Event Driven Control of Voltage and Current Gradients of Medium Voltage IGBTs

Abstract – Medium voltage IGBTs are fast switching de-

Ls iD R
vices that require low gate drive power. They inherently gen-
uD
erate high voltage and current gradients during switching +
L
transitions. These are generally limited by retarding the chan- ud
iC
ges of the gate charge. Additional gate resistors are usually uCE
installed for this purpose. The drawback is high switching loss- iC
es. 4 kV iC
A novel method is described in this paper that aims at con- a) 600 A
trolling the voltage and current gradients during switching 2 400
transitions. The gate charge of the input MOS device is con- uCE uCE 200
trolled by injected gate currents. These follow particular com- b) 0 0
mand functions that are predefined and stored in a memory.
The method requires reacting within microseconds which tra- 4 kV
ditional closed loop control cannot do. Specific time events 600 A
uD
are therefore defined to trigger the respective command func- 2 400
tion. These functions depend on the instantaneous values of uD 200
the collector current or the collector-emitter voltage, variables 0 0 iD
that are identified without delay, knowing their predefined iD – 200
gradients and counting the time from the respective event to – 400
reach their final values. c) 0 4 8 12 16 20 ms
Experimental results show the performance of event driv- t
en control. Low current and voltage gradients are enforced Fig. 1 Commutation waveforms: a) equivalent circuit, b) IBGT wave-
forms, c) diode waveforms. Device losses occur in the blue regions.
while the switching losses are reduced.
I. INTRODUCTION - insulation stress on the stator windings,
- overvoltages due to travelling waves in the motor cable,
Integrated gate bipolar transistors (IGBTs) are the preferred
- generation of bearing currents,
power semiconductor devices used for medium voltage in-
- poor EMC compatibility.
verters. Their fast switching transients are controlled through
a MOS gate which requires only low control power. Typical • High current gradients entail
applications are speed variable ac machine drives in the up- - high IGBT losses at turn on,
per megawatt range. - high diode losses at turn off.
The maximum ratings of presently available medium volt- The effect of high current gradients is visualized in the com-
age IGBTs are UCE max = 6.5 kV and IC max = 750 A. These mutation waveforms Fig. 1. Blue areas mark the regions where
high values cause considerable device losses during switch- losses of IGBT and diode occur. The diode reverse recovery
ing transitions. Medium voltage inverters must be therefore current of high amplitude reflects also on the IBGT.
operated at low switching frequencies of only few hundred Voltage and current gradients are conventionally limited
Hertz in order to restrain the switching losses to tolerable lev- by inserting an additional resistor at the gate terminal [5], [6].
els [1]. Optimal pulsewidth modulation technics are then ap- The resistor delays the gradient of the gate voltage, and con-
plied to keep the harmonic current distortion low. This can sequently the respective gradients of the collector emitter volt-
be achieved by synchronous optimal pulsewidth modulation age and the collector current. Values of about 1 kV/ µs and
[2], or alternatively by predictive current control [3]. An ad- 2 kA/µs are a common target. The trade-off is higher com-
ditional option is using multilevel inverter topologies that in- mutation losses.
herently
Manuscriptproduce
receivedlower
June 2, current distortion
2015; revised [4]. accepted Octo-
Aug 21, 2015; Fig. 2 shows that a current gradient of 2.7 kA/µs at turn on
ber 5, 2015. leads to device losses of 3.2 Ws per commutation. A voltage
Copyright © 2015 IEEE. Personal use of this materialdistortion
While these methods provide low harmonic at
is permitted.
low switching
However, frequency,
permission to use thistwo major
material for problems that relate
any other purposes must to
be gradient of 1 kV/µs at turn off produces 4.8 Ws per commu-
obtained from the IEEE by sending a request to pubs-permission@[Link]. tation, Fig. 3. An inverter operating at f s = 200 Hz switching
theJoachim
switching
Holtztransitions remain
is with Wuppertal unsolved:
University, Germany, [Link]@[Link]
frequency then generates losses of 15 kW per device. Heat
• High voltage gradients entail
 4 kV 4 kV
Rg 2
Rg
15 W Rg = 56 W
uCE 8.2 uCE 22
3 3
4.6 6.8

0 0
1.6 kA
1 kV/µs
1.2
0.8 kA
iC 0.8
0.6
0.4 Rg
iC 0.4
0
0.2
2.7 kA/µs
0
4 MW 4 MW

3 3
Rg
Pon Poff Rg
2 2
1 1
0 0
8 Ws 8 Ws
Rg
6 6
Rg
E on 4 E off 4
2 2

0 0
0 5 t 10 ms 0 5 t 10 ms
Fig. 2 Turn-on performance with different gate resitors Rg Fig. 3 Turn-off performance with different gate resitors Rg

dissipation becomes a problem such that the capability of the sents its dynamic behavior. Symbols displaying the step re-
devices cannot be fully utilized. Operating at high gradients sponse are used here for characterization.
requiers installing expensive output filters to protect the mo- The input variable in Fig. 4(b) is the gate-emitter voltage
tor against switching overvoltages and bearing currents [7]. uGE and the output variable is the collector current iC. When
Analytical models have been used to investigate the switch- a positive step of uCE is applied, the gate-emitter-capacitance
ing properties of IGBTs and to control the voltage gradient CGE gets charged and the gate-emitter voltage uGE starts in-
[6]. The interactions between the gate drive and the device creasing. Eventually will this voltage reach the threshold volt-
are described in [8]. The results in [5, 6, 8] refer to low volt- age uth shown in Fig. 4(a). The collector current iC then be-
age IGBTs that operate at high switching frequencies. They gins to rise. The IGBT acts as a voltage controlled device in
are not applicable to medium voltage devices. this mode.
The method described in this paper aims at controlling the The dynamic properties of the IGBT are derived from the
voltage and current gradients during switching transitions and equivalent circuit Fig. 5. The gate control signal acts upon a
make them to follow predefined reference values while si- MOSFET structure which is the input device. The MOSFET
multaneously reducing the switching losses. generates the base current of a bipolar transistor which is the
II. IGBT DYNAMICS output device. A delay occurs during this process while the
gate-emitter capacitance CGE gets charged through the inter-
IGBTs are conventionally described by a static model as
nal gate resistor RG int. Further delay occurs until the base-
shown in Fig. 4(a). The signal flow graph Fig. 3(b) repre-
emitter voltage uBE has reached the threshold level uth. Only
 20 V
1 kA 3
uGE iC
2A 10
0.8
iC iG 0 0 uGE
b) iG
0.6 uGE

0.4 iG CGE uGE 800 V iC 200 A

0.2 uCE
c) 400 100
0 uCE iC
a) 5 uth 10 15 V
uGE
Fig. 4 Characterization of medium voltage IGBTs 0 0
(a) static characteristic. uth is the threshold voltage. (b) dynamic 0 0.6 kV/ms 1 t 2 ms
behavior at gate voltage control, and (c) at gate charge control
Fig. 6 Turn on process with a voltage ramp uGE applied
then starts the bipolar transistor conducting. A sequence of rapid changes during a switching process results in a fast vari-
mutually coupled nonlinear reactions follows: the collector- ation of CGC. The dependency
emitter voltage uCE drops rapidly, which adds to the gate cur- CGC = f (uCE ) (1)
rent a displacement current through CGC. This capacitance
changes as the collector-emitter voltage uCE changes. The cur- is one of the causes that make the switching dynamics non-
rent gain of the bipolar transistor depends on the magnitude linear. Further nonlinear properties are summarized in the
of the collector current. It is schematically indicated by the following.
step response Fig. 4(c) that the IGBT behaves as a nonlinear An internal displacement current ∆iG 1 caused by chan-
higher order oscillatory system.
Earlier investigations of the authors have demonstrated that ges of uCE is injected into the gate through the gate-collector
a low voltage gradient of 0.6 kV/µs can be achieved without capacitance CGC
increasing the switching losses [9]. The IGBT was turned on du
∆i G 1 = CGC CE (2)
by a ramp-shaped gate voltage having two different gradients dt
as shown in Fig. 6. The response shows that the gate current
i G temporarily reduces when the negative gradient of the col- Another component ∆iG 2 that adds to the gate current is the
lector-emitter voltage appears. The effect is owed to a dis- displacement current through CCB
placement current that is internally injected into the gate du
through the gate-collector capacitance CGC. The result sug- ∆iG = CCB CE (3)
2 dt
gests that further research could be worthwhile.
The analysis starts with looking into the IGBT dynamics and hence
during switching transitions. The gate current iG is consid- iG =/ iG* (4)
ered the controlling variable. Fig. 7 shows the device in the
blocking state where the magnitude of the collector-emitter were iG* is the commanded gate current.
voltage uCE determines the width of the depletion region. This
SiO
voltage thus controls the gate-collector capacitance CGC. Its G E
iC C
p+-emitter
14243

UCE UCE depletion region

CGC
uCE
R G int iG D n -base
on G
off CGE
G n +-buffer
+u G –uG S CCB p -collector
uGE

E' C
E Fig. 7 Structure of an IGBT
The width of the depletion region depends on UCE
Fig.5 IGBT quivalent circuit
 Fig. 11 Command function at voltage gradient control

6 8 nF
A/V 8A 4A 4

turn on
5

turn o
iG(t) 6
4 duCE /dt 6 3

n
gm
3 4 i G* i G*
4 2
2 CGC 2
1 turn
2 off 1
0 0
0 0.2 0.4 0.6 0.8 kA 0 1 2 3 4 kV turn off
iC uCE(t) 0 0
0 0.2 0.4 0.6 0.8 kA 0 1 2 3 4 5 kV
(a) the function gm = f (iC) (b) collector gate capacitance iC uCE
Fig. 8 Identified IBGT characteristics
(a) current control (b) voltage gradient control
Fig. 9 Command functions
The gate current iG changes the gate charge QG in the course
of time
identified by measurement. The results serve to deriving the
1
QG =
CGE ∫ iG dt (5) command functions for gradient control.
III. EVENT-TRIGGERED CONTROL
which introduces another nonlinearity. uGE is the gate-emit- 3.1 Extraction of the command funtions
ter capacitance. Note that it is the gate charge that deter- The dynamic properties of the IGBT are measured, sub-
mines the state of conduction of the MOSFET, not the gate jecting the device to a transient condition. A gate current step
voltage. is applied of constant amplitude iG =1 Amp. Recorded from
Finally is the current gain β of the bipolar transistor a non- the response are the signals
linear function of the collector current.
diC / dt
gm ( iC) = C GC
β = f (iC ) (6) iG (7)

Equations (1) through (6) describe the nonlinear dynamics shown in Fig. 8(a), and
of a medium voltage IGBT. They form part of an algorithm iG
to control the dynamic behavior during the switching process. = f (uCE ) (8)
duCE / dt
It is obvious that using the static characteristics alone would
not lead to improvement. It is the internal feedback expressed shown in Fig. 8(b).
by (1), (2), (3) and (5) that makes the IGBT a higher-order Since the external gate current is kept constant, it is only
dynamic system. the displacement current ∆iG 1 through CGC that changes
Operating the IBGT by gate charge control instead of gate
voltage control is the basis of a novel method presented during the transient. Hence the acquired signal (7) permits
[Link] strategy is controlling identifying the gate-cathode capacitance
• at turn on: ∆iG 1 (t )
CGC (t ) = (9)
first the current gradient diC/dt duCE / dt
and subsequently the voltage gradient duCE/dt, Both the left side term and the right side term of this equation
• at turn off: figure at the ordinate axis in Fig. 8(b).
first the voltage gradient duCE/dt
and subsequently the current gradient diC/dt. 3.2 Event driven control at turn on
Following the turn-on command, the gate current is forced
The respective gradients are adjusted by predefined com-
to ig = 2 A to let the gate charge increase until the emitter
mand functions. A constraint is that IGBTs are very fast
voltage uGE has reached the threshold voltage uth. This pro-
switching devices. Switching transitions are completed with-
cess lasts throughout
in fractions of a microsecond. The required bandwidth for gate
charge control is therefore in the Megahertz range. Traditional QG (uth )
Dt1 = (10)
methods would not be suited for closed loop control. Event- iG
triggered gate charge control is used instead. The controlling which interval depends on the amount of gate charge QG that
variable is the gate current, derived from command functions the device needs to get controllable. Dt1 is a device depen-
that are addressed by the actual dynamic state of the device. dent constant interval that triggers event 1 after completion.
The dynamic properties of the respective type of IGBT are
 diC , then: duCE duCE diC
control first:
dt dt
control first:
dt
then:
dt 5
2A 20 V 20 V
uGE uGE
iG iG uGE
0 0 0 0
uGE iG
iG
– 20 – 2A – 20
0 D t1 5 t 10 ms 0 5 t 10 ms
Dt2 Dt1 Dt2
Dt3
Dt3 variables
turn on: variables settle
turn on event 1: Dt1
event 1: Dt1 settle
I (t) ud(t)
event 2: Dt2 (IC) = C event 2: Dt2 (ud) =
(diC /dt)* (duCE /dt)*
ud(t) IC (t)
event 3: Dt3(ud) = event 3: Dt3 (IC) =
(duCE /dt)* (diC /dt)*
Fig. 10 Event-triggered control at turn on Fig. 11 Event-triggered control at turn off

value (uCE-gradient)*, uCE(t) is computed as

From this point on the device is controllable.
The gate current reference for current gradient control is uCE (t ) = ( d uCE / dt )*⋅ Dt3( uCE ) (14)
now obtained from (7) as
where Dt3(uCE) is the time required for uCE(t) to reach its
1 diC
i G* (t ) = C final value, which is the dc link voltage ud, and Dt3(ud) =
gm (iC (t )) GC dt (11)
ud(t) /(duCE/dt)* is the voltage gradient command. Dt3 marks
This signal is computed from the measured characteristic event 3 at which the voltage ramp is completed. A small
Fig. 8(a). It defines the command function Fig. 9(a) that is residual value is then produced by the command function. It
stored in a memory. The capacitance CGC in (11) has a lets the gate voltage rise to its maximum final value, the
constant value since the collector-emittervoltage uCE rests source voltage of the drive circuit. The gate current decays to
near the dc link voltage ud at current gradient control. The zero while the gate-emitter voltage approaches the full source
value of CGC(ud) is available at Fig. 8(b) with the transient voltage of the drive circuit: the device is locked in its con-
oscillations neglected. ducting state.
The argument i C(t) in (11) varies rapidly during a switch-
ing transient. Measuring its instantaneous value is therefore 3.3 Event driven control at turn off
not possible. A command value (iC-gradient)* is defined onto Upon receiving the turn-off command, the gate current is
which the current trajectory is made to follow. The trajectory forced to ig = –2 A to let the gate discharge the gate until the
iC(t) is computed as emitter voltage uGE has come down to the threshold voltage
uth. Dt1 is a device dependent constant interval that triggers
iC (t ) = ( diC / dt )*⋅Dt2 ( Ic ) (12) event 1 after completion. From this point on the device is con-
trollable.
It lasts until Dt2(IC) where Dt2(IC) = Ic(t) /(diC/dt)* is the
Gradient control at turn off starts with controlling the volt-
time required for i C(t) to reach its maximum value IC and
age gradient. Equation (13) determines the gate current refer-
(diC/dt)* is the current gradient command. Dt2(IC) marks
ence with uCE(t), being computed as
event 2 at which the current ramp is completed. The value
Ic(t) is obtained by measurement since it remains almost uCE (t ) = ( d uCE / dt )*⋅ Dt 2( ud ) (15)
constant within Dt2. Control of the voltage gradient is subse- where (duCE/dt)* is the voltage gradient command and Dt2(ud)
quently initiated. is the time required for uCE(t) to reach its final value ud. Dt2
The oscillations at the end of the commutation in Fig. 8(b) marks event 2 at which the voltage ramp is completed.
are neglected to obtain the clean reference value profile Control of the current gradient follows. Gate current refer-
ig*(uCE) for voltage gradient control, shown in Fig. 9(b). It ence (10) is used and iC(t) is computed as
serves as command function
iC (t ) = ( diC / dt )*⋅D t3( Ic ) (16)
iG* = f ( uCE (t )) (13)
at voltage gradient control and is stored in a table. Event 3 at which the current ramp is completed is given as
The argument uCE(t) of (13) undergoes rapid changes dur- Dt3(IC) = IC(t)/(diC/dt)*. The current reference is now set to
ing a switching transient. It cannot obtained by measurement, – 2A. The gate-emitter voltage then decreases, the gate cur-
but since the voltage gradient strictly follows its command rent reduces to zero, and the device is blocked.
 4 kV The sequence of events at turn on is shown in Fig. 10. Fol-
lowing the turn-on command, the gate current is forced to
6
3 uCE
uCE ig = 2 A to let the gate charge increase until the emitter volt-
600 2 age uGE has reached the threshold voltage }th. This process
A lasts throughout
1 iC
QG (uth )
iC 0 Dt1 = (19)
iG
3 kV/ms which interval depends on the amount of gate charge QG that
2 the device needs to get controllable. Dt1 is a device depen-
duCE dent constant interval that triggers Event 1 when uGE has
dt 1 come up to the source voltage of the drive circuit. From this
0 point on the device is controllable. The gate current is then
read from the stored command function Fig. 9(a). While the
–1
gate charge continues increasing, the collector current starts
2 kA/ms ramping up. It is kept at the desired gradient diC/dt = const.
according to (17).
1
diC The command function Fig. 9(a) is addressed by variable
dt 0 i C(t), which is computed from (12) to evade time the delays
associated to measurement. Depending on the actual range of
–1
iC(t), only portions of the characteristics Fig. 9 are used to
–2 address the respective gate current through the time-varying
4 MW abscissae variable.
Event 2 is computed as (diC/dt)*.Dt, where (diC/dt)* is the
Poff command value (17) and Dt counts the time starting from zero
2 at event 1. Event 2 terminates the range of current gradient
control when the collector current has reached the current level
IC of the external load. Since IC changes only slowly a mea-
0 kV sured value can be used.
6 Ws µs The fly-back diode has now changed to the blocking mode.
1 Its forward voltage starts increasing while the diode capaci-
E off 4 tance gets charged. The voltage gradient is actively controlled
1.5
by the IGBT through its gate current reference (18) referring
2 3
to the command function Fig. 9(b). The time interval Dt3 to
0 reach Event 3 is computed as (duCE/dt)*Dt where (duCE/dt)*
0 5 t 10µs is the command value and Dt counts the time starting at event
2. Event 3 sets the end to this process when the collector-emit-
Fig. 12 Event driven control at turn off
command value (duCE/dt)* = par.
ter voltage has reached the level ud(t)/(duCE/dt)* where ud(t)
is the measured dc link voltage and (duCE/dt)* is the com-
mand value (18). The gate current continues following its ref-
The magnitudes of IC and ud during a particular commuta- erence (14). It decreases exponentially while charging the
tion are generally less than the respective maximum values gate-emitter capacitance CGE up to the stationary value uGE
in Fig. 9. It is only portions of the respective command func- that equals the source voltage of the gate driver circuit.
tions that are used. A similar sequence of events is used at turn off. The proce-
3.4 Experimental results dure is shown in Fig. 11. The turn off command generates a
IGBT type Infineon FZ600R65KF1_S2 was used for ex- negative current of constant magnitude to discharge the gate-
perimental verification. The devices were operated at 4 kV emitter capacitance. During a time interval Dt1 is the device
dc link voltage and 0.6 kA load current. The command val- transferred to the controllable region using (19). The dura-
ues were set as tion Dt1 depends on the gate-emitter capacitance CGE and its
charge at the end of the foregoing turn-on process, lastly on
( di C / dt )* = 1kA µs (17) the source voltage of the gate driver circuit. The end of this
for current gradient control and interval defines event 1.
During following time interval the voltage gradient is con-
( duCE / dt )* = 1 kV µs (18) trolled using the gate current reference (14). Event 2 is com-
for voltage gradient control. puted as (duCE/dt)*Dt where (duCE/dt)* is the command val-
0.8 kA 4 kV 4 kV 7
uCE
iC uCE
2
iC 0.4 2
uCE uCE
0
0 0
1 kA/µs 1 kV/µs
0
3 MW iG
iG
2 Pon –2 A
0 1 2 3 4 5 ms
Pon 1 t
Fig. 14 Turn off at voltage gradient conrol, du/dt = 1 kV/µs
0
6 Ws dients produce high switching losses. These problems are com-
4 Eon monly eliminated by limiting the gradients, delaying the
switching transitions through added gate registers. The draw-
Eon 1 back is increasing switching losses and derating of the equip-
ment.
0 As a countermeasure, the device is seen as a dynamic sys-
0 2 4 6 8 10 ms
t tem. It is modelled considering the internal cross-couplings
Fig. 13 Event driven control and Rg-control at turn on between different spatial state variables. To exert an indepen-
dent control, the forcing variable is not the gate-emitter volt-
ue and Dt counts the time starting from event 1. age but the gate current which in turn controls the gate charge.
Event 2 terminates the phase of voltage gradient control Using this approach, voltage and current gradients are made
when uCE has ramped up to the dc link voltage level and the to follow predefined trajectories. The trajectories are charac-
collector current starts from zero. terized by device-dependent command functions. They are nu-
Control of the current gradient follows. Event 3 that termi- merically extracted by experiments, stored in a memory and
nates this phase is computed as I C (t)/{(di C/dt)*} where retrieved online. Particular events serve to call the respective
(diC/dt)* is the command value and IC(t) is measured. This command function. The events themselves depend on the re-
phase ends when iC has come to zero. spective dynamic state of the device. The method enforces
The turn-off process at event driven control with different low gradients while reducing the switching losses. It makes
settings of (duCE/dt)* and (diC/dt)* is shown in Fig. 12. Lower additional inverter output filters obsolete.
voltage gradients lead to higher the turn-off [Link] on at
event driven control with duCE/dt = 1 kV/µs and diC/dt =
1 kA/µs is shown in Fig. 13. Oscillogram Fig. 14 shows the
performance of voltage gradient control with duCE/dt set to 1
kV/µs. The gate current waveform results as a smooth signal
since the gate drive circuit forces the gate current onto its ref- 4 kV 4 kV/ms
erence. It compensates the displacement currents that are in- duCE
dt uCE
jected internally into the gate. Against this, there are switch- uCE
ing transients being transferred to the gate if conventional gate 2 2
drive method is used. This method produces more than 4 kV/ duCE
µs, Fig. 15. Turn-off losses are ccompared in Fig. 16, show- dt
inga 60% reduction with event driven control. The improve- 0 0
ment reduces as the voltage gradient increases. Fig. 17 shows
that it is particularly at lower voltage gradients- that turn-off
losses are reduced. 0
iG
IV. SUMMARY iG
Medium voltage IGBTs are fast switching devices. They –2 A
produce high voltage gradients that generate overvoltages at 0 1 2 3 4 5 ms
t
the connected machines and degrade their bearings. Compen-
sation by AC filters is expensive. Moreover, high current gra- Fig. 15 Conventional turn-off process, du/dt = 4 kV/µs
 4 kV
8
uCE 6
Ws
uCE 5
3 Rg-control
Rg-control
event driven control 4
0 E loss
1.25 kV/µs 3

1.6 kA
2 event driven
1.2
iC 1
iC 0.8

0.4 0
1 2 3 4 kV/µs
duCE
0 dt
4 MW Fig. 17 Conventional and event driven control at turn off

3
IEEE Trans. Power Elec., Vol. 6, No. 2, April 1991, pp. 208-
Poff 219.
2 Poff 6. A. R. Hefner, „An Improved Understanding for the Transient
1 Operation of the Power Insulated Gate Bipolar Transistor
(IGBT)“, IEEE Trans. Power Elec., Vol. 5, No. 4, Oct 1990,
pp. 459-468.
0 7. T. G. Habetler, R. Naik, and T. A. Nondahl, Design and Imple-
8 Ws mentation of an Inverter Output LC Filter Used for DV/DT
Reduction“, IEEE Trans. Power Elec., Vol. 17, No. 3 , May
6 2002, pp. 327-331.x M. Kojima, K. Hirabayashi, Y. Kawa-
bata, E. C. Ejiogu, and T. Kawabata, „Novel Vector Control
E off Eoff System using Deadbeat-Controlled PWM Inverter with Output
4 LC Filter“, IEEE Trans. Ind. Appl., Vol. 9, No. 1, Jan/Feb
2004, pp. 162-169.
2 8. A. R. Hefner, „Analytical Modeling of Device – Circuit Interac-
tions for the Power Insulated Gate Bipolar Transistor ( IGBT)“,
0 IEEE Trans. Ind. Appl., Vol. 26, No. 6, Nov(Dec 1990, pp.
0 2 4 5 6 8 10 ms 995-1005.
t
8. T. Abdelhedi, „Improving the Switching Properties of Insulated
Gate Bipolar Transistors (in German)“, Ph.-DThesis, Wuppertal
Fig. 16 Comparison: Event driven control and Rg-control at turn off
University, 1994.

V. REFERENCES
1. A. Rathore, J. Holtz, and T. Boller, „Synchronous Optimal
Pulsewidth Modulation for Low Switching Frequency Control
of Medium-Voltage Multilevel Inverters“, IEEE Transactions
on Industrial Electronics, 2010, pp. 2374-2381.
2. J. Holtz, G. da Cunha N. Petry, and P. [Link], „Control of Large
Salient Pole Synchronous Machines using Synchronous Opti-
mal Pulsewidth Modulation“, IEEE Transactions on Industrial
Electronics, Vol. 62, No. 6, June 2015, pp. 3372-3379.
3. J. Holtz „Advanced Pulsewidth Modulation and Predictive Con-
trol – An Overview“, IEEE Transactions on Industrial Elec-
tronics, Vol. 6?, No. 6, May 2016, pp.
4. T. Boller, A. Rathore, and J. Holtz, „Generalized Optimal Pulse-
width Modulation of Multilevel Inverters for Low Switching
Frequency Control of Medium Voltage High Power Industrial
AC Drives“, IEEE Transactions on Industrial Electronics, Vol.
60, No. 10, 2013, pp. 4215-4224.
5. A. R. Hefner, „An Investigation of the Drive Circuit Require-
ments for the Power Insulated Gate Bipolar Transistor (IGBT)“,