ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Active disturbance rejection control: Methodology

and theoretical analysis$
Yi Huang, Wenchao Xue n
Key Laboratory of Systems and Control, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, PR China

art ic l e i nf o a b s t r a c t

Article history: The methodology of ADRC and the progress of its theoretical analysis are reviewed in the paper. Several
Received 31 May 2013 breakthroughs for control of nonlinear uncertain systems, made possible by ADRC, are discussed. The
Received in revised form key in employing ADRC, which is to accurately determine the “total disturbance” that affects the output
4 March 2014
of the system, is illuminated. The latest results in theoretical analysis of the ADRC-based control systems
Accepted 13 March 2014
are introduced.
This paper was recommended
for publication by Didier Theilliol & 2014 ISA. Published by Elsevier Ltd. All rights reserved.

Keywords:
Active disturbance rejection control (ADRC)
Nonlinear uncertain system
Stability
Transient response

1. Introduction of brushless DC servo, ALSTOM gasifier benchmark problem, the

This paper is a review of the methodology of ADRC and the
progress of its theoretical analysis. Since the active disturbance
rejection control (ADRC) was originally proposed by Han in 1998
[1],1 ADRC has become quite attractive to applied researchers even
though theoretical justification was lagging behind for quite some
time. This is because its uniqueness in concepts, simplicity in
engineering implementation, and superior performance were read-
ily translated into something valuable in engineering practice: the
ability in dealing with a vast range of uncertainties, great transient
response, easy implementation and energy savings, to name a few.
The range of applications seems broad, covering almost all domains
of control engineering, such as motion control, web tension regula-
tion, DC–DC power converter, chemical processes, micro-electro-
mechanical systems gyroscope, speed control of induction motor
and permanent-magnet synchronous motor, attitude tracking of
rigid spacecraft, robotic system, control for noncircular turning
process, control for superconducting RF cavities, stabilization of
axial flow compressors, flight control, low-velocity compensation
$
Supported by the National Center for Mathematics and Interdisciplinary
Sciences, Chinese Academy of Sciences.
n
Corresponding author.
E-mail address: wenchaoxue@amss.ac.cn (W. Xue).
1
The original papers on ADRC in 1980s–1990s, such as [1,22,28,29], are mainly
published in Chinese. Their main ideas and results are summed up in the English
references [23–25].
boiler–turbine–generator control systems, and the fractional-order
system [2–19]. In the USA, ADRC has been deployed at a Parker
Hannifin Extrusion Plant in North America, resulting in over 50%
energy saving per line across ten production lines [20]. And Texas
Instrument has licensed the ADRC technology and the ADRC algo-
rithms will be embedded in a new generation of control chips [21].
More and more reports about the successful applications of ADRC
stimulate us to further think of what is unique about the concept of
ADRC? What is the contribution of ADRC for control science? What
are the theoretical foundations of ADRC? Making the essence of the
methodology clear will help us to better understand ADRC and to
flexibly use its concept in solving the challenging problems of today.
Moreover, clearing the kernel of ADRC may inspire more innovations.
It is primarily for this reason that this paper is written. The paper is
organized around these topics. The unique ideas to promote the
development of ADRC are illuminated in Section 2. Several break-
throughs in the investigation of nonlinear uncertain systems, made
possible by ADRC, are also discussed in Section 2. The theoretical
analysis for ADRC, which more and more researchers and also
engineers are concerned with, is reviewed in Section 3. The last
section is devoted to conclusions and discussions on further research.
2. Unique ideas inside ADRC
ADRC is developed mainly based on Han's two fundamental
breakthroughs: (1) the canonical form that transcends the

http://dx.doi.org/10.1016/j.isatra.2014.03.003
0019-0578/& 2014 ISA. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
2 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

boundary of linear and nonlinear systems; and (2) the concept of and the state feedback control law
extended state or total disturbance that lumps the vast kinds of
1
uncertainties and disturbances, including the uncertainties in the u¼ ð  aðzÞ þ vÞ; aðzÞ ¼ Lnf hðxÞ; bðzÞ ¼ Lg Lnf  1 hðxÞ; ð3Þ
bðzÞ
plant model and external unknown disturbance. The foundation of
such that the nonlinear system (1) can be exactly linearized to the
ADRC rests on the two best offerings of modern control theory: the
following canonical form of cascade integrators:
idea of canonical form and that of state observer.
(
Z_ ¼ AZ þ Bv
2.1. Canonical form of linear and nonlinear system ð4Þ
y ¼ CX

In his paper on the structure of feedback system in 1981 where

[22], Han pointed out an important fact about systems with 2 3 2 3
0 1 0
feedback: dynamic systems, linear or nonlinear, under some 6 7 6⋮7
6 0 7 6 7
conditions can be transformed into the canonical form of A¼6 7; B ¼ 6 7; C ¼ ½1 0 … 0:
4 ⋱ 15 405
cascade integrators via feedback. Along this line of thoughts,
Han took some big strides forward on controlling of non- 0 1
linear systems in the 1980s–1990s:
(a) The boundary between the linear and the nonlinear system Remark 1. The definition of Lg Lkf hðxÞ is the same with that in [27].
can be broken by control input. With control, a linear system can
become a nonlinear one, and vice versa. More essentially, with the Since the nonlinear system is usually much more complex than
help of control, the invariant elements of a system are some the linear system, the linearization methods, which are invented
integrators and the information channels connecting them. Hence, based on the exact model information, cannot avoid the computa-
for vast kinds of systems, no matter they are linear time-invariant tion burden. More seriously, since uncertainties widely exist in
or nonlinear time-varying or coupled, we can transform them into practical systems, the model-based control design met great
a linear decoupled chain of integrators via designing control input. challenges in engineering practice, especially the problem of
(b) Even for a linear time-invariant system, we can take the robustness. Then, Han turned his attention to an even more
initiative to actively introduce some special kinds of nonlinear significant problem: can this be done without a mathematical
control, for example, non-smooth feedback, to improve the per- model? The second breakthrough is documented in [28,29], which
formance of the closed-loop system. will be illustrated in detail in the next subsection.
The idea (b), designing special kinds of nonlinear controller,
mainly the non-smooth feedback, to achieve remarkable perfor- 2.2. Dynamic linearization via the technique of observer
mance, has been discussed in the papers of Han [23,25] and also in
many other literatures, such as [26]. In this paper, the discussion In [28], Han showed that insisting on having the mathematical
will be concentrated on the idea (a). model of the process for the purpose of controlling is both
During the 1980s–1990s, the nonlinear control system is a hot impractical and unnecessary. He asked if the control laws could
topic. And lots of linearization methods are proposed. If the math- not be derived from the mathematical model, where they came
ematical model of the plant is given, similar ideas can also be found from? These questions and thoughts spurred Han to find his own
in some other linearization methods, for example, the well known path in his subsequent research. Han conceived the brilliant idea
exact linearization via differential geometry [27], whose essence is to of treating the unknown model as a special state (named extended
linearize the nonlinear system into a canonical form of cascade state by Han) and using a special state observer, the extended state
integrators via a nonlinear state transform and nonlinear feedback, observer (ESO), to estimate it in real time. In such a brave and
constructed based on the exact mathematical model of the plant. The elegant stroke, a system, which may be uncertain, nonlinear, and
process can be described by the following example. time varying, is reduced to the canonical form of the linear chain
of integrators. This, in a nutshell, is how the methodology of ADRC
Example 1. Consider the following single-input single-output gradually developed.
(SISO) nonlinear system: The unique features of ADRC will be discussed in detail by the
( following examples.
x_ ¼ f ðxÞ þ gðxÞu
ð1Þ Example 2. Consider the following multi-input multi-output
y ¼ hðxÞ
(MIMO) uncertain nonlinear time-varying system:
where x A Rn is the state vector, yðtÞ A R is the output, uðtÞ A R is the 8_
control input, f(x) and g(x) are the n-dimensional vectors of real- >
> X 1 ¼ X2
>
>
>
valued functions of the state x. <⋮
>
If f(x), g(x) and h(x) are real-valued functions of x with X_ n  1 ¼ X n ð5Þ
>
>
continuous partial derivatives of any order. Moreover, if the > X_ n ¼ DðtÞ þ FðX; tÞ þBðX; tÞUðtÞ
>
>
>
relative degree of the nonlinear system (1) is n, i.e. :
Y ¼ X 1 ðtÞ

(i) Lg Lkf hðxÞ ¼ 0 for all x in a neighborhood of certain point x0 and where X i A Rm ; i A n are states, X ¼ ½X T1 …X Tn T ; Y(t) is the output;
all k o n 1, UðtÞ A Rm is the control input; DðtÞ A Rm is the unknown external
(ii) Lg Lnf  1 hðx0 Þ a 0, disturbance; FðX; tÞ A Rm and BðX; tÞ ¼ ½B1 ðX; tÞ…Bm ðX; tÞ; Bi ðX; tÞ A
Rm are nonlinear and possibly unknown. BðX; tÞ is nonsingular.
then there exists the change of coordinates
2 3 A large number of engineering systems can be described by the
hðxÞ class (5), for example, the fast tool servo in [10], the cavity dynamics
6 L hðxÞ 7 of the superconducting RF cavities in [13]. And in many papers,
6 f 7
z ¼ ΦðxÞ ¼ 6
6
7;
7 ð2Þ where ADRC is introduced, the class (5) is the most general form
4 ⋮ 5
Lnf  1 hðxÞ used to demonstrate the designing process of ADRC. It is worthy to
be indicated that the system (1) can also be described by (5), and f(x),

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

g(x) and h(x) are even allowed to be unknown. In this case, the system (11) is equivalent to
8
_
X ¼ ½y y…yðn  1Þ T
 ; DðtÞ ¼ 0; FðX; tÞ ¼ Lnf hðxÞ; BðX; tÞ ¼ Lg Lnf  1 hðxÞ: >
> x_ ¼ x 2
> 1
< ∂f ∂f ∂f ∂f ∂f
Assume that the nonsingular matrix BðtÞ is an estimation for x_ 2 ¼ 1 þ 1 f 1 ðt; x1 ; x2 ; DðtÞÞ þ 1 D_ þ 1 f 2 ðt; x1 ; x2 ; DðtÞÞ þ 1 bðt; x1 ; x2 Þu
>
> ∂t ∂x1 ∂D ∂x2 ∂x2
BðX; tÞ, which is known. Using the following linear chain of >
:y¼x
1
integrators
8 ð12Þ
>
> X_ 1 ¼ X 2
>
> If ∂f 1 =∂x2 a 0 and bðtÞð a 0Þ is the estimation for ð∂f 1 =∂x2 Þ
<⋮
ð6Þ bðt; x1 ; x2 Þ, the following system
>
> X_ n  1 ¼ X n
>
> (
: X_ ¼ BðtÞUðtÞ x_ 1 ¼ x 2
n
ð13Þ
x_ 2 ¼ bðtÞuðtÞ
as the canonical form of cascade integrators for the system (5), the
remaining part of the plant, which is different from this canonical can be used as the canonical form for the system (12). Then the
form, is remaining part, which is denoted by x 3 as follows:
DðtÞ þ FðX; tÞ þ ðBðX; tÞ  BðtÞÞUðtÞ 9 X n þ 1 ðÞ: ð7Þ ∂f 1 ∂f 1 ∂f ∂f
x3 9 þ f ðt; x1 ; x2 ; DðtÞÞ þ 1 D_ þ 1 f 2 ðt; x1 ; x2 ; DðtÞÞ
Viewing X n þ 1 ðÞ as the extended state or the “total disturbance” of ∂t ∂x1 1 ∂D ∂x2
the system (5), the system (5) can be regarded as a system of  
∂f 1
cascade integrators with a disturbance or an extended state, which þ bðt; x1 ; x2 Þ  bðtÞ uðtÞ; ð14Þ
∂x2
can be described as follows:
8 is treated as the extended state or the “total disturbance” of the
>
> X_ 1 ¼ X 2
>
> system (12). And a similar ADRC law can be designed as
<⋮
ð8Þ 8
>
> X_ n  1 ¼ X n >
> x^_ ¼ x^ 2  g 1 ðx^ 1  yÞ
> < 1
>
: X_ ¼ X
n n þ 1 ðÞ þ BðtÞUðtÞ x^_ 2 ¼ x^ 3  g 2 ðx^ 1  yÞ þ bðtÞuðtÞ
>
>
: x^_ ¼  g ðx^  yÞ
3 3 1
From [25], the following nonlinear extended-state observer (ESO)
is designed for the system (8):
8 uðtÞ ¼  b
1
ðtÞx^ 3 þ u0 ðt; x^ 1 ; x^ 2 Þ
_
>
>
> X^ 1 ¼ X^ 2  G1 ðE^ 1 Þ
>
>
<⋮ where u0 ðt; x^ 1 ; x^ 2 Þ is the control law which can achieve the
_ ð9Þ
> X^ n ¼ X^ n þ 1  Gn ðE^ 1 Þ þ BðtÞUðtÞ
> satisfied performance for the canonical system (13).
>
>
>
: ^_
X n þ 1 ¼  Gn þ 1 ðE^ 1 Þ Remark 2. Although the uncertainty f 1 ðt; x1 ; x2 ; DðtÞÞ in the system
(11) does not satisfy the matched condition, it is not a headache for
where E^ 1 ¼ X^ 1  X 1 . Gi ðE^ 1 Þ is a vector of nonlinear function to be
achieving the control object via ADRC. The key factor which will
designed such that X^ i ðtÞðiA nÞ will approach Xi(t) and X^ n þ 1 ðtÞ will
influence the output is the total disturbance x 3 rather than mere
approach the extended state X n þ 1 ðtÞ. Based on the on-line estimation
f 1 ðt; x1 ; x2 ; DðtÞÞ. Hence, from the point of control, what is needed
of the “total disturbance” or the extended state, an ADRC can be
to be estimated and compensated for is x 3 . It is not needed to
designed as
estimate f 1 ðt; x1 ; x2 ; DðtÞÞ itself.
1
UðtÞ ¼  B ðtÞX^ n þ 1 þU 0 ðt; X^ 1 ; …; X^ n Þ; ð10Þ
We use this example to further reveal the essence of ADRC's
where U 0 ðt; X 1 ; …; X n Þ is the control law which can achieve the methodology: if certain uncertainty or disturbance will not influ-
satisfactory performance for the canonical system (6). ence the output to be controlled, it is not needed to reject it. And
The system (5) was usually misconstrued as the only form on the disturbance or uncertainty which affects the output can
which ADRC can be used. And the wrong conclusion was obtained definitely be observed from the output. In other words, the
that ADRC could only deal with the uncertainties satisfying the extended state or the “total disturbance” is always observable.
matched condition. Actually, it is not the case. Let us study how to That is the fundamental mechanism of ADRC to estimate and
design ADRC for the following SISO system which contains the compensate for the “total disturbance” via ESO [30].
uncertainty not satisfying the matched condition. The ADRC design in Example 3 can be further generalized to the
following nonlinear uncertain system:
Example 3. Consider the system
8 8_
_ > X 1 ¼ F 1 ðt; X 1 ; X 2 ; DðtÞÞ
< x 1 ¼ f 1 ðt; x1 ; x2 ; DðtÞÞ
> >
>
>
>
x_ 2 ¼ f 2 ðt; x1 ; x2 ; DðtÞÞ þ bðt; x1 ; x2 Þu >
< ⋮
ð11Þ
>
:y¼x X_ n  1 ¼ F n  1 ðt; X 1 ; …; X n ; DðtÞÞ ð15Þ
1 >
>
> X_ n ¼ F n ðt; X 1 ; …; X n ; DðtÞ; UðtÞÞ
>
>
>
where xi A R; i ¼ 1; 2 are the states, y(t) is the output to be :
Y ¼ X 1 ðtÞ;
controlled, u(t) is the control input, DðtÞ A Rp is the unknown
external disturbance, f i ðt; x1 ; x2 ; DðtÞÞ ði ¼ 1; 2Þ and bðt; x1 ; x2 Þ are where X ¼ ½X T1 …X Tn T ðX i A Rm ; i A nÞ is the state, Y(t) is the output to
nonlinear functions, possibly unknown. bðt; x1 ; x2 Þ a 0. be controlled, UðtÞ A Rm is the control input, DðtÞ A Rp is the
unknown external disturbance, F 1 ðt; X 1 ; X 2 ; DðtÞÞ; …; F n ðt; X 1 ; …;
In this system, the uncertainty f 1 ðt; x1 ; x2 ; DðtÞÞ does not satisfy
X n ; DðtÞÞ may be unknown. ð∂F i =∂X i þ 1 Þði A n  1Þ and ∂F n =∂U are
the matched condition. To design ADRC, let us first study what
nonsingular.
kind of canonical form of cascade integrators is suitable for the
Many practical nonlinear systems discussed in the existing
system and what is the “total disturbance” to be estimated and
literatures can be described by (15). And the condition of
compensated for. Setting
ð∂F i =∂X i þ 1 Þ ði A n  1Þ and ∂F n =∂U being nonsingular is a guarantee
x 1 ¼ y1 ; x 2 ¼ f 1 ðt; x1 ; x2 ; DðtÞÞ; of some kind of controllability for the system (15).

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
4 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

2.3. Extended discussion ^

 aðxÞÞu in [31]. Though it can be proved that, when (17) and
(18) are properly designed, xn þ 1 and ηn þ 1 are close, the physical
From the above discussion, the basic idea of ADRC is quite meaning of sb and the control law (17) and (18) is more clear in the
straightforward: frame of ADRC.
On the other hand, the contribution of Freidovich and Khalil
(1) select the canonical form of the plant in the form of cascade [31], which should be credited, is to prove that (17) and (18) based
integrators; closed-loop system recovers the performance designed for the
(2) treat the part of plant, that is different from this canonical linear canonical system (19), that is, the performance of the
form, as the “total disturbance”, which may include the system:
external disturbance from the environment and the internal
x_ n ¼ Axn þ Bϕðxn Þ:
disturbance, a term coined by Han, representing the uncer-
tainties and variations in the dynamics of the plant; This analysis demonstrated that the control law (17) and (18) has
(3) reduce the system, which may be uncertain, nonlinear, the capability for not only achieving stabilization or regulation, but
coupled with other process variables, and time varying, to also shaping the transient response. Since (17) and (18) are a special
the canonical form by means of estimating and compensating form of ADRC (9) and (10), this work conversely greatly pushed the
for the “total disturbance” via a special observer conceived by analysis on ADRC. This will be further discussed in Section 3.
Han, the extended state observer (ESO). Let us continuously study the controller in [32], where the
following nonlinear system is considered:
Next, we use the ADRC strategy to discuss the physical 8
> z_ ¼ qðz; yÞ;
>
>
mechanisms of two other methods proposed for nonlinear uncer- >
> x_ 1 ¼ x2 þ δ1 ðz; x1 Þ;
>
>
tain systems. One is the extended high-gain observer (EHGO) >
>
>
> ⋮
based control in [31], and the other is the method in [32], which is <
based on a special observer. x_ i ¼ xi þ 1 þ δi ðz; x1 ; …; xi Þ; ð20Þ
>
>
Freidovich and Khalil [31] considered the following nonlinear >
> ⋮
>
>
>
> x_ n ¼ u  un þ δn ðz; x1 ; …; xn Þ;
uncertain system: >
>
>
:
8 y ¼ x1 ;
_
< x ¼ Ax þ Bðbðx; z; wÞ þ aðx; z; wÞuÞ
>
_z ¼ f 0 ðx; z; wÞ ð16Þ where z A Rn0 ; x A Rn are the states, y(t) is the measured output,
>
: y ¼ Cx;
u(t) is the input to be designed, un is the unknown constant.
qðÞ; δi ðÞði A nÞ are unknown nonlinear function. The z-system is
where ðA; B; CÞ represents the canonical form of cascade integra-
input-to-state stable (ISS).
tors:
Denoting xn þ 1 ¼  un , Jiang and Praly [32] proposed the follow-
2 3 2 3
0 1 0 ing observer and the controller:
6 0 7 6⋮7 8_
6 7 6 7 > x ¼b
b x 2 þ β1 ðx1  b x 1 Þ;
A¼6 7; B ¼ 6 7; C ¼ ½1 0 … 0; >
> 1
4 ⋱ 15 405 >
>
>
<⋮
0 1
x_ i ¼ b
b x i þ 1 þ β i ðx1  b
x 1 Þ; ð21Þ
>
>
n p1
x A R and z A R are the states, y is the measured output, u is the >
> ⋮
>
>
:_
control input, wðtÞ A Rp2 is a disturbance input. aðÞ; bðÞ and f 0 ðÞ are x ¼b
b n x nþ1 þ u þ β ðx  b
n 1 1 x_
x Þ; b ¼β
nþ1 ðx  b
x Þ; :
nþ1 1 1
the nonlinear functions possibly unknown. aðÞ 4 0, and z(t) is
assumed to be bounded.  
y b x2 b
xn
The EHGO and control law were designed as u ¼ b
x n þ 1  r n F col ; ; …; n þ a  1 ;
r ar r ar þ 1 r r

x^_ ¼ Ax^ þ Bðs^ þ bð

^ xÞ ^ xÞuÞ
^ þ að ^ þ HðεÞðy  C xÞ;
^ r_ ¼  rðbr  sðy; rÞÞ; ð22Þ
α 
_
s^ ¼ nþ1
^
ðy C xÞ; ð17Þ where β i ¼ β i r i , sn þ 1 þ Σ
nþ1
β i sn þ 1  i is stable, F, ar, b and the
i¼1
εn þ 1 function s are the appropriately chosen controller parameters.
^ xÞ It is obvious that the control law (21) and (22) is also a special
 s^  bð ^ þ ϕðxÞ
^
u¼ ; ð18Þ form of the ADRC (9) and (10). However, the system (20), which is in
^ xÞ
að ^
a dominant triangular structure and contains the uncertainties
where aðÞ ^ ^
and bðÞ are twice continuously differentiable, globally qðÞ; δi ðÞði A n  1Þ not satisfying the matched condition, is much
bounded functions that model aðÞ and bðÞ, respectively. HðεÞ ¼ more complex than the system (5). What is the mechanism of the
½α1 ε  1 ; …; αn ε  n , ε; αi ði A n þ1Þ are the parameters of EHGO. ϕðxÞ control law (21) and (22)? Can it be affirmed that the outputs
is a twice continuously differentiable state feedback control law such b
x1; bx 2 ; …; b
x n and b
x n þ 1 of the observer (21) are the estimations of
that the origin of the closed-loop canonical system x1 ; …; xn and xn þ 1 ? The answer is NO. To make it more clear, denote
x_ ¼ Ax þ BϕðxÞ ð19Þ x i ¼ xi þ δ i  1 ðz; x1 ; …; xi  1 Þ; i A n;
is locally exponentially stable and globally asymptotically stable. δ 0 ðÞ ¼ 0; δ i ðz; x1 ; …; xi Þ ¼ δi ðz; x1 ; …; xi Þ þ δ_ i  1 ðz; x1 ; …; xi  1 Þ; i A n:
According to the above discussion on the methodology of ð23Þ
ADRC, the control law (17) and (18) is a special form of ADRC (9)
b and a
and (10). Since bðÞ bðÞ are the estimations for the unknown The system (20) can be transformed to the following cascade
functions bðÞ and aðÞ, it is easy to find out that, to achieve integrator system:
controlling of y, the “total disturbance” to be handled is 8
>
>
> x_ 1 ¼ x 2 ;
>
<⋮
bb
xn þ 1 9 bðx; z; wÞ þ aðx; z; wÞu  bð b ðb
xÞ  a x Þu:
: ð24Þ
>
> x_ n  1 ¼ x n ;
Moreover, the output sb of the EHGO (17) is to estimate this total >
>
:_
^ þ ðaðx; z; wÞ
disturbance xn þ 1 rather than ηn þ 1 9 bðx; z; wÞ  bðxÞ x n ¼ x n þ 1 þ u; y ¼ x 1 ;

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

which is disturbed by the total disturbance

x n þ 1 ¼ δ n ðz; x1 ; …; xn Þ  un :
From Example 2, we guess that the observer (21) is actually an
ESO for the system (24). Moreover, the outputs of (21) observe the
states of the system (24) and the total disturbance x n þ 1 ¼ δ n
ðz; x1 ; …; xn Þ  un rather than the states of the system (20) and
x n þ 1 ¼  un .
The guess will be confirmed by the following example.

Example 4. A simplified pitch dynamics in the aircraft motion can

be described as
8
>
> z€ 1 ¼  z1 ;
>
>
>
> Q
< α_ ¼ ωz 
> ða11 þz1 Þα þ a10 g;
mV
ð25Þ
>
> Q Q
>ω
>
>
_ z ¼ u  un þ ða21 þ z1 Þα þ a22 ωz ;
Jz Jz
>
>
: y ¼ α;

where the measured output is the attack angle y¼ α, ωz is the Fig. 1. The outputs of the observer (27), the states of the systems (25) and (26),
pitch rate, m is the mass of the vehicle, Q is the dynamic pressure, x 3 and  un .
V is the velocity of the center of mass, g is the Gravitational
acceleration, Jz is the moment of inertia, u is the control input, un is
Jiang in 1998 [33]. Although the frame of ADRC has been devel-
the external disturbance, a10, a11, a21 and a22 are the coefficients, z1
oped in the early 1990s, in about ten years, the research on the
is the bounded disturbance to the coefficients a11 and a21.
theoretical analysis for ADRC was progressing haltingly amid great
Set x 1 ¼ α; x 2 ¼ ωz ðQ =mVÞða11 þ z1 Þα þ a10 g. The system (25) difficulties. Freidovich and Khalil [31] and Praly and Jiang [32,33]
can be described by the following cascade integrator system: provided some enlightening results for the analysis of ADRC. This
( will be further discussed in Section 3.
x_ 1 ¼ x 2 ;
ð26Þ
x_ 2 ¼ u þ x 3 ðz1 ; z_ 1 ; α; ωz Þ; Remark 3. The methodology of ADRC can be flexibly used. For
some systems whose uncertainties are complex, like the system
which is disturbed by the total disturbance x 3 :
(15), a set of ESOs can be designed to gradually and approximately
Q Q Q transform the system into the canonical form of the cascade
x 3 ðz1 ; z_ 1 ; α; ωz Þ ¼  un þ ða21 þ z1 Þα þ a22 ωz  ð  z_ 1 α þða11 þ z1 Þx 2 Þ:
Jz Jz mV integrators, based on identifying the physical characteristics of
According to (21) and (22), the observer is designed as the system. For example, if x2 of the system (11) is measurable,
8_ then two ESOs can be designed to estimate the uncertainties in
>
> x ¼b
b x 2 þ β 1 ðy  b
x 1 Þ; x1-subsystem and x2-subsystem, respectively. And the canonical
< 1
_x ¼ b
b x 3 þ u þ β2 ðy  b x 1 Þ; ð27Þ form of the cascade integrators is composed by two integrators,
2
>
> which can be dynamically and approximately developed one by
: b_
x ¼ β ðy  b
3 x Þ; 1
3 one, first from x2 to x1, then from u to x2. Xue and Huang [34]
and the control input is designed to be studied this strategy for MIMO lower-triangular systems in detail.

r2 f 1 y r2 f 2 b
x2 The above examples demonstrate that the essence in employ-
u ¼ b
x3   a þ1 : ð28Þ
r ar r r ing ADRC as a tool in dealing with largely uncertain processes,
which may be nonlinear, time-varying and full of interconnected
In the simulation, the parameters of (27) and (28) are chosen as
dynamics, is to characterize the “total disturbance” that affects the
s3 þ β 1 s2 þ β 2 s þ β 3 ¼ s3 þ β 1 rs2 þ β 2 r 2 s þ β 3 r 3 ¼ ðs þrÞ3 ; process output. Regarding the detailed technique for estimating
r ¼ 35; f 1 ¼ 0:1; f 2 ¼ 1:5; ar ¼ 1: the “total disturbance”, lots of methods for designing observers or
filters can be utilized.
The simulation conditions are
8 Unmodeled dynamics, parametric variations as well as external
>
> Q Q Q disturbances widely exist in most realistic control systems. This is
>
> ¼ 0:2; a10 g ¼ 0:01; a21 ¼  17:8; a22 ¼  0:2; un ¼ 0:5;
>
> mV Jz Jz one of the main reasons why disturbance rejection or control of
>
<
10 systems with uncertainty is the most fundamental issue in control
z1 ðt 0 Þ ¼ 0; z2 ðt 0 Þ ¼ 0:8; αðt 0 Þ ¼ ; ωz ðt 0 Þ ¼ 0;
>
> 57:3 science. Substantial efforts have been made and much progress
>
>
>
> 10 has been achieved, such as adaptive control, robust control, DOBC,
>
:bx 1 ðt 0 Þ ¼ ;bx 2 ðt 0 Þ ¼ 0; b
x 3 ðt 0 Þ ¼ 0:
57:3 and sliding-mode control. Next, the breakthroughs in the investi-
ð29Þ gation of uncertain systems, made possible by ADRC, are briefly
discussed.
The outputs of the observer (27), the states of the systems (25)
and (26), x 3 and  un are compared in Fig. 1. It shows that b x2
approaches the state x 2 rather than ωz, and b
x 3 approaches the total 2.4. ADRC and controlling of uncertain systems
disturbance x 3 rather than the external disturbance  un . Hence,
the physical meaning of the control law (21) and (22) is clearly Basically, the uncertainties stem from two sources: internal
revealed in the frame of ADRC. (parametric or unmodeled dynamics) uncertainties and external
The contribution of [32], which should also be credited, is that (disturbance) uncertainties.
the convergence of the closed-loop system was proved in the Adaptive control and robust control are two main areas where
paper. Actually, a related result has been proposed by Praly and internal parametric and/or bounded unstructured uncertainties of

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
6 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
the plant are the main concern of the controller design. An
adaptive controller is usually the controller with adjustable para-
meters and a mechanism for adjusting the parameters [35,36].
Most proposed adaptive controllers focus on linear or nonlinear
systems with linear unknown parameters. How to design the
adaptive controller for nonlinear uncertain systems, especially,
the nonlinear systems with non-parametric uncertainty is still a
challenging problem. In the robust control, a perturbation model
of the plant is assumed and a controller is sought such that the
closed-loop performance and stability are guaranteed for all plant
perturbations not greater than a prespecified bound [37,38]. In the
frame of robust control, the controller itself is not on-line adjusted
according to the uncertainty. It is the worst case approach. Hence
the results are usually conservative and the uncertainties to be
handled are limited.
Dealing with the external disturbance is an old but unceasingly
developing field. In the early development of control theory, the
integrate controller and the PID controller were originally pro-
posed to eliminate the error caused by the constant external
disturbance [39,40]. For more complex disturbance, the invariance
theory (IT) of the 1940s [41], the internal model principle (IMP)
[42] and the disturbance-accommodation control (DAC) of the
1970s [43] were proposed to compensate for the disturbance by
either utilizing the sensor to measure it or the mathematical
model of the disturbance. Since the 1980s, more and more
disturbance rejection methods have been proposed, such as, the
disturbance observer based control (DOBC) [44–46], the time-
delayed estimator (TDE) [47], nonlinear output regulator theory
[48], Embedded Model Control (EMC) [49,50], and nonlinear
disturbance observer based control (NDOBC) [51]. Lots of these
methods share the structure of the two degree of freedom with
one to achieve disturbance rejection and the other to regulate the
closed-loop characteristics. Hence the controllers designed in this
way usually can not only achieve stabilization or regulation, but
also shape the transient response. ADRC also adopts this structure.
However, the paths to deal with the disturbance are quite different
in these methods and ADRC. In most disturbance rejecting
methods, the plant model, and sometimes the model of the
disturbance, is needed to reconstruct the disturbance and com-
pensate for it. ADRC brings a brand new meaning for the concept
of disturbance, which lumps the external disturbance and the
uncertainty in the plant model. The breakthrough in the concept
makes ADRC less dependent on the model. If the disturbance
satisfies the matched condition, Xue and Huang [52] demonstrated
that there exists some similarities between linear ADRC and linear
DOBC. And the two degree of freedom structure in both controller
designs provided the ability to achieve both the stability and the
satisfactory transient performance. Furthermore, from this com-
parison, the theoretical results of ADRC in [52] suggested that
DOBC can safely be generalized for some kind of systems with
both model uncertainty and disturbance. However, if the distur-
bances are not satisfied the matched condition, the strategy of
ADRC and DOBC is quite different. ADRC emphasizes on the “total
disturbance” affecting the output process rather than the distur-
bance entered at its original site. It is in this way, ADRC keeps the
characteristics of less model depending and of the two degree of
freedom controller design to realize both stabilization and shaping
transient response.
Remark 4. In the early research on DOBC, the disturbance to be
estimated is usually assumed to be constant or bounded. In recent
years, the research on DOBC is also active, and the disturbance to
be observed has been extended to include the external disturbance
which satisfies certain kind of model and part of uncertain non-
linear model of the plant [51]. It is reasonable that, if the physical
Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
conditions are permitted, rational using the reliable model infor-
mation in controller design can help to improve the performance
of the control system. Then what kinds of model information are
useful for control design and how to utilize them are the inter-
esting issues needing further investigation.
One other outstanding control method for uncertain system is
the sliding mode control, where the uncertainty is handled by the
high-gain feedback in order to guarantee the accessibility of the
sliding mode. And the system will be insensitive to both model
uncertainty and disturbance,when it is kept on the sliding mode.
The high-gain feedback may cause chattering in practice. ADRC
does not use the high gain feedback. The real value of the “total
disturbance” is estimated and compensated by ESO. Hence, if the
“total disturbance” is large or its dynamic is quick, usually ESO
should be designed having quick response. The quick response can
be realized by the technique of the high-gain observer or non-
linear strategy. The performance difference of these strategies is
also an interesting issue.
Remark 5. As a central issue in control science, lots of theories
and methods have been developed for dealing with uncertainties
in control systems. Most of them are still the dynamic fields. New
progress is continuously appearing by absorbing the advantages
from each other. Therefore, it is not possible to give a survey of the
entire landscape.
From the above discussion, the main outstanding characteris-
tics of ADRC can be concluded as
(1) ADRC can deal with vast range of uncertainties.
(2) ADRC does not heavily depend on the model. Hence the structure
of the controller is simple. And the satisfactory transient response
is hopefully achieved due to the strategy of ADRC.
These characteristics provide the possibility for overcoming the
difficulty of controlling the uncertain system. However, the grand
and ambitious scheme of ADRC also met some doubts since it was
proposed: is it really able to handle the vastly uncertain nonlinear
systems in practice? Is its way of rejecting the “total disturbance”
too rude to get high precision control? Is there any theoretical base
for this idea?
Though theoretical justification was lagging behind for quite some
time, reports of successes roll in from many application fronts. The
range of applications is board, covering almost all domains of control
engineering. [2–21] are a few samples. Huang et al. [2] and Sun et al.
[15] demonstrated ADRC as applied to the high performance flight
control problem where the plant dynamics was nonlinear and was
treated as largely unknown. It was shown convincingly that effective
control of a complex system could be attained in the absence of
detailed and accurate mathematical model. Wu and Chen [10] applied
ADRC for noncircular turning process (NCTP) to test the precision of
ADRC. The experimental results demonstrated that the tracking errors
are less than 3 μm for different cutting parameters. Ohioepolymer
News [20] highlighted the successful deployment of the ADRC-based
control technologies, as seen at a Parker Hannifin Extrusion Plant in
North America, resulting in over 50% energy saving per line across ten
production lines.
Stimulated by the news of increasingly more successful appli-
cations, the theoretical research for ADRC has been gradually set.
Some latest progress is introduced in the next section.
3. Theoretical analysis
The research on the theoretical analysis for ADRC was progres-
sing haltingly amid great difficulties, which can be traced to the
 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
grand scheme of ADRC. The uncertain systems of concern are
allowed to be nonlinear, time-varying, multi-input and multi-
output; the disturbance signal or the reference signal may not
even be continuous; and the state feedback in ADRC is, in general,
nonlinear. Such great ambition is equally matched by the great
challenge in theoretical analysis for the purpose of explaining and
justifying the engineering success: how to rigorously find the
capacity of ADRC in dealing with vastly uncertain nonlinear
systems? How to rationalize the outstanding performance in the
presence of great uncertainties shown in numerous applications?
What are the limits on the scope of applications for ADRC?
In recent years, the theoretical research has been gradually set in
motion, targeting first various components of ADRC, such as the
tracking-differentiator (TD) [53–56] and the extended-state obser-
ver (ESO) [57–61], eventually arriving at, finally, the property of the
entire ADRC-based closed-loop system [34,52,62–66,68].
TD was first proposed in [69] in the general nonlinear form to
provide the filtered version of the input signal and its differentiation.
Unfortunately, the proof in [69] was incomplete. In recent years,
further analysis on the convergence of TD have been obtained. In
[54], the convergence was proved for a second order linear TD with
any differentiable input signal and random perturbation. Xue et al.
[55] surveyed several kinds of differentiators. The comparison
showed that the tracking-differentiator (TD) permits a weaker
condition on the stability of the systems, which can be used to
construct differentiators. Moreover, TD sets weaker constraints on
the input signal. The proofs for two kinds of nonlinear TD were also
given in [55]. In [56], a rigorous proof of the convergence was given
for the nonlinear TD for both two-dimensional and high-dimensional
cases under some additional assumptions. By now, the analysis on TD
mainly emphasized the convergence. Further study on its advantages
in anti-chattering and noise tolerance is still to come.
Since the “total disturbance” to be estimated by ESO may be a
function of both the closed-loop state, the disturbance and the real
control input, the analysis on ESO is intricately connected with
that on the ADRC-based closed-loop systems in which it is used.
Remark 6. ESO can also be used separately from any control loops,
in which case the signals it estimates can be assumed bounded. In
this case, ESO functions as a filter or differentiator and it should be
analyzed as such. The research is still in its early stage.
The theoretical analysis on ADRC has been carried out from
several different perspectives.
To render the idea clear and controller easy to implement and
to tune, a parameterized linear ADRC (LADRC) was proposed in
[72], where nonlinear gains are replaced with linear ones and
tuning is reduced to the adjustment of one variable, the band-
width, which is the common terminology used by engineers.
LADRC has already shown great promise in many areas of control
engineering. The capability of LADRC has been analyzed.
In [67], the capability of LADRC for linear time-invariant SISO
minimum-phase systems with unknown but bounded relative
degrees, and unknown input disturbances, was analyzed. The
result explains why one ADRC with fixed parameters can be
applied to a group of plants of different orders, relative degrees,
and parameters. In [64,65], the analysis results in the frequency-
domain were shown. In [64], the loop gain frequency response was
analyzed for a second-order linear time-invariant plant. The result
showed that the LADRC based control system possesses a high
level of robustness. The bandwidth and stability margins, in
particular, are nearly unchanged as the plant parameters vary
significantly; so is the sensitivity to the input disturbance. Such
characteristics explain why LADRC is an appealing solution in
dealing with real world control problems where uncertainties
abound. Xue and Huang [65] further investigated the frequency
Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
7
properties of LADRC with the reduced-order LESO for a typical
class of n-th order linear time-invariant uncertain system. It was
shown that the phase margin and crossover frequency are almost
unchanged in the presence of some uncertain parameters. And
moreover, different kinds of uncertain parameters have various
influences on the robustness of the ADRC based control system.
The main results of [65] will be introduced in Section 3.2.
Vast successful application research also revealed the capability
of LADRC for nonlinear uncertain systems. However, the theoretical
justification was lagging behind for quite some time. In [63], the
stability of the closed-loop system under LADRC was studied with
the assumption that the “total disturbance” to be estimated and
compensated for is bounded. Since the “total disturbance” to be
estimated by ESO may be a function of both the closed-loop state,
the disturbance and the real control input, the boundness of the
“total disturbance” is hard to be estimated in advance. As what has
been introduced in Section 2, because of the similarity between
LESO and the observers in [31–33], these literatures provided some
enlightening results for the analysis of ADRC, particularly, relaxing
the assumption on the boundness of the “total disturbance” and
analyzing the performance recovery of the closed-loop system. Xue
and Huang [52,66,34] further discussed the transient performance
and the stability of LADRC design for nonlinear uncertain systems in
the cases of SISO systems, discontinuous disturbance and MIMO
block lower-triangular systems, respectively. The main results of
[66] will be introduced in Section 3.3. Furthermore, the idea of
LADRC has been employed to stabilize some classes of the infinite-
dimension systems with external disturbance [70,71].
The parameters of LADRC, implemented digitally, are restricted
by the sampling rate. Xue and Huang [68] studied the parameters
tuning and the capability of LADRC under a fixed sampling rate.
The quantitative relationship among the parameters of LADRC, the
sampling rate and the uncertainty to be dealt with was given.
The original form of ADRC ((9) and (10)) is a much more
general nonlinear controller. The theoretical analysis of the non-
linear ADRC (NADRC) is extremely hard. Guo and Zhao [59,62]
opened the theoretical analysis of nonlinear ESO (NESO) and ADRC
(NADRC) for nonlinear time varying uncertain systems. The con-
vergence of NADRC was proved for a class of MIMO nonlinear
systems with large uncertainty that comes from both dynamical
modeling and external disturbance [62].
The high level of robustness and the superior transient perfor-
mance are the most valuable characteristics of ADRC to make it be
an appealing solution in dealing with real world control problems.
Next, the main results of [65,66], which can explain these
characteristics from the views of the frequency-domain and the
time-domain, respectively, will be introduced. The analysis is
carried out for the reduced-order linear ESO (RLESO) based LADRC.
The structure of RLESO will be introduced first.
Remark 7. The theoretical analysis of ADRC has been well devel-
oped for vast kinds of uncertain systems and from various views.
The details of all related theoretical work can be obtained from
these paper's references and those therein.
3.1. RLESO based LADRC
We consider the system (5) in the case of SISO
8
> x_ 1 ¼ x2 ;
>
>
>
> ⋮
>
<
x_ n  1 ¼ xn ; t Z t 0 ; y; u A R ð30Þ
>
>
>
> x_ n ¼ dðtÞ þ f ðx; tÞ þ bðx; tÞuðtÞ;
>
>
: y ¼ x ðtÞ;
1
8 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

In [72], the linear ESO (LESO), which is based on the nonlinear We study the closed-loop performance of t Z t u when the
form (9), was proposed, i.e., control begins to work.
Denote Y(s) and U(s) as the Laplace transformations of y(t) and
e i Þ ¼ βi b
Gðb ei; iA n þ 1:
u(t), respectively. Combination of (31) and (34) leads to the
Since y ¼ x1 is available, only the estimation of xi ; i Z 2 are needed. transfer function from Y(s) to U(s)
Hence, this paper will use the following reduced-order LESO [66]:
1 ∑i ¼ 0 ð∑q ¼ 0 K i þ q β n  q Þs
n ni ni
8 8 UðsÞ
> <  β z2  β2 x1  β buðtÞ if n ¼ 1 ¼ CðsÞ; CðsÞ ¼ : ð36Þ
>
>
> z_ 2 ¼ 1 1 1 YðsÞ b ∑ni ¼ 01 ð∑iq ¼ 0 K q βi  q Þsn  i
>
>
>
> :  β 1 z2 þ z3 þ ðβ2  β 21 Þx1 if n 4 1
>
> The transfer function of the plant (35) is
>
>
> z_ 3 ¼  β z2 þ z4 þ ðβ  β β Þx1
< 2 2 1 3 b
⋮ ð31Þ YðsÞ ¼ PðsÞUðsÞ; PðsÞ ¼ : ð37Þ
>
> sn þ a1 sn  1 þ ⋯ þ an  1 s þan
>
>
>
>
>
z_ n ¼  β n1 2z þ z nþ1 þ ðβ n  β β Þx
1 n1 1 þ bðtÞuðtÞ
Therefore, the loop transfer function of the ADRC based closed-
>
>
>
>
> z_ n þ 1 ¼  β n z2  β1 βn x1 loop system is
>
: x^ i ¼ zi þ β i  1 x1 ; i ¼ 2; 3; …; n þ 1
LðsÞ ¼ CðsÞPðsÞ:
In order to simplify the tuning of ESO (31), its characteristic Let ωn denote the crossover frequency of L(s) and θn denote the
polynomial is set as phase margin of L(s), then
jLðjωn Þj ¼ 1:
n n
sn þ ∑ βi sn  i ¼ sn þ ∑ ϕi ωie sn  i ¼ ðs þ ωe Þn ; ωe 4 0; ð32Þ ð38Þ
i¼1 i¼1
θn ¼ 1801 þ ∠Lðjωn Þ: ð39Þ
where ϕi ¼ n!=ði!ðn iÞ!Þ, and ωe represents the bandwidth of ESO
(31), being the only parameter to be tuned. The initial condition of ESO Usually, ωn is an important index which reflects the bandwidth
(31) is chosen as x^ i ðt 0 Þ ¼ 0; i Z 2 by taking zi ðt 0 Þ ¼  β i  1 x1 ðt 0 Þ. of the control system and θn implicates how much phase delay the
LADRC is designed as system can tolerate. Hence, the engineers can judge the system
performance according to the information of ωn and θn. Since L(s)
 kn x1 kn  1 x^ 2 ⋯  k0 x^ n þ 1
uðtÞ ¼ gðx1 ; x^ 2 ; …; x^ n þ 1 ; tÞ 9 ; k0 ¼ 1: ð33Þ has the uncertain parameters a, b, we will study the influences of
bðtÞ
a, b on ωn and θn.
When xi ðt 0 Þ  x^ i ðt 0 Þ ¼ xi ðt 0 Þ a 0; i Z2, large ωe may cause the Define
peaking phenomenon in the output of ESO (31). Hence, xi ðtÞ  x^ i ðtÞ
b βn b ωne
may become very large at the initial period. We use the strategy of L0 ðsÞ ¼ ¼ : ð40Þ
b sn þ∑ni ¼ 11 β i sn  i b ðs þ ωe Þ  ωne
n
no control action at the initial period to eliminate the peaking
phenomenon, i.e., Let ω0 ; θ0 denote the crossover frequency and the phase margin of
( L0 ðsÞ, respectively. Then ω0 ; θ0 satisfy
0; t 0 rt o t u ;
uðtÞ ¼ ð34Þ  
gðx1 ; x^ 2 ; …; x^ n þ 1 ; tÞ; t Z t u ;   b
ω0 ¼ s0 ωe ; ðjs0 þ1Þn  1  ¼ 0; ð41Þ
b
where t u ¼ t 0 þ 2n J P~ e J max fln ωe =ωe ; 0g, P~ e is the unique positive
solution of the Lyapunov equation 1
2 3 θ0 ¼ 1801 þ ∠L0 ðjω0 Þ ¼ 1801 þ ∠ : ð42Þ
ðjs0 þ 1Þn  1
 ϕ1 1 … 0
6 7
6  ϕ2 0 … 7 The following theorem reveals the relationship between the
A~ e P~ e þ P~ e A~ e ¼  I; A~ e ¼ 6
T
7:
6 ⋮ ⋮ … 17 frequency responses of L(s) and L0 ðsÞ.
4 5
 ϕn 0 … 0
Theorem 1 (Xue and Huang [65]). Assume A1 and A2 are satisfied.
3.2. Frequency analysis of the LADRC based closed-loop system
There exists ωn such that for 8 ωe Z ωn
8  
 Lðjω0 Þ
In this subsection, the frequency properties of LADRC with the > 

< L0 ðjω0 Þ  1 r γ 1 =ωe
>
>
reduced-order LESO for a typical class of n-th order linear time-
jωn =ωe  s0 j r γ 2 ðωe Þ ð43Þ
invariant uncertain systems will be investigated. It will be shown >
>
>
: jθ  θ jr γ ðω Þ;
that the phase margin and crossover frequency are almost n 0 3 e
unchanged in the presence of some uncertain parameters. And
moreover, different kinds of uncertain parameters have various where γ 1 is a positive independent of ωe and γ 2 ðωe Þ; γ 3 ðωe Þ are
influences on the robustness of the LADRC based control system. positives satisfying
The system (30) in the case of linear time-invariant is con- lim γ ðωe Þ ¼ 0; lim γ ðωe Þ ¼ 0: ð44Þ
ωe -1 2 ωe -1 3
sidered
8
> x_ 1 ¼ x2 ;
>
>
>
> _
< x 2 ¼ x3 ;
> The results (43) and (44) show that the ADRC based closed-
⋮ t Z t0 ð35Þ loop system has the following properties:
>
> (i) LðjωÞ and L0 ðjωÞ are close at the frequency ω0 ¼ s0 ωe , and
> x_ n ¼  aT x þ bu;
>
>
>
:y¼x ; the error jLðjω0 Þ  L0 ðjω0 Þj decreases as ωe increases. Furthermore,
1
because of jL0 ðjω0 Þj ¼ 1, ðω0 ; θ0 Þ can be used to estimate the
where a ¼ ½an an  1 … a1 T and b are uncertain parameters satisfying ðωn ; θn Þ, i.e., the crossover frequency and the phase margin of L(s).
(ii) (41) and (42) show that ωn is proportional to ωe, and θ0 is
A1. ai A ½  α1 ; α1   ð 1; 1Þ; i A n. independent of ωe. Since ðω0 ; θ0 Þ can be used to estimate ðωn ; θn Þ,
A2. b=b A ½α2 ; α3   ð0; 2n=ðn  1ÞÞ, where b is the nominal value of b, larger ωe generates larger crossover frequency but smaller time-
which is known. delay tolerated by the ADRC based control system. Therefore, the

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

relationship revealed by (41) and (42) can help practitioners to Table 1

design the bandwidth of ESO such that the closed-loop system has The sets of ða1 ; a2 ; bÞ.
a satisfactory stability margin.
C1-1: (  0.3,2.0,0.5) C1-5: (  0.3,2.0,1.25) C1-9: (  0.3,2.0,2.0)
(iii) Since L0 ðsÞ is independent of a, by tuning the ESO's band- C1-2: (1.2,  1.0,0.5) C1-6: (1.2,  1.0,1.25) C1-10: (1.2,  1.0,2.0)
width, the crossover frequency and the stability margin of the ADRC C1-3: (1.5,0.1,0.5) C1-7: (1.5,0.1,1.25) C1-11: (1.5,0.1,2.0)
based system can be almost invariant as a varies. This result reveals C1-4: (  0.5,  1.0,0.5) C1-8: (  0.5,  1.0,1.25) C1-12: (  0.5,  1.0,2.0)
that the performance of the closed-loop system is robust to the
uncertainties a ¼ ½an an  1 …a1 T , which is an advantage of ADRC.
(iv) For the 1st and 2nd system, the explicit expression of ω0 60
and θ0 can be easily deduced 40
8

Magnitude (dB)
< ω0 ¼ ωe b=b; 20
n¼1 : π ;
: θ0 ¼ 0
2
8 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi −20
>
>
> ω ω
2
> 0
> ¼ 2 e
1
þ 1 b
 2;
1 −40
< 4 16 2
b
n¼2 : sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi! ð45Þ −60
>
> −90
>
> θ ¼ π  arcctan 1
þ 1 b
2
 1
: L(s), C1−1
>
: 0 2 4 16 2 2 L(s), C1−2
b
−180 L(s), C1−3

Phase (deg)
L(s), C1−4
Eq. (45) expresses the relationship between ðω0 ; θ0 Þ and b=b, −270
L (s), C1−1,2,3,4

which clearly shows the influence of the uncertain parameter b

on ðωn ; θn Þ. −360

In the case of n¼ 1: the larger b leads to the larger ωn which

−450
means the faster dynamics. On the other hand, b has little 10−1 100 101 102 103
influences on the θn. Frequency (rad/sec)
In the case of n ¼2:
Fig. 2. The Bode Plot for L(s) and L0 ðsÞ under C1-1–4.

 If b o b, then the smaller b, the smaller ωn and the larger θn,

100
which lead to a lower dynamics but larger relative stability.
 If b4 b, then the larger b, the larger ωn and the smaller θn, 50
Magnitude (dB)

which meas the ADRC based control system behaves faster but
is vulnerable to be unstable due to uncertainties, for example, 0
phase delay. The result reveals that the stability margin of the
system is more sensitive to b than a. Hence, in designing −50
LADRC, the information on b will be helpful for designing a
better LADRC. −100
−90
Next, we will illustrate the results of Theorem 1 by a 2nd order −135
LTI uncertain system. −180
Phase (deg)

−225
−270 L(s), C1−5
Example 5 (Xue and Huang [65]). Assume in the system L(s), C1−6
−315
L(s), C1−7
YðsÞ b −360
¼ PðsÞ ¼ 2 ; ð46Þ −405
L(s), C1−8
UðsÞ s þ a1 s þ a2 −450
L (s), C1−5,6,7,8

a1 ; a2 ; b are uncertain parameters which satisfy 10−1 100 101 102 103
Frequency (rad/sec)
ai A ½  2; 2; i ¼ 1; 2; b=b A ½0:5; 3; b ¼ 1:
Fig. 3. The Bode Plot for L(s) and L0 ðsÞ under C1-5–8.
According to (36), the LADRC designed for (46) has the transfer
function of 2. The system with the same a1 ; a2 but different b have different
ðωn ; θn Þ. The larger b, the larger ωn and the smaller θn, which
UðsÞ 1 ðωe þ k1 2
2
ωe þ k2 Þs þ ðk1 ω
2
e þ k2 2
2
ωe Þs þ k2 ω
2
e
¼  CðsÞ; CðsÞ ¼ : coincides with the second equation of (45). The result reveals
YðsÞ b s2 þ ðk1 þ 2ωe Þs
that the stability margin of the system is more sensitive to b
ð47Þ
than a. Hence, in designing ADRC, the information on b will be
To illustrate the results in Theorem 1, we proceed to design helpful for designing a better ADRC.
ωe ¼ 50 and take the sets of the uncertain parameters a1 ; a2 and
b in Table 1. The plants in Table 1 can be classified as three groups Remark 8. The frequency characteristics shown in Theorem 1
by different b, that is C1-1–4, C1-5–8 and C1-9–12. C1-1–4 share explain why LADRC is an appealing solution in dealing with real
b¼ 0.5, thus they have different L(s) but the same L0 ðsÞ. Fig. 2 is world control problems where uncertainties abound. These char-
their Bode plot. Similarly, C1-5–8 share b¼ 1.25 and Fig. 3 is acteristics have also been shown in several experiments (see
their Bode plot. C1-9–12 share b¼ 2 and Fig. 4 is their Bode plot. [3,10]).
Figs. 2–4 show that

3.3. Performance analysis of the LADRC based closed-loop system

1. ðωn ; θn Þ belonging to each group are almost the same although
ða1 ; a2 Þ are different. This result reveals that the performance of In this subsection, the capability of LADRC for controlling the
the closed-loop system is robust to the uncertainties ða1 ; a2 Þ. nonlinear uncertain system (30) is discussed. It will be shown that

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
10 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

100 A5. There exist constants w4 ; w5 ; w6 such that 8 t Z t 0

 
bðX; tÞ 2n
A ½w4 ; w5   0; ; 8 t Z t0 : ð50Þ
Magnitude (dB)

50 bðtÞ n1

_
0 jb ðtÞj r w6 : ð51Þ

−50
Theorem 2 (Xue and Huang [66]). Assume A3–A5 are satisfied and
−90 jxi ðt 0 Þj r ρ~ 0 ; i A n. There exist positives ωn ; ρ~ ; ηni ; i ¼ 1; 2; 3 which
are dependent on ρ~ 0 ; wl ; kq and ψ p ðÞ ðl A 6; q A n; p A 7Þ such that
L(s), C1−9
L(s), C1−10
−180 L(s), C1−11 the closed-loop system (30)–(34) satisfies 8 ωe A ½ωn ; 1Þ
Phase (deg)

L(s), C1−12
8
−270
L (s), C1−9,10,11,12
> sup ‖xðtÞ‖ r ρ~ ;
>
> t A ½t 0 ;1Þ
>
>

>
>
< ln ωe 1
−360 sup ‖xðtÞ  xn ðtÞ‖ r ηn1 max ; ;
> t A ½t 0 ;1Þ ω e ω e
>
> 

−450 >
> η n
ln ωe
>
> ‖x^ i ðtÞ xi ðtÞ‖ r 2 ; t A t j þ ηn3 max ; 0 ; t j þ 1 ; i ¼ 2; …; n þ 1; j Z 1:
10−1 100 101 102 103 : ωe ωe
Frequency (rad/sec)
ð52Þ
Fig. 4. The Bode Plot for L(s) and L0 ðsÞ under C1-9–12.

Eq. (52) demonstrates that the LADRC based closed-loop

LADRC can achieve satisfactory transient performance while deal- system ((30)–(34)) has the following properties:
ing with large nonlinear uncertainties.
The quality of control system is usually evaluated by both (i) The states of the closed-loop system are bounded despite
transient and steady performance, which motivates us to study the nonlinear time-varying unknown dynamics and discontinu-
following object: ous external disturbance.
sup J xðtÞ xn ðtÞ J r ε; ð48Þ (ii) The trajectory of the closed-loop system can be close to the
t A ½t 0 ;1Þ trajectory xn ðtÞ in the time region of ½t 0 ; 1Þ. Furthermore, the
error between them has the bound which can be described by
where ε 4 0 is the desired bound of the error between x(t) and Oðmax fln ωe =ωe ; 1=ωe gÞ.
xn ðtÞ ¼ ½xn1 ðtÞ; …; xnn ðtÞT . The transient response of xn ðtÞ can be tuned (iii) The estimation error of LESO converges to be as small as
by the following canonical cascaded integrators system: Oð1=ωe Þ after every discontinuous point of the disturbance
8 n with the process of the length Oðmax fln ωe =ωe ; 0gÞ.
>
> x_ 1 ¼ xn2 ;
>
>
<⋮
ð49Þ Obviously, Theorem 2 shows that the control object (48) can be
>
> x_ nn  1 ¼ xnn ; xn ðt 0 Þ ¼ xðt 0 Þ;
>
> achieved by designing ADRC's parameters. Therefore, the LADRC
: x_ n ¼  kn xn  kn  1 xn  ⋯  k1 xn ;
n 1 2 n based closed-loop system can obtain satisfactory transient perfor-
mance while dealing with large uncertainties.
for the nonlinear uncertain system (30), where k1 ; …; kn are the
parameters for tuning the dynamic response of the canonical Remark 9. Theorem 2 illuminates the relationship between the
system (49). bandwidth of ESO and the performance of the control system, i.e.,
Let the unmodeled dynamics and external disturbance in the the larger ωe, the smaller error between X(t) and X n ðtÞ. This
system (30) satisfy suggests a possible tuning method of ωe for practitioners, i.e.,
A3. The discontinuous points of d(t) are all the first class increasing ωe from a small value until the satisfactory performance
ft i g1 2
i ¼ 1 ; t i ot i þ 1 . There exist positive constants w1 ; w2 ; w3 such is achieved. On the other hand, the bandwidth of ESO must have
that an upper bound which is restricted by physical limitations of the
_ practical systems, such as the uncertain dynamics, the sampling
sup jdðtÞj rw1 ; sup jdðtÞj rw2 ; inf ft i þ 1  t i g Z w3 :
t A ½0;1Þ = ft i g1
t A ½0;1Þ;t 2 i ¼ 1
i rate, delay, and measurement noise. The second property we get
from Theorem 1 suggests that if there exists some delay in the
A4. f ðx; tÞ, bðx; tÞ and bðtÞ are continuously differentiable and system, ωe cannot be designed very large. And the analysis of a
there exist known locally Lipschitz functions ψ i ð J x J Þ ð1 r i r 7Þ sampled-data ADRC for a class of nonlinear uncertain systems was
such that 8 t A ½t 0 ; 1Þ initiated in [68] which quantitatively discussed the relationship







 between ωe and the sampling rate. The research on how to tune
f ðx; tÞ r ψ ð J x J Þ; ∂f ðx; tÞ  r ψ ð J x J Þ; ADRC's parameters, especially ωe, to achieve satisfactory perfor-
  1  ∂x  2
mance of the closed-loop system under the practical restrictions is
 
∂f ðx; tÞ    still in its early stage, and should be an important future work.
  bðx; tÞ r ψ ð J x J Þ;
 ∂t  r ψ 3 ð J x J Þ; 4

    Remark 10. In (9), the order of the ESO is extended one more than
∂bðx; tÞ  ∂bðx; tÞ  that of the plant to estimate the “total disturbance”. The controller
   
 ∂x  r ψ 5 ð J x J Þ;  ∂t  r ψ 6 ð J x J Þ;
((9) and (10)) can be viewed as a basic kind of ADRC. Theorem 2
  proves the capability of this kind ADRC for the uncertain systems
 1 
b ðx; tÞ r ψ 7 ð J x J Þ: which satisfy the conditions A3–A5. As to the systems not
satisfying A3–A5, when the additional dynamics of the ESO should
be increased, is still to be studied. A simple mathematical example,
2
d(t) has the discontinuous point ti of the first class means that both which is rare in practice, is dðtÞ ¼ t m ðm Z 1Þ; f ðx; tÞ ¼ 0 in the
 þ
d ðtÞ9 limt-t  dðtÞ and d ðtÞ 9 limt-t þ dðtÞ exist. system (30). Then a ðn þ mÞth order LESO should be designed.

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

Remark 11. Theorem 2 suggests that similar results can be is the “total disturbance” of the system (53). We design the
obtained for the uncertain system (11) (and its generalized form following linear ADRC:
(15)). If the effect of the uncertainties, which do not in the same 8
channel of the control input, can be included into the “total >
> z_ 2 ¼  2ωe z2 þz3 þðω2e  4ω2e Þx1 þ buðtÞ
>
>
< z_ ¼  β z þ ðω2  2ω3 Þx ; ωe ¼ 200;
disturbance” and the “total disturbance” satisfies the conditions 3 2 2 e e 1
ð56Þ
A3–A5, then the results of Theorem 2 can also be got. However, if > x^ 2 ¼ z2 þ 2ωe x1 ; x^ 2 ð0Þ ¼ 0;
>
>
>
the output of (11) is y ¼ x2 , the problem will be a little different : x^ ¼ z þ ω2 x ; x^ ð0Þ ¼ 0:
3 3 e 1 3
and may be discussed in two cases. (i) If f 2 ðt; x1 ; x2 ; DðtÞÞ and
bðt; x1 ; x2 Þ satisfy the conditions A3–A5, which implies that the (
state x1, determined by some uncertain dynamics and disturbance 0; 0 r t o tu
u¼ ð57Þ
outside the control channel, can be bounded by the state x2, then ð  302 x1 60x^ 2  x^ 3 Þ=b; t Zt u
the results of Theorem 2 hold. (ii) If the state x1 cannot be bounded pffiffiffi
by the state x2, then when and how ADRC or ESO should be where t u ¼ 0:1ð2 þ 2Þmax fln ωe =ωe ; 0g. We proceed to design
evolved is still open. ωe ¼ 200 and take some random values of the uncertain para-
meters ai ; i ¼ 1; …; 7 in Table 2.
Remark 12. ADRC is originally proposed in the general nonlinear Figs. 5 and 6 present the performance of the closed-loop
form ((9) and (10)), i.e., both ESO and feedback control law are system under C2-1–4. In Fig. 5, the system outputs are close to
nonlinear. The performance of NADRC was also studied for a class
of MIMO nonlinear uncertain systems in [62], where it was proved Table 2
that when the nonlinear functions satisfy certain conditions, the The sets of ai ; i ¼ 1; …; 7.
NADRC ((9)–(10)) can stabilize a class of nonlinear time-varying
uncertain systems. Moreover, by tuning the parameters of NADRC, Serial-number of sets a1 a2 a3 a4 a5 a6 a7

J XðtÞ  ZðtÞ J can be proven to be small enough after certain finite C2-1 5  0.03π 6  0.2 0.1π 5 0.8π
time, where Z(t) is the desired trajectory of X(t). C2-2 8 0.03π 12 0.4 0.3π 9 0.4π
C2-3 7  0.02π 8 0.7 0.15π 10 0.1π
Next, we will illustrate the results of Theorem 2 by a typical C2-4 9 0.04π 10 1.0 0.2π 7 1.0π
motor control system.

Example 6 (Xue and Huang [66]). Consider the following motor

control system: x (C2−1)
8 1
_
< x 1 ¼ x2
> 0.05 x (C2−2)
1
x_ 2 ¼ f ðx1 ; x2 ; tÞ þ dðtÞ þ bðtÞu t Z 0 ; ð53Þ x (C2−3)
>
:y¼x
1
1 0.04 x (C2−4)
1
x*
where the output y ¼ x1 is the motor position, the input u is the 1
0.03
motor terminal voltage, d(t) is the unknown disturbance with
x (m)

discontinuous points which may be caused by a change of load,

1

0.02
f ðx1 ; x2 ; tÞ; bðtÞ have uncertain dynamics generated by frictional
and other forces. We assume the region of the initial values of the
system (53) being 0.01

J ½x1 ð0Þ; x2 ð0ÞT J r 0:05 0

and f ðx1 ; x2 ; tÞ; bðtÞ being
8 −0.01
< f ðx1 ; x2 ; tÞ ¼ a1 sin ð80x1 þ a2 Þ þ a3 x2 ;
> 0 0.5 1 1.5 2 2.5 3 3.5 4
bðtÞ ¼ b þ a4 ð1 þ sin ða5 tÞÞ; time(s)
>
: dðtÞ ¼ a signð sin ða tÞÞ
6 7
Fig. 5. The curves of the position and its reference trajectory under C2-1–4.
where b ¼ 1:5 is the nominal value of b(t) and ai ; i ¼ 1; …; 9 are the
uncertain parameters satisfying
a1 A ½5; 10; a2 A ½  0:05π ; 0:05π ; a3 A ½5; 15; a4 A ½  0:2; 1;
a5 A ½0:1π ; 0:3π ; a6 A ½5; 10; a7 A ½0:1π ; 1π :
To guarantee the transient performance of the closed-loop system,
x1 ; x2 are desired to follow the trajectory of
( n ( n
x_ 1 ¼ xn2 ; x1 ð0Þ ¼ x1 ð0Þ;
ð54Þ
x_ n2 ¼  302 xn1  60xn2 ; xn2 ð0Þ ¼ x2 ð0Þ;

in the whole transient process.

Rewrite the system (53) as

8
> x_ 1 ¼ x2
<
x_ 2 ¼ x3 þ bu t Z 0 ; ð55Þ
>
:y¼x
1

where
x3 9f ðx1 ; x2 ; tÞ þ dðtÞ þ ðbðtÞ  bÞu Fig. 6. The “total disturbance” and its estimation under C2-1–4.

Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
12 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
the trajectory xni in the transient process despite various uncer-
tainties existing in internal dynamics and external disturbance.
Fig. 6 clearly shows that ESO can estimate the “total disturbance”
timely, and after each discontinuous point of the disturbance, the
estimation errors of the “total disturbance” converge quickly.
To further investigate the relationship between ωe and the
tracking error x1 ðtÞ xn1 ðtÞ, using the parameters of C2-1, the
following sets of bandwidth for the ESO (56) are considered:
ωe ¼ 200; ωe ¼ 250; ωe ¼ 300:
The corresponding simulation results are given in Figs. 7 and 8.
Fig. 7 shows that the larger ωe, the smaller error x1 ðtÞ  xn1 ðtÞ in the
whole time region. The reason is shown in Fig. 8, i.e., larger ωe
leads to faster convergence of the ESO's estimation error. Hence,
the simulation results are in accordance with the theoretical
results of Theorem 2.
Remark 13. In the past few years, lots of experimental results also
shown the capability of linear ADRC to achieve desired perfor-
mance of the closed-loop system despite uncertainties, such as the
fast tool servo system and motor control system (see [10,16,4]).
Fig. 7. The tracking error under C2-1 with different ωe.
Fig. 8. The estimation error of the “total disturbance” under C2-1 with different ωe.
Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
4. Conclusions and future research
Control of systems with uncertainty is the most fundamental
issue in control science and control engineering. This paper reveals
the uniqueness of ADRC for dealing with the uncertainties. The key in
a successful application of ADRC is how well one can reformulate the
problem by lumping various unknown quantities that affect the
system performance into total disturbance. Proper problem formula-
tion and simplification are perhaps the most crucial steps in ADRC
design. In conclusion, we offer the following suggestions:
(1) Before the controller is designed, one is suggested to carefully
investigate all possible terms in the physical process, not
ignoring complex terms no matter they are nonlinear, time-
varying, or coupled, until the control problem is clearly
identified. This will help one to correctly distinguish, among
many variables in a physical process, which are the inputs that
can be manipulated, which are the outputs to be controlled,
what are the total disturbances to be rejected, and what are
the features of the total disturbances. This is a crucial step in
correctly transforming a complex control problem into a
simple one and designing a suitable ADRC for the system.
(2) When the control problem is clearly identified, one is sug-
gested to explore intelligent use of ESO to deal with the
complex uncertainties. As shown in Example 2, one ESO can
be designed based on the output y ¼ x1 and the input u.
Alternatively, if x2 is also measurable, two ESOs can be
designed, one for x1-subsystem and the other for x2-subsystem.
In this design, the burden for each ESO may be reduced and
better performance may be obtained.
The paper also reviews the theoretical analysis on ADRC, which
more and more researchers and also engineers are concerned with.
Although the theoretical research of ADRC has been developed from
different angles, the research is still in its early stage. Many challen-
ging theoretical problems remain unresolved. For example,
(1) How to tune ADRC's parameters, especially the bandwidth of
ESO, to achieve satisfactory performance of the closed-loop
system under the practical restrictions of sampling rate, delay,
measurement noise?
(2) How to determine the order of the cascade integrators which
is most suitable for the system considered? Usually, for
minimum phase system, the order of the cascade integrators
can be determined according to the relative degree of the plant
[25]. But the principle does not always work. For example, the
flight control problem studied in [73] can be described as
a€ y ¼ c1 a_ y þ c2 ay þ b2 δ€ þ b1 δ_ þ b0 δ
where ay is the accelerator to be controlled and δ is the control
input. For this system, the relative degree is zero. However, after
analyzing the physical process of the system, the second order
cascade integrators from δ to a€ y was chosen in [73]. And the total
disturbance to be estimated and compensated for is c1 a_ y þ
c2 ay þ b2 δ€ þ b1 δ_ . If the system is non-minimum phase, the choice
of the integrators becomes more crucial for ADRC design. And the
analysis on the capacities of ADRC in dealing with the uncertain
non-minimum phase plants also remains to be seen.
(3) Lots of nonlinear design strategies were used in the original
ADRC, proposed by Han. What are the advantages of NADRC
compared to LADRC?
Moreover, we believe by illuminating the kernel of ADRC, more
innovations can be inspired to improve ADRC. For example, in the
 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
traditional ADRC, the prior information for the uncertainties in the
gain of the control input is needed for controller design. To relax
this restriction, Huang and Guo [74] proposed a method by
combining the ESO technique and the projected gradient estimator
for a class of SISO uncertain nonlinear systems. This work provides
some enlightenment for further research: combining the idea of
ADRC with the advantages of other estimation methods to explore
new grounds.
We hope the challenges listed here will lead to ever more
exhilarating research in the future!
References
[1] Han J. Active disturbance rejection controller and its applications. Control
Decis 1998;13(1):19–23 [in Chinese].
[2] Huang Y, Xu KK, Han J, Lam J. Flight control design using extended state
observer and non-smooth feedback. In: Proceedings of the 40th IEEE con-
ference on decision and control, Orlando, USA; 2001. p. 223–8.
[3] Gao Z, Hu S, Jiang F. A novel motion control design approach based on active
disturbance rejection. In: Proceedings of the 40th IEEE conference on decision
and control, Orlando, FL, USA; 2001. p. 4877–82.
[4] Zheng Q, Gao Z. On practical applications of active disturbance rejection
control. In: Proceedings of the 29th Chinese control conference; 2010.
p. 2503–8.
[5] Feng G, Liu Y, Huang L. A new robust algorithm to improve the dynamic
performance on the speed control of induction motor drive. IEEE Trans Power
Electron 2004;19(6):1624–7.
[6] Li S, Liu Z. Adaptive speed control for permanent-magnet synchronous motor
system with variations of load inertia. IEEE Trans Ind Electron 2009;56
(8):3050–9.
[7] Xia Y, Zhu Z, Fu M, Wang S. Attitude tracking of rigid spacecraft with bounded
disturbances. IEEE Trans Ind Electron 2011;58(2):647–59.
[8] Su J, Ma H, Qiu W, Xi Y. Task-independent robotic uncalibrated hand-eye
coordination based on the extended state observer. IEEE Trans Syst Man
Cybern, Part B: Cybern 2004;34(4):1917–22.
[9] Talole SE, Kolhe JP, Phadke SB. Extended-state-observer-based control of
flexible-joint system with experimental validation. IEEE Trans Ind Electron
2010;57(4):1411–9.
[10] Wu D, Chen K. Design and analysis of precision active disturbance rejection
control for noncircular turning process. IEEE Trans Ind Electron 2009;56
(7):2746–53.
[11] Zheng Q, Dong L, Lee D, Gao Z. Active disturbance rejection control for MEMS
gyroscopes. IEEE Trans Control Syst Technol 2009;17(6):1432–8.
[12] Zheng Q, Chen Z, Gao Z. A practical approach to disturbance decoupling
control. Control Eng Pract 2009;17:1016–25.
[13] Vincent J, Morris D, Usher N, Gao Z, Zhao S, Nicoletti A, et al. On active
disturbance rejection based control design for superconducting RF cavities.
Nucl Instrum Methods Phys Res A 2011;643:11–6.
[14] Jiang T. Robust output feedback stabilization of axial flow compressors with
uncertain compressor characteristics. In: Proceedings of the 2012 IEEE con-
ference on decision and control, Hawaii; 2012. p. 7279–84.
[15] Sun M, Chen Z, Yuan Z. A practical solution to some problems in flight control.
In: Proceedings of the joint 48th IEEE conference on decision and control and
28th Chinese control conference, Shanghai; 2009. p. 1482–7.
[16] Sun M, Wang Z, Wang Y, Chen Z. On low-velocity compensation of brushless
DC servo in the absence of friction model. IEEE Trans Ind Electron 2013;60
(9):3897–905.
[17] Huang C, Li D, Xue Y. Active disturbance rejection control for the ALSTOM
gasifier benchmark problem. Control Eng Pract 2013;21(4):556–64.
[18] Yu T, Chan KW, Tong JP, Zhou B, Li DH. Coordinated robust nonlinear boiler–
turbine–generator control systems via approximate dynamic feedback linear-
ization. J Process Control 2010;20(4):365–74.
[19] Li M, Li DH, Wang J, Zhao C. Active disturbance rejection control for fractional-
order system. ISA Trans 2013;52(3):365–74.
[20] Ohioepolymer News. LineStream technologies: advanced control, made sim-
ple. Accessed on October 1st, 2012. At 〈http://www.polymerohio.org/index.
php?option=com_content&view=article&id=352:linestream-technologies-ad
vanced-control-made-simple&catid=1:latest-news&Itemid=61〉.
[21] Schoenberger R. LineStream technologies signs licensing deal with Texas
instruments. The plain dealer; July 12th 2011.
[22] Han J. The structure of linear system and computation of feedback system. In:
Proceedings of the control theory and its application. Beijing: Academic Press;
1980 [in Chinese].
[23] Han J. Nonlinear design methods for control system. In: Proceedings of the
14th IFAC world congress, Beijing; 1999. p. 521–6.
[24] Gao Z, Huang Y, Han J. An alternative paradigm for control system design. In:
Proceedings of the 40th IEEE CDC, Orlando; 2001. p. 4578–85.
[25] Han J. From PID to active disturbance rejection control. IEEE Trans Ind Electron
2009;56(3):900–6.
[26] Haimo VT. Finite time controller. SIAM J Control Optim 1986;24:760–70.
[27] Isidori A. Nonlinear control system. 3rd ed. London: Springer-Verlag; 1995.
Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
13
[28] Han J. Control theory, is it a model analysis approach or a direct control
approach? Syst Sci Math Sci 1989;9(4):328–35 [in Chinese].
[29] Han J. The extended state observer of a class of uncertainty objects. Control
Decis 1995;10(1):85–8 [in Chinese].
[30] Han J. Active disturbance rejection control technique. Beijing: National
Defense Industry Press; 2008 [in Chinese].
[31] Freidovich LB, Khalil HK. Performance recovery of feedback-linearization
based designs. IEEE Trans Autom Control 2008;53(10):2324–34.
[32] Praly L, Jiang ZP. Linear output feedback with dynamic high gain for nonlinear
systems. Syst Control Lett 2004;53:107–16.
[33] Praly L, Jiang ZP. Further results on robust semiglobal stabilization with
dynamic input uncertainties. In: Proceedings of the 37th IEEE conference on
decision & control, FL; 1998. p. 891–6.
[34] Xue W, Huang Y. On performance analysis of ADRC for a class of MIMO lower-
triangular nonlinear uncertain systems. ISA Trans 2014, http://dx.doi.org/10.
1016/j.isatra.2014.02.002, in this issue.
[35] Aström KJ, Wittenmark B. Adaptive control. 2nd ed. Massachusetts: Pearson
Education, Inc.; 2003.
[36] Krstic M, Kanellakopoulos I, Kokotovic P. Nonlinear and adaptive control
design. Boston: Wiley; 1995.
[37] Doyle TC, Francis B, Tannenbaum A. Feedback control theory. 3rd ed. New
York: Macmillan; 1992.
[38] Başar T. Control theory: twenty-five seminal papers (1932–1981). New York:
IEEE Press; 2001.
[39] Fuller AT. The early development of control theory. J Dyn Syst Meas Control
1976;98:109–18.
[40] Bennett S. Nicolas Minorsky and the automatic steering of ship. IEEE Control
Syst 1984;4:10–5.
[41] Preminger J, Rootenberg J. Some consideration relating to control systems
employing the invariance principle. IEEE Trans Autom Control 1964;9(3):
209–15.
[42] Francis BA, Wonham WM. The internal model principle of control theory.
Automatica 1976;12(5):457–65.
[43] Johnson CD. Accommodation of external disturbances in linear regulation
and servomechanism problems. IEEE Trans Autom Control 1971;16(6):
635–644.
[44] Nakao M, Ohnishi K, Miyachi K. A robust decentralized joint control based on
interference estimation. In: Proceedings of the IEEE international conference
on robotics and automation; 1987. p. 326–31.
[45] Umeno T, Hori Y. Robust speed control of dc servomotors using modern two
degrees-of-freedom controller design. IEEE Trans Ind Electron 1991;38(5):
363–8.
[46] Kwon S, Chung WK. A discrete-time design and analysis of perturbation
observer for motion control applications. IEEE Trans Control Syst Technol
2003;11(3):399–407.
[47] Hsia TC, Gao LS. Robot manipulator control using decentralized linear time-
invariant time-delayed controller. In: Proceedings of the 1990 IEEE interna-
tional conference on robotics and automation, vol. 3; 1990. p. 2070–5.
[48] Huang J. An overview on the output regulation problem. J Syst Sci Math Sci
2011;31(9):1055–81.
[49] Canuto E. Embedded model control: outline of the theory. ISA Trans 2007;46:
363–77.
[50] Canuto E, Acuna-Bravo W, Molano A, Perez C. Embedded model control calls
for disturbance modeling and rejection. ISA Trans 2012;51(5):584–95.
[51] Guo L, Chen W. Disturbance attenuation and rejection for systems with
nonlinearity via DOBC approach. Int J Robust Nonlinear Control 2005;15:
109–25.
[52] Xue W, Huang Y. Comparison of the DOB based control, a special kind of PID
control and ADRC. In: Proceedings of the 2011 ACC, San Francisco; 2011. p.
4373–9.
[53] Wang X, Yau S, Huang J. A study of tracking-differentiator. In: Proceedings of
the 39th IEEE conference on decision and control, Sydney; 2000. p. 4783–4.
[54] Guo B, Han J, Xi FB. Linear tracking-differentiator and application to online
estimation of the frequency of a sinusoidal signal with random noise
perturbation. Int J Syst Sci 2002;33(5):351–8.
[55] Xue W, Huang Y, Yang X. What kinds of system can be used as tracking-
differentiator. In: Proceedings of the 2010 Chinese control conference, Beijing;
2010. p. 6113–20.
[56] Guo B, Zhao Z. On convergence of tracking differentiator. Int J Control 2011;84
(4):693–701.
[57] Huang Y, Han J. Analysis and design for nonlinear continuous extended state
observer. Chin Bull 2000;45(21):1938–44.
[58] Yang X, Huang Y. Capability of extended state observer for estimating
uncertainties. In: Proceedings of the 2009 American control conference,
MO; 2009. p. 3700–5.
[59] Guo B, Zhao Z. On the convergence of an extended state observer for nonlinear
systems with uncertainty. Syst Control Lett 2011;60:420–30.
[60] Yoo D, Yau SS-T, Gao Z. Optimal fast tracking observer bandwidth of the linear
extended state observer. Int J Control 2011;80(1):102–11.
[61] Zheng Q, Gao LQ, Gao Z. On Validation of extended state observer through
analysis and experimentation. J Dyn Syst Meas Control 2012;134(2):
024505.1–6.
[62] Guo B, Zhao Z. On convergence of the nonlinear active disturbance rejection
control for MIMO systems. SIAM J Control Optim 2013;51(2):1727–57.
[63] Zheng Q, Chen Z, Gao Z. A practical approach to disturbance decoupling
control. Control Eng Pract 2009;17:1016–25.
14 Y. Huang, W. Xue / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
[64] Tian G, Gao Z. Frequency response analysis of active disturbance rejec-
tion based control system. In: Proceedings of the 16th IEEE international
conference on control applications. Part of IEEE multi-conference on systems
and control, Singapore; 2007. p. 1595–9.
[65] Xue W, Huang Y. On frequency-domain analysis of ADRC for uncertain system. In:
Proceedings of the 2013 American control conference, Washington, DC; 2013.
[66] Xue W, Huang Y. On performance analysis of ADRC for nonlinear uncertain
systems with unknown dynamics and discontinuous disturbances. In: Pro-
ceedings of the 2013 Chinese control conference, Xi'an; 2013.
[67] Zhao C, Huang Y. ADRC based input disturbance rejection for minimum-phase
plants with unknown orders and/or uncertain relative degrees. J Syst Sci
Complex 2012;25:625–40.
[68] Xue W, Huang Y. On parameters tuning and capability of sampled-data ADRC
for nonlinear uncertain systems. In: Proceedings of the 2013 Chinese control
conference, Xi'an; 2013.
[69] Han J, Wang W. Nonlinear tracking-differentiator. J Syst Sci Math Sci 1994;14
(2):177–83 [in Chinese].
Please cite this article as: Huang Y, Xue W. Active disturbance rejection control: Methodology and theoretical analysis. ISA Transactions
(2014), http://dx.doi.org/10.1016/j.isatra.2014.03.003i
[70] Guo B, Jin F. Sliding mode and active disturbance rejection control to
stabilization of one-dimensional anti-stable wave equations subject to dis-
turbance in boundary input. IEEE Trans Autom Control 2013;58(5):1269–74.
[71] Guo B, Jin F. The active disturbance rejection and sliding mode control
approach to the stabilization of the Euler–Bernoulli beam equation with
boundary input disturbance. Automatica 2013;49:2911–8.
[72] Gao Z. Scaling and bandwidth-parameterization based controller tuning.
In: Proceedings of the 2003 American control conference, Denver, CO; 2003.
p. 4989–96.
[73] Zhao CZ, Huang Y. ADRC based integrated guidance and control scheme for the
interception of maneuvering targets with desired LOS angle. In: Proceedings
of the 29th Chinese control conference, Beijing; 2010. p. 6192–6.
[74] Huang C, Guo L. Control of a class of nonlinear uncertain systems by
combining state observers and parameter estimators. In: Proceedings of the
10th world congress on intelligent control and automation, Beijing; 2012.
p. 2054–9.