Professional Documents
Culture Documents
net/publication/241170662
Active noise control in a duct using output feedback robust control techniques
CITATIONS READS
6 163
3 authors:
F. Pourboghrat
Southern Illinois University Carbondale
57 PUBLICATIONS 647 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Resilient Decentralized Estimation and Control for Cooperative Rigid Body Multivehicle Systems View project
All content following this page was uploaded by F. Pourboghrat on 22 November 2014.
Abstract — An active noise control (ANC) system generally (microphone) response to disturbance is minimized by
uses speakers and microphones as inputs and outputs. The adjusting the gain of the controller (speaker) [13]. However,
ANC is effective in regions around the error microphones, noise control using speakers are effective only in small
known as the zones of quiet. In this paper, a linear matrix regions around the microphones, known as the zones of
inequality (LMI) based robust control method is used to quiet. In addition, the undesired noise may generally
enlarge the zones of quiet around the microphones. This is increase outside these regions. . This calls for a control
achieved by considering microphones to have fictitious algorithm which widens the effective zone of quiet around
uncertain locations around their nominal positions, and by the microphones.
designing a controller to guarantee robustness against all these
location-based uncertainties. As a result, noise reduction in a In the late 70’s, introduction of the robust control
wider region is achieved. The control is developed based on a techniques led to significant achievements in the area of
non-minimal state-space realization (NSSR) scheme that allows control synthesis. However, it is not until recent years that
one to generate an equivalent set of states for the ANC system, these control techniques have been considered for ANC
using only the measured inputs and outputs, without any applications [14]–[17]. Specifically, it is known that the H∞
differentiation. Actual hardware experiments using a digital control concentrates on making the system robust to all
signal processor (DSP) have been carried out to show the
bounded destabilizing disturbances affecting the system. In
applicability of the method.
this paper, the problem of enlarging the zones of quiet
around the microphones is handled by H∞ control method
I. INTRODUCTION
derived using linear matrix inequality (LMI) formulation.
In recent years, active noise control (ANC) has become an The LMI techniques offer tractable solutions to convex
important area of control research for reducing the effects of optimization problems that can be efficiently solved by
unwanted audio noise. This is due to the fact that passive various numerical algorithms [18]. We consider the error
noise cancellation (PNC) techniques are ineffective and microphones to have fictitious uncertain locations around
expensive when they deal with low-frequency noise [1]–[5]. their nominal positions and the proposed ANC method is
The methods reported in the literature for ANC can be designed such that the noise is reduced at all these uncertain
classified as feedforward and feedback methods. The basic locations. As a result, due to the robustness of the designed
principle of the feedforward method, patented by Leug in H∞ controller to these uncertainties, noise reduction can be
1936 [6], involve a search for control gains to reject achieved in a larger region, i.e., larger zones of quiet.
undesirable external disturbances. Feedforward ANC
approach, that can deal with both narrowband and broadband It is known that state-feedback control methods are
noise attenuation, still remains an active area of research [7], generally more robust than output-feedback control methods
[8]. However, the main disadvantages of feedforward [5] and that H∞ optimal control provides a robust control
solutions are their inability to deal with structural variations scheme. In order to increase the robustness of the output
and the requirements of at least two sensors (microphones); feedback H∞ controller, a continuous-time non-minimal
one at the disturbance source and one where the noise is to state-space realization (NSSR) scheme is employed as
be attenuated. Moreover, in a real-time experiment, it is not proposed in [5] and [19]. This NSSR technique allows one to
always possible to measure the signal directly from the generate an equivalent set of states for the system, using the
disturbance source [4], [5]. measured inputs and outputs. The method does not require
differentation, despite perturbations, uncertainties and even
Another method for active noise reduction uses feedback unknown parameters in the system model [4, 5, 19].
control schemes [4], [5], [9]–[11]. In 1953, an ANC system
using feedback control was developed by Olson and May This paper is organized as follows. Section 2 presents
[12]. This first practical feedback approach, the output theoretical background and Section 3 presents active noise
control experiments in an acoustic duct as a suitable real-
time platform to test the proposed LMI based approach.
T. Yucelen is with the School of Aerospace Engineering, Georgia Finally, conclusions are summarized in Section 4.
Institute of Technology, Atlanta, GA 30332 (e-mail: tansel@gatech.edu).
A. S. Sadahalli is with the Department of Electrical and Computer
Engineering, Southern Illinois University at Carbondale (SIUC), II. THEORETICAL BACKGROUND
Carbondale, IL 62901-6603 USA (e-mail: arjunsid@siu.edu).
F. Pourboghrat is with the Department of Electrical and Computer
A. Duct ANC system
Engineering, Southern Illinois University at Carbondale (SIUC), Consider a semi open-ended duct ANC system depicted in
Carbondale, IL 62901-6603 USA (e-mail: pour@siu.edu). Figure 1, where the speaker (Sd) is for generating a
disturbance noise d(t) and the speaker (Sc) is for applying the other, due to the location-based uncertainties, they belong to
control signal u(t) to cancel/reduce the noise y1(t) at the error a convex set constrained between some upper and lower
microphone M1. Moreover, M2 and M3 are additional noise bound Bode magnitude plots, as shown in the figure. The
measurement microphones with outputs y2(t), y3(t), problem here is to design a noise reducing control for a
respectively. The objective is to design a robust control u(t) system whose Bode plot varies between these upper and
such that in a large neighborhood of M1 the acoustic noise is lower frequency-domain bounds. If such a control structure
reduced over a wide frequency band. Most ANC algorithms exists, then it would be robust against such location-
result in a small zone of quiet only around the error dependent uncertainties and the undesired w(t) will be
microphone M1. The robust control u(t) in this paper is reduced in the region between locations a and b. We address
designed based on the acoustic measurements at M1, such this problem in Section 2.3.
that the acoustic noise is reduced at M1 and also at M2 and
M3, which results in a zone of quiet stretching effectively
from point a to b, as shown in Figure 1. Note that M1 is the
only microphone to be used in the closed-loop for control
implementation. Microphones M2 and M3 are only for
monitoring the noise levels at their fixed locations and for
the verification of the control performance.
Figure 2. Block diagram of acoustic duct for active noise control
k k
Ak Bk
The situation is depicted in Figure 3 which shows the typical
y = [I 0 " 0] x
Bode magnitude plots for the three models, G1(s), G2(s) or
[a 0, k
I a1, k I " an −1, k I ]Y + y (n)
= Ck ⎡⎣ a0,k I + a1, k Ak + ...
(4)
y (n)
f = P( s)n −1 y f + y = −λ Y f + y T
f (13)
+an−1,k Ak n −1
+ Ak ⎤⎦ x + [ B0, k
n
B1, k ... Bn −1, k ]U Then, we get
where (
y = y (fn ) + λ Tf y f = Cζ , k + ⎡⎣ λ Tf )
0 ⎤⎦ ζ f (14)
(
B0, k = Ck a1, k I + a2, k Ak + ... + an −1, k Ak n−2
+ Ak n −1
) Bk
Now, we can rewrite (9) using (10) – (14), as
(
B1, k = Ck a2, k I + a3, k Ak + ... + an −1, k Ak n −3 + Ak n − 2 Bk )
#
(5)
(
ζ f = Aζ , k − ⎡⎣ 0, Bζ , k λ Tf ⎤⎦ ζ f + Bζ , k u
)
Bn −1, k = Ck Bk Aζ ,k
(15)
And that Y and U are defined as
(
y = Cζ + ⎡⎣ λ 0 ⎤⎦ ζ T
f ) f
T
Y = ⎡y ⎤
T
T
y ... y T ( n −1) Cζ ,k
⎣ ⎦ (6)
T
and also from the first element of equation (10), we have
U = ⎡u T u T ... u ( n −1) ⎤
T
⎣ ⎦
y f = Cζ , k ζ f (16)
It should be noted that according to Cayley-Hamilton
theorem in [21], the coefficient of x in (4) is zero. That is, Equation (15) represents a non-minimal state-space
realization (NSSR) of (1). Here, the NSSR reformulates the
Ck ⎡⎣ a0, k I + a1, k Ak + ... + an −1, k Ak n −1 + Ak n ⎤⎦ = 0 (7) output feedback problem to a state-feedback control problem
with full-state measurement, ζf, which is available even
Let ζ be a new state vector using Y and U as,
when the parameter matrices in system (1) are unknown.
T T
ζ = ⎡⎣Y T ,U T ⎤⎦ = ⎡⎣ y T ... y ( n −1) u T ... u ( n −1) ⎤⎦
T T
(8) C. Robust state-feedback H∞ control using LMI
Then, a new state-space representation can be defined using In this section, a new H∞ control approach for active noise
(4) and (8) as, control with fictitious sensor location uncertainty is
presented using linear matrix inequality. The full state H∞
⎡ 0 I 0 ... ... 0 ⎤ ⎡0 ⎤ control design [23] used here is the state feedback version of
⎢ # % # ⎥ ⎢# ⎥
⎢ 0 ... 0 I 0 ... ... 0 ⎥ ⎢ ⎥ the output feedback H∞ control theory [5], [22]. This
⎢ ⎥ ⎢ ⎥
− a I ... ... − a I B ... ... B (9) method results in a control u (t ) = − K χ for the system (17),
ζ = ⎢ 0, k n −1, k 0, k n −1, k ⎥
ζ + ⎢ ⎥ u (n)
⎢ 0 ... ... 0 I 0 0 ⎥ ⎢ ⎥ so that the inequality (18) is satisfied. That is,
⎢ # % # ⎥ ⎢# ⎥
⎢ ⎥ ⎢0 ⎥ χ = Fk χ + Gk u + Gw,k w
⎢ 0 ... ... 0 I ⎥
⎢ ⎥ (17)
⎣⎢
0 ... ... 0 ⎦⎥ ⎣N
I⎦
z = C z χ + Dz u
Aζ ,k Bζ , k
where ζ, the states of (9) are formed using direct derivatives where γ is the performance bound. Here, Fk = Aζ , k and
of inputs and outputs. However, it is not practical to Gk = Bζ , k are the non-minimal state space system matrices,
differentiate these measured signals [5]. To avoid
χ =ζf is the full-state measurement, and Gw, k = Gk .
differentiation and to eliminate u(n), the system’s inputs and
outputs are filtered by the transfer function 1/Λ(s) [4], where Moreover, C z and Dz are the H∞ control design matrices,
Λ(s)=(s+λ)n, λ>0, is an arbitrary monic and Hurwitz
where it is assumed that Dz DzT = I [22]. Equation (17), with
polynomial of degree n, as
T T
k=1,2, represents the frequency-domain upper and lower
ζ f = ⎡⎣Y fT ,U Tf ⎤⎦ = ⎡ y f T ... y f ( n −1) u f T ... u f ( n −1) ⎤
T T
(10) bound models of the duct ANC system, respectively. The
⎣ ⎦
H∞ control solution is achieved by solving a minimization
where subscript f denotes filtered version of the signal. Also problem with the cost function given as
∞ this experiment that had an approximately flat frequency
J ( χ , u , w) = ∫ ( z T z − γ 2 wT w)dt (19) response in the frequency range from 30 Hz to 18000 Hz.
0 The purpose of this active noise control experiment was to
where one needs to find the positive-definite and symmetric reduce the noise in a region from a to b, using only
matrix P and the control gain K from (20) to realize microphone M1 (see Figure 1), by applying the proposed
u (t ) = − K χ [5], [22], where we must have control to the speaker Sc at the middle of the duct. The other
microphones were used for monitoring and verification.
(Fk − Gk K )T P + P(Fk − Gk K ) + PGk Gk T P
(20)
+ γ −2 PGw,k Gw,kT P + CzT Cz < 0
This inequality is nonlinear in the unknown variables P and
K. To remove the nonlinearity from (20), first let P = P −1 .
Then, after multiplying both sides by P , (20) becomes,
P(Fk − Gk K )T + (Fk − Gk K )P + Gk GkT
(21)
+ γ −2Gw,k Gw,k T + PCzT Cz P < 0 Figure 4. Experimental setup