38 38iEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. A1 -13, NO.

1, JANUARY/FEBRUARY 1977

Optimal Pulsewidth Modulation for

Feeding AC Motors
GIUSEPPE S. BUJA AND GIOVANNI B INDRI MLMBER IEEE

Abstract-A pulsewidth modulation (PWM) technique is proposed, The determination of the switching patterns, as a function
intended particularly for the reduction of the harmful effects of the of the value of the first voltage harmonic and for a given
harmonics in the inverter-fed drive systems. The approach is based on numnber of pulses per period, can be implemiiented directly by a
the minimization of the rms value of the current harmonics in the
motor by a proper choice of the commutation angles. Switching pat- computer. In this study, comilpuitations have beeni performed
terns are calculated for a different number of commutations per period, for the single-phase systeml and for the three-phase system
for both single-phase and three-phase half-bridge inverters. In the case with no common neutral. The expression of the performance
of the single-phase, an analytical procedure is developed which sim- index, which in general is given1 as a trigonometric function of
plifies the computations. Comparisons are made of the results thus the comnmutation angles, can be expressed in analytical form
obtained with those given by other PWM techniques in use.
for the case of the single phase. This makes the calculation
easier and reduces the computationi timles.
1. INTRODUCTION
11. STATEMENT OF THE PROBLEM
ULSEWIDTH modulation (,PWM ) techniques are used in
forced commutated inverters feeding variable-speed ac A periodic waveformn as shown in Fig. 1 is assumed for the
motor drive systems to control the amplitude of the first phase voltage v(a) of the inverter. It consists of a succession of
harmonic of the output voltage. They also permit the reductior pulses of amplitude ±VB/I. The angles ai, with i 1, 2, 3,
of the harmful effects on the motor of other harmonics such M satisfy the following condition.
as additional losses and pulsatinig torques.
Many techniques of PWM are available today [11; thost 0 <-a1-<.t2a < (3. < - °.31 7a/j
< . < ( )
most used are the suboscillation and eliiimination techniques. The waveformii of Fig. 1 has obviously only odd harmonics.
The first technique, in which the commutation angles are Their phase is equal to 00 or 1800, while their peak values are
set at the crossings of two waveforms, is characterized by the aiven by
simple implementation of the inverter control apparatus [21
several commnutations, however, are required to keep down the n
4 s2
harmonic effects [3]. (2)
In the elimination technique, the commnnutation angles are
Vn =-J
7r
v(oz) sin nat doz.
chosen in a straightforward manner, according to the criterion
of eliminating a number of harimionics [41] [5]. The control From (2) and Fig. I there is
equipment is complex, but satisfactory results are obtained M
even with limited numllber of coimmutationis [3j.
In this paper a different approach is suggested which
Vn = S(-1rl nIT
niT
I1-n2
1_
i:
(-I )i+l1COS notii (3)

consists of
where s assumes +l(-l) value if l'(a) has respectively +VB/2,
1) defining a perfornance inidex related to the unwanted or -VB12, value betweein aM and 7r/2. In the following the
effects of the voltage harmonics, two cases are called +(-) modulation, which corresponds to
) achieving its minimization by a suitable choice of the s +1(-l ). =

switching patterns, i.e., of the cotmtmiutation angles. Equation (3) with n- I gives the first harmonic amplitude
The rms value of current harmoinics on the motor is
assumed as a suitable performance index. The additional losses
in the copper windings, if the skin effect can be neglected, are 7r1
2
V, - A- I)" ~ [1 - 2 i (-1 i+1 cos °i], (4)
in fact proportional to the square of this value. This performance
index also gives an approximate estimate of the pulsating which varies between
torques, particularly if they are snmall and if the flux distribu-
tion at the air gap of the motor canl be assumed to be sinusoidal. V 0 1
VB
<4
7i
(5)

Paper TOD-76-5, approved by the Static Power Converter Com- The upper limit corresponds to the rectangular or nonmodulated
mittee of the IEEE Industry Applications Society. Manuscript released wave.
for publication in this TRANSACTIONS October 1, 1976.
The authors are with the Istituto di Elettrotecnica ed Elettronlica, Any desired value of V1 can be obtained by choosing
Universita di Padova, Padova, Italy. suitable values for the M angles in order to satisfy (4). The
BUJA AND INDRI: PULSEWIDTH MODULATION 39

+~VB2

VB
2

I
Fi.e.Phsevltgeo
d
P
C!S
C U
ft
ert
:C
IV
e
t~
is DC suppy
X'
C!
j

d-
t: t"
X-
tF +t
f i verT

g t+
a
AS C3
t+
d
i
v is
:S C!
unda
CSi C!
2Jr 1a" 0t

Fig. 1. Phase voltage of PWM inverter. VB is DC supply of inverter and wo is fundamental angular frequency.

problem exhibits M - 1 degrees of freedom, which are used to sufficient accuracy. Therefore, a different expression of a is
minimize Io, the rms value of the harmonic currents in the required.
motor.
Assuming that these currents are determined only by the Single-Phase Case
leakage inductance and that this is linear, Ig is given by Let us consider first the case of a single-phase half-bridge
inverter, so that the load voltage coincides with the phase
voltage of the inverter.
I = - , (6) The value i0(a) of the current harmonics as a function of
a, is given by
where
i(a°) = cL| [v(a) - V1 sin a] dao
VJ, peak value of the harmonic n of the output voltage,
L leakage inductance of the motor,
wo fundamental angular frequency.
By setting (10)
I= v(o) dot + V, cos ag .
woL -ir
VB/2
coL (7) The lower extreme of integration was set at rr/2 to avoid a
dc term in the solution.
22 13
I', the rms value of i0, is equal to
Vn 2
VB 2 3\n
(8)
I0= woLV27r /2X1 v(a) da - V1 cos a]d
(6) can be written in the form Y2

= Ka. (9) - 2 V 2
I I_ |/v(at)dat --2
dot (I 1)
It can be noted that a is adimensional and depends only on
cooL 9\ 2r /27r I
L-i
the values of a1, a2, --, aM, while K is independent of them.
For each given pair M, V1, the minimization of Ia coincides By comparing equations (I 1) and (7) we have
with that of a. The values of a1, a2, , aeM obtained by the
condition a = min, together with (1) and (4) represent the a=2 1 r a ol2 pi2
required switching patterns. The value of s must also be
specified, as different results are obtained if s = +1 or s =-1.
VB
- I
[|
v(a) dao da
2-, (12)

III. RESULTS
where = 2VI/VB.
It can be shown [31 that the term
In practice, the use of (8) for evaluating a is not convenient 1 f [ _2
because V,n is a trigonometric function of the variables ai and
2]-v(aL
2ir27r}
2
) daj d
a large number of V,n must be computed in order to obtain -JiT -
40 IEEE TRANSACTIONS ON INDUSTRY APPLICA TIONS, JANUARY/FEBRUARY 1977

0.5 1.0
Fig. 2. amin2 versus VI for single-phase case. For co refer to (8) in text.

vi
Fig. 3. Modulation law for M 2, with reference to curve a+ of Fig. 2.

is an algebraic function of the commutation angle of the form The mninimizationi of g under the constraint (15) has been
carried out by means of a digital computer. The resulting
minima values of the square of a, called Umin2 are shown in
ir L (a) da da Fig. 2 (curves a± and a-) as a function of V1. The ±(-) signs
2r}2rVB J2-
_ t +2 2+J3+~A
placed after a refer to the +(---) modulation. An analogous
comnputation procedure can be applied for M > 2. The
= [+ 2 F j aM+1-i 3 + 7fr (_I)jaM+l_j2 computation times, however, increase rapidly with M. The
Omin2 values for M = 3 (curves b+ and b-) and for M 4
(curves c± and c-) are also given in Fig. 2, together with the
M M 1
-4y, z )j1 IM +1 _j . (_l)i+l UM_2 .(13) amin2 values for M = I (curves d+ and d-). The last two
1 i _
curves are shown only as references, since in this case the
commutation angle, and hence Umin2 are determined once
Equation (13) is the one we are looking for. Its algebraic VE is fixed.
form is well suited for the solution of the problem. The following considerations may be derived from the
Let us assume M - 2. Equations (12) and (4) become analysis of Fig. 2.

Ur - - + 2(a13 3 +
2) iT(alt2 -U2) 4U2a121 (14)

4
V1 = s-(1 -2 cos
iT
a,2 COS2). (15)
BUJA AND INDRI: PULSEWIDTH MODULATION 41

Fig. 4. Modulation law for M= 3, with reference to curve b+ of Fig. 2.

Fig. 5. Modulation law for M = 4, with reference to curve c+ of Fig. 2.

goo

(Xi+
60

30

0 0.5 1.0 V,
Fig. 6. Positive and negative modulation laws for M = 1.
42 IEEE TRANSACTIONS ON IN DUSTFRY APPLICA-FIONS, J AN lJ ARY/lEBRI.
RUY 1977

Fig. 7. cmin versus V1 for three-phase case.

90C a2

..._. _

,
...
- ........
~ ~ ......
30
............
.~~~~~~~~~~~~--.... ~ ~ ----..........
~ ~ ~ ~ ~ - ..

0.5 1,0 VI
Fig. 8. Modulation law for M = 2, with reference to curve a+ of Fig. 7.

a) As expected, ,min2 decreases asM increases. The drop is However, equations (7) to (I1) are still valid if reference is
more appreciable for low values of M. made to the line voltage instead of the phase voltage.
b) The plus modulation is more convenient than the minus The line voltage vl 2 is related to the phase voltage v, and
modulation. The benefit is lower for low values of PI and V2 by
decreases as M increases.
c) The curves converge at V1 4f1 to the Omin2 of the Vl 2 (a) =
VI (a) V2 (a), (16)
nonmodulated wave.
Point a) suggests that there is advantageous to increase M. which because of the assumed voltage symmetry iiay be
However, the commutation losses and the turn-off time of written in the formn
thyristors are to be taken into consideration when choosing
the values of M. Vl 2 (a)-v1 (a)- va - 27r/3 ). (17)
The switching angles in correspondence with the curves
a+, b+, c+, which are the more convenient as stated at point Equations (9), (10), and (1 1) become
b), and the angles in correspondence with the curves d+ and
d- are shown in Figs. 3-6, respectively as a function of V1.
Three-Phase Case ia(() = ~-L { [121(a) -- (cv
li - )]d
For a symmetrical three-phase system of PWM voltages
supplying a motor without common neutral, the above results
cannot be used, since the current harmonics of order multiples -
V,v cr)
COSI a + 6
(18)
of three are now zero.
BUJA AND INDRI: PULSEWIDTH MODULATION 43

(Xi
90

a2-
a
(1
60

30 _

0 0.5 1.0 V1
Fig. 9. Modulation law for M = 3, with reference to curve b- of Fig. 7.

a4
90
a3
a

60

30

10
0 0.5 1.0 V,
Fig. 10. Modulation law for M = 4, with reference to curve c+ of Fig. 7.

IU =
IO Of27r [V2I]j4 (a) - vJ - ) ] da| d a- - (19)

2 V12
V(aV) - V1( aldu
)] da (20)
VB \21r 7[2R 3rI -
VB2 1

where V1 is now the amplitude of the first harmonic of v12 a) Om in2 decreases as M increases; the drop is more
V1 is given by appreciable for low values of M.
b) The advantage of the + or modulation depends
-

V1 = s(l)1I 2V* [l -2 i (-1)+cos ai] (21) on both M and V1. For low values of PI the + modulation is
more advantageous if M is even, and the modulation if M is
-

In this case it is somewhat difficult to achieve for an a

odd.
algebraic function of the commutation angles. Therefore, the c) The curves converge at V1 = 4/ir to the amin2 of the
computation of (20) as well as its minimization has here been nonmodulated wave.
performed numerically. The waveshape of vl(a), which is As pointed out in b), the + and curves of Om in2 with the
-

piecewise constant, makes this computation easier. same M cross repeatedly. In order to keep a2 at minimum for
The results obtained are shown in Fig. 7. The curves a, b, c, each Vl, a few jumps between the + and modulation would
-

and d correspond to M equal to 2, 3, 4, and 1, respectively, be required.

while the + and signs refer to the + or modulation. The
- - Wishing to use the same modulation, it is best to choose the
curves d are shown again for comparison. In Fig. 7 V1 is equal modulation corresponding to the curves a+, b-, c+ of Fig. 7
to 2 Vl /3VB. because this is most advantageous at low values of V1, i.e., at
From the analysis of Fig. 7 the following considerations low rotational speeds of the motor. The switching angles are
may be derived. shown in Figs. 8, 9, and 10, respectively, as a function of V1.
44 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, JANUARY/FEBRUARY 1977

Some comnents may be given about the corner points here assumed is the rms value of the current harmonics in the
which appear in the curves of Fig. 7 and which correspond to motor. Of course, different and more sophisticated indices
the discontinuities of the commut4tion angles a1, a2, --, aM could be used without affecting the validity of the method.
of Figs. 8, 9, 10. These points arise from the crossing of two The calculation performed in some typical cases leads to
curves, each referring to a different set of the commutation significant results, even with a small number of commutations
angles which give a local minimum for a2. As an example, per period, which, in turn, implies low commutation losses in
in Fig. 7 the local minimum curves pertaining to the a+ the inverter.
modulation are indicated by dotted lines; the corresponding The comparison with other usual types of PWM show that
angles are shown in the same way in Fig. 8. The corner point noticeable improvements are obtained, not only in the
appears because the curve of the absolute minimum of a2, reduction of the current distortion, but also in the maximum
indicated by the continous line, follows the prevailing local value of the voltage supplied by the inverter.
minimum. The angle values given in the figures allow the practical
utilization of the proposed technique which, however, requires
a more complex electronic control system than that of the
IV. COMPARISON WITH OTHER TECHNIQUES OF PWM
suboscillation method.
The results obtained using the proposed technique will be
compared now with those of other PWM techniques. REFERENCES
First, the purpose of minimizing the current harmonics in [t] B. Mokrytzki, "Pulse width modulatated inverters for AC motor
the motor leads to appreciable improvements, as shown in drives," IEEE Trans. Ind. AppL, vol. IGA-3, pp. 493-503, Nov./
Figs. 2 and 7, in which the curves e and f for two usual PWM Dec. 1967.
[2] A. Schonung and H. Stemmler, "Reglage d'un motor triphase
types are also shown. reversible a I'aide d'un convertisseur statique de frequence com-
The e curves concern the suboscillation technique, where mande suivant le proc6d6 de la sous-oscillation," Revue Brown
the reference is a sine wave and the carrier is a triangular wave Boveri, vol. 51, pp. 555-576, Aouat/Sept. 1973.
with the frequency 9 times greater than that of the reference [3] G. Buja and G. Indri, "Invertitori a PWM per alimentazione di
motori: caratteristiche della tensione di uscita," Universita di
wave. The e curves must be compared with the c curves, Padova, Italy, Rapp. Int. UPee 05/75, Nov. 1975.
corresponding to M = 4, because the number of commutations [4] H. S. Patel and R. G. Hoft, "Generalized techniques of harmonic
elimination and voltage control in thyristor inverter: Part I-
per period is equal. Harmonic elimination," IEEE Trans. Ind. Appl., vol. IA-9, pp.
The f curves, in turn, concern the elimination technique for 310-317, May/June 1973.
M = 3. In the single-phase case of Fig. 2 the third and fifth [51 H. S. Patel and R. G. Hoft, "Generalized techniques of harmonic
harmonics have been eliminated, while in the three-phase case elimination and voltage control in thyristor inverter: Part II-
Voltage control techniques," IEEE Trans. Ind. Appl., vol. IA-10,
of Fig. 7, the fifth and seventh have been eliminated. The f pp. 666-673, Sept./Oct. 1974.
curves must be compared with the b curves, as M is the same.
Both for the e and f curves the comparison shows that the
improvement of o2 is significant, particularly for the higher Giuseppe S. Buja was born in Padua, Italy, on
values of V1. March 5, 1946. He received the Ph.D. degree
Another advantage of the suggested technique is that the in electronic engineering from the University of
maximum obtainable value of V1 is equal to 4/7r. Padua, Padua, Italy, in 1970.
Since 1971 he has been with the Institute of
The performance index (2 remains low up a value of V, Electrical and Electronic Engineering of Univer-
very near the maximum. This means that a large value of first sity of Padua, where he is engaged in the simu-
voltage harmonic can be supplied by the inverter without lation, analysis and control of static converter
and ac motor systems. Presently he is Associate
causing deterioration of the harmonic contents and with a Professor.
good utilization of the voltage ratings of the inverter. As Dr. Buja is member of A.E.I.
known with the suboscillation and elimination PWM types,
lower upper limits of V1 are allowed, and these limits are
attained with greater values of a2, i.e., of the current harmonics. Giovanni B. Indri (M'75) was born in Asolo
(Treviso) Italy, on March 2, 1943. He received
the Ph.D. degree in electrical engineering from
V. CONCLUSION the University of Padua, Padua, Italy, in 1967.
Since his graduation he has been with the
A PWM technique has been described which is suitable in Institute of Electrical and Electronic Engineer-
quasioptimal operating conditions for the motors with regard ing of the University of Padua as Assistant Pro-
fessor and from 1974 as Associate Professor
to the main unwanted effects of the harmonics: additional lecturing Power Electronics. His areas of
losses and pulsating torques. The approach is based on the interest are power electronic applications in
induction heating, in drives, and in U.P.S. and
selection of a suitable performance index and on its minimiza- digital simulation of nonlinear circuits.
tion by a proper choice of the switching pattern. The index Dr. Indri is a member of A.E.I. and of C.E.I.