Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
in [22]–[25]. In term of dynamic control for multi-phase For single phase systems, a single phase Phase-Lock-Loop
DAB converter, a control strategy in stationary αβ-frame was based on the second-order-generalized integrator [31] can be
presented in [26], [27]. In the aforementioned publications, employed. The Adaptive-Notch-Filter [32] is an alternative
only active power is regulated, none of them manages the option as it can also generates the orthogonal signal from the
reactive power caused by the circulating current. measured signal. Nonetheless, in order to ensure the accuracy
Analysis and control in the frequency domain with the of the transformation, the sampling frequency is usually at
fundamental harmonic approximation (FHA) method for DAB least twenty times greater than the fundamental frequency.
converter are presented in some recent publications [28], Therefore, when the fundamental frequency is in the range
[29]. In which, inverter voltages and the primary current of few tens kHz to few hundreds kHz, such techniques cannot
can be approximated by their fundamental harmonic with be used due to the low resolution of the sampled signal.
high accuracy. Compared to the time-domain analysis, FHA Observer-based control system for DAB converters has
method has some distinguished advantages: 1) system model been introduced recently [33]–[35]. The main advantage of
is independent from switching state; 2) active and reactive employing observer is that it does not require too high
powers can be control separately. sampling frequency. The nonlinear observer was reported in
Based on the FHA analysis, the control method introduced [35] to estimate the direct and the quadrature components of
in [28] can maximize the power factor of the converter and the transferred current. Although the accuracy was good, the
enhanced the efficiency about 1% compared to the conven- computing amount of the nonlinear method was high. In [34],
tional SPS method. The objective of the strategy addressed a full-order linear Luenberger observer was designed for the
in [29] is to minimize the root-mean-square (rms) current. same mission. The calculation was simpler but the accuracy
However, lower rms current or higher power factor may not when the converter operates far from the linearization point
ensure a better efficiency. Besides, although block diagrams was quite limited.
for close loop control in both papers are quite simple which This paper further improves the method presented in [33]
needs only the information of voltage ratio, no results about which employed a reduced-order linear observer to estimate
dynamic response were discussed. the current states. Based on the information from the observer,
a decoupled current controller is designed to individually
B. Approach control the two current components, thereby, to separately
The scope of this paper is dynamic control for a single- handle the active power and the circulating power. By adopting
phase-frequency-modulated DAB converter. The targets are: an appropriate command for the quadrature component of
• to stabilize the terminal 2 voltage, and
the transferred current, the load angle can be minimized to
• to reduce the circulating current.
reduce the conduction loss as well as the current stress on
switching devices. As a consequence, overall performance can
The first objective relates to the active power management,
be improved.
whereas the second one requires to handle the phase different
(from now on, the load angle) between the primary current and II. S TEADY STATE ANALYSIS
voltage of the transformer. The bigger load angle, the more A. Instantaneous power derivation
circulating power and vice versa. If that angle is minimized,
The DAB converter is depicted in Fig. 2. Two H-bridge
then conduction loss could be reduced and efficiency would
inverters are located at the primary and the secondary sides of a
be increased. The next challenges are: i, to determine the load
high-frequency isolated transformer with the ratio of n : 1. The
angle; ii, and to handle both the transmission power and the
input DC voltage of the primary and secondary inverters are V1
load angle simultaneously.
and V2 , respectively. The power electronic devices S1−4 and
The approach is as follows:
T1−4 used in this paper are MOSFET. Both H-bridges operate
• Firstly, the converter is analyzed using the FHA method; in 180 degrees conduction mode. The bridge shift angle ψ
• Then, considers all the quantities in a rotation frame; between the two inverters is employed to handle the power
• In the frame, instantaneous power can be calculated and flow in the system.
the load angle can be derived by taking arctangent of the For simplicity, the primary referred equivalent circuit of the
ratio between the quadrature and the direct components converter expressed in Fig. 3 is utilized in the analysis. The
of the transferred current. fundamental component of the voltages across the primary and
• Finally, models the system and designs an appropriate secondary winding of the transformer are denoted as vpe (t)
controller to regulate both variables simultaneously. ′
and vse (t), respectively; the first harmonic component of the
Nevertheless, determining the two components of such transferred current is ipe (t); Rs and Ls are the equivalent
the high frequency, non-sinusoidal current is not easy. The primary referred resistance and inductance of the transferred
control strategy reported in [30] was also for controlling the path. Since both inverters are modulated with the same switch-
converter in a dq-rotation frame. There, the instantaneous AC ing frequency, the output voltages and the transferred current
′
current was sampled then a coordinator transformation was can be seen as rotation vectors ⃗vpe , ⃗vse , and ⃗ipe with the
employed to detach two current components. This technique is same angular speed of ωs (ωs = 2πfs ). The vector diagram
very common for driving low frequency three-phase inverter. demonstrates the relationship of them is depicted in Fig. 4. The
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
′
projections of ⃗vpe , ⃗vse , and ⃗ipe on d-and q-axis of a dq-frame
rotating synchronously with and being aligned along ⃗vpe are
DC components. In that coordinator, the two voltage vectors Fig. 5: Fundamental harmonics analysis.
are:
[ ] [ ]
vpd (t) 4 v1 (t)
⃗vpe = = (1) From (5), assumes that v1 (t) is constant in one switching
vpq (t) π 0
[ ] [ ] cycle, the active power P1 is proportional to the direct current
vsd (t) 4 nv2 (t) cos ψ id (t), and the circulating power Q1 is proportional to iq (t).
′
⃗vse = = (2) The load angle seen at the primary winding is:
vsq (t) π nv2 (t) sin ψ
Q1 iq (t)
where vpd (t), vpq (t), vsd (t), and vsq (t) are the d- and q- tan ϕ = = (6)
components of the corresponding quantities; v1 (t) and v2 (t) P1 id (t)
are the terminal voltages.
Fig. 5 represents the current and voltage waveforms in one
Let id (t) and iq (t) be the projections of the current vector
⃗ipe on two axis of the dq-frame. When Rs is neglected, id (t) switching cycle at the steady state. The load angle indicated
in the figure is the phase difference between the fundamental
and iq (t) can be determined by:
component of the primary voltage and current. Notes that, it is
[ ] [ ]
not necessarily equal to the actual phase distinction between
⃗ipe = id (t) = 1 vsq (t)
(3)
iq (t) ωs Ls vpd (t) − vsd (t) the actual primary voltage and current due to the distortion and
the asymmetry of the waveform. However, as will be examined
The instantaneous active and reactive powers seen at the later in this paper, the error is insignificant that can be accepted
primary side of the transformer are defined by (4): and can be compensated by a high robust control system.
[ ] [ ] From (5), the active power flow can be regulated by con-
P1 1 vpd (t)id (t) + vpq (t)iq (t)
= (4) trolling id (t); and from and (6), the load angle can be handled
Q1 2 vpd (t)iq (t) − vpq (t)id (t) by adjusting iq (t). The next questions are:
Since d-axis of the rotating frame is intentionally aligned i, how to determine the two current components, and
to ⃗vpe , the quadrature component of ⃗vpe is zero, vpq (t) = 0. ii, what is the suitable value of iq (t).
Therefore, (4) is rewritten: As analyzed in the previous section, the answer for the first
[ ] [ ] question is to use a state observer. Since the average power
P1 2v1 (t) id (t) transferred via terminal 1 is calculated by (7), where i1 (t) is
= (5)
Q1 π iq (t) the average terminal 1 current, the direct component current
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
id (t) can be approximated by (8): where Ad and Bd are determined by the discretization using
bilinear [transformation
] [ at] the discrete time of Tz ;
method
P1 = v1 (t) × i1 (t) (7) a11 a12 B1
π Ad = , Bd = .
⇒ id (t) ≈ i1 (t) (8) a21 a22 B2
2
Among two states, x1 (k) = id (k) can be determined
Because the terminal current i1 (t) can be measured easily, by using (8). In order to identify the unmeasurable state
id (t) is calculated without difficulty. Identifying iq (t) is more x2 (k) = iq (k), a reduced-order observer is designed.
complicated since it is not explicit in a single phase converter. Denoting x̂(k) as the estimated state of the system, we have:
In the next section, a reduced-order observer is developed to {
estimate iq (t). x̂1 (k) = îd (k) = y(k)
(12)
x̂2 (k) = îq (k) = Ky(k) + ζ(k)
III. R EDUCED - ORDER O BSERVER
where K is the observer gain and ζ(k) is an added system
A. Observability state determined by:
From the primary referred equivalent circuit depicted in
Fig. 3, the large signal model of the converter can be obtained ζ(k + 1) = F z(k) + Gy(k) + Hu(k) (13)
as: where F , G and H are gains and matrix of the observer model.
d Rs 4 The current observer error is:
dt id (t) = − L id (t) − ωs iq (t) + πL v1 (t) {
s
4n
s x̃1 (k) = x1 (k) − x̂1 (k) = 0
− v2 (t) cos ψ (14)
πLs
(9) x̃2 (k) = x2 (k) − x̂2 (k)
d R 4n
iq (t) = ωs id (t) − s iq (t) − v2 (t) sin ψ The observer error in the next sampling step is:
dt Ls πLs
x̃2 (k + 1) = [a21 − Ka11 + F K − G]x1 (k)
The small signal space state model (10) of the converter
is derived by linearizing the large signal model (9) around + [a22 − Ka12 − F ]x2 (k) + [B2 − KB1 − H]u(k)
the operation point described by (1), (2) and (3). Notes that, + F x̃2 (k) (15)
in (10), small variation of terminal voltages in one sampling
cycle is neglected. Coefficients F , G and matrix H are selected for:
{
ẋ = Ax + Bu a21 − Ka11 + F K − G = 0
(10) a22 − Ka12 − F =0 (16)
y = Cx + Du
B2 − KB1 − H =0
[ ] [ ] [ ]
id (t) fx −Ωp −Ωs Therefore, the observer error equation becomes:
with x = , u = , A = ,
iq (t) ψ Ωs −Ωp
4n x̃2 (k + 1) = F x̃2 (k) (17)
−Ωs Iq V2 sin Ψ [ ]
πLs According to [36], in order to vanish the observer error in a
B = 4n , C = 1 0 ,
Ωs Id − V2 cos Ψ finite time, the eigenvalue of (17) should be placed inside the
πLs unit circle and be at least twice to six times of system pole(s).
[ ] Rs fs
D = 0 0 , Ωs = 2πFs ; Ωp = ; fx = is the Let λ is one of the system poles, the value of F to place the
Ls Fs eigenvalue of (17) at 2λ is:
normalized frequency; Fs and Ψ are the switching frequency
and the bridge shift at the nominal condition; id (t), iq (t), F = Re{2λ} (18)
and ψ are the small signals of the corresponding quantities;
Id and Iq are the current at the nominal operating point.
The observability matrix is defined as:
[ ] [ ]
C 1 0
O= = .
CA −Ωp −Ωs
Since the rank of O is 2, the system is always observable.
B. Design procedure
The time-continuous small signal model (10) can be repre-
sented in the discrete time domain as:
x(k + 1) = Ad x(k) + Bd u(k) (11) Fig. 6: Block diagram of the proposed reduced-order observer.
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
Substituting (18) into (16) and solving the obtained equation, ensure the stability of the system. Therefore, the normalized
we have: frequency fx is employed to regulate îd , while the bridge phase
a22 − F shift ψ is utilized to regulate îq .
K = a
12
G = a21 − Ka11 + F K (19) B. Frequency limitation
H = B2 − KB1 As discussed in [16] and [19], switching frequency should
The diagram of the proposed reduced-order observer is illus- be limited. When the frequency reduces, the peak flux density
trated in Fig. 6, where z is the discrete operator. Three inputs of the transformer increases. If it drops too deeply, the
are: the DC current at port 1, the normalized frequency fx and transformer be saturated. The minimum frequency depends
the bridge shift angle ψ. Matrix H and coefficients F, G and on the maximum flux density. In this paper, the ETD54/28/19
K are determined by (18) and (19). ferrite core from EPCOS with the material of N87 is used.
The nominal flux density is Bnom = 100 mT at fx = 1
IV. C ONTROL SYSTEM DESIGN (fs = 50 kHz), whereas the saturated one according to
A. Input-output pairing datasheet is Bs = 320 mT. By choosing the maximum flux
density is Bmax = 200 mT, the minimum switching frequency
After designing the reduced-order observer, the discrete
is:
system model for regulating îd (k) and îq (k) is derived as: Bnom
{ fx,min = = 0.5 (23)
x̂(k + 1) = Ad x̂(k) + Bd u(k) Bmax
(20)
y(k) = Cd x̂(k) + Dd u(k) The maximum switching frequency depends not only on
[ ] [ ] the switching capability of MOSFETs. In this study, the
1 0 I + Ad 1 0 Bd
where Cd = × , Dd = × , I is the transformer is designed at one frequency (50 kHz), when the
0 1 2 0 1 2
frequency increases, the skin and proximity effects cause the
identity matrix, and y = [îd (k) îq (k)]T .
increment of the AC resistance of the windings. Accordingly,
Model (20) has two inputs and two outputs. The decouple
the conduction loss grows and the overall efficiency might be
technique based on input-output pairing is applied to detach
decreased. Besides, the sampling frequency is fixed. Although
the model into two individual transfer networks. Let us rewrite
we sample a DC quantity (the terminal 1 current) for obser-
(20) in form of transfer function:
( ) vation, the increment of the switching frequency might affect
Y(z) = Γ (z)U(z) = Cd (zI − Ad )−1 Bd + Dd U(z) (21) the accuracy of the observer. For that reason, the maximum
switching frequency is restricted to be twice the nominal one:
The parameters of the DAB system investigated in this paper
are listed in Table I. Substituting those parameters into Γ (z) fx,max = 2 (24)
then calculating the relative gain array [36] of the system, we
have: C. Current Controllers
[ ]
( )T 1.7782 −0.7782 Fig. 7 describes the diagram of the whole observer-based
RGA = Γ (1) × Γ (1)−1 = (22) control (OBC) system. In order to weaken the interaction be-
−0.7782 1.7782
tween id (t) and iq (t), the decoupled network W is employed:
Since RGA(1, 1) = RGA(2, 2) = 1.7782, which are positive,
−1
the pairing strategy: fx (k) → îd (k) and ψ(k) → îq (k) will W = (Γ (0 )) Γ diag (0) (25)
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
D. Voltage controller
The transfer function from the terminal 2 current i2 (s) to
the terminal 2 voltage v2 (s) is (26):
(b) îq control loop.
v2 (s) RLoad
Gvi2 (s) = = (26)
i2,avg (s) 1 + s(RLoad + rc )C2
2 v1 (t)
i2,avg (t) ≈ × id (t) (27)
π v2 (t)
From (26) and (27), the transfer function from id (s) to v2 (s)
is:
v2 (s) 2V1 RLoad
Gvid (s) = = × (28)
id (s) πV2 1 + s(RLoad + rc )C2
(c) Voltage control loop.
The voltage controller is also a PI-modified II compensator.
In order to compensate for the observation error, the phase Fig. 8: Bode diagrams of open loop systems.
margin of the controller is also set to 75 degrees. The crossover
frequency is selected at 30 Hz aiming to suppress the 300 Hz
voltage oscillation reflected from the AC side to the DC bus. E. Reference calculator
Fig. 8(c) describes the Bode diagram of the open voltage Until now, the observer and the control system are designed
loop. At 300 Hz, the gain of the open loop system is -27.4 to solve the first question in the end of Section II. Here, in
that ensures a reduction of 23.5 times in the amplitude of the this part, we solve the second problem: finding the optimal
300 Hz oscillation. reference for îq .
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
V. E XPERIMENTAL R ESULTS
The 500 W laboratory scaled experiment system is depicted
in Fig. 10. Summary of system parameters is shown in Table
I. A programmable power supply is connected to terminal 1,
whereas a DC electronic load configured at constant voltage
mode is connected to the second terminal. The winding ratio
Fig. 9: Reference calculator. of the transformer is 11 : 11. The magnetizing and the leakage
inductances measured at 50 kHz are 530 µH and 10.06 µH,
respectively.
Obviously, when the load angle is zero, in theory, there All the observer as well as the whole control system are
are no reactive power, and zero-voltage-zero-current transition implemented in a TMS320F28335 control card. The terminal
condition is met. Nevertheless, as discussed in [19], due to currents are measured by Hall-effect current transducers, FA-
the effect of the dead-time, there is a limitation for the load 050P, before feeding-back to the DSP. The port voltages are
angle to maintain soft-switching and to keep the conduction sensed by two Hall-effect voltage sensors, LV25-P, then con-
loss small. Referring to [19], the condition for the load angle verted by two other ADC channels. The sampling frequency
ϕ is (29): is fixed at 50 kHz (i.e. the sampling time is Tz = 20 µs). The
{ ( ) } nominal operating point is chosen at Fx = 1.0, ψ = 25 degrees
1 1 π
ϕ ≥ Φmin = max θd , θ d + 1 − (29)
nM 2 M 2
TABLE I: System Parameters.
where θd = fs Td and Td is the dead-time interval.
Accordingly, the load angle should be minimized by reg- Parameter Symbol Value Unit
Port voltage V1 40 ∼ 60 V
ulating around the minimum dead-phase Φmin . In [19], it
Port voltage V2 50 V
is undertaken by using a unify function. In this paper, that Trans. ratio n 11:11
intention is accomplished by regulating the quadrature current Inductance Ls 10.06 µH
component iq (t) around a suitable reference. Resistance Rs 50 mΩ
From (6), the quadrature reference current when ϕ = γΦmin S1−4
MOSFETs CSD19536KCS
is represented by: T1−4
ON resistance RDS 2.3 mΩ
iq (t) = id (t) tan(γΦmin ) (30) Nom. frequency Fs 50 kHz
Nom. phase shift Ψ 25 deg
where γ is the coefficient to handle the depth of soft-switching Sampling time Tz 20 µs
Dead-time Td 500 ns
and Fs is the nominal switching frequency. In this paper, γ is
DC capacitors C1 , C 2 6400 µF
set to 1. The bigger γ, more reserve for soft-switching of the
modulation.
At 50 kHz, Φmin is 9 degrees. If γ is 1 and fx varies
within [0.5; 2], the approximation (31) can be utilized to avoid
trigonometric function calculation.
i∗q (t) ≈ γΦmin i∗d (t) (31)
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
and V1 = V2 = 50 V (i.e. M = 1.0). The eigenvalues ii, due to relatively low sampling frequency (compared to
of the observer are placed at two times of system poles. the switching frequency);
After discretized, the observer gains[ are: F = −0.9413,
] iii, and due to the distortion and asymmetry of the current
K = −0.2467, G = 0.6016 and H = −1.0882 9.9056 . waveform.
The first reason is quite obvious. It can be eliminated by
A. Open-loop verification adopting a nonlinear observer as reported in [35].
The dependence of the observation performance on the
An open-loop experiment is then conducted to verify the sampling frequency is also comprehensible. In order to avoid
accuracy of the observer. Since id (t) is approximated by (8), the requirement of high sampling speed when processing the
and iq (t) is unmeasurable for a single phase inverter, the AC transfered current, the observer technique is employed.
load angle comparison is taken instead. The estimated load However, it does not mean the sampling frequency can be very
angle is calculated by (6) based on the observed îd (t) and low. As seen in Fig. 11, the error curves are flatter when the
îq (t). The “actual” load angle is determined by measuring switching frequency is less than 50 kHz (Fx < 1), and curvier
the time difference between the zero crossing points of the when it is greater than 50 kHz (Fx > 1). When Fx > 1,
transferred current and the primary voltage of the transformer, although a DC signal is sampled, some fast dynamic might
then converting into degree scale. The test is repeated in be ignored due to the low sampling frequency. If a higher
several cases of voltage ratio and switching frequency. Under sampling frequency can be used, this issue can be fixed.
each condition, ψ is increased gradually with the step of The third reason is more sophisticated. When M = 1, the
5 degrees until reaching the rated power. At each step, the error current waveform is, in theory, symmetric and trapezoidal.
between the actual and the estimated load angle is recorded. Therefore, the primary voltage and current should be same
Fig. 11 describes the performance of the designed reduced- phase as each other. The error ranging from -4 degrees to 2
order observer. degrees (Fig. 11(c)) is mainly due to the linearization and the
The load angle error almost vanishes at the linearization sampling issues.
point (Fx = 1, ψ = 25 deg, M = 1.0). Away from that point, When M < 1 (Fig. 11(a) and Fig. 11(b)), the current wave-
its absolute value of the error tends to increase. There are three form is distorted to the left, thus the fundamental component
possible reasons: leads the actual current. In contrary, when M > 1 (Fig. 11(d)
i, because the model is obtained from linearization; and Fig. 11(e)), the current waveform is distorted to the right,
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
thus the fundamental component is lagged from the actual one. Since a simple static decoupled technique was used, the
As a consequence, the error characteristics appear to be shifted interaction between two control channels is not completely
upward or downward when the voltage ratio is less or greater eliminated and causes up-to 10% oscillation in the dynamic
than unity, respectively. of each variable. However, the fluctuation causes only up-to
In all investigated cases plotted in Fig. 11, the maximum 2 degrees swings in the transient state in the load angle. By
load angle error of about 7.6 degrees when the voltage employing a dynamic decoupled strategy, this interaction can
variation is 20%. At 50 kHz (fx = 1.0), it is equivalent to be further reduced.
only 420 ns or 2.1% of a sampling cycle. When the fluctuation 2) Voltage loops: The dynamic performance of the pro-
of terminal 2 voltage is 10%, the maximum error is less than posed observer-based control (OBC) system is examined in
6 degrees or 1.66% of a sampling period. Therefore, it can comparison with a voltage mode controller. The voltage mode
be concluded that, within the voltage variation of ±20%, the control (VMC) system uses the conventional single phase
designed observer is reasonable. The next problem is to keep shift (SPS) modulation at fs = 50 kHz (Fx = 1). The
the voltage regulation within ± 20%, or even better as ± 10%. controller is also designed as a compensator typed II with
B. Closed-loop verification the same criteria as the voltage regulator of the OBC system
(i.e. 30 Hz crossover frequency and 75 degrees phase margin).
1) Current loops: Dynamic responses of the decoupled The voltage responses of the two control systems are shown
current controller are depicted in Fig. 13. Signals sketched in in Fig. 14, Fig. 15 and Fig. 16 in three cases of voltage ratios.
the figures are obtained by using two PWM pins of the DSP In each case, the terminal 2 voltage is regulated at 50 V while
accompanied with two low-pass filters as two DAC channels. the voltage at terminal 1 varies from 45 V to 55 V (thus,
In these experiments, only current loops are examined. voltage ratio M ranges from 0.9 to 1.1). In the figures, voltage
Firstly, a step current test is conducted. References i∗d and i∗q responses obtained by the VMC method are denoted as blue
are stepped up from 0 A to 16 A and 6 A, respectively. The lines and vV M C (t), whereas, ones obtained by the proposed
current responses are reported in Fig. 12. Both the current OBC strategy are expressed by red curves and vOBC (t).
components are well stabilized. There are no overshoot. The When load resistance changes at the voltage ratio of M =
rising time of both are about 0.9 ms. As a consequence, 0.9, if the proposed OBC system is used, the voltage swing
Terminal 2 current is regulated at 9.6 A with an overshoot is about ±5% while that when employing VMC is -10.8%
of approximately 20%. (Fig. 14(a)) and 8.8% (Fig. 14(b)). When M = 1.0, the
Secondly, the experiment explores the impact of changes in responses by the OBC method are also better than that of the
îd (t) on îq (t). At initialization, îd (t) and îq (t) are regulated VMC one: the voltage sag/swell is -6% and +4% compared to
at 10 A and 4 A, respectively. As shown in Fig. 13(a), -8% and +8% (Fig. 15). The similar effect can also be seen in
both current components are regulated very well. When the Fig. 16 for the case of M = 1.1: OBC: -6%∼4% and VMC:
reference Id∗ steps up to 14 A at 2 ms, îd (t) starts increasing. -8%∼+8%.
After the rising time, îd (t) becomes stable around the desired In all cases, the stability of the system controlled by either
value. In the meantime, there is a fluctuation of ±6.25% in OBC or VMC is confirmed. Furthermore, the proposed OBC
îq (t). The variation is +10% when Id∗ falls down from 14 A system can reduced the voltage fluctuation when load changes.
down to 10 A (Fig. 13(b)). Or in other words, the voltage regulation characteristics is
The changes in îq (t) also causes a little fluctuations in îd (t). improved by the proposed OBC strategy. As the voltage
As observed from Fig. 13(c), during the rising time of îq (t) regulation is kept within ±5%, the issue of observation error
to increase from 2 A to 4 A, a swell of 0.78 A in îd (t) is mentioned above is eliminated. And thus, the observed data is
recorded. And when îq (t) steps down from 4 A to 2 A, a sag reasonable and the OBC system is reliable.
of -0.5 A occurs in îd (t) (Fig. 13(d)). Beside the improvement of the regulation characteristics,
the proposed OBC method can also help eliminate the high
frequency ringing in the voltage response. As seen in Fig. 14
to Fig. 16, thanks to the cascade dual-loops structure of the
OBC system, the order of the voltage loop is reduced and thus,
the voltage responses like a second order system. In contrary,
in the VMC method, the model is established as a transfer
function from the phase shift to the output voltage. Thereby,
high frequency ringing is not modeled and processed properly.
C. Steady state comparison
A comparative experiment is conducted to show the effec-
tiveness of the proposed OBC method in reducing the reactive
power in some cases of voltage ratio. In all cases, the terminal
2 voltage is regulated at 50 V when the voltage at the input
Fig. 12: Current step response. port is set at 45 V, 50 V and 55 V, respectively. The load
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
(a) Enloading: RLoad = 15 Ω → 7.5 Ω. (a) Enloading: RLoad = 15 Ω → 7.5 Ω. (a) Enloading: RLoad = 15 Ω → 7.5 Ω.
(b) Load sheding: RLoad = 7.5 Ω → 15 Ω. (b) Load sheding: RLoad = 7.5 Ω → 15 Ω. (b) Load sheding: RLoad = 7.5 Ω → 15 Ω.
Fig. 14: Load change response; Fig. 15: Load change response; Fig. 16: Load change response;
V1 = 45 V, V2 = 50 V (M = 0.9). V1 = 50 V, V2 = 50 V (M = 1.0). V1 = 55 V, V2 = 50 V (M = 1.1).
Notes: − vV M C (t): voltage response by the voltage mode controller
− vOBC (t): voltage response by the proposed observer-based controller
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
is then gradually increased to obtained different transmission The OBC method decreases the frequency at high power,
power. At each point, the efficiency, load angle and RMS therefore the load angle is able to be reduced. For example,
current are recorded. Notes that, the power dissipation for when M = 0.9, the load angle is reduced by 14.8 degrees
the driver, control and cooling circuits are excluded from as shown in Fig. 17(a). As a result, the RMS current can be
calculation. Furthermore, only operation in the soft-switching reduced by about 7% (Fig. 18(a)) and efficiency is improved
area is investigated. Therefore, turn-on loss can be ignored. by 5.3% (Fig. 19(a)). When M = 1.0 and M = 1.1, the load
The remain is mainly conduction loss and turn-off loss. angle reductions are 16.1 and 11.8 degrees (Fig. 17(b) and
1) Load angle and RMS current comparison: The load Fig. 17(c)), respectively and the RMS current is downsized
angle of the two mentioned methods are plotted in Fig. 17. by 8% and 7.7%, respectively, as shown in Fig. 18(b) and
When load changes, ϕ obtained by VMC method grows Fig. 18(c).
gradually as the phase shift increases. That means, when
transferring high power at a constant frequency, the converter At small power range, switching frequency of OBC method
has to suffer from high reactive power. increases to reduced the transmission power. When it reaches
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2018.2810805, IEEE
Transactions on Industry Applications
[6] D. Costinett, D. Maksimovic, and R. Zane, “Design and control for high dc converters in dc-grid applications,” IEEE Trans. Power. Electron.,
efficiency in high step-down dual active bridge converters operating at vol. 28, no. 4, pp. 1880–1889, 2013.
high switching frequency,” IEEE Trans. Power. Electron., vol. 28, no. 8, [27] ——, “Improved instantaneous current control for high-power three-
pp. 3931–3940, 2013. phase dual-active bridge dc–dc converters,” IEEE Trans. Power. Elec-
[7] G. G. Oggier, G. O. Garcia, and A. R. Oliva, “Switching control strategy tron., vol. 29, no. 8, pp. 4067–4077, 2014.
to minimize dual active bridge converter losses,” IEEE Trans. Power. [28] B. Zhao, Q. Song, W. Liu, G. Liu, and Y. Zhao, “Universal high-
Electron., vol. 24, no. 7, pp. 1826–1838, 2009. frequency-link characterization and practical fundamental-optimal strat-
[8] G. G. Oggier, R. Ledhold, G. O. Garcı́a, A. R. Oliva, J. C. Balda, and egy for dual-active-bridge dc-dc converter under pwm plus phase-shift
F. Barlow, “Extending the zvs operating range of dual active bridge high- control,” IEEE Transactions on Power Electronics, vol. 30, no. 12, pp.
power dc-dc converters,” in Power Electronics Specialists Conference, 6488–6494, 2015.
2006, pp. 1–7. [29] W. Choi, K.-M. Rho, and B.-H. Cho, “Fundamental duty modulation
[9] G. Oggier, G. O. Garcı́a, and A. R. Oliva, “Modulation strategy to of dual-active-bridge converter for wide-range operation,” IEEE Trans.
operate the dual active bridge dc–dc converter under soft switching in Power. Electron., vol. 31, no. 6, pp. 4048–4064, 2016.
the whole operating range,” IEEE Trans. Power. Electron., vol. 26, no. 4, [30] A. K. Tripathi, K. Mainali, D. C. Patel, A. Kadavelugu, S. Hazra,
pp. 1228–1236, 2011. S. Bhattacharya, and K. Hatua, “Design considerations of a 15-kv sic
[10] H. Bai and C. Mi, “Eliminate reactive power and increase system igbt-based medium-voltage high-frequency isolated dc–dc converter,”
efficiency of isolated bidirectional dual-active-bridge dc–dc converters IEEE Trans. Ind. Appl., vol. 51, no. 4, pp. 3284–3294, 2015.
using novel dual-phase-shift control,” IEEE Trans. Power. Electron., [31] M. Ciobotaru, R. Teodorescu, F. Blaabjerg et al., “A new single-phase
vol. 23, no. 6, pp. 2905–2914, 2008. pll structure based on second order generalized integrator,” in Power
[11] B. Zhao, Q. Song, W. Liu, and W. Sun, “Current-stress-optimized Electronics Specialists Conference, 2006, pp. 1–6.
switching strategy of isolated bidirectional dc–dc converter with dual- [32] D. D. Nguyen, T. Nguyen-Duc, G. Fujita, and T. Funabashi, “Adaptive
phase-shift control,” IEEE Trans. Ind. Electron., vol. 60, no. 10, pp. notch filter solution under unbalanced and/or distorted point of common
4458–4467, 2013. coupling voltage for three-phase four-wire shunt active power filter
[12] B. Zhao, Q. Song, and W. Liu, “Efficiency characterization and optimiza- with sinusoidal utility current strategy,” IET Generation, Transmission
tion of isolated bidirectional dc–dc converter based on dual-phase-shift & Distribution, vol. 9, no. 13, pp. 1580–1596, 2015.
control for dc distribution application,” IEEE Trans. Power. Electron., [33] D. D. Nguyen, G. Fujita, Q. Bui-Dang, and M. C. Ta, “A reduced-
vol. 28, no. 4, pp. 1711–1727, 2013. order observer-based control system designed in the frequency domain
[13] F. Krismer and J. W. Kolar, “Efficiency-optimized high-current dual for dual-active-bridge converter,” in Sustainable Energy Technologies
active bridge converter for automotive applications,” IEEE Trans. Ind. (ICSET), 2016 IEEE International Conference on. IEEE, 2016, pp.
Electron., vol. 59, no. 7, pp. 2745–2760, 2012. 1–6.
[14] L. Corradini, D. Seltzer, D. Bloomquist, R. Zane, D. Maksimović, and [34] D. D. Nguyen and G. Fujita, “Observer-based decoupling power con-
B. Jacobson, “Minimum current operation of bidirectional dual-bridge trol for frequency modulated dual-active-bridge converter,” in Power
series resonant dc/dc converters,” IEEE Trans. Power. Electron., vol. 27, Electronics and Motion Control Conference (IPEMC-ECCE Asia), 2016
no. 7, pp. 3266–3276, 2012. IEEE 8th International. IEEE, 2016, pp. 754–760.
[15] K. Wu, C. W. de Silva, and W. G. Dunford, “Stability analysis of isolated [35] D. D. Nguyen, M. L. Nguyen, T. Nguyen-Duc, and G. Fujita, “Observer-
bidirectional dual active full-bridge dc–dc converter with triple phase- based nonlinear control for frequency modulated dual-active-bridge
shift control,” IEEE Trans. Power. Electron., vol. 27, no. 4, pp. 2007– converter,” in Energy Conversion Congress and Exposition (ECCE),
2017, 2012. 2016 IEEE. IEEE, 2016, pp. 1–8.
[16] J. Hiltunen, V. Väisänen, R. Juntunen, and P. Silventoinen, “Variable- [36] G. F. Franklin, J. D. Powell, and M. L. Workman, Digital control of
frequency phase shift modulation of a dual active bridge converter,” dynamic systems. Addison-wesley Menlo Park, 1998, vol. 3.
IEEE Trans. Power. Electron., vol. 30, no. 12, pp. 7138–7148, 2015.
[17] X.-F. He, Z. Zhang, Y.-Y. Cai, and Y.-F. Liu, “A variable switching
frequency hybrid control for zvs dual active bridge converters to achieve
high efficiency in wide load range,” in Applied Power Electronics
Conference and Exposition (APEC), 2014 Twenty-Ninth Annual IEEE.
IEEE, 2014, pp. 1095–1099.
[18] G. G. Oggier and M. Ordonez, “High-efficiency dab converter using
switching sequences and burst mode,” IEEE Trans. Power. Electron.,
vol. 31, no. 3, pp. 2069–2082, 2016.
[19] D. D. Nguyen, D. T. Nguyen, and G. Fujita, “New modulation strategy
combining phase shift and frequency variation for dual-active-bridge
converter,” IEEJ Journal of Industry Applications, vol. 6, no. 2, pp.
140–150, 2017.
[20] D. Segaran, B. McGrath, and D. Holmes, “Adaptive dynamic control of
a bi-directional dc-dc converter,” in Energy Conversion Congress and
Exposition (ECCE), 2010 IEEE. IEEE, 2010, pp. 1442–1449.
[21] D. Segaran, D. G. Holmes, and B. P. McGrath, “Enhanced load step
response for a bidirectional dc–dc converter,” IEEE Trans. Power.
Electron., vol. 28, no. 1, pp. 371–379, 2013.
[22] F. Krismer and J. W. Kolar, “Accurate small-signal model for the digital
control of an automotive bidirectional dual active bridge,” IEEE Trans.
Power. Electron., vol. 24, no. 12, pp. 2756–2768, 2009.
[23] S. Dutta, S. Hazra, and S. Bhattacharya, “A digital predictive current-
mode controller for a single-phase high-frequency transformer-isolated
dual-active bridge dc-to-dc converter,” IEEE Trans. Ind. Electron.,
vol. 63, no. 9, pp. 5943–5952, 2016.
[24] W. Song, N. Hou, and M. Wu, “Virtual direct power control scheme of
dual active bridge dc-dc converters for fast dynamic response,” IEEE
Transactions on Power Electronics, 2017.
[25] F. Xiong, J. Wu, Z. Liu, and L. Hao, “Current sensorless control for dual
active bridge dc-dc converter with estimated load-current feedforward,”
IEEE Transactions on Power Electronics, 2017.
[26] S. P. Engel, N. Soltau, H. Stagge, and R. W. De Doncker, “Dynamic
and balanced control of three-phase high-power dual-active bridge dc–
0093-9994 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.