CURRENT-MODE CONTROL, FIVE DIFFERENT TYPES, USED WITH THE THREE BASIC CLASSES

OF POWER CONVERTERS: SMALL-SIGNAL AC AND LARGE-SIGNAL DC CHARACTERIZATION,

STABILITY REQUIREMENTS, AND IMPLEMENTATION OF PRACTICAL CIRCUITS

Richard Redl*, Project Engineer and Nathan O. Sokal, President

Design Automation, Inc.

Lexington, MA 02173-3992, U.S.A.

ABSTRACT

Current-mode control effectively eliminates the - ease of applying output-current feed-

phase lag of the control function, associated with forward, to obtain super-fast correction
the output filter inductor or the energy-storage for load transients, minimizing the dev1-
inductor. The several types of current-mode con- at ions of output voltage
trollers differ slightly in their phase-lag charac-
teristics, but they are all far superior to the inherent insensitivlty to static and
commonly used PWM duty-ratio controllers. Unfortu- dynamic variations of input voltage, for
nately, industry has been slow to appreciate and the buck and forward converters
exploit the considerable advantages of current-mode
control. There seem to be two reasons for that: - ideally suited for power converters which
(a) the unavailability of integrated circuits which must draw sinusoidal input current from
are well-suited to implementing current-mode con- the ac power mains, a subject of increas-
trollers and (b) the lack of detailed, systematic- ing interest, recently.
ally organized, theoretical and practical informa-
tion on current-mode controllers. The objective of Unfortunately, industry has been slow to appreci-
this paper is to fill th1s gap by providing a well- ate and exploit the considerable advantages of
organized compendium of useful information for the current-mode control. There seem to be two reasons
design engineer who wants to understand the opera- for that: (a) the unavailability of integrated
tion of a current-mode-controlled power converter circuits which are well-suited to implementing
and to accomplish an optimal design. Five differ- current-mode controllers (as we shall see in Sec-
ent control methods are considered, used with the tion VII, all of the available ICs have serious
three basic types of regulated rectangular-wave shortcomings) and (b) the lack of detailed, sys-
power converters. tematically organized, theoretical and practical
information on current-mode controllers.

1. OBJECTIVES The main objective of this paper is to fill this

gap by providing a well-organized compendium of
Current-mode control effectively eliminates the useful information for the design engineer who
phase lag of the control function, associated with wants to understand the operation of a current-
the output filter inductor or the energy-storage mode-controlled power converter and to accomplish
inductor. The several types of current-mode con- an optimal design. Five different control methods
trollers differ slightly in their phase-lag charac- are considered:
terist1cs, but they are all far superior to the
commonly used PWM duty-ratio controllers. - hysteretic

The advantages of current-mode control are - constant-"off"-time

- essentially no phase lag from control to constant-frequency with turn-on at clock
inductor current, practically eliminating time
the possibility of low-frequency oscilla-
tion of the closed control loop - constant-"on"-time

- inherent pulse-by-pulse current limiting, constant-frequency with turn-off at clock

making the power converter nearly immune time,
to damage from overloads
in combinat1on with the three basic types of regu-
- ease of paralleling power stages, to lated rectangular-wave power converters: buck,
provide increased output-current capabil- boost, and buck-boost.
ity, with equal current-sharing among the
paralleled stages

*Assistant Professor, on leave from the Technical University of Budapest, Budapest, Hungary

771

0275-9306/85/0000-0771 $1.00 © 1985 IEEE

 II. REVIEW OF TYPES OF CURRENT-MODE CONTROL

h All Control ~ for use with transformer-coupled flyback convert-

ers. Fig. 2 shows the implementation of this con-
All types of current-mode controllers command troller, and the relevant waveforms. (In the fig-
directly the current in the inductor of the power ures, "MMV" means "monostable multivibrator.") A
converter. Hence the inductor current follows basic difference between the hysteretic and
instantaneously the control input (Le., within one constant-off-time controllers is that the hysteret-
period or less). This is the basic mechanism which ic controller sets both the maximum and the minimum
results in the advantages discussed in Section 1. current levels; the constant-off-time controller
The different types of current-mode controllers sets only the maximum level.
control the peak current, the valley current, or
both (in the case of the hysteretic controller);
these differences in the controlled quantity result
i
in different operating characteristics, wh1ch we e
shall elucidate in this paper •

.!h. Hysteretic
The tightest regulation of the inductor current
is produced by the hysteretic controller (Fig. 1). n:ru
V
sw
vo
~
-.-.---.-.---.

ill
iL Fig. 2. Constant-off-time controller.

JlJU
vsw
!h. Turn-On at Clock Time
Fig. 3 shows the block diagram of a constant-
frequency controller, in which the switching fre-
quency is determined by a clock signal. The power
Fig. 1. Hysteretic controller. switch turns "on" when the clock pulse sets the R-S
flip-flop. Then the inductor current (i L) and the
switch current Ci sw ) begin to increase. The R-S
Here, the inductor current is compared with the flip-flop is reset, and the power switch is turned
control signal (in the case of voltage-regulated "off," when the comparator detects that the sum of
converters, this is the amplitied voltage-error the inductor current (or switch current) and the
signal, i e ), and the decision-making comparator compensating waveform (i comp ) has increased to the
turns off the power switch when the increasing value commanded by the control signal Ci). The
inductor current reaches the value commanded by the compensating waveform is needed to obtai~ stable
control signal. The power switch is turned "on" operation when the duty ratio is greater than 50%,
again when the inductor current decreases by a as first advised by Deisch [1]. This controller
chosen amount, AI; AI is a design parameter. Thus, commands the peak value of the current.
free-running operation results, with the switching
frequency being determined by the input and output V
voltages, the inductor value, and the value chosen o
V.
for A!. I
POWER

NOTE: Hysteretic current-mode control is

ST/l,GE I CLOCK

covered by u.s. Patent 4,456,872 issued

to Thomas A. Froeschle, assigned to Bose
Corp., Framingham, MA 01701, U.S.A. A
license should be obtained from Bose
Corp. to make, sell, or use in the U.S.A.
any equipment wh1ch falls within the
claims of the Froeschle patent.

h Constant "Off" Time

Vl/ IL + Icomp

y.:u
The constant-"off"-time controller is also a CLOCK COMPEtJS.
free-running controller. Its operation is similar GEN. WAVEFORM
to the hysteretic controller, except that the power (T) GEN.
switch is turned "off" for a fixed amount of time, T
instead of for the time until the inductor current f.. "\
decreases by the amount AI. The power switch is
turned "on" again after the fixed "off" time has Fig. 3. Constant-frequency controller with
elapsed. The "off" time is a design parameter. compensating waveform and with peak-current
This type of controller is particularly well suited command (turn-on at clock time).

772
h Constant "On" Time of the range of duty ratio, except that here the
compensating waveform is needed for duty ratios
Fig. 4 shows the constant-on-time free-running less than 50%.
controller. Here the comparator triggers a one-
shot multivibrator, which turns on the power switch ~ Variations on the Theme Qf Current-Mode Control
for a fixed amount of time, when the current in the
inductor decreases to the control level, i e . Thl.s ~ Freguency stabilization Qf free-running ~:
method controls the valley current. This control- Di Cesare and Gambarara [2] described a hysteretic
ler helps to make turn-off snubbers "fool-proof": controller which operates at a fixed frequency.
All turn-off snubbers require a minimum "on" time That was accomplished by varying the width of the
for discharging the snubber capacitor; this con- hysteresis band as required to keep the switching
troller guarantees that the required minimum "on" frequency equal to a reference frequency, under the
time will be provided, under all normal and abnor- control of a phase-locked loop. Their technique
mal conditions of input voltage and load. can be adapted for use with constant-on-time and
Va constant-off-time controllers.
V.I POWER One can add a clock-pulse train to the iL signal
STAGE to pull the free-running frequency of a hysteretic
iL~ /' controller to a submultiple of the clock frequency.
___ __ _________ _
Z~ ,_~_~ In similar ways, constant-off-time and constant-on-
time controllers can also be synchronized to a
clock source.

In a planned future paper, the authors intend to

discuss in detail the frequency stabilization of
free-running controllers.

b.Combinations of technigues: One can combine

more than one type of current-mode control into a
single controller. For example, we shall see in
Fig. 4. Constant-on-time controller. Section III that a hysteretic controller cannot
operate in the "light" operating mode (discontinu-
ous inductor current). The cure for that shortcom-
ing will be shown (in Section III) to be a "combin-
ation" controller: hysteretic in the "heavy" oper-
Fig. 5 shows another constant-frequency control- ating mode (continuous inductor current), and
ler; this one commands the valley current. Here, constant-off-time in the "light" operating mode.
the power switch is turned "off" when the clock Fig. 9 shows such a controller.
pulse resets the R-S flip-flop. Then the inductor
current begins to decrease. The power switch is O'Sullivan et ~ [3] described a combination-type
turned "on" again when the sum of the inductor controller which can operate in any of three modes,
current (i L ) and the compensating signal (i com ) depending on the polarity of the clock signal:
decreases to the value commanded by the contrEI constant-frequency with turn-on at clock time,
signal, i e . Similarly to the previous constant- constant-frequency with turn-off at clock time, or
frequency controller, this controller requires a hysteretic if the clock signal is not provided.
compensating waveform to ensure stability in part
Va In a planned future paper, the authors intend to
treat in detail some less-well-known, but useful,
Vi POWER
STAGE i
e~
I0 ICLOC K
I
I
"combination" types of controllers, similar to the
treatment of the other types in this paper.

t~ camp

R-S
FL I P- FLOP
:t'ref~ III. ANALYSIS OF THE DC CHARACTERISTICS OF
CURRENT-MODE-CONTROLLED CONVERTERS
R S
~ Needed Information

1'\1" To produce a "fool-proof" design, one must know

.~
in detail the static characteristics (dc, nonline-
ar) of the power converter with its controller:
I
CLOCK 1. output voltage as a function of control
~H' ~ signal, input voltage, and load resist-
ance
I• T'
,..1
2. behavior under overload conditions (from
Fig. 5. Constant-frequency controller with full-load to short-circuit)
compensating waveform and with valley-
current command (turn-off at clock time). 3. start-up behavior

773
 4. switching frequency of the free-running jected toward the load during one full period of
types. operation. That is very simple for the buck regu-
lator (the injected current is equal to the average
Until now, insufficient attention has been paid to of the inductor current). but it requires the de-
these important characteristics of sw itching-mode termination of the duty ratio for the boost and
power converters. The users have paid the penalty: buck-boost converters. The next step is to substi-
unnecessary catastrophic failures under the occa- tute the relation between the output voltage and
sional abnormal conditions which cannot be avoided. the load current into the expression for the aver-
age injected current flowing into the load. The
~ Model Complexity ~ Insight Into Operation resultant equation has three variables: Ie. Vi' and
V ; specifying any two of them will determine the
In the practice of circuit analysis, there are t~ird. Table 1 gives a tabular summary of the
various levels of refinement of the models. The procedure for deriving the dc function (1) for the
accuracy of the results increases with the refine- case of continuous inductor current.
ment of the models; but at some point, the increase
of accuracy is no longer worth the extra complica-
tion. The additional accuracy will not help us to TABLE 1
make good design decisions it the added complexity PROCEDURE FOR DERIVING DC FUNCTION
causes us to lose insight into the basic physical IN CONTINUOUS MODE OF OPERATION
behavior of the system, and into the interactions
among the parameters. In the case of high- STEP NO. EQUATION
efficiency switching converters, second-order fac-
tors (e.g., loss resistances, semiconductor junc- 1. ILpeak = fl (Ie' Vi' Vo ' c)
tion conducting voltages, and switching delays) can
usually be neglected without significant sacrifice 2. ILvalley = f 2 (Ie' Vi' Vo ' c)
of accuracy; we will omit them in this paper.
3. I Lave = (Il.peak + ILvalley)/2
h Method of Analysis
4. D = f 3 (V i , vol
The complete dc characteristic of a current-mode
controlled switching converter can be presented in 5. 10 = f4(I Lave ' D)
the form:
6. 10 = f 5 (V o )
(1)
7. f4(I Lave ' D) = fS(Vo )
where the meanings of the quantities in (1) are
shown in Fig. 6. The vector c represents those 8. f 6 (I e , Vi' VO ' c) = 0

I
o
URRENT-MODE --t> ~ Discontinuous inductor current
CONTROLLED
CONVERTER A very similar procedure is followed for the case
of discontinuous ("light") mode of operation. The
difference is that here, by definition. the valley
current is zero. Therefore, instead of calculating
the valley current, one must determine the "on"
time Ton' the "off" time Toff' and the time Tf
Fig. 6. Model of controller and power converter for during which the current is falling from the peak
d~ rh~r~~~or;~g~;nn. current to zero. (See Fig. 7 for the inductor-
current waveform.) Table 2 gives a tabular summary
parameters of the system which influence signifi- of the steps for deriving the dc function (1) for
cantly the dc behavior (inductance, frequency, "on" the case of discontinuous inductor current.
time and/or "off" time, current hysteresis, compen-
sating waveform, etc.). Usually the ratio of the
output voltage Vo to the load current 10 is con-
stant and equal to the load resistance RL• Immedi- I
-- Lpeak
ately below, we shall summarize briefly the method
of deriving the dc function (1) for the two differ-
ent cases of continuous and discontinuous inductor
current, and the method of locating the boundary
between continuous and discontinuous operation.

h Continuous inductor current: The f~rst step

is to determine the peak and valley currents in the
inductor, as functions of the control input I~, of
the input and output voltages Vi and VO ' and ot the
system parameters represented as c in (1). The Fig. 7. Inductor current in discontinuous mode of
next step is to calculate the average current in- operation.

774
 TABLE 2 applications. In the typical case of voltage-
PROCEDURE FOR DERIVING DC FUNCTION regulated power supplies, the usual consequence is
IN DISCONTINUOUS MODE OF OPERATION a low-frequency oscillation, whose frequency is
influenced also by the frequency response and tne
STEP NO. EQUATION slew rate of the voltage-error amplifier. This can
be acceptable in most cases. If, however, the low-
1. ILpeak = g I (Ie' Vi' Vo ' c) frequency oscillation must be avoided, additional
circuitry has to be built into the controller,
2. Ton = g2(ILpeak' Vi' Vo ' c) which will turn the power switch "on" again. One
solution is to add a monostable multivibrator as
3. Tf = g3(ILpeak' Vi' Vo ' c) shown in Fig. 9. Here the "off" time of the switch
is limited by the multivibrator, so the step in the
4. Toff = g4(c) transfer function will be eliminated (free-running
operation will be retained at light load , too).
s. 10 = gS(ILpeak' Ton' Tf , Toff )
h Boundary between heavy and light modes: The
6. 10 = fS(V o ) converter can change operating mode if the control
signal, the input voltage, or the load changes. By
7. gS(ILpeak' Ton' T f , Toff ) = fS(V o ) solving, for example, the equation
8. g6(I e , Vi' Vo ' c) = 0 (3)

[where Vo (f6) and Vo (g6) are the output voltages

The controllers which command the valley current obtained from Tables 1 and 21, one can determine
are inherently unable to work in the discontinuous- the boundary between the heavy and light modes of
current mode, when used with the common types of operation (the operating conditions under which the
power converters. (Those controllers are : (a) valley current in the continuous-conduction mode is
hysteretic, (b) constant-on-time, and (c) constant- just zero) .
frequency with valley-current commanding.) When
the valley current is commanded to be greater than
zero, operation is in the continuous-conduction vo
mode, with no difficulty. But if the valley cur- v.I POWER
rent is commanded to be less than llLQ. (i.e.,
STAGE
reverse polarity), the usual type of power convert-
er is unable to obey the command, because it cannot
produce a reverse-polarity inductor current. (If
one would use a bidirectional power switch instead
of the usual recovery diode, as was suggested in
[4], these valley-current-commanding control meth-
ods could be used at any load current, including
reverse current. However, other measures would
normally be preferable, to provide satisfactory
performance in the discontinuous-current mode.)

We shall illustrate the problem with a hysteretic (T

off max)
controller controlling a buck converter. If the
required va Uey current drops be low zero, i.e., if
Fig. 9. Hysteretic controller with off-time
I limiting.
v
the power switch will not turn "on" again, and the
load current will drop to zero. Fig. 8 shows the ~ Results of Analyses
resulting control-to-output transfer function. The
step in the control function is undesirable in some The dc characteristics of the current-mode-
controlled converters are summarized in Tables 3
through 6.
I
o l
Table 3 gives expressions used for calcu-
lating the output voltage as a function of
control signal, input voltage, ripple cur-
rent, and load resistance, in both the
heavy and light operating modes.

tli - Table 4 gives expressions for the ripple

T currents in the heavy mode, and for the
"on" times and current fall times in the
light mode. The ripple current also gives
Fig. 8. Load-current VB. control current for information for determining the boundaries
hysteretic controller and buck converter. between the heavy and light modes.

775
 TABLE 3
OUTPUT VOLTAGE VS. CONTROL SIGNAL, INPUT VOLTAGE,
RIPPLE CURRENT, AND LOAD RESISTANCE

CONTROLLER CONVERTER
MODE

heavy
BUCK

Vo = (Ie - 6I/2)R L
BOOST

Vo =VV.(I
~
i

e - 6I/2)R L Vo = [Vi
BUCK-BOOST

-VV~
.
+ 4(I e - 6I/2)R LVi 112
Hysteretic
light Vo = 0 Vo = Vi Vo = 0
i

-VV~
i
heavy Vo = (Ie - 6I/2)R L Vo =VV'(I
~ e - 6I/2)R L Vo = [Vi + 4(I e - 6I/2)R LVi 1/2
Constant
off-time RLIe Ton + Tf RLIe Tf -RLIe Tf
light Vo = - ( ) Vo = - ( ) Vo = - ( )
2 Ton + Toff 2 Ton + Toft 2 Ton + Toft

-6I/2)R~
i

Constant-
frequency
heavy Vo = (Ie - 6I/2)R L Vo =VVi(I e V0 = [Vi -11 VI + 4(I e - 6I/2)R LVi 112
with peak- RLIe Ton + Tf RLIe Tf -RLIe Tf
current light Vo = -2( T
-) Vo = -2( -
T
) Vo = -
2
(-)
T
command
i i

Constant-
heavy Vo = (Ie + 6I/2)R L Vo =V V· (I e + 6I/2)R L
~ Vo = [Vi -VV~ + 4(I e + 6I/2)R LVi 112

on-time light Vo = 0 Vo = 0 Vo = 0
Constant-
frequency
with valley-
current
heavy

light
V0 = (Ie + 6I/2)R L Vo =V V· (I e + 6I/2)R L
~
i

Vo = [Vi -v VI + 4(I e + 6I/2)R LVi 112

i

Vo = 0 Vo = 0 Vo = 0
command

TABLE 4
CURRENT RIPPLE IN HEAVY MODE, AND CURRENT FALL TIME
AND ON-TIME IN LIGHT MODE

CONTROLLER CONVERTER

MODE BUCK BOOST BUCK-BOOST

heavy 61: predetermined

Hysteretic
light Tf , Ton: not applicable

heavy 6 I = VoToff/L 61 = (Vo-Vi)Toff/L 61 = -VoToff/L

Constant-
off-time Tf = IeL/Vo Tf = IeL/(Vo-Vi) Tf = IeL/(-v o )
light
Ton = IeL/(Vi-Vo) Ton = IeL/vi Ton = IeL/vi
Constant- heavy 61 = (Vi-Vo)VoT/viL 61 = Vi (Vo-Vi)T/VoL 61 = -VoViT/tvi-VO)L
frequency
with peak- Tf = IeL/Vo Tf = IeL/(Vo-V i ) Tf = IeL/(-v o )
current light
command Ton = IeL/(Vi-Vo) Ton = IeL/Vi Ton = IeL/vi
Constant- heavy 61 = (Vi-VO)Ton/L 61 = ViTon /L 61 = ViTon/L
on-time
light Tf , Tun: not applicable

Constant-
frequency heavy 61 = (Vi-Vo)VoT/ViL 61 = Vi (Vo-Vi)T/VoL 61 = -VoViT/tVcVO)L
with valley-
current light Tf , Ton: not applicable
"''''...... n.!

776
 TABLE 5
FREE-RUNNING FREQUENCY
CONTROLLER CONVERTER

BUCK BOOST BUCK-BOOST

f = [Vo(vi-Vo)]/ViLAJ f = Vi(Vo-Vi)/VoLAI f = V0 Vi / (V0 - Vi) LM
Hysteretic if Ie > AI if Ie > AI if Ie > AI

f = 0, if Ie < AI f = 0, if Ie < AI f = 0, if Ie < AI

f = (Vi-Vo)/ViToff f = Vi/VoToH f = Vi/(Vi-Vo)Toff
Constant- if Ie > VoToff/L if Ie > (Vo-Vi)ToH/L if Ie > -VoToff/L
off-time
f = (VCVo)/lIeL + Toff(VCVo) J f = Vi/(IeL + ToffV i ) f = Vi/(Ie L + ToffV i )
,
if Ie < VoToff/L if Ie < (Vo-Vi)Toff/L if Ie < -VoToff/L

f = Vo/ViTon f = (Vo-Vi)/VoT on f = Vo/(Vo-Vi)T on

Constant-
on-time
if Ie > ° if Ie > ° if Ie > °
f = 0, if Ie
< ° f = 0, if Ie
< ° f = 0, if Ie
< °
Table 5 gives the switching frequencies of
the free-running contro llers.

- Table 6 summarizes the behavior of the

different controllers when operating into a
short-circuit load.

TABLE 6 IV. SMALL-SIGNAL AC CHARACTERISTICS OF

SHORT CIRCUIT CURRENT CURRENT-MODE-CONTROLLED CONVERTERS
CONTROLLER CONVERTER ~ Method of Analysis
BUCK, BUCK-BOOST* The small-signal characteristics of regulated
switching-mode dc/dc converters are important to
Hysteretic lsc = Ie -61/2, well controlled. the design engineer for two reasons:
Constant-off- I~C = Ie' actual sIc current 1. These are negative-feedback systems, and
time hlgher, due to time delay, value loop stability is almost always manda-
controlled by parasitic voltage tory. To design a control loop which is
drops. sure to be stable, the engineer should
have an accurate small-signal-ac model
Constant- Isc = Ie' actual sIc current of the converter and the controller.
frequency higher, due to time delay, value
with peak- controlled by parasitic voltage 2. Suddenly varying load current, and/or
current drops suddenly varying source voltage, and/or
command high-speed programmability of a power
supply, calls for a good dynamic re-
Constant- Isc = Ie + Vi T/2L, short-circuit sponse. To design a control loop which
frequency operation will display low- has a good dynamic response, the engin-
with valley- frequency oscillation eer should understand the dynamics of
current the system; i.e., he should have an
command (no accurate small-signal-ac model of the
compensating converter and the controller.
waveform)
There are two, basically different, methods for
Constant- Isc = I~ + ViTon/2L, well- the analysis of dc/dc converters, which are useful
on-time controlled for practical and tutorial reasons. They are: (a)
averaging of the state variables [51, and (b) the
method of injected-absorbed currents (first intro-
* Not applicable for boost converter duced in l61, and discussed and used extensively

777
throughout [7]). The latter method is especially continuous-current mode), an additional quantity
well suited to the analysis of the various current- must be taken into account. This quantity is the
mode controls, and can produce satisfactory results time derivative of the inductor current. It can be
with a minimum amount of eftort [Sl, 191. For that approximated as follows:
reason, and for other reasons described in 171, we
use in this paper the method cf injected-absorbed diL iLn - iL(n-l)
currents. (8)
dt - T
Fig. 10 shows the model used for the analysis.
Both the current injected toward the combination of where T is the period of the switch1ng frequency,
the load and the output filter capacitor (i:i,.)' and and iL and iL(n-l) are the inductor currents at
the current absorbed from the source (i a ), are the en8 and at the beginning of a switching cycle
functions of three variables: the input voltage, (i.e., two successive values of the valley current
the output voltage, and the "controlled variable," in a peak-current-commanding system, or two succes-
represented as x. In functional form: sive values of the peak current in a valley-
current-commanding system).
(4)
Manipulation and Laplace transformation of (6)-
and (8) yield

(5)
(9)
and
In (4) and (5), all quantities are values aver-
aged over one cycle. The controlled variable (x)
(0)
is th.e variable which is controlled by the error
amplifier in such a way that the regulated output
(When the small-signal variables are being repre-
is maintained proportional to the reference input,
in the c 10 sed-loop syst em. In current -mode- sented as time functions, they are shown as lower-
controlled converters, x is the inductor current; case letters, as in (4) - (8). When those vari-
in PWM-controlled converters, x is the duty ratio. ables are being represented as Laplace-transform
functions, they are shown as capital letters, as in
(9) and (0).) (9) and (0) give a complete small-
signal characterization of the switching cell and
provide the basis for the introduction of the gene-
i i. i ral low-frequency equivalent circuit 110,71. Al-
a I 0
---I> -t> --t> ternatively, the characteristic ac functions can be
c>- -~ derived by using block-diagram representations
SHITCHING I LOAD
tV i
0----
CELL
V0
t C
0
(see, e.g., (91).

OUTPUT F I LTER ~ Generating Open-Loop Transfer and Impedance

CAPACITOR Functions
X
There are four significant open-loop functions
Fig. 10. Model for the analysis method of injected which characterize the dynamic behavior of a con-
and absorbed currents. trolled dc/dc converter. They are:

The small-signal models assume linear relation- 1. control-to-output transfer function

ships among the involved quantities. Therefore we
can write the total differentials for the injected 2. input-to-output transfer function
and absorbed currents as:
3. output-impedance function

4. input-impedance function.
di i (0 ii!dx)dx + (oii/dvo)dvo
The last function is important in the occasional
+ (dii/dvi)dvi (6) special cases when the impedance of the power-input
source is not negligible (e.g., an EMI filter with
high output impedance is inserted between the power
dia (dia/dX)dx + (cHa/ovo)dvo source and the power converter, or the power input
to the converter comes from the output of another
+ (oialdVi)dvi (7) power converter). In this paper, we shall not deal
with the input-impedance function, because it is
where the partial derivatives of these currents can needed in only a few special cases. The other
be derived from (4) and (5). three functions are always needed; we shall summar-
ize here the method of generating them.
If the cell input current (absorbed current) and
output current (injected current) depend on the ~ Control-to-output transfer function (open
average inductor current (as is the case, for ex- .l.Q.QJU.: In order to derive this funct ion, one uses
ample, for all PWM-controlled converters in the (10) in conjunction with the following equation

778
 (ll ) TABLE 7
OPEN-LOOP AND CLOSED-LOOP TRANSFER
which simply states that the output voltage is the AND IMPEDANCE FUNCTIONS
product of the load impedance and the current in-
jected toward the load.
FUNCTION OPEN LOOP CLOSED LOOP
The result is the control-to-output transfer
function: Control-to- A/O/Z L + B) -AG/U/Z L + B - AG)
output
(2)
Input-to C/O/Z L + B) C/(l/ZL + B - AG)
~ Input-to-output transfer function (open JQQpl: output
Similar simple considerations yield the input-to-
output transfer function, too: Output l/(l/ZL + B) 1 (1/ ZL + B - AG)
impedance
( 13)

h Output-impedance function (open.l.Q..Q.pl: The constant-frequency controller with peak-current

output impedance of a converter is, by definition, commanding are available in an excellent recently
the negative of the ratio of output voltage to load pub li shed monograph In. The coefficients of the
current. Here we assume a current generator is in valley-current-commanding versions are the subject
parallel with the load impedance. This means that of a planned future paper by the present authors.
instead of (11),
h Hysteretic controller: Determination of the
(14) coefficients for the hysteretic controller is very
simple. Because of the tight control of the in-
is valid, where Io(S) is the current of the load ductor current, there is practically no delay be-
current generator, separated from the current in tween the control signal and the injected current,
the load impedance. The result is the output im- and between variations of input or output voltage
pedance: and the injected current. That means that the
coefficients for the injected current are indepen-
dent of frequency, up to about half of the switch-
ing frequency. Table 8 lists the coefficients A,
~ Generating Closed-Loop Transfer and Impedance B, and C for a hysteretic controller controlling
Functions buck, boost, and buck-boost converters.

Generation of the closed-loop functions is also

straightforward. The additional equation is TABLE 8
SMALL-SIGNAL COEFFICIENTS FOR
HYSTERETIC CONTROLLER

which states that the controlled quantity is equal COEFFICIENT CONVERTER

to the. amplified error signal. In the case of
current-mode-controlled converters, G(s) is a BUCK BOOST BUCK-BOOST
transfer admittance.
A I Vi/Yo V/(Vo-V i )
The general equation which contains all of the
parameters needed for deriving the closed-loop B 0
2 -I e V./(V.-V )2
IeVi/Vo ~ ~ 0
functions is
C 0
2
Ie/Vo IeVo/(Vi-V o )
A(s) G(s) [Vo(s) - Vref(s)]

-B(s)Vo(s) + C(s)Vi(s) on ~ Constant-off-time controller

Table 7 gives the open-loop and closed-loop The expressions for the small-signal coefticients
transfer functions. of the constant-off-time controller are slightly
different, and are shown in Table 9. Experimental
~ Characteristic Coefficients of Current-Mode- results (discussed later, in Section VI) showed
Controlled Converters -- Six Examples that the response of the injected current to the
control signal is not instantaneous for this con-
In this section, we summarize the coefficients troller, as it is for the hysteretic controller.
for six examples of combinations of control methods For the constant-off-time controller, we found a
and converters, and give a reference to a publica- phase difference between the measured control-to-
tion which lists three more. We give here the output function and the function calculated with
small-signal coefficients A, B, and C for the out- the coefficients presented in [8]. The discrepancy
put injected current, for hysteretic control and is caused by a time delay equal to Toff /2 in the
for constant-off-time control, for buck, boost, and response of the injected current to the control
buck-boost converters, operating in the continuous- signal, not accounted for in [8]. Appendix A of
conduction mode. The coefficients for the this paper gives the derivation of this time delay.

779
 TABLE 9
SMALL-SIGNAL COEFFICIENTS FOR
CONSTANT-OFF-TIME CONTROLLER

COEFFICIENT CONVERTER
BUCK BOOST BUCK-BOOST

A -sT off /2 V· -sT off /2 __V·1_ -sT off /2

e -l- e e
Vo Vo-Vi

B Toff /2L 2 - [(1 e + T0 ff V./

(Ie + ToffVi/2L)Vi/Vo 1 2L )V.l/(v.-v
1 1 0 )2

C 0 [Ie - Toff(Vo-2Vi)/2L1/Vo [(Ie + ToffVo/2L)Vol/(Vi-Vo)2

b. Improving Regulation Parameters .Qy Feed-Forward TABLE 10

of Input Voltage and Load Current STABILITY CHARACTERISTICS OF FIVE CONTROLLERS

The response speed of the feedback-control loop CONTROLLER COMMENTS

is a basic limitation on the speed with wh1ch a
regulator can correct for the disturbing effects of ~ysteretic Unconditional stability, free-
transient changes of input voltage or load current. running frequency varies with ac
After a transient occurs, the output filter capaci- loop gain.
tor must hold the output voltage within the re-
quired tolerance band until the regulator and the Constant- Produces hysteretic subharmonic
power converter eventually change their operating off-time oscillation above critical ac loop
point to accommodate the new condition of input gain.
voltage and load current.
Constant- Produces open-loop instability ~
If one senses the input voltage and the load frequency 50% duty ratio, therefore compensat-
current, and can calculate what the controller ~ith peak- ing waveform is required. Produces
output should be at the new input/output condition current continuous or hysteretic subharmonic
after a transient, one can immediately command the command oscillation above critical ac loop
controller to change to the proper new state, with- gain.
out waiting for a sluggish control loop to decide
what to do and then do it, while the output error Constant- Produces open-loop instability ~
builds-up during the long wait. In [9] and [161, frequency 50% duty ratio, therefore compensat-
the authors of this pape·r gave analyses, designs, with valley- ing waveform is required. Produces
and experimental results for current-mode con- current continuous or hysteretic subharmonic
trollers which use such feed-forward techniques. command oscillation above critical ac loop
The result: with only minimal additional hardware, gain.
the controller corrects the output voltage within
only ~ or two cycles of the switching frequency; Constant- Produces hysteretic subharmonic
the transient deviation of output voltage was re- on-time oscillation above critical ac loop
duced .Qy ll. factor of seven for a step load change gain.
with a rise time of 62% of a switch1ng per1od.
(During the one or two cycles that it takes for the
controller and converter to change their operating
point, the output capacitor must hold the output
voltage within the required tolerance band -- but ~ Low-Frequency Instability
that's much better than requiring the capacitor to
hold the output voltage for several tens of cycles Knowing the small-signal characteristics of the
of the switching frequency, as with a conventional various controller-converter combinations the en-
contro ller. ) gineer can predict whether or not ther: will be
low-frequency instability when the voltage-
reg~lating loop is closed. The characteristics of
V. OPEN-LOOP INSTABILITY, CLOSED-LOOP low-frequency instability are (a) the frequency of
INSTABILITIES, AND OTHER SIDE EFFECTS t~e ~p~rious oscillation of the control loop is
IN CLOSED-LOOP OPERATION s1gn1f1cantly lower than the switch1ng frequency,
and (b) the ratio of the two frequencies is not an
Four types of instabilities are known and have integer number, except by coincidence.
been analyzed. Sections A and B below explain the
causes and the symptoms. Table 10 summarizes the The well-known methods of classical control theo-
information. ry apply directly; it is not necessary to discuss

780
that topic in detail here. There is one point, back. There are several different forms of this
though, which should be emphasized. All current- instability.
mode controlling methods produce a loop gain with
approximately 90 0 overall phase lag (assuming an The constant-frequency current-mode-controlled
ideal capacitor in the output filter). This makes converters can produce a subharmonic oscillation
it much easier to design a stable control loop than whose characteristic feature is that it starts and
in the case of duty-ratio-modulated converters. stops at the same ac gain of the error amplifier.
The frequency response of the error amplifier is
~ Subharmonic Instabilities shown in Fig. 11. We assume that the gain is
constant, its value being g at both the switch1ng
As with all sampled-data control systems, frequency and its first subgarmonic. Analysis of
current-mode-controlled converters can have subhar- the propagation conditions of perturbations gives
monic instabilities. Subharmonic instability (as us bounds on the maximum usable gain. The deriva-
opposed to the previously defined low-frequency tion and the results for the buck and the boost
instability) is characterized by the fact that the converters were presented in [121.
ratio of the nominal switching frequency to the
frequency of spurious oscillation is an integer
number. This means that oscillation occurs at
exact ly the nth subharmonic of the switching fre-
quency. In mest cases, n = 2, although higher- g(w)
order subharmonic oscillation is possible. dB

Subhllrmonic instability is the only real insta-

bility problem of current-mode-controlled convert-
ers. It occurs in three forms, wh1ch have ditfer-
ent root causes. All three types of subharmonic
instability can be prevented by choosing the design
parameters properly, as discussed below. Now we
summarize the three forms of subharmonic ins~abil­
-t-----
f 12 f
-
f(log)
ity. Extensive details are given in [8, 11, 121. sw sw

~Open-loop instability of constant-frequency Fig. 11. Frequency response of the error amplifier.
current-mode controllers: It has long been
known that constant-frequency current-mode-con-
trolled converters with peak-current commanding
produce instability at duty ratios above 50%, even This instability is known to occur in constant-
if the voltage-regulating loop is open. (The first frequency converters. We assume that the valley-
reference to this instability is probably [131; a current-commanding version suffers from this insta-
cure was first proposed by Deisch [1].) bility, too, though it has never been analyzed.

It is less well-known, though straightforward to ~ Closed-loop instability: hysteretically aris-

derive, that a similar instability shows up in iB& subharmonic oscillation: Those converter
constant-frequency converters with valley-current topologies where the current being injected toward
commanding. But in this case, the instability the load is pulsating (boost and buck-boost con-
occurs for duty ratios less than 50%. verters) can exhibit a special kind of subharmonic
instability. The characteristic feature of this
These instabilities can be prevented by adding a instability is that it starts an oscillation at an
compensating waveform to the amplified error S~g­ upper level of ac gain of the error amplifier, and
nal, as shown in Figs. 3 and 5. The compensat1ng decreasing the gain eventually stops the oscilla-
waveform has a significant influence on the low- tion at a lower level of gain (see Fig. 12). This
frequency characteristics of the converter. For instability can occur in the constant-frequency,
the most important version -- peak-cur~en~­ :onstant-off-time, and constant-on-time control-
commanding -- the values of the character1st1c lers, but not in the hysteretic controller. De-
coefficients are available [7]. It is also known tailed analyses of the conditions for this kind of
that the slope of the compensating waveform has an subharmonic oscillation are given in [8,15,11,
effect on the attenuation of perturbations: the 12].
optimum slope causes the perturbations to decay at
the fastest possible rate, i.e., within one switch- ~ Frequency Shift of Hysteretic Controllers Caused
ing cycle. The optimum slope for the buck and the Qy Ripple-Voltage Feedback
buck-boost topologies is constant [14, 121; that
means that the optimum compensating waveform is a The hysteretic controllers are free from all
linear ramp. For the boost converter, the optimum kinds of the previously mentioned open-loop and
slope is proportional to the input voltage; hence closed-loop subharmonic instabilities. However,
(as explained in [121) the optimum compensating the amplified and fed-back ripple voltage can shift
waveform is a segment of a parabola. the free-running switching frequency. The shift
can be upward or downward, depending on the topolo-
~ Closed-loop instability: continuously ar1s1ng gy and on the combination of circuit parameters in
subharmonic oscillation: Closed-loop subhar- a specific design. An upward frequency shift can
monic instabilities are the result of ripple feed- increase the switching losses; a downward frequency

781
 f 25 kHz (for PWM and constant-

STATE 1 OSCIlLATORY
frequency peak-current-
commanding controller)

I
20 us (constant-off-time
controller)
1.25 A

I t
STABLE
57 A/ms (for constant-frequency
peak-current-commanding
contro ller)

L-------l- - --- -__

OSCILLATION
STOPS
OSCILLATION
STARTS
9
o
The measurements were made with a BAFCO Type
916XH Frequency-Response Analyzer. Figs. 13 and 14
show the measured loop gains as functions of fre-
quency: Fig. 13 shows the magnitudes, and Fig. 14
Fig. 12. Appearance of hysteretic oscillation vs. shows the phases. (Due to slight inaccuracies in
initial estimates of the needed experimental condi-
ac gain of the error amplifier.
tions, the slope of the compensating ramp for the

shift can cause increased ripple current in the

inductor and in the output capacitor, and larger
output ripple voltage. The only remedy for this
phenomenon is to account for this effect in the
design of the converter (i.e., design the converter
to operate as desired, understanding how the con-
verter and controller work together). Details of
the analysis are given in [8, Ill.

VI. EXPERIMENTAL CONFIRMATION OF

CONTROL-TO-OUTPUT FUNCTIONS

To verify the theoretical derivations, we meas-

ured the magnitude and phase vs. frequency of the
forward gain of the controller/converter combina-
tion (the transfer function from control input to
voltage output), for four different controllers
used with the same power converter. The test ve- Fig. 13. Measured magnitude of loop gain vs.
hicle was a buck converter; the four types of con- frequency.
trollers were:

a. hysteretic,

b. constant-off-time,

c. constant-frequency peak-current-
commanding with compensating ramp to
provide near-minimum decay time of -40 ,
- --i-
transients at all duty ratios (and guar-
anteed stability at duty ratios larger
than 50%), and -60

d. constant-frequency PWM, as a base-line

of comparison which would be familiar to 80
all engineers concerned with switchlng-
mode power conversion. ,I
100
I
The circuit data for the converter and the control- i

I
i
- -r
lers were:
- 120
Yin 22 V !
Vout 10 V !
D 50% -140
1 A 1 2 10
lout
L 176 uH
C 998 uF
ESR 64 mohm Fig. 14. Measured phase of loop gain vs. frequency.

782
constant-frequency peak - current-commanding con- A[d B)
troller was slightly less than the value which
provides optimum attenuation of perturbations: 57
A/ms instead of the optimum 62 .5 A/ms.)

Fig. 14 shows clearly the basic advantage of all

of the current-mode controllers over the conven- -5
tional PWM duty-ratio controller . While the PWM
controller produces 136 0 peak phase lag at 2 kHz, I
all of the current-mode controllers produce 78 0 or -10 1 1- 30
less phase lag at frequencies up to 7 kHz. Only the
constant-frequency type exceeds 90 0 , when approach-
ing half of the switching frequency. With respect
to phase margin, the hysteretic controller is best ,
and the constant-off-time controller is next-best.

To verify the theoretical results presented in

Section IV, we calculated the expected magnitude
and phase for the two best controllers (hysteretic - 60
and constant-off- time), and compared the calculated
values with the measurements. Fig. IS shows the
results for the hysteretic controller; Fig. 16 -30 -70
gives the same information for the constant-off-
time controller. The calculated data are shown as
solid lines; the measured values are shown as tri-
angles and circles . The graphs show that the cal-
culated values are in excellent agreement with the i
measured data; we conclude that the simple small- Il.
- 40 1 -90
signal ac model discussed in Section IV is accurate
2 5 10 100 1000 10,000
f[Hz)
¢[ O Fig. 16 . Calculated and measures magnitude and
~-;r---~--~--r---~--~--~--,---,---~--,LL-~ phase of buck converter with constant-oft-
time con t ro ller .
o
for all practical purposes. (Included in the cal-
-10 culated phase lags are the switching delays in the
comparator, the power switch , and the driver for
I
I the power switch. Their total delay is 1 us,
5 !
---~+--- - producing a phase lag of (0 .00036)£ degrees.)

I
-lilL_ VII. CONCLUSIONS
i

i 1. Current-mode control effectively eliminates the

-J5~ phase lag of the control function, associated with
the output filter inductor or the energy-storage
!
inductor. The several types of current-mode con-
-20 i -50 trollers differ slightly in their phase-lag charac-
teristics, but they are all far superior to the
commonly used PWM duty-ratio controllers.
-2 1 -60
2. The hysteretic type is the best of the current-
mode controllers in three respects: largest phase
-70
-=-9--I
margin, complete freedom from any type of subhar-
monic oscillation, and well-defined and well-
limited short-circuit current (when used with buck
or buck-boost converters).

3 . The constant-off-time controller is nearly as

good as the hysteretic controller in those re -
5 10 1 0 1000 10,000 spects, and is especially we l l-suited for use in
f[Hz) transformer-coupled flyback converters.
Fig . IS . Calculated and measured magnitude and 4. To the best of the authors' knowledge, none of
phase of control-to-output transfer the available "current-mode-controller" ICs can be
function for buck converter with used for either of the two best types of current-
hysteretic controller . mode control: hysteretic or constant-off-time.

783
5. Current-mode control has so many advantages that i
it deserves to have a family of control ICs which e
allows the design engineer to exploit fully those
advantages.

ACKNOWLEDGEMENT

The authors thank Laszlo Balogh of the Technical

University of Budapest for his help in building the
test circuit and carrying out the measurements.
61

APPENDIX A.
DERIVATION OF TIME DELAY BETWEEN CONTROL Fig. AI. Inductor currents of converter with
AND INJECTED CURRENT (COEFFICIENT "A") hysteretic controller (solid line) and
FOR CONSTANT-OFF-TIME CONTROLLER with constant-off-time controller (dotted
line), when the control signal changes.
The formal derivation of the coefticient "A" (the
functional dependence of the injected current on
the control signal) in [8,91 gives the same result
for both the hysteretic controller and the
constant-off-time controller. However, experiment-
al data (discussed in Section VI) showed that the
phase lag of the two control methods is not the 61
same. The phase difference can best be described T
by a time delay introduced in the control path of
the constant-off-time controller. The amount· of
time delay is Toff !2. The proof follows.

Assume that the current ripple in the constant-

off-time controller is the same as in the hysteret-
ic controller. That is,

(AI)

where ~I is the current ripple and m2 is the abso-

lute value of the current slope dur~ng the off-time
of the switch, Toff.
Fig. A2. Enlargement of part of Fig. Al for
With the assumption of (AI), there is no ditfer- calculation of time delay between average
ence between the waveforms of converters controlled currents.
by hysteretic or constant-off-time controllers.

Assume now that the control signal, ie' changes.

The resultant waveforms are shown in Fig. AI. Here
mi and m2 are the slopes of the inductor current, Toff !2. The distance between the turn-off instant
and m is the slope of the control signal. The and point A (the intersection of the falling cur-
solid line shows the current for the case of the rent waveform with the average current for the
hysteretic controller; the dashed line shows the hysteretic controller) is TA• TA is given as
current for the case of the constant-off-time con-
troller. Obviously, the average inductor currents TA = ~I/2(m + m2) (A2)
will be different in the two cases.
Substituting (AI) into (A2) yields
Let the average current be represented by a
straight line which is half-way between the lines (A3)
of the peak and valley currents. The horizontal
distance is the delay time between the two average Knowing the slopes of the two sides of the ABC
currents; it can be derived with the help of Fig. triangle (m and -m2) and the horizontal projection
A2, which shows in enlarged form the central por- of side AB, which is:
tion of the falling current.
(A4)
Assume that the "off" period for both controllers
starts at the same instant. The distance between one can easily determine T d • The height of the
the turn-off instant and point B (the intersection triangle can be expressed with both slopes:
of the falling current waveform with the average
current for the constant off-time controller) is (AS)

784
 Solving (A5) for T and s u bstituti ng (A3) and [11] R. Redl, 1. Novak, "Instabil
d ities in Current­
(A4) into the result, we obtain Mode Control led S witchi ng Voltage Regulators,"
PES C '81 Record (IEEE Publi ca ti on 81CH1652-7),
(A6) pp. 17-28.

T he result is i n d epe n d en t of t he slope of t he [12] R. Redl, 1. Novak, " S t ab i li t y Analysis of

control signal ie• Therefore we can conclude that Con stant - Frequ enc y Curr ent-Mode Con trolled
there is always a difference in the time delays of Switching Voltag e Regu lators Operating Above
the controlling paths in the two types of control­ 50% Duty Ratio," PESC � Record (IEEE Publi­
lers; t h a t difference is Tof /2. And if ther e is a cation 82CH-1762-4), pp. 213-223.
�
tim e d elay in the average 1nductor c u rren t , the
sa me time delay appears in the inject ed curr e nt, [13] R. Red l , "Com parative Analys is of Overload
too. Pr ote ction Method s for Switchi n g -Mode Voltage
Regulators," Pro c . Third ESTEC Spa c e craf t
Power Con dit i oni ng Seminar (ESA Publication S P
126); 1977, pp. 155-164.
REFERENCES
[14] Shi-Ping Hsu et aI, " M od e llin g and Analysis of
11] C. De i s c h , "Simple Switching Control Method S w itc hing DC-to-DC Converters in Constant­
Changes Power C o n v e r t er into'a Cur r e nt Frequency Current-Programmed Mode," PESC '79
Source," PESC '78 R ec ord (IEEE Pu bl ic a t io n R ecord (IEEE Publication 79CH1461-3 A ES ), pp.
78CHI337-AES), � 300-306. 284-301.

[2] D. Di Cesare, M. Gambarara, "Boo s t LC3 at [IS] R. Redl, 1. Novak, " S t a bi li t y A n a ly s is of

Vari able Constant," Proc. Fourth ESTEC Space­ Constant-Frequency Cu r r ent- M ode Controlled
craft Power Conditioning Seminar (ESA Publi­ Power Converter s," Pr oc. Second Internat iona l
cation SP-186), 1982, pp. 87-90. Powerconversion Conf., 1980, Munich, pp. 4B2-1
- 4B2-I7.
[3] D. O'Sullivan , A. Wein berg, D. Levins, J.
3 116] R . Redl, N. o. S o ka l , "Near-Optimum Dynamic
Schreuders, "SLIC : The Sync hronis ed/Limit
Cycle Conducta nce Controller," PESC '80 R ecord Performance of Swi tch ing- Mode Power Converters

(IEEE Publi ca ti on 80CHI529-7), pp. 397-401. Using Feed-Forward of Output Current and Input
Voltage with Curre nt- M ode Control," Proc.
[4] H. Matsuo, K. Harada, "New DC-DC C o n v er ters Fourth ESTEC Spa c e c raft Power Conditioning
with an Ene rgy Storage Reactor, IEEE Transac­ Sem in ar (ESA Publi ca t i on SP-186), 1982, pp.
tions on Magnetics, vol. MAG-13, no.5, Septe m­ 3 3- 42 .
ber 1977, pp . 1211-1213.

[5] R. D. M iddlebrook, S. <:Uk, "A Ge neral Unified

Approach to Modelling Switching-Conve rter Pow­
er S ta ge s , " PESC � R e c o rd (IEEE P u b li ca ti on
76CHI084-3AES), pp. 18-34.

[6] M. C l i q ue, A. J. Fossard, "A G en er a l Model f or

S w itching Converter s," IEEE Trans . Aerospace
and Electronic Systems, July, 197 7 , pp. 397-
400.

[7] A. S. Kislovski, Dyna mical Analysis of S witch ­

ing DC-DC Converters, EWV Engineering, Bern,
Switzerland, Jan. 1985.

[8] R. R e d l, 1. No vak , "Current-Mode Control of

Switching Voltage Regulators, a New Method of
Improving Regulation Parameters," (in Hungar­
ian), Hi rad as�echnika, vol. 29 , no. 11, 1 978 ,
pp. 321-334. (Note: Engl i s h tra nsla t i o n
available upon request from author Redl; pres­
ent address at be g innin g of this text.)

[9] R. Redl, N. o. S ok a l , "Optim1zing Dynam1c

Behavior with In put an d Output Fe e d - F or wa rd
and Current-Mode C o n t ro l , " Proc. Power c on 2,
1980, San Diego, pp. Hl-l - Hl-16.

[IO] A. S. Kislovski, "General Sma ll-Signal Analy­

sis Method for Switching Regulators," PrOc.
PCI '82, March 29-31, 1982, San Fra ncisco, pp.
1-15.

785