Professional Documents
Culture Documents
Ubc 2011 Fall Sucu Mehmet
Ubc 2011 Fall Sucu Mehmet
by
Mehmet Sucu
in
(Vancouver)
October 2011
Modeling of switched-mode DC-DC converters has been receiving significant interest due to
their widespread applications. Averaged modeling is the most common approach (and tool)
that has been used to analyze dynamic performance of converter circuits. Specifically, state-
space averaged models are widely used because of their simplicity and generality. However,
as has been shown in the literature, the challenges of directly applying this approach to
predict the discontinuous variables (states) and include the parasitics and losses have limited
application of this approach to a wider range of converter circuits. The recently introduced
parametric average value models (PAVM) has a potential to overcome this problem.
In this Thesis, first of all a second-order flyback converter has been investigated. An
analytical solution of state-apace averaging and small-signal analysis of the flyback converter
in continuous conduction mode (CCM) and discontinuous conduction mode (DCM) is given
without and with parasitics. The PAVM methodology has been applied to the second-order
model to overcome the problem of discontinuous state during the DCM.
The snubber circuits in flyback converter have also been investigated. Appearance of
snubbers in the model introduces a problem on the output voltage besides improving the
efficiency prediction. It is shown that with the snubbers the conventional state-space
averaging cannot predict the output voltage correctly in CCM and DCM. To solve this
problem the model is partitioned into two different sub-circuits: i) switching sub-circuit
circuit; and ii) non-switching sub-circuit. Thereafter it becomes possible apply the averaging
on the switching sub-circuit only.
Finally, a full-order flyback converter with two RC snubber circuits and all the basic
parasitics is considered. The PAVM methodology has been extended to this class of
switching converter for the first time. It is shown that including the snubbers and parasitics
significantly improves the model accuracy in terms of predicting converter efficiency, which
represents an appreciable improvement over all previously existing average models. The
proposed model has been verified with detailed simulations and hardware measurements.
ii
Table of Contents
Abstract.................................................................................................................................... ii
List of Figures......................................................................................................................... vi
Acknowledgements ................................................................................................................. x
iii
4.3 Parametric Average Value Modeling in CCM and DCM ................................................... 62
4.3.1 Model Implementation.................................................................................................... 62
4.4 Case Studies ........................................................................................................................ 67
4.4.1 Time Domain .................................................................................................................. 67
4.4.2 Frequency Domain.......................................................................................................... 69
4.4.3 Efficiency Results ........................................................................................................... 70
Bibliography .......................................................................................................................... 74
Appendices ............................................................................................................................. 78
Appendix A. The Converters Circuit Parameters .......................................................................... 78
A.1 Second-order Flyback Converter Parameters without Parasitics in CCM ...................... 78
A.2 Second-order Flyback Converter Parameters without Parasitics in DCM ...................... 78
A.3 Second-order Flyback Converter Parameters with Parasitics in CCM ........................... 78
A.4 Second-order Flyback Converter Parameters with Parasitics in DCM ........................... 79
A.5 Fifth-order Flyback Converter Parameters in CCM ....................................................... 79
A.6 Full-order Flyback Converter Parameters in CCM......................................................... 79
A.7 Full-order Flyback Converter Parameters in DCM ........................................................ 80
Appendix B. Flyback Converter Circuit Diagram......................................................................... 81
iv
List of Tables
v
List of Figures
Figure 2.1 (a) Assumed circuit for the second order Flyback converter without
parasitics; (b) Circuit during subinterval 1; (c) Circuit during
subinterval 2. .......................................................................................................... 6
Figure 2.2 The inductor current ................................................................................................ 6
Figure 2.3 Inductor current and capacitor voltage of second order Flyback
converter without parasitics in CCM. .................................................................. 14
Figure 2.4 Capacitor voltage of second order Flyback converter without parasitics
in CCM................................................................................................................. 14
Figure 2.5 Second-order Flyback converter without parasitics during third
subinterval in DCM. ............................................................................................. 15
Figure 2.6 Magnetizing current in DCM for the load R = 2500Ω . ......................................... 16
Figure 2.7 Inductor current and capacitor voltage of second order Flyback
converter without parasitics in DCM. .................................................................. 18
Figure 2.8(a) Second-order Flyback converter with parasitics; (b) Circuit during
subinterval 1 (c) Circuit during subinterval 2. ..................................................... 19
Figure 2.9 Inductor current, capacitor voltage and output voltage of second order
Flyback converter with parasitics in CCM. ......................................................... 31
Figure 2.10 Output voltage of second order Flyback converter with parasitics in
CCM. .................................................................................................................... 31
Figure 2.11 Second-order Flyback converter with parasitics during third
subinterval in DCM. ............................................................................................. 32
Figure 2.12 Inductor current, capacitor voltage and output voltage of second
order Flyback converter with parasitics in DCM. ................................................ 35
Figure 2.13 Variable d3 as a function` of duty-cycle ( d1 ) and the load ( R ) . ....................... 40
Figure 2.14 The correction term m1 as a function of duty-cycle ( d1 ) and the load
resistance ( R ) . ..................................................................................................... 40
Figure 2.15 The correction term m2 as a function of duty-cycle ( d1 ) and the load
resistance ( R ) . ..................................................................................................... 41
vi
Figure 2.16 Implementation of the parametric average-value model. .................................... 42
Figure 2.17 Simulated inductor current, capacitor voltage and output voltage of
the second order Flyback converter with parasitics in DCM. .............................. 44
Figure 2.18 Transients in inductor current, capacitor voltage and output voltage
of the second order Flyback converter due to the step change in load. ............... 45
Figure 2.19 Control-to-output transfer function of the second-order Flyback
converter evaluated at R = 717.05Ω and d1 = 0.381 . ............................................. 46
vii
Figure 4.6 Variable d 3 as a function of the duty-cycle ( d1 ) and the load
resistance ( R ) . ..................................................................................................... 64
Figure 4.7 The correction term m2 as a function of duty-cycle ( d1 ) and the load
resistance ( R ) . ..................................................................................................... 65
Figure 4.8 The correction term m3 as a function of the duty-cycle ( d1 ) and the
Figure 4.9 The correction term m4 as a function of duty-cycle ( d1 ) and the load
resistance ( R ) . ..................................................................................................... 66
Figure 4.10 The correction term m5 as a function of duty-cycle ( d1 ) and the load
resistance ( R ) . ..................................................................................................... 67
Figure 4.11 Measured and simulated output voltage, primary and secondary
current in CCM at constant duty-cycle. ............................................................... 68
Figure 4.12 Simulated output voltage, primary and secondary current during the
transient from DCM to CCM due to the step change in load. ............................. 69
Figure 4.13 Control-to-output transfer function of the full-order Flyback
converter evaluated at R = 717.05Ω and d1 = 0.381 ............................................... 70
viii
List of Abbreviations
ix
Acknowledgements
I would like to express my appreciation to my supervisor, Dr. Juri Jatskevich, whose strong
academic support have been the most precious assets to my studies and research. I am also
very grateful for the partial financial support that has been made available to me through the
NSERC under the Discovery Grant.
I also like to thank Dr. William Dunford and Dr. Shahriar Mirabbasi who have accepted to be
the committee members and dedicated their time and effort for reading this thesis and
providing their constructive and valuable comments.
My special thanks go to previous and current members of the Power and Energy Systems
Group at UBC who have always supported me and gave their valuable insights into my
research.
x
Chapter 1 : Introduction
A detailed model of a flyback converter can be easily implemented using widely available
simulation packages (e.g. Matlab/Simulink, ASMG, PLECS, etc.) [14-16]. Detailed models
are often used during the design process as such models have all the required information to
1
calculate the exact switching transients and component stresses and characteristics. But the
large computation time required for such detailed switching models makes them less
applicable for system-level studies. Instead, the average-value modeling has been used very
effectively for the system-level analysis and studies, wherein the effects of fast switching are
neglected or averaged with respect to the switching interval. A classical state-space averaged
model of a flyback converter [10] considers only the switch losses without any snubber
circuits and has the simplest first-order transformer approximation. A simplified linear circuit
model for obtaining dc and small-signal circuit model is given in [17], which has the basic
parasitics but does not include any snubber circuits and transformer primary and secondary
copper losses. A dc and small-signal circuit model models for a flyback converter operating
in CCM can be found in [12], which has the basic parasitics but again does not include any
snubber circuits and has only a the simplest transformer model without the primary and
secondary losses.
A typical switched-inductor dc-dc converter can operate in two modes. One is the
Continuous Conduction Mode (CCM) in which inductor current never falls to zero, and the
second mode is Discontinuous Conduction Mode (DCM) allowing inductor current to
become zero for a portion of switching period. The DCM typically occurs at light loads and
differs from CCM since this mode results into three different switched networks over one
switching cycle (as opposed to two switched networks in the case of CCM operation).
Models for PWM converters operating in CCM based on well-known state-space averaging
2
technique were first introduced in 1970’s [18]. Since then, several circuit-oriented averaging
approaches have also been proposed [3, 19]. Numerous method have been developed for the
average value modeling of PWM dc-dc converters in DCM such as reduced-order state-space
averaging [20], reduced-order averaged-switch modeling [7], equivalent duty ratio models
[3], loss-free resistor model [10], full-order averaged-switch modeling [21], and full-order
state-space averaging [4].
State-space averaging is based on the classical averaging theory and involves manipulation of
state-space equations of a converter system. First, a state-space representation of converter is
obtained for each topology and subinterval. Then, the obtained piece-wise linear equations
are weighted by the corresponding time subinterval length and added together. State-space
averaging has been demonstrated to be an effective method to analyze PWM converters.
Analytical averaging, however, is based on so-called small-ripple approximation. Most of the
previous works on averaging methods were derived for a specific ideal topology. In addition,
derivation of state-space average-value model, the equivalent series resistance (ESR) of
circuit components are often neglected and the state variables are considered as linear
segments. Such assumptions result in inaccuracy of the corresponding time constants as well
as the waveforms. If the losses due to the switch and/or active elements are taken into
account, whereby the linear shape of the current waveform would change into exponential
form, the analytically derived models would become significantly more complicated and
challenging. The analytical derivation also becomes more complicated when the number of
energy storage elements (inductors and capacitors) is high.
3
1.4 Parametric Average-Value Modeling
Parametric average-value modeling methodology has been set forth by the UBC researchers.
This methodology has been successfully demonstrated for synchronous machine-converter
systems in [22, 23]. The major point of this approach is to use the detailed simulation for
numerically calculating the key relationships needed for constructing the average-value
model of a certain well-defined form. In doing so, the effect of parasitics included in the
detailed model becomes automatically included in the numerically constructed parametric
functions, which are then used for the state-variable-based average-value models. This
approach also reduces the effort of the model developer and avoids many complicated
analytical derivations. This method has been extended to the PWM dc-dc converters in [24-
27] based on corrected full-order averaged models proposed for circuit averaging [28] and
state-space averaging [4] that very accurately capture the high-frequency dynamics of fast
state variables.
This Thesis makes an original contribution and extends the parametric average-value
modeling the flyback converters. The considered converter model includes all the basic
parasitics and high order transformer model with primary and secondary resistances and
leakage inductances. The propose model also includes two RC (resistance and capacitance)
snubber circuits to protect the switch and the diode during the on-off operation. To the best
of our knowledge, this has not been done in any published research on this subject.
4
Chapter 2 : Second Order Flyback Converters
In this Chapter, we consider and approximate (simplified) circuit of the flyback converter,
wherein only two energy storage elements are considered, hence second order converter.
Such approximate converter circuit has been used in the literature for carrying out basic
analysis and averaging methods. A number of modeling techniques have appeared in the
literature, including the current injected approach [19], circuit averaging [7, 21, 29], and
state-space averaging [18] method.
Obtaining a small-signal ac model of a basic switched converter circuit, such as buck, boost
without parasitics, can be readily achieved using analytical derivations. But when the
converter circuit has parasitics, it becomes almost impossible and impractical to derive the
higher order state equations. In this case, the state-space equations can be obtained from the
detailed model by using commercially available simulation packages [16, 30], and then used
the state-space description (matrices) to establish the small-signal model [10] (see
Section7.3.2).
In this Section, a small-signal ac model will be derived for a second-order flyback converters
without parasitics. Based on that, the state-space equations will be derived.
A second order flyback converter without parasitics is shown in Figure 2.1(a). Here, n is the
turn ratio of the transformer ( N1 N 2 ) .
5
ig (t ) i (t ) n
D ic (t )
+
vg (t ) vL (t ) Lm
C R v (t )
Mosfet
(a )
ig (t ) i (t ) n
ic (t )
+
vg (t ) vL (t ) Lm
C R v (t )
(b )
ig (t ) i (t ) n
ic (t )
+
vg (t ) vL (t ) Lm
C R v (t )
(c )
Figure 2.1 (a) Assumed circuit for the second order Flyback converter without parasitics; (b) Circuit
during subinterval 1; (c) Circuit during subinterval 2.
The inductor current i ( t ) in CCM is shown in Figure 2.2. During the first interval, Figure
2.1(b), the inductor stores some energy and transfers this energy to the secondary side during
the second interval, Figure 2.1(c).
1.6
1.4
1.2
d1Ts d 2Ts
I (Amp.)
0.8
Ts
0.6
0.4
0.999991 0.999993 0.999995 0.999997
Time (s)
vL ( t ) = v g ( t ) (2.1)
v (t )
ic ( t ) = − (2.2)
R
ig ( t ) = i ( t ) (2.3)
Applying the small ripple approximation [10] and replacing the voltages and currents with
their respective average values, we obtain
vL ( t ) = v g ( t ) (2.4)
Ts
v (t )
ic ( t ) = −
Ts
(2.5)
R
ig ( t ) = i ( t ) Ts
(2.6)
During the second subinterval, MOSFET is off and the diode conducts, which results in
circuit of Figure 2.1(c). The inductor voltage vL ( t ) , capacitor current ic ( t ) , and converter
vL ( t ) = v ( t ) n (2.7)
v (t )
ic ( t ) = − i ( t ) n + (2.8)
R
ig ( t ) = 0 (2.9)
v (t )
ic ( t ) = − i ( t ) n−
Ts
Ts
(2.11)
R
ig ( t ) = 0 (2.12)
The average inductor voltage now can be found by averaging the subintervals over one
complete switching period. The result is
7
vL ( t ) Ts
= vg ( t ) d1 ( t ) + v ( t ) Ts
nd 2 ( t ) (2.13)
Ts
This leads to the following state equation for the average inductor current
d i (t )
= vg ( t ) d1 ( t ) + v ( t ) nd 2 ( t )
Ts
Lm Ts
(2.14)
dt Ts
The average capacitor current now can be found by averaging the subintervals over one
switching period resulting in the following:
v (t ) v (t )
ic ( t ) =− d1 ( t ) − i ( t ) nd 2 ( t ) − d2 (t )
Ts Ts
Ts Ts
(2.15)
R R
After collecting terms, the average capacitor current becomes
v (t )
ic ( t ) Ts
=−
R
Ts
(14
d (t ) + d (t )) − i (t )
1
4244 3 2 Ts
nd 2 ( t ) (2.16)
1
This leads to the following state equation for the average capacitor voltage
d v (t ) v (t )
=− − i (t ) nd 2 ( t )
Ts Ts
C Ts
(2.17)
dt R
The converter input current now can be found by averaging the subintervals over one
switching period. The result is
ig ( t ) = i (t ) Ts
d1 ( t ) + 0 (2.18)
Ts
The equation (2.14), (2.17), and (2.18) are nonlinear set of differential equations. In order to
construct the converter small-signal ac model, the next step is to perturb and linearize these
equations. Here, we assume that the converter input voltage vg ( t ) and duty cycle d1 ( t ) can
d1 ( t ) = D1 + dˆ1 ( t ) (2.20)
In response to these inputs, and after all transients have decayed, the average converter
variables can also be expressed as quiescent values plus small ac variations,
i (t ) Ts
= I + iˆ ( t ) (2.21)
v (t ) Ts
= V + vˆ ( t ) (2.22)
8
ig ( t ) = I g + iˆg ( t ) (2.23)
Ts
With these substitutions, the large-signal averaged inductor equation (2.14) becomes
(
d I + iˆ ( t ) )=
Lm
dt
(V g ( ) (
+ vˆg ( t ) ) D1 + dˆ1 ( t ) + (V + vˆ ( t ) ) n D2 − dˆ1 ( t ) ) (2.24)
As usual, this equation contains three types of terms. The dc term contain no time-varying
quantities. The first order ac terms are linear functions of the ac variations in the circuit,
while the second order ac terms are functions of the products of the ac variations. At this
point, we make an assumption that the ac variations are small in magnitude compared to the
dc quiescent values,
vˆg ( t ) << Vg
dˆ1 ( t ) << D1
iˆ ( t ) << I (2.26)
vˆ ( t ) << V
iˆg ( t ) << I g
If the small signal assumptions (2.26) are satisfied, then the second-order terms are much
smaller in magnitude than the first-order terms and hence can be neglected. The dc terms
must satisfy
0 = Vg D1 + VnD2 (2.27)
9
Upon substation of (2.19), (2.20), (2.21), (2.22), and (2.23) into the averaged capacitor
voltage state equation (2.17), we obtain the following
d (V + vˆ ( t ) ) (V + vˆ ( t ) ) −
C
dt
=−
R
( I + iˆ ( t )) n ( D 2 − dˆ1 ( t ) ) (2.29)
Here, we neglect the second-order terms. The dc terms of equation (2.30) must satisfy
V
0=− − InD2 (2.31)
R
The first-order ac terms of (2.30) lead to the small-signal ac state equation for capacitor
voltage
dvˆ ( t ) vˆ ( t )
C =− + Indˆ1 ( t ) − iˆ ( t ) nD2 (2.32)
dt R
Substation of (2.19), (2.20), (2.21), (2.22), and (2.23) into (2.18) results in the following:
( )(
I g + iˆg ( t ) = I + iˆ ( t ) D1 + dˆ1 ( t ) ) (2.33)
(
I g + iˆg ( t ) = ( ID1 ) + Idˆ1 ( t ) + iˆ ( t ) D1 +
{ 1442443 ) (14
iˆ ( t ) dˆ ( t ) )
243
1 (2.34)
Dc terms
1st order ac terms ( linear ) 2 nd order ac terms ( nonlinear )
We neglect the second-order terms in (2.34), leaving the following linearized ac equation
iˆg ( t ) = Idˆ1 ( t ) + iˆ ( t ) D1 (2.36)
This result represents the low-frequency ac variations in the converter input current.
The equation of the quiescent values, (2.27), (2.31), and (2.35) are collected below,
10
0 = Vg D1 + VnD2
V
0 = − − InD2 (2.37)
R
I g = ID1
For given quiescent values of the input voltage Vg , and duty cycle D1 , the system of
equations (2.37) can be evaluated to find the quiescent output voltage V , inductor current I ,
and input current dc component I g . The results are then inserted into the small-signal ac
The final step is to construct an equivalent circuit of (2.37) and (2.38) using commercially
available simulation tools.
Let us now apply the state-space averaging method to the second order flyback converter
circuit depicted in Figure 2.1(a). The independent state variables of the converter are the
inductor current i ( t ) and the capacitor voltage v ( t ) , which form the state vector
i (t )
x (t ) = (2.39)
v ( t )
The input voltage vg ( t ) , is an independent source which should be placed in the input vector,
u ( t ) = vg ( t ) (2.40)
To model the converter as a system with input and output, we need to find the converter input
current ig ( t ) . To calculate this dependent current, it should be included in the output vector
y ( t ) . Therefore,
y ( t ) = ig ( t ) (2.41)
Note that it’s not necessary to include the output voltage v ( t ) in the output vector since the
11
Next, let us write the state equations for each subinterval. When the switch is on and the
diode is off, the converter circuit of Figure 2.1(b) is obtained. The inductor voltage, capacitor
current, and converter input current are
di ( t )
Lm = vg ( t ) (2.42)
dt
dv ( t ) v (t )
C =− (2.43)
dt R
ig ( t ) = i ( t ) (2.44)
Similar to (2.1), (2.2), and (2.3), after organizing (2.42), (2.43), and (2.44), these equations
can be written in the following standard state-space form:
di ( t )
0 0 1
i (t )
dt = + Lm vg ( t )
1 (2.45)
dv ( t ) 0 − v ( t )
23 0 1 424 3
24
14 RC3 1 u (t )
{
1 dt 3
424 A1
x(t )
B1
dx ( t )
dt
i (t )
ig ( t ) = [1 0] + [ 0] vg ( t ) (2.46)
123 { v ( t ) { 1 424 3
y(t ) C1 123 E1 u ( t )
x(t )
So, (2.45) and (2.46) define the state-space equation for the first subinterval.
When the MOSFET is off and the diode is on, the converter circuit of Figure 2.1(c) is
obtained. For the second subinterval, the inductor voltage, capacitor current, and converter
input current are given by
di ( t )
Lm = v (t ) n (2.47)
dt
dv ( t ) v (t )
C = −i ( t ) n − (2.48)
dt R
ig ( t ) = 0 (2.49)
After organizing terms, the following state-space equation for the second interval is obtained:
12
di ( t ) n
0 L i ( t ) 0
dt = + vg ( t )
m
(2.50)
dv ( t ) n 1 v ( t ) { 0 1424 3
− − 123 u(t )
424
1 dt 3 14 C4244RC3 x( t ) B2
dx ( t ) A2
dt
i (t )
ig ( t ) = [ 0 0] + [ 0] vg ( t ) (2.51)
123 123 v ( t ) { 1 424 3
y(t ) C2 123 E2 u ( t )
x(t )
The next step is to evaluate the averaged state-space model, which is achieved as follows:
n nD2
0 0 0 0
Lm Lm
A = A1 D1 + A2 D2 = D + D2 = (2.52)
0 − 1 1 n 1 nD2 1
RC − C −
RC − C −
RC
1 D1
0
B = B1 D1 + B2 D2 = Lm D1 + D2 = Lm
(2.53)
0
0 0
C = C1 D1 + C2 D2 = [1 0] D1 + [ 0 0] D2 = [ D1 0] (2.54)
E = E1 D1 + E2 D2 = [ 0] D1 + [ 0] D2 = [ 0] (2.55)
So the final averaged state-space model of the second order Flyback converter without
parasitics in CCM is
di ( t ) nD2
0 D1
Lm i ( t )
+ Lm vg ( t )
dt = (2.56)
dv ( t ) nD2 1 v ( t )
− − 0
dt C RC
i (t )
ig ( t ) = [ D1 0] + [ 0] vg ( t ) (2.57)
v ( t )
To validate the analytically derived averaged state-space model in CCM, a detailed model of
Figure 2.1(a) has been constructed in PLECS [16]. The averaged state-space equations have
been constructed in Matlab/Simulink [30]. A hardware prototype of the subject converter has
13
been built to validate the results. The parameters of the second-order flyback converter
without parasitics in CCM are given in Appendix A.1. The predicted and measured inductor
current and capacitor voltage are shown in Figure 2.3.
1.6
1.4
1.2
I (Amp.)
0.4
-71
-72
Detailed Model
V (Volt)
Actual Average
See
-73 Figure 2.4 Analytical State-Space
Hardware Prototype
-74
Figure 2.3 Inductor current and capacitor voltage of second order Flyback converter without parasitics
in CCM.
-73.894
-73.896
Figure 2.4 Capacitor voltage of second order Flyback converter without parasitics in CCM.
As can be seen in Figure 2.3, the analytically derived state-space averaged model predicts the
averaged inductor current very well. As can be seen in Figure 2.3, there is a difference
between the capacitor voltage waveforms predicted by the detailed model and the
measurements from the hardware prototype, which comes from the simplified detailed model
14
(second-order without parasitics). As can be seen in Figure 2.4, the analytically derived state-
space averaged model predicts the average capacitor voltage very well.
(1 − D1 )
2 2
RL N1
Lm = (2.58)
2f N2
Here, D1 is the duty cycle, RL is the load resistance, f is the switching frequency, and
N1 N 2 is the turn ratio of the transformer.
In addition to 2 subintervals that occur in CCM, in DCM there is another subinterval that
occurs at light loads. The third interval results in the topological instance of the flyback
converter circuit shown in Figure 2.5.
ig (t ) i (t ) n
ic (t )
+
vg (t ) vL (t ) Lm
C R v (t )
Figure 2.5 Second-order Flyback converter without parasitics during third subinterval in DCM.
During the first subinterval [see circuit of Figure 2.1(b)], the MOSFET is on and the diode is
off, and the magnetizing inductance stores the energy. During the second subinterval [see
Figure 2.1(c)], the MOSFET is off and the diode is on, and the energy stored in the
15
transformer field is transferred to the secondary side. This energy then flows through the
filter capacitor and is supplied to the load resistor. If the magnetizing current during the
second subinterval goes to zero before the end of the second subinterval, the converter goes
into another stage (DCM, subinterval 3) before it goes back to the first interval again. The
magnetizing current predicted by the detailed model in DCM is shown in Figure 2.6 for the
load R = 2500Ω .
1.2
0.8
I (Amp.)
Ts
0
For the flyback converter without parasitics, the first and second subintervals in DCM are the
same as in CCM, as described in Section 2.1. The third subinterval comes into account when
the MOSFET and the diode are both off as shown in Figure 2.5. For this topology, the
inductor voltage vL ( t ) , capacitor current ic ( t ) , and converter input current ig ( t ) are
vL ( t ) = 0 (2.59)
v (t )
ic ( t ) = − (2.60)
R
ig ( t ) = 0 (2.61)
Let us now apply the state-space averaging method to third sub interval. The state vector
x ( t ) , the input vector u ( t ) , and the output vector y ( t ) are as define in Section 2.1 in (2.39),
(2.40), and (2.41), respectively. Next, let us write the state equation for third subinterval.
When the switch and the diode are off, the converter circuit of Figure 2.5 is obtained. The
inductor voltage, capacitor current, and converter input current are
16
di ( t )
Lm =0 (2.62)
dt
dv ( t ) v (t )
C =− (2.63)
dt R
ig ( t ) = 0 (2.64)
Similar to (2.59), (2.60), and (2.61), after organizing the terms in (2.62), (2.63), and(2.64),
the following state-space for the third subinterval is formed
di ( t )
0 0
i ( t ) + 0 v ( t )
dt = (2.65)
dv ( t ) 0 −
1 v ( t ) 0 1g
424 3
RC3 1 2 3 { u ()
t
1 dt 3 1424
424 A3
x(t ) B3
dx ( t )
dt
i (t )
ig ( t ) = [ 0 0] + [ 0] vg ( t ) (2.66)
123 123 v ( t ) { 1 424 3
y(t ) C3 123 E3 u ( t )
x(t )
So the next step is to evaluate the state-space averaged model, which goes as following:
A = A1 D1 + A2 D2 + A3 D3
n nD2
0 0 0 0 0 0
Lm
Lm (2.67)
= D + D2 + D =
0 − 1 1 n 1 0 − 1 3 nD2 1
RC − C −
RC
RC − C −
RC
1 D1
0 0
B = B1 D1 + B2 D2 + B3 D3 = Lm D1 + D2 + D3 = Lm
(2.68)
0 0
0 0
C = C1 D1 + C2 D2 + C3 D3 = [1 0] D1 + [ 0 0] D2 + [ 0 0] D3 = [ D1 0] (2.69)
E = E1 D1 + E2 D2 + E3 D3 = [ 0] D1 + [ 0] D2 + [ 0] D3 = [ 0] (2.70)
So the final state-space averaged equation of the second order flyback converter without
parasitics in DCM becomes:
17
di ( t ) nD2
0 D1
Lm i ( t )
+ Lm vg ( t )
dt = (2.71)
dv ( t ) nD2 1 v ( t )
− − 0
dt C RC
i (t )
ig ( t ) = [ D1 0] + [ 0] vg ( t ) (2.72)
v ( t )
Equations (2.71) and (2.72) are the state-space averaged model for the DCM, which is the
same as (2.56) and (2.57). It only happens when there are no parasitics.
To validate the analytically derived state-space averaged model in DCM, a detailed model of
the converter depicted in Figure 2.1(a) has been constructed in PLECS [16]. The state-space
averaged model has been constructed in Matlab/Simulink [30]. The interval d 2 has been
calculated as 0.4409 using (2.58) and RL = R = 2500Ω . The parameters of the second-order
flyback converter without parasitics in CCM are given in Appendix A.2. The resulting
waveforms of inductor current and capacitor voltage are shown in Figure 2.7.
1.2
0.8
I (Amp.)
0.4
Detailed Model
Actual Average
0 Analytical State-Space
-102.541
-102.543
V (Volt)
-102.545
-102.547
1.699991 1.699993 1.699995 1.699997 1.699999 1.7
Time (s)
Figure 2.7 Inductor current and capacitor voltage of second order Flyback converter without parasitics
in DCM.
18
As can be seen in Figure 2.7, the analytically derived state-space averaged model predicts the
magnetizing current with a large error of 18.25% . This error is because the actual inductor
current is discontinuous, which is not properly accounted by the classical state-space
averaging as will be explained in Section 2.5. At the same time, the analytically derived
state-space averaged model predicts the capacitor voltage and the output voltage with a very
small error because these state variables are continuous. The small error in this case comes
from the error in representing the magnetizing current according to (2.71).
2.3 Small-Signal AC Model and State-Space Averaging with Basic Parasitics in CCM
A second order flyback converter with basic parasitics is shown in Figure 2.8(a). Here, n is
the turn ratio of the transformer ( N1 N 2 ) .
+
ig (t ) i (t ) n
D Vd ic (t )
+
vg (t ) vL (t ) Lm C vc (t )
R v (t )
Rc
Mosfet
Rsw
(a )
ig (t ) i (t ) n
ic (t )
+
vg (t ) vL (t ) Lm C vc (t )
R v (t )
Rc
Rsw
(b )
+
ig (t ) i (t ) n
Vd ic (t )
+
vg (t ) vL (t ) Lm C vc (t )
R v (t )
Rc
(c )
Figure 2.8(a) Second-order Flyback converter with parasitics; (b) Circuit during subinterval 1 (c) Circuit
during subinterval 2.
19
During the first subinterval, when the MOSFET conducts and the diode is off, the circuit
reduces to Figure 2.8(b). For this interval, the inductor voltage vL ( t ) , capacitor current ic ( t ) ,
vL ( t ) = vg ( t ) − Rswig ( t ) (2.73)
vc ( t )
ic ( t ) = − (2.74)
R + Rc
vc ( t ) R
v (t ) = (2.75)
R + Rc
ig ( t ) = i ( t ) (2.76)
We next apply the small ripple approximation and replace the voltages and currents with
their respective average values to obtain the following:
vL ( t ) = v g ( t ) − Rsw ig ( t ) (2.77)
Ts Ts
vc ( t )
ic ( t ) = −
Ts
(2.78)
R + Rc
vc ( t ) R
v (t ) =
Ts
(2.79)
R + Rc
ig ( t ) = i ( t ) Ts
(2.80)
During the second subinterval, the MOSFET is off and diode conducts, which results in the
circuit of Figure 2.8(c). For this interval, the inductor voltage vL ( t ) , capacitor current ic ( t ) ,
vL ( t ) = ( vc ( t ) − ic ( t ) Rc − Vd ) n (2.81)
v (t )
ic ( t ) = − i ( t ) n + (2.82)
R
v ( t ) = vc ( t ) − ic ( t ) Rc (2.83)
ig ( t ) = 0 (2.84)
v (t )
ic ( t ) = − i ( t ) n−
Ts
Ts
(2.86)
R
v ( t ) = vc ( t ) Ts
− ic ( t ) Ts
Rc (2.87)
ig ( t ) = 0 (2.88)
The average inductor voltage now can be found by averaging the subintervals over one
complete switching period. The result is
vL ( t ) Ts
= vg ( t ) d1 ( t ) − Rsw ig ( t ) d1 ( t )
Ts Ts
(2.89)
+ vc ( t ) Ts
nd 2 ( t ) − ic ( t ) Ts
Rc nd 2 ( t ) − Vd nd 2 ( t )
This leads to the following equation for the average inductor current
d i (t )
= vg ( t ) d1 ( t ) − Rsw ig ( t ) d1 ( t )
Ts
Lm
dt Ts Ts (2.90)
+ vc ( t ) Ts
nd 2 ( t ) − ic ( t ) Ts
Rc nd 2 ( t ) − Vd nd 2 ( t )
The average capacitor current now can be found by averaging the subintervals over one
switching period, which results in the following:
vc ( t ) v (t )
ic ( t ) =− d1 ( t ) − i ( t ) nd 2 ( t ) − d2 (t )
Ts Ts
(2.91)
Ts
R + Rc Ts
R
This leads to the following equation for the average capacitor voltage
d vc ( t ) vc ( t ) v (t )
=− d1 ( t ) − i ( t ) nd 2 ( t ) − d2 (t )
Ts Ts Ts
C (2.92)
dt R + Rc Ts
R
The converter output voltage now can be found by averaging the subintervals over one
switching period, which results in
vc ( t ) R
v (t ) = d1 ( t ) + vc ( t ) d 2 ( t ) − ic ( t ) Rc d 2 ( t )
Ts
(2.93)
Ts
R + Rc Ts Ts
21
The converter input current can now be found by averaging the subintervals over one
switching period, resulting in the following:
ig ( t ) = i (t ) Ts
d1 ( t ) + 0 (2.94)
Ts
Equations (2.90), (2.92), (2.93), and (2.94) are nonlinear differential equations. Hence, to
construct the converter small-signal ac model, the next step is to perturb and linearize them.
We assume that the converter input voltage vg ( t ) and duty cycle d1 ( t ) can be expressed as
d1 ( t ) = D1 + dˆ1 ( t ) (2.96)
In response to these inputs, and after all transients have decayed, the averaged converter
waveforms can be expressed as quiescent values plus small ac variations as
ig ( t ) = I g + iˆg ( t ) (2.97)
Ts
i (t ) Ts
= I + iˆ ( t ) (2.98)
vc ( t ) Ts
= Vc + vˆc ( t ) (2.99)
ic ( t ) Ts
= I c + iˆc ( t ) (2.100)
v (t ) Ts
= V + vˆ ( t ) (2.101)
(
d I + iˆ ( t ) )=
Lm
dt
(V
g ( ) ( )(
+ vˆg ( t ) ) D1 + dˆ1 ( t ) − Rsw I g + iˆg ( t ) D1 + dˆ1 ( t ) )
( ) ( ) (
+ (Vc + vˆc ( t ) ) n D2 − dˆ1 ( t ) − I c + iˆc ( t ) Rc n D2 − dˆ1 ( t ) ) (2.102)
(
− Vd n D2 − dˆ1 ( t ) )
22
dI diˆ ( t )
Lm +
dt = (Vg D1 − Rsw I g D1 + Vc nD2 − I c Rc nD2 − Vd nD2 )
dt 1444444442444444443
Dc terms
As usual, this equation contains three types of terms. The dc term contains no time-varying
quantities. The first order ac terms are linear functions of the ac variations in the circuit,
while the second order ac terms are functions of the products of the ac variations. At this
point, we make an assumption that the ac variations are small in magnitude compared to the
dc quiescent values,
vˆg ( t ) << Vg
dˆ1 ( t ) << D1
iˆg ( t ) << I g
iˆ ( t ) << I (2.104)
vˆc ( t ) << Vc
iˆc ( t ) << I c
vˆ ( t ) << V
If the small signal assumptions (2.104) are satisfied, then the second-order terms are much
smaller in magnitude than the first-order terms and hence be neglected. The dc terms must
satisfy
0 = Vg D1 − Rsw I g D1 + Vc nD2 − I c Rc nD2 − Vd nD2 (2.105)
This is the linearized equation that describes ac variations in the inductor current.
23
Upon substation of (2.95)-(2.101) into (2.92), we obtain
d (Vc + vˆc ( t ) ) (V + vˆc ( t ) )
C
dt
=−
c
R + Rc
( D + dˆ ( t )) − ( I + iˆ (t )) n ( D
1 1 2 − dˆ1 ( t ) )
(2.107)
(V + vˆ ( t ) )
−
R
(D 2 − dˆ1 ( t ) )
dV dvˆ ( t ) V D VD
C c + c = − c 1 − InD2 − 2
dt dt R + Rc R
14444244443
Dc terms
V dˆ ( t ) vˆc ( t ) D1 Vdˆ ( t ) vˆ ( t ) D2
+− c 1 − + Indˆ1 ( t ) − iˆ ( t ) nD2 + 1 − (2.108)
R + Rc R + Rc R R
144444444444 42444444444444 3
1st order ac terms ( linear )
vˆ ( t ) dˆ1 ( t ) vˆ ( t ) dˆ1 ( t )
+− c + iˆ ( t ) ndˆ1 ( t ) +
R + Rc R
14444444 244444443
2 nd order ac terms ( nonlinear )
Here again we neglect the second-order terms. The dc terms of (2.108) must satisfy
Vc D1 VD
0=− − InD2 − 2 (2.109)
R + Rc R
The first-order ac terms of (2.108) lead to the following small-signal for the ac capacitor
voltage
dvˆc ( t ) Vc dˆ1 ( t ) vˆc ( t ) D1 Vdˆ1 ( t ) vˆ ( t ) D2
C =− − + Indˆ1 ( t ) − iˆ ( t ) nD2 + − (2.110)
dt R + Rc R + Rc R R
R + Rc
( )
R D1 + dˆ1 ( t ) + (Vc + vˆc ( t ) ) D2 − dˆ1 ( t ) ( ) (2.111)
− ( I + iˆ ( t ) ) R ( D
c c c 2 − dˆ ( t ) )
1
24
Vc D1 R
(V + vˆ ( t ) ) = R + R + Vc D2 − I c Rc D2
14444
c
244443
Dc terms
vˆ ( t ) Rdˆ1 ( t )
+ c − vˆc ( t ) dˆ1 ( t ) + iˆc ( t ) Rc dˆ1 ( t )
R + Rc
14444444 4244444444 3
2 nd order ac terms ( nonlinear )
We neglect the second-order terms in (2.112), leaving the following linearized ac equation
Vc Rdˆ1 ( t ) vˆc ( t ) RD1
vˆ ( t ) = + − Vc dˆ1 ( t ) + vˆc ( t ) D2 + I c Rc dˆ1 ( t ) − iˆc ( t ) Rc D2 (2.114)
R + Rc R + Rc
( )(
I g + iˆg ( t ) = I + iˆ ( t ) D1 + dˆ1 ( t ) ) (2.115)
(
I g + iˆg ( t ) = ( ID1 ) + Idˆ1 ( t ) + iˆ ( t ) D1 +
{ 1442443 ) (14
iˆ ( t ) dˆ ( t ) )
243
1 (2.116)
Dc terms
1st order ac terms ( linear ) 2 nd order ac terms ( nonlinear )
We neglect the second-order terms in (2.116), leaving the following linearized ac equation
iˆg ( t ) = Idˆ1 ( t ) + iˆ ( t ) D1 (2.118)
This result represents the low-frequency ac variations in the converter input current.
The equations of the quiescent values, (2.105), (2.109), (2.113), and (2.117) are collected
below as
25
0 = Vg D1 − Rsw I g D1 + Vc nD2 − I c Rc nD2 − Vd nD2
Vc D1 VD2
0=− − InD2 −
R + Rc R
(2.119)
Vc D1 R
V= + Vc D2 − I c Rc D2
R + Rc
I g = ID1
For given quiescent values of the input voltage Vg , the diode voltage drop Vd , and the duty
cycle D1 , the system (2.119) can be evaluated to find the quiescent output voltage V ,
inductor current I , input current I g , capacitor voltage Vc , and capacitor current I c .
However, in this problem there are 5 variables but there are only 4 equations. The fifth
equation can be the following
V = Vc − I c Rc (2.120)
The results are then inserted into the small-signal ac model.
The small signal ac model, (2.106), (2.110), (2.114), and (2.118), is summarized below
diˆ ( t )
Lm = Vg dˆ1 ( t ) + vˆg ( t ) D1 − Rsw I g dˆ1 ( t ) − Rswiˆ1g ( t ) D1 − Vc ndˆ1 ( t )
dt
+ vc ( t ) nD2 + I c Rc nd1 ( t ) − ic ( t ) Rc nD2 + Vd nd1 ( t )
ˆ ˆ ˆ ˆ
dv ( t )
ˆ V d ( t ) vc ( t ) D1
ˆ ˆ Vd ( t ) v ( t ) D2
ˆ ˆ
C c =− c 1 − + Indˆ1 ( t ) − iˆ ( t ) nD2 + 1 − (2.121)
dt R + Rc R + Rc R R
V Rd ( t ) vc ( t ) RD1
ˆ ˆ
ˆv ( t ) = c 1 + − Vc d1 ( t ) + vc ( t ) D2 + I c Rc d1 ( t ) − ic ( t ) Rc D2
ˆ ˆ ˆ ˆ
R + Rc R + Rc
iˆg ( t ) = Idˆ1 ( t ) + iˆ ( t ) D1
The final step is to construct an equivalent circuit of (2.119) and(2.121) using the
commercially available simulation tools.
26
Let us now apply the state-space averaging method to the second order flyback converter of
Figure 2.8(a). The independent state variables as usual are the inductor current i ( t ) and the
i (t )
x (t ) = (2.122)
vc ( t )
The input voltage vg ( t ) , and the diode voltage drop is an independent source, which should
i ( t )
y (t ) = g (2.124)
v (t )
Next, let us write the state equations for each subinterval. When the switch is on and the
diode is off, the converter circuit of Figure 2.8(b) is obtained. The inductor voltage, capacitor
current, output voltage, and converter input current are
di ( t )
Lm = vg ( t ) − Rswi ( t ) (2.125)
dt
dvc ( t ) vc ( t )
C =− (2.126)
dt R + Rc
vc ( t ) R
v (t ) = (2.127)
R + Rc
ig ( t ) = i ( t ) (2.128)
Similar to (2.73)-(2.76), after organizing the terms in (2.125)-(2.128), the result can be
written in the following state-space form
27
di ( t ) − Rsw
L 0 i ( t ) 1 0 v t
dt = m g( )
+ Lm V
(2.129)
dvc ( t ) 1 vc ( t )
0 − 123 1 0 0 1
42d4 3
424
dt 3 144424443
RC + Rc C x (t ) 424 3 u(t )
1 B1
dx ( t ) A1
dt
1 0
ig ( t ) i ( t ) 0 0 vg ( t )
= 0 R + (2.130)
123v (t ) vc ( t ) 0 0 Vd
R + Rc 123 123 1 424 3
y(t ) 14 4244 3 x(t ) E1 u(t )
C1
So the state-space equations for the first subinterval have been identified.
In the second subinterval, when the MOSFET is off and the diode is on, the converter circuit
of Figure 2.8(c) is obtained. For the second subinterval, the inductor voltage, capacitor
current, output voltage, and converter input current are given by
di ( t ) i ( t ) n 2 Rc R vc ( t ) nR
Lm = + − Vd n (2.131)
dt R − Rc R − Rc
dvc ( t ) i ( t ) nR vc ( t )
C =− − (2.132)
dt R − Rc R − Rc
i ( t ) nRc R vc ( t ) R
v (t ) = + (2.133)
R − Rc R − Rc
ig ( t ) = 0 (2.134)
di ( t ) n Rc R
2
nR
RL − R L i t 0 − n
RLm − Rc Lm ( ) v ( t )
dt = m c m
+ Lm g (2.135)
dvc ( t ) vc ( t ) 0
Vd
0 1
nR 1
− RC − R C − RC − R C 123 1424 3
42 43
1 dt 3
424 c c x(t ) u (t )
144444244444 3 B2
dx ( t ) A2
dt
28
0 0
ig ( t ) i ( t ) 0 0 vg ( t )
= nRc R R + (2.136)
v (t )
123 vc ( t ) 0 0 Vd
R − Rc R − Rc 123
144 123 142 43
y(t ) 42444 3 x(t ) E2 u(t )
C2
So the space-space equation of the second interval has also been identified.
The next step is to combine the result and obtain the state-space averaged model as
− Rsw n 2 Rc R nR
L 0
A = A1 D1 + A2 D2 =
m D1 + RLm − Rc Lm RLm − Rc Lm
D
1 nR 1 2
0 − − −
RC + Rc C RC − Rc C RC − Rc C
(2.137)
− Rsw D1 n 2 Rc RD2 nRD2
+
Lm RLm − Rc Lm RLm − Rc Lm
=
nRD2 D1 D2
− − −
RC − Rc C RC + Rc C RC − Rc C
1 n D1 nD2
0 0 − −
B = B1 D1 + B2 D2 = Lm D + Lm D2 = Lm Lm (2.138)
1
0 0 0 0 0 0
1 0 0 0
C = C1 D1 + C2 D2 =
0 R D1 + nRc R
R D2
R + Rc R − Rc R − Rc
(2.139)
D1 0
= nRc RD2 RD1 RD2
+
R − Rc R + Rc R − Rc
0 0 0 0 0 0
E = E1 D1 + E2 D2 = D1 + D2 = (2.140)
0 0 0 0 0 0
Therefore, the final state-space averaged model of the second order flyback converter with
parasitics in CCM becomes
29
di ( t ) − Rsw D1 n 2 Rc RD2 nRD2
+
i ( t )
Lm RLm − Rc Lm RLm − Rc Lm
dt =
dvc ( t ) nRD2 D1 D2 vc ( t )
− − −
dt RC − Rc C RC + Rc C RC − Rc C (2.141)
D1 nD2
− v ( t )
+ Lm Lm g
V
0 0 d
D1 0
ig ( t ) i ( t ) 0 0 vg ( t )
= nRc RD2 RD1 RD2 + (2.142)
v (t ) R − R + vc ( t ) 0 0 Vd
c R + Rc R − Rc
To validate the analytically derived state-space averaged model in CCM, a detailed model of
Figure 2.8(a) has been constructed in PLECS [16]. The state-space averaged model has been
constructed in Matlab/Simulink [30]. The same hardware prototype has been used here to
validate the results. The parameters of the second-order flyback converter with basic
parasitics in CCM are given in Appendix A.3. The predicted and measured inductor current,
capacitor voltage, and output voltage are shown in Figure 2.9.
As can be seen in Figure 2.9, the analytically derived state-space averaged model predicts the
averaged inductor current and capacitor voltage very well. There is a difference between the
detailed model and hardware prototype in terms of the output voltage (see Figure 2.9), which
comes from the simplified detailed model (second-order with parasitics). As seen in Figure
2.10, the analytically derived state-space averaged model predicts the average output voltage
very well.
30
1.5
I (Amp.)
1
-72.492
-72.496
-71.4
-71.8
V (Volt)
Detailed Model
See Actual Average
-72.2 Figure 2.10 Analytical State-Space
Hardware Prototype
-72.6
0.999991 0.999993 0.999995 0.999997
Time (s)
Figure 2.9 Inductor current, capacitor voltage and output voltage of second order Flyback converter with
parasitics in CCM.
-72.49
-72.495
-72.5
Figure 2.10 Output voltage of second order Flyback converter with parasitics in CCM.
31
2.4 State-Space Averaging with Parasitics in DCM
In addition to the two subintervals that occur in CCM, here there is another subinterval that
occurs at light loads. The topology of the flyback converter in this subinterval is shown in
Figure 2.11.
ig (t ) i (t ) n
ic (t )
+
vg (t ) vL (t ) Lm C vc (t )
R v (t )
Rc
Figure 2.11 Second-order Flyback converter with parasitics during third subinterval in DCM.
During the first subinterval depicted in Figure 2.8(b), the MOSFET is on and the diode is off,
and the magnetizing inductance stores some energy. During the second subinterval depicted
in Figure 2.8(c), the MOSFET is off and the diode is on. During this interval, the stored
energy is transferred to secondary side and it flows through the filter capacitor to the load
resistor. If the magnetizing current during the second interval goes to zero before the end of
the second interval, the converter goes to another stage resulting in third subinterval and
DCM, before it goes back to the first interval in the next cycle.
For the flyback converter without parasitics in DCM, the first and second sub intervals
remain of the same as in CCM as described in Section 2.3. The third interval comes into
account when the MOSFET and the diode are off as in Figure 2.11. For this case, the
inductor voltage vL ( t ) , capacitor current ic ( t ) , output voltage v ( t ) , and converter input
current ig ( t ) are
vL ( t ) = 0 (2.143)
vc ( t )
ic ( t ) = − (2.144)
R + Rc
vc ( t ) R
v (t ) = (2.145)
R + Rc
32
ig ( t ) = 0 (2.146)
Let us now apply the state-space averaging method to the third interval of Figure 2.11. The
inductor voltage, capacitor current, output voltage, and converter input current can be written
as
di ( t )
Lm =0 (2.147)
dt
dvc ( t ) vc ( t )
C =− (2.148)
dt R + Rc
vc ( t ) R
v (t ) = (2.149)
R + Rc
ig ( t ) = 0 (2.150)
di ( t )
0 0
i ( t ) + 0 0 vg ( t )
dt = (2.151)
dvc ( t ) 0 −
1 vc ( t ) 0 0 V
144 RC + Rc C 123 123 1 42d4 3
424
dt 3 4 2444 3 x(t ) B3 u ()
t
1 A3
dx ( t )
dt
0 0
ig ( t ) i ( t ) 0 0 vg ( t )
= 0 R + (2.152)
123v (t ) vc ( t ) 0 0 Vd
R + Rc 123 123 1 424 3
y(t ) 14 4244 3 x(t ) E3 u(t )
C3
which defines the state-space equation for the third subinterval. Hence, the next step is to
evaluate the state-space averaged equations, which goes as following:
33
A = A1 D1 + A2 D2 + A3 D3
− Rsw n 2 Rc R nR
L 0 0 0
= m D1 + RLm − Rc Lm RLm − Rc Lm
D + 1 D (2.153)
1 nR 1 2 0 − 3
0 − − − RC + Rc C
RC + Rc C RC − Rc C RC − Rc C
− Rsw D1 n 2 Rc RD2 nRD2
+
Lm RLm − Rc Lm RLm − Rc Lm
=
nRD2 D1 + D3 D2
− − −
RC − Rc C RC + Rc C RC − Rc C
1 n D1 nD2
0 0 − 0 0 −
B = B1 D1 + B2 D2 + B3 D3 = Lm D + Lm D2 + D3 = Lm Lm (2.154)
1 0 0
0 0 0 0 0 0
1 0 0 0 0 0
C = C1 D1 + C2 D2 + C3 D3 =
R D1 + nRc R
R D2 + R D3
0 0
R + Rc R − Rc R − Rc R + Rc
(2.155)
D1 0
= nRc RD2 R ( D1 + D3 ) RD2
+
R − Rc R + Rc R − Rc
0 0 0 0 0 0 0 0
E = E1 D1 + E2 D2 + E3 D3 = D1 + D2 + D3 = (2.156)
0 0 0 0 0 0 0 0
The final state-space averaged model for the second order flyback converter with parasitics
in DCM becomes:
D1 0
ig ( t ) i ( t ) 0 0 vg ( t )
= nRc RD2 R ( D1 + D3 ) RD2 + (2.158)
v (t ) R − R + vc ( t ) 0 0 Vd
c R + Rc
R − Rc
34
To validate the analytically derived state-space averaged model in DCM, a detailed model of
the converter circuit depicted Figure 2.8(a) has been constructed in PLECS [16]. The state-
space averaged model has been constructed in Matlab/Simulink [30].The variable d 2 has
been calculated using (2.58) and set to 0.4409 . The load was assumed as RL = R = 2500Ω .
The parameters of the second-order flyback converter without parasitics in CCM are given in
Appendix A.4. The predicted inductor current and capacitor voltage are shown in Figure
2.12.
1
I (Amp.)
0.5
-101
Detailed Model
Vc (Volt)
Actual Average
-101.5
Detailed Model and
Actual Average Analytical State-Space
-102
-101
V (Volt)
-101.5
Detailed Model and
Actual Average
-102
0.999991 0.999993 0.999995 0.999997
Time (s)
Figure 2.12 Inductor current, capacitor voltage and output voltage of second order Flyback converter
with parasitics in DCM.
As can be seen in Figure 2.12, the analytically derived state-space averaged model predicts
the magnetizing current with a large error of 17.5% . This error is because the current is
discontinuous, while the conventional state-space averaging fails to correctly take this into
35
account. At the same time, the analytically derived state-space averaged model predicts the
capacitor voltage with a very small error because the capacitor voltage is a continuous state
variable. The small error comes from the error in the magnetizing current defined by (2.157).
( ) ( )
•
x ( t ) = q ( t ) A1 + (1 − q ( t ) ) A2 x ( t ) + q ( t ) B1 + (1 − q ( t ) ) B2 u ( t ) (2.159)
where q ( t ) is the switching function, Ak , Bk are the system matrices, and u ( t ) is the input
vector. By definition, the so-called fast average of a state variable x ( t ) over a switching
interval is
t +Ts
1
x (t ) = ∫ x ( t )dt (2.160)
Ts t
where x ( t ) is the actual or the true average of x ( t ) . Since the averaging is commutative
with respect to differentiation, taking the fast average of (2.159) over a switching interval
yields
( ) ( )
•
x ( t ) = d ( t ) A1 + (1 − d ( t ) ) A2 x ( t ) + d ( t ) B1 + (1 − d ( t ) ) B2 u ( t ) (2.161)
This result follows from the fact that the fast average of the switching function over a
switching interval is the duty cycle function
t +Ts
1
d (t ) = ∫ q ( t )dt (2.162)
Ts t
It is also assumed that the average of the product is equal to the product of the averages,
especially
Ax = A ⋅ x (2.163)
Bu = B ⋅ u (2.164)
Assumption (2.163) is acceptable if the original switching variables do not deviate
significantly from their average values and also the system matrices A1 and A2 are
36
commutative [18, 34]. But in general, the matrices A1 and A2 are not commutative [35, 36].
So, if we make assumption (2.163), the equation (2.161) is not an exact solution as
mentioned in [18] and Appendix A, but can only be an approximation. Assumption (2.164) is
usually accepted when the source ripple is neglected.
The DCM operation of PWM converters differs from CCM operation by an additional time
interval in each switching cycle during which the inductor current or capacitor voltage is
clamped to zero (or a constant when there are multiple energy storage elements).
Conventional state-space averaging for converters working in DCM has been summarized in
[4, 10, 18]. For this mode, the direct extension of (2.159)-(2.161) results in
•
x ( t ) = ( d1 A1 + d 2 A2 + d 3 A3 ) x ( t ) + ( d1 B1 + d 2 B2 + d 3 B3 ) u ( t ) (2.165)
which is no longer accurate. In particular, the local average of the magnetizing current in the
third interval is zero, whereas the state-space averaging implies that this value should be d3 i .
Since d3 and i are not zero, the result of the state-space averaging is not zero. That is why
the discontinuous averaged variable in DCM is higher than the actual averaged value that can
be seen in Figure 2.7 and Figure 2.12.
As shown in Sections 2.2 and 2.4, the conventional state-space averaging is no longer
accurate in DCM. The parametric average-value modeling has been proposed to solve this
problem as documented in numerous publications [22, 24-27].
37
To make state-space averaging work properly in DCM, the so-called correction matrix M is
added in (2.166). The analytical derivation of the correction matrix for an ideal topology
(without parasitics) is given in [4]. Since the circuit of Figure 2.8(a) with parasitics has been
considered in this study, it would be very complicated to extend that analytical solution.
Hence, a numerical solution has been adopted in this Thesis.
The correction term is a diagonal matrix, wherein each state variable has its own correction
coefficient. In this implementation presented in this Thesis, instead of a diagonal matrix, a
column correction vector (each state variable has its own correction term as an element of
column matrix) is proposed as
•
3 3
x = ∑ ( d k Ak ) ( x . ∗ M ) + ∑ d k Bk u (2.167)
k =1 k =1
where ( x . ∗ M ) denotes the element wise multiplication. This operation uses less
computational time compare to direct implementation of (2.166).
functions of the duty-cycle ( d1 ) and the average value of the state variables ( x ) . To obtain
the values of d3 ( d1 , x ) and M ( d1 , x ) , the detailed model is run in the operation region of
interest (for example; duty-cycle changes between 0.1 and 0.9, and the load ( R ) changes
from very low load to very high load) whereas the state variables are averaged numerically
over the prototyping switching interval and saved for the future use. In particular, the
average-value of the state vector ( x ) is computed in a steady-state corresponding to a given
38
3 3
0 = ∑ ( d k Ak ) ( x . ∗ M ) + ∑ d k Bk u (2.168)
k =1 k =1
From which an intermediate variable vector p is computed as
−1
3 3
p = ( x . ∗ M ) = − ∑ ( d k Ak ) ⋅ ∑ d k Bk u (2.169)
k =1 k =1
Thereafter the elements of M can be found using the following
M j = p j ⋅ ÷x j (2.170)
where j denotes number of state variables and ( p j ⋅ ÷ x j ) denotes element wise division.
To obtain the functions d3 ( d1 , x ) and M ( d1 , x ) for the desired operation range, the detailed
simulation is run with different values of control variable ( d1 ) as well as the load ( R ) . The
variables resulting from this procedures are d1 , d 2 , d 3 , R, vc , i and the correction vector ( M ) is
computed using (2.169) and (2.170). These variables are stored for future use in lookup
tables, wherein an interpolation/extrapolation may be used as necessary. The real challenge
here is to calculate the d 2 or d 3 at any given operation point. If one can calculate either one,
thereafter it becomes easy to calculate the other one since d1 is the control variable which
can be calculated from the magnetizing current. As can be seen in Figure 2.12, the
magnetizing current is zero during the third subinterval whereas the variable d3 can be
calculated.
The final numerical function for d 3 is plotted in Figure 2.13, which shows that this function
has a flat surface corresponding to CCM, and varies linearly along the d1 . In DCM, the
surface of d 3 becomes non-linear and increases.
39
DCM
0.7
0.6
0.5
CCM
0.4
d3
0.3
0.2
5000
0.1 4000
3000
0 2000
1000 R (ohm)
1 0.9 0.8 0.7 0.6 0
0.5 0.4 0.3 0.2
d1 0.1
function of duty-cycle ( d1 ) and the load resistance ( R ) . As can be seen in Figure 2.14, on
the one hand, the correction term ( m1 ) has a flat surface and value of 1 in CCM, which
implies that the state-space averaging captures the correct value of this current in CCM and
no correction is required. On the other hand, it has values higher than 1 corresponding to the
DCM because the conventional state-space averaging does not capture the correct average
values in this region of operation as explained in Section 2.5, and therefore the correction
effort is required.
2.6
2.4 DCM
2.2 CCM
2
1.8
m1
1.6
1.4
1.2 5000
4000
1 3000
2000 R (ohm)
0.8 1000
1 0.9 0.8 0.7 0.6 0.5 0
0.4 0.3 0.2 0.1
d1
Figure 2.14 The correction term m1 as a function of duty-cycle ( d1 ) and the load resistance ( R ) .
40
The correction term ( m2 ) of filter capacitor voltage ( vc ) is plotted in Figure 2.15 as a
function of the duty-cycle ( d1 ) and the load resistance ( R ) . As can be seen in Figure 2.15,
the calculated value for this correction term is 1 for either CCM or DCM. This value is
consistent with the fact that the capacitor voltage is a continuous state variable with relatively
small ripple, so the conventional state-space averaging predicts the correct average value for
this variable for both modes.
1.01
m2
0.99
5000
4000
3000
2000 R (ohm)
0.98 1000
1 0.9 0.8 0.7 0.6 0.5 0
0.4 0.3 0.2 0.1
d1
Figure 2.15 The correction term m2 as a function of duty-cycle ( d1 ) and the load resistance ( R ) .
Once the parametric functions d3 ( d1 , x ) and M ( d1 , x ) have been calculated and stored, these
functions become available for the model implementation. Finally, the parametric average-
value model is implemented according to the block diagram shown in Figure 2.16. The
system matrices ( Ak , Bk , Ck , Dk ) for each subinterval are calculated numerically. For a given
value of control variable d1 and state vector x , the values of d 3 and correction vector ( M )
are acquired through the lookup tables. Total average system matrices ( AT , BT , CT , DT ) are
computed using the system matrices ( Ak , Bk , Ck , Dk ) and the variables d1 , d 2 and d 3 . These
matrices are then used to build the new continuous non-linear state-space average-value
model that replaces the discontinuous detailed model. Thereafter, this parametric average-
value model can be used for large-signal transient studies as well as for numerical
linearization and subsequent small-signal frequency-domain analysis.
41
x
·
M x = AT (x ×*M ) + BT u
d 2 = 1 - d1 - d3 y = CT x + DT u y
3
AT = å (d k Ak )
d1 d3 (d1 , x ) k =1
M (d1 , x ) d3 3
BT = å (d k Bk )
k =1
3
CT = å (d k Ck )
Ak , Bk , Ck , Dk k =1
3
DT = å (d k Dk )
k =1
u
The previously proposed parametric average-value models [24-27] had problem with the
correcting the output equation. In particular, in the previous formulation the output equation
was defined as
3 3
y = ∑ ( d k Ck ) Mx + ∑ d k Dk u (2.171)
k =1 k =1
However, my simulation results (which are not shown in this study) have shown that if there
is a correction term ( M ) in the output equation, in this case no correction occurs on the
output voltage (V ) . The reason for this is that the state vector ( x ) has already been corrected
in the state equation (2.167), and therefore no additional correction is required in the output
equation. The analytical proof of this argument goes as follow:
During the steady-state, we have
0 = AT Mx + BT u ⇒ x = − M −1 AT−1 BT u (2.172)
When (2.172) is inserted into (2.171), the output equation becomes
y = CT M ( − M −1 AT−1 BT u ) + DT u (2.173)
42
3 3
y = ∑ ( d k Ck ) x + ∑ d k Dk u (2.175)
k =1 k =1
43
1
I (Amp.)
0.5
Actual Average
and PAVM
0
-101
Detailed Model
Vc (Volt)
Actual Average
Detailed Model,
-101.5 Actual Average Analytical State-Space
and PAVM
PAVM
-102
-101
V (Volt)
Detailed Model,
-101.5 Actual Average
and PAVM
-102
0.999991 0.999993 0.999995 0.999997
Time (s)
Figure 2.17 Simulated inductor current, capacitor voltage and output voltage of the second order Flyback
converter with parasitics in DCM.
The accuracy of the PAVM in predicting the large-signal behaviour in time-domain has also
been verified by studying the effect of sudden change in load. In the following study, the
output of the flyback converter was regulated using a PI controller. The PI controller was
designed to regulate the output voltage at −72V by adjusting the duty-cycle ( d1 ) . The
controller parameters are K p = 0.7 and Ki = 25 . In this study, the converter run with a
resistive load of 2000Ω (which results in DCM) until it reaches a steady-state. At t = 0.2s , a
parallel load of 500Ω is added (which results in CCM). The resulting time-domain transients
are shown in Figure 2.18.
44
3
Detailed Model
2
PAVM
I (Amp.)
1
-71.6
Detailed Model
Vc (Volt)
and PAVM
-71.8
-72
-71.6
Detailed Model
and PAVM
V (Volt)
-71.8
-72
Figure 2.18 Transients in inductor current, capacitor voltage and output voltage of the second order
Flyback converter due to the step change in load.
As it can be observed in Figure 2.17, after the load change at t = 0.2s , the converter switches
operation from DCM to CCM. After load changes the control action brings the output
voltage to about the desired −72V after about 0.15s which wasn't shown in Figure 2.18 to
demonstrate the response of the PAVM to load change. At the same time, as it can be seen in
Figure 2.17, through all transients the large signal behaviour of the detailed model is
accurately predicted by proposed PAVM.
45
from the detailed simulation and the PAVM corresponding to full load operation condition
defined by R = 717.05Ω and d1 = 0.381 . The magnitude and phase of the corresponding
control-to-output transfer function are plotted in Figure 2.19.
The transfer function is evaluated up to 150kHz , which is more than one-half of the switching
frequency ( 250kHz ) . Closer to the switching frequency the results become distorted due to the
interaction between the injected perturbations and the converter switching. In general,
considering the frequencies closer to and above the switching frequency has limited use for
the average-value model since the basic assumptions of the averaging are no longer valid.
30
20
Amplitude (dB)
10
0
PAVM
-10
Detailed
-20
0
-10
-20
-30
Phase ( ° )
-40
-50
-60
PAVM
-70
-80
Detailed
-90
2 3 4 5
10 10 10 10
Frequency (Hz)
Figure 2.19 Control-to-output transfer function of the second-order Flyback converter evaluated at
R = 717.05Ω and d1 = 0.381 .
46
Chapter 3 : Analysis of Flyback Converter with Snubber Circuits
Real semiconductor devices experience voltage and current stress during the turn-on and
turn-off transitions which can easily damage the circuit elements. If a power electronic
converter stresses a power semiconductor device beyond its ratings, there are two basic ways
of relieving the problem. Either the device can be replaced with one whose ratings exceed the
stresses, or a snubber circuits can be added to the converter in order to reduce the stress to
safe levels. The final choice always involves a trade-off between the cost and availability of
semiconductor devices with required electrical ratings compared to the cost and additional
complexity of using the snubber circuits [9, 37-40]. In some topologies, such as a transformer
isolated Flyback converter, the snubber circuit may be required to protect the switching
transistor due to non-zero transformer leakage inductance.
Lp Ls D Vd
+
+ Lm C
Vg
R V
Cds Rds Rc
Css Mosfet
Rss
Rsw
300
200
Vs (Volt)
100
-100
(a )
300
200
Vs (Volt)
100
-100
5.983 5.984 5.985 5.986 5.987 5.988 5.989
Time (s) -3
x10
(b )
Figure 3.2 Measured transformer secondary voltage: (a) without the diode snubber; and (b) with the
diode snubber.
As can be seen in Figure 3.2(a), when the snubber is not used on the secondary side, the
voltage spikes up to 310V and the oscillations die pretty much at the end of the switching
interval. This phenomena causes two problems. First, the spike voltage can cause
breakdowns on the switching device. Secondly, the ringing energy will be radiated creating
noise and electro-magnetic interference (EMI) issues with potential for logic and control
errors. This type of ringing is not acceptable and it is necessary to add the snubber circuit
elements to damp the ringing, or clamp the voltage (with RCD clamps), or active snubbers
[40]. In the hardware prototype, a conventional RC snubber is used because of its simplicity
and low part count. Calculation of the RC elements is not the interest of this Thesis, whereas
the reader can find more information in [9, 37-39].
48
It must be emphasized that the snubbers are not a fundamental part of a power electronic
converter circuit. However, when it comes to averaging of PWM converters, the snubbers
significantly add to the complexity of the problem, and to the best of our knowledge this has
not been considered in the prior literature. At the same time, since the snubber circuits have
an important role in overall system dynamics, it is very desirable to take them into account in
the average-value modeling. On the one hand, the benefit of considering the snubbers will be
the improved accuracy of the average model in terms of capturing and predicting the
converter losses and efficiency, which will be used in Section 4.4.3. On the other hand, their
presence in the circuit has introduced more complexity and theoretical problems which are
almost impossible or impractical to resolve using conventional analytical derivations of the
average-value models as will be explained in more detail in Section 3.2.
A detailed analysis of the fifth-order Flyback converter will be given in Chapter 4. In this
Section, we will focus on only the problem caused by the snubbers, not the whole converter
circuit. The problem occurs on the output filter capacitor voltage, which also translates to the
output voltage which has been defined as an output in state-space output equation. To
demonstrate this problem caused by the snubbers, a detailed model has been built using
PLECS [16]. The corresponding converter parameters are summarized in Appendix A.5. The
state matrices Ak , Bk , Ck , Dk have been extracted for each subinterval in CCM. Finally, the
state-space averaged model has been calculated using these matrices and given duty ratio.
The simulated output filter capacitor voltage and the output voltage are shown Figure 3.3.
49
0
-20
-60
-80
0
Detailed Model
-20
Actual Average
V (Volt)
State-Space Averaging
-40 Detailed Model
and Actual Average
-60
-80
0.0499 0.04991 0.04992 0.04993
Time (s)
Figure 3.3 Simulated output filter capacitor voltage and the output voltage of the fifth-order Flyback
converter with snubbers in CCM.
As can be seen in Figure 3.3, the output voltage and capacitor voltage is about −72V , which
corresponds to the considered operating point in CCM. However, the state-space averaged
predicts the very low value of about −1.25V . Since the output voltage and the capacitor
voltage are continuous and slow variables, the conventional state-space averaging should
predict the average value correctly but in this can it does not! The reason why this is
happening can be understood by examining the Figure 3.1. The output filter capacitor voltage
is created by the charging current that flows through the diode. On the one hand, if the
average current is predicted correctly, the capacitor voltage will also be correct. But on the
other hand, if we look at the final state-space averaged equation (3.1), the extracted from the
detailed model diode current is not a state variable. As a result, the output capacitor voltage
does not depend on the diode current in the averaged state-space model. Instead, the output
filter capacitor voltage depends on the secondary side current ( I s ) , the diode snubber
capacitor voltage (VC ds
) , and the diode voltage drop (Vd ) . The diode voltage drop is a small
constant that is not significant in this discussion. Instead, let us look at the averaged
secondary current ( I s ) , and the diode snubber capacitor voltage (VC ds
) shown in Figure 3.4.
50
•
vc
• −205.33 0 0 28393 141.96 vc
vCss 0 −7.946 × 10 7
1.333 × 109 0 0 v
• Css
ip = 2.593 × 10 −1.572 × 10 −1.572 × 10
5 6 7
3.115 × 107 1.557 × 10 ip
5
• −43538
2.6 × 105 2.6 × 106 −5.231 × 106 −26157 is
i
s 3.123 × 107 0 0 3.752 × 109 −3.123 × 107 vCds
• (3.1)
vCds
0 −141.96
0 0
V
+ 2.509 × 106 2.593 × 105 g
Vd
−4.151 × 10 −43543
5
0 3.123 × 107
0.4
Detailed Model
State-Space Averaging
Is (Amp.)
-0.2
-0.4
200
100
Actual Average and
VC (Volt)
-100
0.01232 0.012321 0.012322 0.012323 0.012324 0.012325 0.012326
Time (s)
Figure 3.4 The predicted secondary current and the diode snubber capacitor voltage of fifth-order
Flyback converter in CCM.
As can be seen in Figure 3.4, the averaged state-space model predicts the secondary current
very low at −0.01 , but its actual value must be about −0.1 . The reason being is that the
secondary current is a discontinuous variable even though the converter operates in CCM.
Also, the diode snubber capacitor voltage represents another problem. Here, he averaged
state-space model predicts this value as zero which is also the actual averaged because of the
51
characteristic of the snubber as it releases the energy stored during the ringing. As a result,
the output filter capacitor voltage has a very large error which translates into the large error
in predicted output voltage.
For the purpose of further investigation, let us look at a circuit without a diode snubber,
shown in Figure 3.5. This circuit is a forth-order system.
Lp Ls D Vd
+
n
+ Lm C
Vg
R V
Rc
Css Mosfet
Rss
Rsw
A detailed model of this simplifies fourth-order converter has been built in PLECS [16]. The
corresponding parameters are summarized in Appendix A.5. The state-space matrices
Ak , Bk , Ck , Dk have been extracted for the CCM in each subinterval. Finally, the state-space
averaged model has been implemented using these matrices and the given duty ratio. The
corresponding simulated output filter capacitor voltage and the output voltage are shown in
Figure 3.6. As can be seen in Figure 3.6, the state-space averaged model predicts the output
filter capacitor Vc and output voltage V with a very small error. The reason for this small
error is the discontinuous secondary current, which the state-space averaging cannot predict
this current correctly.
52
-72.24
Vc (Volt)
-72.248
-72.252
-72.256
-72.23
-72.24
-72.25
V (Volt)
-72.26
-72.27
-72.28
0.199991 0.199993 0.199995 0.199997 0.199999
Time (s)
Figure 3.6 The simulated output filter capacitor and output voltage of the forth-order Flyback converter
without the diode snubber in CCM.
Let us now look at the averaged state-space equation (3.2) of the fort-order flyback converter.
The output filter capacitor voltage depends on only the secondary current. Since the state-
space averaging predicts the secondary current with an error, because of the discontinuity,
the output filter capacitor has the error but much less than the fifth-order model.
•
vc −63.383 0 0 28042 vc
•
vCss 0 −8.116 × 107 1.316 × 109 0 vCss
• = 2.561 × 105 −1.548 × 106 −1.548 × 107 23049 ip
ip
• −43000 2.561 × 105 2.561 × 106 −3870 is
i (3.2)
s
0 0
V
+
0 0 g
1.562 × 106 2.561 × 10 Vd
5
−2.561 × 10 −43005
5
As it is shown in this Section, the state-space averaging has a problem when the diode
snubber is present. The way to fix this problem is to introduce the diode current as a variable
53
in the averaged state-space model. To achieve that, the circuit has to be separated into two
circuits as shown in Figure 3.7.
Lp Ls D Vd
+
n
+ Lm C
Vg
R V
Cds Rds Rc
Css Mosfet
Rss
Rsw
Circuit 1 Circuit 2
Figure 3.7 Modified fifth-order Flyback converter circuit.
The Circuit 1 is a switched circuit, and on the other Circuit 2 is non switched and can be
removed from the state-space averaging. After such partitioning, it is easy to extract the state
matrices of the Circuit 1 in CCM or DCM for each subinterval using detailed model. Here,
the diode current must be defined as an output in the state-space model, which will be the
input of the Circuit 2. Then, the state-space equations of the Circuit 2 can either be extracted
from the detailed model or calculated analytically. After extracting state-space model for
these circuits, the overall system can be modelled in CCM as shown in Figure 3.8.
State Variables
State Variable
vCss ip is vCds vc
Vg · ·
x = ( A1d1 + A2 d 2 ) x + (B1d1 + B2 d 2 )u u x = Ax + Bu
u Vd v
y = (C1d1 + C2 d 2 ) x + (E1d1 + E2 d 2 )u id y = Cx + Eu
Circuit 1 Circuit 2
Figure 3.8 Proposed state-space averaged model of the fifth-order Flyback converter using two sub-
circuits and sub-models.
If the proposed state-space averaged model is implemented, it will be seen that the predicted
output filter capacitor voltage vc and output voltage v is same as Figure 3.3, which is still
54
incorrect. The reason is that the diode current is a discontinuous variable, and so the state-
space averaging of the Circuit 1 does not predict the diode current correctly, which translates
into the corresponding error in the predicted output voltage.
55
Chapter 4 : Full-order Flyback Converter
system matrices Ak , Bk , Ck , Dk .
Rp Lp Ls Rs D Vd
+
n
+ Lm C
Vg
R V
Cds Rds Rc
Css Mosfet
Rss
Rsw
The detailed model has been implemented using PLECS [16]. The converter parameters are
summarized in Appendix A.6. The state matrices Ak , Bk , Ck , Dk have been extracted for each
subinterval in CCM. Finally, the state-space model has been implemented using these
matrices and the given duty ratio. The simulated state variables are shown in Figure 4.2.
56
As can be seen in Figure 4.2, the state-space averaging predicts the switch snubber capacitor
voltage (VC ss
) , and the diode snubber capacitor voltage (V )
Cds correctly as expected because
these are continuous state variables. However, the state-space averaging predicts the output
filter capacitor voltage (Vc ) with a large error. Moreover, the transformers primary ( I p ) and
secondary current ( I s ) , are also predicted with an error. First, the currents deviate from their
average values because of the oscillations seen during the topology changes and as a result
the assumption (2.163) is no longer valid as explained in Section 2.5. Second, both currents
are discontinuous variables although the converter operates in CCM. Regardless of the mode
of operation, CCM or DCM, these currents as individual state variables are discontinuous
and (2.161) is no longer accurate. In particular, the local average of the primary current in the
second interval ( d 2T ) is zero (see Figure 4.2), whereas the state-space averaging implies that
this value should be d 2 ip . Since d 2 and ip are not zero, the result of the state-space
As a result, when it comes to state-space averaging of Flyback converters (if the transformer
model has leakage inductances, which means primary and secondary currents are state
variables), it needs special consideration in CCM as well as in DCM which will be explained
in Section4.2.
57
0
-20
Vc (Volt)
-40 Detailed Model
Actual Average
-60
-80
40
30
(Volt)
20
C ss
V
10
Actual Average and
State-Space Averaging
0
3
Detailed Model
Actual Average
2 State-Space Averaging
Ip (Volt)
-1
0.4
0.2
I s (Volt)
-0.2
-0.4
200
State-Space Averaging
100
C ds
V
-100
0.071399 0.071401 0.071403 0.071405 0.071407
Time (s)
58
4.2 State-Space Averaging in DCM
The DCM operation of PWM converters differs from CCM operation by an additional time
interval in each switching cycle during which the inductor current or capacitor voltage is
clamped to zero. In Flyback converters, there is an exception in this case if there is a diode
snubber in the circuit. To demonstrate this point, a detailed model of the converter circuit
depicted in Figure 4.1 has been implemented using PLECS [16]. The corresponding
parameters are summarized in Appendix A.7. The state matrices Ak , Bk , Ck , Dk have been
extracted for each subinterval in DCM. Finally, the state-space averaged model has been
implemented using these matrices and the given duty ratio. The resulting simulated state
variables are shown in Figure 4.3.
59
0
Vc (Volt)
-50 Detailed Model Detailed Model
Actual Average Actual Average
State-Space Averaging
-100
40
VCss (Volt)
20
1.5
1
Ip (Amp.)
0.5
-0.5
0.2
see Fig 4.4
Is (Amp.)
Ts
200
100
C ds
V
-100
0.099959 0.099961 0.099963 0.099965 0.099967
Time (s)
Figure 4.3 Simulated state variables of the full-order Flyback converter in DCM.
60
0.08
Is (Amp.)
0.04
0
d3Ts
-0.04
0.099963 0.099964 0.099965
Time (s)
Figure 4.4 Simulated transformer secondary current of the full-order Flyback converter in DCM.
During the first interval ( d1Ts ) , when the MOSFET is on and the diode is off, the primary
side of the transformer stores the energy in the field. Right after the switch change its state,
the stored energy on primary side is transferred to the secondary side, and this energy is spent
during the second interval ( d 2Ts ) . In CCM, the switches go back to their original position
before this energy being spent. But in DCM this energy is spent before the switches change
their positions. That is where there is an exception. If there is no snubber circuit, after the
energy is spent, the secondary current will stay at zero until switches go back to their original
state. If there is a snubber circuit, after the energy is spent, the energy stored in the diode
snubber capacitor will flow in the other direction until it reaches zero or switches go back
their original state. This special case makes it difficult to identify d 2 and d3 in the model
with the presence of parasitics, which will be explained is Section 4.3.1.
As can be seen in Figure 4.3, the state-space averaging predicts the switch snubber capacitor
voltage (VC ss
) , and the diode snubber capacitor voltage (V )Cds correctly as expected because
these are continuous state variables. However, the state-space averaging predicts the output
filter capacitor voltage (Vc ) with a large error. The reason has already been explained in
Section 3.2. Also the transformer primary ( I p ) and secondary current ( I s ) are predicted with
an error. As explained in Section 4.1, these currents are discontinuous in either CCM or
DCM. As a result, the conventional state-space averaging method also does not produce the
correct results in DCM.
61
4.3 Parametric Average Value Modeling in CCM and DCM
In this Section, the parametric average-value modeling is extended to the full-order converter
operation in CCM and DCM, which to the best of our knowledge has not been done for the
transformer isolated topologies.
M ( d1 , x ) , the detailed model has been run in the operation region of interest (for example;
duty-cycle changes between 0.1 and 0.9, and the load ( R ) changes from very low load to
very high load) whereas the state variables are averaged numerically over the prototyping
switching interval. In particular, the average-value of the state vector ( x ) is computed in a
steady-state corresponding to given operation point. Specifically, in the steady-state from
equation(2.168) an intermediate vector p is computed using (2.169). Thereafter, the
elements of M are found using (2.170).
To obtain the functions d3 ( d1 , x ) and M ( d1 , x ) for the desired operation range, the detailed
simulation is run with different values of control variable ( d1 ) as well as the load resistance
( R ) . The variables resulting from this procedure are d1 , d 2 , d 3 , R, id , vc , vCss , ip , is , vCds . Then, the
correction vector ( M ) is computed using (2.169) and (2.170). These variables are stored for
future use in lookup tables. The real challenge here is to calculate the d 2 or d 3 at any given
operation point. As can be seen in Figure 4.3 and Figure 4.4, there is no easy way to read the
intervals from the state variables' waveforms because of the ringing waveforms caused by
snubbers. To achieve that, the diode current is used, as seen in Figure 4.5. As can be seen in
62
Figure 4.5 , the diode current is zero during the first and third subinterval. Since we know the
first interval, it becomes easy to calculate the third and then second subintervals.
Ts
-0.04
I (Amp.)
d
-0.08
-0.12
-0.16
0.105993 0.105995 0.105997 0.105999
Time (s)
The final numerical function for d 3 is plotted in Figure 4.6. The variable d 3 has a flat surface
corresponding to CCM and varies linearly along the d1 . In DCM, the surface of d 3 becomes
non-linear and increases.
As explained in Section 3.2, the filter capacitor voltage (Vc ) has a large error even though it
is a continuous voltage. To fix this problem a model proposed in Section 3.2 (see Figure 3.8)
has been used here. If the secondary current is corrected, the diode current will be corrected
as well, and as a result the output capacitor voltage and the output voltage will be fixed
without a correction term. In the PAVM, the correction term m1 of filter capacitor voltage
63
0.5
DCM
0.4
0.3
CCM
d3
0.2
0.1 3000
2000
1000 R (ohm)
0
1 0.9 0
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
d
1
Figure 4.6 Variable d 3 as a function of the duty-cycle ( d1 ) and the load resistance ( R ) .
The correction term ( m2 ) of the primary side snubber capacitor voltage (VC ss
) is plotted in
Figure 4.7 as a function of the duty-cycle ( d1 ) and the load resistance ( R ) . The correction
term ( m3 ) of the primary current ( I p ) is plotted in Figure 4.8 as a function of the duty-cycle
( d1 ) and the load resistance ( R ) . The correction term of the primary current has no flat
surface because some correction is needed in CCM and DCM as explained in Section 4.1.
The correction term ( m4 ) of the secondary current ( I s ) is plotted in Figure 4.9 as a function
of the duty-cycle ( d1 ) and the load resistance ( R ) . Similar to correction term the secondary
current ( m3 ) , the current ( I s ) is always discontinuous and needs correction in CCM and
DCM.
64
1.04 CCM and DCM
1.02
2
1
m
0.98
3000
2000
0.96 1000 R (ohm)
0.8 0.7 0.6 0.5 0.4 0
0.3 0.2 0.1
d1
Figure 4.7 The correction term m2 as a function of duty-cycle ( d1 ) and the load resistance ( R ) .
14
12
CCM and DCM
10
8
m3
2 3000
2000
0 1000
0.8 0 R (ohm)
0.7 0.6 0.5 0.4 0.3 0.2 0.1
d1
Figure 4.8 The correction term m3 as a function of the duty-cycle ( d1 ) and the load resistance ( R ) .
65
1.4
1.2
0.8
4
m
0.6
0.4
3000
0.2 2000
1000
0 R (ohm)
0.9 0.8 0
1 0.7 0.6 0.5 0.4 0.3 0.2 0.1
d1
Figure 4.9 The correction term m4 as a function of duty-cycle ( d1 ) and the load resistance ( R ) .
4.10. This correction term has a problem. Since the diode’s snubber capacitor voltage (VC ds
)
is a continuous state variable, the correction term must be equal to 1. In theory, the average
value of the diode snubber capacitor voltage is zero because it stores energy during one
interval and discharge this energy during another interval. As a result, the average value is
should be equal to zero. But when the value of p calculated using (2.170), it doesn’t give
exactly zero but gives very small numbers such as 2 ⋅ 10−13 due to the numerical precision of
calculations. Then this is used to calculate the correction term ( m5 ) . In theory, vC must be ds
equal to zero but in the detailed simulation again we get very small numbers because of the
ringing waveforms. Since the capacitor voltage is a continuous state variable, in the model
implementation the correction term ( m5 ) is set to 1.
66
-5
x10
2 CCM and DCM
1.5
0.5
5
0
m
-0.5
-1
3000
-1.5
2000
1000
-2 R (ohm)
0
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
d1
Figure 4.10 The correction term m5 as a function of duty-cycle ( d1 ) and the load resistance ( R ) .
Once the functions d3 ( d1 , x ) and M ( d1 , x ) are available and stored, the parametric average-
As can be seen in Figure 4.11, the developed PAVM predicts the average-value of the
waveforms well in CCM as compared to conventional state-space averaging which does not.
67
There is a difference between measured output voltage and detailed model’s output voltage.
The reason is that the detailed model doesn’t have all the parameters the hardware model has
such as stray capacitances.
-71.4
Detailed Model
Hardware Prototype
-71.5 Actual Average
V (Volt)
Detailed Model
PAVM
-71.6
-71.7
Actual Average
and PAVM
2
Actual Average
and PAVM
I p (Amp.)
Detailed Model
State-Space Averaging
0.4 Actual Average
PAVM
0.2
Actual Average
and PAVM
Is (Amp.)
-0.2
-0.4
0.009992 0.009993 0.009994 0.009995 0.009996
Time (s)
Figure 4.11 Measured and simulated output voltage, primary and secondary current in CCM at constant
duty-cycle.
The accuracy of the PAVM in predicting the large-signal behaviour in time-domain has also
been verified by studying the effect of sudden change in load. In the following study, the
output of the flyback converter was regulated using the same PI controller designed to
regulate the output voltage at −72V . In the study being considered, the converter initially
operate in DCM with a 2000Ω load. At t = 0.01s , a parallel load of 500Ω is added which
changes the mode to CCM. The resulting time-domain transients are shown in Figure 4.12.
After load changes the control action brings the output voltage to the desired −72V . Through
68
all transients, the large signal behaviour of the detailed model is accurately predicted by
developed PAVM.
-71.7
Detailed Model
PAVM
-71.8
V (Volt)
-71.9
-72
4 Detailed Model
3
2
Ip (Amp.)
-1
PAVM
-2
0.5
0
Is (Amp.)
-0.5
-1
0.0098 0.0102 0.0106 0.011 0.0114
Time (s)
Figure 4.12 Simulated output voltage, primary and secondary current during the transient from DCM to
CCM due to the step change in load.
69
60
40
Amplitude (dB)
20
0
PAVM
-20
Detailed
Model
-40
160
140
120
100
Phase ( ° )
80
PAVM
60
40
20 Detailed
0 2 Model
3 4 5
10 10 10 10
Frequency (Hz)
Figure 4.13 Control-to-output transfer function of the full-order Flyback converter evaluated at
R = 717.05Ω and d1 = 0.381 .
70
Table 4.1 Efficiency comparison of the average-value models
Efficiency
Model Type
(%)
PAVM 98.11
cannot
5nd Full-order State-space Averaging
predict
PAVM 85.41
As can be seen in Table 4.1, the ability of the model to account for the losses and predict the
efficiency improves with increasing the model order from 2 to 5, wherein the traditional
second-order models are significantly off and over estimate the efficiency to the 98%.
Moreover, the classical state-space averaging cannot be simply extended to the 5th order
model and therefore cannot be used for calculating the efficiency correctly. At the same time,
the proposed PAVM of full-order model predicts the efficiency with only a small error
compared to the hardware prototype and its detailed switching model from which it was
established.
71
Chapter 5 : Conclusion
In this Thesis, the recently established parametric average-value modeling methodology has
been extended to the transformer-isolated Flyback converter topology which includes
parasitics and snubbers. It is shown that the developed model captures includes the effect
parasitics and losses, and is therefore capable of accurately predicting the terminal
characteristics of the converter such as efficiency with the accuracy that has not been
attainable by any previously established average-value models. The developed method
overcomes the complexity and challenges common to many previously developed models
when the parasitics of the circuit elements and snubbers are considered. The numerically
constructed model can function in both CCM and DCM. It has been shown that obtaining an
accurate full order average-value model requires extracting the duty-ratio constraint and the
correction term. The functions of the duty-ratio constraint and correction terms were obtained
numerically by running the detailed simulation at desired operation range. Once the model
established, the resulting model is continuous and valid for large-signal time-domain
transient studies as well as for linearization and subsequent small-signal characterization of
the overall system over a wide range.
It has also been shown that direct application of conventional state-space averaging method
is no longer accurate for flyback converters in CCM when the transformer leakage
inductances are taken into account. A detailed analysis of flyback converters working in
CCM has been done to present the problems in CCM. Regardless of the converter operating
mode (CCM or DCM) the primary and secondary currents are discontinuous. As a result, the
conventional assumptions used for DC-DC converters are no longer acceptable for flyback
converters.
There are other kinds of snubber circuits in the practical applications such as RCD
(resistance, capacitance and diode) and active snubbers. They also could be included in
PAVM.
73
Bibliography
[1] K.-H. Liu and F. Lee, “Topological constraints on basic pwm converters,” in Power
Electronics Specialists Conference, 1988. PESC ’88 Record., 19th Annual IEEE.
IEEE, april 1988, pp. 164 –172 vol.1.
[2] D. Maksimovic, “Synthesis of pwm and quasi-resonant dc-to-dc power converters,”
Ph.D. dissertation, 1989.
[3] D. Maksimovic and S. Cuk, “A unified analysis of pwm converters in discontinuous
modes,” Power Electronics, IEEE Transactions on, vol. 6, no. 3, pp. 476 –490, jul
1991.
[4] J. Sun, D. Mitchell, M. Greuel, P. Krein, and R. Bass, “Averaged modeling of pwm
converters operating in discontinuous conduction mode,” Power Electronics, IEEE
Transactions on, vol. 16, no. 4, pp. 482 –492, Jul. 2001.
[5] D. Maksimovic and S. Cuk, “General properties and synthesis of pwm dc-to-dc
converters,” in Power Electronics Specialists Conference, 1989. PESC ’89 Record.,
20th Annual IEEE. IEEE, jun 1989, pp. 515 –525 vol.2.
[6] P. Krein and R. Bass, “Geometric formulation, classification and methods for power
electronic systems,” in Power Electronics Specialists Conference, 1990. PESC ’90
Record., 21st Annual IEEE. IEEE, jun 1990, pp. 499 –505.
[7] R. Tymerski and V. Vorperian, “Generation, classification and analysis of switched-
mode dc-to-dc converters by the use of converter cells,” in Telecommunications
Energy Conference, 1986. INTELEC ’86. International. IEEE, oct. 1986, pp. 181 –
195.
[8] J. Sun and H. Grotstollen, “Averaged modelling of switching power converters:
reformulation and theoretical basis,” pp. 1165 –1172 vol.2, Jul. 1992.
[9] A. Pressman, K. Billings, and T. Morey, Switching power supply design. McGraw-
Hill Professional, 2009.
[10] R. Erickson and D. Maksimovic, Fundamentals of power electronics. Springer
Netherlands, 2001.
[11] S. Ang and A. Oliva, Power-switching converters. CRC, 2005, vol. 123.
74
[12] M. Kazimierczuk and S. Nguyen, “Small-signal analysis of open-loop pwm flyback
dc-dc converter for ccm,” in Aerospace and Electronics Conference, 1995. NAECON
1995., Proceedings of the IEEE 1995 National, vol. 1. IEEE, may 1995, pp. 69 –76
vol.1.
[13] H. Kewei, L. Jie, F. Ningjun, L. Yuebin, H. Xiaolin, and W. Luo, “Modeling analysis
and simulation of high-voltage flyback dc-dc converter,” in Industrial Electronics,
2009. ISIE 2009. IEEE International Symposium on. IEEE, july 2009, pp. 813 –818.
[14] SimPowerSystems User’s Guide Version 5, Hydro-Quebec, The Mathworks Inc.,
2010. [Online]. Available: www.mathworks.com
[15] Automated State Model Generator (ASMG), PC Krause and Associates Inc., 2002.
[Online]. Available: www.pcka.com
[16] Piece-wise Linear Electrical Circuit Simulation for Simulink (PLECS), Plexim
GmbH, 2009. [Online]. Available: www.plexim.com
[17] D. Czarkowski and M. Kazimierczuk, “Linear circuit models of pwm flyback and
buck/boost converters,” Circuits and Systems I: Fundamental Theory and
Applications, IEEE Transactions on, vol. 39, no. 8, pp. 688 –693, aug 1992.
[18] R. Middlebrook and S. Cúk, “A general unified approach to modelling switching-
converter power stages,” International Journal of Electronics, vol. 42, no. 6, pp. 521–
550, 1977.
[19] P. Chetty, “Current injected equivalent circuit approach to modeling switching dc-dc
converters,” Aerospace and Electronic Systems, IEEE Transactions on, vol. AES-17,
no. 6, pp. 802 –808, nov. 1981.
[20] S. Cúk and R. Middlebrook, “A general unified approach to modeling switching dc-
to-dc converters in discontinuous conduction mode,” in in Proc. IEEE PESC’77,
1977, pp. 36–57.
[21] V. Vorperian, “Simplified analysis of pwm converters using the model of the pwm
switch parts i and 11,” IEEE Trans. on Aerospace and Electronic Systems, vol. 26,
no. 3, pp. 490–505, 1990.
[22] J. Jatskevich, S. Pekarek, and A. Davoudi, “Parametric average-value model of
synchronous machine-rectifier systems,” Energy Conversion, IEEE Transactions on,
vol. 21, no. 1, pp. 9 – 18, 2006.
75
[23] ——, “A fast procedure for constructing an accurate dynamic average-value model of
synchronous machine-rectifier systems,” in Power Engineering Society General
Meeting, 2006. IEEE, 0-0 2006, p. 1 pp.
[24] A. Davoudi, J. Jatskevich, and T. De Rybel, “Numerical state-space average-value
modeling of pwm dc-dc converters operating in dcm and ccm,” Power Electronics,
IEEE Transactions on, vol. 21, no. 4, pp. 1003 – 1012, 2006.
[25] A. Davoudi and J. Jatskevich, “Realization of parasitics in state-space average-value
modeling of pwm dc-dc converters,” Power Electronics, IEEE Transactions on,
vol. 21, no. 4, pp. 1142 – 1147, 2006.
[26] A. Davoudi, J. Jatskevich, and P. Chapman, “Computer-aided average-value
modeling of fourth-order pwm dc-dc converters,” pp. 793 –796, May 2007.
[27] ——, “Computer-aided dynamic characterization of fourth-order pwm dc-dc
converters,” Circuits and Systems II: Express Briefs, IEEE Transactions on, vol. 55,
no. 10, pp. 1021 –1025, 2008.
[28] J. Sun, “Unified averaged switch models for stability analysis of large distributed
power systems,” in Applied Power Electronics Conference and Exposition, 2000.
APEC 2000. Fifteenth Annual IEEE, vol. 1, 2000, pp. 249 –255 vol.1.
[29] G. Wester and R. Middlebrook, “Low-frequency characterization of switched dc-dc
converters,” Aerospace and Electronic Systems, IEEE Transactions on, vol. AES-9,
no. 3, pp. 376 –385, may 1973.
[30] Matlab/Simulink 7 User’s Guide, The Mathworks Inc., 2010. [Online]. Available:
www.mathworks.com
[31] S. Howimanporn and C. Bunlaksananusorn, “Performance comparison of continuous
conduction mode (ccm) and discontinuous conduction mode (dcm) flyback
converters,” in Power Electronics and Drive Systems, 2003. PEDS 2003. The Fifth
International Conference on, vol. 2, nov. 2003, pp. 1434 – 1438 Vol.2.
[32] D. Hart, Introduction to power electronics. Prentice Hall PTR Upper Saddle River,
NJ, USA, 1996.
[33] J. Mahdavi, A. Emaadi, M. Bellar, and M. Ehsani, “Analysis of power electronic
converters using the generalized state-space averaging approach,” Circuits and
76
Systems I: Fundamental Theory and Applications, IEEE Transactions on, vol. 44,
no. 8, pp. 767 –770, Aug. 1997.
[34] R. Brockett and J. Wood, “Electrical networks containing controlled switches,” In its
Appl. of System Theory to Power Process. Probl. p 57-79 (SEE, vol. 75, 1974.
[35] C. Meyer, Matrix analysis and applied linear algebra and solutions manual. Society
for Industrial and Applied Mathematics, 2000.
[36] M. Spiegel, Schaum’s outline of theory and problems of advanced mathematics for
engineers and scientists. Schaum’s Outline Series, 1971.
[37] N. Mohan, T. Undeland, and W. Robbins, Power electronics. Wiley New York, 1995.
[38] J. Kassakian, M. Schlecht, and G. Verghese, Principles of power electronics.
Addison-Wesley Reading, MA, 1991.
[39] R. Ridley, “Flyback converter snubber design,” Switching Power Magazine, 2005.
[Online]. Available: http://www.ridleyengineering.com
[40] A. Elasser and D. Torrey, “Soft switching active snubbers for dc/dc converters,”
Power Electronics, IEEE Transactions on, vol. 11, no. 5, pp. 710 –722, Sep. 1996.
[41] L. O. Chua and P. Y. Lin, Computer-Aided Analysis of Electronic Circuits:
Algorithms and Computational Techniques. Prentice Hall Professional Technical
Reference, 1975.
77
Appendices
78
A.4 Second-order Flyback Converter Parameters with Parasitics in DCM
vg = 20.009V , f s = 250kHz , d1 = 0.381, d 2 = 0.4409
Lm = 27 µ H , n = 1 / 6, C = 22 µ F , R = 2500Ω
Rsw = 0.04Ω, Vd = 1.25V , Rc = 0.09Ω
79
A.7 Full-order Flyback Converter Parameters in DCM
vg = 20.009V
Mosfet : 3.8 A,55V , N − chanel , International Re ctifier IRLL 2705
Rsw = 0.04Ω, f s = 250kHz
d1 = 0.381
Css = 470 pF , Rss = 10Ω
Transformer : ICE Components ICA − 0635
Lm = 27 µ H , Lp = 0.2µ H , Ls = 0.8µ H
R p = 210mΩ, Rs = 1.35Ω
n =1/ 6
Diode :1A, 400V , ultra − fast re cov ery diode, Central Semiconductor Corp CMR1U − 04
Vd = 1.25V
Cds = 100 pF , Rds = 200Ω
C = 22µ F ,100V , alu min um electrolytic capacitor , Sanyo100MV 22 AX , Rc = 0.09Ω
R = 2500Ω
80
Appendix B. Flyback Converter Circuit Diagram
81