Académique Documents
Professionnel Documents
Culture Documents
by
in
(Vancouver)
June 2014
Modeling and analysis of basic DC-DC converters is essential for enabling power-
electronic solutions for the future energy systems and applications. Average-value modeling
(AVM) provides a time-efficient tool for studying power electronic systems, including
DC/DC converters. Many AVM techniques including the analytical and numerical state-
space averaging and circuit averaging have been developed over the years and available in
the literature. Average-value modeling of ideal PWM converters neglects parasitics (losses)
to simplify the derivations and modeling procedures, and the resulting models may not be
sufficiently accurate for practical converters.
In this work, first we consider a second-order Flyback converter, which has
transformer isolation and additional parasitics such as conduction losses that have not been
accurately included in the prior literature. We propose three new AVMs using the analytical
state-space averaging, circuit averaging, and parametric AVM approaches, respectively. By
taking into account conduction losses, the accuracy of the proposed average-value models is
significantly improved. The derived (corrected) models show noticeable improvement over
the traditional (un-corrected) models.
Next, we consider the Flyback converter including the snubbers and leakage
inductances in the full-order model. Snubbers reduce electromagnetic interfaces (EMI)
during transients and protect switching devices from high voltage, and therefore are
necessary in many practical converter circuits. Including snubbers into the model improves
accuracy in predicting the circuit variables during the time-domain transients as well as
predicting the converter efficiency. It is shown that conventional analytical/numerical
methods of averaging do not result in accurate AVM for the full-order Flyback converter. A
new formulation for the state-space averaging methodology is proposed that is functional for
higher-order converters with parasitics and result in highly accurate AVM. The new
approach is justified mathematically and verified experimentally using hardware prototype
and measurements. The new model is demonstrated to achieve accurate results in large signal
time-domain transients, and small-signal frequency-domain analysis in continuous
conduction mode (CCM) and discontinuous conduction mode (DCM), which represents
advancement to the state-of-the-art in this field.
ii
Preface
The research results presented in this thesis have been already published in several
conference proceedings and/or ready for submission as a journal article. In all publications, I
was responsible for developing the formulations, implementing models, doing simulations
and hardware tests, compiling results, as well as preparing the majority of the manuscripts.
My research supervisor, Dr. Juri Jatskevich, has provided supervisory comments and
corrections during the process of conducting research and writing the manuscripts. The other
co-authors have also provided constructive comments and feedback.
iii
Table of Contents
Abstract .................................................................................................................................... ii
iv
2.3.1 Parametric Average Value Modeling ............................................................................. 28
2.3.2 Numerical Average Value Modeling Case Studies ........................................................ 32
2.3.2.1 Time-Domain Transient Studies ............................................................................ 32
2.3.2.2 Frequency-Domain Analysis.................................................................................. 37
Bibliography .......................................................................................................................... 68
Appendices ............................................................................................................................. 73
Appendix A. Second-order Flyback Converter State-space Matrices ........................................... 73
Appendix B. Converters Circuit Parameters ................................................................................. 74
B.1 Second-order Flyback Converter Parameters with Basic Parasitics ............................... 74
B.2 Full-order Flyback Converter Parameters with all Parasitics and Snubbers................... 74
Appendix C. Hardware Flyback Converter Prototype Circuit Diagram ....................................... 75
v
List of Tables
Table 2.1 Output voltage and inductor current values and errors as predicted by
different averaged models in DCM and CCM. .................................................... 24
Table 2.2 Eigenvalues of average-value models in DCM and CCM. ..................................... 25
Table 2.3 Efficiency comparison of average-value models. ................................................... 25
Table 2.4 Output voltage and inductor current as predicted by various models for
the two steady state operating points. .................................................................. 34
Table 2.5 Converter efficiency as predicted by various models for the two steady
state operating points. .......................................................................................... 34
Table 2.6 Simulation speed comparison of the detailed and average models in
closed-loop transient study................................................................................... 36
Table 3.1 Eigenvalues of AVMs. ............................................................................................ 56
Table 3.2 Hardware prototype and detailed model comparison in terms of input
current, output voltage and efficiency.................................................................. 57
Table 3.3 Accuracy precision of the proposed PAVM in predicting steady state
variables. .............................................................................................................. 59
Table 3.4 Accuracy precision of the proposed PAVM in predicting converter
efficiency. ............................................................................................................. 59
vi
List of Figures
vii
Figure 2.12 Typical inductor and switch waveform assuming CCM. .................................... 21
Figure 2.13 Output voltage and inductor current transients during load change as
predicted by various models. ............................................................................... 23
Figure 2.14 Control-to-output transfer function magnitude and phase in DCM as
predicted by various models. ............................................................................... 27
Figure 2.15 Control-to-output transfer function magnitude and phase in CCM as
predicted by various models. ............................................................................... 28
Figure 2.16 Correction coefficient m1 for the example second-order Flyback
converter............................................................................................................... 30
Figure 2.17 Correction coefficient m2 for the example second-order Flyback
converter............................................................................................................... 30
Figure 2.18 Duty-ratio constraint d2 for the example second-order Flyback
converter............................................................................................................... 31
Figure 2.19 Inverted matrix condition number for the example second-order
Flyback converter. ................................................................................................ 31
Figure 2.20 Implementation of PAVM. .................................................................................. 32
Figure 2.21 Output capacitor voltage and inductor current transients due to
change in the switch duty cycle as predicted by various models. ........................ 33
Figure 2.22 Closed-loop system of the considered second-order Flyback
converter with PI controller to regulate the output voltage. ................................ 34
Figure 2.23 Closed-loop second-order Flyback converter with PI controller
response to a load change..................................................................................... 35
Figure 2.24 Control to output transfer function magnitude and phase. .................................. 37
Figure 3.1 Averaged detailed circuit of input stage of Flyback converter in steady
state. ..................................................................................................................... 39
Figure 3.2 Input stage of Flyback converter depicting operation of snubber. ........................ 40
Figure 3.3 Averaged detailed circuit of the output stage Flyback converter
without the snubber in steady state. ..................................................................... 41
Figure 3.4 Output stage topologies of Flyback converter without output snubber. ................ 42
Figure 3.5 Averaged detailed circuit of output stage of full-order Flyback
converter in steady state. ...................................................................................... 43
viii
Figure 3.6 Output side of full-order Flyback converter depicting the snubber
operation............................................................................................................... 44
Figure 3.7 Inverted matrix condition number. ........................................................................ 49
Figure 3.8 Correction coefficient m1 for voltage of output capacitor vC . .............................. 50
Figure 3.9 Correction coefficient m2 for voltage of primary snubber capacitor
vCss . ...................................................................................................................... 50
Figure 3.10 Correction coefficient m3 for primary side leakage inductor current
iLpt . ....................................................................................................................... 51
Figure 3.11 coefficient m4 for secondary side leakage inductor current iLst . ......................... 51
Figure 3.12 Calculated function of duty-ratio constraint d2. .................................................. 52
Figure 3.13 Correction coefficient m1 for voltage of output capacitor vC . ............................ 52
Figure 3.14 Correction coefficient m2 for voltage of primary snubber capacitor
vCss . ...................................................................................................................... 52
Figure 3.15 Correction coefficient m3 for voltage of output snubber capacitor
vCds . ...................................................................................................................... 53
Figure 3.16 Correction coefficient m4 for primary side leakage inductor current
iLpt . ....................................................................................................................... 53
Figure 3.17 Correction coefficient m5 for secondary side leakage inductor current
iLst . ....................................................................................................................... 53
Figure 3.18 Block diagram depicting implementation of proposed extended
PAVM. ................................................................................................................. 55
Figure 3.19 Measured and detailed model waveforms for the considered
operating point. .................................................................................................... 57
Figure 3.20 Transients of state variables of fourth-order Flyback converter as
predicted by uncorrected PAVM and the proposed corrected PAVM................. 61
Figure 3.21 Transients of state variables of full-order Flyback converter due to
load change. ......................................................................................................... 62
Figure 3.22 Circuit state variables transients due to intense increase of duty cycle. .............. 64
ix
Figure 3.23 Control-to-output transfer function magnitude and phase in DCM as
predicted by proposed PAVM and detailed model. ............................................. 65
x
List of Abbreviations
xi
Acknowledgements
I consider it an honor to have Dr. John Madden and Dr. Jose Marti as my committee
members who dedicated their time to provide me constructive comments and take part in my
M.A.Sc. examination.
I would like to thank the following friends and colleagues from the Electrical Energy
Systems Lab at UBC for their continued support and helpful insight toward the problems that
I faced in the course of my research project: Francis Therrien, Hamid Atighechi, Mehmet
Sucu, and Mehrdad Chapariha.
Last but not the least, I would like to express my deepest gratitude to my parents for
their patience while I was abroad, and for supporting me spiritually throughout my life.
xii
Chapter 1 : Introduction
Averaged
Detailed
Figure 1.1 A detailed waveform containing switching ripple and the corresponding average-value
1
1.1.1 State-Space Averaging
State-space averaging is a straightforward and general method of average-value
modeling proposed first in [6] and later extended in [7]-[9], as well as many other
publications. The SSA is used for large and small signal analysis of DC/DC converters. In
state-space description of a system, the differential equations describing the converter circuit
are written in canonical form. Extracting these equations manually is complicated for high-
order converters with parasitics. Therefore, simulation and software tools that can
automatically extract the state-space equation (matrices) become very useful. SSA is valid
when natural frequencies (poles) of system are much lower than the switching frequency of
the circuit [10]. Due to a well-defined structure of this method, its equations can be easily
transformed to the frequency-domain. A corrected full-order SSA model for basic converters
has been introduced in [9]. One of the main challenges in this method is calculating the
length of each subinterval in a switching cycle. When parasitics are placed in the system,
calculation of the duty ratio constraint becomes more complicated. However, a good AVM
should include parasitics so as to approximate the hardware prototype with good accuracy
[11]. Including the parasitics in the state-space AVMs analytically has been considered in
[12] and [13] for basic converters. A recently proposed numerical SSA approach is based on
calculating the duty ratio constraint and correction terms numerically.
2
circuit, which improves the model accuracy. In this thesis, a corrected CA model is derived
for the Flyback converter circuit that takes into account the basic conduction losses and
properly respects the energy conservation principle [17]-[19].
3
between the input and output. Galvanic isolation separates grounds with different potentials
and prevents unwanted current flow, e.g. in the case of output short circuit. For miniaturizing
transformer size, converter circuit should operate at high frequency [25], which in turn
results in increase of leakage inductances. Therefore, it becomes necessary to use snubbers in
the circuit in order to alleviate the hard-switching or non-zero current/voltage switching
problem that would happen otherwise. A snubber is not a fundamental part of a power
electronic converter circuit, and adding it to a semiconductor device reduces the stresses to a
level that is tolerable according to that device electrical ratings [37]. The considered full-
order Flyback converter is illustrated in Figure 1.2. This circuit has both primary and
secondary snubbers. The RC snubbers [39] in the circuit help MOSFET and diode operate
under hard-switching conditions. As the converter works with a relatively high switching
frequency (e.g., 250 KHz for the hardware prototype considered in this work), the absence of
snubbers in the circuit would cause failure of MOSFET and generate broadband noise that
can cause problem for data transmission equipment [25]. These dissipative RC snubbers used
dv di
in the circuit reduce the stress on the switches, reduce and , etc. [40]-[41].
dt dt
Figure 1.2 Full-order Flyback converter circuit with snubbers on primary and secondary sides.
4
1.3 Motivations and Objectives
Different power converters have various challenges when it comes to average-value
modeling. In this thesis, I am focusing on the Flyback converter for which only very
simplified average models have been developed in the prior literature. Unlike the classical
DC/DC converters i.e. Buck, Boost, Buck/Boost, Cuk, etc., for which there have been
extensive prior research with advanced results, the Flyback converter includes a transformer
and two RC snubbers in its circuitry, which represents additional challenges for developing
the accurate dynamic models. To the best of our knowledge, these challenges have not been
addressed in the published literature. Thus, the purpose of this work is to develop accurate
and straightforward-to-use AVMs for Flyback converter for both widely used approaches,
CA and SSA. Specifically, the objectives of this research project are:
5
snubber (for protecting diode). The presence of these snubber circuits has been omitted in
most, if not all, prior work on AVM, due to analytical complexity that they represent. It is
very desirable to understand the effects of snubbers for the analytical and PAVM
methodologies, and to develop a generalized methodology that is capable of automatically
including both the conduction parasitics as well as the energy losses due to all possible
snubber circuits.
6
Chapter 2 : Second-order Flyback Converters
As shown in Figure 1.2, the complete full-order Flyback converter circuit includes the
snubbers, the leakage inductors of the transformer, and the conduction losses of the two
switches. Oscillation of electrical energy between inductive and capacitive elements in this
circuit in the form of ringing at switching instants makes the process of deriving the AVMs
particularly challenging. Therefore, in the first step, we consider just very basic conduction
parasitics as depicted in Figure 2.1. The resulting circuit will have only two energy storing
elements, and correspondingly will have a second-order. Such simplified Flyback converter
circuit has been considered in many classical and contemporary research literature sources.
Rpt vfd
Rst
n
iL(t)
Lm Rload
vg(t) C
Mosfet Rc
Rsw
Figure 2.1 Simplified second-order Flyback converter with basic conduction parasitics.
ipk
iL, A
iL
8
iL(t)
load
iL(t)
load
iL(t)
load
iL(t)
load
Figure 2.3 Assumed topological instances for the second-order Flyback converter without parasitics; (a)
original circuit; (b) circuit during subinterval 1; (c) circuit during subinterval 2; and (d) circuit during
subinterval 3.
9
Herein, in-order to accurately represent the dynamics of the converter circuit, a
corrected state-space average-value model (CSS AVM) is considered, wherein the correction
[ ]
matrix, M = diag (d1 + d 2 ) , 1 , is defined based on [9] to compensate for the errors in
1
10
resistors and diode voltage drop are included, the value of the current peak would drop from
i p to ip .
ip
i'p
iL , A
Ts
0
Figure 2.4 Effect of basic parasitics on inductor waveform and its peak in DCM.
Therefore, to include the effect of basic parasitics in the model, the MOSFET ON-
resistance R sw , the transformer primary and secondary side (referred by the turns-ratio a)
resistances R pt and R st , are included using an equivalent resistor depicted the second-order
Flyback converter circuit illustrated in Figure 2.5. Herein, the equivalent resistor is calculated
as
Req = Rsw + R pt + a 2 Rst . (2.6)
v
fd
p'
n
iL(t)
Lm R
load
v (t)
C
g
Rc
Req
Figure 2.5 Simplified second-order Flyback converter with basic conduction parasitics represented as an
equivalent resistor in the primary side.
11
With presence of parasitic resistors in the circuit, the voltage drop across the
magnetizing inductance is no longer equal to von and an equation for the loop on the primary
side becomes
diL
L + Req iL = von . (2.7)
dt
Calculating the local average of (2.7) over the first subinterval, results in the following
d1Ts d1Ts d1Ts
L diL R 1
d + eq iL ( )d = v on ( )d , (2.8)
Ts 0
dt Ts 0
Ts 0
As the waveforms in Figure 2.2 are assumed to be piece-wise linear, (2.9) is further
simplified to
L d1 Req
ip + i p = d1von . (2.10)
Ts 2
Based on (2.10), the new corrected peak of the inductor current shown in Figure 2.2 is
calculated as
d1von
ip = . (2.11)
L d1 Req
+
Ts 2
Substituting (2.11) in (2.2) and solving for d 2 gives the following result
2L
( + Req )i L
d1Ts
d2 = d1 . (2.12)
von
Similar to (2.5) for ideal converter, a general expression of the duty ratio constraint for both
DCM and CCM is obtained as
2L
( d T + Req )iL
d 2 = min 1 s d1 , 1 d 1 . (2.13)
von
12
2.1.1 State-Space Averaging Case Studies
The detailed model of the Flyback converter shown in Figure 2.5 with parameters
summarized in Appendix in B.1. is used as reference for validating the proposed average
value model. Both detailed model and average-value model have been implemented in
Matlab toolboxes, specifically PLECS [42] and Simulink [43] for each model respectively.
Improvements of state-space average-value model are demonstrated in the following
sequential steps.
In the first step, the necessity of using the correction matrix M is shown. After that,
the importance of including parasitics and dissipative elements is explained by efficiency
studies over a wide range of circuit operating conditions. Finally, a proportional-plus-integral
(PI) controller is added to the models, and closed-loop system's ability to regulate the output
voltage is examined.
13
-140
Detailed
SSA
CSSA
-145
v ,V
C
See lower plot
-150
-155
-150.96
-150.97
,V
-150.97
C
v
-150.98
-150.98
-150.99
2
1.5
1
i ,A
0.5
L
-0.5
0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
Time, s
Figure 2.6 Steady state output voltage and inductor current in DCM as predicted by various models: (a)
detailed model; (b) state-space averaged-value model (SS-AVM); and (c) corrected state-space averaged-
value model (CSS-AVM.).
As Figure 2.6 shows, the corrected state-space averaged-value model has a much
better accuracy in computing the average-value of the circuit state variables compared with
the uncorrected version of the model. According to Figure 2.6, the average inductor current
and output voltage predicted by the CSSA-AVM pass through the ripples of the detailed
model as opposed to uncorrected SSA-AVM. Therefore, in the future investigations in this
thesis, we consider the CSSA-AVM as a basis.
14
2.1.1.2 Effect of Including Parasitics on Predicted Efficiency of Converter
Since calculation of efficiency requires several circuit variables, the predicted
efficiency can be considered for evaluating accuracy of different AVMs. In this Subsection,
we summarize the results obtained from many time domain studies for calculating the
efficiency as predicted by each model. Figure 2.7 demonstrates the results obtained by the
CSSA-AVMs with and without considering the transformer and switch equivalent
resistances.
In Figure 2.7, the duty cycle is kept constant at 0.6 and the load is changed from
500 to 2300 in steps of 200 . In Figure 2.8, the load is kept constant at 700 and the
duty cycle is changed from 0.1 to 0.9 . As can be seen in these figures, the CSSA-AVM that
includes the equivalent parasitics resistance predicts the same efficiency as the corresponding
detailed model. At the same time, the CSSA-AVM that does not include the equivalent
parasitics resistance predicts unrealistically high efficiency.
100
95 (b)
Efficiency
(a) Detailed
90 (a),(c) (b) CSS Avg w/o Req
(c) CSS Avg with Req
85
80
600 800 1000 1200 1400 1600 1800 2000 2200
Rload, Ohm
Figure 2.7 Efficiency prediction by various models as a function of the load resistance: (a) detailed model;
(b) CSS-AVM without equivalent parasitic resistance; and (c) CSS-AVM with equivalent parasitic
resistance.
15
100
(b)
80 (a),(c)
Efficiency
(a) Detailed
(b) CSS Avg w/o Req
60 (c) CSS Avg with Req
40
controller is set to regulate output the voltage at 70V . Initially, the converter is assumed to
operate in DCM loaded by 2200 resistor. At t = 0.35 sec , the load is changed by
connecting another resistor of 150 in parallel to the original load.
As it can be observed in Figure 2.10, the inductor current significantly increases and
the converter mode changes from DCM to CCM. Both current and voltage undergo transient
oscillations and settle at a new steady state in CCM. As it can be seen, the considered CSSA-
AVM that includes the same equivalent parasitic resistance as the detailed model predicts the
entire transient response of the detailed model with good accuracy.
16
-67
Detailed
-68
CSSA
v ,V
-69
C -70
-71
10
6
i ,A
L
0
0.35 0.351 0.352 0.353 0.354 0.355 0.356 0.357
Time, s
Figure 2.9 Output capacitor voltage and inductor current responses to the step change in load for a
closed-loop system.
and secondary side (referred by the corresponding turns-ratio) resistances R pt and Rst , are
added up together and represented by an equivalent resistor defined in (2.6). For convenience
of analysis, the element Req is brought outside the switching cell as shown in Figure 2.10.
vfd
p p'
n
c
iL (t)
Lm Rload
vg(t)
C v
(t)
Rc
Req
a' a b
18
iL t [0, d1Ts ]
isw = 0 t [d1Ts , ( d1 + d 2 )Ts ], (2.14)
0
t [( d1 + d 2 )Ts , Ts ]
d1i pk d1iL
isw = = , (2.16)
2 d1 + d 2
Finally, to get the equivalent average circuit cell, the transistor and diode are replaced
by dependent current and voltage sources as shown in Figure 2.11. The values of these
sources are defined by (2.16) and (2.17) respectively. To get the final AVM, the derived
averaged cell of Figure 2.11 is inserted back into the original converter circuit shown in
Figure 2.10. Transformers are designed to transmit AC voltages but no the DC voltages. In
the following figure, the transformer represents a dependent voltage source wherein the
conversion ratio is equal to n ( i.e. the output voltage is equal to n times input voltage).
c p
n
iL(t) vdiode
Lm
isw
a b
Figure 2.11 Resulting equivalent averaged cell for the second-order Flyback converter.
t
1 d1
(isw ) rms = i (t )dt =
2
L i pk . (2.19)
Ts t Ts
3
Therefore, in-order to have the same energy loss in Req as in the detailed model, an
The energy conservation principle implies that, in each switching cycle, the energy dissipated
in the averaged circuit has to be equal to its actual value as it would be predicted by the
, for
detailed switching circuit. We therefore introduce a modified equivalent resistance, Req
which the energy conservation principle holds. Thus, considering (2.18) and (2.19), the
following energy balance should hold:
(isw )2ave Req Ts = (isw )2rms ReqTs (2.20)
which yields the following formula for calculating the modified equivalent resistance,
4 Req
=
Req . (2.21)
3d1
Finally, since a new resistance is introduced to the converter circuit, the equation for
the duty ratio constraint d 2 needs to be modified as well. In particular, when the switch is
closed, the voltage across the inductor is
4 Req iL
von = vg . (2.22)
3 d1 + d 2
20
Substituting (2.22) into (2.4), we obtain
2 Li L
d2 = d1 . (2.23)
4 ReqiL
d1Ts (v g )
3 d1 + d 2
Finally, isolating d 2 on one side, an explicit expression for the new duty ratio constraint is
derived,
2L 4
( + Req )iL
d1Ts 3
d2 = d1 . (2.24)
vg
Equation (2.21) is derived on the basis of DCM. As shown in Figure 2.12, in CCM,
the inductor current never goes to zero and the switching interval is divided into two
subintervals. Therefore, (2.18) and (2.19) are no longer valid for CCM, and new relationships
need to be re-derived.
iL , A
iL
ave d1Ts d2Ts
Ts
0
isw , A
ipk
iL
ave
d1Ts d2Ts
Ts
0
21
The RMS value of the switch current is
2
d1 / 2
i pk
(isw ) rms = t + (iL ) ave dt , (2.26)
d1 / 2
2 d1
which after taking the integration and simplifying manipulations gives
i pk
2
(isw ) rms
= d1 (iL ) ave +
2 . (2.27)
12
In (2.27), i pk denotes the current ripple that is shown in Figure 2.12, and is calculated based
Req
2
i pk
=
Req 1 + . (2.29)
d1 12(iL ) ave
2
Depending on the converter mode of operation, either (2.21) or (2.29) will be the correct
resistor value such that the energy conservation principle accurately predicts/captures the
energy dissipation in both DCM and CCM, respectively.
22
2.2.4.1 Time-Domain Transient
An informative time-domain study for analysis of PWM converters should span
different conduction modes. In the transient study considered in this Subsection, the duty
cycle is kept constant at d1 = 0.381 . Initially, the converter is assumed to operate in a steady
state in DCM under load Rload = 2500 . Then, at t = 0.3s , a parallel resistor R p = 200 is
added to the load of the system. The converter undergoes a large transient and the mode of
operation changes to CCM. The corresponding transient responses produced by the
considered models are superimposed in Figure 2.13. As can be seen in Figure 2.13, the
decrease in load resistance results in decrease of the output voltage. The actual values of the
output voltage and inductor current in steady state in DCM (before transient) and in CCM
(after transient) as predicted by the considered models are summarized in Table 2.1.
-60
-70 (d)
(b)
(a),(c),(e)
-80
vout , V
(a)Detailed
-90 (b) C Avg
(c) CC Avg
(d) SS Avg
-100 (e) CSS Avg
-110
5
(b)
4
3 (d)
iL , A
2 (a)
(c),(e)
0
0.3 0.301 0.302 0.303 0.304 0.305
Time, s
Figure 2.13 Output voltage and inductor current transients during load change as predicted by various
models.
23
Table 2.1 Output voltage and inductor current values and errors as predicted by different averaged
models in DCM and CCM.
Output voltage Inductor current
Model
DCM CCM DCM CCM
Detailed Model Value -101.87 -66.68 0.46 3.49
Value -102.61 -70.26 0.46 3.68
CA-AVM
Error 0.81% 5.36% 0.62% 5.34%
Value -101.35 -66.71 0.45 3.5
CCA-AVM
Error 0.42% 0.03% 0.61% 0.01%
Value -94.17 -66.69 0.48 3.49
SSA-AVM
Error 7.48% 0.01% 4.19% 0.02%
Value -101.79 -66.69 0.46 3.49
CSSA-AVM
Error 0.01% 0.01% 0.18% 0.02%
also indicates that when value of Req is not corrected (CA-AVM), the inductor current takes
a longer time to settle down to its steady state value and it oscillates with large amplitude
during the transient. But in corrected averaged circuit (CCA-AVM), the value of Req is larger
(the energy is dissipated faster), and no oscillations are observed. It can also be noted that
this model response is somewhat slower (more dissipative) than what is predicted by the
detailed model.
The results from conventional state-space averaging (SSA-AVM) method, wherein no
correction matrix is added, and that of the corrected state-space averaging (CSSA-AVM), are
also shown in Figure 2.13 and Table 2.1. As indicated in Table 2.1, the absence of correction
matrix in conventional state-space average-value model causes an obvious error in DCM in
predicting state variables average value even in steady state. This source of error is
significantly reduced when the state-space model is corrected (CSSA-AVM).
To get more insight into the difference between the corrected and un-corrected
circuit-averaged models, their eigenvalues have been extracted in DCM and CCM and are
summarized in Table 2.2. As it can be seen in Table 2.2, both models have very similar real
eigenvalues in DCM, which implies that the system will not have oscillatory behavior, as is
verified in Figure 2.13. However, when the converter is in transient going into CCM, the CA-
24
AVM shows some oscillations. As can be seen in Table 2.2, the eigenvalues in CCM are
complex, and the imaginary part of complex eigenvalues for the CA-AVM is dominant,
which explains the oscillations. At the same time, the CCA-AVM in CCM has very large
negative real part of the eigenvalues, which explains its significantly more damped response.
25
It can also be noted in Table 2.3 that corrected circuit averaging (CCA-AVM) model
over-estimates the losses in CCM. This model also has higher damping in CCM as compared
to the other models based on the transient study in Figure 2.13 and the eigenvalues in Table
2.2. This phenomenon is attributed to the fact that the energy conservation correction
becomes over-compensated (too high) when the
assumed the DCM, and the value of Req
26
100
Detailed
Magnitude, dB
C Avg
50 CC Avg
SS Avg
0 CSS Avg
-50
200
Phase, deg
150
100
50
0 1 2 3 4 5
10 10 10 10 10 10
Frequency, Hz
Figure 2.14 Control-to-output transfer function magnitude and phase in DCM as predicted by various
models.
27
100
Detailed
Magnitude, dB
C Avg
50
CC Avg
-50
200
100
Phase, deg
-100
0 1 2 3 4 5
10 10 10 10 10 10
Frequency, Hz
Figure 2.15 Control-to-output transfer function magnitude and phase in CCM as predicted by various
models.
28
Unlike (2.1) where the correction matrix is derived analytically, here M is a diagonal
correction matrix which is computed numerically; dk is the relative length of subintervals for
each topology within the switching interval; A k and B k are state-space matrices; u is the
input vector; and x is the vector of averaged state variables [44]. In CCM, there is no third
subinterval and k = 1 , 2 .
When the state variables are in steady state condition, their derivatives are equal to
zero. Thus, under steady state condition, (2.30) becomes equal to zero, yielding
k =3 k =3
0 = d k A k M x + d k B k u . (2.31)
k =1 k =1
Solving (2.31) for M x gives
1
k =3 k =3
p = M x = d k A k d k B k u . (2.32)
k =1 k =1
Using (2.32), the following expression can be utilized to calculate M (d1 , iL ) for each
operating point
p( j )
M ( j, j ) = . (2.33)
x( j)
numerical robustness of the procedure depends on the condition number of this matrix. The
29
condition number of this matrix has been calculated and is depicted in Figure 2.19. As can
be seen in Figure 2.19, the condition number remains relatively small over the range of
considered operating conditions, suggesting that the error resulting from the matrix inversion
is negligible.
2
DCM
1.5 CCM
1
m
1
4000
2000
R, Ohms
0 0.6 0.8 1
0 0.2 0.4 d
1
Figure 2.16 Correction coefficient m1 for the example second-order Flyback converter.
1.1
1.05
m2
1
0.95
4000
1
2000 0.5
R, Ohms d
0 0 1
Figure 2.17 Correction coefficient m2 for the example second-order Flyback converter.
30
1
2
d 0.5
0
0
d 0.5 4000
1 2000 3000
1 0 1000 R, Ohms
Figure 2.18 Duty-ratio constraint d2 for the example second-order Flyback converter.
400
Condition number
200
0
4000
1
2000 0.5 d
R, Ohms 1
0 0
Figure 2.19 Inverted matrix condition number for the example second-order Flyback converter.
The output variables in PAVM are calculated on the basis of the following equation
k =3 k =3
y = d k C k M x + d k D k u . (2.34)
k =1 k =1
After all parametric functions have been calculated numerically, the AVM is
implemented according to the block diagram depicted in Figure 2.20.
31
Figure 2.20 Implementation of PAVM.
model of the considered Flyback converter has been implemented in PLECS, and the AVMs,
including PAVM and CA-AVM (Examined in section 2.2), have been implemented in
Matlab/Simulink.
Rload = 2500 . Then, the duty cycle is increased from 0.3 to 0.8 over a period of
0.05 second. After the transition, the converter operates in CCM with the new duty cycle.
Figure 2.21 shows the state variables of the detailed and average models during this
transition. In-order to compare the precision of average models, their predicted inductor
current and output voltage for the DCM and CCM are summarized in Table 2.4. In addition,
the converter efficiency predicted by each model is calculated and summarized in Table 2.5.
As can be observed in Figure 2.21, the results of the detailed model imply that when
converter duty cycle is changed, the magnitude of both output capacitor voltage and inductor
current increases after undergoing some transient. According to plots in Figure 2.21, the state
32
variables predicted by the PAVM is very accurate and completely overlaps with average
values from the detailed model. However, voltage predicted by the circuit averaging model
(CA-AVM) clearly deviates from the detailed model during transients and in steady state at
the end of simulation. Similar discrepancies are observed for the inductor current. In fact,
these deficiencies in the circuit averaging model have two reasons. First, the derived
averaged circuit in [27] is valid for DCM and light CCM operating conditions, while in our
case study after the increase in duty cycle the converter is working in heavy CCM. Second,
the energy conservation principle is not used in this circuit averaging model, and the CA-
AVM dissipates less power than that of detailed model.
Det. Sim.
-100 C Avg
PAVM
vC , V
-200
-300
-400
10
6
iL , A
0
0.3 0.32 0.34 0.36 0.38
Time, s
Figure 2.21 Output capacitor voltage and inductor current transients due to change in the switch duty
cycle as predicted by various models.
33
Table 2.4 Output voltage and inductor current as predicted by various models for the two steady state
operating points.
Output voltage Inductor current
Model
DCM CCM DCM CCM
Det. Model Value -80.23 -332.38 0.33 3.2
Value -80.81 -338.72 0.33 3.25
CA-AVM
Error 0.73% 1.91% 0.59% 1.7%
Value -80.23 -332.38 0.33 3.2
PAVM
Error 0% 0% 0% 0%
Table 2.5 Converter efficiency as predicted by various models for the two steady state operating points.
Efficiency DCM CCM
Det. Model 97.14% 92.04%
CA-AVM 98.17% 94.05%
PAVM 97.46% 92.31%
In the next study, we test the PAVM functionality in a closed-loop system with
commonly-used PI controller [45]. Here, a PI controller with parameters K i = 0.05 and
K p = 110 is chosen. In-order to prevent error accumulation during the rise time, the output of
controller integrator is limited by a saturation block. The controller diagram together with the
converter plant is shown in Figure 2.23.
ki
s
Ref. Converter Model Output
kp
Figure 2.22 Closed-loop system of the considered second-order Flyback converter with PI controller to
regulate the output voltage.
In the following closed-loop study, load is initially set to Rload = 220 and converter
is assumed to operate in CCM. Then, at t = 0.2 sec, the load Rload is increased to 1800 . As
Figure 2.23 illustrates, the controller is able to regulate the output voltage at 70 V within a
34
short settling time and relatively small overshoot. Also, as can be seen in Figure 2.23, the
results predicted by the PAVM are in very good agreement with the detailed model.
-65
Det. Sim.
PAVM
vC , V
-70
-75
4
2
i ,A
L
0.40
0.35
Control signal
0.3
0.25
0.2
0.19 0.195 0.2 0.205 0.21 0.215 0.22
Time, s
Figure 2.23 Closed-loop second-order Flyback converter with PI controller response to a load change.
To give the reader an idea how the simulation speed of the AVM may differ from that
of the original detailed switching model, Table 2.6 compares the simulation speed of models
in the last study. In order to make the comparison reasonable, the same solver Ode23s with
relative/absolute tolerances set to 1e-3 were used for each model. As can be seen in Table
35
2.6, the detailed model handles all switching events and requires very large number of time
steps (90,8313) and takes appropriately long time (70.43s) to complete the study. At the same
time, the PAVM is not switching, which allows it to use larger time steps (taking only 2,397
time steps) and complete the same study much faster (0.35s). This very significant increase in
simulation speed while preserving the slower dynamics at the control and input-output
terminals of the converter suggest that such AVMs can be used very effectively for the
system-level studies where the focus is on the input-output interactions and controller design,
and where the details of the switching waveforms can therefore be neglected.
Table 2.6 Simulation speed comparison of the detailed and average models in closed-loop transient study.
Model # of time steps Elapsed time, s
Det. Sim. 908,313 70.43
PAVM 2,397 0.35
36
2.3.2.2 Frequency-Domain Analysis
To demonstrate the performance of the AVMs in frequency-domain, we consider the
control-to-output voltage transfer function. The considered operating point of interest in
defined by the steady state operating condition with d1 = 0.381 and Rload = 2500 , with
corresponds to the DCM. The corresponding transfer functions have been extracted with
Matlab/Simulink and PLECS toolboxes, and the results are shown in Figure 2.24.
As Figure 2.24 shows, the derived CA-AVM and PAVM are in a good agreement
with the detailed model at frequencies up to around 100 KHz. At frequencies above 100
KHz, the AVMs start deviating from the detailed model a little bit. This is because the
primary assumptions used for deriving the average models are valid below the switching
frequency.
100
Detailed
Magnitude, dB
C Avg
50 PAVM
-50
200
150
Phase, deg
100
50
0
0 1 2 3 4 5
10 10 10 10 10 10
Frequency, Hz
Figure 2.24 Control to output transfer function magnitude and phase.
37
Chapter 3 : Full-order Flyback Converter
Snubbers bring many benefits to the hardware circuit (Figure 1.2) as they enable
semiconductor components to switch in a softer mode and reduce the voltage stresses. At the
same time, snubbers also increase power loss of the circuit and make modeling of power
electronic systems more complicated. The full-order Flyback converter would typically
include two snubbers: one snubber is used on the input side to protect MOSFET; and another
one is used on the output side of converter to absorb spikes resulting from the secondary side
leakage inductance.
In this work, it is shown that due to the presence of RC snubbers of in Flyback
converter, the previously established SSA-AVM method is not capable of presenting an
accurate AVM. Therefore, a new PAVM is developed which corrects the average values of
state variables in a more general different way. The methodology is based on extending the
SSA formulation to include the corrections are of the state variables as well as the state-space
matrices (their specific entries corresponding to snubbers).
38
Rpt iLpt
Output
iLm
vg(t)
Input
vCss vsw
RCss Rsw
Figure 3.1 Averaged detailed circuit of input stage of Flyback converter in steady state.
According to Figure 3.2, in the first topology (k = 1) , when the switch is closed, the
input voltage source transfers energy to the magnetizing and leakage inductors as the current
increases. During this subinterval, the snubber does not play a significant role, and the
capacitor energy is dissipated on the resistors of snubber and switch. During this subinterval
(k = 1) , the capacitor current is
d vCss 1 Rsw
iCss = C ss = vCss + iLpt . (3.2)
dt Rsw + Rss Rsw + Rss
When the switch opens, second topology (k = 2) , the energy in the magnetizing
branch is transmitted into the secondary side to feed the load. In this subinterval, the snubber
capacitor becomes active and is charged with energy from the current in leakage inductor.
Finally when the current of magnetizing branch goes to zero, the diode stops conducting, and
converter enters the third topology (k = 3) . According to Figure 3.2, during these
39
iLpt(t) Rpt Lpt
Output
Lm
vg(t) iCss(t)
Input
Switch ON
(K = 1)
Css
RCss Rsw
Output
Lm
vg(t) Input
iCss(t)
Switch OFF Css
(K = 2, 3)
RCss Rsw
Rsw + (1 d1 ) Rss
vCss = iLpt (3.5)
d1
Equation (3.5) shows that if iLpt is calculated correctly by the SSA-AVM, then
vCss would be calculated correctly too by the AVM and does not need any additional
correction. Therefore, the correction term for vCss should be close to one. This will be proven
with numerical graphs in the following sections.
40
3.2 Effect of Output Snubber in SSA-AVM
The output snubber protects the diode from voltage and current spikes. To investigate
the effect of output snubber in SSA-AVM, we examine and compare the output stage of the
Flyback converter with and without the snubber branch.
iLst Rst vd
vC
Output
Rload
Input
Rc
Figure 3.3 Averaged detailed circuit of the output stage Flyback converter without the snubber in steady
state.
According to Figure 3.3, the average value of capacitor voltage in steady state could
be calculated as
vC = Rload iLst . (3.6)
To calculate the predicted average value of the output capacitor voltage by conventional
SSA-AVM, all topologies occurring during a switching cycle have to be considered.
41
Lst Rst iLst(t)
iC(t)
C
Output
Diode OFF
Rload
Input
(K=1, 3)
Rc
Output
Diode ON
Input
Rload
(K = 2)
Rc
Figure 3.4 Output stage topologies of Flyback converter without output snubber.
In the first and third topology (k = 1, 3) , the current of secondary side leakage inductor
The derivative term in (3.8) goes to zero in steady state. Solving the resulting equation for vC
in steady state gives,
vC = d 2 (Rload + RC )i Lst . (3.10)
42
Equations (3.6) and (3.10) can be used to calculate the correction term for vC
analytically. In particular, (3.10) shows that for fourth-order Flyback converter, if iLst is
calculated correctly, then vC is accurate too, and its correspondent correction term will be
close to one. This is observation will be proven with numerical graphs in the later sections
(Section 3.4).
iLst Rst vd
vCds vC
Output
Rload
Input
RCds Rc
Figure 3.5 Averaged detailed circuit of output stage of full-order Flyback converter in steady state.
According to Figure 3.6, when the diode is ON (k = 2) , the output snubber capacitor
charges with current coming from transformer. In this subinterval, the snubber branch is
parallel with the main output capacitor and it absorbs part of the transmitted energy to the
secondary side. The output capacitor current in this subinterval is
d vC 1
iC = C = vC
dt RC + ( RCds || Rload )
(3.13)
Rload ( RCds || Rload )
+ vCds + i Lst .
[Rload + RC ][RCds + (RC || Rload )] ( RCds || Rload ) + RC
Correspondingly, the snubber capacitor current is
43
d vCds Rload
iCds = C ds = v
dt [Rload + RCds ][RC + (RCds || Rload )] C
(3.14)
1 ( RC || Rload )
vCds + i Lst .
RCds + ( RC || Rload ) ( RC || Rload ) + RCds
The diode turns OFF in the first and third subintervals (k = 1, 3) . In the third
subinterval, the current stored in secondary side leakage inductor results into a voltage spike
that is partly absorbed by the output snubber capacitor. In this subinterval, the output
capacitor current is
d vC 1
iC = C = vC , (3.15)
dt Rc || Rload
and the snubber capacitor current is
d vCds
iCds = C ds = i Lst . (3.16)
dt
Rload
Input
Diode ON
(K = 2) RCds Rc
Diode OFF
Rload
Input
(K=1, 3) RCds Rc
Figure 3.6 Output side of full-order Flyback converter depicting the snubber operation.
Based on SSA-AVM, we take the averages of the output capacitor current and the
snubber capacitor voltage over the entire switching cycle, respectively, which yields
44
d vC (1 d 2 ) d2
iC = C = + v c
dt RC || Rload RC + ( RCds || Rload ) (3.17)
d 2 Rload d ( R || R )
+ vCds + 2 Cds load i Lst ,
[Rload + RC ][RCds + (RC || Rload )] ( RCds || Rload ) + RC
and
d vCds d 2 Rload
iCds = C ds = v
dt [Rload + RCds ][RC + (RCds || Rload )] C
(3.18)
d2 d 2 ( RC || Rload )
vCds + + (1 d 2 ) i Lst .
RCds + ( RC || Rload ) ( RC || Rload ) + RCds
Note that in steady state, (3.17) and (3.18) are equal to zero. Solving (3.15) for vC and (3.16)
for vCds gives the respective average values as predicted by SSA-AVM in steady state
d 2 Rload d ( R || R )
vC = vCds + 2 Cds load i Lst
[Rload + RC ][RCds + (Rc || Rload )] ( RCds || Rload ) + RC
1
(3.19)
(1 d 2 ) d2
+ ,
RC || Rload RC + ( RCds || Rload )
and
(RCds + ( RC || Rload ) ) d 2 Rload
vCds = [R v C
d2 load + RCds ][RC + (RCds || Rload )]
(3.20)
(RCds + ( RC || Rload ) ) d 2 ( RC || Rload )
+ (1 d 2 ) i Lst .
d2 ( RC || Rload ) + RCds
As shown in (3.19) and (3.20), the vC and vCds depend on two state variables.
Specifically, the variable vC is a function of iLst and vCds ; and vCds is a function of iLst and
vC . Moreover, vC and vCds are dependent on each other. This means that for accurate AVM,
both of these state variables will have to be corrected by appropriate functions. Therefore, in
the case of full-order Flyback converter, the correction terms for vC and vCds are not expected
to be close to one. This will be proven numerically in Section 3.4.
45
3.3 Generalized Numerical SSA-AVM
In conventional SSA-AVM [6], state equation has the following form
d x k =3 k =3
= d k A k x + d k B k u, (3.21)
dt k =1 k =1
which is a weighted sum of state equations corresponding to the topological instances of the
detailed converter circuit. In [9], a correction matrix M is added to (3.21) to correct for the
errors in DCM. Therefore, a new formula was proposed which is given in (2.1).
However, for a full-order Flyback converter, even the continuous state variables such
as output capacitor voltage, have to be corrected due to the presence of snubbers. In the
extended methodology, we consider a possibility for correction of all state variables in the
circuit. A vector of state variables of the circuit has the following general form
x = [ x1 x2 ... x n ]T . (3.22)
In the case of full-order Flyback converter, this vector is
[
x = vC vCss vCds i Lpt i Lst ] T
. (3.23)
Next, we define the state vectors of the conventional SSA-AVM as x SSA , and that of the
The state model (3.21) is corrected by inserting x CSSA from (3.25), which yields
d x C SSA k =3 k =3
M = d k A k M x C SSA + d k B k u, (3.26)
dt k =1 k =1
which after multiplying both sides by M 1 gives
d x C SSA k =3 k =3
= M 1 d k A k M x C SSA + M 1 d k B k u. (3.27)
dt k =1 k =1
46
Therefore, (3.27) would provide the correct average values of all state variables if matrix M
is computed appropriately. It is also noted that the AVMs based on (2.1) and (3.27) are
different as these models will have different equivalent state space matrices.
The duty ratio constraint d 2 is calculated on the basis of the period that the diode is
conducting. The averaged state variables and length of switching subintervals are calculated
and saved for construction of the appropriate correction terms.
The correction terms are then calculated using the following procedure. When the
model is in steady state, from (3.27) we have
k =3 k =3
0 = M 1 d k A k M x C SSA + M 1 d k B k u, (3.29)
k =1 k =1
Multiplying both sides of (3.29) by M gives
k =3 k =3
0 = d k A k M x C SSA + d k B k u. (3.30)
k =1 k =1
47
Matrix M and its calculation are also described in [22]. For consistency with the prior work
and for computational convenience, a diagonal correction matrix, namely M is assumed
here. This is done by solving (3.30) for M x and defining vector p as follows
1
k =3 k =3
p = M x = d k A k d k B k u . (3.31)
k =1 k =1
Thereafter, the entries of the diagonal of correction matrix M are calculated as following:
p( j )
M ( j, j ) = . (3.32)
x( j )
It can be seen from (3.31) that computing vector p requires matrix inversion. To
verify that the procedure is not ill-conditioned, it is possible to calculate the condition
k =3
number of the matrix d A
k =1
k k as has been done in [46]. The result is plotted in Figure 3.7.
As illustrated in Figure 3.7, the condition number of this matrix increases for some operating
conditions but remains sufficiently low to cause a numerical problem for the conventional
double-precision arithmetic in Matlab.
To show that the resulting errors will be sufficiently small, a residual equation is
defined as
k =3 k =3
r = d k A k p + d k B k u, (3.33)
k =1 k =1
wherein p is an approximate solution obtained by numerical methods. Due to the round-off
k =3 k =3
d k A k p = d k B k u, (3.34)
k =1 k =1
and
1
k =3
p p = d k A k r . (3.35)
k =1
Choosing a vector norm, an upper bound can be obtained for the absolute error in p .
1 1
k =3 k =3
p p = d k A k r d k A k r . (3.36)
k =1 k =1
48
It is understood from (3.34) that
k =3 k =3
d k A k p d k B k u , (3.37)
k =1 k =1
In other words,
k =3
dk Ak
1 k =1
. (3.38)
=
p k 3
d k B k u
k =1
Combining (3.36) and (3.38), gives a bound for the relative error in p as
k =3
1 dk Ak
p p k =3 k =1
dk Ak r . (3.39)
k =3
p k =1
d k B k u
k =1
As it is seen in (3.39), part of the inequality on right hand side is
1
k =3 k =3
= dk Ak dk Ak . (3.40)
k =1 k =1
10
0
3000
2000 1
0.8
Rload, Ohm 1000 0.6
0.4 d1
0 0.2
Figure 3.7 Inverted matrix condition number.
49
The correction functions for the fourth-order Flyback converter (without the output
snubber) have also been calculated using the same general procedure. The functions
m1 , m2 , m3 and m4 corresponding to the voltage of output capacitor, the voltage of input
(switch) snubber capacitor, the current of primary side leakage inductor, and current of
secondary side leakage inductor, respectively, are plotted in Figure 3.8 through Figure 3.11.
1.5
m1
0.5 1
3000
2000 0.5 d1
Rload, Ohm 1000 0 0
Figure 3.8 Correction coefficient m1 for voltage of output capacitor vC .
1.5
m2
1 1
0.5
3000 0.5 d1
2500
2000
1500
Rload, Ohm 1000 0
500
Figure 3.9 Correction coefficient m2 for voltage of primary snubber capacitor vCss .
50
10
3
m
0
0
1000
Rload, Ohm 2000
0.4 0.2
3000 0.8 0.6
1 d1
Figure 3.10 Correction coefficient m3 for primary side leakage inductor current iLpt .
10
4
5
m
0
3000
2000 1
Rload, Ohm 1000 0.8
0.6 d
0 0.4 1
0.2
Figure 3.11 coefficient m4 for secondary side leakage inductor current iLst .
The calculated duty ratio constraint for full-order Flyback converter is plotted in
Figure 3.12 with respect to d1 and Rload . The diagonal entries of M are calculated
according (3.32) using the element-wise operations. The resulting correction functions
m1 , m2 , m3 , m4 and m5 corresponding to the voltage of output capacitor, the voltage of first
(transistor) capacitor snubber, the voltage of second (diode) capacitor snubber, the current of
primary side leakage inductor, and the current of secondary side leakage inductor,
respectively. These numerically calculated parametric functions are plotted in Figure 3.12
through Figure 3.13.
51
0.8
0.6
2
0.4
d
0.2
0
0 0.5 d
500 1000 1500 2000 2500 1 1
3000
Rload, Ohm
Figure 3.12 Calculated function of duty-ratio constraint d2.
0
1
0.02
m
0.04
3000
2000 1
0.8
R, Ohm 1000 0.6
0.4 d1
0 0.2
1.5
2
1
m
0.5
3000
2000 1
0.8
Rload, Ohm 1000 0.6
0.4 d1
0 0.2
Figure 3.14 Correction coefficient m2 for voltage of primary snubber capacitor vCss .
52
0
m3
0.02
0.04
3000
2000 1
0.8
Rload, Ohm 1000 0.6
0.4 d1
0 0.2
Figure 3.15 Correction coefficient m3 for voltage of output snubber capacitor vCds .
10
m4
0
3000
2000 1
0.8
Rload, Ohm 1000 0.6
0.4 d
0 0.2 1
Figure 3.16 Correction coefficient m4 for primary side leakage inductor current iLpt .
0
m5
0.02
0.04
3000
2000 1
0.8
Rload, Ohm 1000 0.6
0.4 d1
0 0.2
Figure 3.17 Correction coefficient m5 for secondary side leakage inductor current iLst .
53
The numerically calculated parametric functions, including correction terms and duty
ratio constraint, are stored in lookup tables and is used in the AVM in the form of non-linear
functions, wherein interpolation/extrapolation is automatically implemented. The output of
these functions depend on the input voltage v g , the duty cycle d1 , and the average value of
The impedance z diode defined in (3.41) combines three state variables (two of which are fast
state variables). This impedance is used as input argument for the m1 lookup table that
performs the output capacitor voltage correction. All other correction functions are
implemented using the input argument impedance z mid defined in (3.41)-(3.42). Another
advantage of using the dynamic impedances - (3.42) is that they define the converter
operating point independently of the input voltage. For example, when the input voltage is
increased while d1 and Rload are kept constant, all the state variables will increase
proportionally and the value of dynamic impedance that specifies the operating point of the
converter will not change.
Once the state-space matrices and parametric functions are available, the final
extended PAVM is implemented according to the block diagram shown in Figure 3.18. In
each time step, the state-variable-dependent matrix M (and its inverse) containing
correction coefficients is calculated and appropriately multiplied with the previously
extracted system matrices A k , B k , C k , and D k . The resulting state space model is nonlinear
but continuous (has no switching), and can be readily implemented in any state-variable-
based simulation environment.
54
x
-1 -1
u x = M AT M x + M BT u
zdiode (d1 , x ) M y = CT x + DT u y
z M (d1 , zdiode , zmid )
zmid (d1 , x )
d2 (d1 , zdiode , zmid )
d1 d2 , d 3 3 3
d3=1- d1 - d2
AT = (
k=1
d k Ak ), BT = (
k=1
)
d k Bk
3 3
Ak , Bk , Ck , Dk
CT = (
k=1
d k Ck ), DT = (
k=1
dk D k )
Figure 3.18 Block diagram depicting implementation of proposed extended PAVM.
M1AT M is a similarity transformation [46] that does not change the eigenvalues of the
k =3
matrix A T = d k A k , which is the same matrix used in conventional (uncorrected SSA)
k =1
in (2.1). To verify the effect of correction in (3.27), we assume a CCM operating point
defined by d1 = 0.381 and Rload = 717 . The resulting correction coefficients (the diagonal
entries of M) and the eigenvalues of various matrices are summarized in Table 3.1. As is
shown in Table 3.1, the matrix product AT M contains some eigenvalues with positive real
part (which is a result of negative entries present in M). Therefore, if the PAVM is formed
using conventional approach, the resulting model will also contain positive eigenvalues and
be unstable. At the same time, if the PAVM is formed using the proposed generalized
approach (3.27), the matrix product M1AT M as well as the final resulting PAVM will have
eigenvalues with negative real part. It is also important to point out that the eigenvalues of
M1AT M are different from those computed by numerical linearization of the corresponding
PAVM. This is due to the fact that the matrix M is state-variable-dependent, which affects
the models eigenvalues and makes that different from those of M1AT M and
k =3
AT = d k A k (which would obviously be incorrect due to lack of correction in DCM).
k =1
55
Table 3.1 Eigenvalues of AVMs.
56
Table 3.2 Hardware prototype and detailed model comparison in terms of input current, output voltage
and efficiency.
Input current Output voltage Efficiency
Hardware 0.432 -72.67 85.33%
Detailed Model 0.42 -72.63 87.52%
-72.4
Output voltage Hardware measurement
-72.5 Detailed model
-72.6
vout , V
-72.7
-72.8
-72.9
300 Transformer secondary side voltage
200
v Trans.s , V
100
-100
40
Switch snubber voltage
30
20
vSnub.p , V
10
-10
0 0.5 1 1.5
Time, s 5
x 10
Figure 3.19 Measured and detailed model waveforms for the considered operating point.
57
3.7 Precision Evaluation in Steady State
To evaluate precision of the new generalized PAVM, this model is compared against
the detailed model in DCM (d1 = 0.381, Rload = 2500) and in CCM
(d1 = 0.381, Rload = 717 ) operating points. The average-values of all state variables,
including the voltage of snubber capacitors, obtained from the corrected PAVM and the
detailed model is summarized in Table 3.3. As Table 3.3 demonstrates, the new PAVM is
capable of predicting average-value of circuit variables very accurately, with a very small
relative error.
Next, the precision of the new generalized PAVM in terms of predicting the converter
efficiency is considered. For consistency, the same DCM and CCM operating points are
considered here. The calculated results are summarized in Table 3.4. As can be observed in
Table 3.4, the new model works very well in both modes DCM and CCM. Comparing the
results in Table 3.4 corresponding to the 4th and full-order converter circuit, one can also
observe that the converter efficiency reduces when the output snubber is used (due to
additional energy dissipation).
58
Table 3.3 Accuracy precision of the proposed PAVM in predicting steady state variables.
4th-order Flyback Full-order
Variable converter Flyback converter
DCM CCM DCM CCM
Correct Value -102 -70.87 -85.78 -72.63
Output capacitor voltage vC (V) Predicted Value -102.1 -70.87 -85.78 -72.66
Table 3.4 Accuracy precision of the proposed PAVM in predicting converter efficiency.
4th-order Flyback Full-order Flyback
Model converter converter
DCM CCM DCM CCM
Detailed model 94.31% 94.41% 87.88% 87.52%
Generalized/Corrected PAVM 94.31% 94.41% 87.89% 88.05%
59
good agreement with the corresponding experimental measurements from the laboratory
converter prototype. In the following sections, the proposed PAVM is further evaluated in
time-domain transients as well as in frequency-domain.
duty cycle d1 = 0.381 . While the converter duty cycle is kept constant, another 2500
resistor is added in parallel to the load. The converter transient response as predicted by the
PAVMs is shown in Figure 3.20.
As it is seen in Figure 3.20, the predicted steady state values (before and after the
transient) are same for both models, but the transient is predicted somewhat differently with
the uncorrected PAVM showing a delayed response. This is because formulas that the AVMs
are based on ((2.1) for uncorrected PAVM and (3.27) for corrected PAVM) are equivalent
with (3.30) in steady state, and therefore both AVMs predict same averaged-values.
Next, the same study is performed using the full-order Flyback converter model. As
discussed in Section 3.5, the uncorrected PAVM is not able to operate due to positive
eigenvalues, and is therefore not used in this study. The simulations result obtained by the
generalized/corrected PAVM are shown in Figure 3.21. As it is observed in Figure 3.21, the
new PAVM very accurately predicts the steady state values as well as the entire transient
response of the system.
60
-70
-80
Detailed
Corrected PAVM
v ,V
-90 PAVM
C
-100
-110
40
30
,V
20
Css
v
10
0
1.5
1
iLpt , A
0.5
-0.5
0.05
0
-0.05
,A
-0.1
Lst
i
-0.15
-0.2
-0.25
0.3 0.31 0.32 0.33 0.34 0.35
Time, s
Figure 3.20 Transients of state variables of fourth-order Flyback converter as predicted by uncorrected
PAVM and the proposed corrected PAVM.
61
-70
Detailed
Corrected PAVM
-75
vC , V
-80
-85
-90
40
30
vCss , V
20
10
0
150
100
,V
50
Cds
0
v
-50
-100
1.5
1
,A
0.5
Lpt
i
-0.5
0.05
-0.05
iLst , A
-0.1
-0.15
-0.2
0.29 0.295 0.3 0.305 0.31 0.315 0.32 0.325 0.33
Time, s
Figure 3.21 Transients of state variables of full-order Flyback converter due to load change.
62
Another time-domain study is considered using the full-order Flyback converter
model. In the study presented here, the converter undergoes an increase in the duty cycle
which causes a change in the operating mode. The converter is assumed to initially operate in
steady state under load Rload = 717 and duty cycle d1 = 0.5 . Then, at t = 0.3 s, the control
duty cycle is decreased from 0.5 to 0.2 over a period of 0.05s. The resulting transient
responses are shown in Figure 3.22. As it can be observed in Figure 3.22, the converter
initially operates in CCM, but after the transients and duty cycle change, the operating mode
changes to DCM. With the decrease in duty cycle, the transistor on-state becomes shorter in
each switching cycle, and the output voltage amplitude decreases as well. Therefore, the
primary side leakage inductor current (input current) has smaller oscillation amplitude, which
in turn results in decrease of the average value of the input current. In consequence, less
voltage is delivered to the load. This is verified by the output capacitor voltage in Figure
3.22.
63
-20
-40
Detailed
-60
vC , V
Corrected PAVM
-80
-100
-120
60
40
vCss , V
20
-20
200
100
vCds , V
-100
-200
6
4
iLpt , A
-2
1
0.5
,A
0
Lst
i
-0.5
-1
0.28 0.3 0.32 0.34 0.36 0.38 0.4
Time, s
Figure 3.22 Circuit state variables transients due to intense increase of duty cycle.
64
3.8.2 Performance of Proposed PAVM in Frequency-Domain
In the following study, the converter is assumed to operate in DCM operating point
defined by d1 = 0.381 and Rload = 2500 . The control-to-output transfer function is
considered, and the frequency response is calculated by injecting a small amplitude
sinusoidal signals to the control duty cycle of detailed model and corrected PAVM. The
result of this control-to-output transfer function magnitude and phase are shown in Figure
3.23. As it can be seen in Figure 3.23, the proposed PAVM predicts the small-signal transfer
function with a good agreement with the detailed model. At higher frequencies, the results
become less accurate due to interaction of the injected sinusoidal wave with switching [9],
[49].
50 Detailed
Magnitude , dB
Corrected PAVM
-50
150
100
Phase , deg
50
-50 1 2 3 4 5
10 10 10 10 10
Frequency, Hz
Figure 3.23 Control-to-output transfer function magnitude and phase in DCM as predicted by proposed
PAVM and detailed model.
65
Chapter 4 : Summary of Research and Future Work
66
and CCM with transformer isolation, basic parasitics, and snubbers, which to the best of our
knowledge has not been achieved in the prior literature.
67
Bibliography
[1] P. Chrin and C. Bunlaksananusorn, Large-Signal Average Modeling and Simulation
of DC-DC Converters with SIMULINK, Power Conversion Conference, Nagoya,
April 2007, pp. 2732.
[2] M. Plesnik, Use of the State-Space Averaging Technique in Fast Steady-State
Simulation Algorithms for Switching Power Converters, Canadian Conference on
Electrical and Computer Engineering(CCECE), May 2006, pp. 2224-2227.
[3] D. Maksimovic, A.M. Stankovic, V.J. Thottuvelil, and G.C. Verghese, Modeling and
simulation of power electronic converters, Proceedings of the IEEE, Jun 2001, vol.
89, no. 6, pp. 898912.
[4] G. Nirgude, R. Tirumala, and N. Mohan, A new, large-signal average model for
single-switch DC-DC converters operating in both CCM and DCM, Power
Electronics Specialists Conference (PESC), 2001, vol. 3, pp. 1736-1741.
[5] A. Davoudi, J. Jatskevich, and P.L. Chapman, Averaged modelling of switched-
inductor cells considering conduction losses in discontinuous mode, IET, May 2007,
vol. 1, no.3, pp. 402-406.
[6] R. D. Middlebrook and S. Cuk, A General Unified Approach to Modelling Switching
Converter Power Stage, IEEE PESC Record, 1976, pp. 18-34.
[7] R. Tymerski and V. Vorperian, Generation, Classification and Analysis of Switched-
Mode DC-to-DC Converters by the Use of Converter Cells, Telecommunications
Energy Conference (INTELEC), Oct. 1986, pp. 181-195.
[8] J. Chen and K. D. T. Ngo, Alternate forms of the PWM switch model in discontinuous
conduction mode, IEEE Trans. Aerosp. Electron. Syst., 2001, vol. 37, no. 2, pp. 754-
758.
[9] J. Sun, D. M. Mitchell, M. F. Greuel, P. T. Krein, and R.M. Bass, Averaged modeling
of PWM converters operating in discontinuous conduction mode, IEEE Trans. On
Power Electronics, Jul 2001, vol. 16, no. 4, pp. 482-492.
[10] W. M. Moussa and J. E. Morris, Comparison between state-space averaging and
PWM switch for switch mode power supply analysis, Proceedings of the IEEE
Southern Tier Technical Conference, Apr. 1990, pp. 15-21.
68
[11] K. Rustom, W. Qiu, C. lannello, and I. Batarseh, Unified Flyback switching-cell
model including the leakage inductance effects for SPICE simulation, Power
Electronics Specialist Conference (PESC), June 2003, vol. 3, pp. 1215-1219.
[12] C. T. Rim, G. B. Joung, and G. H. Cho, Practical switch based state-space modeling
of DC-DC converters with all parasitics, IEEE Transactions on Power Electronics,
Oct 1991, vol. 6, no. 4, pp. 611-617.
[13] A. Davoudi, and J. Jatskevich, State-space averaging of switched-inductor-cell for
PWM dc-dc converters considering conduction losses in both operational modes,
IEEE International Symposium on Circuits and Systems (ISCAS), May 2006, pp. 830.
[14] R. Ericson, D. Maksimovic, Fundamentals of Power Electronics, 2nd ed., Kluwer
Academic Publisher, Norwell, 2001, pp. 161-165.
[15] A. Reatti, L. Pellegrini, and M. K. Kazimierczuk, Impact of boost converter
parameters on open-loop dynamic performance for DCM, ISCAS, 2002, vol. 5, pp.
513-516.
[16] L.R. Pujara, M. K. Kazimierczuk, and A. N. I. Shaheen, Robust stability of PWM
buck DC-DC converter, Proceedings of the IEEE International Conference, Sep 1996,
pp. 632-637.
[17] G. Zhu, S. Luo, C. Iannello, and I. Batarseh, Modeling of conduction losses in PWM
converters operating in discontinuous conduction mode, Proceedings of ISCAS,
Geneva, 2000, vol. 3, pp. 511-514.
[18] A. Reatti and M. K. Kazimierczuk, Small-signal model of PWM converters for
discontinuous conduction mode and its application for boost converter, IEEE Trans.
On Circuits Syst. I, Jan. 2003, vol. 50, no. 1, pp. 6573.
[19] D. Czarkowski and M. K. Kazimierczuk, Energy-conservation approach to modeling
PWM dc-dc converters, IEEE Trans. On Aerospace Electron. Syst., 1993, vol. 29, no.
3, pp. 1059-1063.
[20] D. Maksimovic and S. Cuk, A unified analysis of PWM converters in discontinuous
modes, IEEE Transactions On Power Electronics, Jul. 1991, vol. 6, no. 3, pp. 476-
490.
69
[21] D. Maksimovic, Computer-aided small-signal analysis based on impulse response of
DC/DC switching power converters, IEEE Transactions On Power Electronics, Nov.
2000 , vol. 15, no. 6, pp. 1183-1191.
[22] A. Davoudi, J. Jatskevich, and T. De Rybel, Numerical state-space average-value
modeling of PWM DC-DC converters operating in DCM and CCM, IEEE
Transactions On Power Electronics, July 2006, vol. 21, no. 4, pp. 1003-1012.
[23] A. Davoudi, J. Jatskevich, and P. L. Chapman, Numerical Dynamic Characterization
of Peak Current-Mode-Controlled DCDC Converters, IEEE Transactions On
Circuits and Systems II, Dec. 2009, vol. 56, no. 12, pp. 906-910.
[24] A. Davoudi and J. Jatskevich, Realization of parasitics in state-space average-value
modeling of PWM DC-DC converters, IEEE Transactions On Power Electronics,
July 2006, vol. 21, no. 4, pp. 1142-1147.
[25] D. Skendzic, Two transistor Flyback converter design for EMI control, International
Symposium on Electromagnetic Compatibility, Aug. 1990, pp. 130-133.
[26] K. Ma and Y. Lee, An integrated Flyback converter for DC uninterruptible power
supply, IEEE Transactions On Power Electronics, Mar. 1996, vol. 11, no. 2, pp. 318-
327.
[27] S. Amini Akbarabadi, H. Atighechi, and J. Jatskevich, Circuit-Averaged and State-
Space-Averaged-Value Modeling of Second-Order Flyback Converter in CCM and
DCM Including Conduction Losses, International Conference On Energy and
Electrical Drives (POWERENG), May 2013, pp. 995-1000.
[28] M. T. Zhang, M. M. Jovanovic, and F. C. Y. Lee, Design considerations and
performance evaluations of synchronous rectification in Flyback converters, IEEE
Transactions On Power Electronics, May 1998, vol. 13, no. 3, pp. 538-546.
[29] N. Golbon, F. Ghodousipour, and G. Moschopoulos A DC-DC converter with stacked
flyback converters, Energy Conversion Congress and Exposition (ECCE), Sept. 2013,
pp. 4894-4899.
[30] S. Dutta, D. Maiti, A. K. Sil, and S. K. Biswas A Soft-Switched Flyback converter
with recovery of stored energy in leakage inductance, India International Conference
on Power Electronics (IICPE), Dec. 2012, pp. 1-5.
70
[31] G. Huang, T. Liang, and K. Chen, Losses analysis and low standby losses quasi-
resonant flyback converter design, IEEE International Symposium on Circuits and
Systems (ISCAS), May 2012, pp. 217-220.
[32] R. T. H. Li and H. S. Chung, A passive lossless snubber cell with minimum stress and
wide soft-switching range, Energy Conversion Congress and Exposition(ECCE), Sept.
2009, pp. 685-692.
[33] R. T. H. Li, H. S. Chung, A. K. T. Sung, Passive Lossless Snubber for Boost PFC
With Minimum Voltage and Current Stress, IEEE Transactions On Power
Electronics, March 2010, vol. 25, no. 3, pp. 602-613.
[34] J. A. Cobos, J. Sebastian, J. Uceda, E. de la Cruz, and J. M. Gras, Study of the
applicability of self-driven synchronous rectification to resonant topologies, IEEE
Power Electron. Specialists Conf. Rec., 1992, pp. 933-940.
[35] D. J. Harper, D. R. Hyde, G. M. Fry, and J. A. Houldsworth, Controlled synchronous
rectifier, High Freq. Power Conversion Conf. Proc., 1988, pp. 165-172.
[36] K. Harada, H. Sakamoto, Switched snubber for high frequency switching, IEEE
Power Electronics Specialists Conference(PESC), 1990, pp. 181-188.
[37] S. Amini Akbarabadi, M. Sucu, H. Atighechi, and J. Jatskevich, Numerical average
value modeling of second-order Flyback converter in both operational modes, IEEE
Workshop on Control and Modeling for Power Electronics (COMPEL), June 2013, pp.
1-6.
[38] N. Mohan, T. M. Undeland, and W. P. Robbins, Power Electronics, Third Edition,
John Wiley & Sons Publisher, 2003.
[39] K. Harada, T. Ninomiya, Optimum Design of RC Snubbers for Switching
Regulators, IEEE Transactions on Aerospace and Electronic Systems, March 1979,
vol. AES-15, no. 2, pp. 209-218.
[40] N. Mohan, T. Undeland, and W. Robbins, Power Electronics: Converters,
Applications, and Design, Third Edition, John Wiley Publisher, 2003.
[41] E. Acha, V. Agelidis, O. Anaya-Lara, and T. Miller, Power Electronic Control in
Electrical Systems, Newnes, 2002.
[42] Piece-wise Linear Electrical Circuit Simulation for Simulink (PLECS), Plexim GmbH,
2009. [Online]. Available: www.plexim.com
71
[43] Simulink: Dynamic System Simulation for Matlab, Using Simulink, Version R2009b,
The MathWorks Inc., 2009.
[44] A. Davoudi, J. Jatskevich, and P. L. Chapman, Parametric Average-Value Modeling
of Multiple-Input Buck Converters, Canadian Conference on Electrical and
Computer Engineering( CCECE), April 2007, pp. 990-993.
[45] K. J. Astrom K. J. and T. H. Hagglund, New tuning methods for PID controllers,
Proceedings of the 3rd European Control Conference, 1995.
[46] U. M. Ascher and C. Greif, A first course in numerical methods,
SIAM, Philadelphia, 2011.
[47] Automated State Model Generator (ASMG), PC Krause and Associates Inc., 2002.
[Online]. Available: www.pcka.com.
[48] W. Gautchi, Numerical Analysis: An Introduction, Birkhauser Publisher, Boston, 1997.
[49] J. Sun, D. M. Mitchell, and D. E. Jenkins Delay effects in averaged modeling of PWM
converters, Power Electronics Specialists Conference(PESC), 1999, vol. 2, pp. 1210-
1215.
[50] M. Sucu, Parametric Average Value Modeling of Flyback Converters in CCM and
DCM Including Parasitics and Snubbers, M.Sc. Thesis, University of British
Columbia, Vancouver, BC, Canada, 2011.
72
Appendices
Req
0 1
L 0
A1 = m , B1 = Lm
0 1
0 0
C ( Rload + Rc )
Rload Rc Rload
n2 L (R + R ) 1
nLm ( Rload + Rc ) 0
A2 = m c
, B2 = nLm
R load 1
n C ( Rload + Rc ) C ( Rload + Rc ) 0 0
0 0 0 0
A 3 = 0 1 , B3 =
C ( Rload + Rc ) 0 0
73
Appendix B. Converters Circuit Parameters
v g = 20V
Mosfet : 3.8 A, 55V , N chanel , International Re ctifier IRRL 2705
Rsw = 0.04
f s = 250kHz
d1 = 0.381
Transformer : ICE Components ICA 0635
R pt = Rst = 210m
Lm = 27 H
n = 1/ 6
Diode :1A, 400V , ultra fast re cov ery diode, Central Semiconductor CorpCMR1U 04
v fd = 1.25V
C = 22 F ,100V , Alu min um electrolytic capacitor , Sanyo100 MV 22 AX
Rc = 90m
B.2 Full-order Flyback Converter Parameters with all Parasitics and Snubbers
vg = 20V
Mosfet : 3.8 A, 55V , N chanel , International Re ctifier IRRL 2705
Rsw = 0.04
f s = 250kHz
d1 = 0.381
Css = 470 pF , Rss = 10
Transformer : ICE Components ICA 0635
R pt = Rst = 210m
Lm = 27 H , L p = 0.2 H , Ls = 0.8H
n = 1/ 6
Diode :1A, 400V , ultra fast re cov ery diode, Central Semiconductor CorpCMR1U 04
v fd = 1.25V
Cds = 100 pF , Rds = 200
C = 22 F ,100V , Alu min um electrolytic capacitor , Sanyo100 MV 22 AX
Rc = 90m
74
Appendix C. Hardware Flyback Converter Prototype Circuit Diagram
75