A GENERAL UNIFIED APPROACH

TO MODELLING SWITCHING DC-TO-DC CONVERTERS

IN DISCONTINUOUS CONDUCTION MODE
I
SLOBODAN CUK and R. D. MIDDLEBROOK

California Institute of Technology

Pasadena, California

ABSTRACT a) b)
buck power stage:
A method 60~ modelling ~w~ehlng eonve4t~ ~n
the dWe.onUnu.o~ conducxion. mode ~ developed, who~e
L v L v
~~aAting po~nt ~ the uni6~ed ~tate-~pae.e ~e~e6en­
t.aA:,.i,on, and who~e end ~uul:t ~ a e.omplete UneM Vg
ciAeuU model ~c.h e.oJtlL.ec:Uy ~ep~uen.t6 aU u~ent<.ai. c R R
6edtUILu, namehj, the ~nput, output, and tJuur..66e~
p~opVLtiu (~t.aA:,.i,e. de. a.6 well a.6 dyna.rni,e. ae. ~ma.U
.6.ignal) • Whi1,e the method .i6 genVLall.y appUc.a.bte to
any ~wUdU..ng e.onveJLt~ opvr.a..ting in the diAe.onilnuoUh boost power stage:
e.onduction mode, U cs ex.te~~vehj .iU.UhtJta:ted 60IL the L V L V
thJLee e.ommon powe~ ~tagu (buck, ooos t; and buc.k-boo.6t) .
The ILuu.U.6 6o~ the.6e e.onveJLteJL6 Me then eOAily tabu-
lated owing t» the ~xed equ<.val~ ci.JLc.ui.t t:opology V
g
~s-c~----' R V,,! R
06 thei.Jr. ClU10n.ie.a.l cilLc.ui.t model.
The ou.t.ti..ned method l.end6 -i.tA el6 ea.6ily to ~nve6t,i­
ga:tion 06 the e:.tL6 e.on.ti..nuOU6 e.onduction mode in mo~e
complex .6~c.:tuIr.e.6 (e.a.6 cade. e.onnee:Uon 06 buck and ooost: buck - boost power stage:
c.onveJLtelL6, 60~ example), in w~c.h mo~e than one ~nducto~
c.ulVLe~ may bee.ome di..6 e.On.U.nUOUh •
S -V -v
M oppo-6ed to otheJL modelling tec.hn.ique6, the
new method eons-eden» the cU...6 eon.t(.nuoUh e.onduction mode c R~ R
a.6 a .6peci..a.l c.a.6e 06 the e.On.U.nUOU6 e.onduc.t.ion mode~.

1 INTRODUCTION F.i.g. 1. Th~ee e.ommon ~w.U:cJU.ng de.-to-de. e.onve-'LteJL6:

a) topolog~cal e.on6igu.JLation ~dependent 06
Switching-mode dc-to-dc converters afford an ~wilc.h JLeaUzatiOYLj b) b~polaJL tJuur..6~tOJL
efficient means of transforming power at one dc voltage .implementati..onOn the ~wUc.h S.
to another. There are many circuit configurations
capable of performing dc-to-dc conversion, of which
the most common are the buck, boost, and buck-boost Consider, for example, the buck-boost converter of Fig. 1.
converters shown in Fig. 1. In each converter, the If the energy stored in the inductor during the first
basic dc-to-dc conversion function is achieved by con- interval DT = D T is completely released to the out-
trol of the switch fractional closed-time (transistor put load be~oJLe th~ switching cycle T has ended, the
on-time), or duty ratio D (0 < D < 1) with constant inductor current becomes zero for theSlast portion
switching frequency f ~ lIT , where T is the switching D
3Ts'
as seen in Fig. 2b.
peri.od. s s s
0) 6)
Two modes of switching converter operation may be inductor current ilt) inductor current i tt ]
distinguished: the e.OJuUnUOU6 e.onduction mode (inductor
current never falls to zero, as in Fig. 2a), and the VglL slope VI L
di.6e.ontinuoU6 e.onduction mode (inductor current becomes " / I
zero for a portion of switching period, as in Fig. 2b).

This work was, supported in part by Subcontract No.

A72042-RHBE form TRW Systems Group under NASA Prime
Contract NAS3-l9690, by Subcontract No. D04803-CFCM from
TRW Systems Group under NASA Prime Contract NAS3-20102,
by NASA through the Jet Propulsion Laboratory of the F~g. 2. 1ndu.ctOJL c.uNLent wave6oJun6 and de6.inli:i..on 06
California Institute of Technology, and by the Naval two conduction mode.6: a.) c.on.UnuoU6 e.on-
Ocean Systems Center through MIPR No. N009537nW090l8. ducxiov: roodej b) cU6eontinuoU6 conduction mode.

36-PESC 77 RECORD
 approaches ] 0 l [ - ] 4 [ ( ) have been proposed. However,
Thus the transition from continuous to discontin- while all these techniques J 0 l [ - ] 4 [ ( ) provide through
uous conduction mode is obtained by either incAJease of various linearization procedures the proper linearized
load R (hence by lowering of the average dc current I) transfer functions (duty ratio modulation d to output
or by decACOSe of inductance L or switching frequency f . g voltage ν and line voltage ν to output voltage ν trans­
In any case, however, the operation in the discontin- fer functions), they are inclpable of representing the
uous conduction mode results in tha.ee different switched input properties of the converter, and hence fail to
networks, as illustrated in Fig. 3 for the buck-boost arrive at the complete linearized converter model. This
converter (as opposed to two switched networks for con- is an entirely analogous situation to that for contin­
tinuous conduction operation). An analogous situation uous conduction mode [ 2 , 3 ] , where these methods could
exists for the other two converters of Fig. 1 as well not model the input properties (open- and closed-loop
as for a number of other switching converters. input impedance, for example) of the converters and
regulators in continuous conduction mode of operation.
a ) INTERVAL DTS- B ) INTERVAL D^TS · C ) INTERVAL D$Ts'- In addition, they stay throughout modelling in the do­
main of equation manipulations only, and thus the use­
ful insight which can be gained from lineal circuit
models (as demonstrated in [ l , 2 , 3 ] is lost. Hence the
primary objective of the development here becomes to
overcome all these difficulties by extending the power­
ful state-space averaging technique of [ 2 ] , together
with its circuit model realizations, to the discontin­
fig. 3. ThAee smXcked neJwonks ioK the buck-boost uous conduction mode of converter operation and finally
conveAteA operating Jjn the discontinuous con- to arrive at the complete linear ciAcuit model of various
duction mode,: a) tna.nsistoK on, diode o^; converters (like, for example, those of Fig. 1 ) .
b) VuansistoK ou, diode on; c) tAansistoK
ou, diode oU* 2.2 New state-space and circuit averaging methods for
switching converters in the discontinuous conduction
In Section 2 an extensive overview of the complete mode
structure of modelling of switching converters and regu-
lators in the discontinuous conduction mode by use of The state-space and circuit averaging methods pre­
the new method is provided. In particular, the steps sented in [ 2 ] are now to be suitably modified to account
leading to the equivalent circuit models that describe for the discontinuous conduction mode of operation, and
both steady-state (dc) and dynamic (ac small signal) the results are summarized in the Flowchart of Fig. 4.
behaviour are briefly explained. The subsequent sections As before for the continuous conduction mode, the star­
then give a detailed and thorough account of the new ting model for the switching converter (block 1 in the
method outlined in Section 2. Flowchart of Fig. 4) is either in terms of the State-
Space description of the switched networks (as in block
First, in Section 3, the procedure for modelling la) , or in terms of lineaA ciAcuit models of the switched
in discontinuous conduction mode is viewed as a special networks (as in block lb).
case of that for continuous conduction mode [l,2,3]
(provided the state-space averaging step of [2] is The difference, however, from the previous descrip­
properly generalized to include three or more structural tion is not only that now there are thAee different
changes within each switching period as shown in Appendix) structural configurations within each switching period,
and additional constraints imposed to model special but also in the fact that instantaneous inductor cur­
inductor current behaviour. Though the results obtained rent is KeStAJLcted in its behavior: it starts at zero
are in terms of linear equations, the useful circuit at the beginning of a switching period and falls to
realizations may be obtained as in Section 4 . The zero current'again even before the switching period
straightforward perturbation and linearization steps has expired (see the instantaneous inductor current
in Section 5 lead to dc and ac ciAcuit models. They waveform in block 1 of Fig. 4).
result for three common converters of Fig. 1 in the
fixed topology, canonical ciAcuit model and are easily It is actually this second difference which clearly
tabulated. Because of the need for complete presen- distinguishes the discontinuous conduction mode of
tation of the theoretical background of the new method, operation, while the first difference, that of having
and lack of space, only cursory experimental verifi- three different structural configurations, appears in
cation is included at the end of Section 6 . Finally, a way to be merely incidental. That is, in Appendix A
in Section 7 the completeness of the obtained converter it is shown that the state-space averaging step of [ 2 ]
circuit models is reemphasized by their direct incor- can be directly extended to include "three-state" con­
poration in S^ûXching regulator models. verters (converters with three structural changes within
each switching period), provided such converters are
Since the method presented here is essentially a operated in the continuous conduction mode, and any
consistent extension of the technique for continuous restrictions on state-space variables (inductor currents
conduction mode [2], the exposition will closely fol- and capacitor voltages) are avoided; Therefore, our
low the format given in [2], such that the common steps objective in modelling converters operating in the dis­
to both methods become immediately transparent, and continuous conduction mode (and exhibiting "three-state"
those that are different clearly distinguished. configuration behavior) becomes that of supplementing
this generalized state-space averaging step for "three-
state" converters by additional constraints which re­
flect the special behavior of one of the state variables,
2 REVIEW OF THE NEW STATE-SPACE MODELLING TECHNIQUE the inductor current. Hence the switching-mode con­
IN THE DISCONTINUOUS CONDUCTION MODE verter operating in the discontinuous conduction mode
(and having three structural changes) may be viewed as
2 . 1 Brief review of existing modelling techniques a Special case of the ordinary "three-state" converters
which are free from any restrictions on state variables.
Owing to the relatively more complicated nature of Thus the primary goal is properly to determine these
the converter operation in the discontinuous conduction additional constraints and to find how they propagate
mode, dynamic (ac small signal) models have been lacking through various paths of the modelling (such as paths
(even though valid models for continuous conduction
mode have already been obtained) until recently several a and b on the Flowchart of Fig. 4).

P E S C 77 R E C O R D - 3 7
 From the Flowchart of Fig. 4 it is immediately clear along path α the general equations (through general
that path a follows a development strictly in terms of matrices A,, A , A 0 0 and vectors b,, b f t > and b ) are
state-space equations, the state-space averaged model­ 2ze the fact that the
retained to emphasize tfie outlined pro­
ling technique, while the other path b proceeds in cedure is applicable to any "three-state" converter
terms of circuit models, circuit averaged modelling. operating in the discontinuous conduction mode, while
Moreover, as before for the continuous conduction mode, along path b a particular example of the boost con-

stale-space equations I ^ J final stole-space averaged model :

perturbed with:
SLEADY STATE (DC) MODEL:

dr-D^d, , d =tvi , x-X + î, 2

X=-Ä'bV, with I-t(V,y,0,LTs)
v s = V s ^ , D,*D , d / . d
DYNAMIC ( A C ) SMALT SIGNAL MODEL:
and constroints perturbed:
i»Aä*bvfr • a , [ ( A - A i X ^ ( U - b ) V >
1 a f

l-o
WITH: £ S Q

SLATE-'.PACE INSCRIPTION OF t h r e e PERLURBATION

SYITCHED networks
2o
basic slale-space a v e r a g e d model:
X-A,X+B,V ; 9 interval T D ,
5

X=A X + bv 5

X-AjX+b^Vj* interval T d s 2

where. A » D,A, • d ^ + d ^ A j siüte-spoce

all transfer
X - A ,X » U^s ; interval \d à
htd,b, + dibi • d j b i
realization functions
and additional constraints: und a d d i t i o n a l constraints:

iiOhiliX * à,) T > 0 , t = i ( v ,

s dX J ) 5

where x - ( i v ) T

state - s p a c e ^ j j * modeling hybrid ^ j , modehnq canonical circuit" model

switching dc-io-dc convener in 2c steady s t a t e (dc) m o d e l :

discontinuous conduction mode ι : M(DJ0_
stole - space equations V

in a circuit recognizable

dynamic (or small-signal) model:

form

(stale - s p a c e realization)

iboost example)

circuit ^ modeling circuit ,Π, recognition

Li
three s w i t c h e d circuit models: 2b 7V
circuit averaged model :
interval T d, : 4 interval T d : s 3

from constraint : $f " 0

circuit
all transfer
internal
equivalent
functions
|:(^(4) d.: I transformation
additional requirement;

and a d d i t i o n a l constraints? i =
2L » d T

( boost example)
ifoM[(d.*<toT;>0 . i-^clTÎ
(boost e x a m p l e ) perturbation ^ j , 4b final a v e r a g e d circuit models
t e a d y - a f a t e ( d c ) model
3b
circuit model perturbed :
with: d-D*ei , ο > 0 ^ ϊ , i-I*î

I : ft: I
dynamic ( a c small-signal) model *

to τ

and contirainX perturbed:

(beut eiomple")

Ftcj. 4, F^owJc/iaVLt Ojj ave^gXng appKQOLch<it> in moddLZLng milking dc-to- dc. ζοηνζηϊο,ηΔ in tine, dUcontcnuoLU>
conduction mode. Path a: gm&iaJL àtcûz-^pacz wnfejUAng; Pcutk b.* cJjicuJjt Vianù^onmation method.

3 8 - P E S C 77 R E C O R D
verter is followed, owing to the requirement for the Finally, both models (block 4a or 4b) can be used
specific converter topology along that path. Speci­ to determine the transfer functions of interest: line
fically, for the boost power stage, - A« φ A« are voltage variation ν and duty ratio modulation d to
2 x 2 matrices, and b^ * b^ ^ 0, b^ « 0 are vectors. output voltage ν (blocks 6a and 6b respectively).
This example will later be pursued in detail along
both paths. 2.3 New canonical circuit model for discontinuous
conduction mode
We now follow path α more closely. The crucial
step is made in going from block la to 2a in that the As for the continuous conduction mode, the cul­
original description through three state-space equations mination of the modelling is again a canonical circuit
(block la) is substituted by a single state-space model (block 5 of Fig. 4 ) , whose fixed topology (though
averaged model (block 2a). This is justified as fol­ different from the one for continuous conduction mode)
lows. The fundamental performance requirement of has all the features necessary to present a COmpteJx,
switching converters (negligible switching ripple) CÜLCuit modal. However, this fixed topology of the
results in natural frequencies ω and f much lower model for discontinuous conduction mode came merely
than the switching frequency f . This, in turn, leads
ß
as a by-product, since for the three converters of
as shown in Appendix A to the generalized state-space Fig. 1 (buck, boost, and buck-boost) the ac small-sig­
averaging step. So far this would be the same aver­ nal models all resulted in the fixed topological struc­
aging step as applied to any ordinary "three-state" ture of the model in block 4b of Fig. 4 without any
switching converter. However, as indicated before, need for equivalent circuit or other transformations.
the inductor current i do&6 not behave as a true state- It does not appear that this canonical circuit topology
space variable in the discontinuous conduction mode could be directly extended to some arbitrary converter.
since it does not have free boundary conditions (but Even though this canonical circuit model is not so
fixed at zero) which is shown to lead to the following general as that for two-state converters [2], a use­
constraint: ful comparison between the two canonical circuit topo­
logies can be made (at least for the common converters
of Fig. 1 in both operating modes).

This immediately reduces by one the order of the basic While in the continuous conduction mode the effect
state-space averaged model (block 2a), since one of of duty ratio modulation d was represented by voZtagt
the dynamic equations (that for inductor current) and CUAAtnt duty ratio dependent generators at the
reduces to a static equation. In addition to this, input port (hence properly representing negative closed-
an expression describing the average inductor cur­ loop input impedance at low frequencies as shown in [2],
rent i can be found directly from the converter it­ here in discontinuous conduction mode there are two
self (block 1) and becomes the second constraint, duty ratio dependent ΟΛΛΛ&ηΧ. generators, one in the in­
termed perturbation equation I, which is put circuit (again, properly to model converter input
properties as shown later in Section 7), and the other
i = i(v , v, d, L, Τ )
a (2) in the output circuit to generate the duty ratio â to
g b

output transfer function.

Thus, the two additional constraints (1) and (2), The salient feature of the canonical circuit model
together with the generalized state-space averaging in block 5 of the Flowchart in Fig. 4 is that both trans­
step, completely determine the converter model in the fer functions are obtained using only the output port of
discontinuous conduction mode. It remains only to the complete canonical circuit model, unlike the situ­
apply the standard perturbation techniques (block 3a) ation for continuous conduction mode where the complete
and (on the basis of the small-signal assumption) the circuit model was necessary to determine them. This is
linearization techniques to both state-space averaged also why other methods which properly represent the trans­
equations and the perturbation equation of block 2a fer function in discontinuous conduction mode ([4]-[l0])
in order to arrive at the final state-space averaged have completely omitted modelling of the converter in­
model (block 4a). This model gives separately both put properties.
dc and ac small-signal descriptions through general
matrices A^, A«, A_ and vectors b^, b2» b^ of the 2.4 Extension to complete regulator treatment
starting switched models (block la) and constraints
corresponding to those of (1) and (2). It will be shown in Section 7 how the linear model
Naturally, we can now proceed from the basic state- of the modulator stage can be obtained. It remains
space averaged model (block 2a) via hybrid modelling and θimple to incorporate the canonical circuit model (block
circuit recognition (block 2c) to arrive at the very use­ 5 in the Flowchart of Fig. 4) to arrive at the linear
ful cÂAcuUX realization (block 2b). Note, however, that circuit model of a closed-loop switching regulator oper­
now the constraint (1) effectively leads to shorting the ating in the discontinuous conduction mode.
inductance L in the circuit model since v^ - L di/dt * 0.
This, for the particular boost circuit example, reduces A word of caution, however, is appropriate here.
the circuit to first order. The other constraint (2) Namely, since the very nature of operation in the dis­
is also easily specified (see additional constraint in continuous conduction mode is that the order of the sys­
block 2b) with the help of the inductor current wave­ tem is reduced at least by one, this would definitely
form (block 1). The same circuit model (block 2b) could, change the dynamics and possible compensation networks
however, be obtained directly from the switched circuit necessary for stable operation of the closed-loop regu­
models (block lb), by following the circuit averaging lator. Furthermore, if both conduction modes are expected
path, provided the circuit averaging step for "three- to take place for the particular application, the com­
state" converters is supplemented by the aforementioned pensation network should be designed to ensure stability
equivalents of the constraints (1) and (2). Again, the of the closed-loop and acceptable transient performance
remaining circuit perturbation (block 3b) and circuit for either of the two modes. Hence canonical circuit
linearization steps are straightforward and result in models for both continuous and discontinuous conduction
the final circuit averaged models (block 4b) separately mode become an invaluable tool in the proper design of
for dc and ac email-signal. . As seen from block 4b, the switching regulators. In addition, comparison of the
dc part of the perturbation equation, current I, together advantages and/or disadvantages between the two modes
with the dc circuit model, completely determines the dc of operation become feasible, and possible trade-offs
conditions, while its ac part î contributes to the between regulator performance and choice of parameters
final ac circuit averaged model. and operating conditions is clearly displayed.

P E S C 77 R E C O R D - 3 9
 In summary, the new method is generally applicable for this particular example and discussed in depth,
to any "three-state" converter operating in the discon­ including determination of the boundary between the two
tinuous conduction mode (block 4a), even though for an modes of converter operation.. From the dynamic (ac small-
arbitrary converter the final circuit model (block 4b) signal) model, the two transfer functions of interest
may have different (more complicated) topology than the (v(s)/v (s) and v(s)/â(s)) are also determined to enable
canonical circuit model for the three common converters compariSon with the corresponding transfer functions
(block 5). We also emphasize the fact that the methods derived from the final circuit averaged model for the
for finding dc and ac small-signal models are consistent boost converter presented in Section 3.3.
with each other. Namely, for both models we need only
the standard state-space or circuit averaging step (de­ Basic state-space averaged model
pending on whether path a or b is chosen) applicable to
any converter with three switched network configurations. We first define the time-domain description of an
Then to distinguish that the converter is operating in arbitrary three-state switching converter operating in
the discontinuous conduction mode, additional restric­ the discontinuous conduction mode with the help of Fig. 5,
tions (1) and (2) are imposed. Now, the dc part of per­ which displays the switch drive (Fig. 5a) and instantan­
turbation equation (2) together with the dc state-space eous inductor current (Fig. 5b) which becomes discontin­
or circuit averaged model completely determines the uous. The definition of the three intervals Τ d , T_d , and
final dc model, while the ac part i of (2) helps in com­ - S -Τ e 2>
, ! » D-,
,
n T d 0

T d« (or corresponding steady-state quantities

s
plete definition of the final ac small-signal state-space T D , and T D^) is also clearly visible on Fig. §.
or circuit averaged model.
g 2
s"3 e

a) dit)
It may seem that the method outlined holds only for
"three-state" converters in discontinuous conduction mode.
This is not so, since it can easily be generalized to
include more complicated schemes of discontinuous con­ dTs
duction mode of operation. As an illustration of this
generality, consider the new class of switching conver­ t Ts 2

ters of Appendix A, the cascade connection of ordinary b)

buck and boost converters, which could also be classified (Vgwg)/L slope

as two-inductor converters ( as opposed, for example, to

the converters of Fig. 1 which are one-inductor conver­
ters) . Suppose also that the two switches are driven D T 3 S

synchronously with the same switch duty ratio D, thus

resulting in a two-state converter for continuous con­
duction operation. If, however, one of the two inductor
currents becomes discontinuous, a three-state converter DTs D Ts cfeTs 2

operating in the discontinuous conduction mode is obtained.

But now the matrices A^, k^> A^ and A would be of 4th ¥ig. 5. Ve^inition oi the tune intervals and pertur­
order (as opposed to 2nd order for the converters of bation quantities: a) transistor smXcW drive;
Fig. 1) and the final sÇate-space or circuit averaged b) instantaneous inductor current.
model would be of the 3rd order (reduction of order by
one due to discontinuity of one of the two inductor As seen from Fig. 5, the "off" interval [t^T ]
currents). Moreover, there is also the possibility that is now subdivided into two intervals Τ d and Τ d~ (or ?
S

both inductor currents could become discontinuous under T D and T D ) . While the first "on" interval $ D is
g

certain operating conditions in which case four-state dictated by the switch drive and is a known quantity
converters are generated. Therefore, the generalized (at least in open-loop converter usage), the second inter­
state-space averaging step (Appendix A) applicable to val T d« (or T D ) , which will be termed the "decay"
s g 2

four-state converters is supplemented with additional interval, is as yet unknown and depends in general on
constraints: for each discontinuous current there will both the length of the first interval and some circuit
•be two constraints imposed analogous to (1) and (2). parameters, and describes how deep in the discontinuous
The immediate consequence of these constraints is that conduction mode the converter is operating. Nevertheless
the fourth-order original converter model becomes only we assume that the decay interval Τ D^ exists (hence the
a second-order final state-space or circuit averaged discontinuous conduction mode) and îeave it to the model­
ling procedure itself to reveal how it is actually
model (with two inductances effectively disappearing
determined.
from the final circuit averaged model).
For each of the three intervals in Fig. 5, there
Despite this demonstration of the generality of the
exists in general a different switched network (compare
method, we will restrict ourselves in the remaining
with Fig. 3 for the buck-boost converter example), which
Sections to the "three-state" converters in the discon­
can be described by a corresponding state-space equation
tinuous conduction mode since all the essential features
as follows:
of the method are present there.
i - A,x + b,v for interval d,T , (0 - t - t.)
3 STATE-SPACE AVERAGING IN DISCONTINUOUS CONDUCTION
MODE 1 1 g Is 1
χ - A x + b v
0 for interval d T ,
0 (t. - t - t ) (3)
0 0

Various paths on the Flowchart of Fig. 4 will now 2 l g 2 s 1 I

be followed in detail, first with general derivation and
k - A x + b.v for interval d T , (t - t - Τ )
then illustrated by examples. 0 Q 0

3 3 g 3 s 2 s
3.1 State-space averaging While for the continuous conduction mode a similar
expression is sufficient to describe the converter, here
In this section, the final state-space averaged in discontinuous conduction mode, (3) does not describe
model (block 4a of Fig. 4) is derived, first in general the switching converter completely. Namely, the instan­
for any three-state switching converter in discontin­ taneous inductor current is restricted in its evolution
uous conduction mode, and then demonstrated on the since from Fig. 5
idealized boost circuit example (parasitic effects not
included). Steady state (dc) conditions are obtained i(0) = i[(d + d ) T ] = 0 and i(t) ξ 0 for t e[t ,T ] (4)
]L 2 g 2 fl

4 0 - P E S C 77 R E C O R D
Therefore (3) together with (4) completely determine the d A )x + (d b + 2 2 + 3 3 g (8) d b ) v
l l 2 2 ( d A + d A 3 3 1 1 d b

behavior of the switching converter. However, directly

from this description, even the determination of the
steady-state (dc) conditions on an e x a c t basis might be with additional constraints
a very difficult (if not insurmountable) task, and more­
over the tremendous complexity of the result may be un­ Ü - 0 (9)
necessary. In addition, the direct perturbation of (3) dt
and (4) to obtain the dynamic response of the converter
i(v ,v, L, (10)
would become by an order of magnitude more difficult if V g

not virtually impossible. Our objective then becomes,

The two additional constraints (9) and (10) modify the
as it was in [2] for the continuous conduction mode, to
ordinary averaged model (8) to account for the discon­
replace the original converter description through three
tinuity of the inductor current. This model (block 2a
state-space equations (3) by a single state-space descrip- . starting point for all
ln t h e F l o w c h a r t o f F i g 4 ) i s t h e

tion which will accuratley represent the evolution of the ate-space and circuit-oriented)
o t h e r d e r i v a t l o n 8 ( b o t h 8t

state-vector at the switching instants. It is also desir-_and represents an averaged model over a single period T .
able that the additional constraint (4) be appropriately
g

accounted for to modify this averaging equivalent, but Note, also from (7) that the caluclation of the
in such a way as to interfere the least possible with its average inductor current i is actually based on the assump­
orderly procedure. tion of the linearity of the inductor current waveform
(triangular waveshape in Fig. 5 ) . However, this does
The first task is accomplished by application of the
generalized state-space averaging step for three-state not pose any limitations at all, since the linearity of
converters (Appendix A) to (3), which results in a single the inductor waveform is again a consequence of the small
state-space description switching ripple requirement and therefore consistent
with the same basic assumption made in the continuous
x - (d A + d A conduction mode.
(5)
1 1 2 2 +
3 3 d
< l l
A
2 2
) x +
3 3 g
d b + d b + d b ) v

We now consider first the simplest possible case,

Note, however, that this continuous description is a determination of the basic dc conditions in the steady
continuous equivalent to the originally derived approx­ state regime. In the steady state all quantities become
imate discrete system [l]. Hence the definition of a dc quantities and are denoted by capital letters, that
discrete derivative [l] transforms the constraint (4) is, d - D «= D, d = D , d - D , ν - V , x = X. The
x 2 3

into average inductor current i Becomes tfte Atzady 4tote

average inductor current I (see Fig. 5b, for example) and
i(T ) i(O)
di (nT ) - s
the steady-state vector Χ ( ΐ V . . . ) . Since then β

- 0 (6)
dt dX/dt ξ 0, the state-space equation (8) reduces to the
linear algebraic system
It follows that the inductor current in the equi­
valent continuous system (5) ceases to be a tAae state- AX + bV = 0 (11)
Space vanlable, since according to (6) it has lost its g
dynamic properties. Nevertheless, despite the zero con­
straints i(nT ) - 0 and di/dt(nT ) » 0 for η = 0,1,..., where
g

a line voltage perturbation ν (as seen in Fig. 5b) does

D A.
cause a perturbation of the i&tantaneous inductor cur­ 1 1 2 3 3 D A
2
+ D A

rent (shown in dotted lines on Fig. 5b) from its steady- (12)
state waveform (heavy line in Fig. 5b), which in turn D b
3 3 2 2 + D b

results in a corresponding perturbation ν of the output

steady-state voltage. Note that there is also pertur­
while the first constraint (9) is automatically satisfied
bation of the average inductor current i (defined in Fig.
5b for interval (d, + <* )Τ when instantaneous inductor and the second constraint becomes
2

current i(t) is different from zero) from its steady-

i(V , V, (13)
state average current I. This is in sharp contrast to g

the situation in the continuous conduction mode where the

average inductor current does not change under any small- It is now interesting to compare these results for
signal perturbation, but rather initial and final con­ dc conditions (11),(12) and (13) with those for the con­
ditions 1(0) and i(T ) change accordingly to accommodate tinuous conduction mode [2]. For easier correlation of
perturbation. Here, i(0) and i(T ) are fixed at zero, these results, the notation d^ = d and * D henceforth
and the average inductor current Is the quantity which will be used interchangeably. The steady state vector
reflects the effect of introduced perturbation. X is the solution of the linear system (11) as it was
before in [2]. Hence storage elements (L's and C's) are
Since the objective in modelling the dynamic per­ proportionality constants in the linear system (11) and
formance of the converter is faithfully to represent it appears as though solution X of (11) is independent
departure from the steady-state, we introduce the average of them and dependent on dc duty ratios and resistances
inductor current as a substitute for the "lost" state- in the original model. However, since D. + D„ + D = 1
variable (the instantaneous inductor current). But, or D , • 1 - (IH-Do) from (11) and (12) it follows that the
rather than change the symbol, we assign to the same steady state vector X is now dependent on two duty ratios
designation i this new meaning. Then from Fig. 5b we D (given) and D (as yet undetermined) as opposed to only 2

obtain D in [2]. The additional constraint (13) which expresses

i the average steady -State inductor current I in terms of
i(v , v, d, L, (7) circuit parameter values can now be used together with
Vg

(11) to solve for the unknown duty ratio D~, and hence to
determine the length of the second interval 2^s* D I n

and designate it peituhbation equation I, for reasons general, then, D is dependent on circuit parameters (such
which will become apparent later. Naturally, the other
2

as L and Τ , for example) and hence dc conditions are also

constraint (6) for this average inductor current i is substantially dependent on switching frequency f and
maintained (as seen also from Fig. 5b) and we finally
ß

inductance L. This is in sharp contrast to the contin­

obtain the basic state-space averaged model for discon­ uous conduction mode [2], where dc conditions are depen­
tinuous conduction mode: dent on duty ratio D and resistances only.

P E S C 77 R E C O R D - 4 1
 In summary, expressions (11) and (13) completely Vynamlc {ac small-signal) model.*
determine the dc conditions in the discontinuous con­
duction mode, and at the same time help to determine χ » Ax+bv +d[(A -A )X+(b -b )V ]+d [(A -A )X+(b -b )V ]
g 1 3 1 3 g 2 2 3 2 3 g

the length of the second interval D T , unknown 2 g

w n i c n w a s

at the beginning of this analysis. (22)

We now undertake to obtain the dynamic model by per­ subject to constraints

turbation of the basic model (8-10).
(23)
PeituAbation
3i - . di - . 3i -
(24)
Suppose that the switch drive duty ratio d changes 3v g g
v +
3^ 3d
V + d

from cycle to cycle, in addition to the line voltage

variation. Hence, the general perturbation equations
where A and b are as given before by (12).
d a
D + d d, = D d , d = D f d,
2 + 2 3 3 3
From (24) it also becomes obvious why (7) was ori­
(14)
ginally called "perturbation, equation I." In addition,
ν = V + ν , χ = X + x, and i = I + i
g g g since χ [di/dt dv/dt ...]
s
the introduction of con­
straint (23) into (22) reduces the first dynamic equa­
introduced into the basic-state space averaged model tion to a static one, from which the unknown modulation
given by (8), (9), and (10) result in d can be determined in terms of ν and i modulations
2

and circuit parameters. ^

i - [(IH-d)A + (D +d )A + (D -d-d )A ](X+x) +
1 2 2 2 3 2 3 The dynamic state-space equation which, because of
(15) (23), became a static one, can now be designated "pertur­
+ [(IM-d)b + (D +d)b + (D -d-d )b ](V +v )
1 2 2 3 2 3 g g
bation equation II," since it helps to determine the
other unknown perturbation quantity a^. Together with
(24) this uniquely defines the line transfer function
with additional constraints y(s)/v (s) and duty ratio modulation transfer function
v(s)/dfs). However, owing to the presence of constraints
= ο (16) (23) and (24) no closed-form expression is available for
dt the transfer functions, unlike the case for the contin­
uous conduction mode.
I + i - i(V +v , V+v, I>fd, L, T )
g g fi (17)
We conclude this section with illustration of these
general results on the boost converter. Both dc and ac
From d + d« + d~ Ξ 1, when perturbed by (14), we got small-signal models are then analyzed in detail and some
D + d + D + d + D + d = 1 or, since also D + D +
2 2 3 3 £
unique insights into the operation of the boost converter
D Ξ 1, we finally arrive at
3
in the discontinuous conduction mode are obtained. Dc
conditions and the determination of the boundary of the
d 3 = - (d+d ) 2 (18) two modes of operation are particularly thoroughly analyzed.

which was then used in (15). Example.: Ideal boost ροωεΛ stage In discontinuous
conduction mode
The perturbed model given by (15), (16), and (17)
is nonlinear owing to the presence of at least second- For the ideal boost power stage of Fig. 1 the three
order terms. switched networks in the discontinuous conduction mode
of operation are shown in Fig, 6.
Linearization and fainal state-space averaged modal
£osi discontinuous conduction mode
a) INTERVAL dis b) i n t e r v a l d T :
2 s c] i n T E R v a l d/T s :

We now make the small-signal approximation, namely L

that the departures from the steady-state values are small
i=0
compared to the steady-state values themselves:

ν d d χ
' £2 '
9
K< λ Κ<1
V* « D X < K 1 ( 1 9 )

g Flg. 6. ThAee switched networks o£ the Ideal boost

conve/iteK ο h Flg. 1 operating In the discon­
Using approximations (19) we neglect all second (or tinuous conduction mode..
higher) order terms, and obtain once again a linear sys­
tem but including duty-ratio modulation d. After sepa­
rating the steady-state (dc) and dynamic (ac) parts of
both state-space equations (15) and constraints (16) and For the choice of state-space vector χ - (i ν) ,
(17) we arrive at the following results for the final the state-space equations of the three linear switched
state-space averaged model. networks in Fig. 6 become:

Steady state [dc] model: £ * A.x + b_v for interval dT

1 1 g
X = -A'HV (20) χ A - X + b„v
β
for interval d T (25)
g 2 2

2 2 g
Subject to constraint χ = A x + b v for interval d T
0 0
J s 0

3 3 g
I = i(V , V, D, L, T ) (21)
g g
where

4 2 - P E S C 77 R E C O R D
 « -i κ Δ 2L Δ 2L .
(33)
RT ΪΓ s f

1_
Ο - ~r This dimensionless parameter Κ plays a key role in the
RC
discontinuous conduction mode since it combines uniquely
all the parameters responsible for such behavior. Ano­

°J [l L° °]
T ther quantity which will frequently appear is the dc
voltage gain V/V , so we define also another dimension-
b
2= b
3 " less parameter M as 8

In addition to this, perturbation equation I (7) is needed. Δ v_

(34)
However, it can easily be found from Fig. 6a as V
g

Finally, by use of (32) and (34) in yet unused dc rela­

tion (29), the quadratic equation for de gain M is ob­
The same result could have been concluded also from Fig. 5b, tained
which actually represents instantaneous inductor current
for the boost converter (or buck-boost converter since M - M - D /K = 0 (35)
2 2

both have the same slope during interval dTj.

Since from (29) the de gain M is positive, only the
Equations (26) and (27) contain now all that is needed positive solution of (35) is meaningful and we obtain
to determine both dc and ac small-signal models by appli­
cation of the general result, equations (20) through (24). + / 1 + 4P /Κ 2

We first analyze in greater depth the steady-state (dc) M = (36)

model.
Finally, the substitution of (36) in (32) determines
the previously unknown duty ratio D as 2
Steady state (dc) model analysis
Κ 1 + / 1 + 4D /K
2

By use of (26) in (20) the following linear algebraic (37)

system results

A X b Hence, we have succeeded in expressing, through (36)

and (37), two important quantities, the de gain M and
duty ratio D , in terms of the driving condition (duty
2
D+D ratio D of the transistor switch), and the Single
2D
τ
2

0 - 1 dimensionless quantity Κ which solely reflects the effect

L L
of circuit parameter values (L and R) and the other oper­
+ 0 (28) ating condition, the switching frequency f , upon the
dc conditions in the discontinuous conduction mode. If
1 desired, the remaining dc quantity, the steady-state
V 0
" RC average inductor current I, may be found in terms of
D and Κ by use of (37) in (30).
in which the quantities A, X and b are clearly identified
and obtained by use of their definition (12). The general All these expressions (36), (37), and (30) are very
remark made previously about the solution of this linear useful in predicting the dc conditions when the switching
algebraic system (28) becomes clearly visible. Storage converter is used alone, that is in an open-loop iasklon,
elements (L's and C s) are indeed proportionality con­
f
since the duty ratio D is given (independently generated)
stants, and the solution of (28) is and the constant Κ may be calculated from element values
with use of (33). However, if the converter is used in
a closed-loop switching regulator the output dc voltage
(29) V is predetermined by the choice of the reference vol­
V A +
D 0

tage and kept constant regardless of any variation of

input dc voltage V , by appropriate self-adjustment
(30) of the dc duty ratio D (internally generated) in a
D
2 R

negative feedback manner. Hence in closed-loop opera­

tion , D and D become dependent
2 on the external dc gain
Hence, the dc conditions depend only on duty ratios D M and the dimensionless parameter K. These dependences
and and resistance R. From (29) we conclude also that can easily be found from (36) and (37) to get, for
the boost converter has even in the discontinuous con­ >
closed-loop consideration:
duction mode the boosting property (dc gain V/V - 1 ) ,
since D, D are by definition positive q u a n t i t i e s . How­
2

ever, the dc conditions are not quite determined since D = /KM(M-l) (38)
D is as yet unknown. But, by use of the additional
2

constraint (21), as further specified in (27) as (39)

V DT
S s

(31)
2L Hence, (36) and (37) conveniently determine dc quantities
for open-loop considerations, while (38) and (39) are
together with (29) and (30), dc conditions (and also D ) 2
likewise useful for closed-loop considerations.
are completely determined. For example, substitution
of (31) into (30) results in It is now interesting to compare the open-loop dc
gain in the discontinuous conduction mode given by (36)
2L MK
D
2 ™" RI (32) with the corresponding dc gain in the continuous con­
R DT V D
s g g duction mode, which, for the ideal boost converter is

where the important dimensionless quantity Κ is defined M (40)

P E S C 77 R E C O R D - 4 3
Hence, the ideal dc gain (40) is dependent on duty ratio has a maximum of 4/27 at D » 1/3. This now enables an
D only and not on circuit parameters (such as L, R) or important conclusion about operating mode to be drawn.
switching frequency f . In sharp contrast to this, the Namely, if the parameters L, R, and f are such that the
de gain M in the discontinuous conduction mode (36) is computed parameter Κ is greater than S/27, expression (46)
dependent also on Κ in addition to D and hence is a is satisfied JuigaAdleSA of duty ratio D. Hence for
strong function of switching frequency f , inductance L, Κ > 4/27 the converter always operates in the continuous
and load R. Nevertheless, when the converter is used conduction mode, no matter what the operating condition
in this mode in a closed-loop regulator, the self- (duty ratio D) is. However, if parameters L, R, and f
correcting feature of the duty ratio D would compensate are such that Κ < 4/27 ^ 0.15 the situation becomes as
any possible changes of load R or switching frequency f shown in Fig. 7a, where the particular example of Κ »
g

and still keep the output voltage relatively constant. 0.08 < 0.15 is chosen. For a certain range of duty ratio
D, that is D . < D < I> (as shown by the shaded area
Another question naturally arises in comparison of in Fig. 7a),cße condition (47) is satisfied and the con­
the two dc gains: when do we calculate dc gain from one verter operates in the discontinuous conduction mode,
(36) or the other formula (40), or, what is the criterion while for the remaining portions of the operating range
to determine in which of the two modes (continuous or (0 < D < D and D < D < 1.0) it again operates in
m i n

discontinuous) the converter is operating? The answer the continuous conduction mode, since then inequality
is provided easily with reference to Fig. 5. When the (46) holds.
second interval D T is smaller than interval (l-D)T ,
2 g g

the converter is operating in the discontinuous con­ a) o p e n l o o p c o n s i d e r a t i o n b) c l o s e d loop consideration

duction mode, and in continuous mode otherwise, so the
!W(D)
criterion becomes ι discontinuous
conduction
continuous conduction mode. M-l
4 /Dfi-D) 2

4
27 27
Do > 1 (41) χ /
K=0.08 K=0.08

discontinuous conduction

D 2 < 1 - D
mode.

(42)
/ Dmin D
max
1.0
D

Hg. VeteAmination of the operating mode (contin­

To obtain a convenient quantitative measure we find, uous OK dis continuous) foK the Ideal boost
first, what happens exactly on the boundary between the
conveAteA of flg. 1.
two modes of converter operation, or

boundary between two conduction modes This discussion has been in terms of open-loop con­
siderations, when duty ratio D is given and externally
D « 1 - D controlled. However, as before for dc conditions, it is
2 (43) desirable to have the boundary condition (45) in terms
of the dc gain M, which is a more suitable quantity for
By use of (37) in (43), the equation to determine the closed-loop considerations. This can easily be done
critical value of parameter K, that is, Κ for which since the de gain M is continuous across the boundary
this happens, is crit (as seen by use of (43) in (29) resulting in (40)), and
thus substitution D » (M-l)/M in (45) gives
K
crit +
« W * « 2 D D
' - K
cri t
(44) M-l
crit (49)
from which M 3

,2 This function Κ 0 0 *s plotted in Fig. 7b, and a

crit DD (45) l t

similar discussion applies. However, now the maximum

of Κ ( M ) of 4/27 is obtained for gain M = 1.5.
it

The solution (45) is the proper solution of (44) since As before, for Κ < 4/27, the converter is in the dis­
2DD» - Κ - 2DD* - D D = 2DD»(2-D») = 2DD (1+D) is
, Z f
continuous conduction mode, but now for de gain M in
always positive regardless of D, resulting in a proper the range M ^ < M < M as shown by the shaded area
positive right-hand side of (44). With this, the cri­ in Fig. 7b. $his reveaîs a potentially serious problem
teria (41) and (42) for determination of the operating if the boost regulator were designed (and compensated)
mode become to operate in the discontinuous conduction mode only.
Namely, during the initial turn-on process, the out­
continuous conduction mode κ > κ (46) put voltage starts from zero, and the converter would
crit
have to pass through the continuous conduction region
discontinuous conduction mode {^ASt (for 1 < Μ < Μ ^ ) , before coining to the dis­
η

κ < κcrit continuous conduction region (shaded area in Fig. 7b).

(47)
This would suggest possible stability problems, if the
closed-loop were not compensated to assure Stable
boundary between two conduction modes operation in the continuous conduction mode as well.

Κ » Κ (48) From the standpoint of the dc gains (as a function

crit
of duty ratio D ) , the situation corresponding to that
where K, as given before by (33), is a function of of Fig. 7 is shown in Fig. 8 for some Κ < 4/27.
parameters L, R, and f , while *e a function of
g
K
c r i t
From the dc gains for both conduction modes shown
the duty ratio D only. in Fig. 8, it becomes obvious that the actual dc gain
will follow the laKgeK of the two gains, thus the mode
We now investigate how these criteria, (46) through of operation will change accordingly as the duty ratio
(48), behave throughout the duty ratio range D e[0,l]. changes from 0 to J. Also in the close vicinity of
<
To facilitate this insight, Κ is plotted as a function gain M = 1 ( 1 - M - M ^ ) , the converter is always
of duty ratio D in Fig. 7. Asseen in Fig. 7a, Κ (D) operating in the continuous conduction mode. Thus, the

4 4 - P E S C 77 R E C O R D
problem of having, for example, infinite when M*l continuous conduction mode
from (39) is only a fictitious one, since (39) is for
the discontinuous conduction mode and hence not appli­ R < R . (53)
crit
cable in the vicinity of and at gain M • 1.
di6continuous conduction mode
R>R (54)

\
crit u

boundary between two modes

1 + Vl + 4 D / K
2

2 R *= R . (55)
crit
Κ < 4' 27

Let us now illustrate this on a numerical example. For

L « 880uF, f - 20kHz we calculate R - 35.2Ω. By the
1 same argument as before (see Figs. 7 and 8, for example), n

0.0 Dmax 1.0 D the converter will always operate in the continuous con­
duction mode if
Fig. 8. Boost conventeK de voltage gains in continuous 27
and discontinuous conduction modes as a function R < (56)
oi duty hatio V.
or for the given numerical example for R < 238Ω. When
R > 238Ω there will be a range of gain M (see Fig. 8)
We conclude this dc analysis with some numerical for which the converter operates in the discontinuous
examples and related quantitative and qualitative signi­ conduction mode.
ficance of the dimensionless parameter K. For example,
for the set of parameters L » 880uH, R » 220Ω and f β
This concludes the extensive dc analysis and we now
20kHz, we compute Κ * 2Lf /R - 0.16. Therefore, since turn to the dynamic (ac small-signal) model analysis of
Κ « 0.16 > 4/27, the converter will with this set of this ideal boost converter example.
parameters always operate in the continuous conduction
mode. However if, for example, the switching frequency dynamic [ac small-signal) model analysis
is reduced to f * 10kHz, this results in Κ - 0.08 < 4/27
and some range of discontinuous conduction operation Before we apply the general result to this ideal
should be expected (see Figs. 7 and 8). Therefore, the boost converter example, let us first put the con­
reduction of parameter Κ below 4/27 causes this transi­ straint (27) into a more suitable form by using the
tion. From the definition of Κ in (33) this reduction steady-state average inductor current I of (31) to get
and change to the discontinuous conduction mode is
qualitatively achieved by three means: increase of ν dT ν d
8 8
-Α­ (57)
load R, decrease of the inductance L or switching fre­ 2L ν D
quency f . There is also a fourth way to enter the dis­ g
continuous conduction mode, and that is to change the
operating condition, the duty ratio D, as illustrated By use of perturbation equation (57), model description
in Fig. 7 and Fig. 8, but only if the condition Κ < 4/27 (26) and definition (12) in the general result given by
is met. (22) through (24), we obtain
Very often, however, out of all these four pos­ dynamic (ac small-signal) model
sibilities, one is mostly interested in how the change
m —
of load R affects the operating mode. Namely, the
parameters L and f are usually design parameters whose
g
di
0
V i
" DfD ~ 2 V
_£
choice may depend on the size and efficiency require­ dt L L L
ments of th^ converter or regulator. On the other hand, at + ν +
the range of variation of duty ratio D, or equivalently 1 g
of gain M, is a design requirement in a closed-loop dv D
2 η n
dt C RC V υ u
implementation since the output voltage V is maintained
constant against the range of variation of input voltage -
V (hence range of M * V/V ) by the action of negative V -V
fledback. The load R also can have a wide range of
S

change depending on the user of the regulator, and is s

often out of the designer's control. Hence, determi­ (58)
nation of the converter operating mode with respect to
I
changes of load R becomes important. This can be easily
accomplished by finding an equivalent of (45) and (49) C
respectively, as
with additional constraints
crit (50)
,2 nom
(59)
β-·
crit (51)
(60)
where R is a design parameter defined by
nom
As opposed to the general result, we can now for
2Lf (52) this specific example enter the constraints (59) and (60)
into dynamic model description (58). The introduction of
The criteria for determination of the operating mode, (59) reduces the first dynamic equation in (58) to a
( 4 6 ) , (47), and (48), then become static one, and after proportionality constant L is re­
moved the dynamic model becomes

P E S C 77 R E C O R D - 4 5
 where
Ο = - D v + (IM-D )v
2 2 + V d + (V -V)d g 2
(61)
2M-1 1_
(67)
M-l RC
v/R + Id (62)
C
dt V 2

and
with additional constraint (60). Note, however, that now 2V / KM
the first static equation (61) actually determines the M, 2M-1
"od 2M- M-l
(68)
J
unknown modulation quantity d (modulation of the second 2

interval d T as shown in Fig. 5, for example) in terms

2 g As seen from (66) both transfer functions have a
of the other dc and ac quantities. In the remaining single pole ω and no zeros. This is qualitatively com­
dynamic equation (62), besides this modulation d which pletely diffefent dynamic behavior than in the contin­
2

we can now express from (61), current modulation i also uous conduction mode where two poles and even a right
appears. But, from the perturbation equation I (60) it half-plane zero are obtained [2] (for the G ^ transfer
is also determined in terms θί the known ac quantities function only). This in turn suggests easier compen­
Λ

(forced modulations ν and d). In general, both equa- sation (even no compensation at all) and reduced sta­
tions(60) and (61) c§uld have both modulation quanti­ bility problems if the converter as a part of a swit­
ties i and d for some arbitrary converter. But, they
2
ching regulator is operating consistently in the dis­
are linear algebraic equations and could be solved for continuous conduction mode. But, a potential danger
i and do in terms of other ac quantities and then sub­ exists there: any significant transient changes (such
stituted in the remaining dynamic description (which could as sudden change of input voltage or temporary substan­
be, for some converter with more than two storage elements, tial change of load R) could move the operating point
higher than the first order model given by (62)). to the continuous conduction region (see Fig. 8) and
cause instability. Another problem is inherent to the
Another general feature, which is hidden in this discontinuous conduction mode. In addition to the out­
model, is that (61) can be considered as a consequence put current, now the input current becomes pulsating
of the equation as well (as shown in Fig. 5) which increases elect­
romagnetic interference problems. Hence, a decision
(d+d W dv (63) on the choice of operating mode becomes a complex one,
9 2
g 2
depending on the particular design requirements. To
which after usual perturbation and linearization steps facilitate that decision, we now undertake the task of
and subtraction of dc terms reduces to (61). Hence, developing useful circuit models of the switching con­
in analogy to (57), equation (63) can now be designated verter operating in the discontinuous conduction mode.
perturbation equation II. The appearance of (63) in
the modelling will become more apparent later in the
hybrid modelling and circuit averaging techniques. But
in any case, the unknown modulation quantities i and d 2

come as the solution of two linear algebraic equations,

which are essentially linearized versions of pertur­
bation equations I and II, (57) and (63) respectively. 4 HYBRID MODELLING IN DISCONTINUOUS CONDUCTION MODE
To complete the dynamic model description we simply We demonstrate in this section how for any specific
substitute (60) and the solution of d from (61) in (62) 2
converter a useful circuit model of the basic state-
to get space averaged model (8) can be found, appropriately
dv / v i 2 m-D 2
modified by inclusion of the constraint (9), and sup­
V + Iv + Id plemented by the additional constraint (10). In terms
c
d F Ä
v-v g
\D V-V
of the Flowchart of Fig. 4 we will proceed from block
g
2a through 2c to arrive at the circuit model in block 2b.
(64) Again this is illustrated on the same ideal boost con­
verter example as in the previous section.
Since this dynamic model has significance only for the
closed-loop regulator, it is convenient to express all When the boost converter description (26) and (27)
dc quantities in terms of M, K, R and output voltage V, is applied to (8), (9) and (10) the following basic
as was explained before in the dc analysis. Hence by state-space averaged model results:
use of (38), (39) and (30) we obtain
di d+d 0

dv 2M-1 1 Λ
M 2M-1 2V 0 - τ ±

d (65) dt
dt M-l R M-l R g V
R / KM (M-l)
(69)
In (65) all proportionality constants would become infi­
nite and meaningless when M - l . However, it was ex­ dv
plained in the dc analysis that in the vicinity of and at dt RC L _J
V

gain M 1, the boost converter always operates in the

s

continuous conduction mode, hence a different dynamic with additional constraints

model applies.
(70)
Is is now easy to obtain from (65) two transfer
functions of interest ν dT
g s

(71)
2L
v(s) =
(66)
vg v (s)
g
og 1 + s/ω It now becomes clear that introduction of (70) into (69)
reduces the first dynamic equation to perturbation equa­
tion II as given before by (63). But, instead of intro­
v(s) ducing this substitution, let us first find the circuit
= G realization of the state-space equations (69) as shown
vd od
d(s) 1 + s/ω in Fig. 9.

4 6 - P E S C 77 R E C O R D
 The constraint (70) leads, in the circuit model of duced. For consistency, the conventional, physically
Fig. 9, to effective disappearance of the inductance L, realizable, ac transformer only, is pictorially repre­
since v, - Ldi/dt « 0. The resulting equality of the sented as in Fig. 11c. Later in Section 5.2, for si­
two voltage generators produces again the perturbation milar purposes, the same overprint glyphs will be used
equation II given by (63). At the same time shorting with resistance symbols.
of the inductance causes reduction of system order by Following the procedure outlined in this section
one, and effectively a single pole transfer function one can easily obtain the basic averaged circuit models
result becomes apparent. of the three common power stages of Fig. 1. These
models for discontinuous conduction mode are summarized
v ~ Ldildt-o
L in Fig. 12,

a) buck power stage

L |v -v)dT /2L
g s

|d + d i v ( i
2 g

* R

fig. 9. CircuU rzalization of the state-space model

(69), with constraint (70) aUo Included.
Let us now put the circuit of Fig. 9 into more
elegant form, by introducing a dc and ac transformer
in place of the two dependent generators in Fig. 9. Also
it is desirable to have source voltage ν effectively
at the input of the converter, rather thin as some
modified quantity as (d+d )v in Fig. 9. However, this
2 SR
is easily accomplished by introduction of another dc
and ac transformer at the input of the converter. In
addition, the true input current into the converter
becomes properly exposed as seen in the basic circuit-
averaged model of Fig. 10. In addition to the circuit
model in Fig. 10 we need the remaining constraint (71)
to complete the description of the converter in dis­
continuous conduction mode (as also displayed in Fig. 10).
As before, the circuit model and the additional pertur­
bation equation are valid for both dc and ac conditions. =fC > R
Hence the two transformers in Fig. 10 are operating
both at ac and dc and the appropriate symbol is introduced.

jjn _ 1 iout V

fig. 12. Summary oi the basic circuit averaged models

{or three common power stages in discontinuous
conduction mode.

(d+d )
2 d : I
2

5 CIRCUIT AVERAGING IN DISCONTINUOUS CONDUCTION MODE

Fig. 10. Basic circuit averaged model, {or the ideal
boost converter in the discontinuous con­
duction mode. In this section the alternative path b in the Flow­
chart of Fig. 4 is followed and the perturbation and
linearization steps corresponding to those in state-
A word about the new transformer symbol introduced space averaging path a are applied to the circuit model
in Fig. 10 is appropriate here. In the modelling of to arrive at the final circuit averaged models, separ­
dc-to-dc converters a need naturally arises to have as ately for steady-state (dc) and dynamic (ac) response.
a convenient modelling tool special types of transfor­
mers: a transformer which operates for both ac and dc We continue with the same ideal boost converter
signals, as for example that in Fig. 10, and also a trans­ example and hence use as a starting model the circuit
former which only works at dc (for which the need will model of Fig. 10. Even though that circuit model was
arise later in Section 5.1). Even though these trans­ obtained by following hybrid modelling, we emphasize
formers are not physically realizable they are, never­ also the other possibility. Namely, it could have been
theless, useful in modelling the basic converter function: obtained directly by averaging the three switched cir­
dc-to-dc conversion. Hence, as an indicator of their cuit models of Fig. 6 using the standard circuit aver­
specific functions, the symbols of Fig. 11 are intro- aging technique and supplementing it by the appropriate
constraints (70) and (71).
a) dc a n d a c t r a n s f o r m e r b ) d c t ra n s f o rm e r c)ac transformer
o-
Perturbation
If the averaged circuit model of Fig. 10 is per­
turbed ^ together with its perturbation equation I ac­
cording to

Tig. 11. definition o{ various transformer symbols used ν = V +v , l+i, Dfd, D d , ν = V+v
in modelling switching dc-to-dc converters. g g g 2 + 2

(72)

PESC 77 RECORD-47
 the nonlinear model of Fig. 13 results. 5.2 Dynamic (ac) circuit model

After the steady-state (dc) quantities are sub­

tracted from the circuit model in Fig. 14 (and per­
(D*rVd.d XI.i) U i (D *d )(V+v) (D *4)(I*i) V+v
turbation equation as well) the ac circuit model in Fig.
2 2 2 2

' 11 / 16 is obtained.

Q 0 τ (D+D )i+(^d )I
2 2

(DHD )y(d*d>)V
2 g
\
Wi=(V w )(D+d)ls-/2L
g g

Fig. 1 3 , Fentunbatton oi the basic avenaged cVicuiZ

model In Fig. 10 hjesutts In this nonlineaA
circuit model. i = Id/D + Iv /V g g

Linearization Fig. 16. Dynamic [ac small-signal) coicult model ioK

the boost conventeK utiXn the consVuxint on
With the small-signal assumption on perturbation, modulation 1 [pentuAbation equation I ) not
that is yet Included In the clKcult model.
d « D, d 2 « D ,
2 i « I, ν « V, ν « V (73)
8 8 From Fig. 16 it is obvious that the two dependent
current generators are functions of two yet undeter­
the second-order terms in Fig. 13 can be neglected and mined modulation quantities cL, and 1, since the other
the linearized model of Fig. 14 obtained. quantities are either already determined from the dc
circuit model (euch as D , I) or are known driving 2

V+v quantities (as D and d). While the current modulation

- + — · — +- is already available through the linearized pertur­

m
1
— 1
I+i \ bation equation I (see Fig. 16), the other modulation
quantity d can easily be obtained from the inside loop
m m
2

fciR of Fig. 16. Namely, since the two voltage generators

(D+D )V 2 g D V 2 D 1
2
in Fig. 16 must be equal, we get

I î = V DTs/2L+
+ g Iâ/D+lv /V g g
(I»D )v + (d+d )V = D v + d V
2 g 2 g 2 2
(74)

Flg. 14. Hödel oi Flg. 13 lineantzed to Include dc and

Note that this is the same equation as the first (static)
ac small-signal models. equation (61) of the state-space averaged model. Now
it is easy to see that (74) and (61) came out actually
The circuit model in Fig. 14 together with the as a consequence of the perturbation and linearization
dc and ac part of the perturbation equation I (also steps applied to the perturbation equation II (63), since
shown in Fig. 14) completely determines both models. the voltage generators in Fig. 16 resulted from the
At this point, we continue to develop separately the perturbation and linearization of the voltage gener­
two circuit models — the steady-state (dc) circuit model ators in Fig. 9, which have been shown to be equal for
and the dynamic (ac small-signal) model. discontinuous conduction mode (owing to di/dt 0 con­ β

straint).

5.1 Steady-state (dc) circuit model The equation (74) can now be solved for the unknown
modulation jpd, together with the perturbation equa­
With all ac quantities set to zero, the dc circuit tion defining i, determines the two current generators
model is obtained directly from Fig. 14, and upon sub­ in terms of the known modulation quantities as follows:
stitution of dc dependent generators by the dc trans­
former symbols, the circuit model in Fig. 15 results. 2VI V (D+D )I
j - (d+d )I + (I»D )i * — d + ψ ^ νg ν (75)
± 2 2
v-v
l=V DT i2L 1 c
g s
g g g

V g - ^ 2V I 2V-V 1
dI + Di g d + & - ν (76)
2 2
V-V V-V R
g g
(D+D2] 0-
2 I
Since the converter dynamic model is usually used in
Fig. 15. Final dc clKcult model Ion the boost conventen. closed-loop regulator applications, we conveniently ex­
In the discontinuous conduction mode. press all dc quantities in terms of M, K, R and output
regulated voltage V (as explained before) to arrive at
This circuit model is also supplemented by the dc
part of the perturbation equation I, which is, of course, • _ 2V / M * M 1 * J
M_ _1 *
(77)
the same as ( 3 1 ) . From the circuit model in Fig. 1 5
3
i R ν/ K(M-l) M-l R g V
M-l R V

the other two dc relations (29) and ( 3 0 ) are obtained.

Hence the dc circuit model leads to the Same dc con­ - m 2V 1
ditions and results discussed at length in Section 3 . 1 ° R V/KM(M-1) M-l R M-l
on state-space averaging.
By use of ( 7 7 ) and ( 7 8 ) in the circuit model of
We now turn to the development of the dynamic (ac) Fig. 1 6 , the circuit model in Fig. 1 7 is generated.
circuit model.

48-PESC 77 RECORD
 An interesting comparison with the circuit model
2\f )IÛ d -M ν M(2M-l)v 2Vd - M v
9
topologies for the continuous conduction mode [l,2] seems
RVKÎMHT (M-I)R (NTÏÏR (M-I)R RVKM(M-I) (M-I)R . appropriate here. While in the continuous conduction
mode the effect of duty ratio modulation d was expressed
through duty ratio dependent voltage and current gener­
ators, here two duty ratio dependent current generators
(one at the input and the other at the output port)
appropriately account for both input and transfer pro­
perties (and output properties, as well). Another dis­
tinction and unique feature of the circuit model of
Flg. 17. Vynamic lac mail-signal) circuit modal oh the Fig. 18 is the presence of ac resistances only (which
boost converter mXh perturbation equation I are in general dependent on an operating condition,
(ior modulation >c) and perturbation equation II the gain M ) , a characteristic not present in the con­
{equality oi the voltage generators ν. and ν ) tinuous conduction mode. But despite these topological
Included In the circuit model. * 0
and qualitative differences, the circuit models for con­
tinuous conduction mode [l,2] and discontinuous con­
The two voltage generators v.^ and ν in Fig. 17 duction mode (Fig. 18) have something very important
are purposely shown in dotted lines to emphasize the in common: they both represent a complete linearized
fact that they are no linger essential, since the in­ circuit model which accurately represents not only
formation provided by them (74) has already been used transfer properties but input and output properties
to find modulation and substituted elsewhere in the as well.
circuit model, Therefore they can now be omitted from
the circuit model. Finally, by modelling the current The method outlined in this section, and illustrated
generators in Fig. 17 which are proportional to vol­ for the boost converter, is applied to the other two
tages across them as ac resistors only, the final cir­ converters of Fig. 1 and results are presented in various
cuit model of Fig. 18 is obtained. tabular forms (including the boost circuit example) in
Section 6 on a canonical circuit model.

6 CANONICAL CIRCUIT MODEL FOR DISCONTINUOUS

CONDUCTION MODE

In this section the canonical circuit model for

discontinuous conduction mode (block 5 in the Flowchart
of Fig. 4 or Fig. 18) is obtained for the three common
switching converters of Fig. 1, and thanks to its fixed
circuit topology, the results are conveniently summarized
in the form of various tables, separately for dc and for
Tig. Π. Final ac small-signal circuit model ior boost ac small-signal circuit models.
converter In the discontinuous conduction mode.
From the dc conditions and by following the deri­
The element values in Fig. 18 are defined as vations presented in Section 3.1, the simple formulas
for determination of the boundary between the two con­
. _ 2V / M M-l _ _M_ 1 duction modes may also be found for the buck and buck-
J
1 ~ R Y K(M-l) 9 r
l S
3 >
K
l " M-l R
g
(79) boost converters. These results, analogous to (45) and
M
(49) through (51) for the boost converter, are again
tabulated for all three common converters of Fig. 1.
2V 1 M-l M(2M-1) 1 This then ultimately determines which of the circuit
R> g (80)
R ^KM(M-l)
9
models (those of [2J or those of Sections 5.1 and 5.2)
M-l R
should be chosen for given parameter values and oper­
ating conditions of a closed-loop switching regulator.
Also since r.. and are ac resistances only, the ap­
An interesting pictorial interpretation facilitating
propriate symbol consistent with that adopted for the
this decision is given in terms of the frequency scale
ideal transformer designation (see Fig. 11, for example)
and position of another "inherent"frequency ω (fre­
is used in Fig. 18. The two current generators inside β

quency defined by converter element values, like ω

the dotted-line box in Fig. 18 are used with square
symbols to emphasize the fact that they are dependent and f before) with respect to switching frequency f .
c s
current generators (on some other quantities in the
circuit). Finally, both dc and ac transfer properties are
experimentally verified on a particular buck-boost con­
From the circuit model in Fig. 18 and by use of verter breadboard and excellent agreement with the pre­
element definitions (79) and (80), the two transfer dictions is observed, thus confirming the high accuracy
functions G , and G can be derived. It can easily be of the circuit models for the discontinuous conduction
mode.
Vd V2
verified that they agree exactly with those obtained 6.1 Derivation of the canonical circuit models for
before, ((66), (67) and (68)), using state-space averaging. discontinuous conduction mode
An interesting observation with regard to the topology
of the circuit model in Fig. 18 can be made. Namely, In this section the canonical circuit models (both
to arrive at these two transfer functions, only the dc and ac small-signal circuit models) for the two re­
elements in the output port j , r^ and g« have been 2
maining converters of Fig. 1 are derived from the basic
used, without any need for input port description. How­ circuit averaged models in Fig. 12.
ever, the input port description becomes mandatory if
the determination of the complete circuit model is de­ Buck converter In discontinuous conduction mode
sired, since it properly models the important input
properties (both open- and closed-loop input impedances, With regard to the dc circuit model derivation, a
for example), as will be illustrated in Section 7-2. general observation seems appropriate here. Namely, the
Moreover, the output port model now does affect the in­ dc circuit model of the boost converter (Fig. 15) could
put properties through the dependent current generator have been obtained directly from the unperturbed cir-
g^v in Fig. 18.

PESC 77 RECORD-49
cuit model in Fig. 12b by simply taking all quantities Buck-boost conveniez in the discontinuous
to be dc quantities and as usual considering the capa­ conduction mode
citance C to be open for dc signals. Hence, as should
have been expected, the circuit models in Fig. 12 to­ The dc circuit model for the buck-boost converter
gether with the additional expressions for the average is obtained directly from the circuit model in Fig. 12c.
inductor current i are valid dc models. But this is After perturbation and linearization of the model, the
exactly why it was previously emphasized that the pre­ dynamic (ac) circuit model in Fig. 20 is obtained.
sented methods for finding dc and ac models are con­
sistent with each other. After all, ac small-signal
models really represent the linearized perturbation Di+dl DÎgtdVg -(D v+â,V) - ( D f c U ^ D 2

around some steady-state (dc) conditions. Hence, by

perturbation and linearization of the circuit models
in Fig. 12, the ac circuit models consistent with the
superimposed dc circuit models result. Therefore, the
dc circuit model for the buck converter is as in Fig. 12a
with dc quantities d D, V i - I. v β

and dc transformers only. g g

After usual perturbation and linearization steps

are applied to the circuit model of Fig. 12a, the dynamic
(ac) circuit model in Fig. 19 is obtained. Hg. 20. Vynamic [ac smaJUL-signal) cixcuit model fon.
the buck-boost conveniez in discontinuous
conduction mode with pentunbation equation I
(ioil) shown explicitly.
DÎ+dl DydVg (D4D )0i-(dtd )V
z 2 tD+D )iMd+d )I
2 2 a

\ \ \ The perturbation equation I is now the same as for

the boost converter (71), and the two current generators
C < R 3 . and 3 in Fig, 20 are as defined in (83) but with
0 J i f f l m * v 0
tne following element values for the buck-boost converter:
« 2lvl R
Ο2 > gl
» 0 (86)
V^R M

Μ ι 8
2M
2 " R
(87)
I/KR M
Hg. 79. Vynamic (ac small-signal) circuit model foK the
buck conveniez in discontinuous conduction mode
Again the same circuit topology of Fig. 18 results, but
with cowueSQonding pentunbation equation I fou
With element values (86) and (87). However, there is a
modulation i.
small distinction from the previous two models since
now, as seen in (86), g - 0. Therefore there is no
x

feedback effect from the output port to the input cir­

The perturbation equation I is different from that cuit model as in the other two converters, and the open-
for the boost converter and is loop input impedance is just r-. But, this is rea­
sonable to expect for the buck=boost converter, since
(v -v)dT (v -v)d
g it is the only converter in which the energy trans­
(81) ferring inductance is present either solely in the in­
2L (V -V)D
g put circuit (interval DT ) or solely in the output cir­
g

cuit (interval D ^ ) . in the other two converters (buck

•After perturbation and linearization of (81) we get and boost), on tne other hand, the output circuit
(including C and R) is at least for a portion of period
i « V -V
ι η . ι - ι (82) Τ connected to the input and represents a loading
g + D V d
- y -V effect on it. Hence the feedback action through cur­
g g
rent generator g ν is to be expected in these two
When the unknown modulation quantity d^ is found from converters.
equality of the two voltage generators in Fig. 19, and The results for all three converters (buck, boost
by use of (82), the two current generators in Fig. 19, and buck-boost) are summarized in the next section.
after expression of dc quantities in terms of closed-
loop parameters M, K, R, and V, become 6.2 Summary of the canonical circuit model results
for three common converters
(83)
j
l â +
V l"Vr ;
*° " J
2 5+ 8
2^8 " * / r
2 In this section the results for both dc and dynamic
(ac) canonical circuit models for buck, boost, and buck-
where boost converter are summarized and, owing to the fixed
1-M j£ 1
(84) circuit model topology, conveniently listed in several
l-M R tables.

. 2V 1 / r 5 „ N D M(2-M) 1 .... In Fig. 21 the polarity of the second transformer

J
2 - r ü j ~ ' r
2 * " a M ) R
' «2 " M ï ( 8 5 )
l:!!« is inverting for the buck-boost converter and
otherwise as shown. The parameters in the dc circuit
Hence the Same topology of the dynamic (ac) model for model of Fig. 21 are defined in the first three columns
the boost converter shown in Fig. 18 is also obtained of Table I, while the remaining two columns tabulate the
for the buck converter .in ttye discontinuous conduction dc relations derived from this circuit model. Note,
however, that this circuit model can be used to deter­
mode, but with the model element values defined by (84)
mine other dc quantities as well, such as the dc input
and (85). durrent I. in terms of the defining parameters,
in

5 0 - P E S C 77 R E C O R D
 STEADY STATE (DC) CIRCUIT MODEL DYNAMIC (AC SMALL SI6NAL ) CIRCUIT MODEL

I : M, I : M 2

Fig. 21. Steady-state, (dc) circuit modeJL {or the con­ Fig. 11. Final ac smalt-signal circuit model {or con­
verters o{ Fig. 1 in the discontinuous con­ verters o{ Fig. 1 in the discontinuous con­
duction mode. duction mode.

converter d e f i n i t i o n of dc model derived quantities type J, Π 9, J 2

Γ
2 9
type 2

M 2 I (average I = M V/R 2 M-M,M 2

buck 2V /l-M l-M ο M 2

I 2VlihM M(2-M! I
( l - M ) R
1 (Vg-V)DTs V D l-M
buck D R V K R R M Ï K l - M R
D +D 2
2L (D + D )R 2 D + D 2

boost 2V1/M R M I 2V M-l ρ M(2M-I) I

1 Vg D T s V D+D RlklM-l) M 3
M-l M-l
boost D + D 2
2 R RVKM(M-I) M R
D 2
2L D R 2 D 2

buck- 2|Vl R 2lVl 2M

buck - 1 Vq D T s boost 0 R
V D
boost D R-VK M 2
R Î K M R
D 2 2L D R 2 D 2

TABLE III. Ve{inition o{ the elements in the canonical

TABLE I. definition of the dc circuit model in Fig. 21 circuit model o{ Fig. 22 {or the three common
{or the three common converter o{ Fig. 1 oper­ converters o{ Fig. 1 operating in the dis­
ating in the discontinuous conduction mode. continuous conduction mode,.

open-loop consideration closed-loop consideration type God

converter Gog
type
M ( D,K ) D (D,K)
2 D ( M,K ) D (M,K)
2

2V(|-Mf' a
2-M I
buck M
2 Κ 2 I / K M 2 Ί/Κ M(2-M) I - M R C
buck K( l - M )
I+YI+4K/D 2

D l*Vl+4K/D 2
V l-M
2V iJKM 2M-I I
boost M
boost l+Vl+4Cp/K Κ l*Vl + 4 D ? / K V K M ( M - I )
l / K M 2M-i \/M-I M-l RC
D 2 VM-I
buck- V 2
boost M
buck - D
VK
RC
1/K MVK
boost
VK
G g
V Ç g Gog i + s / ^ , C V c J a - C d
0 | + s / î J -

TABLE II. Summary o{ dc trans{er properties o{ the three

common conveniens o{ Fig. 1 in the discontin­
uous conduction mode expressed {or open-loop TABLE IV. Summary o{ the ac trans {er properties o{ the
as welt as {or closed-loop considerations. three common converters o{ Fig. 1 operating
in the discontinuous conduction mode.

With use now of the last three columns of Table I

and the procedures outlined in Section 3, the very use­ All the results presented in this section are appli­
ful Table II can be generated, in which the dimensionless cable only to the discontinuous conduction mode of oper­
parameter Κ is defined as before with Κ - 2L/RT - 2Lf /R. ation of these three switching converters. To determine
when these results ought to be applied and when those
g g

for continuous conduction mode [2], the boundary between

The element values of the dynamic (ac) circuit the two modes of operation is determined for these three
model in Fig. 22 for the three converters are shown converters and tabulated in the next section.
in Table III.

Again, as Table II was generated from Table I and

only input-output dc transfer properties obtained, we 6.3 Determination of the boundary between two
can similarly generate from .Table III another, Table IV, conduction modes
in which only input-output ac transfer properties
(transfer functions G and G ,) are listed for the three As explained in detail in Section 3.1 the criteria
vg vd
converters. for determination of the converter conduction mode are

P E S C 77 R E C O R D - 5 1
bounauiy between the two conduction modes them will operate in the discontinuous conduction mode
for a portion of the duty ratio range. With this, and
κ or R = Rcrit (88) the first column in Table II, the dc voltage gain as a
crit
function of duty ratio can be shown as in Fig. 23b for
continuous conduction mode Κ < 4/27, while the corresponding result for continuous
conduction mode is illustrated for comparison purposes
Κ > Κ R < R . _ (89) in Fig. 23a for Κ > 1.
crit or crit
a) c o n t i n u o u s conduction b) discontinuous conduction

discontinuous conduction mode

Κ < Κcrit or R > Rcrit (90)

where Κ is as defined before Κ « 2L/RT » 2Lf /R.

£buck-boost
Following the same procedure outlined In Section 3.1 for
the boost converter example, the parameters K and c r i t

R ^ can easily be found for the other two converters

all resutls are shown tabulated in Table V.

open-loop c o n s i d e r a t i o n c l o s e d - loop c o n s i d e r a t i o n
converter
type
Tig. 23. Comparison of the dc voltage gain chahacten.-
Kcrit (D) RcritlD.R, Kcrit (M) Rcrit(M,Rnom) istics in the two conduction modes fon. the
common conveniens of Tig. 1.
buck I - D I - M In Fig. 23b heavy lines designate the region of
actual discontinuous conduction operation, whereas
dotted lines signify that the continuous conduction
mode takes over and the dc gain characteristics begin
boost D( l-D)
to follow those for the continuous conduction mode
D(l-D) 2
M-l
(see for comparison Fig. 8 ) . From Fig. 23b it is also
evident that in the buck and the buck-boost converter,
buck- «\nom
boost d - D f (M + l f R n o m the transition between the two conduction modes occurs
; I - D I 2
(ΜΗ only once at higher duty ratio D, and not also at the
lower end as it does in the boost converter. There­
TABLE V. Vetenmination of the boundary between the two fore during initial start-up of the converter, when the
conduction modes, expressed fon. open-teop as duty ratio changes from zero to the value required by
well as fon. ctosed-loop considerations. the steady-state gain M, the two converters (buck and
buck-boost) can be designed to stay in the discontinuous
In Table V nominal resistance R is a design para- conduction mode only, even in this transitional period.
nom
meter defined by
We now present another viewpoint, which in an inter­
R = 2Lf (91)
nom s esting pictorial way and a unique frequency interpretation,
illuminates the determination of the converter operating
It has already been demonstrated in Section 3.1 for mode and the basic small switching ripple requirement.
the boost converter that parameter Κ can be chosen Namely, from Fig. 1 it is apparent that the three common
(K > 4/27), such that the converter is always operating converters essentially consist of the single switch S
in the continuous conduction mode regardless of the positioned differently among the source voltage V and
operating point, that is dc duty ratio D, while the three elements, inductance L, capacitance C, and road R.
discontinuous conduction mode can occur only for Κ < 4/27, With only these three elements three different "inherent"
and then only for a portion of the dynamic range of duty frequencies can be defined regardless of the converter
ratio D. The same holds true for the other two conver­ type. Two of them, ω and ω , termed natural frequencies,
ters, and the following criteria can be set: are defined as
1 1 (92)
a) when Κ > the converter is always in con­
2RC /LC
tinuous conduction mode regardless of D.
b) when Κ < discontinuous conduction mode
However, yet another "inherent" frequency can be
can occur,^>ut only for a limited range of
defined by these three elements as
duty ratio D.
R (93)
Parameter I L . is actually the maximum of the duty 3 2L ω S

ratio D dependent function of first column in Table V, The dimensionless parameter Κ, which plays a crucial role
and is for comparison purposes listed in Table VI. in the determination of the conduction mode, can now be
expressed as
buck- f
buck boost boost
Κ = — (94)
4 3 ω

1 1
Km 2.7 Therefore, the position of this new frequency ω with
respect ot the switching frequency f determines the
conduction mode. Hence for R > 1 or ω < f , each of
TABLE VI. Summary of the pahjameten. K , , determining the the three converters will always be in continuous con­
Kegion of unconditional continuous conduction duction mode regardless of D. On the other hand,
fon. thn.ee common conveniens of Fig. 1. ω << f and ω << f are requirements for small switching
From Table VI it is obvious that when Κ > 1 any of rïpple. The information contained in the position of
8

the three converters listed will always operate in the these three "inherent" frequencies ω , ω and f with
continuous conduction mode, and when Κ < 4/27 each of respect to the switching frequency f is concisely sum­
marized in Fig. 24. The diagram in fig. 24, with the

5 2 - P E S C 77 R E C O R D
 I I ρ verter operates in the discontinuous conduction mode
from D - 0 until D = 1-D = 0.72, and the experimental
2

dc gain characteristic is shown in this duty ratio range

high switching ripple on Fig. 25.
*J Ι
small ripple

fs f r e q u e n c y scale
J I l_J —ι ι ι ι I»
IkHz io k H z 100kHz
discontinuous
u>4 c o n d u c t ion 2
conti nuous
conduction

Fig. 24. Frequency interpretation oh the conduction

mode type and smalt smtcklng ripple requirement, ι
help of definitions (92) and (93), displays in a con­
vincing manner the interplay between conduction mode types,
duty ratio D
switching ripple requirement and choice of parameter
values L, C, R and f . For example, increase of load R
y
J L
can cause change to liscontinuous conduction mode with­
out deterioration in switching ripple. However, if in­
ductance L or switching frequency is reduced, change to Fig. 2 5 . Vc voltage gain measurements hor the buck-boost
discontinuous conduction mode can occur, but at the converter in the discontinuous conduction mode.
price of higher switching ripple since separation be­
tween ω and f is also reduced. One would have to in- As seen in Fig. 25, the experimental points follow
c s very closely the theoretical straight line characteristic.
crease capacitance C to remain at an acceptable switching The experimental data, however, are slightly lower than
ripple level. Thus the frequency diagram of Fig. 24 gives the theoretical curve since the transistor saturation vol­
valuable insight, both qualitative and quantitative, tage and diode drop have not been accounted for in the
into the basic relationships inherent to switching con­ theoretical model, although this could easily be accom­
verters. It is interesting that from (92) and (93) a plished. The inductor current waveform was monitored,
very simple relationship follows and confirmed discontinuous conduction operation for
2/ω ω­ (95) D ε[0,0.72] while D„ measured was constant as predicted
ά 3 at D - 0.28. 2
L

which may further facilitate quantitative analysis. Ac trans her hunctlon measurements
6.4 Experimental verification of the transfer properties The duty ratio modulation d to output voltage ν
transfer function G ^ is now measured using the des­
Both dc and ac transfer properties have been exper­ cribing function measurement technique [llj and results
imentally verified on a circuit breadboard of the buck- are shown in Fig. 26.
boost converter shown in Fig. 12c. Because of lack of
space, only cursory experimental verification is included
here.

The buck-boost converter was chosen because of several

unique features which clearly distinguish it form the
other two converters, and which are easy to check. A f s = IOkHz
quick look at Table II, for example, reveals that it is -10
the only converter whose second interval D^T is inde­
pendent of the operating conditions (duty ratio D or
gain Μ ) , but rather is fixed determined by the parameter
Κ only.

Likewise, a look at Table III shows that the ac

resistance r« is also independent of steady-state oper­
ating condition (gain M ) . Therefore, the single pole
of the two transfer funcitons G and G , does not move
with change of operating conditïSn (gain m) as it does
in the other two converters.
100Hz .IkHz f requency

Finally,.the open-loop input impedance of the buck-

2
Fig. 16. Experimental magnitude-irequency response oh
boost converter is R, - R/M since there is no internal G j = 0 / â trans her hwction hor buck-boost
feedback (g_ • 0). Hence the input impedance is purely converter In the discontinuous conduction mode.
resistive, which is not the case for the other two
converters.
The element values used are the same as for the
dc measurements, except that the inductance was in­
The transfer properties have been verified ON the creased four times to L 3.5mH to reduce the superimposed
s

test buck-boost converter with the following switching switching ripple and to reduce the ringing effect in the
components: transistor 2N2880 and diode TRW SVD 100-6. D T interval. Hence for L « 3.5mH, C 12yF, R « 220Ω,
3
35

f = 10kHz, V = 6V we calculate Κ = 1.62 and D = 0.56. 2

Vc gain measurements Tne range of Siscontinuous conduction operation is then

reduced to D ε [0,0.44]. The single pole of the transfer
For the choice of element values L = 890μΗ, C = 12yF, functions G and G ^ (see Table IV) becomes f = l^RC -
R = 220Ω, f = 10kHz and V = 6V we compute Κ = 2Lf /R = 120Hz, whicfi^is in excellent agreement with thi exper­
0.81 and D = »/κ - 0.28. T H E R E F O R E , the buck-boost con-
0
S
imental data shown in Fig. 26.
S

P E S C 77 R E C O R D - 5 3
 The measurements were repeated for several operating The inclusion of the canonical circuit model (Fig.
points in the discontinuous conduction region, namely, for 22) and an appropriate model for the modulator stage (96)
D « 0.1, 0.2, 0.3, and 0,4 but the single pole at f , as into the switching regulator (Fig. 27) results in a com­
predicted, did not move. plete circuit model of a switching regulator in the dis­
continuous conduction mode, as shown in Fig. 28.
The experimental measurements therefore have con­
firmed the high degree of accuracy of the canonical cir­
cuit model (Fig. 22) for the discontinuous conduction
mode of operation.

The question of input properties of switching

converters and regulators, and particularly of open-
and closed-loop input impedances, is thoroughly analyzed
in the next section on modelling of a switching mode
tiegulxxton. in the discontinuous conduction mode.
1
! a Us)
7 MODELLING OF SWITCHING REGULATOR IN
DISCONTINUOUS CONDUCTION MODE
Fig. U. General ac small-signal circuit model ion. ike
This section demonstrates how the canonical circuit Slacking regulaton. oi Fig. 25 operating In the
model for a switching converter operating in the discon­ dis continuous conduction mode.
tinuous conduction mode can easily be incorporated into
the complete switching-mode regulator model. Consider
now a switching-mode regulator as shown in Fig. 27, an The generator symbol for the current generators
illustrative example since the discussion applies to any j,(s)d and j (s)d at the input and output ports, respec­
?

converter. tively, has been changed from a circle to a square to

emphasize that in the closed-loop regulator they have
punregulated input regulated output
become dependent generators (on output voltage variation
ν in particular). A closer look at the circuit model in
switching mode converter | Fig. 28 reveals some unique properties of this negative
feedback circuit. Namely, it has been previously shown
^ΤΠΡ—ι (X- in Section 3 that only the output port network (con­ Ä

sisting of current generators g^v , j«d» resistances

and R and capacitance C) effectivlly takes part in deter­
mination of the open-loop transfer functions G and G
The immediate implication of this is that for X&eal S0Un.ce
voltage ν , the loop gain Τ is defined only with respect
to the output port as shown in Fig.28 · Likewise, the
duty ratio D
output impedance Ζ and line transmission characteristic
F (audio-susceptibîlity) become solely defined in terms
of the output port elements, while the input port takes
modulator
part only in determination of the input impedance Z^.
ampl if ier - This is easily confirmed by analysis of the equivalent
circuit in Fig.28 , which leads to

Fig. 27. Svoitcklng-mode regulato*. Τ - G (s)A(s)f (s)/V

vd m n (97)
Ζ (β)
ep ' v

7.1 Modulator stage modelling and complete regulator (98)

1+T
circuit model
G (s)
So far, we have obtained the canonical circuit model F - (99)
1+T
for the switching-mode converter. The next step in
development of the regulator equivalent circuit is to
obtain a model for the modulator. This is easily done 1 Τ
by writing an expression for the essential function of z7 Ϊ+Τ (100)
the modulator, which is to convert an (analog) control
voltage V to the switch duty ratio D. This expression
can be written D « V /V in which, by definition, V is The first three expressions are rather obvious and
the range of control signal required to sweep the duty are a consequence of the general results of linear feed­
ratio over its full range from 0 to 1. A small variation back theory. They also confirm that Τ, Ζ , and F are
ν superimposed upon V therefore produces a corresponding functions of the output port elements only, since the
variation â ν /V in D, which can be generalized to
s open-loop transfer functions G^ and G ^ are independent
account for a nonuniform frequency response as of input port elements. ^ V V

It should be noted, however, that this peculiar

(96) dependence of some feedback quantities Τ, Ζ , and F on
output port elements only, le a quite Spedil case,^
which is a consequence of the Ideal Source voltage ν .
in which f (0) * 1. Thus, the control voltage to duty If the source voltage had an internal impedance, or
ratio smalT-signal transmission characteristic of the input filter were included in front of the converter,
modulator can be represented in general by the two even the open-loop transfer functions G and G ^ would
parameters V and f (s), regardless of the detailed become dependent on all circuit elementïf the feedback
mechanism by which fhe modulation is achieved. quantities even more so, and this special feature would

54-PESC 77 RECORD
disappear. This once again demonstrates how powerful Again by using element definitions from Table III
these converter equivalent circuit models are, since any in (106) we get for all three converters
of such additional effects can be directly included in
R / _ M\2 2

the circuit model of Fig. 28, owing to its complete, (107)

circuit, representation of the converter properties. in
M \V / 2

We now investigate in more detail the important which correctly predicts the open-loop low-frequency
input properties of the circuit model in Fig. 28· input resistance to be positive.

7.2 Input properties of switching regulators in From these results and the corresponding one for
discontinuous conduction mode continuous conduction mode [l], it follows that the closed-
loop low-frequency input resistance R is given by (104)
As seen in (100) the input impedance is also regardieSS of the conduction mode type and switching
dependent on the input quantities j , r^, and g^. In converter type (buck, boost, or buck-boost). The same is
addition the input duty ratio dependent current gener­ <also true for the open-loop low-frequency input resis­
ator j is now responsible for the negative input impe­
1 tance Ri n given by (107).
dance at low frequencies. Indeed, if j. • 0, and since
at low frequencies T-*», the input resistance R would Hence, this section has confirmed that the canonical
appear to be positive, in obvious conflict witn the circuit model for discontinuous conduction mode (Fig. 28)
actual physical requirement. properly models the regulator input properties (closed-
loop input impedance) in much the same way as the canon­
Let us now verify this for the discontinuous con­ ical circuit model for continuous conduction mode [l,2]
duction mode, and consider first the limiting case of did, through the presence of duty ratio dependent cur­
(100) for high loop gain T-**> (at low frequencies) rent generators at the input of the converter model.
The immediate consequence of this is that the regulator
circuit model (Fig. 28) is a complete, circuit model which
(101) represents all essential properties; input, output and
"vd transfer properties.
From the circuit model in Fig. 28 the converter open-loop
transfer functions G and G . are easily found as 8
vg CONCLUSIONS

A general method for modelling any three-state

vg
g
2 2 N l + sC(rJ|R)
( r R )

switching converter operating in the discontinuous con­

(102) duction mode has been presented. The fundamental step
is in replacement of the state-space descriptions of the
s v d - V r
2 l l > i + sc(r ffiy
R
2
three switched networks (3) by their average (8) over the
single period Τ , the same step as taken for any ordinary
three-state converter. This is then supplemented by
By use of (102) in (101) we finally obtain the additional constraints (9) and (10) which properly
closed-loop incremental resistance R^ as account for the discontinuous conduction mode of operation.

(103) The subsequent perturbation and linearization steps

are applied not only to the state-space or circuit
averaged models but also to the constraints, which then
Using now the definitions of element values j-, j«, provide the additional information needed to define com­
from Table III in (103), we obtain for iUl pletely both dc and ac small-signal models.
D O » and r
g0 > n

tfirree converters (buck, boost and buck-boost) that

2 An extensive analysis of the dc conditions in the
R / V \< discontinuous conduction mode has been given, in Section
(104)3, which then enabled the definition of the boundary
M \V/ 2
between the two operating modes for a specific boost con­
verter example. An easily interprétable formula ((45) or
From (103) it is also evident that despite the (49)) led to simple criteria ((46), (47) and (48)) for
presence of the positive term, the negative term has pre­ determination of the converter, mode of operation.
vailed, correctly predicting the negative closed-loop
input resistance. Analysis of the dynamic (ac small-signal) model
confirmed the general modelling prediction - reduction
Let us now consider the other extreme when the loop of the system order by one. Thus, common converters of
gain is very small, that is T-K) (or equivalently at high Fig. 1 showed a singlerpole frequency response in the
frequencies). Then, the input impedance approaches the discontinuous conduction mode, as opposed to their two-
open-loop input impedance obtained from (100) as pole response in the continuous conduction mode.

Then, following the hybrid modelling path (Section

τ- - g,1 βvg (105)
4) and the circuit averaging path (Section 5), a new
"in circuit model (Fig. 18) with a rather unusual topo­
logical structure is obtained for the boost converter,
The same result could be obtained directly from the open- which provides a complete model for dynamic (ac small-
loop converter model in Fig, 22. From (105) it seems as signal) behavior.
though Z. could be negative owing to this negative in­
ternal efrect of the current generator g-v in the model The canonical circuit model with the same topo­
of Fig. 22. However, this is not true, since the low- logy (Fig. 18), but with different element values, is
frequency value of the open-loop input impedance R., obtained in Section 6 for the other two converters of
in
becomes from (105) Fig. 1, and the results are conveniently summarized in
r, various tables. Experimental verification of dc and ac
in 1 - l l 2 2 l R) (106) transfer properties of a buck-boost converter in dis­
g r g (r
continuous conduction mode are also provided.

PESC 77 RECORD-55
 Finally, the model of the switching-mode KcgulatOK [l2] Slobodan Cuk and R. D. Middlebrook, "A New Optimum
operating in the discontinuous conduction mode is ob­ Topology Switching Dc-to-Dc Converter," IEEE Power
tained in Section 7, and important input properties Electronics Specialists Conference, Palo Alto, Calif.,
(both open- and closed-loop) are thoroughly analyzed. June 14-16, 1977.
[13] Slobodan Cuk and R. D. Middlebrook, "Coupled-Inductor
The outlined method is general and dUiectiy appli­ and Other Extensions of a New Optimum Topology
cable to investigation of the discontinuous conduction Switching Dc-to-Dc Converter, "IEEE Industry Appli­
mode in more complex switching converter structures,
cations Society Annual Meeting, Los Angeles, Calif.,
such as those described in [12,13], involving more than
a single inductor. Oct. 2-6, 1977.

REFERENCES APPENDIX A

STATE-SPACE AVERAGING STEP EXTENDED TO CONVERTERS WITH

[l] Slobodan Cuk, "Modelling, Analysis, and Design of MULTISTRUCTURAL (THREE OR MORE) TOPOLOGICAL CHANGES
Switching Converters," Ph.D. thesis, California
Institute of Technology, November 1976. We derive the state-space averaging step for
switching converters characterized by thKCe structural
[2] R. D. Middlebrook and S. Cuk, "A General Unified
changes within each switching period. Each topological
Approach to Modelling Switching-Converter Power
Stages," IEEE Power Electronics Specialists Con­ structure can be described as before by linear state-
ference, 1976 Record, pp. 18-34, (IEEE Publication space equations, hence
76 CHO 1084-3AES). (Also to be published in
International Journal of Electronics, London). A x_ + b_v for interval d Τ ,
n (0
11 1 g 1 s'
[3] R. D. Middlebrook and S. Cuk, "Modelling and
Analysis Methods for Dc-to-Dc Switching Converters," A x + b v
2 2 2 g for interval ^ 2 · Τ
(A.l)
IEEE International Semiconductor Power Converter 8

Conference, 1977 Record, pp. 90-111 (IEEE Pub­

lication 77 CH 1183-31A). A x_ + b.v
0
3 3 3 g
for interval d T ,
J s 0 (t 2 - t - τ ) 8

[4] R. Prajoux, J. C. Marpinard, and J. Jalade,

"Accurate Mathematical Modelling of PWM Power
Regulators," Translation of Report, Laboratoire 7Wo boundary conditions are now imposed. Since the
d'Automatique et d'Analyse des Systems, C.N.R.S., state-space vector is continuous in transition from first
issued under ESTEC Contract No. 2026 AK, May 74, to second and from second to third regions,
25 pp.
[5] A. Capel, J. G. Ferrante, and R. Prajoux,
x
2^1* * l^l^
x

"Stability Analysis of PWM Controlled DC/DC Regu­ (A.2)

lator with DC and AC Feedback Loops," IEEE Power x (t )
3 2
x (t )
2 2

Electronics Specialists Conference, 1974 Record,

pp. 246-254, (IEEE Publication 74 CHO 863-1-AES).
Solution of (A.l) under the small signal assumption for
[6] A. Capel, J. G. Ferrante, and R. Prajoux, "State ν (where ν - V +v and ν « V ) yields
Variable Stability Analysis of Multiloop PWM g g g g g g
Controlled Regulators in Light and Heavy Mode," At
IEEE Power Electronics Specialists Conference, x (t)
x e (0) + v B (t)b
1
X]L
for t e [0,^] g 1 1

1975 Record, pp. 91-102 (IEEE Publication 75 CHO

965-4-AES.) . A (t-t )
x (t) = e
2
x ( t ) + v^ (t-tfl> 2 1 2 2 for t ε [ t t ] (A. 3) r 2
[7] H. A. Owen, A. Capel, and J. G. Ferrante, "Si­
mulation and Analysis Methods for Sampled Power A (t-t )
3 2
Electronic Systems," IEEE Power Electronics Spe­ x (t)
cialists Conference, 1976 Record, pp. 45-55
3 e J
x ( t ) + v B (t-t )b 2 2 g 3 2 3 for t ε [t ,T ] 2 s

(IEEE Publication 76 CH 1084-AES).

J
where
[8] F. C. Lee, Y. Yu, and J. E. Triner. "Modelling of
Switching Regulator Power Stages with and without B (t) - e άτ i - 1,2,3 (Α. 4)
±
Zero-inductor Current Dwell Time," IEEE Power
ο
Electronics Specialists Conference, 1976 Record,
pp. 62-72, (IEEE Publication 76 CH 1084-3AES). Use of boundary conditions (A.2) in (A.3) gives
[9] A. Capel, J. G. Ferrante, and R. Prajoux, "Dynamic
Behavior and Ζ Transform Stability Analysis for 33-3*.
3 W e . W e A d T

e e x (0) +
DC/DC Regulators with a Nonlinear PWM Controlled *3<V 1

Loop," IEEE Power Electronics Specialists Conference, A_d,T A d T 0 9

1973 Record, pp. 149-157, (IEEE Publication CHO v [e °e B (d T )b + g
3 3 2 2 8
1 1 s 1 +
787-2AES).
[lO] M. Clique-and A. J. Fossard, "Development et Appli­ ο 3 s
cations de Modèles Numériques pour l'Evaluation des + e B (d T )b + B,(d,T )b,] (A.5)
2 2 s 2 U J
3V 3 e

Systems de Puissance," 1er Rapport d'Advancement,

April 1976, Contract ESTEC no. 2590/75AK. With introduction of the linear approximations
[il] R. D. Middlebrook, "Measurement of Loop Gain in A.d.T
Feedback Systems," International Journal of Elec- lis
* I + A.d.T , i - 1,2,3 (A.6)
tronics, 1975, vol. 38, no. 4, pp. 485-512. lis

5 6 - P E S C 77 R E C O R D
into (Α.4) and (Α.5), and after retention of only first- a) S i on ySz off : b) S | on , S 2 on :
order terms (linear in T ) , (A.5) reduces to g L, L 2

x ( T ) = (I+d A +d A +d A )x (0) + (
3 s 1 1 2 2 3 3 1
d
1
b
1
+
d 2
b
2 + d
3 3 b ) v
g ( A
' 7)

This leads to a single continuous linear system

A = A d l x + d A + d A
2 2 3 3 Λ cl S i off , S 2 on
χ = Ax + bv where (A. 8)
g b Ä d i b i + d b 2 2 + d b 3 3
'• •') L, L 2

ι—«ΊΠΓ»—/~W^
It remains, finally, to characterize the state-
space averaging step for the generalized switching con­
verter with η structural changes within each switching
period, namely, one described by

d.T = t -t,_ 1 Fig. Α. 2 \Jάλλους switched networks fan the conventeK in

i - A,x + b.v , _ \ i«l,2,,.. n (A.9) Flg. Α.Ja.
S
s

1 1 8
' « [ t ^ l
On the other hand if the converter is looked upon
for which the corresponding basic state-space averaged
as consisting of cascaded boost and buck converters and
model is each of them has been modelled separately as a "two-state"
converter as in [2], and their models put together, the
Σ
d
l i
A

same result would have been obtained.

i=l (Α.10)
χ - Ax + bv ; η However, for the discontinuous conduction mode, in
g addition to the state-space averaging step (A.8) for
Zii d b
"three-state" converters, other restrictions ((1), (2))
i-1 are imposed to reflect the limited behavior of inductor
current (Fig. 2b) with fixed (zero) boundary values.
As an illustration of a switching converter with But in any case, for either continuous conduction
such multistructural change, consider the converter mode [2], or discontinuous conduction mode, the corre­
shown in Fig. A.la whose two switches S. and S are 0
sponding state-space averaging step is justified on the
driven as specified in Fig. A.lb. The two switches basis of the fundamental performance requirement for
and S are shown in their "on" position in Fig. A.la.
2
switching dc-to-dc converters of small (negligible)
It can easily be recognized that this converter is switching ripple, as follows:
actually a boost converter cascaded by a buck converter
whose switches are driven synchronously but with dif­ switching j - j v natural switching fundamental
ferent duty ratios, d and d^+d^ respectively. ripple small frequencies frequency matrices

a) ^ linear state-space
approximation"*^ averaging step

Vg^r This, together with proper inclusion of the inductor

current discontinuity as additional constraints (1), (2),
enable the extremely simple, powerful and accurate scheme
for modelling and analysis of switching converters in
b) switches c) switches
discontinuous conduction mode to be established.

s 2
dTs
r
~r—
s,1

Si
off
Γ
Off
I F "
|d.d iTs2

Fig. A.I Switching conveniez exhibiting multistniictuAaZ

change: a) boost conventex cascaded by a buck
convehten; b) switch dnive fax. "thiee-state"
behavloK; c) switch dfilve fax. "faunstote"
behavloi.
However, if this converter is looked upon as single
system, the switching action of Fig. A.lb would produce
periodic sequential change among three different struc­
tures (shown in Fig. A.2b,c, and d ) , while that of Fig.
A.l.c would produce periodic sequential change among all
four different switched networks of Fig. A.2. In any
case, it demonstrates the .feasibility of realization of
a switching converter having three or more switched net­
work configurations, even In the continuous conduction
mode of operation.

P E S C 77 RECORO-57