
shown in Figure 1210. In the figure are shown results for vibration control of a cantilevered beam
using a foam damper, inertial actuators (unactivated) in the foam, activated inertial actuators
driven by single channel controllers and activated inertial actuators driven by a multi-channel
controller. It can be seen that for this simple example involving broadband control, the multi-
20
channel controller achieves much better results. Thus there is considerable impetus for the
necessary and has all of the features listed as dot points in Section 2.
Nevertheless the multiple single channel controller case also reflects substantial noise
as convergence speed, there is an increasing trend towards the use of multiple single channel
controllers for cases where their performance is acceptable and the improved noise reduction
possible with a multi-channel controller is either not needed or not possible due to tracking speed
requirements for noise environments that are barely quasi-stationary. As more is understood
about the performance of multiple single channel controllers, they may find application in
industrial situations as a result of their simplicity, low cost, tracking ability and adaptation speed.
As use of feedback control to actively reduce noise levels is impractical outside of ducts and
control can be implemented. It is also impractical to have long runs of cabling from reference
sensors to the active control system, so it will be necessary to develop a system that transmits
the reference signal to the control system without the need for wiring or cabling. Two types of
reference signal may be envisaged: one derived using a microphone close to the noise source for
random noise and one derived from a tachometer signal for periodic noise synchronised to a
rotating shaft.
One application of great interest for which the generation of an adequate reference signal is
difficult is for the control of random noise propagating in a duct containing a relatively high
21
speed air flow. The problem with the presence of the air flow is that it produces turbulent
pressure fluctuations which are not related to the noise to be controlled, as they only propagate
at the speed of the flow and not the speed of sound. Unfortunately, a microphone is incapable
of distinguishing between turbulent pressure and acoustic pressure fluctuations and the presence
of the turbulent pressure fluctuations reduces the correlation between the reference and error
microphone signals, which in turn reduces the capability of the active noise control system to
achieve a reduction in noise. There are a number of approaches that have been used in the past
to try to separate the turbulent pressure fluctuations from the acoustic pressure fluctuations prior
to them being sensed by the microphone. One method commonly used is to place the microphone
at the end of a long porous tube, which results in the acoustic pressure fluctuations being
reinforced because they are travelling at similar speeds inside and outside of the porous tube.
Perhaps a more clever way of achieving the same result is to use a number of reference
microphones distributed around the duct wall at the same axial location and then adding all of
their signals to produce a single reference signal. This has the effect of cancelling the random
turbulent pressure fluctuations (as their phase will be random at all microphone locations) while
at the same time enhancing the acoustic response, as the plane acoustic wave will be in phase at
all microphone locations. Of course, this also has the effect of excluding higher order modes
should any be present. In addition, any phase mismatch between microphones will limit the
When periodic noise is to be controlled and the noise is synchronous with a rotating shaft,
it is common to use the output of a tachometer as a reference signal. Unfortunately this output
is usually a pulse that occurs once per revolution and what the active control system ideally
needs is a periodic reference signal that contains the frequencies to be controlled and the
reference signal amplitude at these frequencies should be directly proportional to the required
22
noise reduction at each corresponding frequency. Thus there are various techniques to convert
the once per revolution pulse from a tachometer into the required multi-sinusoid reference signal.
The first step that is often taken is to attach a gear wheel to the rotating shaft so that the
tachometer (magnetic or optical pickup) puts out a pulse each time a gear tooth passes it and the
However, this still results in many unwanted higher order harmonics in the reference signal (due
to the non-sinusoidal shape of the tacho signal) and further signal processing is needed to
eliminate the unwanted harmonics and generate the wanted harmonics with the optimum relative
amplitude to the fundamental. This may be done using a DSP board as described in Section 3A.
F. Cancellation Path ID
estimates at regular intervals as any given estimate becomes less valid over time. Such a
procedure involves switching off the active noise control system filter weight update process and
then introducing low level random noise for a short time (a second or less) to re-estimate the
cancellation path impulse responses. Then the control filter weight update can be re-started. This
was detected. A smarter system can be further developed as follows: whenever the total error is
out of the normal range, the controller can measure a new cancellation path model and then save
the transfer functions with other relevant parameters such as a fan speed (if in a duct) and
temperature. The next time a similar situation is encountered, the controller can automatically
recall the saved cancellation path transfer functions and use them for the new controller filter
23
weights update. In this way the controller gradually becomes more intelligent and more adapted
to its environment.
Most industrial sources are located in enclosed spaces and in many cases there are a number
of sources contributing to the noise problem in any one location. Global control in enclosed
spaces and in free space is difficult to achieve, even for tonal noise11 and local control is
unsatisfactory as it invariably includes a volume which is too small. One approach that has been
used in the past is to incorporate an ANC system in a headset. While this approach has had some
success in aircraft cabins, success in industry has been limited for two reasons; they are not
particularly effective in high noise level environments or in environments where the noise is
impulsive or rapidly changing, and they are not very comfortable to wear.
addition, work is underway to develop a system that will have the same effect as ear muffs
without any hardware on the ear of the exposed person. This approach involves developing a
steerable, focussed anti-noise signal at a person’s ear, without the need for sensors or active
noise control sources near the ear, and in such a way that the increase in noise levels at other
locations is relatively small. Such a system can be divided into the following parts, all of which
require some shift away from standard ANC technology and all of which are discussed in more
• Virtual sensing, whereby the physical error sensors are remote from the ears, and
incorporating energy density sensing for more gradual sound pressure gradients near the
24
• radio transmission of reference signals from tachometers or microphones close to noisy
equipment;
• focussing of the active noise cancellation field, generated by using planar and non-planar
• steering the sensor system and the active noise cancellation field generator to ensure that the
• developing the third generation electronic control system (described in the previous section)
capable of handling the immense computational load and capable of remote monitoring, reset
It has been shown by a number of authors11-13 that it is not feasible to attain global sound
reduction using active control of large enclosed sound fields. Thus one is left with the possibility
of achieving small, localised regions of reduced sound pressure level. In conventional active
noise control systems, these localised regions are at and close to the error sensors, which clearly
is not practical as the regions of minimum noise level should be at the ears of people who need
to be protected. Another problem often encountered is the large pressure gradient in the vicinity
of the error microphones which is characterised by a large sound pressure reduction at each error
microphone, and a rapidly increasing sound pressure level as the distance from the microphone
increases. The size of the zone within which at least 10 dB of sound reduction can be achieved
is about 1/10 of a wavelength of sound, but the worst problem is the severe pressure gradient
25
within that zone which results in sharp changes of sound pressure level as one moves small
One promising means of reducing the severity of the pressure gradient in the vicinity of the
microphone is to use energy density rather than just pressure as the cost function to minimise14.
An energy density transducer senses pressure gradient (or three dimensional acoustic particle
velocity) as well as acoustic pressure. A first generation energy density sensor, containing 4
microphones is illustrated in Figure 13. However, this design, which relies on the use of low-cost
electret microphones is effective only over a limited frequency range: the low frequency limit
defined by the lack of phase matching between the microphones and the high frequency limit
defined by the microphone spacing no longer allowing an accurate estimate of the pressure
gradient from the pressure difference between adjacent microphones. Recently, a new type of
energy density sensor has been developed by Phone-Or Ltd., an Israeli company. This three
dimensional energy density sensor consists of three optical pressure gradient microphones and
a single optical pressure microphone. For each pressure gradient microphone, the pressure
closed tube, with one small hole in the tube wall on each side of the membrane. A LED in the
control unit sends light through a fibre optic cable to the membrane, the light reflected back from
the membrane is collected using a photo-detector and the doppler shift in frequency determined
to give a measure of the membrane velocity and hence the pressure difference across it. The
pressure microphone operates in a similar manner, except that one of the holes in the closed tube
is very much smaller than the other, so that for incident pressure signals at frequencies above
about 10 Hz, it appears effectively closed. The optical energy density sensor is illustrated in
Figure 14, where it can be seen that it is a much more compact transducer than that illustrated
26
in Figure 13. It has a dynamic range of 80 dB and covers a frequency range from 10 Hz to 4 kHz
with less than 1% harmonic distortion at 94 dB SPL and less than 5% THD at 120 dB (1 kHz).
Energy density sensing results in smaller noise reductions at the error sensors but a much
smaller pressure gradient in the sound field near the error sensor. It also results in an extension
B. Virtual Sensing
The use of pressure measurements at one or more locations to estimate the sound pressure
at another “virtual” location is known as virtual sensing and there are several schemes that have
been used in the past for determining the estimated pressure at the virtual location. The first
scheme15, 16, illustrated in Figure 15, involves placing a microphone at the virtual location and
measuring the transfer function between that microphone and the physical microphone that will
be used for the future pressure estimates at the virtual location. The transfer function is then used
in the control algorithm to drive the control source signal necessary to produce a minimum
pressure at the virtual location. A disadvantage of this approach is that the primary sound field
contribution is assumed to be the same at the actual microphone location and the virtual location.
This is approximately true when the distance between the primary source and actual microphone
location is large compared to the distance between the actual and virtual microphone locations
and the latter locations are in the far field of the primary source. One advantage of this approach
is that it is robust in terms of primary source location. That is, it will give similar results
regardless of where the primary noise is coming from. More recently Popovich17, Roure &
Albarrazin18 and Friot et al.19 modified the previous approach slightly to take into account the
difference in primary sound pressures at the virtual and actual microphone locations and labelled
27
it the remote microphone approach. The method is essentially illustrated in Figure 16.
Unfortunately, none of the previous approaches are easily amenable to being able to compensate
for head movement, which effectively requires continual changing of the virtual location.
An alternative way of estimating the pressure at the virtual location20-23 is to use linear or
a 1-D sensor and 6-9 for a 3-D sensor). This approach is shown in Figure 17 for a 1-D system,
which illustrates both linear and quadratic extrapolation. Such a system can be used to track head
movement as the head location, xv is a variable easily entered into the governing equations as
Perhaps the most successful approach used so far is that suggested by Cazzolato24 which is
a slight variation of the remote microphone technique suggested by Roure and Albarrazin18 and
very similar to the technique for harmonic noise reported by Gawron ans Schaaf25 in an obscure
conference paper in 1992. This technique involves using an adaptive LMS algorithm to adjust
the weights applied to a number of microphone signals, which when combined, provide an
accurate estimate of the sound pressure at the virtual location (see Figure 18).
Some promising work has also been completed on 1-D virtual energy density sensing20. Work
is continuing on the development of 3-D pressure and energy density sensors using the adaptive
scheme described above to estimate the sound pressure at the virtual location. One interesting
discovery is that when a sufficient number of sensors completely surround the virtual location,
excellent results can be obtained in terms of the estimated sound pressure at the virtual location.
At the time of writing, it is not clear whether multiple pressure sensors or multiple energy
density sensors will provide the best results. Preliminary experiments with a mining equipment
cabin26 showed that for broadband excitation it was difficult to distinguish between the overall
performance obtained using pressure sensing as compared to energy density sensing. However,
28
energy density sensing resulted in a smoother controlled sound pressure spectrum for which the
peaks and troughs in the spectrum were smaller than those obtained with pressure sensing.
Other work currently in progress involves the use of structural vibration sensors to estimate
the sound pressure at a particular virtual location when the source of the noise is a vibrating
structure.
It is relatively well known that planar speaker arrays can be used to generate, focus and steer
sound fields. For active control of sound in enclosures, they allow for the possibility of local
sound cancellation near a person’s head and the possibility of steering the cancelling sound to
ensure it continues to envelop the head. The required size of the array for effective operation is
about 2 wavelengths square, which equates to about 3.4 metres at 200 Hz. Clearly this may not
be practical in small enclosures such as mining vehicle cabins. However, if the speakers are
placed in the ceiling and walls of the cabin in a non-planar array, it will still be possible to focus
and steer the cancelling sound with the advantage that a smaller array size will be effective (see
figure 19). To minimise the effects of side lobes it is necessary to use “amplitude shading” so
that speakers near the edges of the array contribute less to the overall sound field than speakers
near the centre of the array - much like using a windowing function when undertaking an Fourier
29
The audio spotlight27-30 is a relatively new invention that uses airborne ultrasonics and the
non-linearity of the air motion generated by the propagating wave to generate sound in the audio
frequency range. The frequency content of the sound wave can be tailored by appropriate
modulation of the ultrasonic wave, which is achieved by passing it through an electronic control
box. The most interesting part is that the sound generated in the audio frequency range is
contained within the confines of the ultrasonic beam and thus shares the same spreading
properties as the ultrasonic beam. Due the high frequency of the ultrasound and corresponding
short wavelength compared to the size (approximately 300 mm) of the transducer generating it,
the beam spreads out at an angle of approximately 3 degrees, allowing for a highly directional
The audio spotlight has similar properties to an end fired speaker array shown in Figure 20,
with the audio sound level gradually increasing until it reaches the “Rayleigh distance” which
is typically 1.5 m from the transducer array for the two commercially available systems
illustrated in Figure 21. The one on the right has had some of the black covering removed to
The measured directivity of an acoustic spotlight for a 500 Hz audio tone and a 48 kHz
Unfortunately, it may be shown (see the following table calculated using the analysis of
Berktay28) that the level of ultrasonic sound pressure needs to be tens of decibels above the audio
sound pressure level that is to be generated and this clearly would be dangerous if reasonably
high industrial noise levels are to be controlled. The other problem is that as ultrasonic levels
become higher, more of the ultrasonic energy is spent heating the air and eventually a point is
reached at which no increase can be achieved in the intensity of the audio field. The level at
which this occurs is not clear at this time but from the following table, it can be seen that
30
excessive ultrasonic levels are needed even to achieve an 80 dB noise level at low frequencies.
Thus this approach to local cancellation is unlikely to be practical for industrial applications,
One of the problems associated with an audio spotlight is the harmonic distortion at low
frequencies. This is illustrated in Figure 23 where both ultrasonic and audio levels are shown as
a function of axial distance from the transducer for a 500 Hz audio signal, where it can be seen
that the measured ultrasonic levels are slightly lower than predicted in Table 1.
E. Steering the Virtual Sensor Location and the Cancelling Sound Field Focal
Point
Whether the sound pressure sensor at the observer’s ear is virtual or real, if the observer’s
ear moves at all, there will be a diminution in the noise reduction achieved and if the ear moves
back and forth, the perception will be a rapidly varying noise which would be very annoying.
Thus, it is necessary to track the observer’s head position and steer the control source beam and
virtual sensor location to keep track. If a physical sensor array is used, the steering of the virtual
sensor location could be achieved with look up tables and a smoothing algorithm to avoid sudden
changes in control parameters. The steering of the controlling sound field, which is generated
31
In order to know where the sound field is to be focussed and where the virtual sensor is to
be located, it is necessary to be able to track the location of the head around which the sound
field is to be reduced. This could be done using three ultrasonic transmitter/receivers embedded
in he cabin ceiling and some sophisticated software that excludes reflections from objects other
than the head or alternatively, a transmitter could be worn by the operator. In practice, a
combination of the two techniques will most likely be the best. The transmitter could be in a
head rest and the ultrasonic transmitter/receivers could be limited in how far from the headrest
transmitter they would track the head. Work using this approach is currently in progress.
6. CONCLUSIONS
Current successful applications of active noise control systems in industry outside of the
defence sector are essentially limited to the control of sound propagating in ducts and principally
to the control of plane wave propagation only. By use of an example, the development effort and
complexities associated with developing and installing a multi-modal active noise control system
in an industrial environment were discussed. It is clear that if active noise control systems are
to be widely used in industry, there needs to be quantum advances made in the following areas:
7. ACKNOWLEDGEMENTS
32
In assembling the material for this paper, the author gratefully acknowledges assistance from
his colleagues, Dr Anthony Zander, Dr Xun Li, Dr Damien Leclercq, Mr George Vokalek, Dr
8. REFERENCES
1.
C.H. Hansen, “Active noise control - from laboratory to industrial implementation,” Proc.
auxiliary path,” IEEE Trans. Acoust., Speech and Sig. Proc., ASSP-28, 454-467 (1980).
9.
C.R. Fuller and J. Carneal, “A biologically inspired control approach for distributed elastic
33
10.
C.R. Fuller, M.R.F. Kidner, X. Li. and C.H. Hansen, “Active-passive heterogeneous blankets
for control of vibration and sound radiation”, Proc. Active 2004, (2004), paper A04-91.
11.
A.J. Bullmore, P.A. Nelson, A.R.D. Curtis and S.J. Elliott, “The active minimization of
harmonic enclosed sound fields, Part II: Computer simulation,” J. Sound Vib. 117, 15-33
(1987).
12.
S.J. Elliott, A.R.D. Curtis, A.J. Bullmore and P.A. Nelson, “The active minimization of
harmonic enclosed sound fields, Part III: Experimental verification,” J. Sound Vib. 117, 35-
58 (1987).
13.
P.A. Nelson, A.R.D. Curtis, S.J. Elliott and A.J. Bullmore, “The active minimization of
harmonic enclosed sound fields, Part I: Theory,” J. Sound Vib. 117, 1–13 (1987).
14.
S. Sommerfeldt and P. Nashif, “A comparison of control strategies for minimising the sound
Proc. 1st International Conference on Motion and Vibration Control, (1992), pp. 1027-1031.
16.
J. Garcia-Bonito, S.J. Elliott and C.C. Boucher, “Generation of zones of quiet using a virtual
(1997).
18.
A. Roure and A. Albarrazin, “The remote microphone technique for active noise control,”
34
21.
C.D. Kestell, C.H. Hansen and B.S. Cazzolato, “Virtual sensors in active noise control,”
115.
25.
H.J. Gawron and K. Schaaf, “Active cancellation of harmonics using virtual microphones,”
Proc. 2nd International Conference on Vehicle Comfort: Ergonomic, Vibrational, Noise and
energy density in a mining vehicle cabin,” Proc. 10th International Congress on Sound and
568 (1975).
28.
H.O. Berktay, “Possible exploitation of nonlinear acoustics in underwater transmitting
35
Table 1. Ultrasonic noise levels (dB re 20 µPa) required to produce 80 dB and 100 dB,
respectively, at the face of the transducer (Courtesy Laura Brooks). fc is the ultrasonic carrier
frequency and the audible frequency is shown in the second row of the table.
Audible 80 dB 100 dB
SPL
Figure 7. Spray dryer system active noise controller front panel layout.
Figure 8. Typical spectra in the community with and without active control switched on.
Figure 10. Multi-rate filtering controller (5 error inputs and 4 control outputs), which allows
the extension of performance to achieve significant noise reductions over a number of octave
bands. Only the details for one octave band are shown - all other bands would be represented
in an identical way and for each output channel, the outputs from each band are combined
prior to being sent to the actuator.
Figure 11. Illustration of controller delays for various real and down sampled rates. The down
sampled rate is shown on the abscissa and the digital divide number is shown parametrically.
Figure 12. Acceleration response of a bare cantilever beam, a beam with a thick layer of foam
containing five embedded, non-driven inertial actuators, with the actuators driven using five
single channel controllers and then driven with a 5-out.6-in fully coupled (MIMO) controller.
Figure 15. Illustration of Elliott and David’s (1992) virtual microphone. The subscript, p/v,
represents the pressure at the virtual microphone due to the primary source, p/a represents the
pressure at the actual microphone due to the primary source. Hat symbols represent estimated
quantities and the S quantities represent transfer functions in the frequency domain between
the controller output and the error sensor input for the actual microphone (subscript a) and
the virtual microphone (subscript v).
Figure 16. Illustration of Roure and Albarrazin’s (1999) virtual microphone. The difference
between this figure and the preceding one is the transfer function H, which affects the
equations as well.
Figure 17. Illustration of linear and quadratic extrapolation from the physical microphones,
p1, p2 and p3 to the virtual location, pv.
Figure 20. Parametric array, showing how it behaves as a virtual end fired speaker array.
Figure 22. Measured directivity of acoustic spotlight at 500 Hz (Courtesy Laura Brooks).
Weld Stainless
Nylon
Electret steel
Holder
Microphone mount
Aluminum
Cooling Air foil
Duct
Vyon
attenuator
Acoustic
foam
Clamp O rings Air
Flow
4,9
40°
1,6
60° 60°
2,7,11 3,8,12
EL 33500
C C
Sect BB Sect DD
EL 32700
D D
30°
EL 32300
BACKPLANE
RS232
13 CHANNELS
TOTAL
Figure 7. Spray dryer system active noise controller front panel layout.
50
35
30
25
20
15
10
100 150 200 250 300 350 400
F re que nc y ( H z)
Cancellation path
transfer function
Digital
filter Σ modeller
weight update
coefficients
20 models
Adaptive
algorithm
Figure 10. Multi-rate filtering controller (5 error inputs and 4 control outputs), which
allows the extension of performance to achieve significant noise reductions over a number
of octave bands. Only the details for one octave band are shown - all other bands would be
represented in an identical way and for each output channel, the outputs from each band
are combined prior to being sent to the actuator.
0
100
at 55 Hz
at 25 Hz
at 110 Hz
time delay (ms)
10 at 200 Hz
Figure 12. Illustration of controller delays for various real and down
sampled rates. The down sampled rate is shown on the abscissa and the
digital divide number is shown parametrically.
Figure 13. 4-microphone energy
density sensor.
Figure 14. Optical, energy density sensor
manufactured by Phon-Or Ltd.
physical virtual
control actuator microphone microphone
signal
xv
qc
pa pv
pˆ c / a + pˆ p / a
Ŝ a ∑
pˆ p / v = pˆ p / a
pˆ c / v
1
+
Ŝ v ∑ +
pˆ p / v
p̂v
pa = p p / a + S a qc pˆ v = pˆ p / a + Sˆv qc
pv = p p / v + S v q c
pp/v = pp/a ⇒ = pa − Sˆa qc + Sˆv qc
Figure 15. Illustration of Elliott and David’s (1992) virtual microphone. The
subscript, p/v, represents the pressure at the virtual microphone due to the
primary source, p/a represents the pressure at the actual microphone due to
the primary source. Hat symbols represent estimated quantities and the S
quantities represent transfer functions in the frequency domain between the
controller output and the error sensor input for the actual microphone
(subscript a) and the virtual microphone (subscript v).
physical virtual
control actuator microphone microphone
signal
xv
qc
pa pv
pˆ c / a + pˆ p / a
Ŝ a ∑
H pp/v ≠ pp/a
pˆ c / v
+
Ŝ v ∑ +
pˆ p / v
p̂v
p a = p p / a + S a qc pˆ v = Hˆ pˆ p / a + Sˆv q s
p p / v = Hˆ p p / a ⇒
pv = p p / v + S v q c
( )
= Hˆ p a − Sˆa q s + Sˆv q s
p1 p1
pv
2h xv h h xv
Figure 17. Illustration of linear and quadratic extrapolation from the physical
microphones, p1, p2 and p3 to the virtual location, pv.
physical
microphone virtual
control actuator array microphone
signal
qc
h h xv
p3a p1a pv
p2a
e = pv − pˆ v ⇒ min [e 2 ]
-10 dB
-20 dB
-30 dB
110
100
Fundamental
90 audible
80
70
First harmonic
60
50
0 0.5 1 1.5 2 2.5 3 3.5
Distance from sound source (m)