2 SPACE VECTOR MODULATION FOR THREE-LEG VOLTAGE

SOURCE INVERTERS

2.1 THREE-LEG VOLTAGE SOURCE INVERTER

The topology of a three-leg voltage source inverter is shown in Fig. 2.1. Because

of the constraint that the input lines must never be shorted and the output current must

always be continuous a voltage source inverter can assume only eight distinct topologies.

These topologies are shown on Fig. 2.2. Six out of these eight topologies produce a non-

zero output voltage and are known as non-zero switching states and the remaining two

topologies produce zero output voltage and are known as zero switching states.

L Ia A
Vg a B
b C
c

Fig. 2.1. Topology of a three-leg voltage source inverter.

4
 p p
p

a a
a b b
b c c
c

n n
n V2(ppn) V3(npn)
V1(pnn)
p p p

a a a
b b c b c
c

n n n
V4(npp) V5(nnp) V6(pnp)

p p

a a
b c b c

n n
V7(ppp) V8(nnn)

Fig. 2.2. Eight switching state topologies of a voltage source inverter.

2.2 VOLTAGE SPACE VECTORS

Space vector modulation (SVM) for three-leg VSI is based on the representation

of the three phase quantities as vectors in a two-dimensional ( α , β ) plane. This has been

discussed in [1] and is illustrated here for the sake of completeness. Considering

topology 1 of Fig. 2.2, which is repeated in Fig. 2.3 (a) we see that the line voltages Vab,

Vbc, and Vca are given by

5
 Vab = Vg
Vbc = 0 ….. (2.1)
Vca = -Vg

This can be represented in the α , β plane as shown in Fig. 2.3(b), where voltages Vab,

Vbc, and Vca are three line voltage vectors displaced 120 ° in space. The effective voltage

vector generated by this topology is represented as V1(pnn) in Fig. 2.3(b). Here the

notation ‘pnn’ refers to the three legs/phases a,b,c being either connected to the positive

dc rail (p) or to the negative dc rail (n). Thus ‘pnn’ corresponds to ‘phase a’ being

connected to the positive dc rail and phases b and c being connected to the negative dc

rail
p

a
Vg b c

n
Fig. 2.3(a). Topology 1-V1(pnn) of a voltage source inverter.

Vbc
Vab = Vg
Vbc = 0 -Vca
Vca = -Vg -Vg

V1(pnn)
Vg
Vca Vab

Fig. 2.3(b). Representation of topology 1 in the α , β plane.

6
 Proceeding on similar lines the six non-zero voltage vectors (V1 - V6) can be

shown to assume the positions shown in Fig.2.4. The tips of these vectors form a regular

hexagon (dotted line in Fig. 2.4). We define the area enclosed by two adjacent vectors,

within the hexagon, as a sector. Thus there are six sectors numbered 1 - 6 in Fig. 2.4.

V3 V2

2
3 1
V4 V1
4 6
5

V5 V6

Fig. 2.4. Non-zero voltage vectors in the α , β plane.

Considering the last two topologies of Fig. 2.2 which are repeated in Fig. 2.5(a)

for the sake of convenience we see that the output line voltages generated by this

topology are given by

Vab = 0
Vbc = 0 ….. (2.2)
Vca = 0

These are represented as vectors which have zero magnitude and hence are

referred to as zero-switching state vectors or zero voltage vectors. They assume the

position at origin in the α , β plane as shown in Fig. 2.5(b). The vectors V1-V8 are called

the switching state vectors (SSVs).

7
 p p

a a
b c b c

n n
V7(ppp) V8(nnn)

Fig. 2.5(a). Zero output voltage topologies.

V bc

Vab = 0
Vbc = 0
Vca = 0
V7, V8

V ca V ab

Fig. 2.5(b). Representation of the zero voltage vectors in the α , β plane.

2.3 SPACE VECTOR MODULATION

The desired three phase voltages at the output of the inverter could be represented

by an equivalent vector V rotating in the counter clock wise direction as shown in Fig.

2.6(a). The magnitude of this vector is related to the magnitude of the output voltage (Fig.

2.6(b)) and the time this vector takes to complete one revolution is the same as the

fundamental time period of the output voltage.

8
 Vbc

V3 V2
2
V
3 1
V4 V1

4 6
Vca
5 Vab
V5 V6

Fig. 2.6(a). Output voltage vector in the α , β plane.

V ab V V bc V ca

1 2 3 4 5 6

Fig. 2.6(b). Output line voltages in time domain.

Let us consider the situation when the desired line-to-line output voltage vector V

is in sector 1 as shown in Fig. 2.7. This vector could be synthesized by the pulse-width-

modulation (PWM) of the two adjacent SSV’s V1(pnn) and V2 (ppn), the duty cycle of

each being d1 and d2, respectively, and the zero vector ( V7(nnn) / V8(ppp) ) of duty

cycle d0 [1]:

d1 V1 + d 2 V2 = V = m Vg e jè ….. (2.3)

d1 + d 2 + d 0 = 1 ….. (2.4)

where, 0 ≤ m ≤ 0.866, is the modulation index. This would correspond to a maximum

line-to-line voltage of 1.0Vg, which is 15% more than conventional sinusoidal PWM as

shown [1].

9
 V2 (ppn)

d2 V
1
V7 (ppp) θ
V1(pnn)
V8 (nnn) d1

Fig. 2.7. Synthesis of the required output voltage vector in sector 1.

All SVM schemes and most of the other PWM algorithms [1,4], use (2.3), (2.4)

for the output voltage synthesis. The modulation algorithms that use non-adjacent SSV’s

have been shown to produce higher THD and/or switching losses and are not analyzed

here, although some of them, e.g. hysteresis, can be very simple to implement and can

provide faster transient response. The duty cycles d1, d2, and d0, are uniquely determined

from Fig. 2.7, and (2.3), (2.4), the only difference between PWM schemes that use

adjacent vectors is the choice of the zero vector(s) and the sequence in which the vectors

are applied within the switching cycle.

The degrees of freedom we have in the choice of a given modulation algorithm

are:

1) The choice of the zero vector -

whether we would like to use V7(ppp) or V8(nnn) or both,

2) Sequencing of the vectors

3) Splitting of the duty cycles of the vectors without introducing additional

commutations.

Four such SVM algorithms are considered in the next section, namely:

1) The right aligned sequence ( SVM1)

10
 2) The symmetric sequence (SVM2)

3) The alternating zero vector sequence ( SVM3)

4) The highest current not switched sequence (SVM4) .

The modulation schemes are described for the case when the reference vector is in

sector 1: all other cases are circularly symmetric.

These modulation schemes are analyzed and their relative performance with

respect to switching loss, THD and the peak-to-peak current ripple at the output is

assessed. The analysis is performed over the entire range of modulation index and for

load power factor angle varying from –90o to 90o.

All space vector modulation schemes presented here assume digital

implementation and, hence, regular sampling, i.e. all duty cycles are precalculated at the

beginning of the switching cycle, based on the value of the reference voltage vector at

that instant.

2.4 MODULATION SCHEMES

2.4.1 RIGHT ALIGNED SEQUENCE (SVM1)

A simple way to synthesize the output voltage vector is to turn-on all the bottom

(or top) switches at the beginning of switching cycle and then to turn them off

sequentially so that the zero vector is split between V7(ppp) and V8(nnn) equally. This

switching scheme is shown in Fig. 2.8 for two sampling periods. The signals in the figure

represent the gating signals to the upper legs of the inverter. The scheme has three switch

turn-on’s and three switch turn-off’s within a switching cycle. The performance of the left

aligned sequence, where the sequence of vectors is exactly opposite to the right aligned

sequence, is expected to be similar to the right aligned sequence.

11
 nnn pnn ppn ppp nnn pnn ppn ppp

a p
n

p
b
n

c p
n
d 0 /2 d1 d2 d 0 /2 d 0 /2 d1 d2 d 0 /2
Ts Ts

Fig. 2.8. Phase gating signals in SVM1.

2.4.2 Symmetric Sequence (SVM2)

This scheme has been shown in previous works [4] to have the lowest THD. This

is because of the symmetry in the switching waveform as can be seen in Fig. 2.9. The

number of commutations in one sampling period is six. Since this scheme has the same

number of switchings as SVM1, with three switch turn-ons and three switch turn-offs,

their switching losses are expected to be similar.

nnnpnn ppn ppp ppnpnn nnn nnnpnn ppn ppp ppnpnn nnn

a p
n
p
b
n
p
c
n d d d d0 d2 d1 d0
0 1 2 d0 d1 d2 d0 d2 d1 d0
4 2 2 2 2 2 4 4 2 2 2 2 2 4
Ts Ts

Fig. 2.9. Phase gating signals in SVM2.

12
2.4.3 Alternating Zero Vector Sequence (SVM3)

In this scheme, known as DI sequence in literature [6], the zero vectors V7(ppp)

and V8(nnn) are used alternatively in adjacent cycles so that the effective switching

frequency is halved, as shown in Fig. 2.10.

pnn ppn ppp ppn pnn nnn

a p
n
p
b n

p
c
n d1 d2 d0 d2 d1 d0
Ts Ts

Fig. 2.10. Phase gating signals in SVM3.

However, the sampling period is still Ts, same as in the other schemes. The switching

losses for this scheme are expected to be ideally 50% as compared to those of the

previous two schemes and THD significantly higher due to the existence of the harmonics

at half of the sampling frequency.

2.4.4 Highest Current Not-Switched Sequence (SVM4)

This scheme, known as DD sequence in literature [6], is based on the fact that the

switching losses are approximately proportional to the magnitude of the current being

switched and hence it would be advantageous to avoid switching the inverter leg carrying

the highest instantaneous current. This is possible in most cases, because all adjacent

SSV’s differ in the state of switches in only one leg. Hence, by using only one zero

vector, V7(ppp) or V8(nnn) within a given sector one of the legs does not have to be

switched at all, as shown in Fig. 2.11 (a).

13
 pnn ppn ppp pnn ppn ppp
a p
n
p
b n
p
c n
d1 d2 d0 d1 d2 d0
Ts Ts

Fig. 2.11(a). Phase gating signals in SVM4.

However, since the choice of the non-zero SSVs is based on the desired output

voltage vector and the phase and magnitude of the current are determined by the load, it

is not always possible to avoid switching the phase carrying the highest current. In such a

case the phase carrying the second highest current is not switched and the switching

losses are still reduced. For example, the choice of zero vectors in sector 1 is determined

using the flowchart in Fig. 2.11(b).

Y
|Ia| > |Ic| d0 V7(ppp)

d0 V8(nnn)

Fig. 2.11(b). Choice of zero vector in sector 1.

2.5 ANALYSIS

In this section the THD, switching losses and peak-to-peak current ripple of all the

schemes are analyzed over the entire range of modulation index, and over varying load

power factor angles.

14
2.5.1 Total Harmonic Distortion

The total harmonic distortion (THD) of a periodic voltage which can be

∞
represented by the Fourier series V = ∑ Vn e jnωt is defined as
n =1

∑V
n=2
n
2

THD = ….. (2.5)

V1

and its weighted total harmonic distortion is defined as

2
∞
 Vn 
∑  
n=2  n  ….. (2.6)
WTHD =
V1

The analysis of THD is done based on a novel algorithm where the Fourier

coefficients of all the pulses of a given line/phase voltage in one fundamental time period

are summed up. The analysis is valid for all integral fs/fo, where fs is the switching

frequency and fo is the fundamental frequency at the output of the inverter.

Fig. 2.12 shows the plot of typical phase voltage in one fundamental time period.

The THD of this waveform is calculated by first decomposing the waveform of Fig. 2.12

into a series is pulses which resemble Fig. 2.13.

Phase
Phase Voltage
Voltage

Ts
To
Time
Time

Fig. 2.12. Typical Phase voltage in one fundamental time period.

15
Then the Fourier components for each of these pulses are calculated and summed up to

get the effective Fourier components of the entire waveform in one fundamental time

period. This procedure is illustrated here for SVM1.

Considering a pulse shown in Fig. 2.13 with fundamental period To = 1/fo (Fig.

2.12) its Fourier coefficients at frequency nfo are given by

Vm 2 nπ 2 nTmπ ….. (2.7)

an = ( sin( ( dm Ts +Tm ) ) − sin( ))
nπ To To

Vm 2 nTmπ 2 nπ ….. (2.8)

bn = ( cos( ) − cos( ( dm Ts +Tm ) ) )
nπ To To
Voltage

Vm

Tm d m Ts
Ts
Time

Fig. 2.13. Output voltage pulse.

where dm is the pulse duty cycle, Tm is the pulse delay time, Vm is the pulse

1
magnitude, and Ts = , is the sampling time period.
fs

Decomposition of the phase voltage switching waveform for SVM1 (Fig. 2.13) to

obtain a series of pulses as in Fig. 2.12 is a two step process. At first the waveform is

decomposed to obtain stepped pulses as shown in Fig. 2.14(a), then this stepped pulse

waveform is further decomposed to obtain pulses, similar to Fig. 2.13, as shown in Fig.

2.14(b) and Fig. 2.14(c).

16
 Consider the load to be wye connected and balanced. Then the phase voltages Va,

Vb Vc with respect to a neutral point (N) chosen such that Va+Vb+Vc=0 in any sector for

Vab - Vca V - Vab V - Vbc

any load can be obtained from Va = , Vb = bc , Vc = ca
3 3 3

The phase voltage (Va) in sector1 during different SSVs (V8,V1,V2,V7) for

SVM1 is shown in Table 2.1

Table 2.1. Phase voltage in sector 1: SVM1.

SSV’s V8(nnn) V1(pnn) V2(ppn) V7(ppp)

duty cycle d0/2 d1 d2 d0/2
Va 0 2Vg/3 Vg/3 0

Fig. 2.14 shows the decomposition of a typical phase voltage in the kth switching

interval from the beginning of the phase voltage period, where

fs
0 ≤ k ≤ -1 ….. (2.9)
fo

2Vg/3 2Vg/3

Vg/3 Vg/3

0 0 0
d0/2 d1 d2 d0/2 d0/2 d1 d0/2+d2 d0/2+d1 d2 d0/2

(k-1)Ts kTs (k-1)Ts kTs (k-1)Ts kTs

(a) (b) (c)

Fig. 2.14. Phase voltage decomposition in one sampling period.

Comparing each of the component pulses with Fig. 2.13 one can find that for the

first pulse

17
  d 
dm = d1, Tm =  k - 1 + 0  Ts ,
 2
….. (2.10)
2Vg
Vm = ,
3
and for the second pulse

 d0 
dm = d2, Tm =  k -1 + + d1 Ts ,
 2  ….. (2.11)
Vg
Vm =
3

The duty cycles d1, d2 and d0 are calculated using (2.3) and (2.4). The Fourier

components of these pulses are obtained from (2.7) and (2.8). Thus the Fourier

coefficients of the phase voltage can be found using:

fs fs
-1 -1
fo fo
Vn = ∑a + ∑b
2 2
k k ….. (2.12)
k =0 k =0

The Fourier coefficients of the line current can be obtained by

Vn
I n line _ current = ….. (2.13)
zn

where z n is the impedance of a given phase.

Alternatively, WTHD of line voltage could represent a THD of line current [2,3].

Fig. 2.15 shows the THD of the simulated line current and line voltage and the WTHD of

line voltage for all the schemes over the entire range of modulation index, for the inverter

parameters Vg = 400V, fo = 60 Hz, fs = 2160, R(load) = 3, L(load) = 1mH. Appendix B

lists the MATLAB code, used to generate the ‘predicted’ results in Fig. 2.15.

18
 THD

0 0.866
Modulation index(m)

(a)
THD
THD

0 0.866
Modulation index(m)

(b)
THD

0 0.866
Modulation index(m)

(c)

Fig. 2.15. Voltage and current distortion as a function of modulation index.

a) THD of line current

19
 b) WTHD of line current
c) THD of line voltage

These results agree with what is found in literature. It is interesting to note that the

THD (line current and voltage) for all the schemes decreases with increase in modulation

index. This is because of the increase in the fundamental component of the

voltage/current with increase in modulation index; the other higher order harmonics being

relatively constant. It can also be seen that SVM2 (symmetic) has the least THD. This

can be associated with the symmetry in the switching waveform. By introducing

symmetry (SVM2), the number of phase voltage pulses in a given switching time-period

is doubled as compared to the other schemes. This would make the converter look as if it

were operating at twice the switching frequency. Hence this results in a reduced THD

and reduced peak-to-peak ripple in the load current.

2.5.2. Switching Losses

The switching losses are assumed to be proportional to the product of the voltage

across the switch and the current through the switch at the instant of switching. Since the

voltage across the switch is the bus voltage, it is considered to be a constant. Thus the

losses are proportional to the current during switching.

As a first approximation, the switching ripple is neglected and the losses are

estimated based on the number of commutations required for each switching scheme and

the current at the instant of switching. Losses for scheme SVM4 are dependent on the

load power factor and their loss performance has been optimized using the flow chart

20
presented in Fig. 2.11(b) for the entire range of load power factor angle. Fig. 2.16, shows

the loss performance characteristics of all the schemes.

From Fig. 2.16 we see that SVM4 has a 50 % reduction in losses at high load power

factors (>0.866) and the savings in the losses reduce to 37% at low load power factors.

Relative switching losses

switching loss
Relative

Loadpower
Load power factor
factor angle
angle

Fig. 2.16. Relative switching losses as a function of load power factor angle.

2.5.3. Peak-to-Peak Current Ripple

The peak-to-peak value of the current ripple, at maximum value of load current, is

important in the design of inductors - to be used as filters. Some simple formulae have

been derived below for the accurate estimation of the ripple for all the schemes.

Maximum output current ripple occurs when the volt-second across the output

inductors is the largest. Consider phase ‘a’ in Fig. 2.1 and assume that output voltage V A

(Fig.2.1) varies slowly with respect to the switching frequency. Then for a passive load,

the maximum volt-second across the inductor in phase ‘a’ occurs when the reference

vector in Fig. 2.7 is collinear with the switching vector V1(pnn). At this instant the duty

cycle d2 = 0 and duty cycle d1 = m, and the resultant phase voltage will be as shown in

Fig. 2.17.

21
 Phase voltage
Va d1 = m
Vm
Vav d0 = 1- m
d1Ts d0Ts
Ts

Time

Fig. 2.17. Phase voltage pulse at its peak low frequency value.

Due to the assumption that the output voltage VA (Fig. 2.1) is constant, the

voltage across the inductor is

VL = Va - V A ≈ Va - (Va )av ….. (2.14)

Since

Vm =
2Vg
3
; (Va )av = d1 Vm ….. (2.15)

and

I pp
(Va )av =L ….. (2.16)
d 0Ts

Thus the peak-to-peak current ripple is given by

2Vg
I pp = (1 - m) m Ts ….. (2.17)
3L

Equation (17) is true for SVM1, SVM3, and SVM4. For SVM2 the duty cycle d 0 is split

in half and hence

Vg
I pp = (1-m) m Ts ….. (2.18)
3L

22
Fig. 2.18 shows the plot of the relative peak-to-peak ripple current, based on the above

analysis and based on the simulation results using SABER , for the entire range of

modulation index.

Relative peak-to-peak ripple

0 Modulation index (m) 0.866

Fig. 2.18. Relative peak-to-peak current ripple.

2.6 Simulation and Experimental Results

The circuit in Fig. 2.1 was simulated using SABER. Appendix A gives the listing

of the space vector modulation template developed in MAST for SVM1. The circuit in

Fig. 2.1 was simulated using ideal switches, RCD snubbers, and ideal diodes, using the

following parameters: Vg = 16 V, Iload = 6.5 A, fo = 108 Hz, fs = 3888 Hz, m = 0.6. The

output current waveforms and their spectra are shown in Fig. 2.19. Fig. 2.20 shows the

waveforms of the output current and spectrum of output current obtained from an

experimental inverter running under similar conditions. The figures show a fairly good

agreement between simulation and experimental results.

23
 SVM1

SVM2

SVM3

SVM4

Fig. 2.19. Simulated line currents and spectrum of line currents ( fs/fo = 36).

24
Fig. 2.20. Experimental line currents and spectrum of line currents (fs/fo = 36).

25
2.7 PERFORMANCE SUMMARY

Table 2.2 summarizes the performance of four space vector modulation schemes.

It can be clearly seen that a scheme with high THD has low losses and vice versa and

these characteristics are load dependent. This could be translated as a trade-off to be made

between the size of the heat sink and size of the filters. At low switching frequencies

SVM2 could be used, since losses are not very critical. At high switching frequencies

SVM4 is preferred, especially at high load power factors due to the 50% reduction in

switching losses. SVM3 could be used at low load power factors.

The right aligned sequence (SVM1) (or the left-aligned sequence) does not seem

to have any particular advantage if the converter is hard switched. However, if soft-

switching is introduced then this scheme is particularly useful because here all the three

legs are being switched at the same time.

It has also been observed that the performance of SVM4 can be improved by

introducing symmetry. This result’s in reduced THD and in reduced peak-to-peak ripple

in the load current; with the switching losses remaining the same.

. Table 2.2. Relative performance of various modulation schemes (three-leg).

Modulation Schemes SVM1 SVM2 SVM3 SVM4

No of commutations in Ts 6 6 3 4
Relative Losses 1 1 0.5 0.5-0.63**
Dominant harmonic fs fs fs/2 fs
THD at low mod index Least
THD at high mod. index Least Highest
Relative peak-to-peak ripple at Imax 1 0.5 1 1
Number of switching states in Ts 4 7 3 3
** Depending on the load power factor
* Important for digital modulator implementation

26