Вы находитесь на странице: 1из 4

Transfer Function Modeling for the Buck

converter
Sanda Lefteriu Cécile Labarre
Ecole des Mines de Douai
Unité de Recherche en Informatique et Automatique
941, rue Charles Bourseul, 59500 Douai, France
Email: sanda.lefteriu@mines-douai.fr, cecile.labarre@mines-douai.fr

Abstract—Static converters are used with increasingly Average modeling is effective in the low-frequency band,
high switching frequencies. Consequently, they impose but it is valid only until half of the switching frequency.
more severe electromagnetic interference (EMI) constraints
in their environment. To study and reduce the conducted In continuous current mode (CCM), the Buck con-
emissions, it is crucial to understand the converters’ be- verter has two linear working modes, hence it can be
havior from low up to high frequencies. Designers typically
use the average small-signal model to derive the transfer modeled by means of the GTF [1], [2], which combines
function, which is only valid until half of the switching the continuous behavior (due to the storage elements)
frequency. This paper investigates two other techniques and the discrete behavior (due to the switches). The
to model the transfer function, namely the Generalized resulting function produces a Bode plot which is valid
Transfer Function (GTF) [1], [2] and the augmented past half the switching frequency.
Modified Nodal Analysis (MNA) technique extended to
Periodic Switched Linear (PSL) systems [3]. Simulation
DC-DC power converters can be regarded as periodic
results show that these two techniques yield the same model
as the average model until half of the switching frequency. switched linear (PSL) circuits. Thus, they can be charac-
However, they are able to predict the resonances present terized by the time-varying transfer function, which takes
at multiples of the switching frequency, which the average into account the time-varying behavior, or alternatively,
model cannot account for. Lastly, we performed the experi- by the bi-frequency transfer function [5]. For circuits
mental validation. For low frequencies, measurements were
of this kind, their steady-state response can be found
taken with an oscilloscope, while for high frequencies, we
used a spectrum analyzer. as the solution of an augmented MNA equation in the
frequency domain [3].
I. I NTRODUCTION This paper compares three transfer function modeling
techniques for computing the magnitude of the frequency
DC-DC switching power converters are commonly response for a Buck converter: the average model, the
integrated into motherboards, aircrafts, satellite commu- generalized transfer function and the augmented MNA.
nication equipment and DC motors, to name only a As a first contribution of our paper, the GTF was re-
few possible applications. They consist of linear time- derived using the ideas in [1], [2] for the special case of a
invariant components (resistors, capacitors, inductors) Buck converter. As a second contribution, the augmented
together with switches (transistors and diodes), whose MNA [3] technique was adapted to our setting. Our paper
operation is controlled to yield the desired voltage extends this technique to cope with transfer function
conversion. As switches are nonlinear and time-varying modeling in the context of a Buck DC-DC converter.
components, they render frequency domain analysis via
the Bode plot impossible for this kind of systems. This paper is structured as follows. In Sect. II, the
Average models describe the system’s dynamics based Buck converter is described in terms of the average
on average values of current and voltage variables, thus model. In Sect. III, the modeling technique using the
neglecting ripple effects due to switching. The state-of- generalized transfer function is re-derived, while the
the-art state-space averaging (SSA) technique, initially extension of the technique proposed in [3] is presented
proposed in [4], employs the weighted combination of in Sect. IV. Sect. V presents simulation results and,
state equations of the different switching phases in pulse- finally, the conclusion and directions for future research
width modulated (PWM) switching power converters. are given in Sect. VI.

978-1-5090-0349-5/16/$31.00 ©2016 IEEE


II. AVERAGE MODEL OF THE B UCK CONVERTER The output observation is y(t) = Vo (t) = VC (t), hence
the output equation y(t) = Cx̃(t) + Du(t) yields C =
The Buck converter is a hybrid dynamical system 
1 0 and D = 0. The line-to-output transfer function
with the continuous behavior dictated by the linear time-
is defined as H(s) = VVin o (s)
(s) , where s is the Laplace
invariant elements (resistor, capacitor, inductor) and the
variable, and its expression based on the average model
discrete behavior given by two switches. Fig. 1 shows the
is given by
topology of a Buck converter which supplies a passive
load resistor with voltage Vo . D
Havg (s) = L
. (4)
1 + s R + s2 LC
III. G ENERALIZED T RANSFER F UNCTION (GTF)
To derive the GTF, we follow the derivation in [1],
[2]. The solution to (1) for the state x(t) attained from
initial state x0 at time t0 using the input u(t) is given
by [6]:
Z t
Figure 1. Topology of the Buck converter x(t) = eA(t−t0 ) x0 + eA(t−τ ) Bi u(τ )dτ, t ≥ t0 . (5)
t0

The switch is controlled by a binary input signal S: Writing (5) for t ∈ [kT, kT + DT ) (phase 1), we obtain
Z t
when S = 1, the switch is closed (conducting), and
for S = 0, the switch is non-conducting (open). We x(t) = eA(t−kT ) x(kT ) + eA(t−τ ) B1 u(τ )dτ. (6)
kT
consider the converter to be operating in CCM, hence At the end of phase 1, the state is
there are two linear working modes. In the first mode,
the switch is closed (S = 1, S = 0) in the time interval x(kT + DT ) = eADT x(kT ) + (7)
Z kT +DT
t ∈ [kT, (k + D)T ], where D is the duty cycle, T is
the switching period and k = ..., −1, 0, 1, .... The second eA(kT +DT −τ ) B1 u(τ )dτ.
kT
mode is defined by the switch being open (S = 0, S = 1) Writing (5) for t ∈ [kT + DT, (k + 1)T ) (phase 2), we
for t ∈ [(k + D)T, (k + 1)T ]. Fig. 2 shows the different have
switching phases. x(t) = eA(t−kT −DT ) x(kT + DT ). (8)
The rest of the procedure follows [1], [2]. We replace
(7) into (8) and evaluate at the end of the period, for
t = (k + 1)T :
Z kT+DT
AT
x((k+1)T )=e x(kT )+e A(k+1)T
e−Aτ B1 u(τ )dτ.
Figure 2. Switching diagram kT
The Buck converter is a DC-DC converter, hence it op-
The circuit can be modeled by differential equations
erates for DC input signals. However, using small signal
involving the capacitor’s voltage VC and the inductor
analysis, we can use an input given by the superposition
current IL . Eq. (1) describes the dynamics of the Buck
of a DC together with a small-amplitude AC component:
converter in each working mode via the state-space
u(t) = u0 +ũejΩt , with u0 >> ũ and Ω, the perturbation
representation (instantaneous model):
frequency of choice. The output at steady-state will also
contain a DC component and an AC component of the
ẋ(t) = Ax(t) + Bi u(t), i = 1, 2, where (1) same frequency Ω, together with other frequencies, as we
   −1 1     
VC (t) 0 0 shall see in Sect. IV. Using the analogy with linear-time
x(t) = , A = RC
−1
C ,B =
1 1 , B2 = ,
IL (t) L 0 L 0 invariant systems, only the component with frequency Ω
is of interest, hence the generalized transfer function for
with u(t) = Vin (t), the input signal. Using the average
the line-to-output response is given by
values of the voltage VC and current IL , we write
−1 AT DT−Aτ
Z
˙
x̃(t) = Ax̃(t) + Bu(t), where (2) HGT F (jΩ) = C ejΩT I−eAT e e B1 ejΩτ dτ,
0
(9)
 
0
A is the same as in (1) and B = D . (3) where I is the identity matrix.
L

2
IV. AUGMENTED MNA FOR PSL CIRCUITS The LTI relationship can be recovered for n = 0, namely
This section shows how the approach proposed in [3] Y (ω) = Hn (ω)U (ω).
for steady-state computation of periodic switched linear
circuits via an augmented time-invariant MNA equation
in the frequency-domain can be adapted for the small-
signal transfer function computation of Buck converters.
Following [5], the concepts of impulse response and
transfer function can be extended to time-varying sys-
tems as follows. The output can be regarded as
Z ∞
y(t) = h(t, τ )u(τ )dτ, Figure 3. Frequency domain representation of the Buck converter [3]
−∞

where h(t, τ ) = R[δ(t − τ )] is the generalized impulse Following [3], the frequency domain representation of
response, with R, the operator describing the system a Buck converter is obtained by substituting the switches
behavior, and δ(t), the Dirac delta function. The variable S and S in Fig. 1 with periodic switching elements Y1
t is the observation time, while τ is the excitation time. and Y2 and by replacing the voltage source by its Norton
The bi-frequency transfer function can be computed as equivalent, yielding the augmented MNA equation
Z ∞Z ∞     
Y1 + Y2 0 I V1 Y1 Vin
H(ω, Ω) = h(t, τ )e−j(ωt−Ωτ ) dtdτ. (10)
−∞ −∞
 0 jωω C + GI −I V2= 0, (16)
Thus, the output in the frequency domain is I −I −jω ωL IL 0
Z ∞
1 where G = R1 and each block is of dimension 2N + 1.
Y (ω) = H(ω, Ω)U (Ω)dΩ, (11) The unknown IL contains the coefficients of the trun-
2π −∞
cated frequency domain representation with 2N +1 terms
where Ω and ω are the input and output frequencies.
Eq. (11) shows that a sinusoidal input with frequency N
X
Ω produces an output with a rich spectra, as opposed IL (ω) = In δ(ω − nωs − ω0 ), (17)
to LTI systems, for which the output contains solely the n=−N

frequency Ω. The time-varying transfer function with ω0 , the excitation frequency. Similar representations
Z ∞ are used for V1 and V2 , the voltage unknowns at nodes
H(t, Ω) = h(t, τ )e−jΩ(t−τ ) dτ, (12) 1 and 2 in Fig. 3. Y1 and Y2 are matrices of the form
−∞
 
is related to (10) by Yi,0 Yi,−1 . . . Yi,−2N
Z ∞  Yi,1 Yi,0 . . . Yi,−2N +1 
H(t, Ω)ej(Ω−ω)t dt. Yi =  ..  , i = 1, 2
 
H(ω, Ω) = (13) .. .. . .
−∞
 . . . . 
For periodically switched linear circuits, H(t, Ω) is a Yi,2N Yi,2N −1 . . . Yi,0
(18)
periodic function of t which can be written as a Fourier
where Y1,n are the Fourier coefficients of the window
series with respect to the switching frequency ωs = 2π
T : function with amplitude 1 when the switch S is on:
n=∞
Z DT
e−jnωs DT−1
X
H(t, Ω) = Hn (Ω)ejnωs t , (14) Y1,n =fs e−jnωs t dt= , n=−2N, . . . , 2N,
n=−∞ 0 −j2πn
with the frequency-dependent Fourier coefficients for Y1 , and Y2,n = Y1n e−jωs nDT , m = −2N, . . . , 2N ,
Hn (Ω) called aliasing transfer functions: for Y2 , namely for S, due to the delay DT . Finally, the
1 T matrix ω in (16) is given by
Z
Hn (Ω) = H(t, Ω)e−jnωs t dt.
T 0 ω = diag

ω0 − N ωs . . . ω0 + N ωs

.
The output in the frequency domain is obtained by
substituting (13) and (14) into (11): Small-signal analysis assumes an input obtained from
n=∞
the superposition of a DC and a small-amplitude AC
Y (ω) =
X
Hn (ω − nωs )U (ω − nωs ). (15) component u(t) = u0 + ũejΩt , with u0 >> ũ and Ω,
n=−∞
the perturbation frequency of choice. This corresponds

3
to a sum of 2 Dirac functions in the frequency domain: coefficients Yi,n of the window function are zero for n
U (ω) = u0 δ(ω) + ũδ(ω − Ω). The linear system in even. The amplitude of the peaks differs between the
(16) should be solved for each value of the excitation GTF, the MNA and the measurements: the experimental
frequency ω0 = 0 and ω0 = Ω, namely for input vectors data are between the two theoretical models. In between
    the peaks, the MNA predicts a response identical to the
0 0
average model, the GTF predicts higher amplitudes than
UDC =  u0  , UAC =  ũ 
the MNA, while the measurements have even a higher
0 0
amplitude than predicted by the GTF.
where the column vector 0 is of size N . The voltage
across the capacitor, given by the voltage at node 2 in
Fig. 3, represents the variable V2 in (16). We have V2 =
V2,DC + V2,AC , where
N
X
V2,DC (ω) = Vn,DC δ(ω − nωs ), (19)
n=−N
N
X
V2,AC (ω) = Vn,AC δ(ω − nωs − Ω), (20)
n=−N

with the coefficients Vn,DC and Vn,AC found by solving


(16) for ω0 = 0 and ω0 = Ω. The transfer function at
Figure 4. Simulation results and experimental validation
perturbation frequency Ω is
(
V0,AC VI. C ONCLUSION AND FUTURE WORK
ũ , if Ω 6= nωs
HM N A (jΩ) = Vn,DC +Vn,AC . (21) The line-to-output transfer function was simulated
ũ , if Ω = nωs
using the average model, as well as the GTF and the
This expression clearly shows that if the perturbation MNA for PSL circuits. This comparison of the various
frequency Ω is an integer multiple of the switching modeling techniques for obtaining the transfer function
frequency ωs , there are components due to the DC input of a Buck converter and the experimental validation
which also contribute to the response. is only a preliminary step. Foremost, the phase of the
transfer function should also be validated by using a net-
V. R ESULTS work analyzer. Moreover, the control-to-output transfer
Simulation results comparing the three modeling tech- function, critical for control purposes, is to be modeled,
niques detailed in the previous sections, namely the simulated and validated experimentally. Lastly, our ul-
average model, the Generalized Transfer Function and timate goal is to use the measurements of the transfer
the extended MNA approach for periodically switched function to derive a model for the converter, valid from
linear circuits are shown in Fig. 4. The plot also shows low up to high frequency.
the magnitude of the Bode plot measured with an Agilent
R EFERENCES
spectrum analyzer operating between 9kHz and 3GHz.
For low frequencies, measurements were performed with [1] D. Biolek, “Modeling of periodically switched networks by mixed
s-z description,” IEEE Transactions on Circuits and Systems I:
an oscilloscope. Details on the experimental setup will Fundamental Theory and Applications, vol. 44, no. 8, pp. 750–
be given in a subsequent paper. The device under test is a 758, Aug 1997.
Buck converter with parameters: L = 1mH, C = 500µF, [2] D. Biolek, V. Biolkova, and J. Dobes, “Modeling of switched
DC-DC converters by mixed s-z description,” in Proceedings of
R = 12Ω, fs = 20kHz= T1 and the duty cycle D = 12 . the IEEE International Symposium on Circuits and Systems, May
As expected, all three modeling approaches yield the 2006, pp. 831–834.
same results for perturbation frequencies inferior to half [3] R. Trinchero, “EMI analysis and modeling of switching circuits,”
Ph.D. dissertation, Politecnico di Torino, 2015.
the switching frequency (in our example, f2s = 10kHz). [4] R. Middlebrook and S. Cuk, “A general unified approach to
Moreover, the GTF and the MNA approaches predict the modelling switching-converter power stages,” in IEEE Power
peaks in the frequency response at odd multiples of the Electronics Specialists Conference, June 1976, pp. 18–34.
[5] L. A. Zadeh, “Frequency analysis of variable networks,” Proceed-
switching frequency, which the average model does not. ings of the IRE, vol. 38, no. 3, pp. 291–299, March 1950.
As the duty cycle is precisely 12 , there are no peaks at [6] A. C. Antoulas, Approximation of Large-Scale Dynamical Systems.
even multiples of the switching frequency, as the Fourier Philadelphia: SIAM, 2005.

Вам также может понравиться