A Small-Signal Analysis of the Current-Mode

Borderline Conduction Mode PWM switch model

Christophe BASSO

On page 196 and fig. 2-104 of Ref. [1], appears a large-signal model of the PWM switch operated in the
so-called quasi-square wave resonant mode. In order to analytically obtain the ac response of a current-mode
flyback converter operated in the quasi-square wave resonant mode (also known as valley- switching operation),
a new small-signal model has to be derived. First, let us try to simplify the large signal model given in Ref. [1].
In this model, the average current flowing in terminal “c” is written to stick to the PWM switch current-mode
original notations: a main current source delivering a peak current equal to the control voltage Vc divided by the
sense resistor Ri to which a source Iµ subtracts some current. However, in a converter operated in Borderline
Conduction Mode (BCM) where the dead time is negligible, the average current flowing through terminal “c” is
simply the peak current value divided by 2. Capitalizing on this fact, the Iµ source can disappear and an updated
version of the large-signal model appears on figure 1:

Ia Ic
a c
d1.Ic Vc/(2*Ri)

Figure 1: the PWM switch model in Borderline Conduction Mode once simplified.

In this model, we have the following source definitions:

Vc L
d1 = (1)
Ri VacTsw

Vc L ⎛ 1 1 ⎞
Tsw = ⎜⎜ + ⎟ (2)
Ri ⎝ Vac Vcp ⎟⎠

Vc
Ic = (3)
2 Ri

I a = d1 I c (4)

If we use Eqs. (1) and (2) to develop Eq. (4), we have:

Vc L Ic
Ia = Ic = (5)
Ri VacTsw ⎛ 1 1 ⎞
Vac ⎜ +
⎜ Vac Vcp ⎟⎟
⎝ ⎠
The classical small-signal analysis requires the introduction of a perturbation on all the variables of the system.
That is to say, we should transform equations Eqs. (3) and (5) by perturbing the following variables:

I c = I c 0 + iˆc (6a)

I a = I a 0 + iˆa (6b)
Vcp = Vcp 0 + vˆcp (6c)
Vac = Vac 0 + vˆac (6d)
Vc = Vc 0 + vˆc (6e)

The 0-indiced terms represent the dc point of the variable whereas the ^ denotes the small ac variations around
that dc point. Unfortunately, when you update the concerned variables into Eqs. (3) and (5), the exercise which
consists of sorting out and gathering all the dc and ac terms becomes a rather tedious operation. One possible
solution is to build an ac-only model, without any bias point capabilities. In this method, the ac terms are
considered and the dc solutions are purposely left away. After all, we can always calculate the dc conditions
either by using equations-based results or simply by running a bias point simulation with Fig. 1 large signal
model. To obtain ac terms only, we can calculate the sensitivity of each current to the variables of concern listed
in equations 6a to 6e. Let us apply this option to Eq. (3):

∂I
iˆc = c vˆc (7)
∂Vc

⎛ 1 ⎞
iˆc = vˆc ⎜ ⎟ = vˆc kc (8)
⎝ 2 Ri ⎠

Where kc is simply:

1
kc = (9)
2 Ri

Let us apply a similar technique to equation (5) now:

∂I ∂I ∂I
iˆa = a vˆcp + a iˆc + a vˆac (10)
∂Vcp ∂I c ∂Vac

Once we derived Eq. (10), we obtain the following terms:

I c 0Vac 0 Vcp 0 Vcp 0 I c 0

iˆa = vˆcp + iˆc − vˆac (11)
(V + Vcp 0 ) Vcp 0 + Vac 0 (V + Vcp 0 )
2 2
ac 0 ac 0

Assuming:

I c 0Vac 0
kcp = (12a)
(V ac 0 + Vcp 0 )
2

Vcp 0
kic = (12b)
Vcp 0 + Vac 0
 Vcp 0 I c 0
kac = (12c)
(V + Vcp 0 )
2
ac 0

we can re-write equation (11) the following way:

iˆa = vˆcp kcp + iˆc kic − vˆac kac (13)

Re-arranging the original large-signal model with these new source definitions, the updated small-signal model
appears in figure 2:

ic
a c
kc=1/(2*Ri)
kcp=Vac*Ic/(Vcp+Vac)^2 v(c,p).kcp ic.kic v(a,c).kac
vc.kc
kic=Vcp/(Vcp+Vac)
kac=Vcp*Ic/(Vcp+Vac)^2

Figure 2: the updated small-signal model uses four current sources.

A BCM flyback converter

Our main application is a flyback converter operated in Borderline Conduction Mode. The
implementation of the BCM small-signal model in this configuration appears in figure 3:
v(c,p).kcp
a
ic.kic

X5
XFMR
RATIO = N
p

Vout
v(a,c).kac

Vin

Resr
ic

vc.kc

Rload
c

Cout

Lp

Figure 3: the small-signal model with a flyback converter.

In this application, observing the various terminal connections, we can update the coefficient definitions:
Vac 0 = Vin (14a)
V
Vcp 0 = out (14b)
N
Vc 0
Ic0 = (14c)
2 Ri

We can thus evaluate the coefficient parameters:

VinVc 0 VinVc 0 N 2
kcp = = (15a)
2 Ri (Vout + NVin )
2 2
⎛ Vout ⎞
2 Ri ⎜ Vin +
⎝ N ⎟⎠
Vout
N Vout
kic = = (15b)
Vout
+ Vin V out + NV in
N
Vout Vc 0
N 2 Ri VoutVc 0 N
kac = = (15c)
2 Ri (Vout + NVin )
2 2
⎛ Vout ⎞
⎜ Vin + N ⎟
⎝ ⎠

vˆout , s =0
Using this configuration, let us find the dc small-signal gain G0 vˆin = 0 = . First, since we are in
vˆc , s =0
dc and Vin is considered constant during the study, terminal “a” goes to ground. Then, all capacitors are opened
and the inductor is shorted: terminal “c” goes to ground as well. Since v(a,c) is null, the associated source
disappears. Then, the load is reflected on the primary side using the transformer turns ratio squared. As a second
observation, we can see that the current source involving kcp, appears between the terminals “c” and “p”.
Therefore, a resistance of value 1/ kcp can be placed across the terminal “p” and ground. Finally, capitalizing on
all these changes leads to the below equivalent schematic:

1 2 Vout
X4
XFMR
RATIO = N

B8 R8 RloadN
Current {1/kcp} {Rload/N^2}
V(Vc)*{kc}*({kic}-1)

Figure 4: the dc small-signal gain is found by shorting the inductor and opening the capacitor. As Vin is constant,
its small-signal value is 0.
The voltage appearing across R8 is equal to the current source value B8, delivering current to the parallel
combination of R8 and the reflected load. Let’s call Req this compound resistor:

Rload 1 Rload
Req = || = (16)
N 2
kcp Rload kcp + N 2

The small-signal output voltage equation is therefore:

vˆout , s =0 = vˆc , s =0 kc (1 − kic ) Req N (17)

In the above equation, the minus sign passed to the transformer turns ratio – our flyback delivers a positive
voltage, whereas a buck-boost gives a negative output – has been included in the expression. From Eq. (17), we
can derive the dc gain of the BCM flyback:

G0 = kc (1 − kic ) Req N (18)

Application example

A simple open-loop flyback has been wired using the PWM switch model described in Ref. [1]. This a
converter delivering 19.2 V to a 10 Ω load from a 100-Vdc input source. The control voltage is set to 1.7 V and
imposes a 1.7 A peak current.

parameters

Vin=100
B2 V2
Rload=10 Voltage 0
N=-0.25 V(Vc)+{Vc} AC = 1
100V 1.70V 0V
ESR=1 7 9
Cout=100u Vc
PWM switch BCM
vc
a

Lp=1m 17.0V ton

ton

Vc=1.7
Ri=1 X2
Fsw (kHz)

25.6V Fsw XFMR

Fsw =25.6k RATIO = N
-76.9V 19.2V
Vin 1 3 Vout
p

Ic=Vc/(2*Ri) {Vin}
c

ESR
Vac=100 0V 5 X1 {ESR}
PWMBCMCM Rload
Vcp=76.9 L = Lp 19.2V {Rload}
Ri = Ri 2

kc=1/(2*Ri) Ic Lp
850mV Cout
kcp=Vac*Ic/(Vcp+Vac)^2 {Lp}
B3 {Cout}
kic=Vcp/(Vcp+Vac) Voltage
kac=Vcp*Ic/(Vcp+Vac)^2 I(Lp)

Figure 5: a large-signal simulation of a BCM flyback converter running open-loop.

From the above dc points, we can calculate our source coefficients:

VinVc 0 N 2 100 ×1.7 × 0.252

kcp = = = 2.72m (19a)
2 Ri (Vout + NVin ) 2 × 1× (19.2 + 0.25 ×100 )
2 2

Vout 19.2
kic = = = 434m (19b)
Vout + NVin 19.2 + 0.25 × 100
 VoutVc 0 N 19.2 × 1.7 × 0.25
kac = = = 2.09m (19c)
2 Ri (Vout + NVin ) 2 ×1× (19.2 + 0.25 × 100 )
2 2

1 1
kc = = = 0.5 (19d)
2 Ri 2 ×1

The equivalent resistor is found to be:

Rload 10
Req = = = 111.49 Ω (20)
Rload kcp + N 2
10 × 2.72m + 0.252

The dc gain is derived using Eq. (18):

G0 = 20 log10 ( kc (1 − kic ) Req N ) = 20 log10 ( 0.5 × (1 − 0.434 ) × 111.49 × 0.25 ) = 17.93 dB

(21)

If we run Fig. 5 simulation template, we obtain the below gain curve:

0 1

18.0

y = 17.9 db(volts)

44
14.0
vdbout in db(volts)
Plot1

10.0

6.00

2.00

1 10 100 1k 10k 100k

frequency in hertz

Figure 6: the dc gain given by the linearized large-signal model sticks to the result given by Eq. (21).

Ac analysis

The control to output ac analysis implies that inductors and capacitors are back in place as in Fig. 3. The
new schematic appears in Fig. 7.
 p2
1 Vout
X5
B6 XFMR
Current RATIO = N
V(Vc)*{kc}
ESR
{ESR/N^2}
c2
Rload
2
B5 {Rload/N^2}
Current
V(c2,p2)*{kcp}+V(Vc)*{kc}*{kic}+V(c2)*{kac}
Lp Cout
{Lp} {Cout*N^2}

Figure 7: the elements are back in place for the ac analysis.

The schematic looks a bit complicated at the first glance, but a careful writing of the equations will lead us to the
correct result rather quickly. In Fig. 7, we can see a current source B5 delivering current to complex impedance
made of the reflected capacitor, its ESR and the load resistor. However, B5 current is diverted into the inductor
by B6. The equation of the output voltage is thus:

⎛ vˆ ( s ) ⎞ ⎛ vˆ ( s ) ⎞
V ( c2 , p2 ) kcp = −V ( p2 , c2 ) kcp = − ⎜ − out − sLp iˆc ⎟ kcp = ⎜ out + sLp vˆc ( s ) ⎟ kcp (22)
⎝ N ⎠ ⎝ N ⎠

⎡ vˆ ( s ) ⎤
− ⎢ out kcp + vˆc ( s ) kc kcp sLp + vˆc ( s ) kc kic + vˆc ( s ) kc kac sLp − vˆc ( s ) kc ⎥ Z ( s ) N = vˆout ( s ) (23)
⎣ N ⎦

vˆout ( s )
−vˆc ( s ) kc ⎡⎣ sLp ( kcp + kac ) + kic − 1⎤⎦ Z ( s ) N = vˆout ( s ) + kcp Z ( s ) N (24)
N

−vˆc ( s ) kc ⎣⎡ sLp ( kcp + kac ) + kic − 1⎦⎤ Z ( s ) N = vˆout ( s ) ⎡⎣1 + kcp Z ( s ) ⎤⎦ (25)

Z (s)
vˆc ( s ) kc ⎡⎣1 − kic − sLp ( kcp + kac ) ⎤⎦ N = vˆout ( s ) (26)
1 + kcp Z ( s )

Factoring 1−kic:

⎡ kcp + kac ⎤ Z ( s )
vˆc ( s ) Nkc (1 − kic ) ⎢1 − sLp ⎥ = vˆout ( s ) (27)
⎣ 1 − kic ⎦ 1 + kcp Z ( s )

The transfer function now comes easily:

vˆout ( s ) ⎡ kcp + kac ⎤ Z ( s )

= Nkc (1 − kic ) ⎢1 − sL p ⎥ (28)
vˆc ( s ) ⎣ 1 − kic ⎦ 1 + kcp Z ( s )

The term kcpZ(s) combines reflected elements on the primary side. Let’s see how to derive it:
 Rload ⎛ RESR 1 ⎞
2 ⎜ 2
+ 2 ⎟
N ⎝ N sCout N ⎠
kcp Z ( s ) = kcp (29)
Rload ⎛ RESR 1 ⎞
2
+⎜ 2 + 2 ⎟
N ⎝ N sCout N ⎠

Developing and re-arranging all terms gives:

Z (s) 1 1 + sRESR Cout

= (30)
1 + kcp Z ( s ) N 2
⎛ 2 N 2 RESR ⎞
kcp + ⎜N + + kcp RESR ⎟
Rload R
1 + sCout ⎜ load ⎟
⎜ N2 ⎟
⎜ k cp + ⎟
⎝ Rload ⎠

Introducing this expression into the small-signal gain we have derived (Eq. (28)) gives:

⎛ kcp + kac ⎞ ⎛ s ⎞⎛ s ⎞
(1 + sRESRCout ) ⎜1 − sLp ⎟ ⎜1 + ⎟ ⎜1 − ⎟
vˆout ( s ) Nkc (1 − kic ) ⎝ 1 − kic ⎠ ⎝ s z1 ⎠ ⎝ s z 2 ⎠
= = G0 (31)
vˆc ( s ) N2 ⎛ 2 N 2 RESR ⎞ ⎛ s ⎞
kcp + ⎜N + + kcp RESR ⎟ ⎜⎜1 + ⎟⎟
Rload R
1 + sCout ⎜ load ⎟ ⎝ s p1 ⎠
⎜ N2 ⎟
⎜ kcp + ⎟
⎝ Rload ⎠

From this expression, we can identify:

Nkc (1 − kic )
G0 = (32)
N2
kcp +
Rload

A left-half plane zero:

1
s z1 = (33)
RESR Cout

A right-half place zero:

1 − kic 1
sz 2 = = (34)
( kcp + kac ) Lp Lp Vc
2 RV
i in

A pole:

1
s p1 = (35)
⎛ 2 N RESR 2
⎞
⎜N + + kcp RESR ⎟
R
Cout ⎜ load ⎟
⎜ N2 ⎟
⎜ k cp + ⎟
⎝ Rload ⎠
Entering Eq. (35) into Mathcad gives us the complete ac picture:

− 20
15

− 40
180
20⋅ log( Gac( i⋅ 2⋅ π ⋅ f ) )10 arg( Gac( i⋅ 2⋅ π ⋅ f ) ) ⋅
π
− 60

5
− 80

0 − 100
3 4 5
1 10 100 1×10 1×10 1×10
f
Figure 8: the small-signal plot of Eq. (31) using Mathcad.

Numerical application

If we now use the numerical values calculated for the various coefficients, based on Fig. 5 schematic,
we can locate the following poles and zeros:

G0 = 17.9 dB (36a)
f z1 = 1.59 kHz (36b)
f z 2 = 18.7 kHz (36c)
f p1 = 199.7 Hz (36d)

If we neglect the ESR contribution to Eq. 36d, the new pole is positioned at f p1 = 228 Hz .

Reference [2] describes the derivation of BCM structures using the loss-free network concept. The given
expressions for the gain, pole and zeros are the following:

Rload N 10
G0 = 20 log10 = 20 log10 = 17.93 dB (37)
2⎛ Vout ⎞ ⎛ 19.2 × 2 ⎞
2N ⎜ 2 + 1⎟ 2 × 0.25 × ⎜ + 1⎟
⎝ NVin ⎠ ⎝ 0.25 × 100 ⎠
19.2
2 +1
1 2M + 1 1
f p1 = = × 0.25 ×100 = 228 Hz (38)
2π Rload Cout M + 1 6.28 ×10 ×100u 19.2
+1
0.25 ×100
R 1 10 1
f z 2 = 2 load = = 18.7 kHz (39)
N 2π L p M (1 + M ) 0.25 × 6.28 ×1m 19.2 ⎛
2
19.2 ⎞
⎜1 + ⎟
0.25 ×100 ⎝ 0.25 ×100 ⎠

Reference [2] did not consider the ESR of the capacitor in the derivation of the models. Eq. (36b) does not
change.
 The full ac simulation, including phase and gain appears in Fig. 9 where we have purposely added in the
background, the Mathcad plot:

0
17.9 dB

fp1=202 Hz − 20
15

− 40 180
20⋅ log( Gac( i⋅ 2⋅ π ⋅ f ) )10 arg( Gac( i⋅ 2⋅ π ⋅ f ) ) ⋅
π
− 60

5
fz1=1.6 kHz − 80
fz2=18.7 kHz
0 − 100
3 4 5
1 10 100 1×10 1×10 1×10

Figure 9: the comparison between the Bode plot givenf by the SPICE simulation of the non-linear BCM model as
applied in Fig. 5 and the Mathcad transfer function.

As one can see, both plots perfectly superimpose on each other showing the validity of the approach described
here.

References

1. Christophe Basso, “Switch-Mode Power Supplies: SPICE Simulations and Practical Designs”,
McGraw-Hill, 2008.
2. J. Chen, B. Erickson, D. Masksimović, “Average Switch Modeling of Boundary Conduction Mode Dc-
to-Dc Converters, Proc. IEEE Industrial Electronics Society Annual Conference (IECON 01), Nov.
2001, vol. 2, pp. 842-849.