High Performance Control of a Single-Phase Shunt

Active Filter
R. Costa-Castelló, R. Griñó, R. Cardoner, E. Fossas

Abstract— Shunt active power filters are devices connected in Several approaches have been proposed for the outer control
parallel with nonlinear and reactive loads which are in charge of loop: selecting the desired switching pattern by optimization,
compensating these characteristics in order to assure the quality the optimization can be based on genetic algorithms [18], [19],
of the distribution network. This work analyzes the dynamics
of boost-converter used as an active filter and proposes a neural networks [14] or Fourier series analysis [13]; using
control system which guarantees closed-loop performance (power Lyapunov Functions [20]; or using a PI controller to determine
factor close to 1 and current harmonics compensation). Proposed the amplitude of the network sinusoidal current. Using a PI
controller is hierarchically decomposed in two control loops, one controller is the most common approach by far but, since the
in charge of shaping the current and the other in charge of plant is nonlinear, this PI controller is usually experimentally
assuring the power balance. Differently from other works both
control loops are analytically tuned. tuned.
The work describes both the analytical development and the This work presents a new controller for a single-phase shunt
experimental results showing the good performance of the closed- active filter that uses the traditional two control loops decom-
loop system. position. In our work the current controller is composed by
Index Terms— Active power filters, digital repetitive control a feedforward action in charge of assuring very fast transient
response and a feedback control law in charge of assuring
closed-loop stability and very good harmonic performance.
I. I NTRODUCTION The feedback control law is based on the use of a repetitive
Active filters are devices which allow to coexist nonlinear odd-harmonic controller [21]. The outer control law is based
loads and good energy quality in distribution networks. A on the exact computation of the sinusoidal current network
principal effort in the design and control of these devices amplitude; in order to improve robustness this computation is
has been developed in the past years. One research line combined with a feedback control law with a PI controller.
deals with topologies and architectures [1], [2], several types One of the contributions of this paper is the analytical tuning
of topologies have been proposed including parallel (shunt of the PI, which is unusual. The complete system results in
active filters), serial connections and hybrid serial-parallel a simple control law which offers very good results, both in
connections. Besides the architecture, the behaviour principle transient and steady-state behavior.
has also been a research topic: passive, active (using switch-
ing converters) and mixed passive-active devices have been
proposed. Additionally, converter based active filters may be II. P ROBLEM FORMULATION

based on a voltage or a current bus [3]. Most used active filters A. Physical model of the boost converter
are connected in parallel and correspond to active components
based on a voltage bus, although a passive serial connected
filter is usually added to compensate switching noise.
Another important research line related with active filters is in il
their control, many approaches have been proposed [4], [5],
+
[6], [7], [8], [9], [10]. Most of the proposed control schemes vn if Generalload
are based on two hierarchical control loops, an inner one in −
charge of assuring the desired current and an outer one in
charge of determining the required sinusoidal shape as well as
the appropriate power balance and converter operation point.
The current control loop needs to be fast and precise in order
to assure the desired energy flow quality. Several approaches u + C1
v1 rC,1
have been proposed: hysteresis based control [11], [12], [13], −
rL L
deadbeat controllers [14], Park transformation combined with
linear controllers [15] and repetitive control [16], [17] are the + C2
most relevant options. v2 rC,2
u −
This work was supported in part by the Ministerio de Educación y Ciencia
(MEC) under project DPI2004-06871-C02-02.
The authors are with the Instituto de Organización y Control de Sistemas
Industriales, Universitat Politècnica de Catalunya, Barcelona, Spain. E-
mail:{ramon.costa,roberto.grino,rafel.cardoner,enric.fossas}@upc.edu Fig. 1. Single-phase shunt filter connected to the network-load system.

1-4244-0755-9/07/$20.00 '2007 IEEE 3350

 Fig. 1 presents the system architecture. A load is connected D. Rewriting the Equations
to the power source, in parallel an active filter is applied in
order to fulfill the required performance, i.e. to guarantee unity
power factor in the network side. A boost converter with the
il
ac neutral wire connected directly to the midpoint of the dc sin (ωn t)
d if + +
bus is used as active filter. The averaged (at the switching + current
α
Boostconverter in
control
frequency) model of the boost converter is given by −
v2 v1
dif
L = −rL if − v1 d − v2 (d − 1) + vn (1)
dt Variablechange

dv1 v1
C1 = − + if d (2)
dt rC,1
dv2 v2
C2 = − + if (d − 1) (3) Id voltage ET
dt rC,2 control −
+ ETd
where d is the duty ratio, if is the inductor current and
v1 ,√v2 are the DC capacitor voltages, respectively; vn =
Vn 2 sin(ωn t) is the voltage source that represent the ac-line
Fig. 2. Block diagram of the controller showing the current control loop
source; L is the converter inductor, rL is the inductor parasitic (inner) and the voltage (or energy) control loop (outer).
resistances, C1 , C2 are the converter capacitors and rC,1 , rC,2
are the parasitic resistances of the capacitors. The control It is standard for this kind of systems to linearize the
variable, d, takes its values in the set [0, 1] and represents the current dynamics by the partial state feedback α = v1 · d +
averaged value of the PWM (pulse-width modulated) control v2 · (d − 1). Moreover, the change of variables if = if ,
signal injected to the actual plant. EC = 21 C1 v12 + C2 v22 , D = C1 v1 −C2 v2 makes appear two
more meaningful variables. Namely, EC , the energy stored in
B. Load Description the converter capacitors and D, the charge unbalance between
Due to the nature of the voltage source, the load current, them. Assuming that the two dc bus capacitors are equal
in steady-state, is usually a periodic signal with only odd- (C = C1 = C2 , rC = rC,1 = rC,2 ) the system dynamics
harmonics in its Fourier series expansion, so the current can in the new variables results in:
be written as: dif
∞
L = −rL if + vn − α (5)
 dt
il = an sin(ωn (2 · n + 1) t) + bn cos(ωn (2 · n + 1) t) dEc 2Ec
= − + if · α (6)
n=0 dt rC · C
(4) dD 1
= − D + if (7)
where an , bn ∈ R are the real Fourier series coefficients of the dt rC · C
load current. Hence a zero dc component of il is presumed. It is important to note that the state feedback together with the
change of variables results in a state and input diffeomorphism
C. Control objectives so the linearization is complete and formally correct [22]. In
The active filter goal is to assure that the load is seen as a addition, eq. (7) is linear too.
resistive one. This goal can be stated 1 as in ∗ = Id∗ sin(ωn t), This new system (5)-(7) needs a controller to fulfill the
i.e. the source current must have a sinusoidal shape in phase desired performance. This controller will be defined using a
with the network voltage. Another collateral goal, necessary two step approach, first of all a current controller which forces
for a correct operation of the converter, is to assure constant the sine wave shape, afterwards the sine wave amplitude will
average value of the dc bus voltage 2 , i.e. < v1 + v2√>∗0 = be defined by an outer control loop to fulfill the appropriate
2vd , where vd must fulfill the boost condition (vd > 2vs ). active power balance for the whole system. This balance is
It would also be desirable that this voltage would be almost achieved if the energy 3 stored in the active filter capacitors,
equally distributed among both capacitors (v1 ≈ v2 ). Ec , is equal to a reference value, Ecd .
These two objectives define a non-standard control problem: The full control scheme for the system is depicted in
the second one is a regulation objective for the mean value of Fig. 2. The specific controller designs will be presented in
v1 + v2 , while the first one is not a tracking specification sections III-A and III-B.
because only a shape and not a signal is defined, that is
Id∗ is not known a priori and it must take the appropriate III. C ONTROL D ESIGN
value to maintain the power balance of the whole system.
A. Current Loop
This special form of the problem specifications implies the
particular structure of the controller loops described in the Taking benefit from the fact that current equation (5) is
next section. linear, a linear controller is designed to force a sinusoidal
1 x∗ represents the steady-state value of signal x(t). 3 A similar reasoning can be done with the dc bus capacitor voltages as it
2< x >0 means the dc value, or mean value, of signal x(t) is implied by the change of variables used.

3351
shape in the network current. This controller will be designed Hence, the total energy stored in the converter (ET ) should
in two parts: not suffer variations within a period, i.e.
• A feedforward controller which fixes the desired steady
 t
state : ĖT = 0 (13)
t−Tp
in = Id sin (ωn t) (8)
The stored energy in the converter can be decomposed in
The control action related with this action is computed by the energy stored in the inductors (EL = 12 L(if )2 ), and the
inverting the plant dynamics in steady state and forcing energy stored in the capacitors (Ec = 12 C1 V12 + 12 C2 V22 ).
the output to be the desired one. Additionally, it is important to note that some energy is lost
• A feedback controller which compensates uncertainties in the parasitic resistors of the inductors, the capacitors and
and assures closed-loop stability. This feedback controller the switches.
is designed applying the odd-harmonic repetitive control Noting that if ≈ Id sin (ωn t) − il is an odd-harmonic
technique [23], [21]. signal and without taking into account the parasitic resistance
This control technique allows to obtain perfect steady- of the inductors and capacitors, it can be easily proven that
state tracking/rejection of a certain periodic signal and independently of the value of Id and the load currents the
all its odd-harmonics. variation of energy in the inductors on one period is zero.
The complete control action is obtained by adding both Thus,  
t t
control actions. Under this control action the output is the ĖT = Ėc (14)
desired one also in the case of uncertainties and disturbances. t−Tp t−Tp

In case rL ≈ 0 the next relationship can be stated

 t √
B. Energy Shaping (Voltage Loop) 2Vs π (Id − a0 )
ĖT = ET (t) − ET (t − Tp ) ≈ .
t−Tp ω
il Active component
extraction
This energy balance can be seen as a linear discrete-time
Idf f system (with sampling time Tp ) with an input Id and a constant
+
ETd + Idf b + Id 1 ET disturbance a0 . So, applying the z-transform
PI
T
E 2Cp z−1 √
− 1 2Vs π
ET (z) = (Id (z) − a0 · Us (z)) (15)
z − 1 ωn
where Us (z) is the z-transform of the step signal. The value of
a0 corresponds to the active component of il (the load current).
Fig. 3. Simplified 50Hz model with PI Controller
So, in order to assure the desired energy balance (ET ≈ 0),
a closed-loop system is proposed. The control action will be
√ the source voltage is assumed to be vn =
As composed of two main parts:
Vn 2 sin (ωn t) the desired network current is Id sin (ωn t), ff
• A feedforward term : Id = a0 .
with Id locally constant. As a consequence the desired power
• A feedback term which is in charge of compensating
flow seen from the net is:
the dissipative terms effects and the uncertainties in the
√
pn (t)  vn (t) · in (t) ≈ Id 2Vn sin2 (ωn t) (9) system. Thus, a classical PI controller will regulate ET
to the desired value ETd without steady-state error, i.e.
Additionally, the active filter goal is not consuming power so  
fb z+1
ideally the following relationship is desired: Id (z) = kp + ki (ETd (z) − ET (z)) (16)
√ z−1
pn (t) = Id 2Vn sin2 (ωn t) ≈ vn il  pl (t) + pf (t) (10) Fig. 3 shows the complete closed-loop scheme.
where pl (t) and pf (t) are the instantaneous power of the load
and the filter, respectively. However, as Id is designed locally IV. E XPERIMENTAL SETUP AND IMPLEMENTATION ISSUES
constant the power requirements can not be fulfilled instan- The experimental setup used to test the designed controller
taneously. What can be achieved is an energy compensation has the following parts:
within one period • Active filter: half-bridge boost converter (split-capacitor
 t  t dc bus) with IGBT switches (nominal current 100 A)
pn ≈ pl + pf , (11) and the following parameters: r = 0.3 Ω, L = 0.8 mH,
t−Tp t−Tp C1 = C2 = 9900 μF and rC = 8200 Ω. The switching
that yields the Id ideal value. From the power flow point of frequency of the converter is 20 kHz and a synchronous
view the active filter redistributes the power flow within one (regular) centered-pulse single-update mode pulse-width
period in order to assure the stated power balance: modulation strategy is used to map the controller’s output
 t to the IGBT gate signals (see Figure 4 ).
• Rectifier (non-linear load): Full-wave diode rectifier with
pf ≈ 0 (12)
t−Tp a filter capacitor C = 4500 μF. The active power with

3352
 nominal dc resistor is P = 4.56 kW and its reactive
power is approximately zero. Fig. 5 shows the shape of
the ac mains voltage and current and Fig. 6 the harmonic
content of the voltage and the current for the rectifier
with the nominal dc resistor. It is worth to remark that the
total harmonic distortion 4 (THD) of this current is about
63.9% and its maximum derivative is about 70kA/s.
• Analog circuitry of feedback channels: the ac mains
voltage, the ac mains current and the dc bus voltages are
sensed with a voltage transformer, a hall-effect sensor
and two isolation amplifiers, respectively. All the signals
from the sensors pass through the corresponding gain
conditioning stages to adapt their values to A/D converter
input taking advantage of their full dynamic range. In
Fig. 4. Power Converter Picture addition, all the feedback channels include a first order
low-pass filter with unity dc gain and 4.3 kHz cutoff
frequency 5 .
• Control hardware and DSP implementation: the control
vn board has been internally developed and is based on
an ADSP-21161 floating-point DSP processor with an
il ADSP-21990 fixed-point mixed-signal DSP processor
that acts as coprocessor, both from Analog Devices. The
ADSP-21161 and the ADSP-21990 communicate each
other using a high-speed synchronous serial channel in
DMA mode. The ADSP-21990 deals with the PWM
generation and the A/D conversions with its integrated
14 bits eight high-speed A/D channels (Figure 7).
The controller has been implemented running at the IGBT
switching frequency. Technologically this swithching fre-
quency is limited to 20 kHz so this frequency has been
Fig. 5. Nonlinear load: voltage and load current (92 V/div, 19.2 A/div). selected as the sampling one.
• The nominal voltage of the ac mains is Vn = 230 V RMS
and its nominal frequency is 50 Hz.

V. E XPERIMENTAL RESULTS
This section shows some of the experimental results ob-
tained for the active filter operation with the designed control
system. The results are presented by means of oscilloscope
and power analyzer screen dumps of the ac mains electrical
variables and the active filter semi-bus dc voltages when
Fig. 6. Nonlinear load: voltage and load current (RMS and THD).
necessary.
Apart from the selected experiments collected in this sec-
tion, a lot of numerical simulations, including mainly capac-
itive or inductive loads, have been carried out showing the
same good performance as it will be shown below. Also, it
is worth noting that several numerical simulations including
loads that work as generators at some time periods 6 (thus
imposing a negative active power flow to the source) have
been carried out without problems. The voltage loop of the
overall controller assures the active power balance and, after
a transient, in steady state the input to the AM modulator
4 In this work the THD figures and the harmonic content are always taken
with respect to the fundamental harmonic (50 Hz) and they have been obtained
using a Power Quality Analyzer Fluke 43 instrument.
5 The oscilloscope screens in the figures of this section and the following
show the voltages and currents after the corresponding analog low-pass filters.
6 This problem was established as a hard one by Depenbrock and
Fig. 7. Control Setup Picture Staudt [24].

3353
is negative giving a current reference shifted π rad from the
network voltage that the current loop tracks without difficulty. vn v1 , v2

A. Active filter operation with no load

vn in
v1 , v2

in

Fig. 10. Active filter with the nonlinear load: voltage, current and semi-bus
dc voltages (92 V/div, 19.2 A/div and 74.5 V/div, respectively).

Fig. 8. Active filter with the no load: voltage, current and semi-bus dc
voltages (92 V/div, 19.2 A/div and 74.5 V/div, respectively).

Fig. 11. Active filter with the nonlinear Load : RMS and THD %f values
of mains current in and cos φ, PF.

order harmonics. Fig. 10 shows the current that appears with

a good sinusoidal shape and in phase with the grid voltage.
This figure also shows the values of each semi-bus of the
Fig. 9. Active filter with no load: RMS and THD values of current in (t) active filter dc bus. As it can be seen in Fig. 11 the THD of
and P , Q, cos φ and PF. the current is very low (0.6 %) and the power factor is 1.

This subsection presents some results of the no load op- C. Active filter transient response
eration of the active filter. Fig. 8 shows the network voltage
and current and the semi-bus dc voltages. The RMS value of This section presents the results for the following experi-
the current is about 0.68 A and its THD value is 6.8%. Then, ments:
the resulting active power consumed by the filter to cover its 1) the full nonlinear load is applied to the network with the
losses without compensating any load is about 0.15 kW. It is active filter in operation (Fig. 12);
worth to note that the fundamental component of the current 2) the full nonlinear load is disconnected from the ac mains
is in phase with the voltage (cos φ = 1), see Fig. 9. So, almost with the active filter in operation (Fig. 13).
no reactive power is consumed by the filter. The low power In each case, the overshoot in the dc bus voltage is almost im-
factor (PF) is due to the high value of the switching ripple perceptible. Therefore, there is no problem with the maximum
with respect to the fundamental component of the current. load variations expected in the system.

B. Active filter operation with the nonlinear load VI. C ONCLUSIONS

In this experiment the diode rectifier previously described The paper shows the design and implementation of a
is connected to the network. This nonlinear load has not controller for a current active filter. The controller consists
reactive power at the fundamental frequency, however the of a current control loop and an outer dc bus voltage control
active filter must work to compensate all the generated higher loop. The current references for the inner control loop is

3354
 [3] M. Salo and H. Tuusa, “A new control system with a control delay com-
pensation for a current-source active power filter,” IEEE Transactions on
v1 , v2 Industrial Electronics, vol. 52, no. 6, pp. 1616–1624, December 2005.
[4] J. Wu and H. Jou, “Simplified control method for the single-phase active
power filter,” IEE Proc. Electrical Power Applications, vol. 143, no. 3,
pp. 219–224, may 1996.
[5] S. Choi, “A three-phase unity-power-factor diode rectifier with active
input current shaping,” IEEE Transactions on Industrial Electronics,
vol. 52, no. 6, pp. 1711–1714, December 2005.
[6] S. Buso, L. Malesani, and P. Mattavelli, “Comparison of current control
techniques for active filters applications,” IEEE Trans. on Industrial
Electronics, vol. 45, pp. 722–729, october 1998.
[7] P. Mattavelli, “A closed-loop selective harmonic compensation for active
filters,” IEEE Trans. on Industry Applications, vol. 37, no. 1, pp. 81–89,
january 2001.
in [8] B. Singh, K. Al-Haddad, and A. Chandra, “A new control approach
to Three-Phase active filter for harmonics and reactive power compen-
sation,” IEEE Trans. on Power Systems, vol. 13, no. 1, pp. 133–137,
february 1998.
[9] S. Rahmani, K. Al-Haddad, and F. Fnaiech, “Reduced switch number
single-phase shunt active power filter using an indirect current control
Fig. 12. Source current (in ) and semi-bus dc voltages (v1 ,v2 ): from no-load technique,” in Proc. of the 2003 IEEE Int. Conf. on Industrial Technol-
to full nonlinear load (19.2 A/div and 74.5 V/div, respectively). ogy, vol. 2, december 2003, pp. 1107–1112.
[10] S. Rahmani, K. Al-Haddad, H. Kanaan, and B. Singh, “Implementation
and simulation of modified pwm with two current control techniques
applied to single-phase shunt hybrid power filter,” IEE Proc. on Electric
Power Applications, vol. 153, no. 3, pp. 317–326, may 2006.
v1 , v2 [11] J.-G. Hwang, Y.-J. Park, and G.-H. Choi, “Indirect current control of
active filter for harmonic elimination with novel observer-based noise
reduction scheme,” Electrical Engineering, no. 87, pp. 261–266, 2005.
[12] H. Kömürcügil and Ö. Kükrer, “A robust current control strategy for
in single-phase shunt active power filters,” in Proc. of the 29th Annual
Conference of the IEEE Industrial Electronics Society (IECON’03),
vol. 3, november 2003, pp. 2277– 2281.
[13] L. Zhou and Z. Li, “A novel active power filter based on least compensa-
tion current control method,” IEEE Transactions on Power Electronics,
vol. 15, no. 4, pp. 655–659, July 2000.
[14] K. Nishida, M. Rukonuzzman, and M. Nakaoka, “Advanced current
control implementation with robust deadbeat algorithm for shunt single-
phase voltage-source type active power filter,” IEE Proceedings - Elec-
tronics Power Applications, vol. 151, no. 3, pp. 283–288, May 2004.
[15] T. Al Chaer, J. Gaubert, L. Rambault, and C. Dewez, “Linear model
and h∞ control of shunt active power filter,” International Review of
Electrical Engineering, pp. 57–65, february 2006.
[16] R. Costa-Castelló, R. Griñó, and E. Fossas, “Odd-harmonic digital repet-
itive control of a single-phase current active filter,” IEEE Transactions
on Power Electronics, vol. 19, no. 4, pp. 1060– 1068, July 2004.
Fig. 13. Source current (in ) and semi-bus dc voltages (v1 ,v2 ): from full [17] S. Hamasaki, K. Fujii, M. Tsuji, and S. Chen, “A novel method of active
nonlinear load to no-load (19.2 A/div and 74.5 V/div, respectively). filter control using repetitive control,” in Proc. of the Eighth Int. Conf.
on Electrical Machines and Systems (ICEMS 2005), vol. 2. IEEE,
september 2005, pp. 1252– 1256.
[18] M. Welsh, P. Mehta, and M. Darwish, “Genetic algorithm and extended
created passing the output of the voltage controller through an analysis optimisation techniques for switched capacitor active filters -
AM modulator that uses as a carrier a filtered version of the compartive study,” IEE Proceedings - Electronics Power Applications,
network voltage. Both, the current and the outer controllers vol. 147, no. 1, pp. 21–26, January 2000.
[19] M. E.-H. M. Darwish, “A new control technique for active power
are based on the combination effect of a feedforward and filters using a combined genetic algorithm/conventional analysis,” IEEE
a feedback control law. This combination allows to obtain Transactions on Industrial Electronics, vol. 49, no. 1, pp. 58–66,
almost perfect response both in transient and steady-state February 2002.
[20] H. Kömürcügil and Ö. Kükrer, “A new control strategy for single-phase
operation. shunt active power filters using a lyapunov function,” IEEE Transactions
As a conclusion the proposed control scheme constitutes a on Industrial Electronics, vol. 53, no. 1, pp. 305–312, February 2006.
[21] R. Griñó and R. Costa-Castelló, “Digital repetitive plug-in controller
step forward in the active filter control. Both the transient and for odd-harmonic periodic references and disturbances,” Automatica,
steady-state behaviour are very good in the network current vol. 41, no. 1, pp. 153–157, january 2005.
shape and the active filter semi-bus dc voltages. [22] R. Marino and P. Tomei, Nonlinear Control Design: Geometric, Adaptive
and Robust, ser. Prentice Hall Information & System Sciences Series.
Prentice Hall, April 1995.
[23] T. Inoue, M. Nakano, T. Kubo, S. Matsumoto, and H. Baba, “High
R EFERENCES accuracy control of a proton synchroton magnet power supply,” in
Proceedings of the 8th World Congress of IFAC, 1981, pp. 216–220.
[1] H. Akagi, “New trends in active filters for power conditioning,” IEEE [24] M. Depenbrock and V. Staudt, “Stability problems if Three-Phase
Trans. on Industry Applications, vol. 32, no. 6, pp. 1312–1322, november systems with bidirectional energy flow are compensated using the FBD-
1996. Method,” in Proceedings of 8th International Conference on Harmonics
[2] M. El-Habrouk, M. Darwish, and P. Mehta, “Active power filters : A and Quality of Power (ICHQP’98), IEEE/PES-NTUA. Athens (Greece):
review,” IEE Proceedings- Electrical Power Applications, vol. 5, no. IEEE, october 14-16 1998, pp. 325–330.
147, pp. 403–413, September 2000.

3355