LARGE-SIGNAL MODELLING AND ANALYSIS

OF SWITCHING REGULATORS
ROBERT W. ERICKSON t SLOBODAN CUK AND R.D.MIDDLEBROOK
California Institute of Technology
Pasadena, California

ABSTRACT
A I~ge-~ignal ~witehing ~egutato~ model ~
de.JU..v ed, and P'Lomine.YLt ne.atuJlU On the btaYl~ient
nes pon6 e Me de.teJlmine.d. 1n paJt:tieu1.aJt,
ana.lytic.al e.xpJLM~iOn6 Me 60und 6o~ the
e.quilibJtium po-int.6 06 the sus tes« which yield
in6ight into the .eaJtge-~ignai. ~e~pon6e, and
c.ompute.Jt-ge.ne.Jtate.d ~aYL6.ie.nt wave.6oJrm6 Me.
obtained.
~ an example., a boo~t ~egulato~ i~
invutigate.d, and ~ nound to be ~table 60Jt -6mall
~ignaL6 but u~ta.ble 60Jt lange oians cenxs .

1. INTRODUCTION

Switching regulators are inherently nonlinear.

As a result, it is very difficult to design a
stable feedback loop using exact methods since the
resulting differential equations cannot generally F.ig. 1. HtJPothUic.al :tJta j ecto.u.es 06 a -6wUc.hing
be solved. Instead, small-signal methods are ~e.gula.to~. InduetoJt c.uJlJLe.nt i L i~
commonly used, where one linearizes the regulator ploUed vs , caoaccta»: voltage v e- A
model about a quiescent operating point. These ~maU-~ignal model WM used to dUign a
models are very useful to the practicing engineer ~table. 6e.edbac.k loop; indeed, tJta~ie.nt
since he may apply all of the relatively simple -6olu:Uo~ within a JtacUU6 r 06 the
techniques of linear circuit theory such as Bode quiuc.e.nt opeJLa-ting point (VO IO)
II

plots, root locus, etc. The physical insight c.onve.Jtge a.6 des oied. Howe.veA, tJtaru,ient
gained then allows the engineer to intelligently -6 ofuti.o~ ou.tscde: thiA Jtange may be
design his feedback loop and to specify important un6ta.ble.
small-signal specifications such as audio
susceptibility and output impedance.
A regulator which exhibits 'the hypothetical
Unfortunately, because of the small-signal behavior illustrated in Fig. 1 obviously is
approximation, these methods do not ensure the unreliable and unacceptable. The investigation
complete large-signal stability of the quiescent of the effects of switching regulator nonlinearities
operating point. One might conceive of a regulator and the exposition of some techniques for the
which behaves as illustrated in Fig. 1. For small avoidance of large-signal instabilities are the
perturbations, less than some radius r from the subjects of this presentation.
quiescent operating point, the regulator behaves
as predicted by the small-signal model, and 1.1 The Large-Signal Stability Problem:
transients converge as expected. However, for An Example
large perturbations, the nonlinear terms become
significant, and some solutions do not converge Before embarking on a large-signal analysis,
to the desired quiescent point (i.e., some it is necessary before to determine whether the
solutions are unstable). Other large transients effect of the converter nonlinearity is significant.
do converge, but with a large, distorted waveform Is it possible to design a regulator whose response
which may be much larger than predicted by the to large perturbations deviates substantially from
small-signal model and hence unacceptable. ~he response predicted by the small-signal model?

This work was sponsored by the Office of Naval Research, Washington, DC, under contract N00014-78-C-0757.

240
0275-9306/82/0000-0240$00.75 ~ 1982 IEEE
 D(t)
.8A

—<2H
PWM f 2 = .5V
V
_1

f=50kHz
8

ro to
f3=300 integrator lb)

Fig. 2. Three-state boost regulator example.

Even worse, is it possible for the response to

be reasonably stable for small signals but diverge
for large transients? An example is given in this
section which demonstrates that the answer to the
above questions is ye.S; large-signal phenomena
exist which can seriously degrade the performance
of a switching regulator, and these phenomena are
not predicted by small-signal models.

Consider the boost regulator shown in

Fig. 2. The dc gain and efficiency curves are Fig, 3. Vc analyst* of the boost example- [a]
shown in Fig. 3. This regulator was designed to Equilibrium gain M = V / V ; ( b )
c g

operate at a nominal duty ratio of .6 with an Equilibrium efficiency n.

efficiency of 70.5%. The state-space-averaged
small-signal model [l,2,5] is shown in Fig. A.
To stabilize the system in the presence of the l:M(D)

4>
right-half-plane zero which appears in the duty-
ratio-to-output transfer function, the inductor e(s)d
current is fed back in addition to the output
voltage. The integral of the output voltage is (T) j(s)d

also fed back to improve regulation. For the

values chosen, the closed-loop small-signal
response is dominated by two poles with a Q of
.6; thus, the system appears to be quite stable
and nearly critically damped. A computer program,
detailed in Fig. 12 later in this paper, was used d (s)
to investigate the large-signal response of this f!-.8A
regulator. The computer-predicted response to a
small step change in inductor current and
capacitor voltage is diagrammed in Fig. 5. It is
f ».5V

r
_1
indeed well-behaved; the waveforms appear linear, 2
and a very small amount of overshoot occurs in the
control (d) waveform. The small-signal model is
an excellent approximation in this case.

The response to a step change of

f =300
3 1/8 H
intermediate proportions is shown in Fig. 6. The
solid line is the nonlinear response, and the Fig. 4. Small-signal state-space averaged model
dotted line is the response predicted by the used to design the feedback loop of the
small-signal model. Although this response is three-state boost regulator example.

241
 '2~::::::::t._--..................-
.1
2'B~
..._..--.. .-
...................- ....................
B.B.
1.8~
-.1
B.B r" . -T-- • - _. . . . . . ._~_=~I=:::=_
-.2
t nductor current 1nduc to r cu r r.nt

'2~
~::b~'
.1

B.B .......Iiiiii;;;*-.............
- . - -..........iiiiiiiti~~iliiiiiiiiiii&ii;;;l;;;
-l.B /'
-.1
-2.B
-.2
capac t tor va 1tag. -3.B capacttor voltag.

---- ----
~
.8828
.8881
.8882
.eeee~
.B888 -===+=::=:~:::::::::~=+=~~=~~~~~~
-.8881 -.882B

-r'
-.8882
t nt.grator vo 1 tag. -.8848 tntegrator voltage

::~
._ ~i=1~~_ ..................-...-.... ~ ........_.--.......
-:58
Tm1 n- 0.0 m••c Tmax- 2.0 m•• c Tmt n- 0.0 m•• c Tmax- 7.5 maec

F-<-g. 5. Computen-geneAated time domain Lvave6~n~ Fig. 6. Computen-genenated wave6o~ 01 the

06 the nupol16e. to a .6maU peJLtunbatA.-on fLe.6 pon.6 e. to a pefttunbatio Yl. 06 -intVtfr!e.cLi..a.te
Ln c.apacdon voUag e v and Lnduero« pnopo!l..tion.6. Va.6hed line: a.6 r:ne.cli;c.te.d
c.uMent i. The. .6mall-.6~nai. appnoxhna- by the .6mai.l-.6-igna£ mode.l. SaUd li~e:
lion ~ jU.6un-ied -in thAA cas e , and the a6 p!Lec.kc.ted by the. c.omplete iaJl.ge.--6-<-gYlal
!teA pon.6 e. c.onve.nge.6 a.6 pnecU.c.ted. Ylon1,.[neaJt model de..6c.Jtibed in Fig. 1.2.
Peak voltage and c.unnent valuu one.
.6ign-in-icantly lange.1t tha~ the amou~
stable, it is decidedly nonlinear. The peak pne.cU.-cted by the .6maLe-~..{.gnal model.

K
inductor current is larger than the amount 6.B
predicted by the small-signal model, and the 4.8
control (J) waveform overshoots its nominal value
much more than expected. The small-signal 2.8
8.8 ~ I~_ _. . -_ _....._
approximation is unjustified in this case. - ......... az:::4z
......

1nductol" curl"ent

5.8!
The response to a yet larger step change in
inductor current and capacitor voltage is shown B.8 ........ •
in Fig. 7. The system is unstable in this case! -5.B
-18.8
The voltages and currents in the system become
-15.B
large, and the control signal saturates at its

_========::==--.. . . -
-2B.8
maximum value. Obviously, this design is cap_c t tal" vo 1 t_g8

unreliable.

In this example, the small-signal

.elS
B.Be
-.elS
I - _......._ ....

approximation is unjustified for moderate or large -.18

perturbations. Hence, even though the small-signal -.15
tntegl"atol" volt_ge
model predicts that the system response is well-
damped, the effect of the nonlinearity can be of

_::~
significance, causing larger overshoots than
expected and possibly even instabilities.

1.2 Outline of Discussion Tlftt n- 0. e ..ee dCt) Tlftax- 7.5 .... c

From the above example, it is apparent that

Fig. 7. ComputeJt-geneJLated WaVe6oJUn6 06 the
small-signal switching regulator models do not Jte.6 pon6 e to a i.aJr.9 e peJttuJr..ba:tio n, Vab hed
necessarily include all features of importance. Une: a6 pJtedided blJ the -6ma1i.-.6-ignai.
It is of interest, therefore, to model these model. Solid line: a6 pltecUeted by
additional large-signal effects. In Section 2, the l.aJtge-.6ign.ai. nonli.neaJr. model. The
a nonlinear model is derived which predicts the .61J.6 tem .<A ul16 tab le .
behavior described in the boost example above. Next, the saturation of the pulse-width
Two versions of this model are described. First, modulator is accounted for. This phenomenon has
a discrete-time model is derived which is a significant effect on the large-signal response
well-suited for computer simulations. Second, a of the regulator, and hence cannot be ignored.
continuous-time model is found which is sometimes Models are derived which describe the response of
more convenient for analytical calculations. the regulator in the saturated regions.

242
 In Section 3, the implications of these models where D = duty ratio, D 1
= 1 - D.
are examined. First, the equilibrium points of
the system are calculated. The presence of real T g = time of one complete switching period.
equilibrium points in addition to the desired
quiescent operating point indicates the existence K is a matrix usually containing values of
of unstable transient solutions; therefore, these inductance and capacitance.
additional real equilibrium points must be
eliminated. Next, the trajectories or transient x is a state vector, usually comprised of
waveforms are calculated. Peak transient current inductor currents and capacitor voltages.
and voltage levels may then be determined, and the
existence or absence of unstable solutions u is a vector of independent sources.
verified.
These equations may be solved exactly;
The entire procedure is summarized in however, the subsequent analysis is greatly
Section 4. simplified if the "linear-ripple approximation"
[1,2,3,4,5] is made. Specifically, if the natural
frequencies co^ of the converter power stage are
2. LARGE-SIGNAL MODELLING all well below the switching frequency, then
Eqs. (1) and (2) above have approximately linear
In this section, the nonlinear state solutions. This is indeed the case in well-
equations which describe switching regulators designed converters, in which the switching
operating in the continuous conduction mode are ripple is small. In this case, only the terms to
derived, and the regions of their validity are order (u)^T ) need be considered; higher-order terms
s

identified. Two versions of the model are useful. are negligible.

First, a discrete-time model is found which is
suitable for computer simulation. Second, an The result of this approximation is the
additional approximation is made which yields a following solutions for x(t) during the two
continuous-time model; this model is often more switched intervals:
convenient for analytical calculations. Both
models incorporate the key "linear-ripple x(t) = (I + A t) x(0) + B^t ± (3)
approximation" used in previous small-signal
methods [1,2,3,4,5]. However, no small-signal during 0 < t < DT
assumption is made; as a result, the method is
valid for large signals.
0
— — s

Next, the saturation characteristics of the x(t) = (I + A (t-DT ) x(DT ) + B u (t-DT ) (4)
2 g g 2 g

pulse-width modulator are accounted for. The duty

ratio is always restricted to lie in a range no during D T <_ t <_ T g g

greater than [0,l], and the effect of this Combination of Eqs. (3) and (4), and elimination
restriction on the stability of the regulator is so of the second-order terms which appear, yields the
profound that no large-signal analysis can ignore following expression for x(T ): g

it. The result of this section is the formulation

of basic state equations which model the switching x(T ) = [i + T (DA +D A )]x(0)+T (DB +D B )u
g 1
,
2 g ]L
f
2

regulator and which are used in the subsequent

sections to determine stability regions and peak (5)
transient component stress levels.
Furthermore, after n+1 switching intervals,
2.1 Unsaturated Region Model
x((n l)V =+ - [I+VD^+D^)]^
The first step is to write the state equations
of the regulator system during the two switched
intervals D T and D T . During each interval, the
S
f
g + T [D B D;B ]u
s n 1+ 2 (6)
system may be described by a set of linear
differential equations:
This is the basic difference equation which
during interval D T S (transistor ON) describes the response of the system. It contains
a constant term; hence, the system is not in
dx(t) equilibrium at the origin. It is convenient in
K — = A, x(t) + B.u (1) the analysis which follows to redefine the axes
so that the quiescent operating point of the
regulator is at the origin. The quiescent operating
during interval D'T S (transistor OFF) point occurs at:
x
_ ^ = x

—n
=

—0 D n+1
^ = D n= D 0 (7)
dx(t) n

K — = A x(t) + B u
2 2 (2)

243
Insertion of Eq. (7) into Eq. (6) and solution for
dx(t) X
yields
(12)
^ = - (D^+D^r 1
(D B D'B )u
0 1+ 2 (8)
dt

This approximation is valid if all natural

The axes may now be shifted by use of the frequencies of the closed-loop regulator are much
substitution below. less than the switching frequency. With this
approximation, Eq. (10) becomes
(9)
—0 --n n 0 n
dx(t)
= (D A +Dj A )x(t) + [(A -A )X
Q 1 ) 2 1 2 () + ( B ^ B ^ u J d (t)
dt

4- d(t) (A -A )x(t)
x 2 (13)
+
WW4 (10)

with d(t) = - f x(t)

Eq. (10) arises from the substitution of
Eq. (9) into Eq. (6), and is the difference This is a continuous-time state equation
equation which describes the regulator while which describes the regulator while operating in
operating in the normal, unsaturated mode. No the normal, unsaturated mode. It can be used to
small-signal assumption has been made; hence, infer the nature of the response to large
Eq. (10) is valid for large signals. perturbations and the existence of instabilities.
Owing to the presence of the d(A-^-A2)x term, this
The duty ratio is usually a linear function state equation contains quadratic nonlinear terms.
of the regulator states. In this case,
Thus, the equations which describe the
_T.
d = f x (ID response of switching regulators during large
n
transients have been found for the case where the
pulse-width modulator is unsaturated and the
where f_ = feedback gain vector. system operates in its usual mode. The linear-
ripple approximation was made; this has the
As a result of the linear dependence of d desirable effect of simplifying the analysis while
on x , the x _ u (A-
v " i -A )x term in Eq. (10) is
i L
-
2
ignoring the usually negligible effect of the high
—n s —n l ^ —n frequency switching ripple. A set of discrete
quadratic, and the difference equation is
nonlinear. This nonlinearity can seriously state equations with quadratic nonlinearities is
degrade the transient response of the regulator, the result; these equations are well-suited for
possibly causing instabilities. computer evaluation of the large-signal transient
response. An additional approximation may be made
The discrete form of Eqs. (10) and (11) makes which yields a continuous-time version of the state
them ideally suited for evaluation by computer. equations. This is often desirable for analytical
It is a simple matter to implement these equations design.
on a small desktop computer and obtain the large-
signal transient response of any switching 2.2 Saturated Region Models
regulator operating in the unsaturated region.
In view of the difficulty of obtaining a closed- So far, it has been assumed that the regulator
form analytical solution to Eq. (10) under always operates in its usual unsaturated mode.
transient conditions, this is an attractive No account has yet been made of the inherent limits
alternative. The time domain waveforms illustrated on the duty ratio: the duty ratio must always lie
in this paper were obtained in this manner. within the range [0,l]; often, the limit is even
more restrictive. Outside the usual operating
range, the duty ratio remains constant at its
saturated value, and the regulator behaves as an
open-loop linear system. This can significantly
Although the discrete equations above are affect the large-signal stability of the regulator;
well-suited for digital computer evaluation, they in consequence, these additional modes of operation
are sometimes inconvenient for analytical cannot be ignored. Some aspects of this effect
calculations. It is then preferred to obtain a have been previously considered for the buck
continuous-time model which contains familiar R s , f
regulator [6,7]; the phenomenon is further
L s , C s , and nonlinear devices, and hence yields
? ?
investigated here.
physical insight into the design problem. This
has previously been accomplished for the small- As an example, consider the two-state boost
signal case [l], and the same technique is regulator of Fig. 8. In order to stabilize the
applicable here. In particular, one uses the system in the presence of the right-half plane
Euler forward-differencing approximation to zero which appears in the duty-ratio-to-output
estimate the continuous-time derivative of the transfer function, the inductor current is fed
state vector, as below: back in addition to the output voltage. The

244
15V

D=Dmin
;
D(t)

D=D -f 0 x

D=Dmax

PWM
f =50kHz
- ( L f =.5V 2

V 0

D = .6
0

Fig. 8. Twos tote boost regulator example.

Fig. 9. The saturated and unsaturated regions
of the state plane. IQ and V Q are the
expression for the control signal (duty ratio) is quiescent Inductor current and
therefore capaclton voltage.

D(t) DQ + d(t)

d(t) - - f l i L - f.v. = f X (14) The continuous-time state equations are

again found by use of the forward-differencing
approximation. The result for the D = D_
where D^ = quiescent duty ratio saturated region is sat
f^ = current feedback gain
f 9 = voltage feedback gain dx(t)

dt
If D(t) is limited to the range [ D , ^max]
ax M I N
D
m
t h e n

the unsaturated region is the section of the phase

where (17)
+
C sat l
D B + D
sat 2^ B

< D (15) This is a system of linear differential

D
o"i equations which describes the response in the
To the left of this region, the duty ratio is fixed D = D saturated region. Since the systems
s a t

and equal to D^ and to the right of this region, described by Eq. (16) and (17) are linear their
the duty ratio is fixed and equal to Dmin. The solutions are exponential in nature. Furthermore,
situation is illustrated in Fig. 9. they contain exactly one equilibrium point. This
point occurs at
The state equations in the saturated regions
are easily found. When -^ x is greater than T

D - D Q , then D(t) is constant and equal to Dmax*

m a x *0 ^sat*! +
v'sat^' 1
* sat l
[D B + D
sat 2^
B

Likewise, D(t) is constant and equal to Dmin when

-f x is less than D ± - D Q . Substitution of
T
m n
(18)
D = D into Eq. (6) yields
This equilibrium point may or may not occur within
n s a t

i i
the saturated region.
x - - [i + T (D «_A_+D A )]x + T [D B-+D Bju
-n+1 s sat 1 sat 2 -n s sat 1 sat 2 — 0

Thus, three modes of operation have been

identified. This change of operating mode occurs
where D = either D or D (16) owing to the saturation of the pulse-width
sat max mm modulation process. The regions in the state
plane in which each mode occurs have been found,
1 - D
sat sat and the relevant state equations have been
derived. In the next section, this knowledge is
This is a system of linear difference equations applied to the study of the large-signal transient
which describe the response of the system in the waveforms and instabilities which are likely to
saturated regions. occur in switching regulators.

245
3, LARGE-SIGNAL ANALYSIS A is the small-signal continuous-time closed-loop
system matrix, and B is a matrix which describes
The objective of this section is the the nonlinear term. Equation (20) describes the
construction of the system response and the positions of the equilibrium points in the un­
identification of sources of potential instability. saturated region. In addition to the trivial
First, the equilibrium points of the system ar- solution x* = () (the desired quiescent operating
calculated. It is possible for more than one point), a number of extra solutions may exist.
equilibrium point to exist, and this can lead to One may easily find these other equilibrium points
instability under large transient conditions. by first solving Eq. (21)for d*, the value of the
Second, the trajectories of the states of the control (d) at x*.
regulator are determined, either by hand or by
computer. The salient features are identified, det [A + d*B] = 0 (21)
and it then becomes apparent how to modify the
system in order to obtain an acceptable response. Once the values of d* are known, the solution
The analysis is demonstrated on the boost of Eq. (20) for x* is straightforward. One can
regulator of Fig. 8. then see how to design the regulator such that
these additional equilibrium points lie
3.1 Equilibrium Points sufficiently far outside of the unsaturated region.
Their influence on the transient response of the
The equilibrium points of a system are the regulator can then be made small.
most prominent features of the state-plane
portrait of a nonlinear system. Calculation of
the equilibrium points is a useful tool for pin­ For the boost example of Fig. 8, the
pointing the source of instabilities and for quantities in Eq. (20) are
constructing trajectories.
i* -R x -n x

Two types of equilibrium points can exist for

a given region of the state plane. The first x* = A =
type, known as a "real equilibrium point," is an v* n 2 -1/R 2

equilibrium point which occurs inside the given

region. The presence of real equilibrium points
in addition to the desired quiescent operating
point guarantees the presence of at least one
unstable solution. Hence, such points must be
avoided. The second type is called a "virtual f_ = B =
equilibrium point" and occurs outside the given
region. The system is not actually in equilibrium f J 2 - 1 0
at a virtual equilibrium point; nonetheless, these (22)
points can influence the transient response of t
the regulator. The presence of virtual equilibrium where R.,
points guarantees neither the existence nor the *L +
Vl l "0 V
n D
+
2
absence of unstable transient responses.
+ R = R (-I f )
n
2 - 0 D
V1 2 Q 2

Equilibrium points occur where

IQ = quiescent inductor current
X* . = X * (discrete) (19a)
~TI+1 —n
V Q = quiescent output voltage
dx*
= 0 (continuous) (19b)
Eq. (21), the expression for the control (d*) at
dt
the equilibrium points, then becomes
Since the Euler forward differencing approximation,
Eq. (12), has been used, Eqs. (19a) and (19b)
are equivalent. d*(n + n ) + L 2 ^ + 5 ^ (23)

Insertion of Eq. (19b) into Eq. (13) yields This equation is quadratic and has two roots;
hence, two equilibrium points may exist for the
0 = [A + d*B]x* (20) unsaturated region in addition to the quiescent
operating point _x = 0. For the values specified
where A = + - (A^A^X^ 1
- (B^B^u f T
in Fig. 8, the two roots are

.236, 13.64 (24)

B = A -Ax 2

The equilibrium point at d* - .236 is well within

d* = - fV the unsaturated region. It causes the large-signal
transient response to differ significantly from
the value of x at the desired that predicted by small-signal models; in fact,
equilibrium point. some solutions are unstable.

246
 The root d* = 13.64 represents a virtual The next step is the investigation of the
equilibrium point. It lies outside the unsaturated equilibrium points of the saturated regions. Since
region where Eq. (23) is valid; in consequence, in these regions the response is linear, exactly
the system' is not actually in equilibrium at this one equilibrium point occurs for each region. If
point. Nonetheless, it is possible for a virtual either of these points is real, then unstable
equilibrium point to influence the response of a responses exist. Therefore, the equilibrium
nonlinear system. In this case, however, the points of both saturated regions must be virtual
distance it lies away from the unsaturated region in a globally-stable regulator.
is sufficiently large that its effect on the
trajectories is negligible. The equilibrium point for the D = Dmax
saturated region is given by Eq. (18); a similar
The positions in the state plane of the expression can be found for the D = Dmin
equilibrium points are now found by solution of saturated region. This equation is now solved to
Eq. (20). For this example, one obtains find the positions of the saturated region
equilibrium points. For the boost example, one
obtains
i*
(25)
V
sat
V
-...L-
, • --_..
1
_-
'2
(27)
D l+RL/DsatR
sat
v* f
l
£2 + R (d*-n 1)
1 sat
V
For the values specified in Fig. 8, these I
sat
-,-- (28)
expressions yield D R
sat
1~* 2.0A, v"'* -3.7V at d* .236 where
(26)
1~* 1.9A, ~* -24.2V at d* 13.64 V equilibrium output voltage for D D
sat sat
The unsaturated region equilibrium points saturated region
are summarized in Fig. 10. Two real equilibrium
points exist: in addition to the desired I equilibrium inductor current for
quiescent operating point at x
= 0, an unwanted
sat
real equilibrium point occurs-as shown; hence, D = D saturated region.
sa t
unstable responses are expected to occur. A
virtual equilibrium point also occurs, but has One possible design strategy is to limit the
little effect on the response of the system. duty ratio to a range sufficiently smaller than
[O,lJ, thereby improving the response by moving the
saturated region equilibrium points well outside
of their respective regions. Hence, it is of
quiescent interest to determine the locus of Vs a t and I s a t in
operating the state plane for various values of Ds a t .
undesirable
real ----\-~.. point Combination of Eqs. (27) and (28) yields
eQuilibrium
point 12
sat
R~
-L
- I
sat
RV
g
+ V2
sat
o (29)

Thus, the locus is an ellipse. Eq. (29) is plotted

in Fig. 11. It can be seen that the D = 0 point
is well outside of the D = Dmin saturated region;
consequently, the choice of Dmin = 0 poses no
apparent problems. However, the D = 1 point lies
inside the D = Dma x saturated region for the case
illustrated. As a result, unstable solutions are
virtual guaranteed to exist for the choice of Dmax = 1 .
•~ f-- eQ~ilibrium
pornt
One way to avoid obtaining a real equilibrium
point in the D = Dmax saturated region is to choose
Dma x sufficiently less than one, thereby moving
the equilibrium point outside the region. A
ELg. 10. Summevty 06 the unsatiuuued fte.g-i.on second way is to lower the ratio f2/fl (i.e.,
e.quifJ.bte"<UYn po-i~ 60te the ocos : e.xampie increase the amount of current feedback in relation
-i.n Fig. 8. 1Yl addition. to the lL6ua.i to the output voltage feedback). This moves the
quie-6cent ope~ating po~nt, a neal D = Dmax saturated region boundary to the left,
equiUbteA.URl po.i nt ex.i.oa at d * = .236. past the equilibrium point. Both strategies are
A (-iYL6-i.gn--i.6-i.c.ant) v-!!L:tual e.quA.tibJUum effective in eliminating the presence of unstable
point occun-6 at d* = 13.64. transient solutions.

247
 Vg
1\ 0=1
State Plane Time Doeain
Tra1ectories Waveforms

plot !" plot i

-fl
VS. nT
S

yes

VgR
R+RL
D • Ds a t Region
06
Unsaturated Region
F~9. 11. The.
iOC.U6 ~a;twutted Jr..eglon equiubJtium
~+l D~A;z)] ~ ~+l [1+Ts(Dsat~ + D'satA2)](~ ~)
po~~ ~n the ~tate plane ~~ a po~,on • [I + Ts(DOA 1 + • +

o 6 an eli.ip~ e• The ~ aiunated Jr..egi on + Ts[ (AI-A2) ~ + 0'1 -82 )~] do + Ts[Ds a t Bl + D~at B2] ~ - ~
boundan-i.~ aJr..e ~upeJr..impo~ed. The + T d
s n(A I-A;z)
~

Jr.egui..tLtoJr mt.L6t be du-i..gned ~uc.h :that:

both -6atuJr.ated Jr..egion equ-i.UbJtiwn poin:U
aft e v-ULt uai .

Thus, the positions of the equilibrium points

can be calculated without difficulty. The presence
of real equilibrium points in addition to the F~9. 12. Ftowc.haJLt 06 the c.omputVt pJr..ogJr..aJn 6o!l.
desired quiescent operating point indicates that th« c.ai..cu!.ailOYl 06 tJc.ajec.toJLiu and .time
the regulator is not globally stable; therefore, domain tJr..aYl-6~ent wave6oJr..m-6.
these points must be eliminated. As an example,
the equilibrium points of a two-state boost on most computers. The state-plane trajectories
regulator were found. In addition to the quiescent illustrated in this paper were plotted by a small
operating point, one real equilibrium point existed desktop computer programmed in BASIC.
in the unsaturated region and another in the D = 1
saturated region. Hence, this design was not Figure 12 contains the flowchart of the
globally stable. The response could be improved program. The initial state ~ of the system is
by moving these equilibrium points well outside of given as input, and then the computer iteratively
their respective regions; they would then become calculates ~, the values of the state variables
virtual equilibrium points. at the switching instants t = nT s' as follows:
at the n-th switching interval, the duty ratio dn
3.2 Construction of Trajectories is evaluated from Eq. (11). The region of
operation is determined from Eq. (15). If the
Knowledge of the existence and positions of system is in the unsaturated region, then Eq. (10)
the equilibrium points of the regulator yields a is used to find Bn+l- If the system is in one of
great deal of insight into the large-signal the saturated regions, then Eq. (16) is used to
operation of the system; however, this knowledge find 2n+l = (~+l - ~), with Ds a t taken to be
is not complete. It is also necessary to calculate either Dma x or Dmi n, depending on the region. The
peak transient currents and voltages, and to verify program then increments n and repeats the procedure.
that all possible responses are well-behaved and State-plane trajectories may be obtained, where
stable. Therefore, it is desirable to construct the values of two of the states are plotted in a
the system trajectories. plane, or time domain waveforms may be found.

A number of ~ethods exist for the analytical The computer-drawn state plane trajectories
construction of trajectories [8,9], such as the for the two-state boost regulator example are
vector-field method or the method of isoclines. shown in Fig. 13. The peak values of inductor
For second-order systems, these methods work well. current and output voltage during any given
However, since the dimension of the state space transient are apparent_ The effect of the
is equal to the number of states of the system. unsaturated region real equilibrium point at
these analytical methods become impractical for i* = 2.0A, v* = -3.7V can also be seen: some
regulators with many states, and it becomes unstable solutions occur which bend away in the
necessary to resort to computer simulation. The vicinity of the equilibrium point and head towards
state equations describing the transient response a large negative value of This equilibrium v.
of the regulator, Eq. (10), are easily implemented point is a saddle point. As explained in the

248
 This system may be made globally stable by
sufficient increase of the ratio of current feed-
back to voltage feedback. The system then has
four virtual equilibrium points in addition to the
real quiescent operating point. The trajectories
are plotted in Fig. 14 for the values f l = 0.8,
f2 = 0.08. It can be ~een th~t all solutions
converge to the point i = O. v = O. and a well
behaved, globally stable response is obtained.

Transient waveforms may also be plotted vs.

time. This was done for the boost with integrator
example in Figs. 5-7. In this way, the actual
response of systems with more than two states may
be predicted, and the existence or absence of
unstable solutions verified.

Thus, the nonlinear models of Section 2 may

be used to investigate the large-signal response
of a switching regulator. Equilibrium points may
be calculated analytically, and positioned
properly to obtain a well-behaved response. The
Fig. 13. ComputeJt-pJteciicted .otate-plane actual state-plane trajectories or time-domain
:tJt.ajeetoJtiu 6o/t the two--6tate boof>t waveforms may be calculated easily by a small
ex.ample. Some bLaYl.-6ient6 conveJLge to desktop computer; this allows the informed design
the du-Vi.ed quiuc.ent opeJta4ing po.{.VLt of the regulator and ensures a globally stable
x = 0 , while otheJt-6 Me un.otable and system.
eonveJtge to the /teal. equ.iUbJUW11 iYL
the D = Dmax satun are.d Jtegion. The neat.
equiUbJtillm point at d * = .236 .(}.) a 4. CONCLUSIONS
-6table point.
Because the small-signal switching regulator
models are linear and hence easily applied to most
practical design problems, and because of the
insight they afford into the operation of the
regulator, they are indispensable tools for the
design of a switching regulator. However. because
of the small-signal assumption, these models do
not describe the behavior of the regulator during
large transients. As evinced by the example in
Section 1.1, it is possible for a regulator to be
stable for small perturbations but not for all
large perturbations. A design of this type is
unreliable.

It is of interest, therefore, to formulate

models which are valid for large signals and to
v determine the nature of these large-signal
instabilities. It would then be possible to ensure
10 that the large-signal transient response is well-
behaved. A set of large-signal models is describ£d
in Section 2. The key linear-ripple approximation
of the small-signal state-space averaging method
[1,2,4,5] is employed, but no small-signal
assumption is made. The effect of the saturation
of the pulse-width modulator is also included, as
Fig. 14. The quiuceYl.t opeJuLting point 06 the its influence is substantial.
two -.0 tate bo 0-6t JtegulatoJe. example
becomu globally -6table when the ILaA:A.,O The equilibrium points of a system are
06 f 1 to f 2 cs inc.JtelU ed a .ou66ic.ie.YLt prominent features of its state plane portrait,
amount. and the knowledge of their number and positions
can yield a great deal of insight into the large-
signal system behavior. The equilibrium points
previous section, another real equilibrium point of switching regulators are studied in Section
exists in the D = 1 saturated region. This point 3.1. Analytical expressions are found which may
is a stable equilibrium point, and all solutions be used to place the equilibrium points at
which do not converge to the desired quiescent suitable locations in the state plane, thereby
operating point converge to this additional improving the large-signal transient response.
equilibrium point. This is demonstrated for the example of a two-

249
stable boost regulator. It is found that this [7] D. B. Edwards and T. K. Caughey, "Global
regulator has four equilibrium points in addition Analysis of a Buck Regulator," IEEE Power
to the quiescent operating point. Only two of Electronics Specialists Conference, 1978
these points may have a serious detrimental effect Record, Syracuse, New York, June 1978,
on the response, however. With proper circuit pp. 12-16 (IEEE Publication 78CH1337-5 AES).
design, this example may be rendered globally
stable.
[8] M. Vidyasagar, Nonlinear System Analysis,
Prentice-Hall, 1978.
The actual state-plane trajectories or
time-domain transient response may be found. This [9] J. J. Stoker, Nonlinear Vibrations, John
is easily accomplished by the computer evaluation Wiley and Sons, 1950.
of the models of Section 2. In this way, the
existence of unstable solutions may be observed, [10] Robert W. Erickson and R. D. Middlebrook,
and peak values of transient response waveforms
"Origins of Harmonic Distortion in Switching
calculated.
Amplifiers," Proc. International PCI/M0T0RC0N
'82 Conference, San Francisco, California,
A number of effects have been neglected March 29-31, 1982.
here. Additional modes of operation may exist,
such as transient discontinuous conduction mode [ll] Robert W. Erickson and R. D. Middlebrook,
or current limiting modes. Also, more analysis "Large-Signal Modelling and Analysis of
is possible, such as the prediction of limit Distortion in Switching Amplifiers," Power
cycles and the analytical estimation of stability Electronics Group, California Institute of
regions. Nonetheless, the most basic aspects of Technology, January 1982.
the nonlinear phenomena which occur have now been
described, and the informed large-signal design
of most switching regulators is now possible.

REFERENCES

[l] R. D. Middlebrook and Slobodan Cuk, "A

General Unified Approach to Modelling
Switching-Converter Power Stages," IEEE
Power Electronics Specialists Conference,
1976 Record, pp. 18-34 (IEEE Publication
76CH1084-3 AES); also International J. of
Electronics, vol. 42, no. 6, pp. 521-550,
June 1977.

[2] Farhad Barzegar, Slobodan Cuk, and R. D.

Middlebrook, "Using Small Computers to Model
and Measure Magnitude and Phase of Regulator
Transfer Functions and Loop Gain," Proc.
Eighth International Solid-State Power
Conversion Conference (Powercon 8 ) , Dallas,
Texas, April 27-30, 1981, pp. H-l.1-28.

[3] Arthur R. Brown and R. D. Middlebrook,

"Sampled-Data Modeling of Switching
Regulators," Power Electronics Specialists
Conference, 1981 Record, Boulder, Colorado,
June 1981, pp. 349-369, (IEEE Publication
81CH1652-7).

[4] Slobodan Cuk, "Modelling, Analysis, and

Design of Switching Converters," PhD thesis,
California Institute of Technology,
November 1976.

[5] R. D. Middlebrook and Slobodan Cuk, Advances

in SuuAcned-Mode Vomer Conversion, Vol. I,
TESLAco, 1981.

[6] K. Harada and T. Nabeshima, "Large-Signal

Transient Response of a Switching Regulator,"
IEEE Power Electronics Specialists Conference,
1981 Record, Boulder, Colorado, June 1981,
pp. 388-394 (IEEE Publication 81CH1652-7).

250