Field of the Invention
[0001] The invention relates to the area of active noise and vibration control. Specifically,
the invention relates to a feedforward active noise and vibration control system using
a reference input.
Background of the Invention
[0002] Active noise and vibration control (ANVC) systems generally utilize input or reference
sensors which provide signals indicative of the source of disturbance (noise or vibration),
error sensors which provide signals indicative of residual noise or vibration to be
canceled, and adaptive processing of the input and error signals in order to arrive
at output signals. These output signals drive output transducers thereby producing
antinoise or antivibration with the result of reducing the residuals at the error
sensors, thereby reducing the noise or vibration thereat. It is known that the Single-Input
Single-Output (hereinafter SISO) ANVC variant is a subset of the more general Multi-Input
Multi-Output (hereinafter MIMO) ANVC system. See IEEE paper, Vol. ASSP-35, No. 10,
Oct. 1987, by Elliott, Stothers, and Nelson, entitled "A Multiple Error LMS Algorithm
and Its Application to the Active Control of Sound and Vibration" for a description
of one MIMO system. It is desired to have a reference signal which is only indicative
of the source of disturbance to be canceled, if possible. However, in many real-world
applications, the input signal may be corrupted with background noise which is not
correlated with the disturbance to be canceled. A corrupted input signal is likely
to corrupt the output signals, as any noise contained therein may be passed on directly
to the output transducers (speakers, Active Vibration Absorbers (AVAs), or other actuators),
and may even be amplified. The presence of this background noise detracts from the
effectiveness of the ANVC system and contributes to increase the overall level of
noise detected by the error sensors.
[0003] Therefore, there is a recognized need for a method of reducing the background noise
present in input signals, i.e., a method of increasing the signal-to-noise ratio.
In particular, there is a dire need for such a method for application to tonal ANVC
systems where a reference signal contains one or more dominant tones that are being
controlled, yet the input signal including the tones, also includes unwanted broadband
noise.
[0004] Adaptive line enhancement (ALE) methods are known for separating a signal into a
periodic component signal and a random component signal. One known example application
of ALE techniques is for canceling the maternal heartbeat in fetal electrocardiography
as described in
Adaptive Signal Processing, pp. 334-337, by Widrow and Stearns, 1985. Another known example application of ALE
techniques is for removing periodic interference from training noise in an ANVC system
identification process as described in
Investigations in Active Line Enhancer Techniques by Shawn Steenhagen, MS Thesis, Univ. of Wisconsin, Madison, 1993.
[0005] Prior Art Fig. 1 describes one conventional prior art ALE implementation. The conventional Finite
Impulse Response (FIR) ALE
20a processes an ALE input signal
s+n, 21a, containing the periodic component signal
s plus the random component signal
n with the objective of separating the periodic component signal from the random component
signal. The ALE input signal
s+n is delayed Δ>0 samples by the delay operator
28a to produce the delayed input signal
29a. The delayed input signal is processed by an FIR filter
25a to produce an estimate
s' for the periodic component signal
22a. The periodic component signal estimate is subtracted from the input signal by the
error summation block
24a to produce an estimate
n' for the random component signal
23a. One or both of these component estimates may be used depending on the requirements
of the application. The random component signal
23a is also interpreted as an ALE error signal
ek, given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0001)
where
xk is the ALE input
21a, and
yk is the adaptive FIR filter output
22a given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0002)
The ALE input signal
21a and the broadband component estimate signal
n' are processed by an adaptive update block
26a, using a gradient descent method such as LMS
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0003)
to provide updated filter coefficients
27a to the FIR filter
25a.
[0006] Prior Art Fig. 2 describes a second conventional prior art ALE implementation. The conventional Infinite
Impulse Response (IIR) ALE
30b processes an ALE input signal
s+n, 21b, containing the periodic component signal
s plus the random component signal
n with the objective of separating the periodic component signal from the random component
signal. The ALE input signal
s+n is delayed Δ>0 samples by the delay operator
35b to produce the delayed input signal
36b. The delayed input signal is processed by an FIR A-filter
Ak, 31b, to produce an output
39b. The estimate
s' for the periodic component signal
22b is processed by an FIR B-filter
Bk, 32b, to produce an output
40b. The A-filter output is combined with the B-filter output in the path summation block
41b to generate the periodic component signal estimate
22b. The periodic signal component estimate is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0004)
The periodic component signal estimate is subtracted from the ALE input signal by
the error summation block
24b to produce an estimate
n' for the broadband component signal
23b. One or both of these component estimates may be used depending on the requirements
of the application. The ALE input signal
21b and the broadband component estimate signal
n' are convolved by an A-filter correlator block
33b which is used in the LMS gradient descent method to provide A-filter adjustments
34b to the FIR A-filter
31b. The periodic component signal estimate
s' and the broadband component estimate signal
n' are convolved by a B-filter correlator block
37b which is used in the LMS gradient descent method to provide B-filter adjustments
38b to the FIR B-filter
32b. These filter coefficient updates are given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0005)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0006)
when the LMS algorithm is used. However, none of the above address the need for an
ANVC system with an enhanced reference signal.
Summary of the Invention
[0007] Therefore, in light of the problems associated with prior art ANVC systems identified
above, the present invention, in one aspect thereof, is directed to a method and means
for reducing unwanted background noise that is present in a reference (input) signal
in a feedforward-type ANVC system. In particular, the present invention is directed
to adaptive gradient descent means for reducing the noise present in the reference
(input) signal of an ANVC system. More specifically, an Adaptive Line Enhancer (ALE)
is preferably placed in the input path of the ANVC system. The adaptive means, such
as an ALE, for enhancing the reference signal can reduce broadband noise from the
input signal if there are one or more dominant tones therein to be controlled, or
conversely, tones may be reduced if a broadband input is required for the input to
the ANVC system.
[0008] The ALE may include IIR, FIR, parametric or adaptive inverse implementations, or
the like. The ALEs may be adaptively controlled via any of the known gradient descent
algorithms. The use of ALEs in the input path is beneficial to any of the known ANVC
systems, including Active Noise Control (ANC), Active Structural Control (ASC), and
Active Isolation Control (AIC). Further, the ALE enhanced reference signal described
herein can be used with any of the known ANV control processes, such as those including
FIR and IIR adaptive filters. In another aspect, multiple ALEs may be used in parallel
or cascaded relationship to provide, for example, a reference signal with contributions
from multiple tones. Further, the enhanced reference signal provided by the ALE, besides
being used as an input to the control process, may be used as an input to auxiliary
components, such as an Engine Vibration Monitor (EVM) which is used for monitoring,
and/or, displaying the vibration of aircraft engines.
[0009] The main advantage of the invention is that it provides a reference (input) signal
to the ANVC system that has substantially higher Signal-To-Noise (SNR) ratio.
[0010] It is another advantage of the invention that the power requirements to drive the
output transducers can be reduced.
[0011] It is another advantage of the invention that it may reduce the computational requirements
for the adaptive control of the ANVC system, for example, the number of taps in a
multi-tap control algorithm may be reduced.
[0012] It is another advantage of the invention that it may enhance the performance and
convergence of the ANVC system.
[0013] The abovementioned and further novel aspects, features and advantages of the invention
will be apparent from the accompanying descriptions of the preferred embodiments and
attached drawings.
Brief Description of the Drawings
[0014] The accompanying drawings which form a part of the specification, illustrate several
key embodiments of the present invention. The drawings and description together, serve
to fully explain the invention. In the drawings,
Fig. 1 is a schematic diagram of a prior art FIR ALE,
Fig. 2 is a schematic diagram of a prior art IIR ALE,
Fig. 3 is a block schematic view of an Active Noise Control (ANC) system using a microphone
input sensor, an ALE for reducing noise in the reference signal which is uncorrelated
with the disturbance to be canceled and a loudspeaker as an output transducer,
Fig. 4 is a schematic view of another Active Noise Control (ANC) system using an accelerometer
input sensor, an ALE in the input path, and a loudspeaker as the output transducer,
Fig. 5 is a schematic view of an Active Structural Control (ASC) system using an accelerometer
input sensor, an ALE on the input, and an Active Vibration Absorber (AVA) as the output
transducer,
Fig. 6 is a schematic view of an Active Isolation Control (AIC) system using an accelerometer
input sensor, an ALE in the input path, and an active mounting as the output transducer,
Fig. 7 is a schematic block diagram of one embodiment of the present invention ANVC system
including a FIR ALE in the input path and an FIR process control for tonal control,
Fig. 8 is a schematic block diagram of another embodiment of the present invention ANVC
system including a IIR ALE in the input path and an FIR process control for tonal
control,
Fig. 9 is a schematic block diagram of another embodiment of the present invention ANVC
system including a IIR ALE in the input path and an II R process control for tonal
control,
Fig. 10 is a schematic block diagram of a parametric ALE,
Fig. 11 is a schematic block diagram of an adaptive inverse ALE,
Fig. 12 is a schematic block diagram of another embodiment of the present invention ANVC
system including a parametric ALE in the input path and an FIR process control for
tonal control,
Fig. 13 is a schematic block diagram of another embodiment of the present invention ANVC
system including a adaptive inverse ALE in the input path and an FIR process control
for tonal control,
Fig. 14 is a schematic block diagram of another embodiment of the present invention ANVC
system including a FIR ALE and FIR process control for broadband control,
Fig. 15 is a schematic block diagram of another embodiment of the present invention ANVC
system including multiple Low-Order IIR ALEs,
Fig. 16 is a schematic block diagram of another embodiment of the present invention ANVC
system including a High-Order FIR ALE,
Fig. 17 is a schematic block diagram of another embodiment of the present invention ANVC
system including multiple cascaded ALEs,
Fig. 18 is a schematic block diagram of another embodiment of the present invention ANVC
system including using the ALEs output for an input to an auxiliary component,
Fig. 19 is a figure illustrating a reference signal having multiple tones and broadband background
noise, and
Fig. 20 is a figure illustrating an output signal from the Band Pass Filter (BPF) and also
from the ALE within the input path showing the resulting further reduction in unwanted
broadband background noise in the reference signal.
Detailed Description of the Preferred Embodiments
[0015] With reference to the figures herein, where like reference characters are employed
where possible to indicate like parts, there is shown in
Fig. 3, an ANVC system, and in particular, an ANC system
50c comprising an input sensor
52c for providing a reference signal indicative of the disturbance acoustic noise or
vibration causing the disturbing acoustic noise. In the ANC case, the disturbance
may be an acoustic noise emanating from a noise source, such as an aircraft engine
or the like. In most applications, the reference signal generally will include some
unwanted noise therein. By the term noise, what is referred to is background noise
which is uncorrelated with the disturbance which is sought to be canceled. In the
ANC system
50c, adaptive means are provided for reducing the noise present in the reference signal.
[0016] In particular, an Adaptive Line Enhancer (ALE)
54c, which includes an adaptive filter and update means for updating the coefficients
or weights of the adaptive filter, is preferably is used. An error sensor
62c is provided for generating an error signal indicative of the residual acoustic noise
at the point adjacent where a quiet zone is desired. The means for processing the
error signal and the reference signal and producing an output signal is provided in
the control process
58c, in this case, ANC control, which includes a filter taking the form of a IIR or FIR
filter structure with adaptive feedforward control. An output transducer
60c, in this case, a loudspeaker, is dynamically driven responsive to the output signal.
The output transducer
60c produces antinoise which preferably minimizes the noise at the point of interest
to produce a quiet zone. The controller
56c includes both the adaptive means, such as ALE
54c, for reducing the unwanted noise on the input signal and the control process
58c therewithin. It should be understood that the reference enhancement means is based
upon an adaptive gradient descent method and is generally implemented within the software.
It should also be understood that there may be an optional filtering/conditioning
step before the reference signal is provided to the ALE
54c.
[0017] Fig. 4 represents another ANC system
50d which is identical to the system of
Fig. 3 except that the reference sensor is an accelerometer
52d. One ANC system for which the ALE used within the input path as described herein
may be useful is discussed in commonly assigned US Application Serial No. 08/553,227
to G. Billoud entitled "Active Noise Control System for Closed Spaces Such As Aircraft
Cabins" filed Sept. 25, 1995. Other ANC systems are described in US Pat. No. 4,562,589
to Warnaka et al. entitled "Active Attenuation of Noise in a Closed Structure" and
US Pat. No. 4,473,906 to Warnaka et al. entitled "Active Acoustic Attenuator."
[0018] Fig. 5 represents another ANVC system, and in particular, an ASC system
50e which is identical to the system of
Fig. 4 except that the output transducer is an AVA
60e. Active systems including AVAs are described in PCT Patent Application Serial No.
PCT/US95/13610 entitled "Active Systems and Devices Including Active Vibration Absorbers
(AVAs)" and US Pat. No. 4,715,559 to Fuller entitled "Apparatus and Method For Global
Noise Reduction." In aircraft ASC systems, AVAs are attached directly to the interior
surface of the aircraft's fuselage and dynamically shake the fuselage wall to generate
canceling noise in the aircraft's cabin.
[0019] Fig. 6 represents another ANVC system, and in particular, an AIC system
50f which is identical to the system of
Fig. 5 except that the output transducer is an active mount
60f. Active mounts are taught in commonly assigned US Pat. No. 5,174,552 to Hodgson et
al. entitled "Fluid Mount with Active Control" and US Pat. App. Serial No. 08/260,945
entitled "Active Mounts for Aircraft Engines." Further descriptions may be found in
a Lord paper entitled "Frequency-Shaped Control of Active Isolators" by D. A. Hodgson.
Active mounts
60f attach between an engine and the structure the engine is attached to and are dynamically
actuated (driven) to control vibration therebetween or noise at a remote location.
[0020] Fig. 7 represents a detailed block diagram of a tonal ANVC system
50g including a reference sensor
52g, an optional band pass filter
53g, an ALE
54g, and output transducer
60g, and error sensor
62g and an adaptive control process
58g which may include system identification, hereinafter referred to as ID
42g. The ID
42g may be accomplished in an on-line or off-line fashion. Further, filtering or other
signal conditioning is commonly used on the output signal and error signal paths,
however, they are not shown for the sake of clarity in all figures described herein.
The ALE
54g herein includes a FIR filter structure as is described fully with reference to
Fig. 1. The reference signal
51g from reference sensor
52g is preferably band-pass filtered and received at the ALE input
21g. The output of the ALE
22g is provided to the control process
58g for the ANVC system
50g. In this embodiment, the ALE output
22g is comprised of the periodic component of the reference signal
51g, i.e., the one or more tones that are indicative of the noise or vibration
49g generated by the source of disturbance
48g.
[0021] In this embodiment, the control process
58g is achieved by an FIR filter
59g including update means, such as filtered-x LMS, or the like. A further discussion
of a Single-Input Single-Output (SISO) Filtered-x LMS control with system ID can be
found in "Adaptive Control of a Two-Stage Vibration Isolator Mount" by S. D. Sommerfeldt
and J. Tichy, Journal of the Acoustical Society of America, Vol. 88, No. 2, pp. 938-944,
1990. A further discussion of Multiple-Input Multiple-Output (MIMO) systems including
FIR filters is described in US Pat. No. 5,170,433 to Elliott et al. entitled "Active
Vibration Control." In General, the ALE output
22g is filtered through an error path model
63g (sometimes referred to as the X filter) representative of the transfer function between
each output transducer
60g and error sensor
62g pair. The filtered ALE output is referred to as the filtered regressor
65g. Error information in line
64g represents the error signal
55g and any filtered training noise from training block
42g. Error information
64g and the filtered regressor
65g are inputted to an update method and means
61g, such as Filtered-x LMS, for determining the new filter weights to be passed to the
FIR control filter
59g. The output
22g from the ALE
54g is filtered by the control filter
59g to arrive at the output signal
57g used to drive the output transducer
60g, e.g. a loudspeaker, AVA, active mount or the like. It should be understood that
the use of the ALE
54g in the input path provides a cleaner reference signal to the control filter
59g, therefore, the output signal
57g to the output transducer
60g is also cleaner and the ANVC system
50g will do a better job at cancellation of the noise or vibration.
[0022] Fig. 8 represents another tonal feedforward ANVC system
50h including a reference sensor
52h, an adaptive gradient descent means for reducing the uncorrelated noise present in
the reference signal
51h, such as an ALE
54h, an adaptive control process
58h including an FIR filter
59h which may include system identification
42h, an output transducer
60h, and an error sensor
62h. The ALE
54h used herein is fully described with reference to
Fig. 2 and represents a IIR filter ALE. Again, the output
22h from the IIR ALE
54h is used as an input to the control process
58h which is identical to that described with reference to
Fig. 7.
[0023] Fig. 9 represents another tonal feedforward ANVC system
50j including identical elements as described with reference to
Fig. 8 except that the control process
58j includes a IIR filter instead of a FIR filter. Therefore, the ANVC system
50j is a combination of a IIR control filter process
58j and a IIR ALE
54j. Control process
58j may include system ID
42j which may be implemented in an on-line or off-line fashion. IIR control filter structures
are described in US Pat. No. 4,677,676 to Eriksson entitled "Active Attenuation System
with On-Line Modeling of Speaker, Error Path, and Feedback Path" and US Pat. No. 4,677,677
to Eriksson entitled "Active Sound Attenuation System with On-Line Adaptive Feedback
Cancellation."
[0024] Fig. 10 is a schematic showing one embodiment of a parametric IIR ALE. In this configuration,
the adaptive filter is a IIR filter similar to the prior art
Fig. 2; however, the parametric IIR ALE described herein does not require a delay operation
as in
35b, and the filter coefficient update process is greatly simplified with a constrained
adaptation process due to the parametization. The parametric IIR ALE
69m processes an ALE input signal
s+n, 21m, containing the periodic component signal
s plus the random component signal
n with the objective of separating the periodic component signal from the random component
signal. The ALE input signal is processed by an FIR A-filter
A(ωk),
31m, to produce an output
39m. The estimate s' for the periodic component signal
22m is processed by a FIR B-filter
B(ωk),
32m, to produce an output
40m. The A-filter output is combined with the B-filter output in the path summation block
41m to generate the periodic component signal estimate
22m. The periodic component signal estimate is subtracted from the ALE input signal by
the error summation block
24m to produce an estimate
n' for the broadband component signal
23m. One or both of these component estimates may be used depending on the requirements
of the application.
[0025] Each of the IIR filter coefficients in
A(ωk) and
B(ωh) are explicitly parametized by the center frequency ωk. This parametization is
accomplished by first selecting a desired frequency response for the IIR filter. A
preferred response is a second-order band-pass filter which is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0007)
The filter bandwidth (
BW) and the sharpness of resonance (
Q) are given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0008)
where ξ is the damping ratio, and ω
o is the center frequency. Using the pre-warped bilinear transform, the three non-zero
digital filter coefficients may be analytically expressed in terms of the center frequency
and bandwidth. This parametization is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0009)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0010)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0011)
where
T is the sample period, and ωk is the time varying center frequency. For a constant
bandwidth, the digital filter coefficients are parametized by center frequency only.
[0026] There are many methods available for adaptively updating the center frequency. One
preferred approach is to use an LMS gradient descent method which minimizes the instantaneous
squared ALE error
ek, represented by
23m, which is equivalent to
n'. The ALE error is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0012)
where
xk is the ALE input
21m, and
yk is the adaptive IIR filter output
22m. For the second-order band-pass filter response, the adaptive IIR filter output is
given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0013)
Taking the derivative of the instantaneous squared ALE error with respect to center
frequency ωk
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0014)
the update expression for the center frequency becomes
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0015)
where µ is a constant step size which controls the rate of convergence. The derivatives
are also analytically available for the second-order band-pass filter response, and
are given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0016)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0017)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0018)
[0027] The parametric FIR filter coefficients
31m and
32m are updated with each new estimate of the center frequency
77m. The center frequency is updated by the center frequency update
76m wherein the gradient estimate
75m is added to the previous center frequency value. The gradient estimate is the output
of a gradient summation block
74m which sums the A-filter gradient contribution
71m and the B-filter gradient contribution
73m. The ALE input signal is convolved with
n' in the A-filter correlator block
33m which produces an A-filter convolution vector
78m. The A-filter convolution vector is multiplied by the gradient of the A-filter coefficients
with respect to center frequency, in the A-filter parametric gradient product block
70m, which produces the output A-filter gradient contribution
71m. The periodic component signal estimate
s' is convolved with
n' in the B-filter correlator block
37m which produces a B-filter convolution vector
79m. The B-filter convolution vector is multiplied by the gradient of the B-filter coefficients
with respect to center frequency, in the B-filter parametric gradient product block
72m, which produces the output B-filter gradient contribution
73m.
[0028] The parametric IIR ALE may be further simplified in several ways. One improvement
to the invention is to restrict the operational range of the center frequency, which
is easily accomplished when the center frequency is adapted explicitly. The restriction
is accomplished by not allowing the adaptation process to select a center frequency
outside of some prescribed range. Upon inspection of the partial derivatives of the
filter coefficients with respect to center frequency, it may be observed that some
of the derivatives are several orders of magnitude smaller than the others. These
small derivatives may be assumed zero and the corresponding parameter values may be
assumed constant over the restricted frequency range. The transcendental parametric
expressions for the coefficients whose derivatives are not negligible may also be
preferably replaced with polynomial curve fit, or other expressions which are simpler
to evaluate in real time.
[0029] Fig. 11 is a schematic showing one embodiment for an adaptive inverse ALE. In this configuration,
the adaptive filter is an FIR filter similar to the prior art
Fig. 1; however, the adaptive inverse ALE uses the adaptive filter coefficients in a novel
way for separating the periodic component signal from the random component signal.
The adaptive inverse ALE,
80p, processes an ALE input signal
21p where reference numerals
21p, 28p, 29p, 34p, 33p, 25p, and
24p are exactly as described in reference to
Fig. 1. Signal
81p is an auxiliary periodic component signal and signal
82p is an auxiliary random component signal which are only used in the adaptation of
the FIR A-filter
Ak.
[0030] The ALE input signal is further processed by a IIR filter which is constructed from
the modified FIR filter coefficients
Ak, and an additional filter Gk as described below. The ALE error signal ek is given
by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0019)
where
xk is the ALE input
21p, and
yk is the adaptive FIR filter output
81p. The preferred FIR A-filter is a two-coefficient filter whose output is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0020)
The LMS gradient descent update for each of these two coefficients is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0021)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0022)
In order to understand how the adaptive inverse ALE operates, we first look at an
example input signal which is a pure tone as represented in complex notation by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0023)
where
X is an arbitrary non-zero amplitude, ω
o is the radian frequency,
k is the sample index,
T is the sample period, and φ is an arbitrary phase angle. Substituting this result
into the expression for the ALE error signal defined above
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0024)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0025)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0026)
Under perfect convergence conditions, the ALE error signal will be zero and thus
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0027)
This equation may be solved for the ideal or optimal A-filter coefficients as
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0028)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0029)
[0031] It can be shown that the LMS gradient descent update described above will drive the
A-filter coefficients toward the optimal values given above. A further simplification
can be achieved by selecting Δ=1. The ALE input signal is input to a FIR feedforward
filter
Gk, 31p, which produces a G-filter output
83p. The periodic component signal estimate
22p is produced from the path summation
41p whose inputs are the G-filter output and the modified A-filter output
84p. In the preferred embodiment, the G-filter is a single coefficient which appropriately
scales the ALE input signal. The signal
s' is delayed by the delay block
28p' to produce the delayed output
29p' which is input to the modified A-filter
85p. The modified A-filter produces the modified A-filter output
84p. If the A-filter coefficients from
25p were used directly in
85p without modification, the resulting digital IIR filter would have undamped poles
at the frequency of the input excitation. In practice, damping is added to the poles.
In the preferred embodiment, the transfer function
G(z) of the resulting IIR filter is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0030)
and the filter output
rk, which is equivalent to
s', is given by
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0031)
where 0<<γ<1. The poles of the resulting IIR filter are
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0032)
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0033)
and thus we see that the modification of scaling
A[1] by γ and scaling
A[2] by γ
2 preserves the center frequency while adding damping. The G-filter must generally
be selected such that the resulting IIR filter has the appropriate gain at the center
frequency. In the preferred embodiment, the gain is unity and the resulting scalar
G-filter coefficient must be
![](https://data.epo.org/publication-server/image?imagePath=1998/01/DOC/EPNWA2/EP97302679NWA2/imgb0034)
thus we see that the G-filter is parametized by center frequency.
[0032] The estimate
s' for the periodic component signal
22p is subtracted from the ALE input
21p in the error summation
24p' to produce an estimate for the random component signal
23p. One or both of these component estimates may be used depending on the requirements
of the application.
[0033] The adaptive inverse ALE may be further simplified in several ways. By choosing Δ=1,
A[2]=-1 it does not require adaptation, and
A[1]=2cos(ωo
T). Since
A[1] is parametrically related to the center frequency ωo, the adaptive update expression
may be written in terms of an adaptive center frequency from which A[1] is then computed
using the known parametric relationship. The scalar G-filter coefficient may then
easily be computed using its known parametric relationship. One improvement to this
aspect of the invention is to restrict the operational range of the center frequency,
which is easily accomplished when the center frequency is adapted explicitly. The
restriction is accomplished by not allowing the adaptation process to select a center
frequency outside of some prescribed range. When the operating frequency range is
restricted, the transcendental parametric expressions for A[1] and the G-filter coefficient
may be preferably replaced with polynomial curve fit or other expressions which are
simpler to evaluate in real time.
[0034] Fig. 12 represents another tonal feedforward ANVC system
50m including a reference sensor
52m, adaptive gradient descent means for reducing the uncorrelated noise present in the
reference signal
51m, such as ALE
54m, an adaptive control process
58m including a FIR filter
59m which may include system identification
42m, an output transducer
60m, and an error sensor
62m. The ALE 54m used herein receives its ALE input
21m in the form of the band-pass-filtered reference signal. The ALE output
22m which represents a more refined tonal signal, as compared to the reference signal
51m and the band-pass-filtered reference signal, is provided to the control process
58m. The ALE
54m is fully described with reference to
Fig. 10 and represents a parametric ALE. The simulated performance of this parametric embodiment
is illustrated in
Fig. 20. It should be understood that, although a FIR control process
58m is shown, a IIR control process as described with reference to
Fig. 9 could also be employed in combination with the parametric ALE
54m.
[0035] Fig. 13 represents another tonal feedforward ANVC system
50p identical to that described with reference to
Fig. 12 except that the ALE
54p is an adaptive inverse ALE. The ALE
54p used herein is fully described with reference to
Fig. 11 and represents an adaptive inverse FIR ALE. It should be understood that although
a FIR control process
58p is shown, a IIR control process as described with reference to
Fig. 9 could also be employed in combination with the adaptive inverse ALE
54p. As described with previous embodiments, the ALE includes its ALE input
21p and ALE output
22p within the reference path and is used to provide a refined tonal signal with the
broadband uncorrelated noise reduced.
[0036] Fig. 14 represents a broadband feedforward ANVC system
50r including an ALE
54r. The ANVC system is identical to that described with reference to
Fig. 7 except that the ALE
54r is used to reduce unwanted uncorrelated periodic noise (tones) in the reference signal
path and leave only the broadband noise, which is desired to be controlled. Since
broadband noise is desired to be reduced by the broadband control process, then any
uncorrelated tones present in the reference signal will detract from the effectiveness
of the broadband control. The ALE
54r used herein is fully described with reference to
Fig. 1 and represents a conventional FIR ALE. It should be understood that although a FIR
control process
58r and conventional FIR ALE
54r are shown, that other combinations are possible. For example, the output from the
conventional IIR ALE
30b in
Fig. 2 at
23b may be used as a broadband input to any ANV control such as the FIR ANV Control described
with reference to
Fig. 8. Similarly, the broadband output from the parametric ALE
69m described in
Fig. 10 may be the broadband input to an FIR control similar to that described with reference
to
Fig. 12. Finally, the broadband output from the adaptive inverse ALE
80p at
23p (Fig. 10) may be used as an input to an ANV control, such as the control process described
with reference to
Fig. 13.
[0037] Fig. 15 represents a tonal feedforward ANVC system
50t including multiple ALEs
54t and
54t' providing multiple signals to the ANV control. The ANVC system
50t generally represents a MIMO system for controlling multiple tones emanating from
the source
58t which produce unwanted noise or vibration at a point of interest, for example within
an aircraft cabin. In this embodiment, the reference sensor
52t picks up the disturbance signal containing therein the multiple tones to be controlled.
The signal is band-pass filtered to separate into two frequency ranges,
fr1 and
fr2, the signals containing the tones of interest. For example, BPF
53t passes the lower range of frequencies and BPF
53t' passes only the upper range of frequencies.
[0038] The ALEs
54t and
54t' in this embodiment are IIR ALEs as described with reference to
Fig. 8 and are arranged in a parallel relationship. The ALEs
54t and
54t' may also be low-order IIR, FIR, parametric, or adaptive inverse ALEs. Low-order implies
they each have only a low number of taps. ALE
54t passes a lower frequency tone, for example, a signal indicative of the
N1 engine rotation frequency of an aircraft engine to the ANV control
58t. Similarly, ALE
54t' passes a higher frequency tone, for example, a signal indicative of the
N2 engine rotation frequency of an aircraft engine to the ANV control
58t. The ANV control process
58t then drives output transducer(s)
60t responsive to the error signals from the error sensor(s)
62t to preferably cancel the unwanted tones present at the point of interest that are
caused by the source
48t. It should be understood that in the tonal case, the ALEs
54t and
54t' reject the uncorrelated broadband noise within each frequency range of operation
fr1 and
fr2 and thereby provide an enhanced signal representative of the tones of interest to
the ANV control
58t.
[0039] Fig. 16 represents a tonal feedforward ANVC system
50v including a high-order FIR ALE
54v providing a signal to the ANV control
58v. By the term high-order it is meant that the FIR filter therein has a high number
of taps.
[0040] Fig. 17 represents a tonal feedforward ANVC system
50w including a pair of cascaded ALEs
54w and
54w' providing reference signals to the ANV control
58w. In this embodiment, the first ALE
54w is used to enhance only the first tone of interest, while the second tone is enhanced
by the second ALE
54w'.
[0041] Fig. 18 represents a tonal feedforward ANVC system
50z including an ALE
54z for providing a clean signal to the ANV control
58z. In this embodiment, the output from the ALE
54z is also used as an input to an auxiliary component
90z. For example, the output from the ALE
54z may be used in an auxiliary component
90z such as an Engine Vibration Monitor (EVM) including a peak detect
92z, signal processing means
94z, and display means
96z.
[0042] Fig. 19 and
Fig. 20 represent the raw reference signal
51m from the reference sensor
52m and the filtered output from the BPF
53m, i.e., the ALE input
21m, and the tonal-enhanced ALE output
22m from the ALE
54m for the parametric embodiment of
Fig. 12. The raw reference signal
51m provided from the reference sensor
52m is indicative of a reference signal (vibration) that is picked up by an accelerometer
mounted on a jet engine. The engine has multiple vibrations, for example, at
N1 and
N2, which produce tonal noise within the aircraft's cabin. Present in the reference
signal
51m is unwanted broadband background and some periodic noise
66m. It is desired to have a separated reference signal that has contributions only at
the tones that are to be controlled. The cleaner (less broadband) the reference signal
51m the better, as any broadband noise present in the input to the FIR control filter
59m may be passed onto the output transducers
60m. The ALE input
21m represents the reference signal once it has been filtered through the band pass filter
53m.
[0043] In this case, the range of the band pass filter
53m is centered around the nominal of
N1 and rejects broadband noise below and above its operating range. The ALE
54m further enhances the input signal
21m and produces an ALE output signal
22m which is significantly more indicative of the tone of interest, in this case
N1. It should be understood that similar results may be obtained for the conventional
FIR ALE of
Fig. 1, the conventional IIR ALE of
Fig. 2 and the adaptive inverse ALE of
Fig. 11. Through the use of an ALE on the input path in a feedforward ANV control system,
similar results can be achieved if one desires to remove the tones and leave the broadband
noise as is described with reference to
Fig. 14. This may be desirable for a system where the input is from a road wheel in a vehicle
and broadband road noise is being controlled in the vehicle compartment or cabin.
[0044] In summary, the present invention is directed to a method and means for reducing
unwanted background noise present in a reference (input) signal of a feedforward-type
ANVC system. More specifically, an Adaptive Line Enhancer (ALE) is preferably placed
in the input path of the ANVC system and enhances the reference signal by reducing
broadband noise or tone(s) contained therein. The ALE may include IIR, FIR, parametric
or adaptive inverse implementations, or the like. Further, the use of ALEs in the
input path is beneficial to Active Noise Control (ANC), Active Structural Control
(ASC), and Active Isolation Control (AIC) systems. In another aspect, multiple ALEs
may be used in parallel or cascaded relationship and, further, the ALE output may
be used as an input to auxiliary components, such as Engine Vibration Monitors (EVMs).
EVMs are described in SAE paper 871732 entitled "The V-22 Vibration, Structural Life,
and Engine Diagnostic System, VSLED" by M. J. Augustin and J. D. Phillips.
[0045] While the preferred embodiments of the present invention have been described in detail,
various modifications, additions, alternatives, changes, and adaptations to the aforementioned
may be made without departing from the spirit and scope of the present invention defined
in the appended claims. It is intended that all such modifications, additions, alternatives,
changes, and adaptations be considered part of the present invention.