A Fast Model Predictive Control with Fixed

Switching Frequency Based on Virtual Space

Vector for Three-Phase Inverters
Chao Jiang, Guiping Du, Fada Du, Yanxiong Lei
School of Electric Power
South China University of Science and Technology
Guangzhou, China
[email protected], [email protected], {1027693753, 904853421 }@qq.com

Abstract—This paper proposes an improved finite control methods, they are summarized in simple penalization [6], [7],
set model predictive control (FCS-MPC) algorithm based on sliding window [8], and notch filter [9], which are elaborated
virtual space vector to reduce computational burden, and in [10]. However, these solutions can only ameliorate
achieve a constant switching frequency for two-level three- harmonic distribution to some extent, they do not achieve
phase inverters. The voltage vector reference is calculated in a true fixed switching frequency like what pulse-width
60°coordinate system according to the principle of deadbeat modulation (PWM) realizes. To get a concentrated harmonic
(DB) technique. A large number of virtual voltage vectors distribution without modulator, a simple and effective
synthesized by discrete space vector modulation (DSVM) is approach is to add a period detector into cost function [3],
utilized to approach the voltage vector reference. To avoid
[10]–[14]. Nevertheless, the feasibility of those solutions is
enumerating all the virtual voltage vectors from a look-up
table, an algebraic way using the voltage vector reference is
based on that the sampling frequency is much higher than the
adopted to calculate the three candidate vectors. Then, a cost desired switching frequency, which burdens processor
function is defined to select the optimal voltage vector among heavily. Besides, the frequency weight factor of the detector
them. This new method greatly improves the computation influences the control effect significantly, and it is
efficiency since plenty of multiply operation is eliminated in the complicated and iterative to obtain a suitable value. Hence,
60°coordinate system. Moreover, a fixed switching frequency the other, which is easier and intuitive, to get constant
is achieved since the durations of each switch states has been switching frequency is to combine the modulator with FCS-
calculated. Experimental results under a two-level three-phase MPC [15]–[19]. The key step in this kind of method is to
inverter verify the effectiveness of the proposed control calculate the operation durations. An easiest way is to
strategy. calculate it as space vector PWM does [15]. Another
effective way is to utilize a mathematical methodology,
Keywords — finite control set model predictive control (FCS- which searches the optimal duty cycle that could minimize
MPC), constant switching frequency, virtual space vector, three- the defined weighted error, to calculate the duty cycle [16],
phase inverter [17].
The conventional FCS-MPC only implements one
I. INTRODUCTION
switching vector in a sampling period, hence, the complete
Finite control set model predictive control (FCS-MPC) control region of converters is narrow and not precise enough.
implements directly the switch vector to the system by If one sampling period is divided into N equal parts, then the
minimizing a predefined cost function. Since the cost larger N, the larger range of control region. This is described
function is defined arbitrarily, controller can be designed to detailedly in [18]. However, a challenge brought by a large N
include system constraints and multiple control objectives [1], is the numerous calculation because conventional FCS-MPC
[2]. However, a tremendous challenge coming along is the enumerates every virtual space vectors. A more efficient
random switching frequency because every actuation solution proposed in [20] is to calculate the candidate voltage
determining process is irrelevant. The variable switching vectors by an algebraic way based on the voltage vector
frequency produces a wide range of harmonic spectrum, reference. Nevertheless, the calculation method is
weakening the performance of the system in terms of power complicated and inefficient in traditional αβ coordinate
quality [3]–[5]. Furthermore, the designs of filters become system.
complex and inefficient because a typical filter design relies
on which harmonic frequencies that should be eliminated. In this paper, a constant switching frequency FCS-MPC
Hence, conventional FCS-MPC is not suitable in some based a 60°coordinate system called gh coordinate system to
application where harmonic distribution is a significant calculate the candidate voltage vectors is proposed. The
criterion. discrete space vector modulation (DSVM) is used to
synthesize a large number of virtual voltage vectors. To
There are plenty of strategies to keep FCS-MPC operate reduce computational burden, an algebraic way is adopted to
at a fixed switching frequency at present [3], [6]–[19]. calculate the three candidate virtual voltage vectors which
Basically, according to the existence or inexistence of are adjacent to the voltage vector reference. Once the
modulator, they can be divided into two typical types. One is candidate voltage vectors is determined, a cost function is
to adjust cost function properly without relying on modulator defined to select the best one. Since all the virtual space
to get fixed switching frequency [3], [6]–[14]. According to vectors are synthesized by real space vectors, it is easy to
the different alterations of cost functions used in control calculate the operation durations.

Supported by Guangdong Natural Science Foundation, 2017B030312001.

978-1-5386-6054-6/18/$31.00 ©2018 IEEE

 The rest of this paper is organized as follows. Section II
introduces the discrete models of the two-level three-phase
S1 S2 S3
inverters. Then, the principle and implementation of
proposed control strategy are presented in Section III. Later
ia L R
Section IV provides the simulation and experimental results a
of the proposed control technique and other FCS-MPC Vdc ib
methods. Finally, a conclusion of this work is summarized in b
Section V. ic
c

II. DISCRETE MODEL OF THREE-PHASE INVERTER S1' S2' S3'

The topology of two-level three-phase inverter with
linear load is presented in Fig.1. The system dynamics can be
described in three-phase static abc coordinate system as
Fig. 1. Topology of two-level three-phase inverter with linear load.
follow:
Then, the prediction of the future output currents at
instant k + 1 can be expressed by transforming (5) as follow:
ia  U an  ia 
d    
L ib   U bn   R ib  (1)
dt i (k  1)  Ts U (k ) 
ic  U cn   ic  RTs i (k ) 
i (k  1)   U (k )   (1  )  (6)
  L    L i (k ) 
where ia, ib, and ic are the phase current; L is filter inductance
and R is the load; Uan, Ubn, and Ucn are output voltage of each According to the principle of deadbeat control, the output
phase, and they can be calculated as follow: currents should reach their references at the end of one
sampling period, that is:
U an   2 1 1  Sa 
U   Vdc  1 2 1  S  (2) i (k  1)  i* (k  1) 
 bn  3    b i (k  1)    *  (7)
U cn   1 1 2   Sc    i (k  1) 

where Vdc is dc power supply; Sa, Sb, and Sc are switch states Combining (6) and (7), the inverter desired output
of each phase. voltage reference can be obtained as follow:
For a balanced load, (1) can be transformed into two
independent systems in αβ coordinate system:
U* (k )  L i* (k  1)  L i (k ) 
 * = *   (R  )   (8)
U  (k )  Ts i (k  1)  Ts i (k ) 
d i  U  i 
L i   U   R i  (3)
dt       
III. PROPOSED CONTROL METHOD
The corresponding transformation matrix from three- A. gh Coordinate System
phase static abc coordinate system into the static αβ
coordinate system is: Universally, the FCS-MPC with modulator calculates the
operation durations of each switch state in the αβ coordinate
system. However, the computational speed is not fast enough
 1 1  since the processor has to perform a large amount of
 1   multiplication. A 60° coordinate system, also called gh
2 2 2 
Cabc /    (4) coordinate system as described in Fig.2, whose axes are
3 3 3 located at zero and sixty degrees respectively, is introduced
0 2 
2  to simplify the calculation [21], [22]. The corresponding
transformation matrix from αβ coordinate system into gh
Since the sampling period is relatively small compared coordinate system is:
with the period of output current, the discrete mathematic
model of the inverter can be derived by using Euler forward
method as follow:  1 
 1 
3 3 
C / gh   (9)
2 2 
d i (k  1)  i (k )  U (k )  i (k )  0 
L i (k  1)  i (k )   U (k )   R i (k )  (5)  3 
dt         
Combining (8) and (9), the inverter desired output voltage
where Ts is the sampling period. reference in gh coordinate system can be obtained as follow:
 h
h U3 U2
h
U3 U2
U3(-1,1)
U2(0,1)
II
U*(k) U4 U0 U1
U4 U0 U1
III I U7 g U7 g

U0(0,0) U1(1,0)
U4(-1,0)
U7(0,0) g U5 U6 U5 U6
(a) (b)
IV VI Fig. 3. Control region with virtual space vectors. (a) N =2; (b) N =3.

V
h
U5(0,-1) U6(1,-1) U3(-1,1) II U2(0,1)

Candidate space vectors

Fig. 2. gh coordinate system III
I Reluctant space vector
Vlu Vlu Vuu Selected virtual space vector
U*(k+1)
U0
U g* (k )  L ig* (k  1)  L ig (k )  U4(-1,0)
 * = *   (R  )   (10) U7 Vll Vul U1(1,0) g
U h (k )  Ts ih (k  1)  Ts  ih (k ) 
IV
VI
B. Delay Compensation
Due to the intrinsic property of digital implementation, a U5(0,-1) V U6(1,-1)
control delay always exists in actual control process. To Fig. 4. Relationship between U* and candidate space vectors
alleviate the control error caused by this delay, a simple
solution that determining the voltage vectors at the (k + 1)th
C. Virtual Space Vectors
instant is applied [23], [24]. Hence, the predictive voltage
model is calculated using (10) shifted one step forward in The conventional FCS-MPC selects the optimal real
time: space vector in every sampling period. For example, if the
voltage vector reference is located at the subinterval I as
Fig.2. Then, the controller will choose the best switch vector
U g* (k  1)  L ig* (k  2)  L ig (k  1)  from the three nearest candidates that are V1, V2, and V0 (V7).
 * = *   (R  )   (11) However, the error between real space vector and voltage
U h (k  1)  Ts ih (k  2)  Ts  ih (k  1)  vector reference, which always exists in practice, impairs the
control performance. Hence, that combining several real
voltage vectors in one sampling period to synthesize the
Based on linear interpolation theorem, the current virtual voltage vector is used to reduce the control error since
reference at time k + 1 can be calculated by the past current they occupy more vector space [18], [25]. For a two-level
references as follow: three-phase inverter with linear load, dividing one sampling
period into N equal parts, the control region is not the limited
eight real space vectors anymore. In other words, the control
ig* (k  1)  ig* (k )  ig* (k  1)  ig* (k  2)  region is enlarged by the introduction of virtual voltage
*   3 *   3 *  *  (12) vectors. Fig.3, in which the intersection of two lines is the
ih (k  1)   ih (k )   ih (k  1)  ih (k  2)  end point of virtual space vector, gives the control region
when N = 2 and N = 3. It can be deduced that the total
Similarly, shifting (12) one step forward, the current number of feasible virtual space vectors is as follow:
reference at the (k + 2)th instant can be calculated as follow:

ntotal  3N 2  3N  2 (15)
ig* (k  2)  ig* (k  1)  ig* (k )  ig* (k  1) 
*   3 *   3 *    *  (13)
ih (k  2)  ih (k  1)   ih (k )  ih (k  1)  Obviously, the larger N is, the larger the control region is.
Nevertheless, the conventional FCS-MPC enumerates all the
Then, the cost function based voltage model can be voltage vectors, hence, the amount of calculation is very
expressed as follow: huge, especially when the N is big enough. Therefore, a new
mission is to reduce the computational load.

g  U g* (k )  Vg (k )  U h* (k )  Vh (k ) (14) D. Calculation of Candidate Space Vectors

Dividing each real space vectors by Vdc, then, the length
where Vg(k) and Vh(k) are the candidate inverter voltage of them are 1 as shown in Fig.4. In the meanwhile, the
vectors. sampling period is divided into N equal parts, hence, we can
assume that the coordinates of voltage vector reference is:
 U g* (k  1) / Vdc   x / N 
 *   (16)
U h (k  1) / Vdc   y / N 

where |x| ≤ N and |y| ≤ N. Ugh (k+1) Sab c

Cost function
On the other hand, the coordinates of each virtual space mini mization
vectors can be described as follow:
Candidate vectors iabc (k)
calculation V1,V2,V3
Vg / Vdc   m / N 
V / V    n / N  (17) [U*g(k+1),U*h(k+1)]T
 h dc    calculation

where Vg and Vh are the g component and h component of the i*abc (k) abc i*g h (k) [ig(k+1),ih(k+1)]T , igh (k) abc
virtual space vectors; m and n are two integers. [i*g(k+2),i*h(k+2)]T
gh gh
calculation
Multiplying (16) with N, and then rounding up and down
respectively, we can get four candidate space vectors as Ugh (k)
follows:
Fig. 5. Control block diagram of the proposed control method

Vuu    x  ,  y  
 TABLE I. EXPERIMENTAL PARAMETERS
 Vll    x  ,  y  
 (18) Parameters Description Value
Vlu    x  ,  y   Vdc DC power supply 250 V
V  x , y
 ul      
R Loads 11 Ω
L Filter inductance 10 mH
Ir Reference current magnitude 5A
fo Output current frequency 50 Hz
where   denotes rounding it up, and   denotes fs Sampling frequency 24 kHz
rounding it down. td Dead time 200 ns
N Subsection Number 200
Obviously, the Vul/N and Vlu/N are always the candidate
space vectors. The selection of the third candidate vector can IV. SIMULATION AND EXPERIMENT RESULTS
be done by evaluating the sign of the expression: The proposed fast FCS-MPC has been verified by
simulation and experimental results under steady-state and
transient conditions. The MATLAB/Simulink R2014a is
 x  y     x    y   (19)
utilized to simulate the proposed control method. A 32 bit
DSP (TMS320F28335) from Texas Instruments is used in all
If the sign is positive, then Vuu/N is the third candidate the experiments. Switches used in all the experiments are
vector. Otherwise, Vll/N is the third candidate vector. IRGP20B60PD from Infineon Technologies. The main
parameters using in simulation and experiments are listed in
E. Calculation of Operation Durations Table I.
A virtual space vector minimizing the cost function The steady-state simulation results of the conventional
defined in (14) is selected to implement to the converter. FCS-MPC and the proposed FCS-MPC are shown in Fig. 6
Now that m and n are both integers, therefore, the operation and Fig. 7. Compared with the former, it is clear that the
durations of each real space vectors, which synthesizes the latter presents a better output current harmonic distribution
corresponding virtual space vector, is integral multiple of Ts and lower total harmonic distortion (THD).
= N. Since the coordinates of each virtual space vectors are
on the basis matrix [U1 U2] as shown in Fig.4, in order to Fig. 8 shows the dynamic simulation waveforms of two
calculate the operation durations, we have to transform to different control strategies when the reference current is
another corresponding basis matrix. For example, if a changed. Obviously, the two control methods can track the
selected virtual space vector is located at subsection III as reference current fast.
shown in Fig.4, it has to left multiply the basis matrix [U3 In order to validate the efficiency of the proposed FCS-
U4]-1 to get the durations. MPC, an input/output (I/O) port of DSP is set to measure the
duration of control process. When the processor begins to
F. Implementation Process execute the control program, the I/O port outputs a high level.
The control block diagram of the proposed control While the control program is completed, the I/O port outputs
method with fixed switching frequency is shown in Fig.5. a low level. Measurement results are shown in Fig. 9.
Compared with the conventional FCS-MPC and other fixed
switching frequency FCS-MPC techniques, the key are those
blocks about reference calculation that has been colored with
green and the block about candidate vectors calculation that
has been colored with orange.
 As shown in Fig. 9, the execution time for the
conventional FCS-MPC is 14.51 μs. For the control strategy
proposed in this paper, it requires 27.01 μs for the same
process. It is clear that the conventional needs less execution
time. However, the proposed control method has much better
performance.
Fig. 10 shows experimental steady-state results of the
conventional FCS-MPC when the reference current peaks at
(a) 5A. The corresponding waveforms of output line voltage Uan,
Uab, and phase current ia are depicted in Fig. 10a, and the
harmonic spectrum of phase A output current is presented in
Fig. 10b. The same experiment test results but with different
control strategies are shown in Fig. 11.
As show in Fig. 10b and Fig. 11b, on the one hand, the
harmonic spectrum distribution is improved significantly. It
can be seen that the harmonic spectrum of the proposed
control strategy is mainly gathered around the switching
frequency and its multiples, while the conventional FCS-
(b)
MPC obtains a scattered harmonic spectrum. On the other
Fig.6. Steady-state simulation results of the conventional FCS-MPC. (a) hand, the THD is reduced significantly with 1.39%.
Load current, phase A voltage, and line voltage; (b) Phase A harmonic Moreover, the value achieved at the fundamental frequency
spectrum.
of proposed control method is 5.096 A, which indicates the
tracking error is 1.92% from the reference of 5 A. In
comparison, the conventional FCS-MPC obtains a THD with
6.03% as shown in Fig. 11b. Moreover, the counterpart
current value at fundamental frequency is 5.124, which
means a tracking error of 2.48%.

(a)

(b) (a)
Fig. 7. Steady-state simulation results of the proposed FCS-MPC. (a) Load
current, phase A voltage, and line voltage; (b) Phase A harmonic spectrum.

(a)

(b)
Fig. 9. Execution time of different FCS-MPC strategies. (a) The
conventional FCS-MPC; (b) Proposed FCS-MPC.
(b)
Fig. 8. Output current simulation behaviors with different strategies. (a)
Conventional FCS-MPC; (b) Proposed FCS-MPC.
 (a)

(a)

(b)
Fig. 12. Output current experimental behaviors with different strategies. (a)
Conventional FCS-MPC; (b) Proposed FCS-MPC.

TABLE II. EXPERIMENTAL RESULTS COMPARISON

Parameter Conventional FCS-MPC Proposed FCS-MPC

Execution time (μs) 14.51 27.01

THD 6.03% 1.39%

Fundamental (A) 5.124 5.096

(b) Tracking error 2.48% 1.92%
Fig. 10. Steady-state experimental results of the conventional FCS-MPC. (a)
Load current, phase A voltage, and line voltage; (b) Phase A harmonic
spectrum. To test the dynamic performance of the proposed FCS-
MPC, we change the reference current while system is
running to inspect the input current behavior. The reference
current is changed from 5 A to 3 A, The experimental results
are shown in Fig. 12. Compared with the conventional FCS-
MPC as shown in Fig. 12a, the control strategy proposed
presents a similar fast tracking ability no matter how the
magnitude of current reference changes. Therefore, the
proposed control method has a very good dynamic
performance.
For a better understanding, the aforementioned
experimental results of the conventional FCS-MPC and the
proposed FCS-MPC are summarized in Table II.

(a)
V. CONCLUSION
This paper proposes a fast FCS-MPC with fixed
switching frequency based on virtual space vectors. The
control method is applied in a two-level three-phase inverter
with linear load. The effectiveness is compared with other
FCS-MPC methods, the execution time of the proposed
control method is longer, however, the steady-state
performance is improved greatly since the THD is much
lower and the harmonic spectrum is concentrated. In the
meanwhile, it keeps the same good dynamic performance.
Hence, this proposed FCS-MPC control method has better
performance than the conventional FCS-MPC.

REFERENCES
(b)
[1] J. Rodriguez and P. Cortes, Predictive Control of Power Converters
Fig. 11. Steady-state experimental results of the proposed FCS-MPC. (a) and Electrical Drives. John Willey-Sons, 2012.
Load current, phase A voltage, and line voltage; (b) Phase A harmonic [2] S. Vazquez, J. I. Leon, L. G. Franquelo, J. Rodriguez, H. A. Young, A.
spectrum. Marquez, and P. Zanchetta, “Model predictive control: A review of its
 applications in power electronics,” IEEE Industrial Electronics Magazine, discrete space-vector modulation for pmsm drives,” IEEE Transactions on
vol. 8, no. 1, pp. 16–31, March 2014. Industrial Electronics, vol. 64, no. 5, pp. 3537–3547, May 2017.
[3] M. Tomlinson, T. Mouton, R. Kennel, and P. Stolze, “Model predictive [21] N. Celanovic, Space Vector Modulation and Control of Multilevel
control with a fixed switching frequency for a 5-level flying capacitor Converters. Virginia, 2000.
converter,” in 2013 IEEE ECCE Asia Downunder, June 2013, pp. 1208– [22] J. Y. Zheng, Z. L. Shen, J. Mei, and L. F. Wang, “An improved neutral
1214. point voltage balancing algorithm for the npc three-level inverter based on
[4] L. Tarisciotti, P. Zanchetta, A.Watson, J. Clare, S. Bifaretti, and M. virtual space vector pwm,” in 2010 International Conference on Electrical
Rivera,“A new predictive control method for cascaded multilevel and Control Engineering, June 2010, pp. 3283–3287.
converters with intrinsic modulation scheme,” in IECON 2013 - 39th [23] F. Sebaaly and H. Y. Kanaan, “New voltage vector generation method
Annual Conference of the IEEE Industrial Electronics Society, Nov 2013, fora mpc algorithm with constant switching frequency operation,” in
pp. 5764–5769. 2017IEEE 26th International Symposium on Industrial Electronics
[5] R. Mikail, I. Husain, Y. Sozer, M. S. Islam, and T. Sebastian, “A (ISIE), June2017, pp. 1692–1698.
fixed switching frequency predictive current control method for switched [24] J. H. Lee, J. S. Lee, H. C. Moon, and K. B. Lee, “An improved finite set
reluctance machines,” IEEE Transactions on Industry Applications, vol. model predictive control based on discrete space vector modulation
50,no. 6, pp. 3717–3726, Nov 2014. methods for grid-connected three-level voltage source inverter,” IEEE
[6] T. M.Wolbank, R. Stumberger, A. Lechner, and J. Machl, “Novel Journal of Emerging and Selected Topics in Power Electronics, pp. 1–
approach of constant switching-frequency inverter control with optimum 1,2018.
current transient response,” in SPEEDAM 2010, June 2010, pp. 803–808. [25] D. Casadei, G. Serra, and K. Tani, “Implementation of a direct control
[7] M. Rivera, S. Kouro, J. Rodriguez, B. Wu, V. Yaramasu, J. Espinoza, and algorithm for induction motors based on discrete space vector
P. Melila, “Predictive current control in a current source inverter operating modulation,” IEEE Transactions on Power Electronics, vol. 15, no. 4, pp.
with low switching frequency,” in 4th International Conference on Power 769–777, Jul 2000.
Engineering, Energy and Electrical Drives, May 2013, pp. 334–339.
[8] C. A. Rojas, M. Aguirre, S. Kouro, T. Geyer, and E. Gutierrez, “Leakage
current mitigation in photovoltaic string inverter using predictive control
with fixed average switching frequency,” IEEE Transactions on Industrial
Electronics, vol. 64, no. 12, pp. 9344–9354, Dec 2017.
[9] P. Cortes, J. Rodriguez, D. Quevedo, and C. Silva, “Predictive current
control strategy with imposed load current spectrum,” in 2006 12th
International Power Electronics and Motion Control Conference, Aug
2006,pp. 252–257.
[10] M. Aguirre, S. Kouro, C. A. Rojas, J. Rodriguez, and J. I. Leon,
“Switching frequency regulation for fcs-mpc based on a period control
approach,” IEEE Transactions on Industrial Electronics, vol. 65, no. 7, pp.
5764–5773,July 2018.
[11] M. Norambuena, H. Yin, S. Dieckerhoff, and J. Rodriguez, “Improved
finite control set model predictive control with fixed switching frequency
for three-phase npc converter,” in PCIM Europe 2016; International
Exhibition and Conference for Power Electronics, Intelligent Motion,
Renewable Energy and Energy Management, May 2016, pp. 1–8.
[12] M. Tomlinson, H. d. T. Mouton, R. Kennel, and P. Stolze, “A fixed
switching frequency scheme for finite-control-set model predictive
control–concept and algorithm,” IEEE Transactions on Industrial
Electronics, vol. 63, no. 12, pp. 7662–7670, Dec 2016.
[13] O. Abdel-Rahim, H. Funato, and J. Haruna, “An efficient mppt technique
with fixed frequency finite-set model predictive control,” in 2015 IEEE
Energy Conversion Congress and Exposition (ECCE), Sept 2015,
pp.6444–6449.
[14] P. Nikhil, K. Sonam, M. Monika, and S. Wagh, “Finite control set model
predictive control for two level inverter with fixed switching frequency,”
in 2018 SICE International Symposium on Control Systems (SICE
ISCS),March 2018, pp. 74–81.
[15] F. Sebaaly, H. Vahedi, H. Y. Kanaan, N. Moubayed, and K. Al-Haddad,
“Model predictive controller with fixed switching frequency for a 3l-
npcinverter,” in IECON 2016 - 42nd Annual Conference of the IEEE
Industrial Electronics Society, Oct 2016, pp. 6536–6511.
[16] Y. Yang, H. Wen, and D. Li, “A fast and fixed switching frequency model
predictive control with delay compensation for three-phase inverters,”
IEEE Access, vol. 5, pp. 17 904–17 913, 2017.
[17] M. Rivera, “A new predictive control scheme for a vsi with reduced
common mode voltage operating at fixed switching frequency,” in
2015IEEE 5th International Conference on Power Engineering, Energy
and Electrical Drives (POWERENG), May 2015, pp. 617–622.
[18] S. Vazquez, J. I. Leon, L. G. Franquelo, J. M. Carrasco, O. Martinez,J.
Rodriguez, P. Cortes, and S. Kouro, “Model predictive control with
constant switching frequency using a discrete space vector modulation
with virtual state vectors,” in 2009 IEEE International Conference on
Industrial Technology, Feb 2009, pp. 1–6.
[19] J. Lei, L. Tarisciotti, A. Trentin, P. Zanchetta, P. Wheeler, and A.
Formentini, “Fixed frequency finite-state model predictive control for
indirect matrix converters with optimal switching pattern,” in 2016 IEEE
Energy Conversion Congress and Exposition (ECCE), Sept 2016, pp. 1–8.
[20] Y. Wang, X. Wang, W. Xie, F. Wang, M. Dou, R. M. Kennel, R.
D.Lorenz, and D. Gerling, “Deadbeat model-predictive torque control with