Title: Disturbance Observer

Name: Hyungbo Shim

Affil./Addr.: Department of Electrical and Computer Engineering,

Seoul National University, Seoul, Korea

+82-2-880-1745 / hshim@snu.ac.kr

Disturbance Observer
arXiv:2101.02859v1 [eess.SY] 8 Jan 2021

Abstract

Disturbance observer is an inner-loop output-feedback controller whose role is to reject

external disturbances and to make the outer-loop baseline controller robust against

plant’s uncertainties. Therefore, the closed-loop system with the DOB approximates

the nominal closed-loop by the baseline controller and the nominal plant model with

no disturbances. This article presents how the disturbance observer works under what

conditions, and how one can design a disturbance observer to guarantee robust stability

and to recover the nominal performance not only in the steady-state but also for the

transient response under large uncertainty and disturbance.

Keywords

robust stabilization; robust transient response; disturbance attenuation; singular per-

turbation; normal form

Introduction

The term disturbance observer has been used for a few different algorithms and methods

in the literature. In this article, we restrict the disturbance observer to imply the inner-
 2

loop controller as described conceptually in Fig. 1. This is one that has been employed in

many practical applications and verified in industry, and is often simply called “DOB.”

Fig. 1. Disturbance observer as an inner-loop feedback controller

The primary goal of DOB is to estimate the external disturbance at the input

stage of the plant, which is then used to counteract the external disturbance as in Fig. 1.

This initial idea is extended to deal with the plant’s uncertainties when uncertain terms

can be lumped into the external disturbance. Therefore, DOB is considered as a method

for robust control. More specifically, DOB robustifies the (possibly non-robust) baseline

controller. The baseline controller is supposed to be designed for a nominal model of

the plant that does not have external disturbances nor uncertainties. By inserting the

DOB in the inner-feedback-loop, the nominal stability and performance that would have

been obtained by the baseline controller and the nominal plant, can be approximately

recovered even in the presence of disturbances and uncertainties. There is effectively

no restriction on the baseline controller as long as it stabilizes the nominal plant.

When there is no uncertainty and no disturbance with a DOB being equipped,

the closed-loop system recovers the nominal performance completely, and as the amount

of uncertainty and disturbance grows, the performance degrades gradually while it is

still close to the nominal performance. This is in contrast to other robust controls

based on the worst-case design, which sacrifices the nominal performance for the worst

uncertainty.
 3

Disturbance Observer for Linear System

Let P (s) be the transfer function of a unknown linear plant, Pn (s) be its nominal

model, and C(s) be a baseline controller designed for the nominal model Pn (s). Then,

the closed-loop system with the DOB can be depicted as Fig. 2. In the figure, Q(s)

is a stable low-pass filter whose dc gain is one. The relative degree (i.e., the order

of denominator minus the order of numerator) of Q(s) is greater than or equal to the

relative degree of Pn (s), so that the block Pn−1 (s)Q(s) is proper and implementable. The

signals d and r are the external disturbance and the reference, respectively, which are

assumed to have little high frequency components, and the signal n is the measurement

noise.

Fig. 2. Disturbance observer for a linear plant

From Fig. 2, the output y in the frequency domain is written as

Pn P C
y(s) = r(s)
Pn (1 + P C) + Q(P − Pn )
Pn P (1 − Q) P (Q + Pn C)
+ d(s) − n(s). (1)
Pn (1 + P C) + Q(P − Pn ) Pn (1 + P C) + Q(P − Pn )

Assume that all the transfer functions are stable (i.e., all the poles have negative real

parts), and let ωc be the cut-off frequency of the low-pass filter so that Q(jω) ≈ 1 for

ω  ωc and Q(jω) ≈ 0 for ω  ωc . Therefore, it is seen from (1) that, for ω  ωc ,

Pn C
y(jω) ≈ r(jω) − n(jω) (2)
1 + Pn C
 4

and for ω  ωc where d(jω) ≈ 0 and r(jω) ≈ 0,

PC P PC PC
y(jω) ≈ r(jω) + d(jω) − n(jω) ≈ − n(jω).
1 + PC 1 + PC 1 + PC 1 + PC

The property (2) is of particular interest because the input-output relation recovers

the nominal closed-loop system without being affected by the disturbance d.

The low-pass filter Q(s) is often called ‘Q-filter’ and it is typically given by

a0
Q(s) =
(τ s)ν + aν−1 (τ s)ν−1
+ · · · + a1 (τ s) + a0

where ν is the relative degree of Pn (s), the coefficients ai are chosen such that sν +

aν−1 sν−1 + · · · + a0 be a Hurwitz polynomial, and the constant τ > 0 determines the

cut-off frequency.

Robust stability condition

The beneficial property (2) is obtained under the assumption that the transfer functions

in (1) are stable for all possible uncertain plants P (s). Therefore, it is necessary to

design Q(s) (or, choose τ and ai ’s) such that the closed-loop system remains stable for

all possible P (s).

In order to deal with uncertain plants having parametric uncertainty, consider

the set P of uncertain transfer functions

sn−ν + βn−ν−1 sn−ν−1 + · · · + β0

 
P = P (s) = g : αi ∈ [αi , αi ], βi ∈ [β i , β i ], g ∈ [g, g]
sn + αn−1 sn−1 + · · · + α0

where the intervals [αi , αi ], [β i , β i ], and [g, g] are known, and g > 0.

Theorem 1 (Shim and Jo (2009)) Assume that (a) the nominal model Pn belongs

to P and the baseline controller C internally stabilizes Pn , (b) all transfer functions in

P are minimum phase, and (c) the coefficients ai of Q are chosen such that

g
pf (s) := sν + aν−1 sν−1 + · · · + a1 s + a0
g∗
 5

is Hurwitz for all g ∈ [g, g], where g ∗ ∈ [g, g] is the nominal value of g, then, there

exists τ ∗ > 0 such that, for all τ ≤ τ ∗ , the transfer functions in (1) are stable for all

P (s) ∈ P.

The value of τ ∗ can be computed from the knowledge of the bounds of the

intervals in P (Shim and Jo, 2009), but it may also be conservatively chosen based on

repeated simulations in practice. Smaller τ ∗ implies larger bandwidth of Q-filter, which

is desired in the sense that the property (2) holds for larger frequency range.

The proof of Theorem 1 proceeds by observing the closed-loop system in Fig. 2

has 2n + m poles where m is the order of C(s). Then one can inspect the behavior of

those poles by changing the design parameter τ . For this, let us denote the poles by

λi (τ ), i = 1, . . . , 2n + m. Then, it can be shown that

lim τ λi (τ ) = λ∗i , i = 1, . . . , ν (3)

τ →0

where λ∗i , i = 1, . . . , ν, are the roots of pf (s), and

lim λi (τ ) = λ∗i , i = ν + 1, . . . , 2n + m
τ →0

where λ∗i , i = ν + 1, . . . , n, are the zeros of P (s), and λ∗i , n + 1, . . . , 2n + m, are the

poles of the nominal closed-loop with Pn (s) and C(s).

This argument shows that, if there is no λ∗i on the imaginary axis, the conditions

(a), (b), and (c) are also necessary for robust stability with sufficiently small τ . It is

also seen by (3) that, when pf (s) is Hurwitz, the poles λi (τ ), i = 1, . . . , ν, escapes to

the negative infinity as τ → 0. Therefore, with large bandwidth of Q(s), the closed-

loop system shows two-time scale behavior. In fact, it turns out that pf (s) is the

characteristic polynomial of the fast dynamics called ‘boundary-layer system’ in the

singular perturbation analysis with τ being the parameter of singular perturbation.

 6

Design of Q(s) for robust stability

In order to satisfy the condition (c) of Theorem 1, one can choose {ai : i = 1, · · · , ν −1}

such that

sν−1 + aν−1 sν−2 + · · · + a2 s + a1

be a Hurwitz polynomial, and then pick a0 > 0 sufficiently small. Then, the polynomial

pf (s) remains Hurwitz for all variation of g ∈ [g, g].

This can be justified by, for example, the circle criterion. With G(s) := a0 /(sν +

aν−1 sν−1 + · · · + a1 s) and a static gain (g/g ∗ ) that belongs to the sector [g/g ∗ , g/g ∗ ],

the characteristic polynomial of the closed-loop system becomes sν + aν−1 sν−1 + · · · +

a1 s + (g/g ∗ )a0 . Therefore, if the Nyquist plot of G(s) does not enter nor encircle the

closed disk in the complex plane whose diameter is the line segment connecting −g ∗ /g

and −g ∗ /g, then pf (s) is Hurwitz for all variation of g ∈ [g, g]. Since G(s) has one pole

at the origin and the rest poles have negative real parts, its Nyquist plot is bounded

to the direction of real axis. Therefore, by choosing a0 sufficiently small, the Nyquist

plot is disjoint from and does not encircle the disk.

Disturbance Observer for Nonlinear System

Intuitive introduction of the DOB for nonlinear system

DOB for nonlinear systems inherits all the ingredients and properties of the DOB

for linear systems. The DOB can be constructed for a class of systems that can be

represented by the Byrnes-Isidori normal form in certain coordinates as


y = x1 , ẋi = xi+1 , i = 1, · · · , ν − 1,







P : ẋν = f (x, z) + g(x, z)(u + d),





ż = h(x, z, dz )


 7

where u ∈ R is the input, y ∈ R is the measured output, x = [x1 , · · · , xν ]T ∈ Rν and

z ∈ Rn−ν are the states of the n-th order system, and unknown external disturbances

are denoted by d and dz . The disturbances and their first time-derivatives are assumed

to be bounded. The functions f and g, and the vector field h contain uncertainty. The

corresponding nominal model of the plant is considered as


y = x1 , ẋi = xi+1 , i = 1, · · · , ν − 1,







Pn : ẋν = fn (x, z) + gn (x, z)ū





ż = hn (x, z, 0)



where ū represents the input to the nominal model, and the nominal fn , gn , and hn are

known. Let us assume all functions and vector fields are smooth.

Assumption 1 (a) A baseline feedback controller


∈ Rm ,

 η̇ = Π(η, y)

C:

 ū = π(η, y)


stabilizes the nominal model Pn .

(b) The zero dynamics ż = h(x, z, dz ) is input-to-state stable (ISS) from x and dz to

the state z.

The underlying idea of the DOB is that, if one can apply the input udesired to

the plant P , which is generated by

z̄˙ = hn (x, z̄)

(4)
1
udesired = −d + (−f (x, z) + fn (x, z̄) + gn (x, z̄)ū)
g(x, z)
then the plant P with (4) behaves identical to the nominal model Pn plus ż = h(x, z, dz ).

Then, the plant P with (4) interacts with the baseline controller C, and so, it is sta-

bilized, while the plant’s zero dynamics ż = h(x, z, dz ) becomes stand-alone. However,

the zero dynamics is ISS, and so, the state z is bounded under the bounded inputs x

and dz .
 8

This idea is, however, not implementable because d, f , and g, as well as the

states x and z in (4), are unknown. It turns out that the role of the DOB is to estimate

the state x and the signal udesired . In the linear case, the structure of the DOB in Fig. 2

actually does this job (Shim et al, 2016).

Implementation of the DOB

The idea of estimating x and udesired is realized in the semi-global sense. Suppose that

S0 ⊂ Rn+m be a compact set for possible initial conditions (x(0), z(0), η(0)), and U be

a compact subset of region of attraction for the nominal closed-loop system Pn and C,

which is assumed to contain a compact set S that is strictly larger than S0 . Let Sz

and Ux be the projections of S and U to the z-axis and the x-axis, respectively, and

let Md and Mdz be an upper bounds for the norms of d and dz , respectively. Uncertain

f , g, and h are supposed to belong to the sets F, G, and H, respectively, which are

defined as follows. Let H be a collection of uncertain h. Then there is a compact set

Zh ⊂ Rn−ν to which the state z(t) belongs, where z(t) is the solution to each ISS

system ż = h(u1 , z, u2 ), h ∈ H, with any z(0) ∈ Sz and any bounded inputs u1 (t) ∈ Ux

and ku2 (t)k ≤ Mdz . Assume that H is such that there is a compact set Z ⊃ ∪h∈H Zh .

The set F is a collection of uncertain f such that, for every f ∈ F, there are uniform

bounds Mf and Mdf such that |f (x, z)| ≤ Mf and k∂f (x, z)/∂(x, z)k ≤ Mdf on Ux × Z.

The set G is a collection of uncertain g such that, for every g ∈ G, there are uniform

bounds g, g, and Mdg such that k∂g(x, z)/∂(x, z)k ≤ Mdg and

0 < g ≤ g(x, z) ≤ g, ∀(x, z) ∈ Ux × Z.

The DOB is illustrated in Fig. 3 and is given by

 9

Fig. 3. Disturbance observer for a nonlinear plant. The state of Q(s) is p.


z̄˙ = hn (satx (q), z̄) ∈ Rn−ν






a0


∈ Rν


 q̇ = A(τ )q + By
τ 

D: 
a0 1 1
ṗ = A(τ )p + B φ − ∗ (φ − satφ (φ)) + ∗ w ∈ Rν





 τ g g


 u = satφ (φ) + 1 w



g∗
where

1 h a0 a1 aν−1 i 1 a0
φ = p1 + ∗ ν
, ν−1
, · · · , q − ∗ νy
g τ τ τ g τ

w = fn (satx (q), z̄) + gn (satx (q), z̄)ū

   
 0 1 ··· 0  0
   
 .. .
. . . .
.   .. 
 . . . .  .
A(τ ) =  , B= 
   
   
 0 0 ··· 1  0
   
   
−a0 −a1 −aν−1
τ ν τ ν−1
··· τ 1

and g ∗ is any constant between g and g, ū is the output of the baseline controller C,

and two saturations are globally bounded, continuously differentiable functions such

that satx (x) = x for all x ∈ Ux and satφ (φ) = φ for all φ ∈ Sφ where the interval Sφ is

given by
 
1 1 f (x, z)
Sφ = − ∗ (fn (x, z̄) + gn (x, z̄)π(η, x1 )) − −d
g(z, x) g g(x, z)

: z ∈ Z, (z̄, x, η) ∈ U, |d| ≤ Md , f ∈ F, g ∈ G .
 10

Suppose that (z̄(0), q(0), p(0)) ∈ S0z × Sqp where S0z is the projection of S0 to the

z-axis, and Sqp is any compact set in R2ν .

In practice, choosing the sets like U , Z, and Sφ may not be easy. If so, conserva-

tive choice of them works. Based on repeated simulations or numerical computations,

one can take sufficiently large compact sets for them.

If everything is linear, then the above controller D becomes effectively the same

as the state-space realization of the linear DOB in Fig. 2.

Robust stability

Assumption 1 (c) The coefficients ai in the DOB are chosen such that sν−1 +aν−1 sν−2 +

· · · + a1 is Hurwitz and a0 > 0 is sufficiently small such that the Nyquist plot of

G(s) := a0 /(sν +aν−1 sν−1 +· · ·+a1 s) is disjoint from and does not encircle the closed

disk in the complex plane whose diameter is the line segment connecting −g ∗ /g and

−g ∗ /g.

Under this assumption, the polynomial

g(x, z)
pf (s) := sν + aν−1 sν−1 + · · · + a1 s + a0
g∗

is Hurwitz for all (x, z) ∈ Ux × Z and for all uncertain functions g ∈ G, as discussed in

the section ‘Design of Q(s) for robust stability.’

Theorem 2 Suppose that the conditions (a), (b), and (c) of Assumption 1 hold. Then,

there exists τ ∗ > 0 such that, for all τ ≤ τ ∗ , the closed-loop system of P , C, and D

with d ≡ 0 and dz ≡ 0 is stable for all f ∈ F, g ∈ G, and h ∈ H, and the region of

attraction for (x, z, η, z̄, q, p) includes S0 × S0z × Sqp .

To prove the theorem, one can convert the closed-loop system into the standard

singular perturbation form with τ being the singular perturbation parameter. Then,

it can be seen that the quasi-steady-state system on the slow manifold is simply the
 11

nominal closed-loop system Pn and C without external disturbance d plus the actual

zero dynamics ż = h(x, z, dz ). Since the quasi-steady-state system is assumed to be

stable by Assumption 1, the overall system is stable with sufficiently small τ if the

boundary-layer system is also stable. Then, it turns out that the boundary-layer sys-

tem is linear since all the slow variables such as x(t) and z(t) are treated as frozen

parameters, and the characteristic polynomial of the linear boundary-layer system is

pf (s). For details, see (Back and Shim, 2008).

It can be also seen that, on the slow manifold, all the saturation functions in the

DOB become inactive. The role of the saturations is to prevent the peaking phenomenon

(Sussmann and Kokotovic, 1991) from transferring into the plant. Without saturations,

the region of attraction may shrink in general as τ gets smaller, as in (Kokotovic and

Marino, 1986), and only local stability is achieved. On the other hand, even if the plant

is protected from the peaking components by the saturation functions, some internal

components must peak for fast convergence of the DOB states. In this regard, the role

of the dead-zone nonlinearity in Fig. 3 is to allow peaking inside the DOB structure.

Robust Transient Response

Additional benefit of the DOB with saturation functions is robustness of transient

response. If the baseline controller C is designed for the nominal model Pn to achieve

desired transients such as small overshoot or fast settling time for example, then, similar

transients can be obtained for the actual plant P under external disturbances by adding

the nonlinear DOB to the baseline controller. This holds true also for linear plants.

Theorem 3 (Back and Shim (2008)) Suppose that the conditions (a), (b), and (c)

of Theorem 2 hold. For any given  > 0, there exists τ ∗ > 0 such that, for each

τ ≤ τ ∗ , the solution of the closed-loop system denoted by (z(t), z̄(t), x(t), η(t)), initiated

in Sz × S, is bounded and satisfies

 12
   

z̄(t)
  z̄N (t) 


   
x(t) − xN (t) ≤ , ∀t ≥ 0
   
   
   

η(t) ηN (t)

where the reference (z̄N (t), xN (t), ηN (t)) is the solution of the nominal closed-loop sys-

tem of Pn and C with (z̄N (0), xN (0), ηN (0)) = (z̄(0), x(0), η(0)).

Since y = x1 , this theorem ensures robust transient response that y(t) remains

close to its nominal counterpart yN (t) from the initial time t = 0. However, nothing

can be said regarding the state z(t) except that it is bounded by the ISS property of

the zero dynamics.

Theorem 3 is basically an application of Tikhonov’s theorem (Hoppensteadt,

1966).

Discussions

• The baseline controller C may depend on a reference r, which is the case for

tracking control. Theorems 2 and 3 also hold for this case (Back and Shim,

2008).

• If uncertainties are small in modeling so that f ≈ fn , g ≈ gn , and h ≈ hn , then

the DOB can be used to estimate the input disturbance d. This is clearly seen

from (4).

• Regarding a design of DOB for multi-input-multi-output plants, refer to (Back

and Shim, 2009), which requires well-defined vector relative degree of the plant.

For the case of extended state observer (ESO), this requirement has been relaxed

in (Wu et al, 2019).

• If the external disturbance is a sum of a modeled disturbance, that is generated

by a known model, and a unmodeled disturbance that is slowly varying, then the
 13

modeled disturbance can be exactly canceled by embedding the known model

into the DOB structure while the unmodeled disturbance can still be canceled

approximately. This is done by utilizing the internal model principle, and the

details can be found in (Joo et al, 2015).

• For linear systems with mismatched disturbances:

ẋ = Ax + bu + Ed, x ∈ Rn , u ∈ R,

y = cx, d ∈ Rq , y ∈ R,

one can transfer the disturbances into the input stage by redefining the state

combined with the disturbances, if the disturbance d is smooth in t. Indeed,

there is a coordinate change [xT , z T ]T = Φx + Θ[dT , ḋT , · · · , (d(ν−2) )T ]T from x

to (x, z) with a nonsingular matrix Φ, and the system becomes

y = x1 , ẋi = xi+1 , i = 1, · · · , ν − 1,

ẋν = Fx x + Fz z + gu + gd

ż = Hx x + Hz z + dz

where d and dz are linear combinations of d and its derivatives (Shim et al,

2016).

• Robust control based on the linear DOB is relatively simple to construct, which

is one of the reasons why it is frequently used in industry. Stability condition

is often ignored in practice, but as seen in Theorem 1, all three conditions

are often automatically met. In particular, with small amount of uncertainty,

the condition (c) tends to hold with g ≈ g ∗ for any stable Q-filter because of

structural robustness of Hurwitz polynomials.

• In the case of linear DOB, there is another robust stability condition based on

the small-gain theorem (Choi et al, 2003).

 14

• Basic philosophy of DOB is to treat plant’s uncertainties and external distur-

bances together as a lumped disturbance, and to estimate and compensate it.

This philosophy is shared with other similar approaches such as extended state

observer (ESO) (Freidovich and Khalil, 2008) and active disturbance rejection

control (ADRC) (Han, 2009). The DOB also has been reviewed with comparison

to other similar methods in (Chen et al, 2015).

Summary and Future Directions

Underlying theory for the disturbance observer is presented. The analysis is mainly

based on large bandwidth of Q-filter. However, there are cases when the bandwidth

cannot be increased in practice because of limited sampling rate in discrete-time imple-

mentation. Further study is necessary to achieve satisfactory performance for discrete-

time design of DOB.

Cross References

References

Back J, Shim H (2008) Adding robustness to nominal output-feedback controllers for uncertain non-

linear systems: A nonlinear version of disturbance observer. Automatica 44(10):2528–2537

Back J, Shim H (2009) An inner-loop controller guaranteeing robust transient performance for uncer-

tain MIMO nonlinear systems. IEEE Transactions on Automatic Control 54(7):1601–1607

Chen WH, Yang J, Guo L, Li S (2015) Disturbance-observer-based control and related methods—an

overview. IEEE Transactions on Industrial Electronics 63(2):1083–1095

Choi Y, Yang K, Chung WK, Kim HR, Suh IH (2003) On the robustness and performance of

disturbance observers for second-order systems. IEEE Transactions on Automatic Control

48(2):315–320
 15

Freidovich LB, Khalil HK (2008) Performance recovery of feedback-linearization-based designs. IEEE

Transactions on Automatic Control 53(10):2324–2334

Han J (2009) From PID to active disturbance rejection control. IEEE Transactions on Industrial

Electronics 56(3):900–906

Hoppensteadt FC (1966) Singular perturbations on the infinite interval. Transactions of the American

Mathematical Society 123(2):521–535

Joo Y, Park G, Back J, Shim H (2015) Embedding internal model in disturbance observer with robust

stability. IEEE Transactions on Automatic Control 61(10):3128–3133

Kokotovic P, Marino R (1986) On vanishing stability regions in nonlinear systems with high-gain

feedback. IEEE Transactions on Automatic Control 31(10):967–970

Shim H, Jo NH (2009) An almost necessary and sufficient condition for robust stability of closed-loop

systems with disturbance observer. Automatica 45(1):296–299

Shim H, Park G, Joo Y, Back J, Jo N (2016) Yet another tutorial of disturbance observer: Robust

stabilization and recovery of nominal performance. Control Theory and Technology 4(3):237–

249, Special Issue on Disturbance Rejection: A Snapshot, A Necessity, and A Beginning,

(preprint = arXiv:1601.02075)

Sussmann H, Kokotovic P (1991) The peaking phenomenon and the global stabilization of nonlinear

systems. IEEE Transactions on Automatic Control 36(4):424–440

Wu Y, Isidori A, Lu R, Khalil H (2019) Performance recovery of dynamic feedback-linearization

methods for multivariable nonlinear systems. to appear at IEEE Transactions on Automatic

Control