IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 18, NO.

1, JANUARY 2003 309

Digital Repetitive Controlled Three-Phase

PWM Rectifier
Keliang Zhou and Danwei Wang, Member, IEEE

Abstract—In this paper, a digital repetitive control (RC) vides a zero tracking error solution for CVCF PWM converters.
strategy is proposed to achieve zero tracking error for con- It was first applied to single-phase inverters with good prelim-
stant-voltage constant-frequency (CVCF) pulse width modulation inary results in [7], which attracts more research in [8]–[10]
(PWM) converters. The proposed control scheme is of “plug-in”
structure: a plug-in digital repetitive controller plus a conventional for single-phase inverter, [11] for 3-phase rectifier, and [12] for
controller (e.g., PD controller). The design of the plug-in repetitive 3-phase inverter. In this case, the robustness analysis and design
learning controller is systematically developed. The stability of repetitive controllers are neither systematic nor complete.
analysis of overall system is discussed. A repetitive controlled In this paper, a digital repetitive control scheme is proposed
three-phase reversible PWM rectifier is given as an application to achieve zero tracking error for CVCF PWM converters. It is
example. Near unit power factor and constant output dc voltage
are ensured under parameter uncertainties and load disturbances. of “plug-in” structure: a conventional feedback controller and a
Simulation and experimental results are provided to testify the plug-in feedforward repetitive controller. The design of a dig-
effectiveness of the proposed control scheme. ital repetitive control scheme is systematically developed with
Index Terms—PWM converters, repetitive control. complete stability and robustness analysis. A digital repetitive
controlled three-phase reversible PWM rectifier is illustrated in
details. Simulation and experimental results are provided to tes-
I. INTRODUCTION tify the effectiveness of the proposed control scheme.

O NE BASIC requirement in control systems is the capa-

bility to regulate controlled variables to their reference
inputs without steady-state errors against unknown and un-
II. DIGITAL REPETITIVE LEARNING CONTROL
A. Internal Model Principle
measurable disturbances. Control systems with this property
are called servo systems [1]. In servo system design, the Exact compensation of certain types of deterministic distur-
internal model principle proposed in [2] plays an important bance and exact tracking of particular reference signals are two
role. According to the internal model principle [2]–[4], zero important issues in the control system design [4]. A particular
error tracking of any reference input, in steady-state, can be disturbance and/or a reference signal of interest here are those
accomplished if a generator of the reference input is included in that can be described as
a stable closed-loop system. For example, a type 1 closed-loop
system with an integrator [or in discrete time (1)
domain], i.e., the generator of unit step function, in the loop
offers asymptotic tracking of a step input with zero steady state
where is the numerator polynomial; is the denom-
error. Repetitive control (RC) [1], [5], [6] is a special case of
inator which represents the reference and/or disturbance gener-
the internal model principle in control systems with periodic
ating polynomial.
signals. In discrete time domain, RC ensures zero tracking
For example, a constant signal can be described by the model
error at least at sampling points. Moreover, integral control is a with initial condition . And its corresponding
special case of repetitive control with unit period. Laplace transform is .
In fact, constant-voltage–constant-frequency (CVCF) power Fig. 1 shows a closed-loop control system, where is the
converters are servo systems with periodic sinusoidal and/or controller, is the plant with its input , is the
dc (step) reference commands. CVCF PWM power converters, disturbance, and is the tracking error between the refer-
such as dc-to-dc boost and buck converters, dc-to-ac inverters, ence and the output . From Fig. 1, the tracking error
and ac-to-dc rectifiers are widely employed in various power can be derived by
supplies. A good CVCF power supply should have the nominal
constant output in the presence of disturbances and uncertain-
ties, good dynamic response to disturbances, and remain stable (2)
under all operating conditions. Hence, repetitive control pro-
For steady state exact disturbance compensation and exact
Manuscript received January 12, 2002; revised September 28, 2002. Recom- reference tracking of the closed-loop control system in Fig. 1, a
mended by Associate Editor S. B. Leeb. sufficient condition is that the reference generating polynomial
The authors are with the School of Electrical and Electronic Engi- and the disturbance generating polynomial be included as part
neering, Nanyang Technological University, Singapore 639798 (e-mail:
[email protected]; [email protected]). of the denominator of the controller . This is known as the
Digital Object Identifier 10.1109/TPEL.2002.807150 Internal Model Principle (IMP) [4].
0885-8993/03$17.00 © 2003 IEEE
310 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 18, NO. 1, JANUARY 2003

Fig. 1. Closed-loop control system.

In discrete time cases, the Internal Model Principle can be Fig. 2. “Plug-in” repetitive control system.
used to achieve some very interesting results, such as discrete
repetitive control for system with periodic signals. A periodic C. Plug-In Repetitive Controller
signal of samples can be modeled by
1) Controller Design: Based on the Internal Model Prin-
ciple, in order to achieve zero tracking error, the internal model
(3)
for periodic signal is plugged into the control system as shown in
where Fig. 2 [6]. The plug-in part in Fig. 2 is called the “repetitive con-
; . Obviously, when and trol.” The transfer function of the plug-in repetitive controller
, is a step function generator. Thus, a dc signal is
can be regarded as a special case of periodic signals. Equation
(3) is the unified generator of both periodic signals and dc sig- (6)
nals. Using the Internal Model Principle, any period reference
signal can be exactly tracked (at least at the sample points) by where is the control gain; is a low-pass filter.
including in the denominator of the controller. This idea For periodic ac reference inputs, is usually chosen in
is the basis of the digital repetitive control. the following form [6]:

B. Problem Formulation (7)

Consider a repetitive control system shown in Fig. 2. The
transfer function for the control system without plug-in where with being the known number of pure
repetitive controller can be written as time step delays; is obtained from with
replaced by ; is a scalar chosen so that ;
(4) is the order of , and makes the filter realizable.
Equation (7) is a pole-zero cancellation or Zero-Phase compen-
where is the known number of pure time step delays; and sation [14] as a filter design for .
are the output and input of the plant , respectively; For a dc reference input, its period is 1 with and
is the disturbance with a known period of with . can be designed as (7). But in most cases,
, being the frequency of , and being for dc signal is simply chosen as follows .
the sampling frequency; , with , is the Therefore, the integral controller is also a special case of
tracking error; is a conventional feedback controller, e.g., plug-in repetitive controllers (6). For a dc reference input, PID
PD controller. controller is one special case of the plug-in control scheme
Without loss of generality, it is assumed that (4) defines an which includes a feedback PD controller and a plug-in integral
asymptotically stable system: i.e., the roots of controller.
are all inside the unit circle. This assumption implies that 2) Stability Robustness: In practice, model uncertainty ex-
the closed-loop control system is asymptotically stable. It is ists in plants. Let be the modeling error for unmodeled
also assumed that and are coprime (i.e., dynamics. It is assumed that there exists a constant such that
, , where denotes . The relation between the true system transfer func-
the largest integer ). This assumption is necessary for tion and the nominal system transfer function can
the asymptotic convergence of to zero. be written as
Factorize as
(5) (8)

where and are the cancelable and uncan- where all poles of are inside the unit circle.
celable parts of , respectively. Thus comprises From Fig. 2, the error transfer function for the overall system
roots on or outside the unit circle and undesirable roots which is
are inside the unit circle, and comprises roots of
which are not in [13].
The design objective is to find a robust zero error tracking
control law which assures the asymptotic stability
of overall system and the asymptotic zero error tracking:
. (9)
ZHOU AND WANG: DIGITAL REPETITIVE CONTROLLED THREE-PHASE PWM RECTIFIER 311

From (9), it can be concluded that the overall closed-loop

system is stable if the following two conditions hold:
1) the roots of are inside the unit circle;
2)

for all (10)

In view of (5)–(8), it can be derived that

(11)

Then, (10) yields

(12)

Obviously, if the overall closed-loop system shown in Fig. 2

is asymptotically stable and the angular frequency of the ref-
erence input and the disturbance approaches
, [ for even and
for odd ], then ,
, and thus

(13)

Equation (13) indicates that, if the frequency of the reference

input or disturbance is less than half of the sampling frequency, Fig. 3. Modified repetitive control system. (a) Modified repetitive controller.
zero steady-state tracking errors for both dc and ac reference (b) Stability bounds for modified repetitive controller.
inputs are ensured using the plug-in controller [10], even
in the presence of modeling uncertainty. A first order filter is generally suf-
ficient in most cases. The repetitive controller with
D. Modified Repetitive Controller is named the “modified” repetitive controller. However, since
For the repetitive control system, the larger the control gain high frequency periodic disturbances are not perfectly canceled
is, the faster the tracking error converges. If the stability by this controller, a trade-off is made between tracking preci-
region for can be expanded beyond , it will be sion and system robustness in this case [17].
more flexible to design system performance. In order to achieve
such a goal, as shown in Fig. 3(a), a low-pass filter E. Discussion
is added into as follows [13]:
In the proposed scheme, the plug-in repetitive controller
and the feedback controller are complementary. The conven-
(14) tional feedback controller offers fast response and robustness.
However, a feedback controller has no memory and will re-
where, with coefficients ( ; peat, in all subsequent cycles, any imperfection existing in the
) to be designed performance of the feedback controller. Whereas, the plug-in
repetitive controller is essentially an intelligent controller using
previous information stored in memory, and ensures steady-state
(15) zero error tracking by repetitive learning. Because the plug-in
controller adjusts its output once per period, it will take longer
time to force the tracking error converge gradually. The pro-
And correspondingly, (10) is modified as follows [15]: posed plug-in control scheme comprises of the advantages of
both feedback controllers and plug-in repetitive controllers:
the fast dynamic response of feedback controllers and the high
precision tracking ability of repetitive controllers.
for all (16) Since reference inputs for CVCF PWM converters are con-
stant dc signals or CVCF sinusoidal ac signals, the plug-in
Note that is a moving average filter that has zero digital repetitive plus a conventional feedback controller (e.g.,
phase shift. Fig. 3(b) shows that the right-hand side of (14) for PD controller) constructs the robust zero error tracking (RZET)
three choices of [16] and . The control scheme for CVCF PWM converters. Since PD controller
stability region is substantially larger when is used. is simple, popular and effective for most control systems, it is
312 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 18, NO. 1, JANUARY 2003

Fig. 5. Repetitive controlled three-phase reversible rectifier system.

and
(18)

where the neutral point M of dc bus is referred as the ground;

phase currents , , and dc bus voltage are the state vari-
ables; , , are the PWM modulated input voltages at port ,
, ; is the PWM modulated input current; , , are the
Fig. 4. PWM rectifier and waveform of switching function. (a) Three-phase
reversible PWM rectifier with nominal parameters. (b) PWM switching known three-phase sinusoidal voltages, and ;
waveform. , , and are the nominal values of the components;
is the of the load.
The output equation can be expressed as
usually adopted as the conventional feedback controller in our
proposed “plug-in” control scheme. And for the dc reference
input, our proposed control scheme is PID control. (19)
In the following section, the digital repetitive control scheme
is applied to a three-phase reversible PWM rectifier to testify
its validity and effectiveness. The PWM rectifier is a typical On the instantaneous basis, the PWM modulated control in-
application example for digital repetitive control, because it re- puts ( ) and can be written as
quires high-precision periodic ac current and constant dc voltage
tracking. (20)
(21)

III. REPETITIVE CONTROLLED THREE-PHASE REVERSIBLE

PWM RECTIFIER where are switching functions. For a two-level
PWM modulator, the switching functions are
The dynamics of the three-phase reversible PWM rectifier (as defined as: , when the switch is on and the switch
shown in Fig. 4) can be described as is off; , when the switch is on and the switch
is off. As shown in Fig. 4, each PWM switching waveform
at port is a pulse of magnitude “ 1” with width
being in the sampling interval and the active duty ratio
( )( ) being
.
The objective for the PWM rectifier system with a RC con-
troller is to achieve unit power factor, low current harmonics and
constant output dc voltage.

A. Controller Design
(17)
As shown in Fig. 5, a control scheme of double-loop structure
is proposed for the PWM rectifier: an inner ac current loop; and
an outer dc voltage loop.
ZHOU AND WANG: DIGITAL REPETITIVE CONTROLLED THREE-PHASE PWM RECTIFIER 313

1) Current-Loop Controller: The corresponding sampled-

data state equations for (17) can be expressed as

Fig. 6. Current-loop control scheme.

(22)

where is the sampling period;

( ); ; ; and are the Fig. 7. Voltage-loop control scheme.
nominal value of and , respectively.
It is clear that (22) can be treated as three independent phase
subsystems. The current-loop control scheme for each phase
subsystem is shown in Fig. 6. From (22), the sampled-data equa-
tion for each phase subsystem with nominal parameters is

(23)

Then the nominal transfer function for each phase sub-

system is

(24)

If the current controller is chosen for the plant (23) as

(25)

is obtained. Then the transfer function for

each phase current-loop control system without repetitive con-
troller is . Equation (25) is a deadbeat controller
which is referred as the predictive controller in [18], [19]. The
deadbeat controller offers fast response with only one sampling
period delay.
However, the deadbeat current controller is based on an
accurate nominal model of the reversible PWM rectifier. In
practice, there are uncertainties in the converter parameters,
such as , and .
Therefore, even with phase angles feedforward compensation,
zero tracking phase error can not be achieved. Based on the
Fig. 8. Simulation steady-state ac-side response under load 100
(Vertical:
repetitive control theory in Section II, a repetitive controller 10 V/div for E , 10 A/div for i ). (a) Deadbeat controlled E and i .
is proposed as follows for the reversible rectifier to overcome (b) Deadbeat plus repetitive controlled E and i .
the uncertainties and improve the current tracking
where ; is the frequency of ;
is the sampling frequency; .
(26)
In order to enhance the robustness, can chosen as
314 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 18, NO. 1, JANUARY 2003

Fig. 9. Simulated transient response under load change: (a) dc bus voltage
v (t) under load change from 50
to 100
(Vertical: 5 V/div) and (b) dc bus
voltage v (t) under load change from 100
to 50
(Vertical: 5 V/div).

. If , in sampled-data form,
the repetitive controller can be expressed as Fig. 10. Experimental ac-side current response under load 100
(Voltage:
, which is the same as the anticipatory 12 V/div; Current: 1 A/div). (a) Deadbeat controlled E and i . (b) Deadbeat
learning control law [20]. plus repetitive controlled E and i . (c) Transient current tracking error e(t) =
In practice, converter parameters are , i 0 i when the repetitive controller is plugged in.
. Therefore, the transfer function for each actual
subsystem is, with , The transfer function from to can be approximately
treated as
(27)
(29)
Therefore, as shown in Fig. 6, when a deadbeat controller (25)
is designed for the plant (27), the transfer function for the where . From (18), the transfer function from to
subsystem without repetitive controller is can be obtained as

(28) (30)

When and , a deadbeat response If the pole in (29) is far away from the origin compared to
is achieved. According to the stability analysis in Section II, that in (30), i.e., , the transfer function from
the overall system is stable if to can be simplified as
and .
2) Voltage-Loop Controller: From (21) and Fig. 5, we have
(31)
.
ZHOU AND WANG: DIGITAL REPETITIVE CONTROLLED THREE-PHASE PWM RECTIFIER 315

Because is a dc signal, a zero tracking controller—PI

controller is employed as

(34)

where gains and are designed to ensure a stable system

with a satisfactory dynamic response.
The voltage-loop control subsystem is shown in Fig. 7. It is
easy to analyze the stability of the closed voltage-loop control
system using Routh–Hurwitz Criterion.

IV. SIMULATION AND EXPERIMENT

A three-phase reversible PWM rectifier system with repeti-
tive learning controller, shown in Fig. 5, has been implemented
with an experiment setup.
For both simulation and experiment, system parameters are
selected as follows: mH; mH; ;
; F; V; three-phase si-
nusoidal voltages are 50 Hz 30 V (peak);
KHz; ; ,
, , .
With these parameter values, the pole of is obtained as
0.19 (inside the unit circle) and thus the system is stable. The
maximum value of in frequency domain is 0.9786.
According to the stability condition for
repetitive control design, the system with repetitive controller is
stable if . We choose .
Since , (31) can be
approximately written as .
From (32) and (33), the steady-state is equal to 0.56. And
the transfer function from to can be approximately
expressed as

(35)

Based on the Routh–Hurwitz criterion, the system of (35) is

stable if and . Equation (35) is a
second-order system. When the damping ratio of (35) is equal
to 0.4, we have and .
Fig. 11. Experimental response under sudden load change (Voltage: 12 V/div; Fig. 8 shows the -phase simulation steady-state current
Current: 2.5 A/div): (a) dc bus voltage v (t) and current i (t) under load change response of the reversible PWM rectifier with/without the
from 50
to 100
and (b) dc bus voltage v (t) and current i (t) under load repetitive current controller. As shown in Fig. 8(a), the zero
change from 100
to 50
.
crossing points of both phase voltage and phase current
show that there is a lag between and for the
Moreover, according to the principle of energy balance [11], deadbeat controlled rectifier. Fig. 8(b) shows that the repetitive
the steady-state can be approximately derived as controller force the phase displacement between and
approaches zero successfully. Therefore, the power factor
approaches unity. The response of other phase currents are
(32) similar to that of the -phase subsystem.
Fig. 9 shows the simulation dc bus voltage response with
where , and are the peak values of , sudden load changes between 50 and 100 . Both cases show
and , respectively. that returns to the reference value 80 V after about 120 ms.
Since , steady-state can calculated as With PI controller, can track the reference dc voltage with
near zero error.
Fig. 10 shows the -phase experimental current response of
(33)
the reversible PWM rectifier with/without the plug-in repetitive
316 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 18, NO. 1, JANUARY 2003

current controller. Obviously, Fig. 10(a) and (b) show that repet- [9] Y. Y. Tzou, S. L. Jung, and H. C. Yeh, “Adaptive repetitive control of
itive controller reduces the phase displacement between PWM inverters for very low THD AC-voltage regulation with unknown
loads,” IEEE Trans. Power Electron., vol. 14, pp. 973–981, Sept. 1999.
and to almost zero. And Fig. 10(c) shows that the peak of [10] K. Zhou and D. Wang, “Periodic errors elimination in CVCF PWM
current tracking error is reduced from 0.25 A to about 0.04 A DC/AC converter systems: A repetitive control approach,” Proc. Inst.
after 0.7 s. Therefore, the power factor approaches unity. The re- Elect. Eng., vol. 147, no. 6, pp. 694–700, 2000.
[11] S. L. Jung, H. S. Huang, and Y. Y. Tzou, “A three-phase PWM AC–DC
sponse of other phase currents are similar to that of the -phase converter with low switching frequency and high power factor using
subsystem. DSP-based repetitive control technique,” in Proc. IEEE PESC’98 Conf.,
Fig. 11 shows one experimental system response with sudden Fukuoka, Japan, May 1998, pp. 517–523.
[12] K. Zhou and D. Wang, “Digital repetitive learning controller for three-
load changes between 50 and 100 . In both cases, after about phase CVCF PWM inverter,” IEEE Trans. Ind. Electron., vol. 48, pp.
110 ms, returns to the reference value 80 V. With PI con- 820–830, Aug. 2001.
troller, can track the reference dc voltage with near zero [13] C. Cosner, G. Anwar, and M. Tomizuka, “Plug in repetitive control for
industrial robotic manipulators,” in Proc. IEEE Int. Conf. Robotics Au-
error. tomat., 1990, pp. 1970–1975.
[14] M. Tomizuka, “Zero phase error tracking algorithm for digital control,”
Trans. ASME: J. Dyn. Syst., Meas., Contr., vol. 109, no. 2, pp. 65–68,
V. CONCLUSION 1987.
[15] K. K. Chew and M. Tomizuka, “Digital control of repetitive errors in
In this paper, a digital repetitive control scheme is proposed
disk drive systems,” in Proc. Amer. Control Conf., 1989, pp. 540–548.
for CVCF PWM converters to achieve zero error tracking. As [16] K. Chew and M. Tomizuka, “Steady-state and stochastic performance of
an example, the proposed digital repetitive control scheme is a modified discrete-time prototype repetitive controller,” Trans. ASME:
applied to a three-phase PWM rectifier. Simulation and exper- J. Dyn. Syst., Meas., Contr., vol. 112, 1990.
[17] H. L. Broberg and R. G. Molyet, “Reduction of repetitive errors in
iment results show that the tracking errors in ac current loop tracking of periodic signals: Theory and application of repetitive
and dc voltage loop are quickly eliminated by the proposed control,” in Proc. 1st IEEE Conf. Contr. Applicat., Dayton, OH, Sept.
digital control scheme. The digital repetitive controlled PWM 1992, pp. 1116–1121.
[18] R. Wu, S. B. Dewan, and G. R. Slemon, “A PWM AC to DC converter
rectifier has a near unit power factor and constant dc output with fixed switching frequency,” IEEE Trans. Ind. Applicat., vol. 26, pp.
voltage under disturbances and parameter uncertainties while 880–885, Sept./Oct. 1990.
[19] , “Analysis of a PWM AC to DC voltage source converter under
maintaining good response characteristics. Simulation and ex- the predicted current control with a fixed switching frequency,” IEEE
perimental results verify the effectiveness of the proposed con- Trans. Ind. Applicat., vol. 27, pp. 756–764, July/Aug. 1991.
trol scheme. [20] D. Wang, “On D-type and P-type ILC designs and anticipatory ap-
proach,” Int. J. Contr., Special Issue Iterative Learning Contr., vol. 73,
The developments and results for a three-phase rectifier in no. 10, pp. 890–901, 2000.
this paper, a single-phase inverter in [10], and a three-phase
inverter in [12] show that the proposed control scheme is a
robust zero tracking error control scheme for CVCF PWM
converters. The proposed digital repetitive control scheme Keliang Zhou received the B.S. degree from the
Huazhong University of Science and Technology,
provides a simple and high-performance control solution for Wuhan, China, in 1992, the M.S.E degree from
CVCF PWM converters. Wuhan University of Transportation, Wuhan, China,
in 1995, and the Ph.D. degree from Nanyang
Technological University, Singapore, in 2002.
REFERENCES His research interests include power electronics
[1] S. Hara, Y. Yamamoto, T. Omata, and M. Nakano, “Repetitive control and electric machines drives, advanced control
theory, and applications. He has authored around
system: A new type servo system for periodical exogenous signals,”
IEEE Trans. Auto. Contr., vol. 33, pp. 659–667, July 1988. 20 published technical articles in the relevant areas
of repetitive learning controlled PWM converters,
[2] B. A. Francis and W. M. Wonham, “The internal model principle for
linear multivariable regulators,” Applicat. Math, vol. 2, pp. 170–194, PWM modulation strategy, and electric motor drives.
1975.
[3] , “The internal model principle of control theory,” Automatica, vol.
12, pp. 457–465, 1976.
[4] G. C. Goodwin, S. F. Graebe, and M. E. Salgado, Control System De- Danwei Wang (M’90) received the B.E. degree
sign. Englewood Cliffs, NJ: Prentice-Hall, 2001. from the South China University of Technology,
[5] T. Inoue, “High accuracy control of servomechanism for repeated con- Guangzhou, China, in 1982, and the M.S.E. and
touring,” in Proc. 10th Annu. Symp. Increamental Motion Contr. Syst. Ph.D. degrees from the University of Michigan, Ann
Devices, 1981, pp. 258–292. Arbor, in 1985 and 1989, respectively.
[6] M. Tomizuka, T. Tsao, and K. Chew, “Analysis and synthesis of dis- Since 1989, he has been with the School of Elec-
crete-time repetitive controllers,” Trans. ASME: J. Dyn. Syst., Meas., trical and Electronic Engineering, Nanyang Techno-
Contr., vol. 110, pp. 271–280, 1988. logical University, Singapore, where he is currently
[7] T. Haneyoshi, A. Kawamura, and R. G. Hoft, “Waveform compensation an Associate Professor and Deputy Director of the
of PWM inverter with cyclic fluctuating loads,” in Proc. IEEE Power Robotics Research Center. His research interests in-
Electron. Spec. Conf., 1987, pp. 745–751. clude robotics, control theory and applications. He
[8] Y. Y. Tzou, R. S. Ou, S. L. Jung, and M. Y. Chang, “High-performance has authored more than 100 published technical articles in the areas of itera-
programmable ac power source with low harmonic distortion using tive learning control and applications, manipulator/mobile robot dynamics, path
DSP-based repetitive control technique,” IEEE Trans. Power Electron., planning, robust control, adaptive control of such systems, and position/force
vol. 12, pp. 715–725, July 1997. control systems.