Documente Academic
Documente Profesional
Documente Cultură
David C. Hamill
Abstract - A general method is presented for modeling multi- voltage to constant current, or be nonlinear, so a linear re-
port dc-dc converters. It copes with multiple inputs, multiple sistive load is a special case. With other loads, the small-
outputs and bidirectional ports, and is, based on an averaging signal damping will differ from that predicted using an
formulation. A matrix description is adopted, so the technique equivalent load resistance. Worse, the load could be induc-
can be extended to converters with any number of ports. As an
tive or capacitive, and this will greatly affect the overall
example, a three-port forward-flyback converter is analyzed
using symbolic computation software (Maple V). system dynamics. For these cases, models developed using a
stiff voltage source and a resistive load will give misleading
results. Instead, the aim should be to model the converter in
I. INTRODUCTION isolation from its surrounding circuit. Provided its terminal
characteristics are properly represented, such a model can
ompared with the attention paiid to two-port conver-
C
subsequently be embedded within a complete power system,
ters, little consideration has bjeen given to modeling allowing the interactions to be assessed with ease. This is the
multi-output dc-dc converters. Important characteris- basis of two-port circuit theory, adapted here to multiple
tics such as static and dynamic cross; regulation have been ports.
little explored in the literature, and then usually only on an First, a generalized model of an open loop dc-dc converter
ad-hoc basis - e.g. [l], [2]. Furthermore, very little work is developed, assuming an averaged description. Next, the
has been done on generalized multi-port dc-dc converters. characteristics are linearized around a quiescent operating
For example, a personal computer power supply could com- point. Finally, the open loop model is embedded within a
prise a single converter with multiple dc outputs as usual, feedback control loop. The method is subject to the usual
but two dc inputs: one fed from rectified ac mains, the other limitation of linear models: it may be inaccurate for large
from a secondary battery. The battery port could be bidirec- signals. Nevertheless, linearized average models have proved
tional, to allow recharging. A generalized dc-dc converter popular with engineers because they allow the application of
can have multiple input ports, multiple output ports and standard linear systems control theory.
bidirectional ports.
This paper describes a general method for small-signal II. THE OPEN LOOP CONVERTER
modeling of any multi-port dc-dc converter. It is based on
A dc-dc converter has two or more power ports. If power
matrices, for several reasons: they provide a compact nota-
always flows into the converter it is an input port; if power
tion; the results can be applied to converters with any num-
always flows out, it is an output port; if power can flow in
ber of ports; numerical matrix computations are easily
either direction, it is bidirectional. However, for generality,
programmed using standard linear algebra packages, work-
the standard circuit-theory sign convention is adopted here:
sheets such as MathCad, and even spreadsheets; and matrix
the reference direction of current is into the positive terminal
algebra can be automated with symbolic computation pack-
of each port, as shown in Fig. 1. No distinction is made
ages such as Maple, Mathematica and Macsyma.
between input and output ports; if the power at a port is
Most analyses of two-port dc-dc converters start by assum- positive, it is acting as an input; if negative, as an output.
ing a stiff voltage source at the input amd a resistive load at
the output. For good reasons the proposed approach does not. A. The State Vector
At its input, a converter is often fed from a filter, or at least The model presented is based on four essential vectors, the
via some line impedance decoupled by a capacitor. At the first of which is the state vector x(t).A general description of
output, the load's dc characteristic many vary from constant a dynamical system is:
Q 1997 IEEE
0-7803-3840-5/97/$10.00 42 1
+
where distinguishes a particular system. The dynamics of
an mth-order converter can be characterized by m internal
dc-dc
state variables, usually the inductor currents and capacitor
converter
voltages. These state variables can be formed into an instan-
taneous state vector, x,,Jt) E R. This gives an exact de-
scription of the time varying, nonlinear circuit including, for
example, ripple at the switching frequency.
L- N-i
422
D. Large-Signal Model loop stability, the converter will settle to a steady state where
The open loop converter can be characterized by two non- x(t) = const = X and y(t) = const = Y. Substituting for U, w,
linear vector equations linking the essential vectors U, w, x x and y in (2) and (3), and setting dxldt = 0 (since X =
and y: const), X may be found in terms of U, W and Y. This gives
the steady state operating point, Q = {U,W, X, Y >.
Now consider small perturbations around (2. Adapting the
usual notation for dc-dc converters to the vector case, let
x(t) = X + g(t) , etc., where k(t) is a small perturbation from
Y(0 = Ur[X(t)> w,W ( 4 l (3)
the steady state equilibrium. Provided they are smooth, the
The state equation. (2), governs the (converterslarge-signal nonlinear functions 4 and y may each be expanded by a
dynamics as it reacts to the control signals and independent multivariable Taylor series; truncating after the linear terms,
port variables. Equation (3), the response equation (output (2) and (3) become
equation in control terms), describes how the dependent
port variables respond. Functions 4 and y are attributes of a d
- act) = A i ( t ) + B G(t) + E G(t) (4)
particular converter. dt
TABLE
I:NOMENCLARJKE MODEL
FOR THE GENERALIZED
SCALARS
423
m
Fig. 2: Block schematic of the open loop small-signal Fig. 3: The model of Fig. 2 may be reduced to two
model of a dc-dc converter. frequency dependent blocks and an adder.
(I is the N x N identity matr)ix. In (sa) the internal state 2N port variables, there are (2N)!/(2N!)possible combina-
vector ?(s) has been eliminated and the constant matrices A tions, excluding row swaps. This number determines the
to F have been replaced by frequency dependent matrices. different forms of matrix G. (E.g. there are six varieties of
Equation (sa) is shown in block schematic form in Fig. 3, a hvo-port parameters, but 60 varieties of three-port parame-
complete black box small-signal model of a generalized ters.) In the analysis of large multi-port networks, the usual
open loop N-port dc-dc converter. The entries of the N x M choice is to group all the currents into one vector and all the
matrix H(s) are transfer hnctions from the A4 control signals voltages into the other. Depending on which vector is chosen
to the N dependent port variables, and the entries of the N x as independent, the resulting matrix contains either ini-
N matrix J(s) are transfer functions from the N independent pedances (2-parameters) or admittances (y-parameters).
port variables to the N dependent port variables. However, these are not very useful for the present purposes,
because the y and z-parameters for an ideal converter are all
111. THE CLOSED LOOP CONVERTER infinite.
The two-port g-parameters have been suggested by Mand-
An objective of a dc-dc converter is to regulate against hana [ 111 as a small-signal model for dc-dc converters;
disturbances: to maintain certain components of y constant
despite variations in w. Ideally, i ( s ) = 0 . This goal is ap-
proached by applying closed loop control. In the feedback
system of Fig. 4, i ( s ) is compared to zero to produce an error
vector, which is processed by controller K(s) to produce the where g , , is the converters input admittance, g,, its reverse
control vector ii(s),thus closing the loop. current gain, g,,its forward voltage gain, and g2, its output
impedance. This description can be extended to the Ordinary
G(s) = -K(s) $(s) (10) Case of a multi-port dc-dc converter if the port variables are
In general K(s) allows cross coupling, so that any component segregated into dependent and independent vectors as sug-
of y can influence any or all components of U. From (9) and gested above. with port 1 taken as the input. Then (1la) may
(10) the closed loop small-signal model is obtained: be written as
1p(T) 1 1
VNW
= A&) I
I
Zo&) \I;*?I
iN(S)
(13)
A condition for validity is that I + H(s)K(s) must be nonsin- Here $(SI and +(s) have been partitioned into voltage and
gular, which is usually true. A block schematic is shown in current parts, and the matrix G(s) partitioned accordingly.
Fig. 5 . Note that if K(s) = 0, ( l l a ) reduces to G(s)= J(s). Scalar YJs) = g , , = i ,/cl is the converters input admittance,
A. The Ordinary Case Row vector Ai(s) = [g,, ... g,,J comprises reverse current
A , . * A
424
Open loop converter
Fig. 5 : The model of Fig. 4 can be reduced to a single
Fig. 4: Block schematic of the closed loop small-signal frequency dependent block, forming a small-signal black
model of a dc-dc converter. box model of a multi-port dc-dc converter.
how input voltage changes affect the output voltages, i.e. the discontinuous mode, so both the duty ratio and the switching
dynamic line regulation of the converter frequency affect its output voltage. For this example n = 5
(audiosusceptibility). Ideally all its entries would be zero. (state variables), N = 3 (ports) and A4 = 2 (control signals).
Matrix Z,,(s) = [g,], i, j = 2.. .N, comlprises self and mutual Let the four essential vectors be:
impedances c i i j . The leading-diagonal elements (i = j ) are
the source impedances of each output jport, and describe the State vector x = P L I iL2 vc1 vc2 Val
dynamic load regulation of the converter. The off-diagonal Independent port vector w = [vI io, i,IT
terms (i j ) are mutual impedances relating the port-i cur- Dependent port vector y = [i, vo, vmIT
rent to the port-j voltage, and describe the dynamic cross Control vector U = [6 fl
regulation. Ideally all the entries of Z,,i(s) would be zero.
For cases other than the Ordinary Case, different ways of where i,, is the current in L , , etc., 6 is the duty ratio andf, is
choosing the independent and dependent variables might be the switching frequency. Other quantities are defined in Fig.
more appropriate, and they should be considered on their 6. The circuit has the following parameter values: L , =50pH,
merits. L, = 600pH, Lp = 60pH, C, = 47@, C, = 470pF, C3 = 470pF,
Fs = 49kHz, A =z 0.3, N,,IN, = 20114, y/ = 28V, I,, = -1A, I ,
Thus the N x N transfer function matrix G(s)gives a com-
= -lA. The output voltages are intended to be Yo, = 12V, V,
plete small-signal description of the dy.namics of any N-port
= 12v.
dc-dc converter. For an open loop convierter (with its control
signals held constant), G(s) is identical to J(s). For a closed The low frequency state equation may be found by a
loop converter under the control scheme shown, G(s) is naive averaging process as:
given by (lla). Other forms of control (e.g. innerlouter -
loops) will result in a different form for (1lb).
N. EXAMPLE
As an example, an Ordinary Case converter is analyzed
using the symbolic computation packagie Maple V [ 121. The
method can easily be applied to more complicated conver-
ters, the computer handling the increased complexity of the which corresponds to the state equation (2). Likewise,
algebra.
A. Open Loop Model
The three-port converter shown in Fig. 6 [I31 can be sepa-
rated into two semi-converters: it is basically a forward
converter but, instead of the normal energy-recovery reset corresponds to the response equation (3).
winding, the transformer has a flyback winding. The forward
semi-converter operates in continuous mode, so its output
voltage depends on the duty ratio but is independent of the
switching frequency. The flyback semi-converter operates in
425
For space reasons only the 5 x 5 A matrix is shown here B
is 5 x 2, C is 3 x 5, D is the 3 x 2 zero matrix, E is 5 x 3,
o1 and F is the 3 x 3 zero matrix
- The H(s) and J(s) matrices are found by computing (10)
and (11) in Maple
+ Hrcrat.=evalm(Cmat&*imse(s* &* () -Amat) &*Bmat+Dmat):
Jmat:=evalm(Cmat&*imse(s * &* (1 -Amat)&*Emat+lFmat) ;
02
-
H(s) is 3 x 2, J(s) is 5 x 5
The full expressions for H(s) and J(s) are too complex to
reproduce here but, substituting numerical parameter values
and setting s = 0, their dc values are found as
I
4.286 -0.875 x
Fig. 6 : Two-output forward-flyback converter used in the
example. H(0) = 40 0
80 -2.45 x
0.0153 -0.4286 0
426
d I As in [13], the duty ratio was employed to regulate the
forward semi-converter and the switching frequency was
used for the flyback semi-converter. The two control loops
interact, see Fig. 7. For better results the loops should be
decoupled; this will be discussed in a planned future paper.
The full Maple listing of this example is available from
the author on request.
V. CONCLUSION
A technique has been presented for modeling a general-
ized N-port converter, in isolation from its sources and loads.
The starting point is an averaged large-signal state equation,
obtained by any applicable method. The outcome is a full
dosed loop open loop model. This can be embedded within a control
loop to give a complete small-signal dynamical model, G(s),
for any multi-port dc-dc converter. The matrix formulation is
particularly suited to automatic computation, either numer-
ical or symbolic.
REFERENCES
A 2000 4000 6000 -00 K. Harada, T. Nabeshima and K. Hisanaga, State-space analysis of the
cross-regulation,Power Electronics Specialists Con$, San Diego, C A
June 1979, pp. 186-192
Fig. 7: Dynamic line regulation (audiosusceptibility) of the
example converter, in open loop and closed loop. Y.T. Chen, D.Y. Chen, Y.P. Wu and F.Y. Shih, Small-signal modeling
of multiple-output forward converters with current-mode control, IEEE
Trans. on PowerElectronics, vol. 11, no. 1, pp. 122-131, Jan. 1996
r 1
G.W. Wester and R.D. Middlebrook, Low-frequency characterization
of switched dc-dc converters, IEEE Trans. Aero. and Elec. Systems,
vol. 9, no. 3, pp. 376-385, May 1973
R.D. Middlebrook and S. Cuk, A general unified approach to model-
The 3 x 3 G(s)matrix can now be calciulated from (14): ling switching-converterpower stages, Power Electronics Specialists
Qnat := evalm(inverse( &* () + Hmat &* Kmat) &* Jmat) ;
Con$, Cleveland OH, June 1976, pp. 18-34
A.S. Kislowski, R. Red1 and N.O. Sokal, Dynamic Analysis of
The resulting (extremely large) expression may be simplified Switching-Mode DC/DC Converters, New York: Van Nostrand Rein-
hold, 1991
by substituting numerical values for the parameters. Inverse
V. Vorperian, Simplified analysis of PWM converters using model of
Laplace transformation can be used to find the response to PWM switch, 2 parts, IEEE Trans. Aero. and Elec. Systems, vol. 26,
steps and other functions, or the matrix may be evaluated in no. 3, pp. 490-505, May 1990
the frequency domain by setting s =.io. Fig. 7 shows the P.T. Krein, J. Bentsman, R.M. Bass and B.L. Lesieutre, On the use of
dynamic line regulation of the two semi-converters, in open averaging for the analysis of power electronic systems, IEEE Trans. on
PowerElectronics, vol. 5, no. 2, pp. 182-190, Apr. 1990
loop and closed loop with K , = 5 , K2 =: -50. (No evaluation
B. k h a n and R.M. Bass, Switching frequency dependent averaged
of stability was made, but in practice this must be done.) models for PWM DC-DC converters, IEEE Trans. on PowerElectron-
To see the effect of infinite dc loop gain (e.g. by including ics, vol. 11, no. 1, pp. 89-98, Jan. 1996
ideal integrators in the controller), s was set to zero and the J. Sun and H. Grotstollen, A symbolic computation package for aver-
aged analysis of power electronic systems, Applied Power Electronics
limit taken as K,-+ CO, K,+ -W. The mw value of G is
1
Con$, San Jose, CA, Mar. 1996, vol. 1, pp. 96-102
J. Van de Vegte, Feedback control systems, 2nd edition, Englewood
-0.0306 -0.4286 -0.4286 Cliffs, NJ: Prentice-Hall, 1990
G(0) = 0 0 0 (21) 0.P Mandhana and R.G. HoA, Two-port characterization of dc to dc
[ o 0 0 resonant converters, Applied Power Electronics Conf, Los Angeles,
CA, Mar. 1990,737-745
which may be compared with the equivalent open loop ma- B.W. Char et al., First Leaves; A Tutorial Introduction to Maple V,
trix, J(0). The iilfnite loop gain has made the input conduc- New York: Springer-Verlag, 1992
tance negative, g,, = -30.6mS. The reverse current gains are J. Sebastian, J. Uceda, M. Rico, M.A. Perez and F. Aldana, A complete
study of the double forward-flyback converter, Power Electronics
now both -0.4286. The line, load and cross regulation are Specialists Con$, Kyoto, June 1988, vol. 1, pp. 142-149
perfect, as shown by the zeroes in rows 2 and 3.
427