Chapter 1 Introduction
1.3 Dissertation Overview
In this dissertation, five chapters in total are included, and each of them is briefly
described as follows.
In Chapter 2, the conventional feedback circuit for isolated switch-mode power
converters will first be reviewed. After the operating principles are explained, the
essential difficulty suffered by the conventional circuit will be indicated. Also, in this
chapter, previous related researches will be analyzed, and their advantages and
disadvantages will be pointed out.
Chapter 3 presents the proposed feedback network to address the power loss issue.
The proposed solution aims at minimizing its standby power consumption while
ensuring feasible compensation of control loop. The central concept is provided first
and then is followed by system and circuit design considerations. At last, the power loss
and the control loop compensation method of the proposed feedback network will be
analyzed, respectively.
All of the materials related to experiments will be given in Chapter 4. Contents
include the design and fabrication of integrated circuits, the implement of the proposed
and the conventional systems for testing and comparison, measurement approaches,
experimental results, and discussions on the outcomes.
Finally, Chapter 5 summarizes this dissertation and gives ideas for future work.
Chapter 2
Conventional Isolated Feedback Network and Previous Researches
2.1 Introduction
This chapter comprises two major parts. The first part introduces the basic
knowledge of the conventional isolated feedback network, including the operating
principles and the compensation method. Since what we care the most about is the
power loss when a power converter operates under very light/no-load conditions, the
power analysis of the feedback network will be carried out as well. The second part
includes three recent techniques that can help reduce the power dissipation of the
feedback network under very light/no-load conditions. Both their advantages and
disadvantages will be discussed.
2.2 Conventional Feedback Network 2.2.1 Architecture
Generally, in order to meet the safety regulations (e.g., IEC 60950) for safety
concerns, the outputs of power supplies must be kept insulated from inputs to ensure
galvanic isolation. For power stages, it will not be a problem since we can easily choose
those transformer-isolated topologies, such as flyback and forward topologies where
their secondary sides are inherently isolated from their primary sides. But, for the
control loop to feedback the output information and acquire a stable system control,
additional efforts and cost should be paid to prevent electrical contact between grounds
on the input and output sides. Among contactless signal transmission techniques, the
magnetic flux coupling through a transformer and the AC (alternating current) signal
coupling through a capacitor are not favorable because in this case it is supposed to
feedback a very-low-frequency analog signal. Instead, optical coupling through an
optocoupler proves to be a cost-effective approach to transmit such a feedback signal. A
typical optocoupler is most likely composed of an infrared light-emitting-diode (LED)
and a phototransistor, which are encapsulated into one same package. The strength of
the emitted light from the LED will be determined by the current flowing through it,
and the phototransistor will convert the light that reaches its base terminal into its
collector current.
Fig 2.1 shows a transformer-isolated power converter with a conventional feedback
network where an optocoupler is used as an interface of signal transmission. A shunt
regulator rather than an operational amplifier is placed in series with the optocoupler to
pull down an error signal of current for flowing through the LED inside that optocoupler.
In comparison to a standard operational amplifier, a shunt regulator is a low-cost single
IC with simply three pin connections such that it is overwhelming for applications in
Fig. 2.1. Conventional feedback network in a transformer-isolated topology.
power conversion. Its internal circuit structure can be viewed as an operational amplifier
with its output driving an npn bipolar transistor, which makes its output capable of
sinking current only. The inverting terminal of the internal operational amplifier is
connected to a built-in reference voltage. When the voltage on the non-inverting
terminal is below the reference voltage, the npn transistor remains open-circuit and the
shunt regulator is transparent to the circuit. As long as the voltage exceeds the reference,
the transistor will begin to conduct.
In Fig. 2.1, the input voltage VIN can be from the previous power-factor-correction
stage or directly from the rectified AC line. A PWM controller is used to receive the
feedback signal from the phototransistor inside the optocoupler and, in response to the
feedback information, output switching pulses to control the ON/OFF of the power
switches in the primary-side power stage. The entire feedback network, which consists
of R1, R2, CC, RLED, CP, a shunt regulator (commercially well-know as TL431), and an
optocoupler, delivers the output voltage VOUT information to the PWM controller while
maintaining galvanic isolation between the primary and the secondary sides.
2.2.2 Operating Principle
Fundamental operating principles of the feedback circuit are delineated as follows.
In Fig. 2.1, VOUT is divided by the voltage divider which is composed of R1 and R2. The
shunt regulator compares the divided output voltage with its built-in reference voltage,
and an error signal ILED is drawn according to their difference. The current ILED, sunk by
the shunt regulator, will flow through RLED and the LED inside the optocoupler. With
the help of the optocoupler, ILED is transferred to the primary side by a current transfer
ratio CTR. A resistor RP which will generally be integrated in the PWM controller
connects the phototransistor collector to an internal supply voltage VLO, and the induced
primary-side current IFB will be converted to a voltage form VFB. VFB will next be
modulated by the PWM modulator to produce gate-driving signals, and finally the gate
driver outputs the modulated pulses to switch the power devices in the primary-side
power stage.
Overall speaking, when VOUT drops and the divided output voltage is lower than
the built-in reference voltage in the shunt regulator, which means, in the system’s point
of view, the converted energy is insufficient for supporting the present output current
request, I and I will be decreased to raise V . A higher V results in a higher
inductor current limit and therefore makes the modulator increase the duty cycle of the
driving pulse, and eventually more energy is delivered to the output. In contrast, when
the converted energy exceeds the output request and VOUT grows up, ILED and IFB will be
increased to reduce VFB. Due to the lower current limit caused by the lower VFB, the
modulator decreases the pulse duty cycle, making less energy converted by the
converter in a switching period.
The modulator in today’s green-mode PWM controller may be somewhat more
complicated than just described. Fig. 2.2 portrays a probable control scheme
arrangement in commercial products. In order to reduce switching losses, it is very
common that when VFB drops to a green-mode threshold voltage VGR, the modulator
starts using pulse-frequency modulation (PFM) to decrease the switching frequency in
stead of keeping trying to reduce the pulse width for regulation. Besides, burst mode
[39], [40] is widely adopted to control a converter under very light/no-load conditions
(i.e., VFB is lower than VBU). In the later discussion where the standby power is analyzed,
we will describe the burst mode operation in more details.
The above principles are not limited to any converter topology, that is, this
conventional feedback circuit is applicable to many kinds of transformer-isolated
topology. Fig. 2.3 gives two examples, one of which is a flyback converter and the other
one is a forward converter. Although they differ in the configurations of power stage,
Fig. 2.2. The relationship of VFB versus VOUT and the control scheme division.
(a)
(b)
Fig. 2.3. Conventional feedback network in (a) flyback and (b) forward topologies.
the functions of their feedback circuits are exactly identical.
2.2.3 Control Loop Compensation
The conventional feedback circuit also provides frequency compensation for
stabilizing the control loop. To have deeper insights into how the compensator works,
we can simply perform the small-signal analysis on the feedback circuit. Fig. 2.4
illustrates the small-signal equivalent circuit of the feedback network from VOUT shown
in Fig. 2.1 to VFB. The shunt regulator can be modeled as a voltage-controlled current
source with a transconductance Gm, and the optocoupler is treated as a
current-controlled current source with a current gain of CTR. The internal pole of the
optocoupler is considered by including COPT. Note that the dynamic resistance of the
light emitting diode is much smaller than RLED and therefore is omitted from the
following analysis.
By observing Fig. 2.4, we can first recognize that IC is the difference of currents
through R1 and R2. That is,
Fig. 2.4. Small-signal equivalent circuit of the conventional feedback circuit.
Equating (2.2) with (2.3), substituting (2.1) into it, and rearranging that, we can obtain
V1 as a function of VOUT:
we can finally arrive at the overall transfer function by substituting (2.1), (2.4), and (2.5)
into (2.6):
C
From equation (2.7), we can find that this network exhibits a two-pole one-zero
characteristic. Since a current-mode control power stage has only one dominant pole at
low frequencies of interest, this conventional feedback network thus can be easily
utilized for the type-1 or type-2 compensations [46]. The dominant pole ωp1 is created
by the Miller effect capacitor CC, and the second pole ωp2 is formed basically by the
internal capacitor of the optocoupler and can be adjusted by varying capacitor CP.
To design a type-2 compensator, the very first step starts from drawing the Bode
plot of the well-designed power stage that is going to be compensated, as shown in Fig.
2.5. Then, we have to choose a crossover frequency fC for the final loop gain. Regarding
how to select fC, previous literature [47] has given a method to analytically derive the
crossover point depending on the specification of the maximum undershoot. Now, since
the final loop gain has to cross the 0-dB line at fC, we can design the midband gain GMID
of the compensator to cancel out the extra gain of the power stage at fC. The midband
Fig. 2.5. An example of compensator design.
gain can be derived as
LED P
MID R
CTR
G R ⋅
= . (2.12)
Note that it has nothing to do with CC and, for system designers, the only way to adjust
GMID is to vary RLED. After GMID is defined, the actual locations of fz and fp2 can be
selected based on how much phase boost is required at fC and thus CC and CP can be
calculated out [48]. In this example, the Bode plot of the final loop gain is sketched in
Fig. 2.6. As for the type-1 compensation, it can be done by making fz and fp2 coincident
to leave fp1 alone.
Although the compensator in the conventional feedback network suffices for the
Fig. 2.6. Bode plot of compensated loop gain.
realize this restriction, we observe the circuit structure drawn in Fig. 2.7(a). We can find
that since there requires a certain amount of IFB flowing through the phototransistor
collector for dropping down VFB, RLED will inherently have an upper limit to allow of a
large-enough ILED. The resulting difficulty indicated by equation (2.12) is that the type-2
compensator will suffer from a minimum midband gain limitation, which implies the
design freedom to boost or attenuate the power-stage gain curve at the selected
crossover frequency is also limited [48]. The reason why it causes this phenomenon is
that the only means for system designers to adjust the ratio of the first pole to the zero is
to vary RLED. As shown in Fig. 2.7(b), a larger RLED will result in a lower midband gain
without moving the zero. Therefore, to be more precise, the restriction in choosing RLED
(a)
(b)
Fig. 2.7. (a) Circuit that limits RLED and (b) the effect of RLED on midband gain.
actually limits how far the first pole and the zero can be separated, leading to a trouble
achieving the desired midband gain.
2.2.4 Power Loss Analysis
In Section 2.2.2, since the whole ideas about the operation of the feedback network
have been introduced, we now focus on the power loss that caused by this network.
Refer to Fig. 2.1, we observe that there exist three current branches. The first one is the
current consumed by the resistor divider (R1 and R2), while the second and the third one
are ILED and IFB, respectively. If we temporarily do not consider the loss inside the
power stage, then, based on the observation, we can formulate the power loss of the
feedback network as
Note that VCC is the supply voltage of the control IC. Since ILED can be expressed as IFB
divided by CTR, we can rewrite (2.13) as
⎟⎠
This equation is an ideal approximation, where many non-idealities are not taken into
account. For instance, if we consider voltage drops of diodes in a flyback converter as
shown in Fig. 2.3(a), (2.16) can be modified as
⎟⎠
Remember that we still do not consider the transformer loss and switching loss in (2.17)
for simplicity. The second term of (2.16) or (2.17) is essentially caused by currents
flowing through the optocoupler, and it gives us a hint that the steady-state voltage of
VFB will determine the actual power loss. In view of this, we proceed to discuss how
much does it really contribute to the standby power when the system operates under
very light/no-load conditions.
We have known from Section 2.2.2 that a typical controller will adopt the burst
mode to control a converter when its output demands very little current. Under this
circumstance, what does VFB behave like? Fig. 2.8 gives simulated waveforms of a
typical flyback converter operating in the burst mode. VG is the gate driving signal for a
switching power device. If the current request at the output is drastically reduced, VFB is
going to drop continuously due to excessive power delivery. When it falls below the
burst-mode threshold voltage VBU set in the controller, the output switching pulses will
be blocked. After the converter’s output voltage drops down and VFB recovers to exceed
VBU, the driving signals will be released again. As suggested by the name, burst mode,
this blocking-and-releasing mechanism makes VG look like a periodic burst of
consecutive pulses and causes VFB to move around the burst-mode threshold voltage
VBU. Therefore, when a system operates in the burst mode, we can estimate (2.16) or
(2.17) as
⎞
⎞⎛
⎛V −V V
V2
Fig. 2.8. Simulated waveforms of a conventional flyback converter in burst mode.
1 V as an example. Assume that CTR is 100% and forward voltages of diodes are both
0.5 V. In the burst mode, the second term in (2.21) will result in a 23-mW power loss,
leading to an obstacle to the low-standby-power target.
Through the previous discussion, we have known that a higher VFB will correspond
to a lower VOUT and the modulator should increase its output pulse width for keeping a
constant output voltage. Fig. 2.9 shows the relationship between VFB and the output
power in a conventional flyback converter. As a higher inductor peak current is
demanded by a heavier load, VFB should stay at a higher level to have a larger inductor
current limit. Therefore, IFB and hence ILED are smaller under this condition. In contrast,
when the load gets lighter, VFB drops to a lower value and both ILED and IFB become
Fig. 2.9. The relationship between steady-state VFB and the output power.
larger. This means the power loss expressed by (2.16) or (2.17) will increase while the
output power becomes smaller, and the worst case happens when there is no output load
applied. Although this amount of loss looks small in value, it evidently degrades the
light-load efficiency and, more importantly, occupies a significant portion of the total
standby power. Since most of the time, power supplies operate only in the light to
medium load range [49] or just remain plugged in but idle, this conventional feedback
topology seems to be unfavorable from an energy-saving point of view. Of course, one
can reduce steady-state ILED and IFB by raising the value of RP. However, a minimum
current ILED,min. is still required to supply the shunt regulator for proper functioning, and
this current will cause a minimum voltage drop equaling IFB,min.RP across RP, as
indicated in Fig 2.9. Thus, using a too large RP here will leave VFB a very small voltage
dynamic range and result in poor noise immunity for the feedback path. Besides, Fig.
2.10 shows the normalized frequency response of a commercial optocoupler [50] with
different R values. A larger R gives rise to a lower-frequency pole, which means the
Fig. 2.10. Frequency response of a commercial optocoupler with different RP values.
design freedom of the second pole ωp2 given in (2.11) is limited. Therefore, the
maximum value of RP is also limited by the desired crossover frequency of a converter.
2.3 Previous Solutions
We have introduced the whole background knowledge about the feedback network
in previous sections, and also the power loss disadvantage has been pointed out and
explained. In recent years, there have been some companies issuing patents to address
this problem. With an attempt to obtain and learn some experiences, prior published
techniques toward the standby power loss issue are summarized and compared with
each other in this section.
2.3.1 Primary Sensing Technique
The primary sensing [51]-[54] means the output information is not feedback
through the explicit signal path. Instead, it tries to extract the output voltage from the
information already existing on the primary side. Hence, the entire feedback network
can be removed. Not only the cost can be largely saved from this technique, but also the
losses due to current branches in the conventional feedback network disappear. Fig. 2.11
shows a simplified primary-side-control flyback converter, where there is no any direct
signal path for feedback except for the flux coupling through the flyback transformer.
To show how the voltage extraction technique takes effect, the operating
waveforms of the converter in Fig. 2.11 are illustrated in Fig. 2.12. To obtain the VOUT
information, we first recognize that the ratio of the transformer winding voltages is
proportional to that of their turn numbers. That is,
A
the ground. The flyback transformer is charged with VPW equal to VIN, making the
primary-side inductor current ILP continuously climb up. Due to the flux coupling, the
auxiliary winding (tertiary winding) then reflects a voltage of −(NA/NP)VIN. Next, in t2
when the power MOS is turned OFF, the transformer starts discharging through the
secondary winding. The secondary winding sees a voltage VSW equal to VOUT plus the
diode voltage VD1, so the secondary-side inductor current ILS declines gradually with a
slope of −V /L . In the meantime, V reflects a voltage of
Fig. 2.11. Primary-side control flyback converter.
Fig. 2.12. Operating waveforms of primary-side-control flyback converter.
(
OUT D1)
S A
AUX V V
N
V = N + . (2.23)
VPW also reflects a voltage, and VDRAIN can be expressed as
(
OUT D1)
S P IN
DRAIN V V
N V N
V = + + . (2.24)
The above two equations thus inspire us that they contain the output information in this
period of time. Although theoretically it is possible to obtain VOUT from VDRAIN [51], the
quite high voltage there which would probably cause troubles and inconvenience makes
it a worse choice. Therefore, VAUX is mostly chosen for extracting VOUT for feedback
and the controller IC generally samples the divided VAUX. In t3, the energy in the
transformer is empty. The parasitic capacitance at the drain of the power MOS and the
transformer is empty. The parasitic capacitance at the drain of the power MOS and the