191
Original scientific paper
Journal of Microelectronics, 
Electronic Components and Materials
Vol. 53, No. 3(2023), 191 – 202
https://doi.org/10.33180/InfMIDEM2023.306
How to cite:
K. Saksida et al., “A Comprehensive Strategy for Modelling the Conducted EMI of an Integrated Motor Drive in the Time Domain", Inf.
Midem-J. Microelectron. Electron. Compon. Mater., Vol. 53, No. 3(2023), pp. 191–202
A Comprehensive Strategy for Modelling the 
Conducted EMI of an Integrated Motor Drive in 
the Time Domain
Kristjan Saksida
1 ,2
, Marko Jankovec
2
1
Pre-development Mechatronics, MAHLE Electric Drives Slovenija d.o.o., Šempeter pri Gorici, 
Slovenia
2
Laboratory of Photovoltaics and Optoelectronics, Faculty of Electrical Engineering, University of 
Ljubljana, Ljubljana, Slovenia
Abstract: The paper introduces a comprehensive strategy for modelling the conducted electromagnetic interference (EMI) of a 48 V 11,2 
kW permanent magnet motor drive with an integrated three-phase inverter in the time domain to comply with the CISPR 25 standard. 
The strategy is based on transient electrical simulations using LTspice® and EMI receiver modelling by signal post-processing in MATLAB®. 
To extract AC lumped component parameters for electrical simulations of the inverter, the Ansys® Q3D Extractor® environment was 
employed, while the electrical parameters of the motor side were determined from an impedance analyzer measurement. Two distinct 
transient simulations were performed and compared, namely a static PWM and a Space Vector Pulse Width Modulation (SVPWM) control 
strategy, where for the latter the simulation time was chosen to capture one electrical revolution of the motor. For the EMI receiver 
simulation, a digital Short-Time Fourier Transform (STFT) algorithm, compliant with CISPR 1611, was implemented. The simulation of one 
electrical revolution of the motor lasted 43 minutes, resulting in the average spectral deviation between the measured and simulated 
spectrum bandwidth of 3,33 dBµV and 2,58 dBµV for peak and average detectors, respectively. The introduced strategy proposes a 
straightforward approach for the extraction of parasitic elements. Furthermore, it ensures accurate EMI prediction without compromising 
simulation time, which is crucial to stay within the timeframe when developing an automotive product.
Keywords: EMI modelling; integrated motor drive; FFT EMI receiver; SVPWM; transient analysis
Celostni pristop k modeliranju prevodnih 
elektromagnetnih motenj integriranega 
elektromotorskega pogona v časovni domeni
Izvleček: Članek predstavlja celovito strategijo modeliranja prevodnih elektromagnetnih motenj (EMI) elektromotorskega pogona 
s trajnimi magneti (48 V, 11,2 kW), ki ima integriran trifazni pretvornik, v skladu s standardom CISPR 25. Strategija temelji na uporabi 
orodja LTspice® za analizo v časovnem prostoru in na modeliranju EMI sprejemnika s post-obdelavo signalov v okolju MATLAB®. Za 
pridobivanje parazitnih parametrov, ki so potrebni za električne simulacije pretvornika, je bilo uporabljeno okolje Ansys® Q3D Extractor®, 
medtem ko so bili električni parametri na strani motorja določeni z impedančnim analizatorjem. Izvedeni in primerjani sta bili dve različni 
časovni simulaciji. Prva je temeljila na statičnem PWM krmiljenju mostiča, druga pa na pulzno-širinski modulaciji v vektorskem prostoru 
(SVPWM), pri čemer je bilo za slednjo določeno trajanje simulacije na podlagi zajema enega električnega obrata motorja. Pri simulaciji EMI 
sprejemnika je bila v skladu s CISPR 16-1-1 implementirana časovno kratka Fourierjeva transformacija (STFT). Simulacija enega električnega 
obrata motorja je trajala 43 minut. Povprečno odstopanje med merjenim in simuliranim spektrom na frekvenčnem območju zanimanja 
je za detektor vršne vrednosti znašalo 3,33 dBµV, za detektor povprečne vrednosti pa 2,58 dBµV. Predstavljena strategija predlaga preprost 
pristop za določanje parazitnih parametrov. Poleg tega zagotavlja zadovoljivo napoved EMI, vendar ne na račun podaljšanja trajanja 
simulacije, kar je ključno za upoštevanje časovnih omejitev pri razvoju izdelka za avtomobilski trg.
Ključne besede: EMI modeliranje; integriran elektromotorski pogon; FFT EMI sprejemnik; SVPWM; analiza v časovnem prostoru
* Corresponding Author’s e-mail: kristjan.saksida@mahle.com

192
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202
1 Introduction
During the development phase of an automotive 
product, strict project timelines and budget limitations 
must be considered and not taken separately from 
the technical aspect. Having fewer iteration loops also 
means using fewer project resources. 
Development of an integrated motor drive is a good 
example, where failing to reach electromagnetic com-
patibility (EMC) requirements could lead to major cor-
rection loops of the entire design. An accurate and 
fast electromagnetic interference (EMI) prediction is 
therefore crucial to stay within the project timeline and 
budget boundaries. To mitigate the simulation time, re-
searchers mainly propose a frequency domain instead 
of a time domain approach, simply claiming that the 
time domain simulation takes too long to complete 
[1]–[3]. However, they do not reveal actual simulation 
times [4]–[6].
Regardless of the simulation domain, many papers 
focus on methodologies for the extraction of lumped 
parameters. Frequently, they focus intensively on a par-
ticular component of the system, such as a switching 
device, PCB, or motor, while oversimplifying other ele-
ments. However, they often lack support from a sen-
sitivity study addressing the omitted parameters that 
influence the EMI spectrum.
In terms of modeling switching devices, authors main-
ly focus on adequately describing current and voltage 
transition slopes, recognizing these phenomena as the 
main source of the EMI [4], [5], [7], [8]. 
The determination of lumped parameters of a PCB is 
typically approached in two ways. One method in-
volves analytical techniques, particularly suitable for 
straightforward geometries. Alternatively, for more 
intricate structures, such as multilayer configurations, 
parasitic extraction software is often employed [5].
When modelling a motor, the finite element method 
(FEM) yields good results; however, managing mechan-
ically complex electrical machine geometries is a time-
consuming task [9]. Based on this fact, many research-
ers propose a method, where a complex motor model 
is built from different impedance measurements [10]–
[15]. Keeping the rotor position in mind, the method is 
suitable for induction and permanent magnet motors.
Last but not least, in many cases the EMI receiver mod-
els are either not implemented or the specifics not 
documented or referenced, resulting in a questionable 
comparison of a simulated and measurement results 
[1]–[6].
This paper proposes a comprehensive strategy for 
modelling the conducted emissions of an integrated 
motor drive, ensuring that spectrum accuracy is not 
compromised by simplifications that lead to simula-
tion time mitigation. The paper indicates the lumped 
parameters with significant effect to the EMI spectrum 
and objectively addresses the simulation time dura-
tion, aiming to keep it below 1 hour. In conjunction 
with non-symmetrical parasitic elements, it also evalu-
ates the impact of a motor control strategy implemen-
tation on the EMI spectrum. Furthermore, it indicates 
the minimum simulation time needed to capture all 
EMI phenomena of an integrated motor drive within 
the dwell time of the EMI receiver.
The equipment under test (EUT) was MAHLE 48 V 11,2 
kW permanent magnet motor drive (see Fig. 1) with an 
integrated 3phase inverter that must comply with the 
CISPR 25: 2016 standard for Conducted and Radiated 
Emissions [16], [17]. The standard prescribes an EUT to 
operate under typical loading conditions. To simplify 
the test setup, the EUT was used without a load in a 
fixed operating point.
Figure 1: MAHLE 48 V 11.2 kW integrated motor drive 
with test setup according to CISPR 25.
2 Frequency & time domain approach
SPICE based simulators are commonly used for elec-
tronic circuits simulation. Modelling of EMI phenom-
ena consists of modelling the EUT itself as well as the 
test setup specifics according to an applied standard. 
Many authors propose to use frequency domain ap-
proach which gives almost an instant simulation result, 
that is spectral density on the measurement node [1]–
[3]. This approach usually gives adequate results when 
simulating low power motor drives where the dimen-
sions and with that all parasitic phenomena are limited 
to the components themselves.

193
However, when the dimensions increase and with that 
nonsymmetrical parasitic elements are introduced, the 
simplification of the di/dt and du/dt transients with cur-
rent and voltage sources respectively, deteriorates simu-
lation accuracy. High power motor drives usually consist 
of multiple printed circuit board assemblies (PCBAs), 
where additional parasitic elements are introduced with 
interfaces between these boards (see Fig. 2).
Figure 2: Exploded view of the EUT.
To accurately capture all EMI effects that are influenced 
by the rotation of the motor itself in the form of differ-
ent current paths through the non-symmetrical geom-
etry, a transient analysis must be performed, inevitably 
resulting in longer simulation times. A comparison be-
tween both approaches could be seen in the Table 1.
Table 1: Frequency vs. time domain.
Frequency domain Time domain
(-) EMI receiver func-
tionality (limited imple-
mentation)
(+) Complete EMI receiver 
functionality (IF filter, STFT, 
detectors, dwell time)
(-) Circuit analysis 
presumes stationary 
conditions
(+) Space Vector Pulse 
Width Modulation
(-) Complex non-linear 
SPICE models not pos-
sible to use
(+) Possibility to use official 
non-liner SPICE models 
(e.g., MOSFETs)
(+) Instant simulation 
results with high con-
vergence
(-) Time consuming simula-
tion with convergence 
issues
To mitigate long simulation time and convergence is-
sues, a good understanding of the EM spectrum sensi-
tivity to different parasitic components is beneficial. By 
that, to simplify the model and thus shorten the simu-
lation time, we can exclude those with negligible effect 
to the spectrum.
3 Modelling
According to the Fig. 3, the LTspice® environment was 
used to construct an appropriate schematic for tran-
sient analysis. Signal post-processing, corresponding 
to the EMI receiver, was done using MATLAB®.
Figure 3: Modelling flow.
The EUT has a dedicated 12 V power supply line for the 
logic board, which is separate from the main 48 V bat-
tery power supply in terms of grounding. This setup en-
ables us to observe and address conductive emissions 
for each part separately. In this paper, our focus will be 
solely on the 48 V power stage.
3.1 CISPR 25 test setup
The baseline for the test setup modelling is a reference 
ground plane with two grounded Artificial Networks 
(Ans), where the EUT’s housing is also grounded with a 
copper braid (see Fig. 4). This configuration establishes 
a return path for common-mode (CM) emissions. The 
power supply lines were 40 cm long, and their induct-
ances are a part of the path for differential mode noise. 
The mentioned inductances were calculated analytically 
based on their geometry according to CISPR 16-2-1 [18].
Figure 4: Top level EMI model including inverter (DC link, power board), motor and test setup according to CISPR 25.
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

194
 
2
2l n0 ,5 0, 22  
lb c
Ll
bc l
 
 



  (1)
3.2 DC link
The DC link PCB assembly (PCBA) consists of four cop-
per layers, as shown in the Fig. 5, with the internal two 
layers serving as DC positive and DC negative power 
planes. The two external layers contain multiple poly-
gons to ensure a homogeneous load on each capacitor. 
This configuration forms a matrix of parasitic induct-
ances that was modeled along with electrolytic and 
ceramic capacitors. Parasitic capacitances between 
the planes were neglected due to their small values in 
comparison to the capacitances of the capacitors. 
The parasitic inductances, extracted with boundary ele-
ment method (BEM), were recalculated into an average 
inductance per unit length (IPL) to establish a founda-
tion for simplifying the matrix. This simplification results 
in fewer components and nodes for SPICE simulation.
Figure 5: DC link PCB (Ansys® SIwave™).
The DC link matrix was modeled with 5 different hori-
zontal inductances and 2 vertical ones. Each induct-
ance has its own series resistance (see Table 2). The 
pads connecting the DC link to the power board are 
not symmetrically distributed, as shown in the Fig. 6.
Table 2: DC link parasitic parameters.
Parameter Inductance [nH] Series resistance [mΩ]
Lh1 0,88 0,22 
Lh2 1,76 0,44
Lh3 2,4 0,6
Lh4 3,28 0,82
Lh5 1,4 0,35
Lv1 0,44 0,11
Lv2 1,64 0,41
3.3 Power board
The same approach to extract the parasitic elements 
from the Table 3 was used for the power board (see Fig. 
7), where we sought the values of the parasitic induct-
ances between the MOSFETs soldered onto the copper 
layer of the insulated metal substrate (IMS) [19]. 
Table 3: Power board PCB parasitic parameters.
Parameter Inductance [nH] Series resistance [mΩ]
L_pwr1 1,75 0,1 
L_pwr2 3,5 0,2
L_phase 3,0 0,15
Parasitic capacitances of the polygons towards the sub-
strate were calculated analytically. Capacitances of the B+ 
and B- polygons towards the substrate did not have an 
influence on the EMI spectrum due to the much higher 
capacitance of the Y-capacitors C1 and C2 from the Fig. 
8. There was also a negligible influence of the parasitic in-
ductances of the gate traces, whose values vary with the 
distance from the connector to the most distant MOSFET. 
To model the MOSFET used in our experiment (Fig. 8), 
we applied the official SPICE model provided by the 
manufacturer. As the accurate switching element sim-
ulation is crucial for accurate EMI modelling, we veri-
fied the MOSEFET model adequacy by comparing the 
measured and simulated switching voltage transition 
shapes on a clamped inductive load according to Vr-
tovec et. al. [8]. Fig. 9 shows the setup for switching of 
the high side MOSFETs where the power board model 
is simplified for figure clarity.
C1
{C_mlcc}
C2
{C_elko}
L1
{Lh2}
L2
{Lh1}
C3
{C_mlcc}
C4
{C_elko}
L3
{Lh2}
L4
{Lh1}
L5
{Lh3}
L6
{Lh4}
C5
{C_mlcc}
C6
{C_elko}
L7
{Lh1}
L8
{Lh2}
C7
{C_mlcc}
L9
{Lh2}
L10
{Lh1}
C8
{C_elko}
L11
{Lh1}
L12
{Lh2}
C9
{C_mlcc}
C10
{C_elko}
L13
{Lh4}
L14
{Lh3}
L15
{Lh1}
L16
{Lh2}
C11
{C_mlcc}
C12
{C_mlcc}
C13
{C_elko}
L17
{Lh1}
L18
{Lh2}
C14
{C_mlcc}
C15
{C_elko}
L19
{Lh1}
L20
{Lh2}
L21
{Lh4}
L22
{Lh3}
C16
{C_mlcc}
C17
{C_elko}
L23
{Lh2}
L24
{Lh1}
C18
{C_mlcc}
L25
{Lh1}
L26
{Lh2}
C19
{C_elko}
L27
{Lh2}
L28
{Lh1}
C20
{C_mlcc}
C21
{C_elko}
L29
{Lh3}
L30
{Lh4}
L31
{Lh2}
L32
{Lh1}
C22
{C_mlcc}
C23
{C_mlcc}
C24
{C_elko}
L33
{Lh2}
L34
{Lh1}
C25
{C_mlcc}
C26
{C_elko}
L35
{Lh2}
L36
{Lh1}
L37
{Lh3}
L38
{Lh4}
C27
{C_mlcc}
C28
{C_elko}
L39
{Lh1}
L40
{Lh2}
L41
{Lh5}
L42
{Lh3}
C29
{C_mlcc}
L43
{Lh2}
L44
{Lh1}
C30
{C_elko}
L45
{Lh1}
L46
{Lh2}
C31
{C_mlcc}
C32
{C_elko}
L47
{Lh4}
L48
{Lh3}
L49
{Lh1}
L50
{Lh2}
C33
{C_mlcc}
L51
{Lv1}
L52
{Lv1}
L53
{Lv2}
L54
{Lv2}
L55
{Lv2}
L56
{Lv2}
L57
{Lv2}
L58
{Lv2}
L59
{Lv2}
L60
{Lv2}
L61
{Lv2}
L62
{Lv2}
L63
{Lv2}
L64
{Lv2}
L65
{Lv1}
L66
{Lv1}
L67
{Lv1}
L68
{Lv1}
L69
{L_port}
L70
{L_port}
C34
{C_mlcc}
L71
{L_port}
L72
{L_port}
L73
{Lh5}
L74
{Lh5}
L75
{Lh5}
L76
{Lh5}
L77
{Lh3}
L78
{Lh5}
L79
{Lh3}
L80
{Lh5}
L81
{Lh5}
L82
{Lh5}
L83
{Lh5}
L84
{Lh5}
L85
{Lh5}
L86
{Lh3}
L87
{Lh5}
L88
{Lh3}
L89
{Lh5}
L90
{Lh5}
L91
{Lh5}
L92
{Lh5}
L93
{Lh3}
L94
{Lh5}
W2-
V2-
U2-
W2+ V2+ U2+
W1- V1- U1-
W1+
V1+
U1+
DC+
DC-
DC-1 DC+1
DC-2 DC+2
DC+1
DC-1
DC+2
DC-2
Figure 6: LTspice® model of the DC link.
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

195
In the Fig. 10 we observe a good matching between 
the measured and simulated switching behavior of the 
high side MOSFETs (Phase V). It’s crucial to note that 
the voltage-dependent capacitances of a MOSFET con-
sistently influence the switching characteristics, along-
side the inductances L
d
 and L
s
 originating from the PCB 
itself [8]. The latter is probably the cause for a slight 
phase shift between the graphs.
Figure 10: Switching sequence – high side ON. V
GS
 has 
an offset of +50 V for figure clarity.
Figure 7: Power board PCB planes (Ansys® SIwave™).
Figure 8: Power board single phase LTspice® model.
Figure 9: Test configuration featuring the clamped 
inductive load with V
gg
 controlling the high side MOS-
FETs. The measurements were taken using an inductive 
load of 40 µH and 6.66 Ω, achieving a current of 400 A
DC
 
with the correct duty cycle, the V
DC
 was 50 V.
Figure 11: Common mode test configuration with mo-
tor equivalent impedance Z
CM-1
 and differential mode 
test configuration with Z
DM-1
 [13].
3.4 Motor
As already stated in the introduction, many researches 
propose an approach where lumped motor parameters 
are extracted through motor impedance measure-
ments [10]–[15]. With this technique, a high-frequency 
model distinguishes common mode (CM) and differen-
tial mode (DM) impedances (see Fig. 11).
The Z
CM-1
 measurement results in the Fig. 12 indicates 
a capacitive response in the low-frequency range, 
characterized by a single resonance point where in-
ductance becomes predominant. It turns out that the 
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

196
angular orientation of the rotor, representing the mag-
netic field of the permanent magnets with respect to 
the stator winding, exhibits minimal impact within the 
capacitive range of the Z
CM-1
.
Figure 12: Z
CM-1
 measurement results and approxima-
tion with RLC model.
From the high inductance of the windings, it also follows 
that the Z
DM-1
 itself does not have a major influence on 
the di/dt produced during the switching sequence. Pro-
ceeding from this claim we can simplify the motor model 
where the Z
CM-1
 represents CM propagation path and is in-
dependently combined with the winding characteristics. 
To model the impedance behavior from the Fig. 12, we 
employed a simplified Z
CM-1
 model with three basic RLC 
circuits, disregarding variations in rotor positions.
1
 
3
cm
CM
Z
Z

     (2)
The proposed motor model from the Fig. 4 was evalu-
ated with an AC analysis according to the Fig. 13 result-
ing in an adequate matching to the measured Z
CM-1
 be-
havior (see Fig. 12). 
4 EMI receiver
The characteristics of an EMI receiver must also be con-
sidered if we want to have an appropriate comparison 
between measured and simulated results of an EM 
spectrum. For historical reasons, the instrument char-
acteristics are based on analogue super-heterodyne 
EMI receivers that sequentially scan frequency range of 
interest [20]. This approach is time-consuming and has 
been replaced with new digital Fast Fourier Transform 
(FFT) based instruments that comply with CISPR 16-1-1 
[21], [22]. A basic block diagram of such a digital instru-
ment can be found in the Fig. 14.
Figure 14: Block diagram of the FFT based EMI receiver 
[20].
4.1 MATLAB implementation of the EMI receiver
Within the SPICE transient simulation, the time step is 
varied according to the dynamics of the simulated re-
sults, thus an interpolation was performed to obtain a 
time vector with a constant time step. Generally, inter-
polation methods can introduce signal artifacts that 
could result in unrealistic spectral components. Thor-
ough testing, we concluded that for the applied simu-
lation conditions, the only method among the avail-
able within the MATLAB® environment that does not 
produce random overshoots between two simulated 
points was the Piece-wise Cubic Hermite Interpolating 
Polynomial (PCHIP) method.
CISPR 16 does not provide normative specifications for 
the parameters of the FFT. However, emerging from the 
Table 4, an appropriate windowing function must be 
used to meet the frequency response of the applied In-
termediate Frequency (IF) filter, which originates from 
the superheterodyne principle [21]. Conventional EMI 
receivers generally apply a Gaussian window where the 
standard deviation of the windowing function is de-
fined with the IF bandwidth for each frequency band. 
The windowing function also prevents spectral leakage 
caused by the discontinuity of the sampled signal.
Figure 13: AC analysis model of the RLC circuit togeth-
er with winding characteristics.
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

197
Table 4: Bandwidth requirements for measuring re-
ceivers according to CISPR 16-1-1.
Frequency range [MHz] f
BW
 [kHz]
0,15 – 30 (Band B) 9
30 – 1000 (Bands C & D) 120
1000 – 18000 (Band E) 1000
In the time domain, the Gaussian window function 
is transformed by the Short-Time Fourier Transform 
(STFT) into a Gaussian measurement bandwidth. This 
transformation results in discrete, overlapping meas-
urement bandwidths in the frequency range. If a sine-
wave carrier is positioned exactly between two meas-
urement bandwidths, referred to as frequency bins, the 
result is a picket-fence effect (PFE) level error [20].
CISPR 16-1-1 requires accuracy better than ± 2 dB (± 2,5 
dB above 1 GHz) for a sine-wave voltage measurement 
with 50 Ω resistive source impedance [21]. An overlap 
of more than 75% between the STFTs (see Fig. 15) en-
sures that the level measurement uncertainty for the 
pulse amplitude remains less than ± 1,5 dB [23]. In the 
used R&S ESR EMI test receiver, the STFT’s overlap is at 
least 93%. The maximum level error is ± 0,4 dB, and the 
average level error just ± 0,1 dB [20].
Higher overlapping means longer computational time. 
To stay on the theoretical limit for the required error, an 
overlap of 75% was implemented, where M represents 
the number of applied windows (i.e., the number of 
STFT computations), including the measures for spec-
trum corrections [24]:
- dividing by the window length N
win 
,
- multiplying by the factor of  2 to calculate single 
side spectrum,
- dividing by the factor of 2 to represent the meas-
ured power on 50 Ω input of the EMI receiver since 
the STFT algorithm returns the peak voltage value,
- dividing by the coherent amplification of the ap-
plied window G
C
.
 
 
1
1
 
win
N
n
win
Gc wn
N



   (3)
STFT algorithm produces an array of complex num-
bers ( ) , Xkm, where the k th row corresponds to the 
amplitudes and phases of the frequency spectrum at 
frequency K 
.
 f
res
.
By analyzing each individual row m of the complex ar-
ray ( ) , Xkm (see Fig. 16), it is possible to calculate the 
peak and average values at the observed frequency. 
The peak value is determined as the maximum value in 
the respective row.
Figure 16: STFT algorithm without envelope detection.
 
  
,, 1, 
PK
XkXkmm M   (5)
The average value for a certain k is calculated as the av-
erage of all absolute values in a certain row.
 


1
,
M
m
AV
Xkm
Xk
M



   (6)
Figure 15: STFT overlapping.
  
 
2
1
2
, 11  
100 2
win
win
nk
j
N
proc N
win
n
win
O
Xkmx nm Nw ne
NG c
 




  
 






(4)
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

198
5 Control strategy and results
The complexity of models, the number of nodes, and 
simulation parameters, have a direct impact on the du-
ration of transient simulations and spectrum accuracy. 
The EUT employs the Space Vector Pulse Width Modu-
lation (SVPWM) control strategy. To understand the 
influences of its implementation in the SPICE environ-
ment, we first implemented a simplified strategy where 
the PWM remains static.
5.1 Static PWM
The circuit from the Fig. 17 simulates a freeze-frame of 
a switching sequence, including dead-time. We ana-
lyzed EMI at 2 arbitrarily chosen static PWM switching 
sequences according to the Fig. 18, denoted by t
1
 and 
t
2
. The first PWM1 switching sequence from the Fig. 19 
was set at t
1
 when I
Phase U
 goes through zero, while the 
second static PWM2 sequence was set at t
2
 when I
Phase U
  is at its peak value of 280 A.
Figure 19: Static PWM1 switching sequence at t
1
.
Here we show only the simulation setup at t
1
 in detail, 
similarly was done for t
2
. The switching frequency was 
10 kHz, and the DC bus voltage was 48 V.
The approximation of this method lies in the inability 
to completely recreate PWM conditions at any SVPWM 
operating point. Therefore, it is important to set ap-
propriate initial conditions to match the phase current 
values.
The input for the STFT algorithm cannot be infinitely 
short and must encompass at least a few overlapping 
windows to provide an adequate result. A time range of 
± 0,5 ms around t
1
 = 0,51 ms resulted in the application 
of six windows M = 6 (see Fig. 20). 
Figure 20: Simulated phase currents based on static 
PWM1 (± 0,5 ms around t
1
 = 0,51 ms).
Figure 17: Static PWM1 implementation in LTspice®.
Figure 18: Simulated phase currents based on the SVP-
WM control strategy (t
START
 is the starting time for post-
processing, t
1
 and t
2
 are times of the arbitrarily chosen 
static PWM switching sequences).
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

199
The simulated peak and average spectra for both se-
quences can be seen in the Fig. 21 and Fig. 22. The aver-
age spectral deviation for the switching sequence at t
1
 
is 7,34 dBµV and 3,79 dBµV, at t
2
 is 6,35 dBµV and 3,65 
dBµV for peak and average detectors, respectively. The 
transient analysis took 2 min to apply M = 6.
Figure 21: Comparison between measured and pre-
dicted peak detector spectrum (static PWM).
Figure 22: Comparison between measured and pre-
dicted average detector spectrum (static PWM).
5.2 Space Vector Pulse Width Modulation
To evaluate the effect of the motor rotation on the 
overall spectrum accuracy, the complete SVPWM con-
trol strategy for the inverter system was implemented 
in the LTspice® (see Fig. 23). The corresponding control 
strategy diagram is shown in the Fig. 24.
Figure 23: SVPWM implementation in the LTspice®.
Figure 24: SVPWM control strategy (the chosen se-
quence is for figure clarity only and does not match the 
operating point at 1000 RPM).
The simulated peak phase current reached a value of 
280 A
PEAK
 (200 A
RMS
) by adjusting the appropriate modu-
lation factor and other switching parameters according 
to the Table 5. On the EUT, the crossover distortion ef-
fect is mitigated within regulation algorithm.
Table 5: LTspice® switching parameters.
 Parameter Description
.param fm=66.6 Motor frequency
.param m=0.095 Modulation factor
.param Tdead=1.5e-6 Dead time
5.3 Transient analysis
The complexity of models, the number of nodes, and 
simulation parameters, have a direct impact on the du-
ration of transient simulations. CISPR 25 requires operat-
ing the EUT under typical loading and other conditions 
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

200
like those in a vehicle, ensuring that the maximum emis-
sion state occurs. The dwell time was 100 ms for 9 kHz 
and 10 ms for 120 kHz IF bandwidth, which must be ap-
propriately translated into the simulation environment. 
To capture the entire spectrum of possible emissions 
produced by the motor drive, which in our case has 8 
poles, we must set the simulation time to observe one 
electrical revolution of the motor at 1000 RPM.
 
2
15 
 
2
TRAN
m
tm s
p



    (7)
As stated in the previous chapter, the LTspice® environ-
ment does not have the option to fix the time step due 
to convergence reasons, but it does have an option to 
limit its maximum value. In this case, the Nyquist theo-
rem must be considered, with 108 MHz as the max fre-
quency scale of interest.
2 2 108   
NYQUISTS
ff MHz     (8)
 1
 4,63   
stepMAX
NYQUIST
tn s
f

   (9)
In addition to the previously discussed minimum dura-
tion of the transient analysis (15 ms), we added 7.5 ms 
(total 22.5 ms) to ensure that all transient phenomena 
in LTspice® expired before t
START
, including all phase cur-
rents reaching 280 A
PEAK
, as it seen in the Fig. 18. The 
simulation setup was according to the Table 6.
Table 6: Simulation setup.
Parameter Value
Processor, max. threads Intel®Core™ i7-12700H, 20
Solver Normal
Analysis .tran 0 22.5m 7.5m 4.63n
The comparison of the measured and simulated EM 
spectrum for both detectors is shown in the Fig. 25 and 
Fig. 26, including the absolute deviation. The results 
show 3,33 dBµV and 2,58 dBµV of the average spectral 
deviation for peak and average detectors, respectively. 
The transient analysis took 43 min to capture one elec-
trical revolution of the motor.
Figure 26: Comparison between measured and pre-
dicted average detector spectrum (SVPWM).
6 Conclusions
The proposed comprehensive strategy for modelling 
the conducted EMI of an integrated motor drive in the 
time domain was developed with the aim of finding a 
balance between overall simulation duration and spec-
trum accuracy.
For modelling the DC link and power board, we em-
ployed a direct approach to extract parasitic induct-
ances between points of interest on the circuits, cre-
ating simplified matrices. This approach differed from 
using an automated Ansys® Q3D Extractor® feature, 
which generates a large matrix of lumped RLGC param-
eters. When extracting lumped motor parameters, we 
measured CM impedance through frequency sweep. 
Other parasitic elements were calculated analytically. 
This approach allowed us to maintain absolute control 
over the complexity of the matrices (i.e., to control the 
number of nodes and parasitic elements), directly ad-
dressing convergence issues and simulation duration 
in the LTspice® environment.
The implementation of the SVPWM control strategy 
improved the overall spectrum accuracy, considering 
the design of the EUT, with its large geometry result-
ing in different current paths during one electrical rota-
tion of the motor. Neglecting this phenomenon leads 
Figure 25: Comparison between measured and pre-
dicted peak detector spectrum (SVPWM).
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

201
to the spectral value of a randomly chosen static PWM 
sequence deviating from the measured spectrum of 
a rotating motor. An additional improvement in the 
simulated spectrum was achieved with a proper EMI re-
ceiver implementation. The built-in response of the IF 
filters significantly impacted the spectrum, especially 
between multiple frequency ranges of the CISPR 16-1-
1 standard.
An overview and comparison of the main factors of the 
SVPWM and static PWM methods can be seen in the 
Table 7.
Table 7: Comparison between static PWM and SVPWM 
simulation results.
Transient  
analysis 
Static 
PWM1
Static 
PWM2
SVPWM
Simulation 2 min 43 min
.tran duration 1.01 ms 22.5 ms
Spectrum delta 
peak
@150 kHz- 108 
MHz
Average = 
7,34 dBµV
Average = 
6,35 dBµV
Average = 
3,33 dBµV
Spectrum delta 
avg.
@150 kHz- 108 
MHz
Average = 
3,79 dBµV
Average = 
3,65 dBµV
Average = 
2,58 dBµV
The proposed method with the SVPWM algorithm took 
43 minutes to cover one electrical revolution of the mo-
tor, including the duration of the transient phenome-
na. Based on our experiences with developing multiple 
integrated motor drives, the resulting simulation time 
(i.e., < 1 hour) allows multiple iterations of model opti-
mization without introducing bottlenecks in the devel-
opment process. Simulation loops that take multiple 
hours or even days have limited repeatability, and their 
results are usually treated with low confidence since 
there is not enough time for parameterizing the para-
sitic elements to observe their impact on the spectrum.
The presented comprehensive strategy has a potential 
to be extended to other automotive products or even 
to other fields of interest where the accuracy of EMI 
prediction is highly dependent on changing current 
paths during the required dwell time.
7 Conflict of interest
The authors declare no conflict of interest.
8 References
1. B. Revol, J. Roudet, J. L. Schanen, and P. Loizelet, 
‘Fast EMI prediction method for three-phase in-
verter based on Laplace transforms’, in IEEE 34th 
Annual Conference on Power Electronics Specialist, 
2003. PESC ’03., 2003, vol. 3, pp. 1133–1138 vol.3, 
 https://doi.org/10.1109/PESC.2003.1216608.
2. B. Revol, J. Roudet, J.-L. Schanen, and P. Loizelet, 
‘EMI Study of Three-Phase Inverter-Fed Motor 
Drives’ , IEEE Trans. Ind. Appl., vol. 47, no. 1, pp. 223–
231, Jan. 2011, 
 https://doi.org/10.1109/TIA.2010.2091193.
3. E. Gubia, P . Sanchis, A. Ursua, J. Lopez, and L. Mar-
royo, ‘Frequency domain model of conducted EMI 
in electrical drives’ , IEEE Power Electron. Lett., vol. 3, 
no. 2, pp. 45–49, Jun. 2005, 
 https://doi.org/10.1109/LPEL.2005.848730.
4. J.-L. Kotny and N. Idir, ‘Time domain models of the 
EMI sources in the variable speed drives’ , in 2010 
IEEE Energy Conversion Congress and Exposition, 
2010, pp. 1355–1360, 
 https://doi.org/10.1109/ECCE.2010.5618276.
5. D. Zhuolin, Z. Dong, F. Tao, and W. Xuhui, ‘Predic-
tion of conducted EMI in three phase inverters by 
simulation method’, in 2017 IEEE Transportation 
Electrification Conference and Expo, Asia-Pacific 
(ITEC Asia-Pacific), 2017, pp. 1–6, 
 https://doi.org/10.1109/ITEC-AP .2017.8080876.
6. K. Wang, H. Lu, C. Chen, and Y. Xiong, ‘Modeling 
of System-Level Conducted EMI of the High-
Voltage Electric Drive System in Electric Vehicles’ , 
IEEE Trans. Electromagn. Compat., vol. 64, no. 3, pp. 
741–749, Jun. 2022, 
 https://doi.org/10.1109/TEMC.2022.3147521.
7. A. Apelsmeier, C. Rettner, and M. Marz, ‘Model 
for Conducted Emission of SiC Power Modules 
for automotive traction Inverter-Comparison to 
behaviour-based Model’, in 2020 IEEE 21st Work-
shop on Control and Modeling for Power Electronics 
(COMPEL), Aalborg, Denmark, 2020, pp. 1–8, 
 https://doi.org/10.1109/COMPEL49091.2020.9265805.
8. R. Vrtovec and J. Trontelj, ‘Advanced gate control 
system for power MOSFET switching losses re-
duction with complete switching sequence con-
trol’ , Inf. MIDEM, vol. 46, no. 4, pp. 238–249, 2016, 
https://ojs.midem-drustvo.si/index.php/InfMI-
DEM/article/view/289.
9. Y. Kwack et al., ‘EMI modeling method of interior 
permanent magnet synchronous motor for hy-
brid electric vehicle drive system considering 
parasitic and dynamic parameters’, in 2015 Asia-
Pacific Symposium on Electromagnetic Compatibil-
ity (APEMC), 2015, pp. 78–81, 
 https://doi.org/10.1109/APEMC.2015.7175390.
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202

202
10. G. Vidmar and D. Miljavec, ‘A Universal High-Fre-
quency Three-Phase Electric-Motor Model Suita-
ble for the Delta- and Star-Winding Connections’ , 
IEEE Trans. Power Electron., vol. 30, no. 8, pp. 4365–
4376, Aug. 2015, 
 https://doi.org/10.1109/TPEL.2014.2352452.
11. F. Zare, ‘Practical approach to model electric mo-
tors for electromagnetic interference and shaft 
voltage analysis’ , IET Electr. Power Appl., vol. 4, no. 
9, p. 727, 2010, 
 https://doi.org/10.1049/iet-epa.2009.0305.
12. O. Magdun and A. Binder, ‘High-Frequency Induc-
tion Machine Modeling for Common Mode Cur-
rent and Bearing Voltage Calculation’, IEEE Trans. 
Ind. Appl., vol. 50, no. 3, pp. 1780–1790, May 2014, 
 https://doi.org/10.1109/TIA.2013.2284301.
13. N. Idir, Y. Weens, M. Moreau, and J. J. Franchaud, 
‘High-Frequency Behavior Models of AC Motors’, 
IEEE Trans. Magn., vol. 45, no. 1, pp. 133–138, Jan. 
2009, 
 https://doi.org/10.1109/TMAG.2008.2006006.
14. A. Rahimi and K. Kanzi, ‘High-frequency model-
ling of permanent magnet synchronous motor 
for conducted EMI studies’ , IET Electr. Power Appl., 
vol. 14, no. 11, pp. 2027–2036, 2020, 
 https://doi.org/10.1049/iet-epa.2019.0773.
15. S. Yang, L. Zhai, and G. Hu, ‘Measurement-based 
Modeling of High Frequency Equivalent Circuit of 
Permanent Magnet Synchronous Motor of Elec-
tric Vehicle’ , in 2022 Asia-Pacific International Sym-
posium on Electromagnetic Compatibility (APEMC), 
2022, pp. 719–721, 
 https://doi.org/10.1109/APEMC53576.2022.9888573.
16. ‘CISPR 25: Vehicles, boats and internal combus-
tion engines – Radio disturbance characteristics – 
Limits and methods of measurement for the pro-
tection of on-board receivers’ . IEC, Geneva, 2016.
17. G. Ergaver and J. Trontelj, ‘Prediction of Radiated 
Emissions of Automotive Electronics Early in the 
Design Phase based on Automotive Component 
Level Testing’ , Inf. MIDEM, vol. 46, no. 1, pp. 42–52, 
Mar. 2016, https://ojs.midem-drustvo.si/index.
php/InfMIDEM/article/view/191.
18. ‘CISPR 16: Specification for radio disturbance and 
immunity measuring apparatus and methods 
–  Part 2-1: Methods of measurement of distur-
bances and immunity – Conducted disturbance 
measurements’ . IEC, Geneva, 2017.
19. T. Skuber, ‘Modeling and Optimization of Power 
Module for 48V High Power Inverter’ , Inf. MIDEM, 
vol. 51, no. 4, pp. 243–251, Dec. 2021, 
 https://doi.org/10.33180/InfMIDEM2021.404.
20. B. Wangard, ‘MIL-STD-461 Testing Advantages Us-
ing Time Domain Scan EMI Receivers’. Rohde & 
Schwarz, 2019, https://scdn.rohde-schwarz.com/
ur/pws/dl_downloads/premiumdownloads/pre-
mium_dl_application/MIL-STD-461_Testing-Ad-
vantages-Using-Time-Domain-Scan-EMI-Recei-
vers.pdf.
21. ‘CISPR 16: Specification for radio disturbance and 
immunity measuring apparatus and methods –  
Part 1-1: Radio disturbance and immunity mea-
suring apparatus – Measuring  apparatus’. IEC, 
Geneva, 2015.
22. F. Krug and P . Russer, ‘Signal processing methods 
for time domain EMI measurements’ , in 2003 IEEE 
International Symposium on Electromagnetic Com-
patibility, 2003. EMC ’03., 2003, vol. 2, pp. 1289-
1292 Vol.2, 
 https://doi.org/10.1109/ICSMC2.2003.1429156.
23. S. Braun, A. Frech, and P. Russer, ‘CISPR specifica-
tion and measurement uncertainty of the time-
domain EMI measurement system’, in 2008 IEEE 
International Symposium on Electromagnetic Com-
patibility, Detroit, MI, 2008, pp. 1–4, 
 https://doi.org/10.1109/ISEMC.2008.4652078.
24. G. Ergaver, ‘Raziskava elektromagnetnih seval-
nih emisij elektronskih naprav za avtomobilsko 
področje’, PhD diss., University of Ljubljana, Lju-
bljana, 2017, https://repozitorij.uni-lj.si/IzpisGrad-
iva.php?lang=slv&id=92362.
Arrived: 03. 01. 2024
Accepted: 06. 02. 2024
K. Saksida et al.; Informacije Midem, Vol. 53, No. 3(2023), 191 – 202
Copyright © 2023 by the Authors. 
This is an open access article dis-
tributed under the Creative Com-
mons Attribution  (CC BY) License (https://creativecom-
mons.org/licenses/by/4.0/), which permits unrestricted 
use, distribution, and reproduction in any medium, 
provided the original work is properly cited.