Page 1 of 10 2022-PCIC-0540

MATRIX CONVERTERS: CHALLENGES AND SOLUTIONS

Copyright Material IEEE
Paper No. PCIC-

Galina Mirzaeva Yuan Liu

Senior Member, IEEE Postgraduate Research Student
The University of Newcastle The University of Newcastle
Callaghan NSW2308 Callaghan NSW2308
Australia Australia
[email protected] [email protected]

Abstract – The advantages of high efficiency and reliability, is practically sinusoidal, with only high-frequency (switching)
as well as weight and loss benefits, make matrix converters an harmonics present. Other advantages include fully regenerative
attractive option for AC/AC power conversion. This paper operation and the ability to control (typically, at unity) the supply
reviews the existing AC/AC converter solutions in mining side power factor.
industry and provides a detailed discussion of the main features
and techniques associated with matrix converters. The paper
explains the fundamental difficulties in satisfying the load side AC Input
Lf
and the supply side performance objectives, simultaneously. vU va AC
The paper then discusses and compares several control vV vb
options for matrix converters, including the recently proposed vW vc MOTOR

Comprehensive control. The latter shows advantages over iU,V,W ia,b,c

other options and can be recommended for demanding industry
applications. Findings of the paper are supported by detailed
simulations. (a) Standard VSD topology: diode rectifier and controlled inverter

Index Terms — AC/AC converter, matrix converter, power AC Input

converter control, mining Industry vU Lf
vA va v AC
vV vB b
I. INTRODUCTION vW vC vc MOTOR

iU,V,W ia,b,c
A large number of mining applications require AC/AC power
conversion from one (usually, standard) to another (usually,
variable or adjustable) frequency and amplitude. These (b) Regenerative VSD topology: AFE and controlled inverter
applications include electric shuttle cars and loaders, trucks and
excavators, continuous miners and conveyors, grinders and Fig. 1 Common AC/AC power converter topologies
crushers, fans and pumps, and many others [1]-[3]. A typical
example of such applications is an AC variable speed drive Note that both VSD topologies shown in Fig.1(a) and Fig.1(b)
(VSD), referred further in this paper as VSD. include a decoupling stage in the form of a DC-link capacitor.
A traditional solution to the AC/AC power conversion has This capacitor constitutes larger part of the VSD weight and
been to utilize a two-step conversion: AC/DC via a controlled or volume. It is also known to be the first component to fail [4]. From
uncontrolled rectifier and DC/AC via a controlled inverter. The reliability and size/weight reduction viewpoints, removing the DC
middle stage of the AC/DC/AC conversion constitutes a large link capacitor is desirable. Therefore, direct AC/AC power
DC-link capacitor, which decouples the two AC sides and their conversion is of interest.
respective control.
B. Direct AC/AC Converters
A. Standard AC/DC/AC Converters
The concept of direct AC/AC power conversion has been
A typical industrial VSD is based on an uncontrolled (diode) known for decades. Its realization in the form of thyristor-based
rectifier and a voltage source inverter (VSI), as shown in Fig.1(a). cycloconverters can be found in high-power, low-speed mining
The use of diode rectifier causes a significant low-frequency applications such as grinding mills [5], [6].
distortion of the supply side current. Also, when a motor driven The cycloconverter topology is shown in Fig.2(a). Each of the
by this VSD is braking, regenerative power cannot be returned to output AC lines is connected to all three input AC lines via
the supply and is dissipated on a chopping resistor (not shown in thyristors with controlled firing angles. Using anti-parallel
Fig.1(a)). thyristors (see, for example 𝑆!"# and 𝑆!"$ in Fig.2(a)) provides
Another popular VSD topology is shown in Fig.1(b). It is known conduction in both directions.
as a four-quadrant VSD, or a VSD with Active Front End (AFE). Recent developments in power electronic technology [7] have
In this topology, diode rectifier is replaced by a controlled voltage enabled a modern version of a ‘cycloconverter’ - a matrix
source rectifier (VSR). Due to this, current drawn from the supply converter (MC). The MC concept first appeared in the literature

1
/20/$31.00 © 2022 IEEE
 2022-PCIC-0540 Page 2 of 10

in the 1970s [8]. Effective modulation schemes for the MC were of the thyristors. The output voltage must have lower frequency
devised in the 1980s [9]-[10] and further developed in the 1990s- than the input voltage, i.e., 𝑓' < 𝑓( . The resulting output
2000s [11]-[12]. The existing MC applications include aerospace, waveforms are highly distorted and include low frequency
power quality compensation, renewable generation, etc. [13]. harmonics at integer multiples of 𝑓' .
Electric transportation is one particular application that can fully Fig.3(b) shows how the three phase output voltages of MC are
utilize the MC advantages of compactness (owing to the lack of accurately reproduced, on average, by fast switching between
the DC-link capacitor), high efficiency and reliability. Recently, the three-phase input voltages. With using MC, any output
the MC is being introduced to mining industry, especially, electric frequency can be obtained, provided that 𝑓' ≪ 𝑓%& .
haulage vehicles [2].

AC Input L
f vA VA VB VC V*
a
V*
b
V*
c
Va Vb Vc

vB
vC

SAa+
SAa- vb
va AC vc
MOTOR

(a) Thyristor AC/AC converter: cycloconverter

AC Input Lf vA
vB
vC 0 0.02 0.04 0.06 0.08 0.1

Cf
SBa SCa SAb SBb SCb SAc SBc SCc (a) Cycloconverter: synthesis of 𝑉!"# from 𝑉$% (𝑓! < 𝑓$ )
SAa
vb
va AC vc
MOTOR

(b) Switched AC/AC converter: matrix converter

vsA isA Lf , R f iiA A

~v SAa SAb SAc
sB isB viA iiB B
~v isC viB iiC SBa SBb SBc
sC
~ viC
C S
Ca SCb SCc
Cf
a b c
ioa iob ioc (b) Matrix converter: switching at 𝑓&' ≫ 𝑓$ to form 𝑉!"# from 𝑉$%
voa vob v
Alternative oc
connection Ll , R l
AC
Motor
Bidirectional switch
(c) Common representation of a direct matrix converter

Fig. 2 Topologies of direct AC/AC power converters

Fig.2(b) shows that the basic topology of an MC is very similar

to that of a cycloconverter, save for using different switches, such
as IGBT or MOSFET, operating at high switching frequencies
𝑓%& , in the order of 10kHz. Fig.2(c) shows an equivalent MC
representation that is more common in the literature. It also
shows the two options for the MC input filter.
Fig.3 illustrates principles of power conversion corresponding
to the AC/AC converter topologies of Fig.2. Fig.3(a) shows how
each phase of the output voltage of a cycloconverter is
(c) Matrix converter: increasing 𝑉!"# /𝑉$% from 0.5 to 0.866
synthesized from segments of all three-phase input voltages at
the standard supply frequency (50Hz or 60Hz), via commutation Fig. 3 Output voltage waveforms of direct AC/AC converters

2
/20/$31.00 © 2022 IEEE
Page 3 of 10 2022-PCIC-0540

As shown in Fig.3(b), the ’envelope’ of the output voltages unbalanced, resulting in nonzero common-mode voltage (CMV)
must fit inside the ’envelope’ of the input voltages, which limits 𝑉)* defined as
the voltage transfer ratio to 0.5. However, the transfer ratio can 3
be increased to 0.866 when adding a common-mode voltage 𝑉)* ≜ 1𝑣" (𝑡) + 𝑣/ (𝑡) + 𝑣) (𝑡)6 (2)
.
(CMV), at 3𝑓( and 3𝑓' , to all three output voltages, as shown in
Fig.3(c). Despite the apparent distortion of the average output The CMV 𝑉)* will appear at the neutral point of a wye
voltages (e.g., 𝑣" + 𝑣)* ), the CMV is cancelled between the connected load. To minimize the CMV, some authors considered
phases of the load and produces no current. constraining the set of the switching states to the rotating vectors
Another way to increase the voltage transfer ratio to 0.866 is to only [15]. However, this approach has severe limitations on the
use Space Vector Modulation (SVM), which will be discussed in magnitude and frequency of the output voltage. In a vast majority
the next section. of applications, it suffices that the CMV averages to zero over a
full space vector revolution.
II. MATRIX CONVERTER FUNDAMENTALS A. Modulation Strategies for Matrix Converter
The notations shown in Fig.2(c) are used throughout this
1) Direct solution approach: Historically, this was the first
paper, namely, 𝑣% and 𝑖% for the supply voltage and current; 𝑣( and approach, introduced by Alesina-Venturini in early 1980s [9]. Its
𝑖( for the MC input voltage and current; 𝑣' and 𝑖' for the MC
objective was to find a time varying switching function 𝑺(𝑡) such
output voltage and current; (A, B, C) for the input phases and (a, that
b, c) for the output phases.
𝑣' = 𝑺(𝑡)𝑣( 𝑎𝑛𝑑 𝑖( = 𝑺4 (𝑡)𝑖' (3)
The MC output typically feeds an inductive load (defined by
resistance 𝑅+ and inductance 𝐿+ ). The input side includes an LC where 𝑥 denotes a vector (in a matrix sense).
filter (𝐿, , 𝐶, ) with a damping resistor 𝑅, , which can be connected Assuming that
in series or in parallel to 𝐿, . The parallel connection is usually 𝑐𝑜𝑠(𝜔( 𝑡)
preferred due to lower 𝐼- 𝑅 losses. -5
𝑣( = 𝑉( B𝑐𝑜𝑠(𝜔( 𝑡 − )
. H (4)
The direct MC shown in Fig.2(c) includes 9 bidirectional -5
switches arranged in a 3×3 matrix. The available switching states 𝑐𝑜𝑠(𝜔( 𝑡 + )
.
are constrained by two fundamental limitations, namely, avoiding
a short circuit of the input voltage, and avoiding an open circuit 𝑐𝑜𝑠(𝜔' 𝑡 − 𝜑' )
-5
of the output current (since an inductive load acts like a ’current 𝑖' = 𝐼' B𝑐𝑜𝑠(𝜔' 𝑡 − − 𝜑' )H (5)
.
source’). These limitations result in that one and only one switch -5
should be ON in each column of the matrix. This gives 33 = 27 𝑐𝑜𝑠(𝜔' 𝑡 + − 𝜑' )
.
possible switching states including:
then, indeed, a closed-form solution can be found for 𝑺(𝑡) such
- 18 states in which two out of three output phases are that applying (3) results in:
connected to the same input phase and the remaining output 𝑐𝑜𝑠(𝜔' 𝑡)
phase is connected to a different input phase, resulting in the so- -5
called active vectors, each having magnitude related to 𝑣( and 𝑣' = 𝑞𝑉( 𝑐𝑜𝑠 𝜑( B𝑐𝑜𝑠(𝜔' 𝑡 − )
. H (6)
constant orientation; 𝑐𝑜𝑠(𝜔' 𝑡 +
-5
)
.
- 3 states in which all output phases are connected to the
same input phase, resulting in zero vectors; 𝑐𝑜𝑠(𝜔( 𝑡−𝜑( )
- 6 states in which all output phases are connected to -5
𝑖( = 𝑞𝐼' 𝑐𝑜𝑠 𝜑' B𝑐𝑜𝑠(𝜔( 𝑡 − .
−𝜑( )H (7)
different input phases, resulting in the so-called rotating vectors, -5
each having magnitude related to 𝑣( and varying orientation. 𝑐𝑜𝑠(𝜔( 𝑡 + −𝜑( )
.
One useful concept to understand the effect of each switching 6$ ;%
state on the MC input and output is ’space vector’. Space vector where 𝑞 = 6 = is transfer ratio.
% 789 :% ;& 789 :&
describes combined effect of all phases of an AC quantity (e.g.
voltage), and is defined as [14]: In fact, two closed-form solutions for 𝑺(𝑡) were found, namely,
𝑺𝟏 (𝑡) at frequency 𝜔*3 = 𝜔' − 𝜔( resulting in 𝜑( = 𝜑' ; and 𝑺𝟐 (𝑡)
-
𝑣⃗ ≜ . 1𝑣" (𝑡) + 𝑎𝑣/ (𝑡) + 𝑎- 𝑣) (𝑡)6 = 𝑣∠𝛼. (1) at frequency 𝜔*- = −𝜔' − 𝜔( resulting in 𝜑( = −𝜑' . A desired
!"
value of the input displacement angle 𝜑( can be achieved by
0#
where 𝑎 = 𝑒 . combining both solutions with appropriate weights.
The initial direct solution [9] was limited to a maximum transfer
Appendix A lists space vectors, calculated for all 27 switching ratio of 𝑞*"> = 0.5 (as illustrated in Fig.3(b)). A modified solution
states of the MC. Several examples appearing at the bottom of then followed [10], where an increased 𝑞*"> = 0.866 was
Appendix A, assist in understanding of the space vector achieved (as shown in Fig.3(c)).
calculation process. Despite its elegance, the direct solution approach has found
For any switching state, the sum of input currents 𝑖! + 𝑖1 + limited practical application. This is mainly due to the fact that its
𝑖2 = 0 and the sum of output line-to-line voltages 𝑣"/ + 𝑣/) + time scope corresponds to low-frequency periods of frequencies
𝑣)" = 0. This is because the load neutral point is isolated (hence 𝜔( and 𝜔' , which are assumed constant. In this sense, the SVM
𝑖" + 𝑖/ + 𝑖) = 0) and the supply represents a balanced voltage approach, described next, has an advantage, as its time scope
source (hence 𝑣! + 𝑣1 + 𝑣2 = 0). However, for the majority of amounts to a single sampling period, and the constant frequency
the switching states, the output phase voltages (𝑣" , 𝑣/ , 𝑣) ) are assumption does not hold.

3
/20/$31.00 © 2022 IEEE
 2022-PCIC-0540 Page 4 of 10

system of 4 equations (for 𝑣⃗'D , 𝑣⃗'"∗ , 𝚤⃗(D and 𝚤⃗(" ) with 4 unknowns (the
3 2 3 2 dwell times). Unfortunately, such a system is singular and has no
±1,±2,±3 ±3,±6,±9 solution. This reflects the fact that 𝑣⃗'AA and 𝚤⃗( cannot be controlled
±7,±8,±9 ±2,±5,±8 independently and must satisfy the instantaneous balance of real
vo' voLL* ii' ii* powers: 𝑝(F = 𝑝'GH . Once this condition is added then the system
4 1 4 1 reduces to 3 independent equations with 4 unknowns, and still
αo βi cannot be solved.
v"o ii" A common approach is to impose a sensible fourth condition.
±4,±5,±6 ±1,±4,±7 For example, each of the two current combinations, 𝑖;; 𝛿 ;; +
5 6 5 6 𝑖;6 𝛿 ;6 and 𝑖; 𝛿 ; + 𝑖;;; 𝛿 ;;; , may be required to have the same
direction as the reference current 𝚤⃗(∗ [12]. Then the four dwell
times can be obtained as
a) Construction of output voltage (b) Construction of input current "
M & $ N 789KO
789KL
"
P% $ N
-
𝛿; = (−1)I(#I% 𝑞 #
789 :%
#

Fig. 4 Space Vector Modulation of matrix converter √.

" P% # N "
- M & $ N 789 KO
789KL
2) Space Vector Modulation (SVM) approach: This 𝛿;; = (−1)I(#I%#3 𝑞 #
789 :%
#
√.
approach was introduced in 1990s by Huber-Borojevic [11] and " P% $"N
(8)
- M & # N 789KO
789KL
was further developed in 2000s by several researchers including 𝛿;;; = (−1)I(#I%#3 𝑞 # #
√. 789 :%
[12]. " P % #" N
M & # N789 KO
- 789KL
Assume that, in every sampling period, references for the 𝛿;6 = (−1)I(#I% 𝑞 #
789 :%
#
√.
output voltage space vector 𝑣⃗'∗ and the input current space vector
𝚤⃗(∗ are given. These references are pictured on a complex plane where 𝑘Q and 𝑘( are the sector numbers for the voltage and
in Fig.4(a) and Fig.4(b), respectively. For the sake of symmetry current space vectors, respectively.
between voltage and current, the line-to-line voltage reference
The sign of each dwell time found from (8) determines which
𝑣⃗'AA∗ is preferred to the phase voltage reference 𝑣⃗'BC∗ . vector, positive or negative, should be used. For example, if 𝛿 ; >
In this example, 𝑣⃗'AA∗ and 𝚤⃗(∗ both belong to the respective 0 then use vector +1; if 𝛿 ; < 0 then use vector −1.
sectors number 1. However, any combination of the two sectors Once the dwell times of the active vectors are determined, the
can exist because the space vectors 𝑣⃗'AA∗ and 𝚤⃗(∗ rotate, as remaining part of the cycle is distributed between zero vectors
general case, at different frequencies. Notations 𝛼N' and 𝛽P( denote (see vectors 0" , 0/ , 0) in Appendix A). Then all the vectors are
angles relative to the middles of the respective sectors. Both arranged in a symmetrical pattern as shown in Fig.5. For smooth
references 𝑣⃗'AA∗ and 𝚤⃗(∗ are then projected onto the directions of commutation, the order of the vectors inside the SVM pattern is
" "
the nearest available switching states (e.g., 𝑒 0 ' and 𝑒 $0 ' ). As a such that only one phase switches at a time.
result, 𝑣⃗'AA∗ = 𝑣⃗'D + 𝑣⃗'"∗ and 𝚤⃗(∗ = 𝚤⃗(D + 𝚤⃗(" . Expressions (8) define the standard SVM dwell time
From the data presented in Appendix A, the switching vectors calculation. Variations and extensions have been proposed to
corresponding to the above directions can be identified. There deal with overmodulation [16] and supply unbalance [17]. Carrier
are indeed, three such vectors in each direction. Namely, for 𝑣⃗'AA PWM for MC are also described in the literature [18].
"
these are: ±1, ±2, ±3 in the 𝑒 0 ' direction and ±4, ±5, ±6 in the B. Matrix Converter Input and Output Models
" "
$0 '
𝑒 direction. For 𝚤⃗( these are: ±3, ±6, ±9 in the 𝑒 0 ' direction
"
and ±1, ±4, ±7 in the 𝑒 $0 ' direction. For the control purpose, the three phase quantities on either
It can be shown that, for any sector combination, there are side are usually converted into the equivalent two orthogonal
phases, 𝛼𝛽 or 𝑑𝑞, via Clarke or Park transform, respectively. Say
always four vectors that work for both 𝑣⃗'AA∗ and 𝚤⃗(∗ , in both
that 𝛼𝛽-phase representation, determined relative to a stationary
directions. In the example illustrated in Fig.4 these are: ±1, ±3,
frame, is used for both sides.
±4, ±6. Vectors of both signs are kept at this stage because
The load side model, with parameters 𝑅+ , 𝐿+ , is given by:
vectors of both directions may need to be utilized.
)
The objective is now to find the dwell times (𝛿 ; , 𝛿 ;; , 𝛿 ;;; , 𝛿 ;6 ), R((& ) U
= −(𝑅+ ⁄𝐿+ )𝑖' + (1⁄𝐿+ )𝑣'
U
𝜙 ∈ {𝛼, 𝛽} (9)
for the four vectors (e.g. ±1, ±3, ±4, ±6,) such that, on average, RH

they reproduce the desired 𝑣⃗'AA∗ and 𝚤⃗(∗ . This requires solving a

SBa SAa SCa SAa SBa

a
b SBb SAb SCb SAb SBb
c SBc SAc SCc SAc SBc
0B +1 -4 0A +6 -3 0C 0C -3 +6 0A -4 +1 0B

δ0B/2 δI/2 δIII/2 δ0A/2 δIV/2 δII/2 δ0C/2 δ0C/2 δII/2 δIV/2 δ0A/2 δIII/2 δI/2 δ0B/2
Ts/2 Ts/2
Fig. 5 Example SVM pattern when both references 𝑣⃗'AA∗ and 𝚤⃗(∗ are in sector 1

4
/20/$31.00 © 2022 IEEE
Page 5 of 10 2022-PCIC-0540

A discrete-time equivalent of model (9) can be obtained by vs vi ii vo io

using standard Euler discretization as: Grid Input Matrix
Load
Filter Converter
U (𝑘 U (𝑘) U (𝑘)
𝑖 + 1) = 𝐴' 𝑖 + 𝐵' 𝑣 𝜙 ∈ {𝛼, 𝛽} (10)
ωr
* , * ,
$ .+ - 3 $ .+ - SVM
where 𝐴' = 𝑒 + ; 𝐵' =
V+
c1 − 𝑒 + d;
Δφf αo
𝑇% is sampling period. vo* Feedback
βi |vo*| control
The supply side model often assumes an ideal AC voltage
U U αs
source 𝑣( . This is because the MC input voltage 𝑣( is very close
U (a) Conventional control: PI at load side with PWM
to the supply voltage 𝑣% . However, the presence of the input RLC
filter (with parameters 𝑅, , 𝐿, , 𝐶, ) adds resonant behaviour. vs vi ii vo io
Control schemes that aim to suppress the input filter resonance, Grid Input Matrix Load
include its full model. If 𝑅, and 𝐿, are connected in series, then Filter Converter
the supply side model is given by ωr
U SVM
R 𝑖% 𝑖U 𝑣%U
RH
f Ug = 𝐴( f %U g + 𝐵( f Ug 𝜙 ∈ {𝛼, 𝛽} (11) ii* β i αo
𝑣( 𝑣( 𝑖( ii* vo* Feedback
αs
3 3 3 3 pi=po control
−V −A
/ 2/ / A/ V/ 2/
where 𝐴( = h 3 i; 𝐵( = h 3 i.
0 0 −2 (b) Conventional control shifted to input side
2/ /

If 𝑅, and 𝐿, are connected in parallel, then the model structure is vc ii vo io

Grid Input Matrix Load
is the same as in (11) but the definitions are: Filter Converter
3 3 3 3
−V 2 −A A/ V/ 2/
ωr
/ / /
𝐴( = h 3 i; 𝐵( = h 3i;
0 0 −2 Model predict.
2/ /
p p
->ioα, ioβ
U
and the supply voltage 𝑣% is replaced by a modified supply Speed Control
U pi=po ->ioα
* ,i *
oβ
voltage 𝑣%* , which is scaled by k1 + (𝜔( 𝐿, /𝑅, ) and phase
shifted by a positive angle 𝜑%* = tan$3 (𝜔( 𝐿, /𝑅, ) relative to the (c) Model Predictive Control with cost function
U
supply voltage 𝑣% .
Fig. 6 Common control schemes for matrix converter
A discrete-time equivalent of model (11) is given by

𝑖%U (𝑘 + 1) 𝑖%U (𝑘) 𝑣%U (𝑘) Conventional MC control scheme [18] is illustrated in
f U g = 𝐴R f U g + 𝐵R f U g 𝜙 ∈ {𝛼, 𝛽} (12) Fig.6(a). The load side is controlled in a similar way to a standard
𝑣( (𝑘 + 1) 𝑣( (𝑘) 𝑖( (𝑘) inverter via closed-loop current control. This control (for example,
where 𝐴R = 𝑒 !%4- ; 𝐵0 = 𝐴1 (𝑒𝐴𝑖𝑇𝑠 − 𝐼)𝐵1 ;
23 of PI or PR type) determines the desired output voltage, in both
magnitude 𝑣'∗ and angle 𝛼' . The remaining degree of freedom
I is identity matrix. (angle 𝛽( ) can be used to achieve the desired displacement
power factor angle 𝜑( at the input side.
C. Control Strategies for Direct Matrix Converter For example, using the RLC filter model given by (11),
expressed in Fourier domain under sinusoidal steady state, it can
In section II-A it was understood that both 𝑣⃗'AA and 𝚤⃗( cannot be shown that (note that 𝑥 hereafter denotes a phasor):
be controlled independently, due to their interconnection via
instantaneous power balance. When the power balance 𝑖( = 𝑖% 11 − 𝜔- 𝐿, 𝐶, 6 − 𝑗𝜔𝐶, (𝑣% − 𝑅, 𝑖% ) (13)
condition is imposed, it is possible to obtain magnitude and
direction of one of these references, and the direction of the where phasor 𝑣% = 𝑣% ∠0 is taken as reference, and 𝑖% = 𝑖% ∠0
other. Then the following three options can be explored: to satisfy the alignment condition for the unity power factor.
- full control over the output voltage space vector (by From (13), the angle ∆𝜑, of the phasor 𝑖( relative to 𝑣% at the
magnitude 𝑣' and direction 𝛼' ) and partial control over the input fundamental frequency 𝜔( can be determined as
current space vector (by direction 𝛽( only); Y% 2/ (Q- $V/ (- )
- full control over the input current space vector (by ∆𝜑, = tan$3 r ( ! s (14)
- Z3$Y% A/ 2/ [
magnitude 𝑖( and direction 𝛽( ) and partial control over the output
voltage space vector (by direction 𝛼' only); Subtracting ∆𝜑, from the supply voltage angle 𝛼% generated by
phase lock loop (PLL), as shown in Fig.6(a), yields an estimation
- full control of both quantities is pursued but these of 𝛽( such that it corresponds to the unity power factor at the
objectives are weighted against each other in a cost function. supply. Magnitude 𝑣% used in (14) is provided by the
The corresponding three control schemes are shown in Fig.6. measurement, and 𝑖% can be found by solving a quadratic

5
/20/$31.00 © 2022 IEEE
 2022-PCIC-0540 Page 6 of 10

equation corresponding to the power balance: is vc ii vo io

∗ Grid Input Matrix Load
∗
𝑣% 𝑖% − 𝑅, 𝑖%- = 𝑝( = 𝑝' = 𝑅𝑒 t𝑣' ∙ 𝑖( v (15) Filter Converter
∗ ∗
where instantaneous output power 𝑝' = 𝑅𝑒 t𝑣' ∙ 𝑖( v is provided ωr
by the load side control. SVM
ii * βi q αo
An alternative form of the conventional control is shown in vo* Observer
Fig.6(b). In this scheme the reference to SVM comes in the form vs, is Reference generator
of 𝑖(∗ , 𝛽( and 𝛼' , where 𝑖(∗ is calculated from 𝑣'∗ using the power vo, io Dynamic
vc, ii p ) control
balance equation [19]. The advantage is that ad-hoc power factor
correction and/or active damping can be then included in the 𝑖(∗ Fig.7 Comprehensive control for matrix converter
reference generation. This control option, due its limited value, is
not explored in the sequel. All the above issues are overcome in the recently proposed
Cost function approach illustrated in Fig.6(c) is typically Comprehensive control of matrix converters [22], which is
associated with Model Predictive Control (MPC) [20]. Under this illustrated in Fig.7. Its main features include the following.
approach, each control objective is represented by a respective - Both load and supply side models are complemented
term in the total cost function. Since the objective of the load side by disturbance models. State variables and disturbances on both
∗ sides are estimated using Observers.
control is usually to track a given current reference 𝑖' then the
relevant cost can be expressed as the Euclidean distance of the - State variables are driven to their respective references
∗ using Dynamic control based on state feedback. The control is
predicted output current to the reference 𝑖' . Current predictions
B designed, on the load side - to give the desired dynamics, on the
𝑖' can be determined, using model (10), as: supply side - to suppress the input filter resonance and
UB U
𝑖',0 (𝑘 + 1) = 𝑎' 𝑖' (𝑘) + 𝑏' 𝑣',0 (𝑘)
U
𝜙 ∈ {𝛼, 𝛽} (16) harmonics.
- References for the state feedback control are derived
U
where 𝑣',0 (𝑗 = 1 … 𝑁) is output voltage corresponding to each by Reference generator from the power balance equations, given
of the available switching states listed in Appendix A; the power factor. The estimated disturbances are included in
UB these calculations, which provides immunity against parameters
𝑖',0 are the corresponding output current predictions. uncertainties and measurement errors.
To account for the objectives of resonance and harmonic - The control inputs, 𝑣' and 𝑖( , are forced to satisfy the
suppression, and power factor control, at the supply side, a instantaneous power balance. Based on these inputs, the
sinusoidal reference can be imposed on the supply current [21]. modulation index 𝑞 and the angles 𝛼' , 𝛽( that satisfy both side
Its magnitude can be found from the power balance equation (15) objectives, are calculated and commanded to SVM.
and its angle equals to that of 𝑣% , for the unity power factor.
∗ In mining applications, control of matrix converters faces
For the purpose of comparison with the desired current 𝑖% , the additional challenges, associated with power factor control and
B
predicted current 𝑖% can be found by using model (12), as: limitations on the sampling frequency. These challenges can be
UB (3,3) U (3,-) U (3) U U addressed by providing an optimal power factor under all
𝑖%,0 (𝑘 + 1) = 𝐴R 𝑖% (𝑘) + 𝐴R 𝑣( + 𝐵R 𝑣% + 𝑏' 𝑖(,0 (17) operating conditions [23] and by utilizing accurate modelling of
U system behaviour inside a sampling interval [24].
where 𝑖(,0 (𝑗 = 1 … 𝑁) is the MC input current corresponding to
each of the available switching states listed in Appendix A. III. SIMULATION RESULTS AND DISCUSSION
The total cost function for the 𝑗-th switching state is given by:
LB - OB O∗ - A simulation model was developed of a matrix converter
𝑔0 = 1𝑖',0 (𝑘 + 1) − 𝑖'L∗ 6 + r𝑖',0 (𝑘 + 1) − 𝑖' s
(18) driving a load that emulates an AC motor. The following
- -
LB OB
+𝜆[1𝑖%,0 (𝑘 + 1) − 𝑖%L∗ 6 + r𝑖%,0 (𝑘 + 1) − 𝑖% s ]
O∗ parameters representative of a VSD application were used in the
simulations. Load: 𝑅+ = 20Ω; 𝐿+ = 52.3mH. Input Filter: 𝐶, = 14.2µF;
where 𝜆 is weighting factor that determines the trade-off 𝐿, = 2.5mH; 𝑅, = 2.5Ω (when in series) or 𝑅, = 6.8Ω (when in
between the input and the output control objectives. parallel). Frequencies: 𝑓( = 50Hz, 𝑓' = 68Hz; 𝑓]^% = 844.7Hz.
Then the switching state that minimizes cost function (18) is The control options studied by simulation, included:
selected and applied to the power converter. conventional control, Model Predictive control and
While the cost function approach addresses both side control Comprehensive control. For fair comparison, the average
objectives simultaneously, it is not immune to a potential conflict switching frequency (per switch), which is related to switching
between these objectives. Such a conflict can arise from losses, was maintained equal. Consequently, conventional and
inaccuracies of the underlying assumptions. Comprehensive schemes that utilize SVM were run at the
For example, in practice, real powers 𝑝' and 𝑝( are not exactly sampling frequency of 𝑓% = 10.417kHz (𝑇% = 96µs), while the MPC
matched due to losses; the input filter parameters 𝑅, , 𝐿, , 𝐶, scheme was run 12 times faster, at 𝑓% = 125kHz (𝑇% = 8µs).
values may be known imprecisely or vary with time; supply It can be noted that performing all the MPC calculations within
voltage 𝑣% may be measured with an error or not measured at all, 𝑇% = 8µs is not feasible with the existing real-time processors.
etc. An additional issue is the selection of weighting 𝜆 and the Therefore, an alternative MPC option was also implemented,
need for its adjustment during operation. namely, a Modulated MPC, or MMPC [25].

6
/20/$31.00 © 2022 IEEE
Page 7 of 10 2022-PCIC-0540

(a) Conventional control with SVM at 𝑇& = 96𝜇𝑠 (b) Model Predictive Control at 𝑇& = 8𝜇𝑠

(c) Modulated MPC at 𝑇& = 96𝜇𝑠 (d) Comprehensive control with SVM at 𝑇& = 96𝜇𝑠

Fig. 8 Simulation results: comparison of control schemes for matrix converter

Under MMPC, cost functions for all states are evaluated Following that, the selected switching states are arranged in a
according to (18), and the N states with the lowest cost function symmetrical pattern similar to SVM, as per Fig.5.
values 𝑔( (i = 0...N) are selected, including the zero state. Dwell Simulation results for the four schemes (including MPC and
times 𝛿( of these states are assumed inversely proportional to MMPC) are presented in Fig.8. In all four cases, the load side
their respective costs 𝑔( , and should sum up to 1. The following reference undergoes a step change from 2A to 3A at t = 0.03s
logic is then applied [26]: and back to 2A at t = 0.06s.
3 I 3 I
Under conventional control (see Fig.8(a)), each step change
∑_ _
(`a 𝛿( = 1 ; 𝑘 = 1•∑(`a ; 𝛿( = = •∑_
(`a (19) is accompanied by resonant oscillations in the supply current (𝑖% ),
b% b% b% b%
real power (𝑃% ) and reactive power (𝑄% ). On the stepdown, the

7
/20/$31.00 © 2022 IEEE
 2022-PCIC-0540 Page 8 of 10

load current (𝑖' ) takes longer time to reach the reference (𝑖'∗ ). of high power grinding mill drives in mining applications,” in
These are the consequences of insufficient passive damping (by Industry Applications Conference, 2003. 38th IAS Annual
𝑅, ) and no active damping (by the control action). Steady state Meeting. Conference Record of the, vol. 3, 2003, pp. 1629–
tracking performance is very good (𝑖' appears practically on top 1636.
of 𝑖'∗ ). [6] G. Francisco Silva, T. Luis Moran, T. Miguel Torres, and V.
Under MPC (see Fig.8(b)), the transient oscillation is still Christian Weishaupt, “A method to evaluate
present at each step change but is quickly damped. The steady cycloconverters commutation robustness under voltage
state tracking is very good. Under the more feasible MMPC and frequency variations in mining distribution systems,”
option (see Fig.8(c)), transient oscillations appear to be damped IEEE Transactions on Industry Applications, vol. 54, no. 1,
better but the steady state tracking is poor. pp. 858–865, 2018.
Finally, under Comprehensive control, transient oscillations [7] T. Friedli and J. W. Kolar, “Milestones in matrix converter
are instantly and effective damped. Tracking performance is research,” Industrial Applications, IEEJ Journal of, vol. 1,
excellent. Power factor remains exactly at unity throughout the pp. 2–14, 2012.
simulation. This can be confirmed by the observation that the [8] L. Gyugi and B. Pelly, Static Power Frequency Changers:
supply side reactive power 𝑄% = 0. Theory, Performance and Applications. New York: Wiley,
Based on the presented comparison, matrix converter under 1976.
Comprehensive control can be recommended as a suitable [9] A. Alesina and M. Venturini, “Solid-state power conversion:
AC/AC power converter option for demanding industry A Fourier analysis approach to generalized transformer
environments, such as mining. synthesis,” Circuits and Systems, IEEE Transactions on,
vol. 28, no. 4, pp. 319–330, 1981.
IV. CONCLUSIONS [10] A. Alesina and M. Venturini, “Analysis and design of
optimum-amplitude nine-switch direct AC-ac converters,”
The paper has presented a thorough review of the history, Power Electronics, IEEE Transactions on, vol. 4, no. 1, pp.
main features and techniques, as well as the recent 101–112, 1989.
developments, around the matrix converter modulation and [11] L. Huber and D. Borojevic, “Space vector modulated three-
control. It has also included the relevant theory that assists in in- phase to three-phase matrix converter with input power
depth understanding of this type of power converter. factor correction,” Industry Applications, IEEE Transactions
The paper has discussed challenges associated with the tight on, vol. 31, no. 6, pp. 1234– 1246, 1995.
interconnection between the MC input and output sides. It [12] D. Casadei, G. Serra, A. Tani, and L. Zarri, “Matrix
showed how these challenges have been addressed by converter modulation strategies: a new general approach
traditional and more advanced, modern control, approaches. based on space-vector representation of the switch state,”
Detailed simulations are included in the paper, to illustrate Industrial Electronics, IEEE Transactions on, vol. 49, no. 2,
performance of different control schemes and draw a comparison pp. 370–381, 2002.
between them. The paper has shown that the recently proposed [13] J. Zhang, H. Yang, T. Wang, L. Li, D. G. Dorrell, and D. D.
Comprehensive control is able to meet all the control objectives Lu, “Field-oriented control based on hysteresis band
in an effective and mutually non-contradicting way. current controller for a permanent magnet synchronous
The Comprehensive control scheme, enhanced by power motor driven by a direct matrix converter,” IET Power
factor optimization and other embellishments, can be Electronics, vol. 11, no. 7, pp. 1277–1285, 2018.
recommended for the most demanding industry applications, [14] P. Vas, Sensorless Vector and Direct Torque Control.
including but not limited to, mining. Oxford University Press, 2003.
[15] H.-N. Nguyen, M.-K. Nguyen, V.-Q.-B. Ngo, T.-T. Tran, J.-
H. Choi, and Y.-C. Lim, “Input power factor compensation
V. REFERENCES strategy for zero CMVSVM method in matrix converters,”
IEEE Access, vol. 8, pp. 175805– 175814, 2020.
[1] J. V. Pulgar, M. A. Valenzuela, and C. Molina Vicuna, [16] Y. Li, L. Qiu, Y. Zhi, G. Yan, J. Zhang, J. Ma, and Y. Fang,
“Correlation between power and lifters forces in grinding “An overmodulation strategy for matrix converter under
mills,” IEEE Transactions on Industry Applications, vol. 55, unbalanced input voltages,” IEEE Access, vol. 9, pp. 2345–
no. 4, pp. 4417–4427, 2019. 2356, 2021.
[2] G. Mirzaeva, D. Carter, M. M. Seron, and G. C. Goodwin, [17] J. D. Dasika and M. Saeedifard, “An online modulation
“Matrix converters with input resonance suppression for strategy to control the matrix converter under unbalanced
electric haulage vehicles,” IEEE Transactions on Industry input conditions,” IEEE Transactions on Power Electronics,
Applications, vol. 57, no. 5, pp. 5527–5536, 2021. vol. 30, no. 8, pp. 4423–4436, 2015.
[3] M. Ramezani and S. Tafazoli, “Using artificial intelligence in [18] F. Blaabjerg, D. Casadei, C. Klumpner, and M. Matteini,
mining excavators: Automating routine operational “Comparison of two current modulation strategies for matrix
decisions,” IEEE Industrial Electronics Magazine, vol. 15, converters under unbalanced input voltage conditions,”
no. 1, pp. 6–11, 2021. Industrial Electronics, IEEE Transactions on, vol. 49, no. 2,
[4] R. Zhu and M. Liserre, “Lifetime estimation of dc-link pp. 289–296, 2002.
electrolytic capacitor for smart transformer lv side inverter,” [19] H. M. Nguyen, H. H. Lee, and T. W. Chun, “Input power
in 2019 IEEE Energy Conversion Congress and Exposition factor compensation algorithms using a new direct-SVM
(ECCE), 2019, pp. 4174–4180. method for matrix converter,” Industrial Electronics, IEEE
[5] J. Rodriguez, J. Pontt, P. Newman, L. Moran, and G. Transactions on, vol. 58, no. 1, pp. 232–243, 2011.
Alzamora, “Technical evaluation and practical experience

8
/20/$31.00 © 2022 IEEE
Page 9 of 10 2022-PCIC-0540

[20] J. Lei, S. Feng, P. Wheeler, B. Zhou, and J. Zhao, “Steady- [26] B. Sanabria, S. Toledo, D. Caballero, E. Maqueda, J.
state error suppression and simplified implementation of Rodas, R. Gregor, and M. Rivera, “Modulated predictive
direct source current control for matrix converter with model voltage control of a direct matrix converter,” in 2020 IEEE
predictive control,” Power Electronics, IEEE Transactions Congreso Bienal de Argentina (ARGENCON), 2020, pp. 1–
on, pp. 3183–3194, 2020. 7.
[21] M. Rivera, J. Rodriguez, P. W. Wheeler, C. A. Rojas, A.
Wilson, and J. R. Espinoza, “Control of a matrix converter VI. VITAE
with imposed sinusoidal source currents,” Industrial
Electronics, IEEE Transactions on, vol. 59, no. 4, pp. 1939– Galina Mirzaeva received BEng degree in electronic
1949, 2012. engineering in 1990 and PhD in electrical engineering in 1997
[22] G. Mirzaeva, M. Seron, and G. Goodwin, “Novel from the South Urals State University, Russia.
comprehensive control of matrix converters,” in 2021 IEEE From 2004 to 2010, she worked at CRC-Mining, Australia.
Energy Conversion Congress and Exposition (ECCE), Since 2010, she has been with the School of Engineering at the
2021, pp. 2432–2438. University of Newcastle, Australia, first as Senior Lecturer and
[23] G. Mirzaeva, M. Seron, and D. Carter, “Grid-side power from 2017 - as Associate Professor.
factor optimisation for matrix converters in mobile mining Her research interests include power electronics, electric
vehicle applications,” in 2021 IEEE Industry Applications drives, renewable energy integration and electric transportation.
Society Annual Meeting (IAS), 2021, pp. 1–6. Dr Mirzaeva currently serves as Chair of the IEEE IAS Mining
[24] G. Mirzaeva, M. Seron, and D. Carter, “Advanced hybrid Industry Committee.
models for control of matrix converters in mining vehicle
applications,” in 2021 IEEE Industry Applications Society Yuan Liu received BEng degree in electrical engineering from
Annual Meeting (IAS), 2021, pp. 1–6. Shanghai University of Electric Power, China, in 2017, and MS
[25] L. Tarisciotti, J. Lei, A. Formentini, A. Trentin, P. Zanchetta, degree in electrical engineering from the University of New South
P. Wheeler, and M. Rivera, “Modulated predictive control Wales, Australia, in 2020. He is currently pursuing a research
for indirect matrix converter,” IEEE Transactions on degree at The University of Newcastle, Australia.
Industry Applications, vol. 53, no. 5, pp. 4644–4654, 2017. His research interests include power converters and
renewable energy integration.

9
/20/$31.00 © 2022 IEEE
 2022-PCIC-0540 Page 10 of 10

APPENDIX A
MATRIX CONVERTER SWITCHING STATES
Output voltages Input currents CMV
V# State
va vb vc vab vbc vca voph αph
o voLL αLL
o iA iB iC ii βi Vcm

2 √2 vAB π 2
+1 vA v B vB vAB 0 -vAB v
3 AB
0 6
ia -ia 0 √ ia - π6 1
v
3 BC
3 3

2 √2 vAB
-1 v B vA v A -vAB 0 vAB v
3 AB
π - 56
π
-ia ia 0 2
√ ia 5π
6
-1v
3 CA
3 3

2 2 π 2 π 1
+2 vB v C v C vBC 0 -vBC v
3 BC
0 √ v 0 ia -ia √ ia v
3 BC 6 3 2 3 CA

2 2
-2 vC vB vB -vBC 0 vBC v
3 BC
π √ v - 56
π
0 -ia ia 2
√ ia - π2 -1v
3 BC 3 3 AB

2 √2 vCA π 2
+3 vC vA vA vCA 0 -vCA v
3 CA
0 6
-ia 0 ia √ ia - 56
π 1
v
3 AB
3 3

2 √2 vCA
-3 v A vC vC -vCA 0 vCA v
3 CA
π - 56
π
ia 0 -ia 2
√ ia π
6
-1v
3 BC
3 3

2 2π √2 vAB 5π √2 ib
+4 vB v A vB -vAB vAB 0 v
3 AB 3 6
ib -ib 0 - π6 1
v
3 BC
3 3

2
-4 v A vB v A vAB -vAB 0 v
3 AB
- π3 √2 vAB - π6 -ib ib 0 √2 ib 5π
6
-1v
3 CA
3 3

2 2π 2 5π √2 ib π 1
+5 vC vB v C -vBC vBC 0 v
3 BC
√ v 0 ib -ib v
3 3 BC 6 3 2 3 CA

2
-5 v B vC v B vBC -vBC 0 v
3 BC
- π3 2
√ v - π6 0 -ib ib √2 ib - π2 -1v
3 BC 3 3 AB

2 2π √2 vCA 5π √2 ib
+6 v A vC v A -vCA vCA 0 v
3 CA 3 6
-ib 0 ib - 56
π 1
v
3 AB
3 3

2
-6 v C v A vC vCA -vCA 0 v
3 CA
- π3 √2 vCA - π6 ib 0 -ib √2 ib π
6
-1v
3 BC
3 3

2
+7 vB v B vA 0 -vAB vAB v
3 AB
- 23
π √2 vAB - π2 ic -ic 0 2
√ ic - π6 1
v
3 BC
3 3

2 π √2 vAB π 2 5π
-7 v A vA v B 0 vAB -vAB v
3 AB 3 2
-ic ic 0 √ ic 6
-1v
3 CA
3 3

2
+8 vC vC vB 0 -vBC vBC v
3 BC
- 23
π 2
√ v - π2 0 ic -ic 2
√ ic π 1
v
3 BC 3 2 3 CA

2 π 2 π 2
-8 v B vB vC 0 vBC -vBC v
3 BC
√ v 0 -ic ic √ ic - π2 -1v
3 3 BC 2 3 3 AB

2
+9 v A vA vC 0 -vCA vCA v
3 CA
- 23
π √2 vCA - π2 -ic 0 ic 2
√ ic - 56
π 1
v
3 AB
3 3

2 π √2 vCA π 2 π
-9 vC vC vA 0 vCA -vCA v
3 CA 3 2
ic 0 -ic √ ic 6
-1v
3 BC
3 3

1
0a vA v A vA 000 0 × 0 × 000 0 × v
3 A

1
0b v B vB v B 000 0 × 0 × 000 0 × v
3 B

1
0c vC vC v C 000 0 × 0 × 000 0 × v
3 C
√
R1 vA vB v C vAB vBC vCA viph αph
i 3viph αph π
i +6 ia ib ic io βo 0
√
R2 vA vC vB -vCA -vBC -vAB viph -αph
i 3viph -αph π
i +6 ia ic ib io -βo 0
√
R3 vB v A v C -vAB -vCA -vBC viph -αph 2π
i + 3 3viph -αph 5π
i + 6 ib ia ic io -βo + 23
π
0
√
R4 vB v C vA vBC vCA vAB viph αph
i − 2π
3
3viph αph
i − π
2
ic ia ib io βo + 23
π
0
√
R5 vC vA vB vCA vAB vBC viph αph 2π
i + 3 3viph αph 5π
i + 6 ib ic ia io βo − 2π
3
0
√
R6 vC vB vA -vAB -vBC -vCA viph -αph
i −
2π
3
3viph -αph
i −
π
2
ic ib ia io -βo − 2π
3
0
Selected calculation examples
! " ! "
AB ! 6
+1: ⃗voLL = 2 v AB − a 2v
AB = 2
v AB 1 − a 2 = √2 v π ⃗ii = 2 (ia − aia ) = 2ia (1 − a) = √2 ia !- π
3 3
! π 2π3 " ! 3 " 3
! " 3 !6 "
+9: ⃗vo = - 3vCA (1 − a) a = √ vCA ! - 6+ 3 -π
LL 2 2 ⃗
ii = 2 3
-ic + a2ic = - 2
3c
i 1 − a2 = √2 ic ! π6-π
! " 3 ! " ! " ! 3 "
R5: ⃗voph = 2 3 C
v + avA + a2vB = 2 a a2vC + vA + avB R6: ⃗ii = 2 ic + aib + a2ia = 2a2 aic + a2ib + ia
3!
ph
3 3 "
= a⃗vi,pos = viph !αph
i + 3
2π
= a2⃗io,neg = io ! -βo - 23
π

10
/20/$31.00 © 2022 IEEE