AIAA Atmospheric Flight Mechanics Conference AIAA 2010-7502

2 - 5 August 2010, Toronto, Ontario Canada

Practical Aeroservoelasticity In-Flight Identification and

Adaptive Control

Xiaohong L. Li1
Integrated Aerospace Engineering, Irvine, California, 92620

Martin J. Brenner2
NASA Dryden Flight Research Center, Edwards, California, 93523
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

When an adaptive system changes to respond to off-nominal aircraft configuration, the

changes in control can affect the structural mode attenuation levels which are very difficult
to predict a priori. A practical approach for in-flight aeroservoelastic (ASE) suppression is
proposed. In this paper, the model reference adaptive gain-tuning flight control law is
illustrated for flexible aircraft longitudinal control. Since one of the most unexpected ASE
couplings usually shows as low frequency limit cycle oscillation (LCO); using the in-flight
narrow bandpass filter (NBWF) detection, the adaptive notch filter will be inserted in the
actuation command path to attenuate the LCO resonance on the closed loop flight dynamics.

I. Introduction

H istorically, separate control systems have been designed to augment the rigid-body dynamics and the structural
dynamics. When a vehicles structure exhibits sufficient rigidity, wide frequency separation results between
the rigid-body flight dynamic modes and the remaining natural modes which are dominated by the elastic degrees of
freedom. As a result, aircraft dynamics modeling, analysis, and synthesis generally treat the rigid-body dynamics or
the structural dynamics separately, seldom together. For example, in the development of a flight dynamics model
for application to performance, stability and control analysis, or flight control synthesis, a rigid-body assumption is
frequently made. Elastic degrees of freedom are generally absent in the traditional flight performance model and
control design system.
The term of rigid body dynamics is employed here as assumed that the aircraft has no elastic deformation during
its maneuver. In reality, all the aircraft due to its distributed mass and structure is flexible by nature. The underline
assumption for the above flight control synthesis approach is the wide frequency separation or loose coupling
between the rigid body dynamics and flexible modes. Another aspect of this approach is the limited authority in the
legacy control actuation system which inheritably imposes a bandwidth separation between the rigid body flight
mechanics and the excitation capability of flexible structural modes.
In the field of structural dynamics, the elastic degrees of freedom are employed for dynamic load calculation,
vibration and flutter analyses, and not all of the rigid-body degrees of freedom are included in this type of study.
Flutter (aeroelasticity) analysis as its own study is a stability analysis and usually incorporate aerodynamics and
structural dynamics coupling without the closed loop flight control system.
For the flight vehicles with significant flexibility which results in narrow frequency separation between the rigid-
body modes and the elastic modes, the existing solutions for airplane control usually include roll-off (limited
bandwidth and authority of actuation) and notch filters to avoid aeroservoelastic (ASE) interactions. These structural
mode suppression filters are designed to provide 8 dB of gain attenuation at the structural mode frequency.
Analytically, these notch filters with bandwidth limited flight control system in the format of transfer functions
including the sensor transfer functions are augmented into the finite element flutter model to verify the
aeroservoelasticity stability vs. the aircraft critical speed altitude envelope. At the test stage, Ground Vibration
Testing (GVT) to verify the correct modeling of the aircraft structural model, Structural Mode Interaction (SMI)
1
Consultant, lindali.iae@gmail.com. AIAA Member.
2
Aerospace Engineer, Aerostructures Branch. AIAA Member.

Copyright 2010 by Xiaohong Li and Martin Brenner. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
 testing, and finally full scale flight testing (with autopilot in the loop) are performed to verify that no adverse
aeroservoelastic interactions occur.
The above systemic design and test approach usually provides adequate protection and acceptable aircraft
performance for normal designed operation range.
If an abnormal air vehicle configuration occurs, a human pilot will usually take over the autopilot. Most likely,
after a quick assessment of the situation, the pilot will try to land the air vehicle safely as soon as possible. In this
scenario, the pilot acts as an adaptive intelligent controller in the flight dynamic control loop with human
identification skill to the new abnormal flight configuration. Depends on the abnormal condition, it is usually very
challenging for the human only piloting, especially for nowadays highly augmented flight by wire control system.

A. Objective and Interest

NASA has initiated an Integrated Resilient Aircraft Control (IRAC) effort under the Aviation Safety Program
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

(Reference 1). The main focus of the effort is to advance the state-of-the-art technology in adaptive controls to
provide a design option that allows for increased resiliency to failures, damage, and upset conditions. These
adaptive flight control systems will automatically adjust the control feedback and command paths to regain stability,
maneuverability, and eventually a safe landing. One potential consequence of changing the control feedback and
command paths is that an undesired aeroservoelastic interaction could occur. The resulting limit cycle oscillation
could result in structural damage or potentially total loss of vehicle control.
When an adaptive system changes to respond to off-nominal rigid body behavior, the changes in control can
affect the structural mode attenuation levels. Examples of abnormal configurations include control surface and
actuator hardover, disconnect, signal oscillatory failure, control and monitoring sensor failure. The upset and
damage condition includes aircraft attached ordnance drop, passage parachute out, fuselage damage, wing or control
surfaces absent, etc.
The combination of changing structural behavior with changing control system results in a system with a
probability of adverse interactions that is very difficult to predict a priori. Thus, an onboard, measurement based
method is needed to ensure that the system adjusts to attenuate any adverse ASE interaction before a sustained limit
cycle and vehicle damage are encountered. Thus, the solution is relying on efficient and accurate real time
measurement, detection and identification technology.

B. Real Time System Identification Technical Challenge

In general, the adaptive control schemes incorporate real time system identification (SID) method to estimate the
unknown plant model and design adaptive control based on estimated model. There are two basic approaches for
plant model estimation: time domain and frequency domain. The fundamental theory of time domain system
identification is the linear least square estimation method. The other is based on the fast Fourier transformation
(FFT) to calculate system transfer function from input and output signals. Elaborated solutions are expanded for
different noise issues with both approaches. Of those, time domain system identification SID has more applications
in Multiple Input and Multiple Output (MIMO) system and there are some promising application cases of system
identification technology in aerospace industry (Reference 7).
The following features are summarized as traditional in-flight system identification approach using model based
approach:
1. Prefers or requests persist excitation input from actuator or test equipment (for example, flutter test
shaker) during flight maneuver in order to excite as much as possible the dynamics of the system such
that the derived model incorporate the accurate and broader dynamics. This type of aggressive input
is not suitable for a damaged air vehicle with unknown or unstable flight dynamics.
2. Both system identification methods need to collect sufficient length of input and output signals. The
methods involve discretizing the signal into digital format and using FFT or LSE Jacobian matrix to
estimate the system transfer functions or model. Convergences and matrix calculations are not
computational efficiency for real time response.
3. Most adaptive control with system identification research efforts focus on comprehensive mathematic
modeling. The resulted control law architecture is usually complex to implement in-flight for
emergency response.
 II. Methodology and Theory

Since one of the most detrimental results with adverse ASE interaction is in the form of sustained limit cycle
oscillation at and beyond flutter/ASE onset with actuation system in the loop. To exit this type of LCO, either the
pilot or autopilot needs to eliminate the actuation input to the system completely if not adaptively, thus usually result
the lost control of the aircraft. The key issue for ASE related LCO suppression is to detect the excitation path and
frequency of the limit cycle oscillation, and then avoid coupling of LCO while still maintain the aircraft control.
As stated, a complete system identification using LSE or FFT is computation intense and not efficient for quick
response, especially for suppression of LCO. In this project, authors using self tuning adaptive flight control law
with model reference architecture for flight dynamics control and with adaptive notch filter for the adverse ASE
suppression. The system transfer function is not real time identified. The system is assumed stabilizable even with
failures. This assumption did not impose any restriction of the solution since the system stabilizability is the
minimum requirement for any control effort employed; the adaptive control system will only focus on LCO
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

detection and suppression from closed loop control. The adaptive gain tuning is used to improve the system
dynamic damping while the LCO is suppressed by notch filter. The adaptive filter is designed based on Narrow
Band Filtering system identification technology outlined in the following section.
It is worth mention some previous works for flutter onset predication based on flight test data which were
summarized in Reference 3. The methodologies using damping extrapolation, envelope function, Zimmerman-
Weissenburger flutter margin method are test data based approaches for flutter detection. Since this paper only
focuses on ASE in-flight LCO suppression and has its own limitation; it can be integrated with the above data based
flutter detection and predication method for flutter prevention which also includes flutter divergence mitigation.

C. Theory of Operation for Swept Tuned Spectrum Analyzer

To introduce in-flight Narrow Band Filtering system identification technology, the practice of swept tuned
spectrum analyzer in signal processing will be reviewed first.
A French mathematician and physicist, named Jean Baptiste Joseph Fourier, who lived from 1768 to 1830
discovered that time history signals are comprising of the fundamental sine wave as well as third, fifth, seventh, etc.
harmonics in the correct proportions and developed the theory of signal frequency analysis.
To enable the measurement of the frequency contents of a signal, the spectrum analyzer is an essential tool.
There are two types of spectrum analyzer:
Swept or superheterodyne spectrum analyzers: The operation of the swept frequency spectrum analyzer is
based on the use of the superheterodyne principle, sweeping the frequency that is analyzed across the required band
to produce a view of the signals with their relative strengths. It is most widely used for radio frequency signal
communication field for its high frequency detection and analysis efficiency.
Fast Fourier Transform, FFT analyzers: These spectrum analyzers use a form of Fourier transform known as a
Fast Fourier Transform, FFT, converting the signals into a digital format for digitally analysis. These analyzers are
more expensive and often more specialized.
As stated in previous section, FFT analyzer technology is also the foundation for frequency domain system
identification technology. The visible difference from user point of view between the two measurement techniques
is that using a frequency sweep approach, the result will seen as sweep progresses, when an FFT measurement is
made, the result cannot be displayed until the sufficient length of data is collected and FFT processing is complete.
The following concept of operation of swept tuned spectrum analyzer is illustrated here (Reference 2).
 fs Signal Range LO Range

IF Filter

Threshold

Sweep Generator
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

Figure 1 Operation Theory of Swept Tuned Spectrum Analyzer

In a swept tuned spectrum analyzer, an Intermediate Filter (IF) is utilized as bandpass filter to detect the signals.
Its bandwidth window is called the resolution bandwidth (RBW) of the analyzer; the narrower of the filter
window, the higher the resolution.
An input signal is combined with the local oscillator signals (LO) through the mixer, to convert it to an
intermediate frequency signals which composed of four frequency contents at (fLO, fLO+fs, fLO-fs, fs). These signals
are then sent to the IF filter. The output of this filter is detected, indicating the presence of a signal component at the
analyzer's tuned frequency.
Starting the local oscillator with the low end of spectrum analyzer, for example 3.6 GHz, the output of the mixer
has four signals after the signal mixer, one of which is at 3.6 GHz (fLO). Notice that the IF filter is fixed at 3.6 GHz
for this example. After passing through the IF filter, the so called "LO Feedthrough" signal is display noticeably at
0 Hz on the resulted frequency analyzer plot. Then, the sweep generator is command the LO to sweep higher in
frequency. As the LO sweeps, so too will three of the mixer output signals (the input signal fs is stationary). As the
LO Feedthrough moves out of the IF filter bandwidth, the shape is taper off on the display. As soon as the
difference frequency (fLO-fs) comes into the skirt of the IF filter, it becomes obvious. When it is at the center (e.g.
3.6 GHz), the full amplitude of this signal is on the display. And, as it moves further to the right, it leaves the filter
skirt, and no signal is seen on the display. With known LO frequency and IF filter, the input signal at frequency fs is
identified on the frequency analyzer plot.
This concept of intermediate frequency filtering is extended as the narrow bandwidth filter technology with
model reference signal to detect the ASE adverse dynamic response.

D. Narrow Bandwidth Filter for Real Time Identification

As stated in previous section, the narrow bandpass filtering (NBWF) is commonly used for signal frequency
content detection in spectrum analyzer with the local oscillatory signal generator to sweep the frequency range.
For ASE adverse dynamic coupling detection, multiple filter windows are utilized simultaneously to cover the
frequency range, namely 1Hz to 10 Hz range, which is a simpler design utilization comparing to local oscillatory
and mixer circuit design. Similar, the bandwidth of NBWF is the resolution bandwidth of the signal detection. In
this project, 1 Hz is defined as resolution bandwidth.
The filtered signals are then compared with nominal signal generated by reference model, also NBWF filtered
to evaluate if significant frequency component and strength are presented in the system response as abnormal failure
occurred. This expected nominal signal is the model response using actuation signals either commanded or
measured for threshold standard in real time.
Assume a given flexible aircraft pitch rate transfer function from elevator input is represented as the singular
values shown in Figure 2 (Reference 4). There are 6 eigenvalues (modes) in this transfer function of which the first
 two are rigid body modes. The narrow band filter is applied to detect the response component with most sharp
resonant frequency such as the first flexible aircraft mode at 12.40 rad/sec. The nominal system response will
assume that the first flexible mode is not presented in the pitch rate response as the brown dash line indicated (no
peak around 12.40 rad/sec). The brown line indicated nominal configuration is used as the model reference for
excessive resonance frequency detection.

S in g u la r V a lu e s
First flexible mode
0
Full Order Real Sys
F O M ra d / se c
R Model
O M 8 t hReference
F S M C raSys
d / se c
MNarrow
O D E L cBand
2 Filter
-1 0
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

-2 0
S i n g u l a r V a l u e s (d B )

-3 0

-4 0

-5 0

-6 0

-7 0
-2 -1 0 1 2
10 10 10 10 10
F re q u e n c y (ra d / se c )

Red Window is narrow band filter with Detection Threshold

Figure 2 Concept of Narrow Band Filter for LCO Detection

E. Lyapunov Based Model Reference Adaptive Control

Figure 3 outline the MIT rule model reference adaptive control which is applicable for adaptive control of real
process plant with forward gain unknown or gain variation. It is further developed into Lyapunov rule based model
reference adaptive control. The theoretical detail is not covered here.

Reference Model

K 0*G(s)
ym

1
Kadp
s
Integrator e
Gain

Real Process Subtract

K*G(s) 1
1 ur yr
y
uc Product

Figure 3 Concept of MIT Model Reference Feedforward Gain Adaptive Control

 The adaptive control using the Lyapunov rule based model reference is utilized in this paper for its ASE control
design as in Figure 4 to demonstrate the adaptive control design for a system with forward gain variation in the real
process plant. The referenced model is assumed that the aircraft is rigid at nominal operation with no failure.

cmd cmd q_m

q_m

cmd q
Ref Sys

err
1
Kadp K *u
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

Kbeta

Matrix
Kunc cmd
Multiply
q
q_r
SWH
1
0 nbwF_SWH Real Sys

Figure 4 Simulation Model for ASE Model Reference Gain Adaptive Control

III. Simulation

Two representative ASE simulation models are setup. One is nominal aircraft in designed configuration and the
other is abnormal condition with significant ASE dynamics limit cycle signal presented in pitch rate gyro response.
A persistent sine wave disturbance is injected to the rate gyro sensor to represent the LCO. This type of pitch LCO
was similarly to that Residual Pitch Oscillation (RPO) observed in B2 flight test (Reference 6).
The effectiveness of NBWF is evaluated first. The Lyapunov based adaptive control with NBWF is evaluated
later for the combined failures: LCO presented in rate gyro and forward loop gain variation of Flight control system.

A. ASE Model and Previous Designed Robust Control

The following fictitious ASE model is used here for technology demonstration. This longitudinal dynamics
model of an elastic aircraft is an augmented state space model of a rigid flight dynamics model.
The aircraft model is inherited with static instability. It is assumed that the elastic effects of an aircraft in
longitudinal motion are adequately represented by eighteen aeroelastic modes. Thus, the state space equations have
40 state variables. The additional information of the matrices A, B, C and D are provided in Reference 4.

x& = Ax + Bu, (1)

y = Cx + Du , (2)
 where, u (in / s )
w (in / s )

q(deg/ s )
x= , (3)
(deg)
& i i = 1,...,18 (1 / s 2 )

i i = 1,...,18 (1 / s )

(4)
u = elevator ,

q(deg/ s ) (5)
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

and, y= ,
q XS (deg/ s )
18
(6)
where, q XS = q ( x )& ,
j =1
'
j S j

d j ( xs ) j (x s ) = C j x S .
'j ( xs ) = , (7) and (8)
dxs

C j = j th row of C , [
xS = x13
S x12
S L xS 1 . ]
T
(9) and (10)

Equation (6) represents the outputs from pitch sensors. q is the ideal pitch rate sensor output at the center of
gravity location which is not contaminated by the deformation and elastic motion of the aircraft and used as nominal
aircraft response. qxs represent the two sensor outputs that depend on the mode shapes of the flexible structural
modes. x s is the displacement in inches of the fuselage station measured from the nose of the aircraft. The mode
shape j ( x s ) is interpolated as the 13th order of polynomial expression of the displacement x s , given in Equation
(8) of linear vector expression of xS of Equation (10). Assuming that two pitch rate sensors are measuring the
motion in the direction of the slope of the mode shape given by Equation (7).
The closed loop control for nominal aircraft configuration was developed based on previous work using reduced
order robust controller (Reference 4). It is employed here as the Golden reference model to provide the nominal
operation response, shown in Figure 5. For failure condition, a persistent sine wave disturbance is injected to the
rate gyro sensor to represent the LCO impact (Figure 6).
FCS cmd signal
1

0.5

0
0 5 10 15
Ideal pitch rate signal, nominal without LCO (deg/sec)
1

0.5

-0.5
0 5 10 15

Figure 5 Closed Loop Response of Nominal System (without LCO)

 Actuator signal with LCO
5

-5
0 5 10 15
Ideal pitch rate signal with LCO (deg/sec)
1

0
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

-1
0 5 10 15
Sensor #1 and #2 pitch rate signal with LCO (deg/sec)
5

-5
0 5 10 15

Figure 6 Closed Loop Response of Real System with LCO

B. LCO Onset Detection and Suppression

The adaptive control goal is to eliminate the coupling of this LCO from flight control system using notch filter.
No control architecture change is applied, other than the proper notch is inserted in the sensor feedback loop.
As shown in Figure 7, the actuator does not respond to this LCO signal which is still existed in rate sensor when
NBWF is activated. This is different from previous response of Figure 6. However, actuator is still responding to
pilot/autopilot command for rigid aircraft command. In this case, pitch rate damping augmentation is still effective
in closed loop. There are two sensors in the system, the average of the signal is also provided as feedback signal for
pitch axis control.
The suppression of LCO is seen as signal progresses in time history within one frequency cycle when NBWF is
turned on. Due to the filter technology, there is no need for repeated excitation to collect long history of the data.
 FCS cmd signal
1

0.5

0
0 5 10 15 20 25 30 35 40 45 50
Actuator signal
5

0
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

-5
0 5 10 15 20 25 30 35 40 45 50
Ideal pitch rate signal deg/sec
2

-2
0 5 10 15 20 25 30 35 40 45 50
Sensor #1 and #2 pitch rate signal deg/sec
5

-5
0 5 10 15 20 25 30 35 40 45 50
Switch of NBWF On=1/Off=0
1

0.5

0
0 5 10 15 20 25 30 35 40 45 50

Figure 7 LCO Detection and Suppression using NBWF and Notch Fitler

C. LCO Suppression and Adaptive Gain Tuning Control

Three simulation cases are evaluated:

i) The pitch axis flight control forward gain is increased 3x with LCO presented in rate sensor. The adaptive
gain control is not used and the NBWF is not turned on. Shown in Figure 8, the magnitude of pitch rate signal is
excessive comparing to the nominal reference model and LCO is persistent presented.
ii) The pitch axis flight control forward gain is increased 3x with LCO presented in rate sensor. The adaptive
gain control is used and the NBWF is also turned on. Shown in Figure 9, the magnitude of pitch rate signal is
compatible with the nominal reference model and the LCO is reduced in the pitch rate response. This is similar to
the simulation results in Figure 7. The noise and sharp edge of signal due to the transient switch of the notch filter
can be reduced using signal fade-in strategy in future improvement.
 iii) The pitch axis flight control forward gain is increased 3x with LCO presented in rate sensor. The adaptive
gain control is used and the NBWF is partially turned on. Shown in Figure 10, the applied adaptive control alone is
not effective for LCO suppression.
Reference model response q (rad/sec), No Adaptive, No NBWF

1.5

0.5

-0.5
Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

0 5 10 15

Real plant response q (rad/sec), No Adaptive, No NBWF

-2

-4
0 5 10 15

Figure 8 Real System Response with Gain Variation without Adaptive Control, No NBWF

Reference model response q (rad/sec), With Adaptive, and NBWF

1.5

0.5

-0.5
0 5 10 15

Real plant response q (rad/sec), With Adaptive, and NBWF

-1
0 5 10 15

Figure 9 Real System Response with Gain Variation with Adaptive Control and NBWF
 Reference model response q (rad/sec), With Adaptive, No NBWF

1.5

0.5

-0.5
0 5 10 15

Real plant response q (rad/sec), With Adaptive, No NBWF--

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on January 18, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2010-7502

nbwF switch
2

-2

-4
0 5 10 15

Figure 10 Real System Response with Gain Variation with Adaptive Control, Effects of NBWF

IV. Conclusion

The adaptive control with adaptive notch has effectively attenuated the adverse ASE coupling in the simulation.
The Narrow Bandwidth Filtering (NBWF) technology can be applied to quickly identify the unexpected structural
resonant frequency and its signal strength.
The presented solution is by nature faster than FFT and least square time domain system identification method
and practical to implement for real world application of ASE LCO suppression.
The optimization of filter allocation in the interested frequency range can be developed in the future.

References
1
National Aeronautics and Space Administration Small Business Innovation Research & Technology Transfer 2009 Program
Solicitations, A 1.07, Adaptive Aeroservoelastic Suppression, 2009.
2
Hewlett-Packard Company, Spectrum Analysis Basics, Application Note 150. 1989.
3
Rick, L., Brenner, M., Flight Test Evaluation of Flutter Prediction Methods, 43rd AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics and Materials Conference, 22-25 April 2002, Denver, Colorado, AIAA-2002-1649.
4
Li, X., Agarwal, R., Application of Reduced-Order-Models to Robust Control of the Dynamics of a Flexible Aircraft,
AIAA-2003-5504. AIAA Guidance, Navigation, and Control Conference and Exhibit, Austin, Texas, Aug. 11-14, 2003.
5
strm and Wittenmark, Model-Reference Adaptive Control-From Theory to Practice, edited by Hans Butler, Prentice-
Hall, 1992.
6
Britt, R. T., Volk, J. A., Dreim D. R. and Applewhite, K. A., Aeroservoelastic Characteristics of the B-2 Bomber and
Implications for Future Large Aircraft, RTO AVT Specialists' Meeting on "Structural Aspects of Flexible Aircraft Control",
Ottawa, Canada, 18-20 October 1999, and published in RTO MP-36.
7
Morelli, E., Klein, V., Application of System Identification to Aircraft at NASA Langley Research Center, Journal of
Aircraft, Vol. 42, No. 1, January-February 2005.