Reducing Passive Filter Sizes with Tuned Traps for

Distribution Level Power Electronics

Jeffrey M. Bloemink, Timothy C. Green
Imperial College London
Exhibition Road
London, United Kingdom
Phone: +44 (20) 7594-6287
Fax: +44 (20) 7594-6282
Email: j.bloemink@imperial.ac.uk
URL: http://www.imperial.ac.uk/controlandpower

Acknowledgments
This work is supported by the FLEXnet project, part of the Engineering and Physical Sciences Research
Council (EPSRC) Supergen Consortium.

Keywords
<<Passive filter >>, <<Distribution FACTS (DFACTS)>>, <<Passive component>>, <<Voltage
Source Converter (VSC)>>

Abstract
This paper proposes a filter topology intended for use with medium voltage grid-interfaced power elec-
tronics as an alternative to the widely-utilized LCL filter topology. The proposed topology utilizes mul-
tiple shunt RLC trap filters with parameters optimized to minimize the overall size of the filter while
meeting performance requirements for connection to the distribution grid. Results indicate a potential
for reduction of twenty-five to thirty percent in total filter volume for the scenario studied. In addition,
total power losses taking place within the filter are slightly reduced.

Introduction
The implementation of power electronics at the distribution level for utility applications or to interface
distributed generation brings about a new set of challenges for the designer. If installing in urban, or
suburban areas where geographical space can be at a premium, multi-megawatt three-phase voltage-
sourced converters (VSCs) and their associated passive components will have limits on their physical
volume.
The three primary components which comprise the bulk of the total converter volume are the valves,
the passive components (filters) and the cooling system [1]. This paper addresses the issue of filter
component size and proposes an alternative topology for consideration.
For grid connection applications, the filter must be designed in order to meet regional grid connection
requirements [2] while accounting for device limitations, such as the converter DC bus voltage, overall
rating, and passive filter losses. Despite their simplicity, purely inductive filters are not utilized for these
applications due to the high level of inductance required in order to meet these grid coupling require-
ments. The overall size will be large and the reactive power output capabilities for a given maximum
converter output voltage (restricted by the DC bus voltage) are limited by this large inductance [3]. A
widely used topology for high power, medium voltage, motor drives is the LCL filter [1], which offers
similar performance with a much lower size when compared with a purely inductive filter [3]. The trade-
offs for this reduction in size are an increased component-count as well the requirement for damping
the resonance inherent in this topology. Both active and passive options for damping can be utilised
[4]. In order to actively damp via control of the inverter, it is necessary to keep the significant resonant
frequencies of the plant within the controller bandwidth [5] which puts further restrictions on component
  
  


 
 
  
  


 


  


 
 


 

 

 




     


(a) LCL filter with trap tuned to resonant fre- (b) Proposed filter topology with n shunt RLC traps
quency (benchmark)
Figure 1: Filter topologies under study

selection. In addition, some passive damping is required for times in which the VSC is disabled, as there
is potential for excitation of this resonance from grid-side events such as faults or tap changes at the
primary substation.
The LCL design that will serve as the benchmark is shown in Figure 1a. This topology includes an
additional RLC branch tuned in order to reduce the peak gain at the resonant frequency of the LCL
circuit, thereby providing the necessary passive damping.
The proposed alternative topology, shown in Fig. 1b, makes use of multiple (n) RLC shunt trap filters
between the ripple inductor, Li, and grid-coupling inductor, Lo in order to increase the attenuation of
high frequency components sourced by the sinusoidal pulse-width modulated waveform (SPWM) from
the VSC. The use of multiple traps allows the shunt capacitor to be removed entirely while still meeting
performance requirements, and offers a reduction in peak stored energy, and hence size, for all filter
components relative to the benchmark case (LCL-type).

Component Selection and Size Estimation

Optimal selection of filter components is achieved using a custom genetic algorithm with the fitness
(cost function), of a given individual (feasible point in solution space) evaluated according to the size
estimation method to be described. It was observed in preliminary studies that the optimization routine
converged to a global minimum for a given number of traps when the traps are tuned to multiples of the
switching frequency, as opposed to being tuned to arbitrary frequencies across the spectrum. Results for
a two-trap and three-trap filter are presented in this paper as they demonstrate the desired properties and
offer a more significant potential size reduction than that of a single trap filter. The resulting optimal size
was observed to increase after n=3, thus filters with greater than three traps are not considered.
Precise size estimation is challenging due to variations in manufacturer processes and other design pa-
rameters, as well as taking into account how the cooling system scales with losses and volume. To
generate an initial estimate however, the peak energy stored in a reactive component can be considered
proportional to the size of the component. Some techniques were considered for the estimation of induc-
tor and capacitor volume based on peak energy requirements [6]; however it was deemed appropriate to
use manufacturer data for passive components of similar rating and for similar applications in order to
come up with mean constants of volume versus energy. While final realizations will likely not follow
this linear scaling, it was critical to the component selection (during the optimization procedure) that
the relative scaling of inductor volume with peak energy versus capacitor volume with peak energy is
realistic. If the fitness was solely based on peak reactive energy levels, results would vary from those
3 3
presented. The scaling factors utilized, are 1.11 mkJ for high power inductive elements and 0.54 mkJ for
capacitive elements, a ratio of approximately 2 to 1. By taking into account this difference in energy/size
proportionality when performing the optimization, a better component selection, with respect to filter
volume, is achieved.
To perform an initial size estimate with a low computational cost, the system is characterized in the
frequency domain for the spectrum of interest; that is, at all significant harmonics of the SPWM VSC
terminal voltage waveform (an example waveform is shown in Figure 8), including all side-band compo-
nents. This allows the significant peak currents and voltages in all reactive components to be calculated
in order to give an estimate of peak energy, utilizing the relationship Ec = 12 CV 2 and EL = 21 LI 2 . During
the optimization procedure, a single VSC operating point is used to calculate the peak energy in all re-
active components also to reduce computation time; specifically, the modulation index and phase angle
reference for the SPWM unit is set such that the VSC delivers its rated apparent power (12 MVA) with a
power factor of 0.85.

Filter Volume for Peak Energy Requirements Worst Case Passive Component Losses
80
2.6 % 3.4 %
20 70

30.1 % 60
36.2 %
15
50

Losses (kW)
Volume (m3)

40
10
30

20
5

10

0 0
LCL (opt) n=2 (opt) n=3 (opt) LCL (opt) n=2 (opt) n=3 (opt)

Figure 2: Size based on peak energy, passive component losses

After the optimization procedure converges, a more accurate estimate of peak inductor currents and peak
capacitor voltages is obtained by considering all feasible VSC operating points, i.e. all inverter terminal
voltage phase angles which do not exceed the specified rated inverter output current or maximum inverter
fundamental output voltage for a given modulation index, ma. In addition, the peak values are obtained in
the time-domain based on a real SPWM voltage waveform. Out of these operating points, the largest peak
current/voltage for each reactive component of the filter is selected and this figure is used for attaining a
size estimate.

Design Constraints
In the benchmark LCL filter of Fig. 1a, Li, Lo and C are selected in order to meet the grid connection
requirements specified by the IEEE 1547 grid interconnection standard [2] while minimizing size and
losses. The single RLC trap is tuned to the resonant frequency of the LCL set.
The proposed multiple-trap topology has additional parameters available for tuning including the res-
onant frequency of the trap, the damping resistance (Rtn), and capacitance (Ctn) . These parameters
were selected to meet grid-interconnection requirements while meeting design constraints. Violation of
any constraints is interpreted as a large penalty to the fitness of a given individual during the genetic
algorithm procedure. The equivalent series resistance (ESR) of the capacitor components is considered
negligible and an X/R ratio of 50 for all inductive components is assumed. A DC bus voltage of 20 kV
is used, which is adequate for grid-coupling utilizing SPWM at the medium voltage (11 kV) level. A
2-level VSC with a switching frequency of 1950 Hz is considered. Note that if a multi-level converter
topology is considered as opposed to a 2-level, it is possible to reduce the size of the filter further due to
the reduced high frequency harmonic content and reduced requirement for harmonic attenuation.
Constraints involved in parameter selection during the optimization procedure are as follows:
• the maximum reactive power that can be sourced for grid support (for a fixed DC bus voltage)
- limits have been placed such that for a given DC bus voltage, the converter can source reac-
tive power with a power factor of 0.75. The maximum supplied reactive power is limited by a
combination of the DC bus voltage as well as the total series inductance in line with the point of
grid coupling, thus Li and Lo should be selected to allow for a reasonable level of reactive power
support in this application
• the total harmonic distortion (THD) for high order harmonics, calculated based on a 2-level VSC
switching at 1.95 kHz utilizing sinusoidal pulse-width modulation. THD is calculated based on
dominant harmonics associated with this PWM waveform during the optimization procedure to
reduce computational time
• the peak gain at any filter resonant frequency is limited to below the gain at the fundamental
frequency (50 Hz) in order to ensure filter will be appropriately damped
In order to speed up the genetic algorithm process, an initial population set is generated such that the
components (R, L,C) are limited to within realistic (non-zero to a reasonable finite level) values. Also,
the total “worst case” losses in the passive components of the filter for a given individual is limited to
within a reasonable level. The worst case loss scenario is found through consideration of a range feasible
operating points.
It is worth noting that the ripple inductor (Li) is often selected as an initial design step to meet a specific
current ripple requirement, accounting for the the rated converter output current, Ii, plus a crest factor
which varies with the frequency and DC bus voltage being utilized [7]. Optimization and simulation
results indicated a greater reduction in size (as well as a stable closed-loop control system) when allowing
the ripple inductance to be varied, and thus this initial design step was not considered for most of the
results presented. Results for a fixed ripple inductance of 10 mH are given, however, in Table I for
comparison purposes. This ripple inductance at the specified switching frequency and DC voltage level
results in a ripple magnitude of 20% of the nominal output current.
It is emphasized that when allowing the tuning of the ripple inductance during the optimization proce-
dure, the peak current including all harmonic components is still considered, i.e. accounting for crest
factor in components. Results favoured the allowance of a larger ripple current magnitude for an overall
size reduction. The switching frequencies associated with utility-scale power electronics must be kept
low to reduce switching losses, and thus the effect of a large high frequency ripple current, in terms of
magnetic core losses, are not as significant in the filter design as compared with high frequency switch-
mode power supply design.

Optimization Results

Table I: Optimization results (with and without constant ripple inductance)

Parameter Description LCL n=2 n=3 LCL n=2 n=3

Li Ripple inductor 3.9 mH 3.9 mH 2.1mH 10 mH (fixed) 10 mH (fixed) 10 mH (fixed)

C Shunt filter capacitance 108 µF N/A N/A 80µF N/A N/A

Lo Grid coupling inductor 5.0 mH 2.7 mH 3.4mH 2.6 mH 1.6 mH 1.2 mH

Rt Shunt trap resistance 4.88 Ω 0.1Ω,0.87Ω 0.2Ω,4.85Ω,0.18Ω 24.6Ω 0.1Ω,0.78Ω 0.1Ω,0.31Ω,1.39Ω

Lt Shunt trap inductance 23.6 mH 590µH, 60µH 410µH,102µH,33µH 16.5 mH 675µH,64µH 1.3 mH, 422µH,85µH

Ct Shunt trap capacitance 10 µF 13µF, 28µF 16.3µF,16.3µF,22.4µF 10µF 9.9µF,26µF 5µF,3.9µF,8.7µF

V Size estimate 20.72m3 13.65m3 12.45m3 26.63 m3 21.97 m3 20.36 m3

Pl Worst case losses 78 kW 72 kW 71 kW 144 kW 115 kW 107 kW

Results of the optimization procedure are presented in Table I. The resulting parameters, as well as size
and loss estimates are given for the case in which the ripple inductor is assigned a constant value as an
initial design step. Figure 2 shows the results of Table I in graphical form. The resulting waveform THD
for various frequency bands, obtained via time-domain simulation, is given later in this paper (Fig. 9b).

LCL Topology Two Trap Topology Three Trap Topology

3% 10%
12%
Li
Lo 8%
Lt1,2 7%
34% 34%
32% Ct1 < 1%

Ct2 9%
Li
52% Lo < 1%
Lt
Li 1,2,3
Lo Ct1
Lt 29%
< 1% Ct
2
C
Ct3
Ct
31% 38%

Figure 3: Size breakdown of filter reactive components for topologies being compared

The worst case overall filter losses amongst all feasible operating points are also shown in Table I. The
total losses depend primarily on the number of traps as well as the damping present in each trap. In
addition, the losses associated with the filter when the inverter is off, or exporting no power, are reduced
for the multiple-trap case.
It is acknowledged that the additional shunt branch in the LCL topology will contribute to the overall
volume, but emphasized that, for utility applications, the filter should be passively damped to avoid large
transients from grid events. Another alternative to provide passive damping would be either to increase
the resistance of the inductive branches, thereby markedly increasing the total losses in the filter, or in-
creasing the resistance in series with the shunt capacitor. Using a resistor in series with the capacitor
could provide the necessary damping without the additional L, C components, but as the shunt capaci-
tative branch represents a much lower impedance to the fundamental component when compared with
a tuned RLC with a reasonably high quality factor, the resistive losses will be much greater. Increasing
the resistance in series with the capacitance also reduces the effectiveness of the filter and increases the
sensitivity to the grid impedance [8]. It is for these reasons that it was decided to benchmark against a
LCL system passively damped with a tuned RLC trap.
The percentage size of each reactive component is shown in Figure 3. The inductive elements of the
tuned traps for both the benchmark and proposed topology account for an insignificant portion of the
total estimated size. This is due to the fact that the peak currents in these shunt branches are relatively
small and thus the peak stored magnetic energy will also be quite small. Since the size is estimate as
proportional to the peak stored energy, this results in a small size estimate for these inductors. It is likely
that a physical realization of these inductors would be larger than the given estimate, i.e. due to voltage
clearances, winding sizes, etc. Secondary effects, such as a dead-time in the switched voltage waveform,
will introduce low order harmonics to the system [9], as shown in Fig. 9a, which could be absorbed by
the shunt trap filters; this is especially true for the benchmark LCL topology as the RLC trap is tuned to a
low-order frequency. Thus, the peak currents in Lt will be higher if taking into account this phenomenon.
For this reason, the estimation technique was considered adequate for the purposes of comparison.
Also observed in Fig. 3 is that the RLC trap accounts for less than 4% of the volume of the benchmark
case. This suggests that including the RLC trap damp the LCL circuit resonance does not significantly
affect the overall size of this topology.

Output Current Verus VSC Terminal Voltage with Varying Grid Coupling Impedance
0
LCL (zero)
n=2 (zero)
n=3 (zero)
−20 LCL (moderate coupling)
n=2 (moderate coupling)
n=3 (moderate coupling)
LCL (weak coupling)
n=2 (weak coupling)
−40
n=3 (weak coupling)
Gain (dB)

−60

−80

−100

Fund mf 2mf
−120

2 3 4
10 10 10
Frequency (Hz)

Figure 4: Frequency Response for the LCL and multi-trap filters for a range of grid impedances

Sensitivity Analysis
Fig. 4 shows a comparison of frequency responses for the three filter topologies being compared for the
optimal parameter values specified in Table I. The transfer function, VIot (s) , is the output current to the
VSC terminal voltage. Also included in the figure are the frequency responses obtained when additional
series impedance of the grid is added at the output. Grid inductances considered were 0.5 mH and 5 mH
with a X/R ratio of 2. This filter is being considered for distribution level applications, and thus a low
X/R ratio is used.
The results indicate some sensitivity of the filter to the level of grid coupling; however, overall attenuation
tends to increase, thus ensuring that a zero grid impedance represents the worst case scenario for THD.
In addition, the resonant peaks move towards a lower frequency, potentially further within the control
bandwidth if designing the control system to perform active damping [10].
Modelling and Control
This section will demonstrate the control of a VSC utilizing the filter topologies under study in order to
demonstrate that a stable current control system can be achieved with the proposed multiple trap filter.
A generalized state space model of the multiple trap filter is given in Eq. 1 with the symbols referring to
those quantities shown in Fig. 1b. This model is utilized for control design and to present the frequency
response of the filter. By transforming this model into dq-form, control design in the rotating reference
frame can take place. This transformation is straight-forward and thus is excluded for brevity.

  ˙      
−Li −Lo ··· 0 0 ii Ri Ro ··· 0 0 ii −1 1
 1 −1 ··· 0 −1   i˙o   0 0 ··· 0 0  io   0 0  
 vi
       
[h!]  .. .. .. .. ..  . .. .. .. .. .. .. .. ..
  .. = +
     
 . . . . .    . . . . . 
 .   . .  vo

 0 0 ··· 0 −Ltn   v˙tn   1 −Ro ··· 1 Rtn  vtn   0 −1 
0 0 ··· Ctn 0 ˙
itn 0 0 ··· 0 0 itn 0 0
(1)

Control of the inverter output current can be achieved with a decoupled rotating frame current control. A
simplified model of the controlled utilized is shown in Fig. 5. The controller consists of a PI controlled
to regulate the d and q-axis currents as well as a decoupling term to allow for independent control of
these two quantities. In this case, the decoupling term neglects the mathematical coupling present in the
shunt filter elements, i.e. capacitor and trap in LCL filter and trap components in the multiple trap filters
with the assumption that the components are of sufficiently high impedance to the 50 Hz component.

   



  







 

Figure 5: Decoupled dq-axis current control utilized for closed-loop control design and testing

Fig. 6a shows the resulting open-loop phase/gain for the control system with the chosen controller gains
(K p, Ki) for a two-trap filter (n=2). A stable system is achieved with a phase margin of approximately
85 degrees.

Open Loop Transfer Function Freq. Response (n=2) Closed Loop Transfer Function Freq. Response (n=2)
80 20

60
0
40
−20
Magnitude (dB)

20

0 −40
−20 Closed Loop
−60 Bandwidth
−40
approx 25 Hz
−60 −80

−80
G.M.: 4.51 dB −100
−100 Freq: 3.82e+003 rad/s
Stable loop
−120 −120
−90 0

−45
−135
Phase (deg)

−90

−180 −135

−180
−225
−225
P.M.: 84.9 deg
Freq: 153 rad/s
−270 −270
−1 0 1 2 3 4 5 6 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/s)

(a) Control system open-loop frequency response (b) Control system closed loop frequency response
Figure 6: Frequency response associated with chosen gains for current control system (n=2)
The closed loop transfer function is shown in Fig. 6b, showing a control bandwidth of 25 Hz, which is
adequate for the control of the fundamental waveform in the rotating reference frame. When utilizing ac-
tive damping, further control design will be necessary to allow control of higher frequency components;
however, for the purposes of demonstrating a stable closed loop control system utilizing a multiple-trap
filter, this control bandwidth is adequate.

dq Current Control Step Response − Model vs Simulation Comparison

Step Response Magnitude (d−axis Current)

0.8 Approx Overshoot

6% Approx settling Time
80 ms

0.6

Approx Time Constant

5 ms
0.4

n=2 (simulation)
n=2 (model)
0.2
n=3 (simulation)
n=3 (model)
Reference
0
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19
Time (s)

Figure 7: Step response comparison between simulation and control system state space model

Simulation Results
Results from a time-domain simulation of the controlled system step response using Matlab Simulink
are shown in Fig. 7. In this figure, results for two filter types (n=2 and n=3) are shown, both utilizing the
same controller gains. Also shown is the step response from the analytical state space model presented
earlier in this paper. Response time is similar for both model and simulation, with discrepancies mainly
accounted for by the lack of perfect decoupling achieved with the controller utilized, which is not ac-
counted for in the state space model. Also observed is that both filter types have a similar response for
the same controller gains. This is expected, as the shunt elements should have little effect on the control
of the 50 Hz component.
Results of simulations demonstrating the control of an inverter utilizing the proposed multiple-trap filters
are presented in this section. Fig. 8 shows the relevant stationary-frame waveforms obtained in simula-
tion utilizing the controller of Fig. 5. As mentioned previously, the VSC is switched at 1950 Hz utilizing
SPWM with a DC bus voltage of 20 kV. The rating of the VSC is 12 MVA corresponding to a nominal
phase current of approximately 850 A.
One of the effects of switching dead-time is to introduce lower order harmonics to the input voltage
waveform, thus the effects of a dead-time of 10 µs is considered. It is noted that this is a much larger
than normal dead-time if considering insulated-gate bipolar transistor (IGBT) switching components,
and thus the effects on voltage waveform harmonics will be exaggerated.
The resulting output current is shown in Fig. 9a. Here it is seen that the low order harmonic resonance
peak, observed in Fig. 4, is being excited by the low-order harmonics present in the VSC terminal voltage
waveform due to the switching dead-time. Fig. 9b shows the THD for each section of the spectrum,
corresponding to the limits proposed in [2]. It can be seen that the presence of a large dead-time is
sufficient to increase the harmonic levels (for the n=3 case) beyond allowable limits.
 Current and Voltage Waveforms during Step Response (n=2)

VSC Voltage, Viab (kA)

20

10

−10

−20

1
VSC Curr., Ii (kA)

0.5

−0.5

−1
1
Output Current, Io (kA)

0.5

−0.5

−1
0.09 0.1 0.11 0.12 0.13 0.14 0.15
Time (s)

Figure 8: Waveforms for controlled response to step in d-axis reference to rated current

Controlled Phase Current with and without 10 us deadtime Total Harmonic Distortion by Frequency Band
1
4
n=2 (no deadtime)
0.8 n=2 (deadtime) IEE 1547 Limit
LCL (deadtime) 3.5 n=2 (no deadtime)
0.6 n=2 (deadtime)
THD (% by frequency band)

3 n=3 (no deadtime)

0.4
n=3 (deadtime)
LCL (no deadtime)
0.2 2.5
LCL (deadtime)
Ioa (kA)

Zoom @ t=5ms
0
2
−0.2
1.5
−0.4

−0.6
1

−0.8 0.5

−1
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0
Time (s) h<11 11<=h<17 17<=h<23 23<=h<35 h>=35

(a) Effect of large dead-time on output current (Io) wave-(b) Percentage THD by frequency band and cor-
forms responding limits according to IEEE 1547 stan-
dard for grid interconnection for odd harmonics
Figure 9: Effects of dead-time on output current waveform and THD by frequency band

Conclusions
Using estimation based on a stored energy per unit volume, multiple trap filters have been shown to
have potential for a decrease in power losses as well as a size reduction of 25-30% for the given system
constraints and parameters in comparison with an LCL filter topology. The results of a genetic algorithm
based optimization procedure are intuitive, in that they have suggested that tuning the multiple traps
to the most significant harmonic components of the voltage-sourced converter waveform results in the
smallest filter size while still meeting grid connection requirements. While this paper has given results
based upon a sinusoidally pulse-width modulated waveform, the same idea can be applied in order to
filter dominant high frequency harmonics from other switching strategies.
The proposed filter topology has been verified to be stable in time-domain simulations for both two
and three trap topologies when utilizing a basic decoupled rotating-frame current control system. By
extension, it is then possible to achieved the controlled export of real and reactive power, thus allowing for
the interfacing of distributed generation or other types of real and reactive power compensation devices.

References
[1] A. Rockhill, M. Liserre, R. Teodorescu, and P. Rodriguez, “Grid Filter Design for a Multi-Megawatt
Medium-Voltage Voltage Source Inverter,” Industrial Electronics, IEEE Transactions on, no. 99, pp. 1–
1.,2011
[2] “1547.2-2008 IEEE Application Guide for IEEE Std 1547, IEEE Standard for Interconnecting Distributed
Resources with Electric Power Systems,” tech. rep., 2009.
[3] M. Liserre, F. Blaabjerg, and S. Hansen, “Design and control of an LCL-filter-based three-phase active
rectifier,” Industry Applications, IEEE Transactions on, vol. 41, no. 5, pp. 1281–1291., 2005
[4] V. Blasko and V. Kaura, “A novel control to actively damp resonance in input LC filter of a three-phase
voltage source converter,” Industry Applications, IEEE Transactions on, vol. 33, no. 2, pp. 542–550., 1997
[5] M. Liserre, A. Dell’Aquila, and F. Blaabjerg, “Genetic algorithm-based design of the active damping for an
LCL-filter three-phase active rectifier,” Power Electronics, IEEE Transactions on, vol. 19, no. 1, pp. 76–86.,
2004
[6] P. Imbertson and N. Mohan, A method for estimating switch-mode power supply size. IEEE, 1989.
[7] M. Prodanovic and T. Green, “Control and filter design of three-phase inverters for high power quality grid
connection,” Power Electronics, IEEE Transactions on, vol. 18, no. 1, pp. 373–380., 2003
[8] M. Liserre, A. Dell’Aquila, and F. Blaabjerg, “Stability improvements of an LCL-filter based three-phase
active rectifier,” Power Electronics Specialists Conference, 2002.pesc 02. 2002 IEEE 33rd Annual , vol.3,
no., pp. 1195- 1201 vol.3, 2002 IEEE.
[9] J. Seung-Gi and P. Min-Ho, “The analysis and compensation of dead-time effects in PWM inverters,” IEEE
Transactions on Industrial Electronics, vol. 38, pp. 108–114, Apr. 1991.
[10] N. Abdel-Rahim and J. Quaicoe, “Modeling and analysis of a feedback control strategy for three-phase
voltage-source utility interface systems,” in Proceedings of 1994 IEEE Industry Applications Society Annual
Meeting, vol. 18, pp. 895–902, IEEE, 1994.