THE BELL SYSTEM

TECHNICAL JOURNAL
DEVOTED TO T H E SCIENTIFIC AND ENGINEERING
ASPECTS OF ELECTRICAL COMMUNICATION

October 1980

Volume 59

Number 8

Copyright 1980 American Telephone and Telegraph Company. Printed in U.SA.

Two-Port Analysis of SC Networks with

Continuous Input Signals
By C . F. K U R T H
(Manuscript received December 19,1979)
Switched-capacitor
(sc) networks are comprised of capacitors
in
terconnected
by an array of periodically
operated
switches.
Such
networks are particularly
attractive in light of the high circuit
density
possible with MOS circuit technology and hybrid integrated
circuits
using thin-film and silicon technology.
Their implementation
and
analysis have received increasing attention over the last years, and
economical
designs find applications
in communication
and elec
tronic equipment.
Previous publications
by the author and G. S.
Moschytz
have shown that sc networks can be analyzed
as timevariant sampled-data
networks
by using nodal charge
equations.
This led to four-port equivalent circuits in the domain, thus model
ing the sc network with a time-invariant
network. The
four-port
equivalent circuit was reduced to a two-port, and a two-port
transfer
function for the entire sc network was derived. This previous
work
assumed that the network is fed by a staircase input function. In this
paper, we show how the theory can be extended to continuous
input
functions. The results obtained are similar to those obtained by Liou
and Kuo, but, since we use traditional
two-port theory, the
derivation
is different. This derivation
makes it simple and more
intuitively
accessible. The interpretation
of the analytical expression leads to a
bypass circuit that has the properties
of a classical ring
modulator
added to the two-port equivalent circuit mentioned above. We present
examples.
I. INTRODUCTION
Previous publications by the author and G. S. Moschytz showed
that s c networks can be analyzed in the time domain by using nodal
1297

charge equations with time-varying coefficients.

This led to a rep
resentation of s c networks as time-variant sampled-data networks,
which can be modeled by a discrete, time-series, sampled-data system
in cascade with sample-and-hold devices. Ultimately, it was shown
that any s c network can be represented by a four-port equivalent
circuit in the Z-domain. Throughout this analysis, it was assumed
that the s c network is fed by a staircase voltage function to be
compatible with the inherent staircase nature of a voltage across any
internal capacitor of the s c network. T h e charge transfer occurs at the
end of the closing time of the switches; consequently, the magnitude
of the voltage at every node of the network must be constant over the
previous inactive interval of the switches. This leads to a staircase
approximation of a compatible input function, as shown in Fig. 1. T h e
closing times and of the two sets of switches are assumed to be
different. One can define, however, a common period of 2 = + .
With this common period of 2, it is possible to model the enitre s c
network with a sampled-data system of a sampling rate 1/r followed
by sample-and-hold circuits for restoring the physically realistic stair
case output function. T h e following analysis forms the link to the
previously explained ideas in Refs. 1 and 2. It is a prerequisite to the
subsequent analysis with continuous input functions.
XX3

II. E X A C T ANALYSIS WITH STAIRCASE INPUT

B y continuing the ideas of a sampled-data system model of scnetworks as introduced in (1) and (2), the staircase input function
shown in Fig. 1 can be expressed as

Fig. 1Staircase approximation of input function to switched-capacitor network.

1298

THE BELL SYSTEM TECHNICAL JOURNAL. OCTOBER 1980

where (*) stands for convolution and /l(r) and f {t) are the proper
rectangular pulse functions of length and , respectively, consistent
with the ones used in section III of Ref. 2. T h e y convert a train of
impulses into a staircase function. A s can be noted, eq. (1) consists of
two terms, a term with n' = even sampling times and a term with n"
= odd sampling times. T h e second term with n" = odd can be expressed
with n ' = even by introducing " = n' + 1. With a further substitution
of variables,
2

] T2

t = t + j \

and

= - ,

eq. (1) can be rewritten as

vfU) = JiMO

j^Sit

')^)

+ I vdt + , - - ) - S(t - ' - - ) l*/i(r). (2)

|_
'-0
J
Now in the term representing the odd samples, can be considered a
shift of the baseband signal t>i(f) with respect to the sampling time '
, i.e., the instantaneous value of >('). r is a delay of the entire
signal by one-half the previously defined period of 2 = + , and
is a delay of the entire series of impulses after sampling. With a
frequency variable of after sampling and a periodic variable of
e*" (corresponding to a Z-transform with a period of = 2-n/Qc), eq.
(2) can be expressed in the frequency domain:
2

\'(, ) = V , ( z ) - F ( ) + ^ - ' e ^ ^ . F . M .
2

(3)

V\(z ) represents the input signal in the Z-domain sampled at a rate of

' .
T h e frequency in eq. (3) corresponds to the baseband frequency
of the sampled signal (vAt) <-> V,(wo} before sampling. It is related to
the frequency variable by
2

= + mfio/2

m = 0, 1, 2, 3, ,

(4)

which later will be used in the spectral interpretation of eq. (3).

N o t e that e'" ' is a phase shift for iMf) before sampling. T h e
frequency functions F (<j) and F (u)) are the functions corresponding
to f\(t) and f (t) in the time domain. T h e y disperse the samples of f i )
to rectangular pulses in the direction of the negative time axis and can
be expressed as
T

2 sin /2 ^

TWO-PORT ANALYSIS OF SC NETWORKS

( 5 a )

1299

and
Fi(w) =

2 sin / 2
2

e*" .

(5b)

T2/2

In pursuing the analysis of an s c network with a staircase input,

only the discrete samples of the sampled-input function At) are
necessary to be considered as an input function, provided that the s c
network has been modeled as a sampled-data system with a sampling
rate of l / (or a period of 2 and 50 percent duty cycle). This was
explained in Refs. 1 and 2. It leads to a general block diagram of the
entire s c network in the time or frequency domain as shown in Figs.
2 and 3.
T h e Z-transformed part of the sampled input function can be ex
tracted from eq. (3) as
= VMz'Ml + z" V * * " ) .

V\'(z) = VAz ) + VAz )z~ e>^

2

(6)

T h e physical delay of e~
is applied to the odd samples at the output
of the sampled-data system, which brings the sampling back to the
nonequal sampling times and as shown in Fig. 3. F\ () and /^()
at the output convert the series of discrete samples into a staircase
function, as it appears physically at the output of any s c network.
Finally, the even and odd samples have to be added into one
consecutive string of pulses with alternating even and odd time slots.
With the four transfer functions, "(), "(), /() and i f ( z ) , as
they appear in the model of the s c network, the overall output signal
can be expressed by inspecting Fig. 3 as
JuliT

V?(co) = ()() + VS(z)F,(u>)

VUz) = VAz )[H"(z)
2

VS(z) = V (z )[ir(z)
2

+ H (z)z- e ^]
ot

(8a)

+ IT"(z)z- e ']e- " ,

l

JUor

(7)

&T

(8b)

( ! - n

S(t-n TI

SAMPLED-DATA
SYSTEM MODEL
FOR
SC NETWORK
PERIOO 2T
50% D U T Y C Y C L E

(t-n T)
e

'rT,, T,

T|-T,

Fig. 2Time domain equivalent circuit for switched-capacitor network with unequal
switching times , and r .
2

1300

THE BELL SYSTEM TECHNICAL JOURNAL. OCTOBER 1980

Fig. 3Frequency-domain equivalent circuit for switched-capacitor network with

unequal switching times , and .

and after substituting (8) into (7),

V' '(u) = V , ( 2 ) { [ i f (*) + / T ( 2 ) 2 " V ' ] .F (u>)
2

+ [H"'(z) + H"(z)z- e" ']e- ^-F (u)).

l

JnT

(9)

T h e four transfer functions H(z) have to be determined from the fourport equivalent circuit in the Z-domain for any individual s c network
under investigation. T h e derivation of the four-port equivalent circuit
has been described in Refs. 1 and 2. In the latter part of this paper,
examples will demonstrate the process of using it.
T h e significance of eq. (9) is that it describes the overall transfer
characteristics of the entire s c network fed by a staircase input in one
expression. This expression was derived in the frequency domain from
a time-invariant model of the time-variant s c network. It might be
pointed out that this time-invariant model has some similarity to the
time-invariant models of periodically time-variant networks as de
scribed in Refs. 4 and 5.
In the following section, a correction term is introduced that extends
the analysis from a staircase input to a continuous input function.
III. ANALYSIS FOR CONTINUOUS INPUT FUNCTIONS
B y approximating the continuous input function i>i(f) with a stair
case function as indicated in Fig. 1, an error was made as shown in Fig.
4. This error can simply be calculated as
e,(r) = -,() - '().

(10)

It is important to notice that the error is always zero at sampling

times, since the instantaneous values at sampling times have already
been processed with the staircase input analysis. Consequently, what
is left to be considered is the error function input signal between
sampling times. From a physical inspection of any s c network, it can
be determined that the network is time-invariant between two samTWO-PORT ANALYSIS OF S C NETWORKS

1301

,(tl = V , ( t ) - V ? ( t |

^M^Hi^^r^

Fig. 4Error function of the input of the switched-capacitor network.

pling instances. It behaves like a capacitive voltage divider that might

have some active devices imbedded. T h e gain or loss of this capacitive
voltage divider will change periodically when the switches alternate
between positions for and . Thus, any error voltage appearing
within the closing times or will be attenuated (or amplified)
according to the constant loss or gain of the network within these
periods. T h e result is that a periodically modulated error signal will be
superimposed on the output signal.
T h e modulating function can be determined by calculating the static
dc gain of the s c network for the two different switching positions. Ai
is the gain for the switch position n , and A corresponds to
consequently, a switching function S(t) can be defined as shown in Fig.
5. It can be separated into a constant term Ao and a function A(t) as
shown in Fig. 6, which alternates between a positive and negative
value. This makes it possible to interpret the switching operation of an
s c network as that of a conventional ring modulator with unequal
switching times and a gain of A\ Ao, offset by a dc path of Ao. T h e
resulting output error function e (t) is the simple product of the input
error e\(t) and the switching function
2

, S(t)-A|tl + A
A,T,
A

A T
2

II

II

A II
2

'

I 1
7 2

I I
3

II

10

,.

Fig. 5Periodically changing gain of sc network between sampling times.

1302

THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980

All)

,
1

2
4

10

Fig. 6Modulating function of sc network between sampling times.

e (t) = e (t).S(t),
2

or with eq. (10) and S(t) = A

+ A(t)

e (t) = v (t)-[Ao
2

(11)

+ A(t)]

- wf(i).S(f).

(12)

Substituting into (12), the expression for u f (r) from eq. (2) and knowing
that S(t) = Ai for any time interval and S(i) A for any time
interval r results in
2

i>,(f).A(r)

e (f) - t;,(<)-A +
2

A |i>,(f).
[

-A,j(/,(f

"

+ , - - )

/
J

n'-O

(-'--)|,().

(13)

With the previously introduced baseband frequency spectrum V (uo)

and by
before sampling and the Fourier transform of A(t) as &{A(t)},
using (3), eq. (13) can be expressed in the frequency domain:
t

Ei(u)

AoVi(ud)

+ ,()*.^{()} - A V (z )

.F {u)

- A, V,U )2-V'
2

U0T

'e- *
>

,AT

.F,(u),

(14)

where * is complex convolution.

Finally, the output in response to a continuous input function
becomes the sum of eqs. (9) and (14).
V () = V '() + (),
2

(15)

which yields
V M =.,() +

VA<*>)*?{A(t))

+ VAz )[H"(z)

-A-i

+ VAz )[H "(z)

+ (H'-(z)

H" (z)z- e ^]F (w)

r

- Ai)2 V^'Je-^-Fitu).
_

(16)

T h e additional terms in eq. (16) can be included into the model by a

slight modification of Fig. 3. This is shown in Fig. 7, which represents
TWO-PORT ANALYSIS OF SC NETWORKS

1303

the entire s c network with a continuous input signal. T h e terms A

and A2 appear as frequency independent bypasses added into the
stream of discrete samples. Ao is a direct through path for the baseband
frequency spectrum. T h e path through the modulator represents the
term of the baseband signal multiplied by A(t).
It may be appropriate to point out here that not all s c networks
have a through path A\ and A during the closing times , and r ,
respectively, of the switches. In many cases, and Ai may be zero.
T h e n Fig. 7 reduces to Fig. 3. Other cases may exist where A\ (or A2)
only is zero. T o better understand the physical meaning of eq. (16), it
is necessary to interpret its spectral contents over the entire frequency
range . This is the subject of the following section.
t

IV. SPECTRAL INTERPRETATION O F T H E O U T P U T SIGNAL

Assuming the spectrum Vi(too) of the input signal v,{t) is given as
indicated in Fig. 8, the first term in eq. (16) will generate the same
spectrum at the output of the system, scaled, however, by the constant
factor Ao. It is shown on line 1 in Fig. 8. T h e spectrum is nonperiodic;
consequently, m = 0 in eq. (4) and = for this part of the output
spectrum.
T o evaluate the spectrum of the second term in eq. (16), it is
necessary to know the spectrum of A(t) as shown in Fig. 6. It is a
0

(t-n T)
e

v,u i

H (2|
M

- -

HO
|'(

S/H

\:

Ol

hW ij
v

u>
2

MODULATOR
A(tl

(!
2

Fig. 7Frequency domain model for sc network with continuous input and unequal
switching times , and -.
1304

THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980

LINE

, INPUT SIGNAL
0

DC T H R O U G H PATH

Bs.
2
MODULATOR
SPECTRUM

EVEN TERM
SPECTRUM

4 -2^

ODD T E R M
SPECTRUM

12
-
*3

Fig. 8Output spectrum distribution of sc network.

periodic signal; thus, its spectrum must be discrete. T h e function A(t)
can be expressed in the time domain as a series of impulses convolved
with the function f\(t ) , which is the previously introduced function
fi(t) delayed by . It disperses the impulses to pulses of finite width
T| in the positive time direction. With a correction term (A )
/( + ), which assures a zero dc term, it follows:
t

A(t) = (A\ -

A)
2

SU - m2r) *Mt - ,)
- (Ai -

A)
2

Tl

(17)
T

With the corresponding function of Fi(w) as shown in eq. (5a), the

spectrum of eq. (17) results in
*{A(t))

= y

(,

-A )
2

2 sin /2

( - mo/2)

- 2w(i4, - )

->UT|/2

*M,

(18)

TWO-PORT ANALYSIS O F SC NETWORKS

1305

Tl +

Tj

which can also be written as

&{A{t)}

= sin m

= 2( - )

'

(-,/2)

Jm

m oc

- 2 TTL4, - A )

'
().
+ T

(19)

N o w we can convolve eq. (19) with the baseband signal Vi(w) to obtain
the spectrum of the second term in eq. (16).

m=-i

It

. ,( - mo/2) - (A, - A )

V,().

Ti + T

(20)

A s can be noticed, the spectrum {)*&{)

repeats periodically
with mSlo/2 over the entire frequency spectrum oo < < +oo with a
declining magnitude for higher m. This corresponds to the well-known
performance of an ideal ring modulator with unequal switching times.
B y studying the declining magnitude of the periodic spectrum, it can
be observed that the spectrum for even m is inherently smaller in
magnitude. It will disappear completely for = = , which is also
a typical property of the ring modulator. T h e spectrum is indicated in
Fig. 8, Unes 2 to 5.
Finally, the spectrum of the last two terms of (16) can be explained
by interpreting the spectrum of V i ( z ) , which was the Z-transform of
the sampled baseband signal Vi(t)- m - - " 8(t m2r).
With the spectrum of the sampling function

+00

-f oo

<5( - mo/2) ~

/oo

8(t - m2r),

(21)

/ 00

the spectrum of Vxiz ) can be expressed by convolution

2

V^z ) ,()*-.

(-/2)

= e~ " \
2

(22)

m<

which is equal to
)=^-

-/2)

(23)

rn on

Consequently, the last two terms in eq. (16) can be interpreted as the
sum of the periodically spread-out baseband spectra shaped by a
transfer characteristic depending on and m. With eq. (23), e~
JUT

1306

THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980

and - /2 substituted into eq. (16), we obtain:

T e r m 3 = ^ - Vi ( - m&>/2)
2
[#"()

+ H ( )e- e ot

T e r m 4 = ^ - ,( 2

jUT

^-F ( ),

(24)

i,M mSh/2)T

2 u

mo/2)

[//"() + ( () - A , ) e - " e " - '

y

> <

n n

/ 2 , T

'>- " .F ( ).
>

A T

(25)

T h e spectra corresponding to eqs. (24) and (25) are indicated in Fig.

8 on lines 6-9 and 10-13. T h e functions () and F (w) cause a decline
of the magnitude of the spectra towards higher frequencies with a
(sin x)/x roll-off.
T h e results of the above interpretations of the spectral parts can
now be combined to a new expression for eq. (16) where the term A
( )/(
+ ) A is being used (see Fig. 5 for Ao):
2

V (<o) - A , V i ( ) + T
Vi(

+ ["() -A

J y
1

*
m

*****

H^e-^e' "-^^']
1

2
+ [ ~ ( ) + {-() - A i } e -

/ U T

e "-"
/ (

, O 0 / 2 , T

'] - ^

.^^|.

(26)

Equation (26) can be used for calculating the power in any particular
frequency range < < of the entire spectrum by taking the
integral of | V (to) | over that range. Although this is a laborious
process, it will be useful for the noise analysis of s c networks. For such
noise analysis, ^) can be considered as the noise source at any given
point of the network. T h e '*() are the transfer characteristics of the
network from the noise source to the output. Via superposition, the
noise power from all sources can be accumulated at the output.
In closing, it might be pointed out that eq. (26) also can be expressed
in terms of the baseband frequency by substituting eq. (4). In
addition, one can take advantage of the following observations:
2

z = e '
2

= e <"

Ju2

+mn

"

/2,2T

= e^V"

1 2

" = e'^

for all m.

This means that all even functions in (or functions in z only) like
H"(z) and H"z~ are periodic, and the following holds for all m:
2

"( + mflo/2) = "()

(27a)

and
TWO-PORT ANALYSIS O F SC NETWORKS

1307

""( + m i 2 / 2 ) e - "
> (

u + m a

/ 2 , 2 T

= H'(wo)e- .

(27b)

JaaT

Furthermore, the odd functions can always be considered as a function

in z multiplied by z . Consequently, with
2

the odd functions in eq. (26) can be expressed as

/( + nulo/2) = ( - l ) / f ( w o )
m

(28a)

TO

and
( + m /2)e-

= (-1)"'' ()" " .

jWo+mSki/2)T

00

(28b)

With this and resubstituting eq. (4) into (26), the final result is
V (<oo + /2) = 7( )
2

.{a

2 +

!!^I -M,/4
e

mCD

+ ,()
[#"(<*,)

-A

+ / / ( - ^ ^ ' ] ^-

2T

^ ( + //2)

1 OD

Vi((Oo)

[#"(<*>) + () - Atf-^]

(-l) e- ^ ^ "F ((o

m

ma

2
+ mo/2).

(29)

Equations (26) and (29) describe any s c network in general terms and
are equivalent to the slightly different expressions derived by M. L.
Liou and Y. L. K u o . B o t h eqs. (26) and (29) are equally valid.
Whichever is more convenient in a particular case can be used. It
should be remembered that in eq. (29) each part of the spectrum is
calculated relative to the frequencies mo/2. This should be convenient
for calculating just the spectrum relative to one particular frequency
mo/2 by knowing the baseband spectrum Vi(con).
11

1308

THE BELL SYSTEM TECHNICAL JOURNAL, O C T O B E R 1980

V. EXAMPLES
S. 1 Double sample-and-hold

circuit

As a simple first example, the analysis of a double sample-and-hold

circuit shown in Fig. 9 is demonstrated. B y applying the technique
described in Ref. 1, a four-port equivalent circuit as shown in Fig. 10
can be established in the Z-domain. T h e "link two-ports" ( L T P S )
correspond to the storage capacitors Ci and C in Fig. 9. For a
convenient further reduction of the circuit in Fig. 10, some properties
of link two-ports may be recapitulated. As derived in Ref. 2, a conven
ient equivalent circuit for an L P T is the one shown in Fig. 11. If the L T P
is driven by a voltage source, Thevenin's theorem yields an equivalent
circuit as shown in Fig. 12. This is an important equivalent circuit, and
it can be used advantageously to reduce complex configurations. An
other important circuit element is an open-ended L T P whose input
impedance frequently occurs in s c network equivalent circuits. Al
though derived in Ref. 2, its equivalent circuit is shown in Fig. 13 for
the convenience of the reader. Finally, it is shown in Fig. 14 how an
L T P reduces to two "non-linked" resistors by letting z ' = 0. T h i s
special case is convenient for calculating the through-path coefficients
and A from the four-port equivalent circuit in the Z-domain.
2

After this short introduction, the circuit in Fig. 10 can be reduced to

the one in Fig. 15 by using the equivalent circuit in Fig. 12.

C O N T I N U O U S INPUT

STAIHCASE O U T P U T

Fig. 9Double sample-and-hold circuit.

TWO-PORT ANALYSIS O F SC NETWORKS

1309

Fig. 10Four-port equivalent circuit for double sample-and-hold circuit.

T h e four transfer functions can b e established by inspecting Fig. 15.
VI
H" = - = 0

H" =

Vi

Vi
Vi

H = = 0-

V?
,
- = = e~ "
J

V\
vi
Vi

H" = = z~ = e~
e

JUT

It also can be seen that t h e circuit does not have any static feedthrough; consequently,
= 0

and A = 0.
2

A s a result, w e obtain from eq. (26)

V (w) = ,( - mno/2)e-'" V ' -
z

<

ma

/2)T

F (u)
'
2
2

+ ,( - mo/2)e- eJur

Ja&T

,()

. (30)

k--Cz-'-H

Fig. 11Equivalent two port for link two ports.

1310

THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER 1980

-Cl

C=

fi

Fig. 12Thevenin's equivalent for LTP driven with voltage source.

T h e output spectrum consists virtually of all terms for all m including

m = 0. () and F (w) introduce shaping towards higher frequencies.
A special case where = r-> = is of interest. Under this condition,
Fi () = F (w) and (30) reduces to
2

V () =

- mQo/2){\

+ (-1)"}

or, with eq. (5a), = :

1 - e ~""

VM - mi2o/2){l + (-1)"*).
|

V-)

JU2T

(31)

For odd m, the spectrum is zero, which indicates that the spectrum
is related to a sampling period of r. T h e double sample-and-hold circuit
is of practical value, since it can be used to convert a continuous input
signal into a staircase function, without any feed-through. T h e result
is a spectrum that lacks all the extra terms shown in Fig. 8 on lines 2
to 5. T h e circuit can be used conveniently as driving circuit with
resistive source impedances in the domain, as shown in Fig. 15. Any
subsequent circuit can be designed exactly as a sampled-data system
with a sample-and-hold circuit in tandem. Noise due to feed-through
does not occur, regardless of potential feed-through paths in the
subsequent SC network. It is not necessary to decouple the double
sample-and-hold circuit from a following circuit via an active device.
5.2 First-order high-pass

section

A s a second example, a first-order high-pass section as shown in Fig.

16 will be analyzed. Again applying the technique described in Ref. 2,
9

Fig. 13Open circuit input impedance of LTP.

TWO-PORT ANALYSIS O F SC NETWORKS

1311

FOR

" -
1

C=1/R

- R - 1/C--<

Fig. 14LTP reduced to nonlinked two port for Z~ ' 0.

the equivalent four-port circuit as shown in Fig. 17 can be derived in
the Z-domain. The four transfer functions can be derived as
V = 0.

(a)

This results in V = 0, Vu = 0, Vi = 0. N o current flows in the even

path. Consequently,
22

v%

(32a)

" ~

(32b)

(b)

V? = 0.

N o current flows in the even path, which corresponds to an openended L T P in the even path. Consequently,
V|

i?2

V'

R + Zu
2

Cl
C, +

<V(1 -

z~ )
2

where Zu was substituted with the expression in Fig. 13.

Applying the equivalent circuit in Fig. 12 yields V = V z~ .
can be calculated from Vf via voltage division
x

22

Fig. IiiReduced four-port equivalent circuit for double sample-and-hold circuit.

1312

THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER

1980

V II)

CONTINUOUS

Fig. 16First-order high-pass section.

which yields
-z~ C
l

Vf
With = C /(Ci
2

Vf

+ C) H

c,(i - - ) +

vf

cY

and if can be rewritten more conveniently:

ee

-)( -)
2

H-

eo

(33a)

- 1+
2

(33b)

z ' - l +

Finally, the coefficients and A of the through path can be calculated

by substituting z~ = 0 into the L T P , as was explained in Fig. 14. This
2

E V E N PATH

rh

ODD PATH

I"

Fig. 17Four-port equivalent circuit for high-pass section.

TWO-PORT ANALYSIS O F S C NETWORKS

1313

eliminates all memory and separates the even and odd paths com
pletely. T h e static value of the storage capacitor C i , which corresponds
to a resistor, remains in each path. This is indicated in Fig. 18, from
which and A2 can be derived by inspection [A\ corresponds to odd
samples and A to even samplessee eq. (13)].
2

A, = / r " ( 0 ) = - r p - = l

(34a)

y1

, . - , - ^ - . ^ - - .

(34b)

Substituting eqs. (32), (33), and (34) into eq. (26) results in the following
expression for the entire output spectrum:

V ( ) = (-))
2

+ -
It

V.< -

mft,/2)

8 m m

T l / 4
e

^W<

4- 2, Vi ( - mS2o/2)

-'""~^ Fi ()

^ - ^ ) / .

^ .

(35)

As we can see, virtually all possible parts in the spectrum are

present. Notice that parts of the direct terms Vi(<o) will be cancelled

Fig. 18Four-port equivalent circuit for high-pass section for ' = 0 .

1314

THE BELL SYSTEM TECHNICAL JOURNAL. OCTOBER

1980

by the baseband content of the other terms. T h e example, although a

relatively simple circuit, shows how complex the analysis can become
when the input signal is not a staircase function but rather a continuous
function. This particular result in eq. (35) is an example of where a
transformation into the form shown in eq. (29) might be advantageous.
It allows an easy calculation of every spectral component around /
2 separately. T h e same example was treated by T s i v i d i s . After
further reduction, eq. (35) leads to the expression obtained by Liou
and K u o ' .
910

1 1

VI. CONCLUSIONS
We have shown how the two-port analysis of s c networks, which
was based on staircase inputs, can be extended to continuous input
functions. This extension still allows us to model the s c network as a
four-port representing a sampled-data system in tandem with sampleand-hold circuits. After a slight modification of this model, continuous
input signals can be treated. For convenience or for practical reasons,
most s c network designs will be based on the staircase approximation
of the input signal. For any noise analysis or the introduction of
nonlinear terms, an analysis with a continuous input signal is neces
sary. T h e method shown demonstrates how an s c network designed
for a staircase input can be post-analyzed with respect to its response
to a continuous input signal. As explained, the numerical analysis can
be achieved with computer programs based on mesh or nodal analysis
principles and a post-processor for the extension to continuous input
signals.

VII. ACKNOWLEDGMENT
T h e author acknowledges the free exchange of ideas with Y. L. Kuo,
M. L. Liou, and Y. Tsividis, who were working at the same time on the
same problem by taking different approaches to the solution. T h e
author also appreciates . I. Jury's study of the manuscript and his
subsequent comments.
REFERENCES
1. C. F. Kurth and G. S. Moschytz, "Nodal Analysis of Switched-Capacitor Networks,"
IEEE Trans. CAS, 26, No." 2 (Feb. 1979), pp. 93-105.
2. C. F. Kurth and G. S. Moschytz, "Two-Port Analysis of Switched-Capacitor Net
works Using Four-Port Equivalent Circuits," IEEE Trans. CAS, 26, No. 3 (March
1979), pp. 166-180.
3. K. R. Laker, "Equivalent Circuits for the Analysis and Synthesis of Switched
Capacitor Networks," B.S.T.J., 58, No. 3 (March 1979), pp. 729-769.
4. C. A. Desoer, "Steady-State Transmission Through a Network Containing a Single
Time-Variant Element," IRE Trans. Circuit Theory (Sept. 1959), pp. 249-252.
5. C. F. Kurth, "Steady-State Analysis of Sinusoidal Time-Variant Networks Applied

TWO-PORT ANALYSIS O F SC NETWORKS

1315

6.
7.
8.
9.
10.
11.
12.

to Equivalent Circuits for Transmission Networks," IEEE Trans. CAS, 24, No. 11
(Nov. 1977).
A. Papoulis, The Fourier Integral and its Applications, New York: McGraw-Hill,
1962.
. I. Jury, Theory and Application of the Z-Transform Method, New York: John
Wiley, 1964; republished by E. Kreiger Pub. Co., 1973.
G. Doetsch, Guide to the Applications of the Laplace and Z-Transforms, London:
Van Nostrand Reinhold, 1971.
Y. Tsividis, "Analytical and Experimental Evaluation of a Switched-Capacitor Filter
and Remarks on the Resistor/Switched-Capacitor Correspondence," IEEE Trans.
CAS, 26, No. 2 (Feb. 1979).
Y. Tsividis, "Analysis of Switched-Capacitor Networks," IEEE Trans. CAS. (Nov.
1979).
M. L. Liou, Y. L. Kuo, "Exact Analysis of Switched-Capacitor Circuits with Arbitrary
Inputs," IEEE Trans. CAS, 26, No. 4 (April 1979), pp. 213-223.
Y. L. Kuo, M. L. Liou, J. W. Kasinskas, "An Equivalent Circuit Approach to the
Computer-Aided Analysis of Switched Capacitor Circuits," IEEE Trans. CAS, 26,
No. 9 (Sept. 1979). pp. 708-714.

1316

THE BELL SYSTEM TECHNICAL JOURNAL, OCTOBER

1980