Proceedings of the American Control Conference
Arlington, VA June 25-27, 2001
A Single-Input Two-Output Feedback Formulation for
ANC Problems
V. Toochindatl, T. Klawitted, C.V. Hollott and Y. Chaitt
SMIE Dcpartmcnt, tECE Dcpartmcnt, Univcrsity of Massachusetts, Amhcrst,MA 01003
Abstract
This paper explores inherent feedback limitations of ac-
tive noise control in ducts by using a singleinput, two-
output framework and observing properties of closed-
loop transfer functions. Performance is assessed using
the plant and disturbance alignment angle. We show
that the sound levels at the measurement microphone
axe amplified when attenuating acoustic energy at the
error microphone. We also show that the stability mar-
gins can be improved over feedforward control using
measurements from two sensors.
1 Introduction
In recent years, the potential benefit of using active
noise control (ANC) in commercial applications has
driven recent academic research; e.g., see [l]and [2]. In
contrast to passive techniques, a typical ANC scheme
wcs additional sccondary sourccs and adaptivc alge
rithms to cancel noise from the original primary source
by, roughly speaking, introducing "anti-noise" - an ex-
act but outdf-phase copy of the noise. The level of
cancellation depends critically on the ability t o produce
such anti-noise in the face of uncertain system dynam-
ics and noise properties. Such is the motivation for the
introduction of adaptivc canccllation tcchniqucs [I] as
well as the more recent feedforward/feedback methods
[3, 4,61.
This paper is a continuation of some of our previous
work [3]-[6]where we explore the use of non-adaptive,
fixed-filter schemes for ANC in ducts. Our motivation
for using hed-flters lies in the simplicity of implemen-
tation and availability of tools for aualyzing stability
and performance. Other work using the non-adaptive
approach can be found in 171 and (81 and the references
contained therein.
The main goal of this paper is to analyze a non-
adaptive ANC design reported in [4], within the frame
work of singleinput, two-output (SITO)feedback con-
trol previously developed in [9]. This design used two
sensors (measurement and error microphones) and a
'The 6rst author is supported by mculty of Engineering,
Kasetsart University, Thailand.
Email: vtoochinQacad.umass.edu
0-7803-6495-3/01/$10.00 02001 AACC
so-called feedfonvard/feedback configuration. In Sec-
tion 3 wc will show that thc placcmcnt of thc sccondary
source, relative to the disturbance source, plays an im-
portant role in performance. In particular, we will
demonstrate, via analysis and examination of experi-
mental data collected in 141, that noise attenuation at
the error microphone must come at the expense of am-
plification at the measurement microphone. We believe
this is thc first timc this obscrvation is madc in thc lit-
erature. Our second result, discussed in Section 4, ar- -
gues that the role of feedback in ANC is not for noise
attenuation, but for stability robustness. In Section 2
we begin our discussion by describing the ANC duct
setup and the design results of 141.
2 Description of ANC setup and design results
Figurc 1 illustratcs a basic configuration of an ac-
tive noise control problem. It consists of a duct with
two loudspeakers and two microphones mounted on it.
The speaker located upstream simulates a disturbance
source that injects acoustic "noise" into the duct. A
measurement microphone detects the disturbance near
the source and the downstream error microphone mea-
surcs thc lcvcl of noisc canccllation a t a point in thc
'duct where noise attenuation is desired. The ANC sys-
tem uses the information provided by these two micro-
phones to generate a signal and send it to the cancel-
ing loudspeaker. The objective of the controller is to
minimize the acoustic energy at the error microphone.
In the ANC literature, the action taken on the mea-
surcmcnt microphonc signal y, i s rcfcrrcd to as "fccd-
forward" control, while action taken on ye is called
"feedback" control. While such terminology can be am-
biguous in the presence of acoustic feedback, we will
nevertheless retain this terminology for sake of consis-
tency. The ANC configuration used in 1.11 is similar to
that shown in Figure 1. The measurement microphone
is colocated with the disturbance source, and the can-
celling speaker is colocated with the error microphone.
The duct-length is about one meter with a diameter
of 0.1 meters. The two microphones are separated by
'The term feedforward comes from the fact that the meas-
ment microphone is located upstream i.e. closer to the dstur-
bance source. Unlike a conventional feedforward control, how-
ever, the acoustic feedback from the control speaker could deet
closed-loop stability.
923

Figure 1: ANC system in a duct.
approximately 0.8 meters of duct. Motivation for this
placement of measurement microphone and cancelling
speaker came from [8] ,where only U,,, was considered as
input to their ANC controller. Interestingly,we will see
in Section 3 (see Remark 1) that it may be preferrable
to reconsider colocating the cancelling speaker with the
disturbance - a collfiguration not recommended in [8].
With the con6guration shown in Figure 1, a formal
design of feedforward and feedback controllers was ex-
ecuted in [4] using both Ifo, synthesis and QFT loop-
shaping techniques. A comparison of the experimental
open and closed-loop performance is shown in Figure
2. This figure plots the magnitude frequency response
of ye due to sinusoidal excitation at d. Note that this
ANC design achieves broadband attenuation from ap-
proximately 180 - loo0 Hz with no amplification below
2000 Hz. Our subsequent analysis is driven by a desire
to understand the roles of feedforward and feedback in
this particular design.
"il
Figure 2: Comparison of open loop and clmed-loop fre-
quency response
Our empical observations were twofold
1. While noise attenuation was achieved at ye, as
shown in Figure 2, amplification occurs at y,,,. In
the next section, we w i l l show why this occurs and
prove that this amplification is not a function of
the structure or value of the controllers,but is due
to the duct dynamics coupled with the location of
924
the microphones and placement of the cancelling
speaker.
2. The feedback controller Ce did not contribute
to the level of noise attenuation achieved at ye
- this W ~ Sprimarily m p l i s h c d by thc fccd-
forward controller C ,
. Indeed, the attenuation
curve in Fre;ure 2 is only slightly perturbed when
we switched-out C,. Hawever, the robustness of
this performance to variations in C,,, were signif-
icantly impraved by the presence of the feedback
element C,.
3 Performance analysis
In this section we analyze the performance of the ANC
design in [4]. We start by treating the duct (or plant)
as a SIT0 system P(s) where the input is the can-
celing speaker input U and the outputs are the micre
phone signals y,,, and ge, respectively. Conversely, we
view the controller as a two-input, singleoutput system
C(s) with inputs y, and ye and output U. Their in-
terconnection forms the feedback control system shown
in Figure 3. The objective of this section is to analyze
the performance in ym and ye. We will show that ye
is attenuated at the expense of amplified U,,,. Most
importantly, this finding will be shown independent of
the value or structure of the controllers (as long as ye is
attenuated). It will depend only on the configuration
of microphones and canceling speakers.
To start, considcr Figurc 3, whcrc P,d(s), P,,(s),
Pmd(s), and Pmu(s) are the transfer functions from the
disturbance speaker to the error microphone, the can-
celing speaker to the error microphone, the disturbance
speaker to the measurement microphone, and the can-
celing speaker to the measurement microphone, respec-
tively. Let Cm(s) denote the feedfonuardcontroller and
C,(s)thc feedbackcontrollcr. Furthcr,-dcfinc thc plant,
controller, and disturbance transfer functions as:
I
I'
Figure 3 ANC as a singlc input, two output fccdback
system
Assume zero input and initial conditions for the plant
transfer function, the open loop response is then
Associated with this feedback system are several im-
portant transfer functions. These are the input and
output loop transfer functions: Ll(s) = C ( s ) P ( s ) and
LO(.) = P ( s ) C ( s ) , the input and output sensitiv-
+
ity functions: Sr(s) = (1 Lr(s))-' and So(s) =
+
(I Lo(s))-l, and the input and output complemen-
tary sensitivity functions: Tl(s)= L l ( s ) ( l + LI(s))-'
+
and To(s) = Lo(s)(I Lo(s))-'. The dimension of
transfer functions at the plant input are 1x1, while
those at the plant output are 2 x 2 . In our context,
the closed-loop response of interest is:
We then define the attenuation factor as
In the sequel we will show that there are situations,
experienced in the setup [4] for which Q ( W ) must be
greater than one. First,we d&e the notion of align-
ment angles introduced in [9].
Definition 1: The plant-controller alignment angle (at
frequency w ) is
while the plant-disturbance alignment angle is
The plant and controller (plant and disturbance)
are said to be perfectly aligned if q5Jjb) = 4 and 5 that both ym(jw) and y e @ ) are small at
0" (4pd(jw) = O"), and completely misaligned if
& ( j w ) = 90" (&&U) = 90"). From [Proposition
9,9], we have the following upper and lower bounds on
the attenuation factor:
925
Proposition 1: If lye(ju)l = 0 and #pd(jwo) = No,
then
From Proposition 1 we see that under perfect cancel-
lation of ye&), ym(jw) must be amplified if the plant
and disturbance transfer functions are completely mi%
aligned. Furthermore, if IPed(jLd)I >> IPmd(jw)Ir then
Iy,,, ( j w )I could be unacceptably largez.
Example 1: Using data from (41, wc plot
& d ( j w ) , W { and the lower bound (5) in Figure 4.
At w = 2n(174) rad/sec, q5pd(jw) FZ 90, and
cxpcricnccs its pcak. From Proposition 1, wc would
then predict that the closed loop response Igm(jw)l is
large. This is verified in Figure 5. Thus, while lge(jw)l
was attenuated in the closed loop, it appears to come
at the expense of amplifying lym(jw)I.
1m . . . . * . . . . ,
4 .
1' . . ' . ' ' ' 1
Figure 4: q5pd(jw), I=+, and lower bound (5)
Remark 1: It is interesting to observe from Figures
frequencies where the plant and disturbance are well-
aligned; i.e., = 0". Clearly, at these frequen-
cies both the upper bound (4)and lower bound (5)col-
lapse to ISr(jw)l and thc disturbancc attcnuation pcr-
formance becomes a sensitivity minimization problem.
2For example, it may be unacceptableto have the disturbance
amplified at any point in an HVAC system. The duct is a sim-
plied model of such a system.

Figure 5: Open loop and closed loop comparison of
V m W and zlediw)
In other words, when &&U) = O", the disturbance
affects the outputs in the same ay as does the control
signal, and that this is favorable for disturbance rejec-
tion. Otherwise, one can only choose one of the two
outputs for disturbance attenuation, and the control
signal used to do 60 will act as a disturbance to the
other output.
4 Stability margins
In the last section we concentrated on performance
analysis and presented a situation where lye(jw)l was
attenuated at the expense of an amplified Iym(jb)l.
Specifically, we saw that Iym(jw)l peaks at the f r e
quency where &&w) w 90" and, the ratio =I\
has a maximum. So, we could expect a smaller stabfity
margin (or a large I IS0 ( j w ) 1I) at this frequency. There-
{m\
fore, our subsequent analysis will devote special atten-
tion to the condition &&U) = 90" and
occurring when w = 2r(174) rad/sec . In this sec-
>> 1
tion we study the output sensitivity So(jw), whose
size IlSo(jw)ll provides one measure of stability mar-
gin;e.g. see [lo].
Case 1: (feedforward only; C ( j w ) = [Cm(jw) 01) In
this case,the system's closed-loop response is given by
Now, suppose Iye(jw)l = 0. Then, fiom Lemmas 1 and
2 in Appendix A we have
Thus, with &{# > 1, it follows from the funda-
mental algebraic constraint S&w) +
T . ( j w ) = l that
ISr(jw)l >> 1. This in turn implies that llSo(jw)ll >>
926
1. Thus, whenever we use only feedforward con-
trol, C ( j w ) = [Cm(jw) 01, attenuation of Iye(jw)l
at a frequency where both 1-5 > 1 and plant-
disturbance are misaligned necessarily leads to a r e
duced stability margin.
Case 2: (feedforward and feedback control; C ( j w ) =
[ C m ( j w ) C e ( j ~ ) ]USihg
) Straightforward manipular
tions, the output sensitivity and complementary sen-
sitivity functions can be written as
and where So,j(jw) and Toij(jw)are the (ij)th elk
ment of S o ( j w ) and T o ( j w ) , respectively. Note the
similarities and differences to (7). The proof of the
next propositon can be found in Appendix B.
Proposition 2: If Iye(jw)I = 0, then
Proposition 2 shows that for perfect cancellation,
lToll(jw)ldepends not only on plant and disturbance
transfer functions but aLs0 on both controllers. There
fore, unlike Case 1, the ratio4-1 does not impose
a constraint on lToll(jw)I when g e ( j w ) = 0.
+
Since Soll(jw) Ton(jw) = 1, then IToll(jw)l >> 1
implies ISoll(jw)l 3 1, which, in using (Q),leads to
l l S ~ ( j w ) l>
l 1 and smaU stability margins. Note also
that ]Toll(jw)l >> 1 implies poor robust stability to
multiplicative uncertainw in P,,.
plot the magnitude ratios m
y
Example 2: Returning to the experiments in [4],we
and
Figure 6. The magnitude of each element of the out-
in
put sensitivity S&w) is shown in Figure 7 for the
case CGw) = [C,(jw) 01. At w = 2r(174) rad/=,
q&@d) w 90" (seeFigure 4), and #$$# = 3.7 (from
Figure 6). So, by (8), IT'(jw)I $3 (3.7)' = 13.7. From
the algebraic constraint, ISr(jw)l is commensurately
large. The data yields llS0ll = 13, which compares
favorably with the preceeding analysis.
Now consider C(jw) = [Cm(jw) C e ( j w ) ] . Figure 8
shows the magnitude of the four elements of So(jw).
 tor configurations 80 that we can choose the setup that
gives the best overall performance.

6 Appendices

A. Proof of Proposition 1: First we need some lem-

mas.
4

0
0 m
mQm
Imp L e m m a 1: Given w , y,(jw) = 0 if and only if

Figure 6: Magnitude plots of S o ( j w ) for case C ( j w ) =

[Cm(jw) 01
For the special case when C e ( j w ) = 0:

Proof: From (1) wc havc

Figure 7: Magnitude plots of S o ( j w ) for c s e C ( j w ) =

[CmW) C=(jw)l Letting g e ( j w ) = 0 yields (11). Equation (12) follows
from Ce = 0 which, in turn, imp lie^ Tola(jw) = 0 and
T021(jw) = - m T I ( j w ) - 0
We ean compute lT'll(jw)I directly using Proposition
2, though it is easier to see from (9) that IT011(jw)l= Lemma 2: Suppose q5pd = 90'. Then,
J$&#lS~21(jw)l. At w = 2~(174)rad/sec, the data
from Figure 6 and Figure 8 gives #$$\ = 0.4 and
15021(jw)I = 1.7, resulting in lTol1(jw)l = = 4.25.
This implies that the size of lSoll(jw)l is compara-
ble. This is mnfirmed in Figure 8, where ISOl,(jw)l =
l 6; a signscant improvement over llsoll = 13
l l S ~ lNN
using just fccdforward control.

5 Conclusions

In this paper we have analyzed stability and perfor-

mance of ANC in a duct using the SIT0 formulation.
We derived a lower bound on closed-loop sound pres-
sure level that depends only on open-loop transfer fun0
tions. We proved that using only f e e d f o m d controller
imposes inherent limitations on the achievable closed-
loop performance. We also show that this limitation, in
terms of stability margin,can be alleviated when both
f e e d f o m d and feedback controllers are used. At this
point; we have not discussed how to design the con-
trollers to achieve the performance objective. This will
be a topic for further research. Another interesting
research direction is to study the relation of plant -
disturbance alignment angle to the sensor and actua- 30verbar denotes complex conjugate.

927
Lemma 3: suppose = 90 and ge(jw)= 0. Then, and (6) follows. 0

ProoE We will only show for the more general case

Cbw) = [Cm(jw) Ce(jw)]. The situation C ( j w ) = Cepe, Pmu
[Cm(jw) 01 would then follow directly by setting = (1 --Toll)--
CmPmu
Ped
pmd p e u -
. . = 0. The statement in Lemma 3 is equivalent
Ce(jw)
to _ _ gives
Fkuranging _
Ye(jw) = 01 4 p d = 90
Ped pmu
* ITr0w)l tan&(iw) -
and (10) follows. 0
To simplify notation, the dependence on w is omitted.
Thcn,

ITrl tan#, = ITA-

I6:PlHI
= ITrl
I - Cmpeu + CepmuI. References
ICPl ICmPmu + CePeuI [l] S.M. Kuo and D. R. Morgan, Active Noise Con-
trol System, Wiley-Interscience, New York, 1996.
[2] P.A. Nelson and S. J. Elliott, Active Control of
Sound, Academic Press, London, 1992.
[3] P.G. Metha, Y. Zheng, C.V. Hollot and Y.
Chait, Activc N o h Control in Ducts: Fccd-
.forward/Feedback Design by Blending H, and
QFT Methods,Prou?edings of the 13th IFAC World
Congress, San Francisco, 1996.
Applying Lcmma 1 and rcwanging tcrms givcs [4] T.Klawitter, Linear Active Duct Noise Control:
Feedfomard/Feedback Controllers Designed with lm
and Quantitative Feedback Theory, M.S. Thesis, Uni-
versity of Massachusetts, Amherst, 2000.
[5] L. Li, Experimental Inverstigation of LTI Feed-
forward Active Noise Control Design h m Experimen-
tal Zhpency Responses, M.S. Thesis, University of
Now ,at #pd = 90, me use Lemma 2 to obtain Massachusetts, Amhcrst, 1999.
[6] P.G. Metha, Fixed-Filter Designs for Active
Noise Control in Ducts, M.S.Thesis, University of
Massachusetts, Amherst, 1996.
[7] A.J. Hull, C.J. RadcliEc, and S. C. Southward,
Global Active Noise Control of a One Dimensional
Acoustic Duct Using a Feedback Controller,J. Dun.
Sys. Meas. Contr-, Vol. 115, pp. 488 - 494, 1993.
[8] J. Hong and D.S. Bernstein, Bode Integral Con-
straints, Colocation, and Spillover in Active Noise
and Vibration Control, IEEE rrcmS. Control Systems
We can now prove Proposition 1. Without loss of gen- Technology, Vol. 6(1), pp. 111-120, 1998.
erality, assume ld(ju)l = 1. From (I), ye@) = 0
impliCS 191 J.S. Fkeudenberg and R.H. Middleton, Proper-
ties of Single Input, Two Output Feed back Systems,
International Journal of Control, Vol. 72(16) pp. 1446-
1465, 1999.

Applying Lemma 3 and 4pd = 90 to the lower bound [lo] J.S. Freudenberg, C.V. Hollot, and D.P.Looze,
A First Gradaate Course in Feedback Control,
(5) gives Course Notes, EECS 565, University of Michigan,Win-
ter, 2000.

928