10 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. IE-30, NO.

1, FEBRUARY 1983

Advances in Switched-Mode Power Conversion

Part I
SLOBODAN CUK, MEMBER, IEEE, AND R. D. MIDDLEBROOK, FELLOW, IEEE

Abstract-A new concept in the design of switched-mode power can efficiently raise or lower dc levels, and, with minor modi-
conversion circuitry is presented. Because of its extreme simplicity, fication, bidirectional power flow can be easily achieved, al-
flexibility, and efficiency, it has the potential to replace some con- lowing the roles of power source and load to be arbitrarily in-
ventional electrical power processing methods currently in use. A
thorough discussion of the background, development, and operation terchanged without physically switching their connections. A
of the Cuk converter and several of its extensions is included. The switching power amplifier based on the new converter requires
coupled-inductor technique further improves efficiency and reduces only a single power supply-, enabling efficient dc-to-ac power
size and weight. The basic converter and its extensions to two- conversion. The improved performance of the new switching
quadrant and four-quadrant operation are attractive for many appli- power amplifier design permits the use of a lower switching
cations, i.e., solar (dc) to utility line (ac) power conversion, variable
speed dc or ac motor drives, switching power supplies, uninterrupt- frequency and simple circuits for both the drive and feed-
ible power supplies, and many others. back.
We will begin with a discussion of switching converter de-
I. INTRODUCTION sign, leading to the development of the new converter. Next,
the design of a switched-mode power amplifier using the new
WITCHING POWER supplies and regulators ("switchers") converter and a discussion of its performance will be pre-
.Jhave come into widespread use in the last decade. Because sented.
of their much higher efficiency, smaller size and weight, and
relatively low cost, they are displacing conventional linear
power supplies, even at low power levels (about 25 W). The A. A Review ofSwitching Converter Design
design of switching converters has been extensively studied, In the basic design of the linear power supply shown in Fig.
and it is commonly believed that the designs in commercial 1, the output voltage V0 is regulated to a multiple of the refer-
use today employ the simplest possible switching structures ence voltage Vref. The difference between the unregulated in-
for dc-to-dc level conversion. put voltage and VO results in power dissipation in the transis-
However, even though high conversion efficiency has been tor, which, in high current supplies, can result in considerable
achieved, present switching converter designs possess several energy loss and heating. Consequently, the unregulated voltage
desirable characteristics. Rapid switching of the input or out- must be kept as low as possible while still allowing adequate
put currents can cause severe electromagnetic interference regulation.
(EMI) problems, requiring the addition of appropriate filters Switching power supplies are based on the principle that,
that increase both the complexity and cost of the circuit. Also, by alternately switching the transistor completely off and on,
the implementation of the power transistor switch requires its power dissipation can be held to a minimum. Passive energy
complex drive circuitry in configurations where the emitter storage elements, inductors, and capacitors can then be used to
voltage "floats" above ground. In switched-mode power ampli- transfer energy from the source to the load, performing the ap-
fier designs, additional problems are imposed by the require- prop-riate level conversion in the process.1
ments for dual (bipolar) power supplies, high switching fre- Fig. 2 shows the most commonly used switching configura-
quencies, and complex feedback circuitry. tion, referred to as the "buck" converter. The ideal switch S
A revised look at present converter designs and an analysis can be realized by the combination of bipolar transistor and
of switching circuit topology has led to the discovery of a new commutating diode as shown. In operation, the input voltage
design that retains all the desirable properties of conventional is connected by S to charge inductor L to the output current
designs, with none of their undesirable attributes. Both input necessary to produce the desired voltage V across load R. Once
and output currents are essentially nonpulsating dc, and, in V is attained, S disconnects the input and provides an alternate
fact, ripple can be reduced to zero. Furthermore, the new path for the inductor current, which then begins to decay. The
topology may be implemented with fewer parts than com- output capacitor C helps reduce the residual voltage ripple
parable solutions, and thus may be said to be an "optimal" caused by the switching. The cycle repeats at a fixed rate, and
design. the average voltage gain is equal to the duty ratio D (the frac-
Because of its simplicity and generality, the new converter
1 Recall from basic physics that energy is stored by an inductor in
the magnetic field produced by the current through its winding. A volt-
Manuscript received October 15, 1982. This article is reprinted from age applied across it either increases or decreases this current to ener-
Robotics Age Magazine. gize or deenergize the field. Conversely, the current through a capacitor
The authors are with the California Institute of Technology, Pasa- either charges or discharges the energy stored in an electric field, re-
dena, CA 91125. spectively raising or lowering the voltage across its terminals.

0278-0046/83/0200-0010$01.00 (© 1983 IEEE

 CUK AND MIDDLEBROOK: SWITCHED-MODE POWER CONVERSION PART I 11
+ VO

Vg- -Ci C

vg -
I2
U R

BOOST BUCK
Fig. 4. Cascade combination of a boost and a buck power stage.
Fig. 1. Conventional linear power conversion.
ples of simplification and optimum interconnection while
V9~
L v
~
L v
simultaneously maximizing performance. The details of the
analytical technique, described more completely in [1], will
c R
R~~~~~~~~
S
be abbreviated here.
Consider the cascade connection of a boost power stage
(a) (b) followed by a buck power stage, resulting in the converter
Fig. 2. The basic "buck"-type switching converter configuration, shown in Fig. 4. This configuration retains the desirable prop-
showing (a) the topology of the circuit, using an idealized switch
S, and (b) the implementation of the switch using a bipolar transistor erties of the low input current ripple of the boost converter
and a commutating diode. and the low output current ripple of the buck stage. The volt-
age gain of the converter is the product of the gains of the two
L v L
stages. Assuming that S, and S2 are synchronized so that both
switch from position I to 2 (and back) simultaneously, the
V

Vg= L3RCRR resulting ideal gain is D/D', where D' = 1 - D. Thus, the same
converter can be used for both level reduction (D < 0.5) and
(a) (b)
increase (D > 0.5).
Fig. 3. The common "boost' converter configuration. (a) Circuit
The undesirable output ripple of the boost converter is now
topology, using
an ideal switch, and (b) implementation. isolated between the stages, and, in fact, the capacitor serves
as the sole energy transfer mechanism between the stages. To
see this, note that during the part of the cycle when both
tion of the switching cycle that the transistor is on). switches are in position 2, C1 is isolated from the output cir-
Note that the input current to the regulator alternates be- cuit and is charged by the input current through L1 . In the
tween the full output current and zero. This abrupt variation rest of the cycle, C1 is completely transferred to the output
in the input energy flow causes severe EMI, and invariably re- circuit and is discharged by the output current through L2,
quires the presence of an input filter to smooth out the sub- reenergizing L2 while LI stores energy from the input. Posi-
stantial current ripple component at the switching frequency. tion 2 of S2 provides a path to sustain the output current
Also, since the emitter floats above ground, isolated drive cir- (from energy in L2) while C1 is being recharged by L 1.
cuitry (not shown) is required to switch the transistor. A The issue remains, however, as to whether or not this de-
further limitation of this design is that only voltage reduction sign represents the optimal configuration of this attractive
is possible, since D is less than 1. boost-buck cascade. It is apparent that the two inductors
For applications requiring a voltage step-up, some of the are essential to the continuity of the input and output cur-
drawbacks of the buck configuration can be avoided by using rents, and that the intermediate capacitor is required for
the "boost" converter configuration, shown in Fig. 3. In this energy transfer. The issue of optimality thus resolves to the
design, the inductor is always in series with the input, so that questions, can the number of switches in this cascade configu-
the supply current is continuous. The inductor stores energy ration be reduced from two to one and still achieve capacitive
when grounded by S, and then the stored energy is released to energy transfer?
the output. The output voltage of the boost converter is al- The answer to this question may at first seem surprising.
ways greater than the supply, with an ideal gain equal to /(1 Switches SI and S2 may indeed be combined, resulting in a
D). Note also that the grounded emitter of the transistor new optimal converter configuration. The solution is found by
switch simplifies the drive circuit. The drawback of the con- considering the topological properties of known converter
figuration, however, is that during the inductor charging in- types and their cascaded combinations [1], [2] . A key to the
terval, all the output current must be supplied by discharging solution is that inversion of the converter's output voltage is a
the output capacitor, resulting in considerable output ripple. necessary feature of the new design. Since both of the basic
converter types are noninverting, the only way to achieve this
II. DEVELOPMENT OF THE NEW CONVERTER is by reversing the polarity of the charged energy transfer capa-
TOPOLOGY citor when it is switched into the output circuit. Attention to
this issue alone leads to the solution shown in Fig. 5.
The discovery of the new converter design resulted from Even though a true dc-to-dc transformer is physically im-
the objective of retaining the desirable properties of both possible, the new converter can functionally be considered as
types of converters. This goal was realized by examining com- such, since both its input and output voltages and currents are
binations of the two basic types, with the underlying princi- very close to true dc quantities, owing to the negligible switch-
12 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. IE-30, NO. 1, FEBRUARY 1983

NEW CONVERTER TOPOLOGY

Li C, L2 -V
A -
R
ii 2
Vg tA B 0 R

Fig. 5. The new switching topology, employing capacitive energy

transfer with polarity inversion of the output voltage.

NEW SWITCHING DC-TO-DC CONVERTER

Li Ci L2 -V
ii ~~~~~~~~~~~2
Vg 02R R Fig. 7. The Coupled Inductor Extension of the Cuk converter. With
appropriate transformer design, the output current ripple can be
SWITCH DRIVE
reduced to zero.

DC VOLTAGE INPUT CURRENT

r~ ~ t DC CURRENT state balance condition), the two waveforms must be identical
GAIN GAIN
V D' L t 12 D
[3]. This means that the two inductors may be coupled by
Vg iD OUTPUT CURRENT 11 D being wound on the same core, without affecting the basic dc
12
t conversion property, provided that the resulting transformer
Fig. 6. Hardware realization of the Cuk converter using a transistor
has a 1:1 primary to secondary voltage ratio. This is easily
and a diode to implement switch S. Ideal dc voltage and current achieved by setting L1 = L2 (same number of turns in each),
gains are as shown. with the direction of coupling as shown in Fig. 7.
The resulting circuit not only eliminates one inductor core
ing ripple. Moreover, due to the advantages of capacitive en- (for further component savings), but has an interesting prop-
ergy transfer, the actual conversion efficiency of the new cir- erty that profoundly affects the performance of the new con-
cuit is substantially greater than that of conventional designs verter. Since the coupled inductors form a transformer, induc-
[1]. The grounded emitter of the switching transistor also al- tive energy transfer between the two windings alters the effec-
lows the most simple drive circuitry to be used. tive inductance of each. For a 1:1 turns ratio, both induct-
In this design, the single switch S alternately grounds the ances are approximately doubled, so that both input and out-
opposite ends of the capacitor, effectively switching it from put current ripples are about one half those of the uncoupled
the input to the output circuits. C1 is charged by the input converter. Even more significant, however, is that by slightly
current to a positive voltage, as viewed from left to right, with altering the turns ratio, one of the effective inductance values
S in position B. With S switched to A (during the charging in- can be made arbitarily large, dramatically reducing the cur-
terval of L1), the "positive" side of C1 is now connected to rent ripple on that side. When the effective turns ratio (pri-
ground, and its "negative" side (since the voltage across it does mary to secondary) matches the inductive coupling coeffi-
not change) effectively "pulls down" the voltage level at termi- cient of the transformer,2 the output current ripple can be
nal B. Thus, current flows from the grounded load to discharge completely eliminated, resulting in pure dc to the load [3] . In
C1 through L2, causing a negative voltage drop across the load, this matching condition, no output capacitor is needed, since
hence a negative output voltage. the load voltage is also constant.
The hardware implementation of the new converter, using a The coupled-inductor converter thus has the simplest possi-
transistor/diode combination to implement the ideal switch S ble structure, consisting of a single transfonner, commutation
as in the other converter configurations, is shown in Fig. 6. capacitor, and a single switch (realized by two semicon-
With the transistor off (open), C1 is charged by the input cur- ductors), and yet it achieves the maximum performance (both
rent through the forward biased diode. When the transistor input and output current nonpulsating with one of them even
turns on, it provides a charging path for L1, but also drops the being pure dc with no ripple) in a topology which offers the
positive terminal of C1 to (very near) ground. As in the ideal smallest possible size and weight and highest efficiency.
case, this pulls down the output voltage, reversing the bias on It is interesting to note that the idea of coupling the in-
the diode (turning it off) and transferring the stored energy to ductors is a new concept which can be implemented with simi-
L2 in the form of increased current. Note that the dual role of lar benefits to many other switching converter configurations.
switch S, its inclusion in the input and output circuits simul- Nevertheless, the greatest potential of this configuration is ob-
taneously, requires that the transistor and the diode, when tained through its application to the new converter topology,
conducting, must carry both the input and the output cur- on which it was originally conceived.
rents. Yet, this still seems to fall short of the ultimate objective,
the functional realization of the ideal dc-to-dc transformer
A. Coupled Inductor Extension of the New Converter shown schematically in Fig. 8, which has constant current at
It would seem from the above discussion that the simplest both input and output. In addition, the dc isolation between
possible converter circuit has been obtained, but this is not the
2 Like the effective turns ratio, the inductive coupling coefficient is
case. Consideration of the voltage waveforms across the two
a physical property of the transformer construction, with a value ranging
inductors L1 and L2 over the switching cycle reveals that, for from 0 (separate inductors) to very near 1 (windings closely wound on
the average dc voltage across each inductor to be zero (steady- a single ferromagnetic core).
 +

Vin
°
ii
--

,_. , ~ ~ ~I_
CUK AND MIDDLEBROOK: SWITCHED-MODE POWER CONVERSION PART I

I -

1
t

M (D)
i2
t
+

Vout

Fig. 8. The ultimate objective of switching conversion-A dc-to-dc

transformer with constant input and output currents and an elec-
tronically controllable effective turns ratio for voltage level change.

practical applications, is not present in any of the converter

extensions discussed so far. This, however, poses no insur-
mountable problems, since there is an elegant way of intro-
ducing dc isolation into the new converter by the addition of a
singe ac transformer and an extra capacitor, as described in
detail in [4]. Furthermore, the same idea of coupling in-
BIDIRECTIONAL POWER FLOW

r5
-5V
0-
r
go__

DUTY RATIO
CONTROLLED SWITCH

input and output shown in Fig. 8, often required in many Fig. 9. A symmetrical implementation of the Cuk converter, capable-
of power transfer in either direction.

L1:
s s
13

ductors in this dcisolated extension leads also to zero current N CHANNEL L. -x _ P CHANNEL
ripple on one side.
The latest generalization of the new coupled-inductor con-
verter concept has resulted in what may be considered the ulti- Fig.
10. Bidirectional switching converter using MOSFET power
switches.
mate solution: a dc-isolated switching dc-to-dc converter with
zero current ripple at both input and output terminals, which a load normally
requires the transfer of power to the load.
functionally emulates the ideal dc-to-dc transformer of Fig. However, when two sources of stored energy are coupled by a
8.3 bidirectional converter, precise control of the duty ratio is es-
sential to restrict the resulting power flow to an acceptable
B. Bidirectional Power Flow in the New Converter
limit. The bidirectional current switch implementation is
Note that the configuration of the new converter shown in equally applicable to the coupled-inductor extension (Fig.
Fig. 5 is completely symmetrical with respect to the designa- of the new converter. 7)
tion of the input and output terminals. Because the ideal Since both of the transistors in Fig. 9 are referenced to
switch S allows current flow in either direction, either termi- ground, the complementary switching of the pair can be ac-
nal of the converter can behave as a current source or as a cur- complished by a single drive source as shown.
rent sink. However, when the switch is implemented by a because the two base junctions are tied
Moreover,
together, the circuit
single transistor and diode, only unidirectional current flow is also automatically prevents the simultaneous turnon of both
allowed. The addition of a single p-n-p transistor and a diode transistors (and thus prevents shorting out capacitor C), in
removes this constraint and results in bidirectional current and spite of the presence of transistor switch storage time.
power flow as shown in Fig. 9. Another implementatation of the bidirectional current
The converter circuit is thus symmetrical, and the input and switch is made possible by recent technological advances in
output terminals can be arbitrarily designated (as long as the metal oxide semiconductor field effect transistors
voltage polarities are respected). The configuration shown in FET's). Formerly limited mostly to small signal applications, (MOS-
Fig. 9 is ideal for battery charger/discharger applications, since newer MOSFET devices are capable of switching to a relatively
both functions are accomplished by this single converter cir- low ON resistance, with a substantially higher current rating. It
cuit [5]. The direction of power flow through the converter is not widely known that the power MOSFET's are capable of
is determined by whether the duty ratio is greater or less than bidirectional current flow. Owing to the device's internal con-
the value that gives a voltage gain equal to the ratio of the bus struction, there is effectively an inherent diode connected be-
to battery voltages. In general, attempting to reduce the volt- tween drain and source which provides the alternate
age to a load that stores energy causes power to be transferred current
(opposing)
path. Hence, each transistor/diode pair can be replaced
from the load,4 just as attempting to raise the voltage across by a single power MOSFET device, resulting in the bidirect-
ional converter shown in Fig. 10. If both p-channel and
3 Technical details of this latest development have not yet been n-channel MOSFET's are
publicly released, but are contained in a patent application. The basic used, the source terminals of both
Cuk converter configuration and many of its improvements, as well as devices may be grounded, allowing a simple driving scheme
new switching power amplifier configurations, are protected by a series similar to that used for the grounded emitter transistors in Fig.
of patents [91-[12].
4 Editor's note: A very useful application of this is to use a dc drive 9-
motor as a generator for regenerative braking, recovering some of the Besides reducing the component count, power MOSFET's
energy stored in a robot's momentum or provided by an external source have a number of advantages over bipolar transistors which
(as when rolling down a slope). make them especially attractive for switching converter appli-
14 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. IE-30, NO. 1, FEBRUARY 1983

VL
cations [6]. Whereas bipolars are current-controlled devices, L V
FET's are voltage controlled. As a consequence, they require V.
much lower drive currents and are capable of very fast switch-
ing speeds. Also, unlike bipolars, several MOSFET's can be +Vgf0 T-g c=. R

easily paralleled to distribute switching current. Their switch- Vg-V

ing speed (5 ns typical) is one or two orders of magnitude - V =D-D'
V9
faster than comparable bipolars (200 ns typical), significantly Fig. 11. Modified buck power stage with output voltage of either
reducing power loss during switching. Although older power polarity.
MOSFET's had a high saturation voltage compared to bi-
polars (3 V at 10 A versus 2 V or less for a bipolar), the newest v
devices have as little as 0.05 5 Q2 RSD, making their use in con- +1.0i Vg
verter designs extremely attractive, especially for applications
requiring a high switching frequency.
D
0.5 1.0
III. APPLICATION OF THE NEW CONVERTER CIRCUIT
TO SWITCHED-MODE POWER AMPLIFIER DESIGN
-1.0
Although a substantial effort has been made in recent years
toward the development of complex switching power supplies, Fig. 12. DC voltage gain of the power stage in Fig. 11.
substantially less attention has been devoted to their natural
outgrowth-switching power amplifiers. It is quite natural, D2
then, that the principles of the operation of switching ampli- Vi L V+
fiers are not widely known. We will begin, therefore, with a re-
view of present designs, which are all based on a modification +Vg 02 D Rl
of the "buck" power stage of Fig. 2. This review reveals sub-
stantial performance deficiencies originating from the buck
converter itself.
The new circuit topology of Fig. 5 eliminates most of the Fig. 13. Practical implementation of the converter in Fig. 11.
problems associated with earlier switching converter designs, +Vg
providing greatly improved performance with fewer com- SIGNAL POWER
OUTPUT
INPUT COMPARATOR
ponents. This, together with the extension of the new design z2 / xN v
D
L
to allow bidirectional current flow, suggests that the design of
switching amplifiers should be reexamined in light of this new
development. Indeed, we will show that a power amplifier de- T DRIVER 02 D0 C R
sign based on the new converter topology provides a much 14Th ITCh -vg LOAD-
superior solution. CLOCKED
RAMP
PWM SIGNAL POWER STAGE

A. SwitchingAmplifier Principles of Operation Fig. 14. Open-loop buck-type switching power amplifier.
The major difference between the power stages of a switch-
ing amplifier and a switching power supply is that the former To apply this dual-polarity power stage in a switching
must be capable of producing an output of either polarity. In power amplifier, it is necessary to control the duty cycle so
all- present designs, this is accomplished by modifying the basic that the output voltage varies in proportion to the input signal.
buck power stage to use two power supplies +Vg and -Vg as The linear voltage gain of the buck power stage facilitates this
input, with the switch S switching between positive and nega- function, and is incorporated into the open4oop amplifier con-
tive supplies as shown in Fig. 11. figuration shown in Fig. 14. The design is the same as that for
The resulting voltage gain, shown in Fig. 12, is a linear an open-loop dc-to-dc converter operated at a constant switch-
function of the duty ratio D. For D greater than 0.5, the out- ing frequency f = l/Ts, with the only difference that a time
put voltage is positive, while for D less than 0.5, it is negative. varying (sinusoidal, for instance) input signal is used at the
Note that since the load voltage may be negative as well as comparator input, instead of a dc reference voltage. When the
positive, the implementation of switch S shown in Fig. 2 is input signal is positive, a pulse with D greater than 0.5 is gen-
inadequate. Since the output voltage is determined by the erated, producing positive output, while for negative input, D
inductor current, the hardware implementation of the switch is less than 0.5 and negative output is produced. In fact, com-
must permit bidirectional current flow, as shown by the parison of the low-frequency input signal and the high-fre-
arrows in Fig. 11. This is readily accomplished by the two- quency sawtooth (clocked ramp), generates a pulse width
transistor, two-diode circuit shown in Fig. 13. This bidirec- modulated (PWM) signal, whose low frequency spectrum is,
tional implementation is similar to that required for the sym- in effect, recovered by low-pass filtering with the inductor.
metrical converter in Fig. 9, but with the significant difference Hence, a close replica of the input signal is generated at the
that neither transistor is referenced to ground, necessitating output, but at a high power level.
the use of isolated drive circuitry to accomplish the comple- The comparison of this switching amplifier approach with
mentary switching action. conventional linear designs with respect to the two foremost
CUK AND MIDDLEBROOK: SWITCHED-MODE POWER CONVERSION PART I 15
Vg. -v POWER FLOW
L Ci L

i2 A2 , 2 B V2T R2

L Ci L
- POWER FLOW
Fig. 16. Two bidirectional Cuk converters operating in parallel from
COMPARATOR YYJ, -,LOOP COMP - a single power supply.
SAWTOOTH
Fig. 15. Closed-loop buck-type switching power amplifier.

constraints in power amplifier design, efficiency, and distor-

tion, now becomes apparent. In terns of efficiency, this ap-
proach boasts the usual advantages of switching power supplies
over linear-significantly lower power dissipation. Namely, its
theoretical 100-percent; efficiency is usually only slightly de-
graded (often it is over 90 percent) by losses due to nonzero
transistor saturation voltage and switching time and parasitic
resistances of storage elements in the power path.
Distortion, however, becomes a function of the switching
frequency, rather than being dependent on the linearity of the
transistor gain curve. Specifically, for low distortion, the
switching frequency has to be an order of magnitude or so
higher than the signal frequency to avoid overlapping side-
bands in the PWM signal and to minimize the effects of out- Fig. 17. Differential voltage gain V/Vs for the power stage in Fig. 16.
put ripple. On the other hand, with increased switching fre-
quency, the switching time of the transistor may represent a voltage of either polarity, depending upon a switch duty ratio
significant portion of the duty cycle, introducing further dis- D. The design should preserve all the good properties of the
tortion and degrading efficiency as well. Other distortion new converter previously described and possibly add some
sources arise from nonlinearities in the clocked ramp (which more. Since the use of dual power supphes, as in the buck con-
also increases with switching frequency) and variations in the verter power amplifier, adds to the expense and complexity of
power supply voltages. the design, the additional feature of a single power supply is
To reduce the effects of these sources of distortion requires desired. Consideration of this requirement led to the discovery
the use of negative feedback, as shown in the block diagram in of the push-pull type topology shown in Fig. 16, in which two
Fig. 15. Feedback allows a reduced switching frequency and bidirectional switching converters are operated in parallel
improves linearity at the expense of increased circuit com- from a single supply.
plexity, but the amplifier in Fig. 15 still has drawbacks that Let us assume that the two converters are operated switch-
onginate directly from the use of the buck power stage. As in ing out of phase, that is, with complementary drive ratios.
the buck power supply design, the input currents pulsate, at
the switching frequency, between zero and the output current, Namely, when switch S, is in position A1 for interval DTs,
causing severe EMI. The design requires two power supplies of switch S2 is in position B2 for the same interval. Assuming
opposite polarity, and complex drive circuitry is required for
normal operating conditions, the ideal voltage gain of the top
the transistors, both to translate dnrve signals to the un- stage is DID', and that of the bottom is D'/D. Thus, the two
output voltages are equal only for a duty ratio of 0.5, while
grounded emitters, and to prevent simultaneous turn on of the one or the other becomes greater for other values of D. Eval-
transistors and shorting of the power supplies. Finally, a rela-
uating the difference of the two output voltages V - V1 -
tively high switching frequency (300 kHz or so) is still neces- gives a differential voltage gain of
sary to reduce switching ripple.
As in the case of power supply design, the introduction of V D-D
the new switching converter topology solves all of these prob- Vg DD'
lems, but the objective of achieving maximum performance
(wide bandwidth, low noise and distortion, small (or zero) which is plotted as a function of duty ratio D in Fig. 17 (heavy
switching ripple, and low switching frequency), for the mini- line). The individual converter gains are shown as dotted lines.
mum number and size of parts, is by no means an easy one. As seen in the figure, the differential gain is just the one
needed for a switching power amplifier, since it has the same
B. A New Push-Pull SwitchingPower Amplifier required polarity change property as that of the modified buck
We now pose the problem of inventing a power stage, based power stage (Fig. 11). The only trouble, however, is that the
on the new converter of Fig. 5, which will produce an output load is not across the two converter outputs. Thus, an interest-
16 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. IE-30, NO. 1, FEBRUARY 1983

L Ci L
POWER FLOW (SOURCE)
Fig. 18. The Cuk push-pull switching power stage.

ii Ci

I I I, vv
PWM SiGNAL SAWTOOTH -SIGNAL
AUDIO INPUT
Fig. 20. Cuk push-pul switching power amplifier.

L Ci L

Fig. 19. Hardware implementation of the Cuk power stage.

SWITCHING DC-TO-DC
CONVERTER WITH
BIDIRECTIONAL CURRENT
ing question arises. Is it possible to connect a load between the CAPABILITY
OUT OF
two outputs without violating any basic circuit laws or disturb- PHASE Vg.
ing the proper operation of the converters? SWITCH
DRIVES
The answer to this question is affirmative, and is a key to _ SWITCHING DC-TO-DC
CONVERTER WITH
the success of the new push-pull, switching power amplifier de- BIDIRECTIONAL CURRENT
CAPABILITY
sign. With the two loads in the converter of Fig. 16 replaced
by a differential ("floating") load R, the new push-pull power
stage of Fig. 18 is obtained. Fig. 21. New generalized concept of constructing a switching amp-
In Fig. 16, each converter operates independently,withuni- lifier power stage from any dc-to-dc converter type.
directional current and power flow as shown. However, this is
not so in the new push-pull power stage. Owing to the differ- The most advantageous configuration is obtained, however,
ential load, current originating from one converter must be when the coupled-inductor extension of the power stage in
"sunk" by the other, resulting in equal and opposite current Fig. 19 is used. The resulting design is shown in Fig. 20, which
flow, as shown by the solid arrows in Fig. 18. With the op-- also represents a complete block diagram of a closed-loop
posite polarity of the output current, (dotted lines) the roles switching power amplifier using the new push-pull configura-
are reversed. Thus, the switches S1 and S2 must permit this bi- tion. When the two transformers (coupled-inductors) are de-
directional flow, depending upon the duty ratio D. In other signed to satisfy the matching condition, as described earlier,
words, a part of the energy delivered by one converter is con- the output current ripple, and consequently the need for out-
sumed by the load, and the remainder returned by the other to put capacitors, is completely eliminated. Removing the capa-
the source. This bidirectional flow is easily accomplished, as citors results in extremely favorable phase-frequency response
discussed previously, and as shown in Fig. 19. and permits closing the feedback loop directly, even without
It may now become evident that the new amplifier power any compensation network, and yet with a high degree of sta-
stage may be called a true push-pull power stage. Namely, bility. Also, there is no longer any need for an excessively high
while the lower converter pushes the current (and energy) switching frequency to reduce output ripple, thus resulting
through the load, the upper converter pulls if from the load in further improvement.
and vice versa. This is quite unlike the conventional push-pull It should be emphasized that the new switching amplifier of
class B linear amplifiers for which a push-pull configuration Fig. 19 is actually based on a new concept of significantly
would be a more appropriate designation. broader scope. Namely, this novel technique of connecting a
In addition to the advantages of the bidirectional new con- single dc source to a true push-pull configuration with a dif-
verter already described, another very desirable feature de- ferentially connected load may be implemented using any
rives from this true push-pull configuration itself. Note that other converter type, as illustrated in Fig. 21, provided that
since one converter operates as a power sink, the current the converters are capable of bidirectional current flow and are
drawn from the source by the other is reduced by the return driven with complementary (out of phase) switching drives
current (i2 in Fig. 18) which is increasing when S2 is in posi- [10]. In particular, when this configuration is implemented
tion B2 as shown. (The capacitor is discharging return power using two buck-type converters, a single-source power ampli-
into the power supply inductor of converter 2.) Since current fier is obtained even with the buck power stage.
il is increasing at the same time, the current ripple through the
power supply is reduced. In fact, if the two inductors on the C. Performance of the New Converter and Amplifier
supply side are equal, the current drawn from the power sup- The new switching amplifier design retains all of the opti-
ply is pure dc with no ripple at all. mal characteristics of the new converter topology, while simul-
CUK AND MIDDLEBROOK: SWITCHED-MODE POWER CONVERSION PART I 17

DC GAIN k V (D)
5
4
, /~ ~ ~ ~ ~ ~ ~,
IDEAL VOLTAGE GAIN
5 - V9
4 IDEAL

3 3
* ** MEASUREMENT
DATA 2
2
REAL 1
1 VOLTAGE GAIN D
i .l t t D
U
II
0.5
A c 15
-11 .u 05 1.0
-1
Fig. 22. Theoretical and actual dc voltage gain characteristics of a
typical test circuit. Experimental measurements confirm the pre- -2
dicted gain curve.
-3
REAL DC GAIN
-4 (PARASITICS INCLUDED)
taneously offering improvements due to the push-pull config- /a, - Q.0 a2 0.01
=

uration itself. However, one important property of the new de- -5

sign, the linearity of the voltage gain, has not been discussed. Fig. 23. Effective differential gain curve when parasitic resistances
Note that although the gain curves of the individual converters are considered.
in the new design are highly nonlinear, as shown by the dotted
lines in Fig. 17, the differential voltage is very linear near the for the ideal case (both a, and a2 zero) is fairly linear about
duty ratio D = 0.5, just where it is needed the most-in the the center of the operating region D = 0.5. For example, if the
center of its useful dynamic range. Nonetheless, sources of dis- variation of the duty ratios in the PWM signal is limited to
tortion, as well as the effects of nonideal circuit elements, ±0.1 (around D = 0.5), the resulting total harmonic distortion
must now be addressed to more accurately characterize the (THD) in the output will be limited to only 1 percent [8].
performance of the new power stage. However, a variation of ±0.2 produces 4-percent distortion,
The new converter design has been extensively analyzed and increasingly higher THD values result from broader ex-
and experimentally verified. Limitations to the theoretical dc cursions.
voltage gain arise due primarily to the effects of the parasitic Furthermore, as can be seen from the figure, the parasitic
resistance in the windings of the input and output inductors. resistances tend to further linearize the gain around D = 0.5.
Consideration of these tenns in the analytical circuit model In fact, the values of these resistances can be adjusted to mini-
leads to a predicted voltage gain curve of mize the distortion over the entire operating region. By add-
ing a small resistance in series with the input inductor so that
V DF 1 the value of a1 matches the optimality criterion a, = 0.0714 X
Vg D' (1 + Za2), THD can be reduced to well below 0.1 percent over
+a,(D) ;a2 the range D = 0.5 + 0.2 [8]. Thus, almost perfect linearity in
the dc gain curve is achieved even without the use of feedback.
where a1 and a2 are the ratios of the series resistances of the Although adding an input resistor results in slightly degraded
input and the output inductors to the load resistance, respec- efficiency, test circuits still perform in the 90-percent range if
tively, and R is the equivalent resistance of the load. A plot of the variation in the duty ratio is suitably limited. This need
this curve for a test circuit is shown in Fig. 22, and conforms pose no unreasonable restriction, however, if the transformer-
well to the measured experimental results. isolated version of the converter is used [4].
The current gain is unaffected and remains equal to D'ID Equally significant is the effect of the series resistance on
as before. Thus, the power conversion efficiency for a given the frequency response of the new converter circuits. Al-
duty ratio is equal to the bracketed term in the above expres- though each converter stage has three energy storage elements,
sion. As the duty ratio approaches one, the power dissipated in which could theoretically introduce complications due to mul-
the inductor grows significant and reduces the operating ef- tiple resonances, the damping provided by the input resistance
ficiency of the circuit. The power loss due to the output re- effectively produces a highly favorable "single-pole" frequency
sistance is independent of the operating point, and becomes response [8]. This completely eliminates feedback stabiliza-
the most significant detrimental factor when D is less than 0.5 tion problems and allows closing the loop without a compensa-
(voltage reduction). tion network, greatly simplifying design and resulting in even
This model does not consider the effects of other resistive further savings in circuit size, weight, and cost.
losses that occur because of nonzero voltage drops across the
semiconductors (when operating) and the finite switching time IV. CONCLUSIONS
they require to become fully conducting. These losses may be- We have shown how a fundamentally new design in switch-
come significant in applications involving relatively low input ing regulator circuit topology has resulted in dramatically im-
or output voltages or high switching frequencies. Nevertheless, proved performance. The new converter not only offers higher
all these effects can easily be included analytically by use of efficiency, low or zero output ripple, and greatly reduced EMI,
the modelling methods described in [7] . but at the same time achieves the general conversion func-
When the effects of the parasitic resistances on the perform- tion-it is capable of either increasing or decreasing the output
ance of the push-pull power stage are included, the differential voltage depending upon the duty ratio of the switching transis-
gain curve of Fig. 23 results. As mentioned earlier, the curve tor. The new topology uses capacitive energy transfer between
 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. IE-30, NO. 1, FEBRUARY 1983
Ci L

ONE QUADRANT
(a)
bidirectionc power flow

K -V

I~~~~~~~~~~~~~~~~~~~~~~~~~
.

Fig. 24. Prototype version of a Cuk switched mode audio power amp-
(b)
lifier. The circuit delivers 40 W rms to an 8 S1 speaker load, with a
flat frequency response over the 20-Hz-20KHz range, and over 90
percent efficiency. A recently built model has been reduced to less
than one quarter this size.
-I

the input and output stages rather than the inductive energy
transfer of other converters, resulting in nonpulsating input -V L
and output current. Its implementation requires fewer parts (C)
and simpler drive circuitry, with attendant savings in circuit Fig. 25. (a) Basic Cuk converter with one-quadrant classification. (b)
size, weight, and cost. Battery charger/discharger with two-quadrant classificatiox). (c) Bi-
The excellent frequency response characteristics of the new directional power amplifier with four-quadrant classificationi.
design allow highly stable feedback regulation to be achieved
with simple circuitry. As a result, the basic converter may be Simply by doubling the number of power devices in the
used in a new switching power amplifier design that offers per- Cuk converter of Fig. 25(a), the output capability of a dc-to-
formance comparable to linear amplifiers but at much higher dc converter can be extended so that current can flow in either
efficiency and lower cost-offering, for the first time, efficient direction (bidirectional current switch irnplementation), and
dc-to-ac power conversion. the system becomes a two-quadrant converter as shown in
Thus, the new optimum topology converter superior
is to Fig. 25(b). Thus, INPUT and OUTPUT can be arbitrarily inter-
any of the currently known switching converters, out. perform- changed, and the converter is capable of bidirectional power
ing them in every respect (Fig. 24). flow.
Finally, by the new general concept of converter topologi-
APPE.NDIX cal interconnection illustrated in Fig. 21, a two-quadrant con-
A GENERAL AND UNIFIED APPROACH TO POWER verter is extended into a four-quadrant converter in which the
CONVERTER TOPOLOGY directions of both voltage and current can change independ-
A. Modular Concept ently as shown in Fig. 25(c). Hence, a true ac output can be
A large number of switching configurations are currently obtained together with bidirectional power flow.
available for a multitude of power conversion functions. Here- Although this vertical integration has been conceived for
tofore, some circuit configurations have been used solely to the family of Cuk converters, it can be directly applied to
perform the dc-to-dc conversions function (converters), others other converter types, such as buck, boost, or any other
for dc-to-ac conversion (inverters), and yet another for power switching configuration.
amplificiation, with their circuit configurations having nothing
or very little in common. With the family of Cuk converters, C. Applications
however, a vertical integration has been made, and for the first The Cuk converters have a broad range of applications in all
time a logical extension from a single-quadrant to a two-qua- aspects of power processing. Robotics, in particular, will bene-
drant converter (battery charger/discharger) and finally to a fit from the excellent performance of Cuk converters in motor
four-quadrant converter (bidirectional power amplifier) has contTol. Its high efficiency and low noise make it an obvious
been realized. choice as a dc motor servo drive, with regenerative braking,
B. Quadrant Classification in either the two- or four-quadrant configurations. In fact, the
Power processing systems may be classified according to the high performance of the Cuk bipolar amplifier makes it possi-
nature of their output capabilities. The simplest is a dc-to-dc ble to design servo-systems employing ac induction motors,
converter in whieh output current is delivered in one direction capable of optimum motor drive, as well as regenerative brak-
at one output voltage polarity. This means that on a graph of ing.
output voltage V versus output current 1, only one quadrant is
available, as shown for positive V and I in Fig. 25(a). This is ACKNOWLEDGMENT
structurally the simplest switched-mode converter, and can be Special thanks go to R. Erickson and W. Behen, members
realized with only one transistor and one diode, as in the Cuk of the Power Electronics Group at the California Institute of
converter illustrated. Technology, for many devoted hours spent in building the first
 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. IE-30, NO. 1, FEBRUARY 1983 19

prototype of the new switching power amplifier and the later output extensions of a new optimum topology switching dc-to-dc
improvements. converter," in 1978 IEEE Power Electronics Specialists Conf.
Rec., pp. 256-264.
[51 R. D. Middlebrook, Slobodan Cuk, and W. Behen, "A new battery
REFERENCES charger/discharger converter," in 1978 IEEE Power Electronics
Specialists Conf. Rec., pp. 251-255.
[1] Slobodan Cuk and R. D. Middlebrook, "A new optimum topology [6] L. Shaeffer, "VMOS-A breakthrough in power MOSFET tech-
switching dc-to-dc converter," in 1977 IEEE Power Elec. Spe- nology," Siliconix Applications Note AN76-3, Santa Clara, CA,
cialists Conif. Rec., pp. 160-179. 1976.
[2] Slobodan Cuk, "Modelling, analysis, and design of switching [7] R. D. Middlebrook and Slobodan Cuk, "Modelling and analysis
converters," Ph.D. thesis, California Inst. Technology, Nov. methods for dc-to-dc switching converters," (invited review paper)
1976. Also, NASA Rep. CR-135174. in 1977 IEEE Int. Semiconductor Power Converter Conf. Rec., pp.
[31 Slobodan Cuk and R. D. Middlebrook, "Coupled-inductor and 90-111.
other extensions of a new optimum topology switching dc-to-dc [8] Slobodan Cuk and Robert W. Erickson, "A conceptually new high-
converter," in 1977 IEEE Industry Applications Soc. Ann. Meet. frequency switched-mode amplifier technique eliminates current
Rec., pp. 1110-1126. ripple," in Proc. Fifth Nat. Solid-State Power Conversion Conf.,
[4] R. D. Middlebrook and Slobodan Cuk, "Isolation and multiple May 1978, pp. G3.I-G3.22.

Advances in Switched-Mode Power Conversion

Part II
SLOBODAN CUK, MEMBER, IEEE, AND R. D. MIDDLEBROOK, FELLOW, IEEE

Abstract-A number of important practical extensions to the basic avoiding the electrical noise problems associated with switch-
Cuk converter are presented. They include dc isolation, multiple- ing either the input or output current and resulting in in-
output power sources, and a physical realization of the sought for creased conversion efficiency. The new converter has the de-
hypothetical dc-to-dc transformer, a device which converts from pure
dc (no voltage or current ripple) at one terminal, to pure dc (at a sirable property that its output voltage can be either higher or
different voltage) at the other terminal. The application of the circuit lower than that of the input supply, as determined by the duty
in a highly efficient amplifier for the servo control of a dc motor or ratio of the switching transistor (the fraction D of the switch-
other loads is also presented. ing period (Ti) that the transistor is turned on).
Also presented in Part I were several important extensions to
1. INTRODUCTION the basic converter design. These include adding the capability
IN PART I of this series we provided a review of the basic for bidirectional power flow between input and output, coup-
types of switched-mode power converters and showed how ling the input and output inductors via a single transformer
the effort to solve the characteristic problems of these earlier core, and a high-performance switched-mode power amplifier
designs led to the development of a fundamentally new con- configuration using parallel converters driving a differentially
verter configuration. Fig. 1 shows a summary of the basic con- connected load.
verter types and a physical realization of the new converter An important result of the coupled-inductor extension of
in its simplest forn. The new converter topology embodies all the new converter, apart from the further reduction in the
of the desirable features of previous types while retaining none number of components, is that, by proper adjustment of the
of their liabilities. Whereas other converters rely upon the in- magnetic coupling between the input and the output induc-
ductive coupling of energy between the input and output, the tors, the residual current and ripple at one of the terminals can
new converter uses a capacitor to transfer stored energy be- be reduced exactly to zero, resulting in pure dc. Naturally, this
tween input and output inductors as shown. motivated the search for a converter configuration which
Due to the presence of inductors on both the input and would achieve the desired characteristic of having zero current
output, the current on either terminal remains continuous ripple at both input and output simultaneously, thus resulting
in a physical realization of an ideal dc-to-dc "transformer."
Manuscript received October 15, 1982. (See Fig. 8 in Part I.)
The authors are with the California Institute of Technology, Pasa-
dena, CA 91125. In this article, we will show how, by pursuing the desirable
This article is reprinted from Robotics Age Magazine. property of dc isolation between the input and the output cir-
0278-0046/83/0200-0019$01 .00 © 1983 IEEE