J Control Theory Appl 2012 10 (1) 5055

DOI 10.1007/s11768-012-8221-3

A nonlinear repetitive controller

Xin TANG 1 , An LUO 2 , Chunming TU 2
1.College of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha Hunan 410076, China;
2.College of Electrical and Information Engineering, Hunan University, Changsha Hunan 410082, China

Abstract: A nonlinear repetitive controller is proposed. The new method is mainly composed of a repetitive control
part and a deadband relay. Whenever the input error goes beyond the range of the deadband relay, the control loop is driven
dominantly by the deadband relay to obtain fast dynamic response and meanwhile to avoid the saturation of the repetitive
control part. After the input error falls within the range of the deadband relay, the deadband relay automatically turns off
and the repetitive control alone governs the current control to eliminate the steady state error. A systematic methodology
is established and it is linked to the conventional control system design. The proposed scheme is practically applied to the
current control of active lter. Experimental results veried the feasibility of the proposed method.
Keywords: Nonlinear repetitive control; Active lter; Dynamic response

1 Introduction control algorithm has to nd the trade-off between the sys-

According to [13], repetitive control theory derived tem stability and the dynamic performance for too large
from the internal model principle provides a solution for proportional parameter will cause the system unstable. In
eliminating periodic errors which occur in a dynamic sys- [6, 12, 13], an odd-harmonic repetitive control can introduce
tem. A repetitive controller can be viewed as a periodic high gain only at odd harmonic frequencies and has been
waveform generator augmented within the control loop of a applied to active lters. In this method, however, dynamic
control system, so that the periodic errors can be eliminated. response rarely has been taken into account. In this paper,
References [46] developed the repetitive controller and ap- an alternative nonlinear repetitive control method, which is
plied it to various industrial applications including power mainly composed of a repetitive control part and a deadband
electronic systems such as rectiers, inverters and active l- relay, is proposed to obtain better dynamic response than a
ters. In practice, the standard repetitive controller that only basic repetitive control. The proposed scheme, furthermore,
has a pure internal mode often results in very narrow closed- is practically applied to the current control of active power
loop bandwidth due to the large phase shift of the inter- lter in which the reference and disturbance signals appear-
nal mode. This means that such a repetitive control system ing in control loop, in steady state, are always periodic sig-
may cause sluggish transient response and has poor track- nals. Experimental examples demonstrate the simplicity and
ing performance, although it is able to track the periodic effectiveness of the proposed scheme.
input perfectly at the steady state. Reference [7] introduced
a so-called regeneration spectrum to estimate the locus of 2 Principle of nonlinear repetitive controller
the dominant closed-loop poles of the repetitive control sys- Repetitive control, which can be viewed as a waveform
tem. In fact, in a control system, the design of the repetitive generator based on the internal model principle, can accom-
portion of the controller can be separated from the conven- plish perfect tracking in the steady state if the generator for
tional servo-controller. In this way, the repetitive controller all frequency modes of the periodic signal is included in the
can be added afterwards without disturbing the existing sta- control loop.
ble servo-controller. Reference [8] presented some insights A generator for periodic signals is shown in Fig. 1 (a)
into the tradeoff between system stability and tracking per- where x(t) is input signal and c(t) is output signal. Such a
formance by combining the regeneration spectrum and the delay system has the transfer function 1/(1 esT ), which
sensitivity function. For time-varying systems or systems has an innite number of poles on the imaginary axis at
with many plant uncertainties, references [9, 10] developed jkwn ; k = 0, 1, 2, . . . , where wn = 2/T . It is ex-
adaptive repetitive control schemes to eliminate periodic er- pected from the internal model principle that the asymptotic
rors. Although these methods can track the changing plant tracking for exogenous periodic signals can be achieved
dynamics, they have the drawback that the number of pa- by incorporating the plug-in model 1/(1 esT ) into the
rameters to be estimated is proportional to the frequency closed-loop system [8].
modes selected to be cancelled. Reference [11] presented A typical repetitive controller is shown in Fig. 1 (b) where
the repetitive control combined with conventional PI to im- Kq (s) is a low-pass lter. This low-pass lter cascaded with
prove the dynamic performance. However, the improved the time delay would in principle restrict the bandwidth of
Received 27 November 2008; revised 8 April 2011.
This work was supported by the National High Technology Research and Development of China (No. 2004AA001032), the Planed Science and
Technology Project of Hunnan Province (No. 2010FJ3032), and the Natural Science Foundation of Hunnan Province (No. 11JJ5034).

c South China University of Technology and Academy of Mathematics and Systems Science, CAS and Springer-Verlag Berlin Heidelberg 2012
 X. Tang et al. / J Control Theory Appl 2012 10 (1) 5055 51

the controller and at the same time reinforce the stability of can be designed independently.
overall system. On the other hand, however, Kq (s) only al- 3.1 Design of the repetitive control part
lows the specied harmonics of input signal to pass, subject
The transfer function of the repetitive control part in
to the lter bandwidth, and inuence the steady-state char-
Fig. 1 can be written as
acteristics of the controller. The design of Kq (s) is nom-
1
inally made by achieving the tradeoff of tracking perfor- Grc (s) = , (1)
mance and system stability. 1 Kq (s)esT
The proposed nonlinear repetitive controller is shown in and where T represents the period of the input signal.
Fig. 1 (c) where Gl (s) is a lead compensator. This new Let K(s) be a predesigned controller such that the
method is mainly composed of a repetitive control part and closed-loop system So (s) = 1/(1 + Gp (s)) is internally
a deadband relay which is a relay with a deadband in the stable, and let Kq (s) = 1/(1 + q s) with Kq (s) = 1,
vicinity of zero input. Whenever the input error goes be- where   denotes the H -norm of a stable transfer func-
yond the range of the deadband relay, the current control tion.
loop is driven dominantly by the deadband relay with a near- When the repetitive control part works alone, the closed-
maximum control output, and then obtains fast dynamic re- loop characteristic equation of Fig. 2 is given by
sponse, at the same time avoids the saturation of repetitive 1 + Gp (s) Kq (s)esT = 0. (2)
control part. After the input error falls within the range of
the deadband relay, the deadband relay automatically turns Let the regeneration spectrum is dened as
off and the repetitive control alone governs the control sys- R(w) := |Kq (jw)So (jw)| . (3)
tem to eliminate the steady state error.
By assumption, 1 + Gp (s) = 0 has no zero poles in the
right-half plane. If R(w) < 1 for all w, according to the
small gain theory, the closed-loop system in Fig. 2 without
the deadband controller is stable for T > 0.
Let So  = Ms and |So (jws )| = 1 shown in Fig. 3.
Since Kq  = 1, a condition to ensure R(w) < 1 is given
by
 1  1
 
|Kq (jws )| =  < . (4)
1 + jws q Ms
This characterizes a criterion on the stability for the de-
sign of Kq (s) since ws and Ms can be found from So (s) as
Gp (s) is given.

Fig. 1 The repetitive control.

3 System design methodology

For a controlled plant Go (s), the feedback design is to de- Fig. 3 The frequency characteristics of So (s) and Kq (s).
termine at least a controller K(s) such that the unity feed-
back system is internally stable. Fig. 2 shows a nonlinear 3.2 Design of deadband relay
repetitive control system where Gp (s) = K(s)Go (s) is a According to [14], two-valued relay cannot be applied to
proper rational function, r(t) is reference input, y(t) is sys- high-performance control systems because it often causes
tem output and e(t) is tracking error. limit cycle problems. A deadband relay is a relay hav-
ing three different output values including zero. Fig. 4 (a)
shows the input-output relation of this type relay where d
denotes the relay output amplitude, 2h represents the dead-
band width, x is the input, and z is the output. It can be seen
that the output of the deadband relay is zero when the input
falls within the range of the deadband. Hence, the deadband
Fig. 2 A nonlinear repetitive control system. relay will be out of action and then the limit cycle oscilla-
As the deadband relay will automatically turn off when tions can be avoided.
the tracking error becomes small, the repetitive control part The describing function of the deadband relay can be de-
52 X. Tang et al. / J Control Theory Appl 2012 10 (1) 5055

rived as  reasonable selection may be 510% of the maximum value

4d h of the absolute steady-state error of the closed-loop control
N (X) = 1 ( )2 , (5) system using the repetitive control part alone. This maxi-
X X
where N (X) is the describing function of the deadband re- mum value can easily be estimated via a numerical investi-
lay and X is the sinusoidal input amplitude of the deadband gation because the repetitive control part has already been
relay. designed.
In the analysis of describing nonlinear system function, The task of the lead compensator Gl (s) is to move
the inverse of the describing function is usually employed. the locus of Gl (s)Gp (s)/(1 + Grc (s)Gp (s)) away from
The inverse of the above describing function is denoted by the (1/N (X)) locus, equivalently away from the
1 X ([1/N (X)]min , j0). In the other words, the task of the lead
=  . (6) compensator is to ensure small enough. This is an ordinary
N (X) 4d 1 ( h )2 frequency-domain design problem.
X
The inverse of the deadband relay describing function is If the Gp (s)/(1 + Grc (s)Gp (s)) locus is already away
not monotonous function of X. It has a minimum value at from the (1/N (X)) locus to a satisfactory degree, it is
some point in the admissible set of X: {X| X>h}. This not necessary to include the lead compensator Gl (s). This
minimum point can be calculated as will result in a simplied controller.

X = 3h. (7)
The minimum value of (1/N (X)) is
 
1 3h
= . (8)
N (X) min 4 2d
The minimum value of the inverse of the describing func-
tion is proportional to the ratio of the deadband band to the
relay output amplitude.
The characteristic equation of closed-loop control system
shown in Fig. 2 is
1 + [N (X)Gl (s) + Grc (s)]Gp (s) = 0,
or
Gl (s)Gp (s) 1
= . (9)
1 + Grc (s)Gp (s) N (X)
Because the previously designed repetitive control part
guarantees the stability of the current control loop only
with the repetitive control part, 1 + Grc (s)Gp (s) only
includes zeros in the left-half s plane. Hence, the min-
imum phase property of the transfer function Grc (s)
and Gp (s) results in the minimum phase property of
Gl (s)Gp (s)/(1 + Grc (s)Gp (s)). Therefore, if the Nyquist
plot of Gl (s)Gp (s)/(1 + Grc (s)Gp (s)) does not cross the Fig. 4 Characteristics of the deadband relay.
(1/N (X)) locus as shown in Fig. 4 (b), the closed-loop
control system is stable and there is no limit cycle. 4 Application to active lter
The [1/N (X)]min depends on the ratio of the dead- Power quality is an increasingly important problem that
band width to the relay output amplitude. The relay output affects users of the power grid and has signicant economic
amplitude d is often selected to be the maximum admissi- consequences. According to [6], active power lters as so-
ble control input. Hence, the distance of (1/N (X)) locus lutions to power quality problems have become more and
from the imaginary axis relies only on the deadband width. more important nowadays
Larger deadband width results in more stable robust control A schematic diagram for a single-phase shunt active l-
loop but produces less compensation effect. If the deadband ter is shown in Fig. 5, where L is the lter inductance, C
half-width h is greater than the maximum value of the ab- is the capacitance, R sums up the parasitic resistance of all
solute tracking error, the relay will never take effect. Hence, the converter elements, iF (t) is the lter inductor current,
the selection of h must satisfy is (t) is the supply current, iL (t) is the load current, uc (t) is
< h < |e(t)|max , (10) the DC bus voltage, and us (t) is the supply voltage, respec-
where |e(t)|max is the maximum value of the absolute track- tively.
ing error and is the critical magnitude of the deadband The aim of a shunt active lter is to generate and inject a
half-width h which let Gl (s)Gp (s)/(1 + Grc (s)Gp (s)) lo- suitable current iF (t) into the power system so as to cancel
cus begin to intersect with the (1/N (X)) locus, i.e., the the harmonic content of the distorted load and to obtain a
closed-loop control system is critical stable. supply current in phase with the supply voltage. The idea
Generally, the deadband half-width h is selected to be as can be also described by
small as possible in order to obtain fast dynamic response. A IF iF (t) = iLh + iLfq , THEN is (t) = iL iF = iLfp ,
 X. Tang et al. / J Control Theory Appl 2012 10 (1) 5055 53

where iLfp iLfq and iLh are the real content, reactive content From the energy-balancing point of view, the required
and harmonic content of load current, respectively. supply current ir (t), which is generated by the outer volt-
The nonlinear load is full-wave diode rectier with C = age loop controller will be equal to the real content iLfp of
1500 F in parallel with R = 50 . The nominal AC mains load current, when the DC voltage uc (t) settles down and if
root-mean-square (RMS) voltage is 110 V and its nominal the current loop of active lter has zero steady-state [15] .
frequency is 50 Hz.
4.3 Controller design
The main circuit of the shunt active lter was built by a
full-bridge boost converter with IGBT (insulated gate bipo- From (11), if the DC bus voltage is assumed to be well
lar transistor) switches and a DC capacitor C = 470 F. regulated at UDC , the current dynamics can be approxi-
Although the bang-bang control will cause the variable to mated as
switch frequency, the settling time is short. Therefore, the LiF (t) + RiF (t) = 2UDC d(t) uc (t) us (t).
low-pass lter eliminating the switching frequency can be A block diagram representation of the current loop control
designed only considering the steady-state model. As the is shown in Fig. 7.
frequency of the triangle waveform used as the carrier sig-
nal is 20 kHz, the parameter of the low-pass lter is set as
L = 1 mH.

Fig. 7 Current control loop.

The required DC bus voltage is chosen to be 200 V. The
transfer function, Gp (s), in Fig. 2 can be calculated as
400
Gp (s) = .
0.036 + 0.001s
As the nominal frequency fu is 50 Hz, the input signal pe-
riod of the current control loop is T = 20 ms. As stated in
Section 3, for stabilizing the system, the parameter q in the
Fig. 5 Single-phase shunt active lter. low pass lter Kq (s) is evaluated as
4.1 Modeling q = 0.06.
According to [15], the state-space averaging model of the Since the maximum input voltage of triangle pulse modu-
active lter in Fig. 5 describing the voltage and current dy- lator is 5 V, the output amplitude of the deadband relay is
namics are given by determined as 5 V. The deaband half-width is selected to be

LiF (t) + RiF (t) = (2d(t) 1)uc (t) us (t), 0.3, which is about 10% the maximum value of the absolute
(11) current tracking error obtained from a numerical investiga-
C uC (t) = iF (t),
tion about the repetitive control part working alone. There-
where d(t) is the duty ratio. fore, two parameters of the deadband relay are determined
4.2 Control structure as
For proper operation of the inverter, it is necessary to d = 5, h = 0.3.
maintain a DC bus voltage uc (t) greater than the peak of By means of the CONTROL SYSTEM Toolbox, the
the supply voltage us (t). Thus, the shunt active lter need Nyquist plot of Gp (s)/(1 + Grc (s)Gp (s)) can be con-
be decomposed into a voltage control loop and a current veniently obtained and does not intersect with the
control loop. The control block diagram is shown in Fig. (1/N (X)) locus. As a result, there is no need to include
6, where is a constant, ur (t) is the required DC voltage, an additional lead compensator for this system, i.e.,
ec (t) is the current tracking error, and ir (t) is the required Gl (s) = 1.
supply current.
The setting of was to make sure that the multiplier output
would not be easily saturated and was set to 0.01 such that
the peak value of us (t)was about 1.56 V.
For the voltage control loop, the PI controller Gcu (s) =
KP1 + KI1 /s is easily designed and the parameters are se-
lected as KP1 = 0.0295 and KI1 = 0.4639.

5 Experimental results
To verify the effectiveness of the proposed method, the
active lter shown in Fig. 5 with different controllers was
tested. Fig. 8 shows the dynamic performance of active l-
ter with a basic repetitive controller. Note that the settling
time took about 10 periods. Fig. 9 shows the dynamic per-
Fig. 6 Block diagram of the current and voltage control loops. formance of active lter with the proposed method. The set-
54 X. Tang et al. / J Control Theory Appl 2012 10 (1) 5055

tling time only took about 3 periods. Fig. 11 shows steady-state performance of active lter
with the traditional PI controller. The high order harmon-
ics had a certain extent been reduced and the current THD
was 9.7%. The power factor of the overall system was 0.91.
Fig. 12 shows steady-state performance of active lter with
the proposed method. There was clearly a signicant im-
provement in the power factor and the current THD. The
high order harmonics had clearly been reduced and the cur-
rent THD was 5.6%. The power factor of the overall system
was 0.98.

Fig. 8 The dynamic performance of active lter with a basic repet-

itive controller.

Fig. 9 The dynamic performance of active lter with the proposed

method.
Fig. 10 shows the load current waveform and the load cur-
rent spectrum. The RMS of load current was 1.97 A. There Fig. 11 Steady-state performance of active lter with the tradi-
were clearly a lot of 3rd, 5th, 9th, etc. high-order harmonics tional PI controller.
and the current total harmonic distortion (THD) was 61.8%.
The power factor of the circuit was 0.77.

Fig. 12 Steady-state performance of active lter with the proposed

Fig. 10 Nonlinear load. method.
 X. Tang et al. / J Control Theory Appl 2012 10 (1) 5055 55

6 Conclusions current waveform correction for shunt active power lter. Transaction
of China Electrotechnical Society, 2007, 22(2): 127 133.
In this paper, a nonlinear repetitive controller was pro-
[12] G. Escobar, P. R. Martnez, J. Leyva-Ramos, et al. A negative
posed to obtain better dynamic response than a basic repet-
feedback repetitive control scheme for harmonic compensation. IEEE
itive controller. The aforementioned results veried the va- Transactions on Industrial Electronics, 2006, 53(4): 1383 1386.
lidity of the proposed method by applying it to current con- [13] R. Grino, R. Cardoner, R. Costa-Castello. Odd-harmonic digital
trol of active lter. Comparison to the adaptive repetitive repetitive control of a single-phase current active lter. IEEE
method, the proposed method is simpler and faster because Transactions on Power Electronics, 2004, 19(4): 1060 1068.

it does not need any parameter identication. In addition, [14] K. M. Tsang, G. Li. Robust nonlinear nominal-model following
control to overcome deadzone nonlinearities. IEEE Transactions on
the repetitive control combined with a PI has to achieve Industrial Electronics, 2001, 48(1): 177 184.
a trade-off between the stability and the dynamic perfor-
[15] K. M. Tsang, W. L. Chan. Design of single phase active power lter
mance, the proposed method is faster. using analogue cascade controller. IEE Proceedings Electric Power
Applications, 2006, 153(5): 735 741.
References
[1] T. Inoue, M. Nakano, S. Iwai. High accuracy control of Xin TANG was born in Hunan, China, in 1975.
servomechanism for repeated contouring. Proceedings of 10th He received his M.S. and Ph.D. degrees from Cen-
Incremental Motion Control System Devices. Oxford: Pergamon, tral South University, Changsha, in 2001 and 2005,
1981: 258 292. respectively. His research interests include elec-
tric power savings and active power lters. E-mail:
[2] T. Inoue, M. Nakano. High accuracy control of a proton synchrotron
tangxin csu@163.com.
magnet power supply. Proceedings of the 8th IFAC World Congress.
Oxford: Pergamon, 1981: 216 221.
[3] S. Hara, Y. Yamammoto, T. Omata, et al. Repetitive control system:
A new type servo system for periodic exogenous signals. IEEE
Transactions on Automation Control, 1998, 33(7): 659 666. An LUO was born in Hunan, China, in 1957. He
received his B.S. and M.S. degrees from Hunan
[4] K. Zhang, Y. Kang, J. Xiong, et al. Direct repetitive control of SPWM University, in 1982 and 1986, respectively, and
inverter for UPS purpose. IEEE Transactions on Power Electronics, Ph.D. degree from Zhejiang University, Zhejiang,
2003, 18(3): 784 792. China, in 1993. He was with the Central South
[5] Z. Zhou. Zero-phase odd-harmonic repetitive controller for a single- University, as a professor between 1996 and 2002.
phase PWM inverter. IEEE Transactions on Power Electronics, 2006, Starting in 2003, he became a professor at Hunan
21(1): 193 201. University. He is engaged in research on power
conversion system, harmonics suppression and re-
[6] R. Grino, R. Cardoner, R. Costa-Castello. Digital repetitive control
active power compensation, and intelligent control theory and application.
of a three-phase four-wire shunt active lter. IEEE Transactions on
He has published over 100 journal and conference articles. He is a recipi-
Industrial Electronics, 2007, 54(3): 1495 1503.
ent of the 2006 National Scientic and Technological Awards of China, the
[7] K. Srinivasan, F. R. Shaw. Analysis and design of repetitive control 2005 Scientic and Technological Awards from the National Mechanical
systems using the regeneration spectrum. Transactions of the ASME. Industry Association of China, and the 2007 Scientic and Technological
Journal of Dynamic Systems, Measurement and Control, 1991, Awards from the Hunan Province of China. He currently serves as the As-
113(2): 216 222. sociate Board Chairperson of Hunan Society of Electrical Engineering. He
[8] M. C. Tsai, W. S. Yao. Design of a plug-in type repetitive also serves as the Chief of Hunan Electric Science and Application Labo-
controller for periodic inputs. IEEE Transactions on Control Systems ratory. E-mail: an luo@hnu.edu.cn.
Technology, 2002, 10(4): 547 555. Chunming TU was born in Jiangxi, China, in
[9] J. Hu, M. Tomizuka. A new plug-in adaptive controller for rejection of 1976. He received his M.S. and Ph.D. degrees
periodic disturbances. Transactions of the ASME. Journal of Dynamic from Central South University, Changsha, in 2001
Systems, Measurement and Control, 1991, 33(6): 1 5. and 2004, respectively. His research interests in-
clude electric power savings and active power l-
[10] Y. Tzou. Adaptive repetitive control of PWM inverter for very low ters. E-mail: chunming tu@263.com.
HD AC-voltage regulation with unknown loads. IEEE Transactions -
on Power Electornics, 1999, 14 (5): 971 981.
[11] P. Geng, K. Dai, X. Wei. The repetitive control algorithm based