14th International Power Electronics and Motion Control Conference, EPE-PEMC 2010

Random Space Vector Modulation for Electric

Drives: A Digital Approach
Hamid KHAN*,**, Youssef TOUZANI*, Khalil El Khamlichi DRISSI**
*IFP, Rueil-Malmaison, France, e-mail: hamid.khan@ifp.fr, youssef.touzani@ifp.fr
**
UBP, LASMEA, Clermont-Ferrand, France, e-mail: drissi@lasmea.univ-bpclermont.fr

Abstract — SVM, with its many advantages over PWM, is before hand. The idea is to use only the zero vector V0 at
gaining popularity. However, only Deterministic-SVM exist. the beginning and end of the period.
Whereas Randomised-PWM with their cleaner harmonic-
spectrum, are gaining interest for industrial applications
required to meet EMC-Standards. Here we present,
Randomised-SVM intended for Electric-Drive oriented
HEV, which has the advantages of SVM & clean harmonic-
spectrum of Randomised-PWM.

Keywords — SVM, RSVM, PWM, Harmonics, HEV, Vector

Control, VSI, FOC, Electrical Drives, Variable speed drive,
Brushless Motor, Power converters for EV, EMC/EMI, DSP.

I. INTRODUCTION

The issue with Electric/HEV is the weak autonomy; to

improve that, the battery should be used efficiently and the
power conversion system must be as light as possible. The
objectives of RSVM are : reducing the weight by
lightening the filtering effort to meet the EMC/EMI Fig. 1. Voltage Space Vectors
standards, optimal use of the battery voltage and minimal
Whereas PWM techniques are based on comparison of
switching losses. Moreover we can place the electronic
components at proximity with the power circuit and the reference signals and the carrier signal, one cannot
reduce the volume of the power conversion chain and avoid the insertion of the vector V7 at the middle of the
period [3]. This would be clear from Fig. 2; switching
could avoid using a faraday cage to curb the radiated noise
function of leg A, when only the vector V0 is used to
(Weight, volume and cost constraints). The electric
traction drive under consideration consists of a Permanent complete the modulation period, remains zero during one-
Magnet Synchronous Motor (PMSM) fed by a 3-phase third of the fundamental cycle. Which explains 33.3%
reduction in the switching losses i.e. only 4 commutations
inverter. The Field Oriented Control (FOC) is used to
instead of 6 per modulation period [4]. It can be seen as an
regulate the torque.
intrinsically or naturally discontinuous-PWM technique
A. Space Vector Modulation i.e. without having to add a zero sequence voltage to the
reference. An analogy can be drawn with the Generalized
SVM is a digital Power Converter PWM technique Discontinuous Modulator (GDPWM) where the zero
where the duty cycle of inverter switches are calculated sequence voltage is given by following expression:
directly using mathematical transformations [1]. Fig. 1
depicts the vectors representing all possible 3-phase Vdc
inverter states that form a circle on the αβ-plane. The three u0    vmin (1)
bit binary subscript denotes the state of upper switch of 2
the inverter leg corresponding the three phases 'a, b and c' Where vmin  min (va* , vb* , vc* ) . So the new reference
in the same order. '0' and '1' represent the off and on state
respectively. The upper and lower switch states of a leg signals become:
are complimentary to avoid short circuiting the voltage
x  vx  u0
v** *
source. The voltage space vectors circle is divided in 6 (2)
sectors in a way to make the duty cycle calculations as
For a balanced three phase sinusoidal system:
simple as possible.
SVM was chosen over ordinary PWM techniques for  
the following reasons: better DC voltage utilization [2] va*  sin ( )  (3)
and decreased switching losses. Reduced switching losses    

can be explained by the freedom that one has in vb   A sin (  2 ) 
*
 *  3 
generating the pulses when the duty cycles are known vc    
sin (  4 ) 
 3 

978-1-4244-7854-5/10/$26.00 ©2010 IEEE T8-20

 14th International Power Electronics and Motion Control Conference, EPE-PEMC 2010

The new reference signals would be saturated to B. EMC/EMI Problem

Vdc for 120 i.e. one-third of the fundamental period In spite of all the advantages of SVM, its drawback is

2 the harmonics at the switching frequency. RPWM is
and out of phase by 2  , this technique is also known as commonly used to tackle this problem. The concept is to
3 randomize the modulation function parameters, like the
DPWMMIN. pulse position [5], the switching frequency to spread the
Fig. 2 shows the switching function for maximum frequency spectrum around the switching frequency and
utilization of the DC bus, i.e. maximum achievable value hence remain under the permissible EMI limit.
V There are quite a few papers on RPWM. The most
of the reference voltage is dc and not Vdc in the linear common RPWM is the randomization of the switching
3 2 frequency, which improves the voltage and current
zone, as explained in the next section. Therefore 'da' is the harmonics [6]. There are other techniques such as dual
image of V in percentage Vdc .
a
randomization [7], which may spread the spectrum a little
3 bit more, whereas only randomizing the pulse position is
not very effective and has a discontinuous Power
Spectrum Density (PSD) [8].

II. RSVM

A. Random Space Vector Modulation

Fig. 2. SVM: Switching Function leg A
The modulation index is defined as:
V fundamental  PWM
mi 
V fundamental  six step
Fig. 3 shows an increase of 15.47% in phase voltages
compared to ordinary PWM techniques. There exist
techniques such as harmonic injection (3rd and its
multiple), to increase the linearity of PWM techniques [4].
However these techniques are not possible for dynamic
systems where the fundamental frequency varies with time
randomly. Furthermore in vector control, instantaneous
values are treated and the complete form of the reference
signal has no significance.
Random Space Vector Modulation (RSVM), a new
solution is proposed to conserve the advantages of SVM
while adding those of the RPWM. The proposed method
is to vary the switching frequency randomly and
calculating the corresponding duty cycles.
There are many random distribution laws; it has been
shown that the uniform distribution, which is the simplest
random distribution, is as good as any other complex
distribution law that can be used for harmonic dithering
linked to switching frequency [9].
B. RSVM Switching Function
The switching (or modulating) function should
(4) correspond to the randomly generated frequency at every
cycle. Fig. 4 and Fig. 5 show the variation of the
modulation function after every random SVM-period.
It can be seen that completing the time period with
vectors V0 and V7 or just by V0 does not change the
average value of the resultant vector generated. “da-db”
gives the image of the line voltage which is conserved in
both the cases, as is clear from the two figures.

mi
SVM
0.907
PWM
Over modulation

0.785

Fig. 4. RSVM: Switching Function (using V7 and V0)

1 1
Vx
Vdc Vdc
2 3

Fig. 3. Increased linearity of Modulation Index (adv of SVM)

978-1-4244-7854-5/10/$26.00 ©2010 IEEE T8-21

 14th International Power Electronics and Motion Control Conference, EPE-PEMC 2010
Fig. 5. RSVM: Switching Function (using only V0)
C. State of the Art
There are about half a dozen papers on RSVM; reading
them thoroughly we find contradictions in basic principles
of SVM. [10] Proposes a 3-level inverter with RSVM to
reduce acoustic noise, however the generation of gate
signals is done through comparison of the reference signal
by a triangular carrier signal, and this technique is
commonly known as PWM. [11] Proposes a random SVM
technique, which is not really random, only two switching
frequencies are used with scalar control, however the
pulse generation remains ambiguous. Similarly there are
other papers such as [12], [13] that treat the problem in a
similar way, which in our opinion are required to be
treated further, which is the purpose of the paper to draw a
line between RSVM and RPWM.
III. SIMULATION RESULTS
All the simulations are done on MATLAB/SIMULINK.
The inverter, the Permanent Magnet Synchronous
Machine (PMSM) and the DC source are taken from the
“SimPowerSystems” library.
A. Open-loop Simulations
Three sine-waves of 20 Hz out of phase to form a
balanced 3-phase system are taken as reference signals.
The open loop results are shown in this section. The
switching frequency for SVM is kept constant at 1 kHz,
whereas for RSVM it is varied randomly between 0.75
and 1.25 kHz and therefore has an expected mean
frequency of 1 kHz.
Fig. 6. Frequency Spectrum Contrast (Open Loop)
Frequency spectrum of the two signals are shown in
Fig. 6. We can see a peak at 1 kHz for SVM and for
978-1-4244-7854-5/10/$26.00 ©2010 IEEE T8-22
RSVM we see a small hump around 1 kHz while the rest
remains the same. The peak value of the switching
harmonics is 10 times lower than the peak value for an
SVM at a constant frequency.
SVM intrinsically adds third-order harmonics in the
phase voltages, which could be seen in Fig. 6 marked by
f3, f6. Higher multiples are not marked but can be seen on
the chart. These are zero sequence harmonics and
therefore disappear from the line voltage. The simulation
result validates the concept in open loop as we can see the
low frequency spectrums perfectly superimpose each
other. The open loop test consists of measuring the phase
voltages with respect to the DC mid point at no load.
B. Closed-loop Simulations
A torque drive is simulated using a PMSM supplied by
a battery via a 3 phase inverter. The control strategy used
is Field Oriented Control (FOC), with the proposed
modulation technique RSVM and SVM. The simulation
models are completely discretized, with the simulation
step being 10µs and a torque of 3 Nm is commanded in
both the cases. The modulating frequency varies between
0.75 and 1.25 kHz for RSVM. In both cases it was
observed that the torque generated at the motor shaft
follows the torque reference, hence the two control
strategies are validated. Now that the models are validated
we continue our analysis on the stator currents and
voltages. Fig. 7 and Fig. 8 show the stator current evolution
in the - plane for SVM and RSVM respectively. We get
anticipated results. The curve for the RSVM is a little
blurry and wavy (darker inner circle) which represents the
randomness added to the system.
Fig. 7. SVM: Stator Current Evolution
Fig. 8. RSVM: Stator Current Evolution
The outer circle represents the starting of the motor,
when it draws in the maximum current and the inner most
circles represent low currents when the machine reaches
 14th International Power Electronics and Motion Control Conference, EPE-PEMC 2010

its steady state and draws only the load current and a part
to meet the mechanical losses.
The frequency spectrum of the line voltage ‘Vab’ is shown
in Fig. 9 and Fig. 10, SVM in red and RSVM in blue. The
first peak represents the evolution of the fundamental
frequency till it reaches its steady state at 19 Hz, which
corresponds to the steady-state speed of 570 rpm for a 4
pole electric machine.

Fig. 11. Schematic Diagram- Complete System

All the calculations (time sector) and generation of a

Fig. 9. Frequency Spectrum Contrast (Closed Loop)
The peaks in red and the humps in blue represent the
harmonics at the switching frequency and its multiples.
Fig. 10 is a zoom around fs (1 kHz) and it's multiple. We
can see the harmonic peak for RSVM is five times smaller
for a frequency range of 500 Hz.
random frequency are done on the DSP. The randomly
generated frequency would synchronize reading data from
the Acquisition system buffers and would also determine
the execution of the control algorithm and in turn the duty
cycle calculation and pulse generation. The gate signals
would be fed to the “Pulse Conditioning + Drivers” block
where the signals of 0 and +3.3V are brought to a certain
voltage level to ensure safe commutations (-7 and +15V
respectively) i.e. to avoid accidental commutations.
B. Test Bench
The test bench consists of two identical machines
mechanically coupled, Fig. 12, along with the rest of the
apparatus mentioned in system description. One that is
part of the system under observation is controlled and the
other acts as a load.

Fig. 10. Switching Harmonics (RSVM and SVM)

Satisfactory results were obtained for simulations

carried out for higher switching frequencies, 10 kHz and
above. However to stay coherent throughout the document
we've worked with 1 kHz.

IV. EXPERIMETAL SETUP

A. System Description
The experimental setup consists of a 15 kW, 3-phase 2-
level IGBT inverter works on 400 V DC with current and
voltage sensors for stator current and voltage
measurement, 3 kW PMSM with an incremental encoder
(4096 points), a 300 MHz Floating-Point DSP
“TMS320C6727”, high speed acquisition system, a buffer
card for gate signal conditioning with incorporated dead
time. The Fig. 11 shows the schematic diagram of the
system. The RSVM algorithm is validated on SIMULINK
in open and closed loop.
Fig. 12. Test Bench
C. Experimental Results
Open-loop test results are shown in Fig. 13 and Fig. 14
for SVM and RSVM respectively, for a balanced 3-phase
system with fundamental frequency at 20 Hz and the
switching frequency at 1 kHz for SVM and a randomly
varying switching frequency between 0.75 and 1.25 kHz
for RSVM.
The experimental results are completely coherent with
simulation results for open loop, with switching
harmonics 10 times smaller in both cases.

978-1-4244-7854-5/10/$26.00 ©2010 IEEE T8-23

 14th International Power Electronics and Motion Control Conference, EPE-PEMC 2010
Fig. 13. Frequency Spectrum SVM
Fig. 14. Frequency Spectrum RSVM
Due to some technical glitches closed loop
experimental results could not be included. Nevertheless
we intend to present it in the conference.
V. CONCLUSION AND FUTURE WORK
The RSVM concept was validated with harmonics 5
times smaller in closed loop for an electric drive with high
dynamic over a frequency range of 500 Hz while reducing
the number of commutations by 33.3% for a given
switching frequency. We have a custom-made
experimental setup capable of testing the functionality,
feasibility and practicality of the technique. We could also
analyze the different inverter and electric motor variables
(electrical and mechanical) like the phase currents and
voltages, torque and speed.
We plan to quantify the reduction of Electromagnetic
Emissions using RSVM in a reverberating chamber; treat
the problem of torque ripple produced by the harmonic
content of the stator current as well as different acquisition
techniques that could be used when the data is not fed at a
fixed frequency to the control block, such as using anti-
aliasing filters, oversampling techniques etc. Finally
validating it on a IFP HEV prototype.
978-1-4244-7854-5/10/$26.00 ©2010 IEEE
REFERENCES
[1] Simone Buso, Paolo Mattavelli, “Digital Control in Power
Electronics”, Morgan & Claypool 2006 ISBN-10: 1598291122,
LECTURES ON POWER ELECTRONICS, vol. 2 pp.81-91.
[2] Stefan Laurentiu CAPITANEANU, “Optimisation de la fonction
MLI d’un onduleur de tension deux-niveaux", PhD Thesis
submitted to the Institut National Polytechnique De Toulouse,
pp.30-44, November 2002.
[3] Keliang Zhou, Danwei Wang, “Relationship between space-vector
modulation and three-phase carrier-based PWM: a comprehensive
analysis [three-phase inverters] ", Industrial Electronics, IEEE
Transactions, Sch. of Electr. & Electron. Eng., Nanyang Technol.
Univ., vol. 49 pp. 186-196, Feb 2002.
[4] Michael M. Bech, “Analysis of Random Pulse-Width Modulation
Techniques for Power Electronic Converters", PhD Thesis
submitted to the Faculty of Engineering & Science at Aalborg
University, pp.22-26, August 2002.
[5] S. Mahdavi, J. Agah, “Application of Random PWM Technique
for Reducing the Conducted Electromagnetic Emissions in Active
Filters”, Industrial Electronics, IEEE Transactions, vol. 54 pp.
2333 - 2343, Aug. 2007.
[6] V. Blasko, M. Bech, “A new hybrid random pulse width
modulator for industrial drives", Applied Power Electronics
Conference and Exposition, IEEE 2000, New Orleans, LA, USA.
vol. 2 pp.932-938, 2000.
[7] K. El Khamlichi Drissi, P.C.K. Luk, B. WANG and J. Fontaine,
“A Novel Dual-Randomization PWM Scheme for Power
Converters”, 34th IEEE Power Electronics Specialists Conference
(PESC), vol. 2, pp. 480-484, 15-19 June 2003, Acapulco, Mexico.
[8] N. Boudjerda, M. Melit, B. Nekhoul, K. El Khamlichi Drissi and
K. Kerroum, “Spread spectrum in DC-DC full bridge voltage
converter by a dual randomized PWM scheme", Electromagnetic
Compatibility - EMC Europe, 2008, Hamburg, pp. 181-186,
September 2008.
[9] K. El Khamlichi Drissi, P.C.K. Luk and B. Wang, “Effects of
symmetric distribution laws on spectral power density in
randomized PWM ", Power Electronics Letters, IEEE, vol. 1
pp.41-44, June 2003.
[10] Antônio S. de Oliveira Jr., Edison R. da Silva, “Random Space
Vector Modulation for 3-Level Power Inverters and Induction
Motor Drives Current Control ", Power Electronics Specialists
Conference, vol. 36 pp.987-993, June 2005.
[11] M. Zigliotto, A. M. Trzynadlowski, “Effective Random Space
Vector Modulation for EMI Reduction in Low-Cost PWM
Inverters", Power Electronics and Variable Speed Drives 1998,
IEEE, pp.163-168, Sep 1998.
[12] S. Bolognani, R. Conton, “Experimental Analysis of the EMI
Reduction in PWM Inverters Using Random Space vector
Modulation", Industrial Electronics, IEEE, 1996, vol. 1 pp. 482 -
487, Jun 1996.
[13] Yen-Shin Lai, “Random space vector modulation technique for
high switching frequency inverter control ", Electronics Letters
1997, IEEE, vol. A4 pp.108-117, September 2008.
T8-24