You are on page 1of 6

2017 11th Asian Control Conference (ASCC)

Gold Coast Convention Centre, Australia


December 17-20, 2017

Some Fundamental Aspects of Second and Third Order Repetitive


Controllers
Hai-Jiao Guo, Tadashi Ishihara Member, IEEE

Abstract— Repetitive controllers ((RCs) are well known as an V(z)


effective control method that can produce zero error in a control Y*(z) E(z) Y(z)
system tracking a periodic command, or cancel the influence of a G(z)
periodic disturbance. In addition, several studies address that R(z)
using a higher order RC incorporating errors from some -
previous period can improve the robustness to disturbance
period fluctuations. So the higher order RCs show us the more
prospects of RCs. But so many studies of higher order RCs are Figure 1. Basic RC system
developed in the frequency response analysis instead of the time
higher order RC can be effective at minimizing the influence
response analysis. In order to show some abilities of higher order
RCs in both frequency domain and time domain, we believe that
of a fluctuating disturbance. Ref. [3] and [6] show how higher
will be better to study the simplest higher order RCs: second and order RC can widen the notches around the fundamental
third order RCs. So, this paper studies some basic aspects of frequency and its harmonics in the sensitivity transfer function
second and third order repetitive controls from both frequency from disturbance (as an equivalent output disturbance on the
response and time response. We found some interesting results feedback control system output) to resulting error. But so many
and that will find a toehold to study the higher order RCs. studies of higher order RCs are developed in the frequency
response analysis instead of the time response analysis. In
I. INTRODUCTION order to show some abilities of higher order RCs in time
Repetitive control (RC) is well known as a method for response, we believe that will be better to study the simplest
tracking a periodic command as well as eliminating the higher order RCs: second and third order RCs. So, this paper
influence of a periodic disturbance. Many applications using studies some fundamental aspects of second and third order
RC have been studied and show this method is very useful and repetitive controls from both frequency response and time
powerful. But RCs have an implicit condition that the both response. We found some interested results and that will find a
periods of command and disturbance should be same and the toehold to study the higher order RCs.
period should be known previously. In some applications, the II. BASIC CHARACTERISTICS OF REPETITIVE CONTROL
disturbance is related to a periodic command being executed,
in which case one can easily stay synchronized. In [1] the First, in order to simplify our discussion, we summarize
disturbance period is the period of rotation of a control moment some basic characteristics as the following.
gyro (CMG) on a spacecraft run using three phase motors, and
A. Stability of the RC
one knows the phase from the commands to the motors. A little
more generally, one can try to measure the period in real time. We consider the block diagram in Fig.1 as the repetitive
Ref. [2] analyze this kind of application. RC can be effective control (RC) system, where 𝑅(z) is the repetitive controller,
provided that the change in period is sufficiently slow that the and 𝐺(𝑧) is a closed loop transfer function of a feedback
RC convergence time keeps the system close to zero error. control system.
Sometimes the period fluctuates, and in this situation, one 𝑁 (𝑧) 𝑁 (𝑧)
wants an RC design that is robust to uncertainty in the G(z) = , 𝑅(𝑧) = (1)
𝐷 (𝑧) 𝐷 (𝑧)
disturbance period. It is this situation, [3] and [4] develop
higher order RC to create improved robustness to the The closed loop transfer function of Fig.1 is
disturbance period. Higher order RC makes use of measured
𝐺(𝑧)𝑅(𝑧)
errors not only from the previous period, but also from one or 𝐺 (𝑧) = . (2)
more earlier period. Ref. [5] and [6] develop an understanding 1 + 𝐺(𝑧)𝑅(𝑧)
of how this approach can improve period robustness, from both And then the stability of the closed loop is well known as the roots of
a frequency response point of view and a root locus analysis the following characteristic equation must be in the unit circle.
approach. From above studies, in order to extend the abilities
1 + G(z)R(z) = 0, or 𝐷 (𝑧)𝐷 (𝑧) + 𝑁 (𝑧)𝑁 (𝑧) = 0 (3)
of RC, higher order RCs will be very extracting subject for the
researchers, and many researchers have studied the subject. Ref. Because the repetitive controller usually has the higher order so how
[3] demonstrated that to determine the stability from (3) is not so easy. An efficient way has
been summarized in [7]. But for visible, the root locus will be used in
*Research supported by JSPS KAKENHI Grant Number 15K05950. the paper.
Hai-Jiao Guo is with the Tohoku Gakuin University, Sendai, 985-8537
JAPAN (phone: +81-22-368-7019; fax: +81-22-368-7019; e-mail: kaku@ B. Tracking Condition of the RC
mail.tohoku-gakuin.ac.jp). From Fig.1, we have
Tadashi Ishihara is with Fukushima University, Fukushima, 960-1296
JAPAN. (e-mail: ishihara@sss.fukushima-u.ac.jp). 𝑌 ∗ (𝑧) − 𝑉(𝑧) = [1 + 𝐺(𝑧)𝑅(𝑧)]𝐸(𝑧)

978-1-5090-1573-3/17/$31.00 ©2017 IEEE 693


𝐷𝑔 (𝑧)𝐷𝑟 (𝑧) + 𝑁𝑔 (𝑧)𝑁𝑟 (𝑧) determine any clear benefits for higher order RC. But [3] and
= 𝐸(𝑧) (4) [4] suggests using higher order RC with negative coefficients
𝐷𝑔 (𝑧)𝐷𝑟 (𝑧)
on the measurements for some previous period errors, in order
𝐷 (𝑧)𝐷 (𝑧) + 𝑁 (𝑧)𝑁 (𝑧) 𝐸(𝑧) to make RC less sensitive to accurate knowledge of the period,

= 𝐷 (𝑧)𝐷 (𝑧) 𝑌 (𝑧) − 𝑉(𝑧) , (5) or to period fluctuations. In addition, we found a way use some
freedom to improve the robustness when the disturbance has
when the forcing function on the right side of (5) becomes zero, some fluctuation [8]. But in that study we just showed one of

𝐷 (𝑧) 𝑌 (𝑧) − 𝑉(𝑧) = 0 , (6) the ability of the freedom included in the higher order RC.
When we studied this problem more detail, we find who to use
in this case the error of (5) converges to zero provided the
characteristic polynomial in brackets on the left of (5) has all these freedoms will be more interesting and more difficult.
roots inside the unit circle. That means this condition is as same These studies let us know that is necessary to study from
as (3) in stabilization analysis. simple higher order RC.
So next we must focus our condition of (6). As a simple way, Suppose that the period of the command (or disturbance, or
if at the frequencies, that we want to track, 𝐷 (z) = 0, equation both when both are present) is pT , where T is the sample time
(6) will be held, references that contain such frequency will be interval, and p is the number of time steps per period.
tracked and disturbances that contain such frequency will be
eliminated. That is well known as the internal model principle The form for Nth order RC corresponding to (1) is
[9]. We will show such thing more straightforward style and
more detail in context. 𝑢(𝑘) = 𝛼 [𝑢(𝑘 − 𝑗𝑝 + 𝜑𝑒(𝑘 − 𝑗𝑝 + 1)], (8)
By the way, it will be better to show another way to
understand of eliminating the disturbance for our following where N is the number of periods one wishes to include, and
discussion. Consider the transfer function from the disturbance 1 ,  2 ,...,  N are the weights that will be chosen by the
V(z) to the output 𝑌(𝑧), when 𝑌 ∗ (𝑧) = 0, we have designer. One may think of this as creating a weighted average,
𝑌(𝑧) = 𝑉(𝑧) + 𝐺(𝑧)𝑅(𝑧)𝐸(𝑧) in which case each weight should be non-negative. However,
as pointed out in [3], there is no need to restrict the weights to
= 𝑉(𝑧) + 𝐺(𝑧)𝑅(𝑧)[−𝑌(𝑧)],
be positive. In practice, it is convenient to design the RC law
[1 + 𝐺(𝑧)𝑅(𝑧)]𝑌(𝑧) = 𝑉(𝑧), in the frequency domain, using the form
𝑌(𝑧) = [1 + 𝐺(𝑧)𝑅(𝑧)] 𝑉(𝑧) = 𝑆(𝑧)𝑉(𝑧) 𝜑 𝛼 𝑧( )
+ 𝛼 𝑧( )
+ ⋯+ 𝛼
𝑅 (𝑧) =
𝐷𝑔 (𝑧)𝐷𝑟 (𝑧) 𝑧 − [𝛼 𝑧 ( ) + 𝛼 𝑧( ) + ⋯+ 𝛼 ]
𝑆(𝑧) = [1 + 𝐺(𝑧)𝑅(𝑧)] = , (7)
𝐷𝑔 (𝑧)𝐷𝑟 (𝑧) + 𝑁𝑔 (𝑧)𝑁𝑟 (𝑧) 𝑁 (𝑧)
= , (9)
where 𝑆(𝑧) is called as sensitivity function. So the disturbance 𝐷 (𝑧)
will be eliminated at the frequencies of 𝑆(𝑧) = 0, that is as 𝑁 (𝑧) = φ 𝛼 𝑧 ( )
+ 𝛼 𝑧( )
+⋯+𝛼 , (10)
same as 𝐷 (z) = 0. The sensitivity function will be used in the
( ) ( )
simulation to show the elimination of the disturbance. 𝐷 (𝑧) = 𝑧 − 𝛼 𝑧 +𝛼 𝑧 +⋯+𝛼 , (11)
From the above analysis, when designing a repetitive control and in addition it is necessary to restrict the choice of the  j
system, the important thing is how to design the denominator to satisfy
𝐷 (z) = 0 . We summarize the above discussions as the
following Property. 1   2     N  1 . (12)
Property 1: A repetitive control must satisfy the following For first order RC, N=1, 𝛼 = 1, we have
conditions.
𝐷 (𝑧) = 𝑧 − 1 .
1. Design the roots of the characteristic equation of the
closed loop in the unit circle. Then the condition
∗ ∗
𝐷 (𝑧)𝐷 (𝑧) + 𝑁 (𝑧)𝑁 (𝑧) = 0 𝐷 (𝑧) 𝑌 (𝑧) − 𝑉(𝑧) = (𝑧𝑝 − 1) 𝑌 (𝑧) − 𝑉(𝑧) = 0

2. Provide 𝐷 (𝑧) 𝑌∗ (𝑧) − 𝑉(𝑧) = 0, design the roots of the holed. Because it is the difference of the value of periodic
denominator 𝐷 (z) = 0 at the desired frequencies in functions at the present time and shifted one period ahead
order to track the references and eliminating the because p is the number of time steps per period.
disturbances. How to determine the parameters
𝛼 , 𝛼 , ⋯ 𝛼 , will be the designing freedom. A similar way as
shown in the follows has been used in conventional studies [4].
III. FUNDAMENTAL ASPECTS OF HIGHER ORDER RC
𝐷 (𝑧) = 𝑧 − 𝛼 𝑧( )
+ 𝛼 𝑧( )
+⋯+𝛼
Higher order RC refers to RC laws that make use of error
information from not just the previous period, but additional = (𝑧 − 1) (13)
previous periods as well [7]. Generally, it has been difficult to

694
We have shown the way [8] how to determine the parameters 1) , that is conventional second order RC. From the discussion
𝛼 , 𝛼 , ⋯ 𝛼 from (13). In addition, instead above, we propose above, we make a summary as Property 2.
a more prospected way to use the parameters
Property 2: For the second order RC (15),
𝛼 , 𝛼 , ⋯ 𝛼 as follows [8],
1) The free parameter 𝑞 must be a real number.
𝐷 (𝑧) = 𝑧 − 𝛼 𝑧( )
+ 𝛼 𝑧( )
+⋯+𝛼
2) No new frequencies of the forcing function compare with
= (𝑧 − 1) (𝑧 − 𝑞 ) ⋯ (𝑧 − 𝑞 ) , (14) 𝑞 = 1,
where 𝑛 + 𝑛 + ⋯ + 𝑛 = 𝑁 . But that will be difficult to Even such restrictions will confine the usability of the
show how to determine all of the free parameters 𝛼 , 𝛼 , ⋯ 𝛼 freedom, but the second order RC is the simplest higher order
and 𝑛 , 𝑛 , ⋯ 𝑛 in (14), we only show the case N=3, and 𝑛 = RC, and it is as good as well using low order RC in the
𝑛 = 𝑛 = 1 in order to show the new proposed design method applications. So clear the ability of the free parameter also is
of (14) have many prospect characteristics. And left so many important. We will discuss that in the simulation section, and
objects for future studies. This paper is a constituted study of show you some interesting characteristics.
above, and will show some basic characteristics that will bring
so many hits for further studies.
Property 1 in above just shows general conditions for design C. Third order RC
the repetitive controller, but in design the specified higher Compare with second order RC, third order RC is another
order RC, we find more difficult situations. So at the paper we interesting controller. For N=3, we have
show some simple cases in order to better to understand the
difficulties to design of (14), then show the prospects for higher 𝐷 (𝑧) = 𝑧 − [𝛼 𝑧 +𝛼 𝑧 +𝛼 ]
order RCs. = (𝑧 − 1)(z − 𝑞 )(z − 𝑞 ). (20)
A. Second Order RC The relations between 𝛼 , 𝛼 , 𝛼 and 𝑞 , 𝑞 can be directly
For N=2, from (14) we have calculated as follows [8] and satisfy 𝛼 + 𝛼 + 𝛼 = 1.
𝐷 (𝑧) = 𝑧 − [𝛼 𝑧 + 𝛼 ] = (𝑧 − 1)(𝑧 − 𝑞 ). (15) 𝛼 =1+𝑞 + 𝑞
Compare the last equation, then the parameters 𝛼 , 𝛼 can be 𝛼 = −(𝑞 𝑞 + 𝑞 + 𝑞 ) (21)
easily determined as follows
𝛼 =𝑞 𝑞 .
𝛼 =𝑞 +1
. (16) In the study [8], we proposed the idea of using (20) instead
𝛼 = −𝑞
of the conventional RC (13) and just discussed one of case for
The condition of (12) also can be checked easily. Then the 𝑞 , 𝑞 , in order to show our proposed new method had some
function of the parameter 𝑞 will be interested. In the general new abilities. But we had not mentioned any constrained
case, it can be selected as conditions about the free parameters 𝑞 , 𝑞 . In this paper, we
will clear all of the constrained conditions.
𝑞 =𝑟𝑒 = 𝑟 (𝑐𝑜𝑠𝜃 + 𝑗𝑠𝑖𝑛𝜃 ) . (17)
Because the controller parameters 𝛼 , 𝛼 , 𝛼 in control role
But considering (8), when implement the controller in the
(8) must be real number in order to implement in practical
application, the parameters 𝛼 , 𝛼 must be real numbers. So
situations, from the relations of (21), one has the following
from (16), the 𝑞 must be a real number. That means the 𝜃 limitations.
must be 0 in (17). From the Property 1 (2), the forcing function
on the right side of (5) becomes zero is necessary. 1) 𝑞 𝑞 must be a real number.
∗ ∗
𝐷 (𝑧) 𝑌 (𝑧) − 𝑉(𝑧) = (𝑧𝑝 − 1)(𝑧𝑝 − 𝑟1 ) 𝑌 (𝑧) − 𝑉(𝑧) = 0 2) 𝑞 + 𝑞 must be a real number.
That also means the frequencies of the command must contain In order to satisfy such limitations, following several choices
in (𝑧 − 1)(𝑧 − 𝑟 ) = 0. can be considered for different combination of 𝑛 , 𝑛 , 𝑛 .
Except the conventional case (𝑛 = 3, 𝑛 = 𝑛 = 0 ), we have
Consider 𝑧 = 𝑟(𝑐𝑜𝑠𝜃 + 𝑖𝑠𝑖𝑛𝜃), 𝑧 = 𝑟 (𝑐𝑜𝑠𝜃 + 𝑖𝑠𝑖𝑛𝜃 )
and make use of the De Moivre’s theorem to write the solution 1)𝑛 = 1, 𝑛 = 2, 𝑛 = 0: (or 𝑛 = 1, 𝑛 = 0, 𝑛 = 2).
of the equation 𝑧 = 𝑧 as
𝐷 (𝑧) = 𝑧 − [𝛼 𝑧 + 𝛼 𝑧 + 𝛼 ] = (𝑧 − 1)(z − 𝑞 ) ,
/ / 𝜃 + 2𝑘𝜋 𝜃 + 2𝑘𝜋
𝑧=𝑧 =𝑟 𝑐𝑜𝑠 + 𝑖𝑠𝑖𝑛 , (18) So the free parameter 𝑞 must be real number that will satisfy
𝑛 𝑛 the restriction 1) and 2). This is similar to the second order RC
/ 𝜃 + 2𝑘𝜋 except with the square of factor that contain the free parameter
𝑟=𝑟 , 𝜃= , (𝑘 = 0,1, ⋯ , 𝑛 − 1). (19) (𝑞 = 𝑞 ). The effect of the real number can be thought as
𝑛
similar as in the discussion of Property 2 and the effect of the
Then for both (𝑧 − 1) = 0 and (𝑧 − 𝑟 ) = 0, we have the square can be thought as similar as in the discussion of the
/
solution with different gains as 𝑟 = 1 𝑎𝑛𝑑 𝑟 = 𝑟 , but same conventional case [4], [7].
angles 𝜃 = , (𝜃 = 0). That means even we add a new 2) 𝑛 = 1, 𝑛 = 1, 𝑛 = 1:
freedom 𝑞 , but we can’t introduce more frequencies compare
𝑝
with the forcing function (𝑧 − 1) 𝑌∗ (𝑧) − 𝑉(𝑧) = 0, (𝑞 =

695
In order to satisfy constrained conditions 1) and 2), one must
select the free parameters 𝑞 , 𝑞 as follows.
Case1: 𝑞 , 𝑞 are different real numbers.
Case 2: 𝑞 , 𝑞 are complex conjugate numbers.
It is clear in Case 1, because if the parameters 𝑞 , 𝑞 are
different real numbers then 𝛼 , 𝛼 , 𝛼 will be real numbers.
From the relation (21), we have the conditions, 𝑞 𝑞 and 𝑞 +
𝑞 must be real numbers, that is the Case 2. Any other
combinations, for example, 𝑞 is real number and 𝑞 is a
complex numbers, and vice versa, or 𝑞 , 𝑞 are different (or
same) complex numbers can’t satisfy the condition 1) and 2).
For Case 1, the effects of the real parameters can be recognized
as similar as discussed in second order higher RC. For Case 2,
we had shown the selecting method in [8] and shown the
abilities for robustness when the disturbance period has some Real Axis
Fig.2 (a) Root locus of the second order RC (q=0.9)
fluctuations. But such discussions are based on frequency
domain, in this paper, we will show some abilities in time
domain at the following simulation section and introduce some
useful knowledge in practical uses. From the discussion above,
[dB]

we make a summary as Property 3.


Property 3: For the third order RC (20), only following cases
are possible for a selection of the free parameter 𝑞 , 𝑞
1) 𝑞 , 𝑞 are different real numbers.
2) 𝑞 , 𝑞 are complex conjugate numbers.
From above discussions, it is not difficult to understand that
[Hz]
will be very complex to find the constrained conditions for the Fig.2 (b) Sensitivity function of the second order RC
general case of higher order RCs (14). But if you understand
the essence of the second order RC and the third order RC
discussed above that will give you some useful introduction for
design higher order RCs. For example, if you would like to use
a forth order RC, then the following will be suggested way.
𝐷 (𝑧) = 𝑧 − [𝛼 𝑧 +𝛼 𝑧 +𝛼 𝑧 +𝛼 ]
= (𝑧 − 1)(z − 𝑞 )(z − 𝑞 )(z − 𝑞 ) (22)
You can chose 𝑞 as a real number as discussed at second
higher order RC and chose 𝑞 , 𝑞 as complex conjugate
numbers as discussed at third higher order RC. That is why the
studies of second or third order RCs are important.

Real Axis
IV. NUMERICAL RESULTS Fig.3 (a) Root locus of the second order RC (q=-0.9)
We have shown some restrictions on how to select the free
parameters for second and third order RC, now we would like
to show some roles of such free parameters from numerical
results. In order to show some law characteristics of the higher
[dB]

order RC, we assume that the closed loop transfer function of


a feedback control system 𝐺(𝑧) = 1 at Fig.1. We use
Matlab/Simulink tool for simulation.
A. Second order RC
As we discussed in Section III, the second order RC cannot
bring a new frequency of the forcing function. But, we will
show some role of the free parameter 𝑞 by simulations. We
have the second order RC as follows, where 𝜑 is a scalar gain [Hz]
parameter.
Fig.3 (b) Sensitivity function of the second order RC

696
𝜑[𝛼 𝑧 + 𝛼 ] 𝜑[𝛼 𝑧 + 𝛼 ]
𝑅 (𝑧) = = (23)
𝑧 − [𝛼 𝑧 + 𝛼 ] (𝑧 − 1)(𝑧 − 𝑞 )
We choose the sampling rate f=100 [Hz], and the reference
signal is periodic with period 𝑝 = 10 time steps. So the
fundament frequency is 10 [Hz], the Nyquist frequency is 50
[Hz].
The root locus plot for the second order RC control system is
defined by
𝜑[𝛼 𝑧 + 𝛼 ]
= −1 , (24)
(𝑧 − 1)(𝑧 − 𝑞 )
where 𝜑 changed from 0 to ∞ . The sensitivity function is [s]
defined by 𝑆(𝑧) = [1 + 𝑅 (𝑧)𝐺(𝑧)] . We chose 𝑞 = Fig.4 (a) Output of the second order RC (q=0.9)
0.9 and 𝑞 = −0.9 , then plot out the root locus and the
sensitivity transfer function (𝜑 = 1) at Fig. 2 and Fig.3. 3

Even the second order RC can’t bring a new frequency of the 2


forcing function, but the free parameter 𝑞 can effete the
sensitivity transfer function very large between the 1
fundamental frequencies (Fig.3(b)). In addition, that is
Output

0
interesting for the step response shown in Fig.4. The free
parameter can be used to improve the transit response. So even
-1
the second order RC not so much contribution at the frequency
domain, but may have some contribution at the time domain if -2
you make a careful choice. Because the study of the higher
order RC in time domain is few and far between, this will be -3
0 0.2 0.4 0.6 0.8 1
new knowledge about higher order RC. [s]
Time(s)
B. Third order RC Fig.4 (b) Output of the second order RC (q=-0.9)
From (20), the third order RC is as follows.
𝜑[𝛼 𝑧 +𝛼 𝑧 +𝛼 ]
𝑅 (𝑧) =
𝑧 − 𝛼 𝑧 +𝛼 𝑧 +𝛼
𝜑[𝛼 𝑧 + 𝛼 𝑧 + 𝛼 ]
= (25)
(𝑧 − 1)(z − 𝑞 )(z − 𝑞 )
The simulation conditions are just as same as above
subsection. If you select the parameters 𝑞 , 𝑞 as the Case 1,
then you can get some image from the discussions for second
order RC. For the selection as Case 2, we have discussed it at
[8] for frequency domain only. So we give some simulation
results for time domain at Fig.5 and Fig.6. Fig.5 show the
conventional case ( 𝑞 = 𝑞 = 1 ) , when the disturbance Fig.5 (a) Output of the old third order RC(disturbance=10Hz)
frequency is fluctuate from 10[Hz] to 11[Hz], the tracking
accuracy will be problem (Fig.5 (b)). But using our proposed
method [8], Fig.6 show even disturbance frequency is 11[Hz],
the tracking accuracy has not been affected (Fig.6 (b)). The
difference is that we take p=10 in simulation of Fig.5 and
p=100 in simulation of Fig.6, so if one need a smooth tracking,
it is better to select a large p. But a large p means very higher
order controller, it will raise the cost at applications. Such
problem will be appeared in time domain, but not appear when
one only study in frequency domain [8].

V. CONCLUSION
The second and third order RCs have been studied very
detail. The constrained conditions in both cases have been
cleared. Both frequency and time domain characteristics Fig.5 (b) Output of the old third order RC(disturbance=11Hz)

697
have been shown. In addition, such results will be very
important when one desires to design higher order RCs. Using
these results will let one finds a new way to use higher order
RCs.

ACKNOWLEDGMENT
The author H. J. Guo wish to acknowledge Prof. R.W.
Longman (Columbia University, Mechanical Department) for
his kind introduction and assistance in this study. This work
was supported by JSPS KAKENHI Grant Number 15K05950.

REFERENCES
[1] E. S. Ahn, R.W. Longman, J. J. Kim and B. N. Agrawal, “Improving
laser communication between formation flying satellites using
repetitive control jitter mitigation.” In 7th International Workshop on Fig.6 (a) Output of new third order RC(disturbance=10Hz)
Satellite Constellations and Formation Flying, 13-15 March, Lisbon,
2013.
[2] T. C. Tsao, X. Y. Qian and M. Nemani, “Repetitive control for
asymptotic tracking of periodic signals with an uncertain period,” J
Dyn Syst Meas Control, No.122, pp. 364-368, 2000.
[3] M. Steinbuch, “Repetitive control for systems with uncertain period,”
Automatica, No.38(12), pp. 2103-2109, 2002.
[4] M. Steinbuch, S. Weiland, J.V.d. Eerenbeemt and T. Singh, “On noise
and period-time sensitivity in high order repetitive control,” In 43rd
IEEE Conference on Decision and Control, pp. 1295-1300, 2004.
[5] C-P. Lo and R. W. Longman, “Frequency response analysis of higher
order repetitive control.” Advances in the Astronautical Sciences,
No.123, 2006, pp. 1183–1202.
[6] C-P. Lo and R. W. Longman, “Frequency response analysis of higher
order repetitive control,” Advances in the Astronautical Sciences, 123,
pp. 1183–1202, 2006.
[7] R.W. Longman, “On the theory and design of linear repetitive control
systems,” European Journal of Control, 5, pp. 447-496, 2010.
[8] H. J. Guo, R.W. Longman and T. Ishihara, “A Design Approach for Fig.6 (b) Output of new third order RC(disturbance=11Hz)
Insensitivity to Disturbance Period Fluctuations Using Higher Order
Repetitive Control,” IFAC, 2016.
[9] B. A. Francis and W. M. Wonham, “The internal model of control
theory,” Automatica 12, pp. 457-465, 1976.
[10] Perry Y. Li, “Advanced control system,” University of Minnesota, ME
8281, pp. 131-147, 2006.

698

You might also like