You are on page 1of 4

358

IEEE JOURNAL

OF SOLID-STATE

CIRCUITS, VOL. SC-15, NO. 3, JUNE 1980

Finite Amplifier Gain and Bandwidth Switched-Capacitor Filters


GABOR C. TEMES, FELLOW, IEEE

Effects in

AbstractThis work discusses the effects of finite operationalamplifier gain and bandwidth on the response of the most widely used switched-capacitor filter section. Formulas are derived for the minimum acceptable values of the dc amplifier gain and the unitygain frequency under specified conditions.

+ 5 2
VI c,
La

I. INTRODUCTION

IG. 1(a) illustrates the circuit diagram of a switchedcapacitor filter (SCF) section which is used as a general building block in almost all SCl?s [1]- [7]. Fig. 1(b) shows the timing diagram, including the location of the sampling points tn.l, tn, etc. The response of the operational amplifier can be described by the Laplace-transform relation
A(s) A ~O(S) = _ ~ _ -Aouo

:; .

v(s)

S+(70
1

s+uo

(1)

Q
-Va+ +

A(s)

(a)

Hence, in the time domain,


U(t) =-+uo(t)~ duo (t) ~t .

(2)

In (1) and (2), A. is the magnitude of the dc gain, U. is the unit y-gain radian frequency, and -0. is the pole radian frequency. It follows from (2) that, unless v(t) contains impulses, v.(f) is a continuous function of time, If U. is infinite, the relation between the z-transforms of the output sequence Uo,. ~ U.(tn) and the input sequences vi,n~ ZJj(tn), i= 1,2, ., lVis

kT/2 1
I

+T/z

+T/z

+T/z

++

I
tn + 1/2 t

T/2 +

I
I

I tn.,

kl-1/2

+3/2

tn+l

(b) Fig. 1. Switched-capacitor filter section. ing diagram. (a) Circuit diagram. (b) Tim-

V()(z) -

&*
a

$jCiV~(z).
i=l

(3)

Uo(t)=u.(t)+ u(t).

(4) requiresl (5)

The corresponding formula will now be derived for the case when co. < CQ. The calculations are straightforward but lengthy, and only the most important steps will be described, Two transients have to be considered. The first occurs after the moving contacts switch to the right (R transient), the second one aft er they switch to the left (L transient ). For simplicity, we consider first the case N = 1; the results will then be generalized to N >1. The analysis will assume that the amplifier is linear, and will neglect such other important limiting factors as the slew rate and settling time of the amplifier. II. R TRANSIENT
This is the transient taking place for fn_l <

Also, for N= 1, charge conservation Ca[%(t) - %,n-l]

= Cl[u(f) - ul, n_l].

From (2), (4), and (5), we obtain


1 kwo
where

C&(f) +
dt

/1+ -_!_ ljo(t)= on-l


\

kAo

(6)

(7) effect becomes negligible if Hence, the required dc gain Ca << Cl. Otherwise, the response will depend Neglecting l/(kAo) in (6) and solving for the
this

The finite-de-gain

A. >>1 + C1/Ca.
t< tn-1/2. From

Fig. l(a) Manuscript received March 9, 1979; revised November 7, 1979.


The author is with the Department of Electrical Sciences neering, University of California, Los Angeles, CA 90024. and Engi-

i.e., is very large if on Ao. output voltage


kAo >>1, and ua, n-l ~

1Here, as throughout ua(tn_~).

paper, VI,n_l ~ ul (tn-1)

0018 -9200/80/0600-0358

$00.75

@ 1980 IEEE

TEMES: SWITCHED-CAPACITOR FILTERS gives2 UO(t)=On_l[l -e


-kw~(t-tn.l)l +vo,n-le-kuo(t-tn-l)< (8)

359

and ~~[1-k(l-e WOT/2)]e-kwOT/2. (16)

Next, (5) and (2) give Us(t) = On_l [1 - (1 - lc)e-kUO(t-fn-l)] +Uo, n_.l(l - lc)e
-kwo(t-tn.l)

(9)

The values of the voltages at the end of the R transient, vo, n_112 and va, n-1/2~ can be obtained from (8) and (9), respectively, by setting t - tn_l equal to T/2. The time constant of the R transient is
_ca+cl

Cauo

(lo)

If co. ~ W, then Va(z) -+ Vo(z) and (14) reduces to (3). Otherwise, two important effects appear. The gain VO/ VI of the section is reduced by a factor of (1 - 8). This corresponds to a relative change equal to -8 in Cl /Ca. Hence, for a given value of and a given permissible tolerance on Cl /Ca, (16) can be solved to get the minimum allowable value of U. T. Note that 6+1 ask-+0. In addition to this effect, the factor z/(z - e) introduces a parasitic pole and zero at e and O, respectively, into the transfer functions. The resulting change in the gain of the stage over the range O < u G n/T is Aa = 20 loglo ~ peak-to-peak; dB phase change is3 (18) (17)

Thus, t_R >> 1/~0 if Ca << Cl, i.e., the time constant of the transient can be much longer than 1/coo. III. L TRANSIENT In the period tn_ ~12 < t< tn, Us(r) remains Ua,n_l/,. Hence, from (2) and (4), 1 duo(t) +
GJo dt

the corresponding

constant

at

Afl = arcsin e rad.

1 1+
() A.

~O(f)= ua, n-1/2.

(11)

Hence, for a given k and a permissible Aa and/or AI?, the minimum vrdue of ~. T can be found from (1 5). In most cases, the frequency range of interest satisfies uT<< 1. Then z-l = e-@Ts 1- jc.JT, and (14) gives
Vo(e~@T) Cl 1-6 - Ca ()l-e

Neglecting 1/A. and solving (11) gives Uo(t)= a, n-l/2 [1 - e


-~o(t-Ga-l/2)

e-jwT
jwT

1
(12)

VI (e@T)

(19)

+ vo,72-q2e
a(t) = a, n-l/2.

wO(t-tn-l/2)

Hence, to a first approximation, only the gain is affected. It is interesting to note that while ~(k) is a monotone creasing function of k, e(k) has a maximum. It occurs for
k max=

de-

The value of Uo,n is obtained by setting t - tn-i/2 = T/2 in (12). Clearly, U~,n = ua,n-l/2The time constant of the L transient is TL=L<7R. Uo IV. TRANSFER FUNCTIONS (13)

2
tio T

(20) Equations (17), (18), and (20) thus give an upper bound for the effect of the added pole-zero pair, even before Cl /Ca is known.

The z-domain tqmsfer functions can now be obtained from the following steps. 1) The values of vo,n_l/2 and va,n-1/2j obtainable from ($3) V. MULTIPLE-INPUT SECTIONS and (9), are substituted into (12). For the general case when N> 1 [Fig. 1(a)], only minor 2) Replacing t- tn-1/2 by T/2 in (12), two linear difference modifications of the above derivation are required. In (5), equations, relating the unknowns uO,~, vo, n.1, va, n, @ va, n -t the RHS is replaced by with each other and with V1, n -1 result. 3) Taking the z-transform of these two difference equations, (21) f Ci[w) - Wn-,)] = cpw) - qJ+4n-J two linear equations are obtained. These can be solved for i=l Fo(z) and Va(z). where The resulting transfer functions turn out to be Vo(z) Vi(z) =-z
a(z) ..

c, (zcl c=

(1 - 6)Z 1)(2-C)
e)z-e j=l

(1-a+ (z-

VI (z)

l)(Z - e)

(14)
p(t) = f

jzl

Civi(t)

where ~ ~ ~ e-(l+k)wo2/2 zln the solution, the contimrity Of Vo(f) at ~n-1 must be sed-

I
jcl
f

Ci.

(22)

(15)

3Equations (17)-(19) assume that the stage is SimPIY u~d as a cascaded twoport in the overall system. As shown in the next section, if there is feedback in the system, the calculation of the gain rolloff is somewhat more complicated.

360

IEEE JOURNAL OF SOLID-STATECIRCUITS, VOL. SC-15, NO. 3, JUNE 1980 .

Therefore, the replacements Cl ~ CP and VI(t) ~ UP(t) should be performed in all formulas. The resulting transfer functions are thus, from (14),
vi) (z)

1
=

(1 - S)z

~
i=l

CjVi(Z)

Ca (z - 1)(Z - e)

F5
I
Cz + VI

Va(z) ~ C, Vi(Z) jzl

1 . -
Ca

(1-a

+e)z-e

(z - 1)(Z - e)

(23)

It should be noted that one of the inputs vi(t) can, in fact, be chosen as the output UO (t). Consider, e.g., the circuit of Fig. 2. Its transfer function, from (23), is
v~ (z) =-cl

(1 - ti)z Ca(z - l)(Z - e)+c~(l - S)z

(24)

v, (z)

If Cl = Cz = Ca, the transfer function of the circuit with an ideal amplifier is that of a negative delay: V./Vi = -z 1. Due to the finite gain and bandwidth, the actual transfer function is, from (24), -(1 - (5)Z Vo(z) . VI (z) z-(e+6)z+e
(25)

i j ~~ J+q+
Fig. 2. Delay circuit.

. c

c.

c1 v

A(s)

.+

Fig. 3. Simulated inductor circuit.

From (25), the actual loss and phase of the section can be predicted. The formulas given in Sections II and III can also be used to derive the transfer function for sections in which some of the capacitors Ci [Fig, 1(a)] are switched in opposite phase. Such sections include the simulated inductors of Hosticka et al. [7] and Temes et al. [8]. As an example, Fig. 3 illustrates the simulated inductor described in [7]. The charge entering during the nth clock interval is
dnz-) - q(nl q = cl(%,n - %/2) (26)

VI. CONCLUSIONS The derivations given in the paper show that the effects of finite dc gain A. and finite unity-gain radian frequency tio will be negligible in the general SCF stage of Fig. 1(a) if the following conditions hold:
kAO >>1

Q[1 -

lc(l - e-wOT2)] e-kwOT2 <<1

(29)

+ C2(l, n - O, n-l/2).

Now vO,n-1/ from (8) and z-transform of and Va(z) from Q(z) =2c1{[z VI (z)

and vn-1/z = vo, n-1/2 - va, n-112 can be found (9) setting t - tn-l= T/2, Then, taking the the resulting relation and substituting VO(z) (14), we obtain +(l-@][z-l][z-e]

+(1 - @)[(l -8 +e)z+el ++(1 6) Z}[(Z - 1)2(Z - C)]-l


(27)

where k ~ Ca/ [Ca + ~~. ~ Ci]. Clearly, the conditions given in (29) become hard to satisfy if C. << Z:, Cl so that k<< 1. If the above conditions are not satisfied, then (23) can be used to predict the effects of the finite gain-bandwidth product on the overall circuit response. It should again be emphasized that the analysis has not taken into account the effects of the slew rate and the settling time of the amplifier. Often these, and not the finite gain-bandwidth product, represent the ultimate limitation on the highest achievable switching rate. REFERENCES
[1] B. J. Hosticka, R. W. Brodersen, and P. R. Gray, MOS sampled

where the usual condition Cl = C2 = 2Ca was assumed [7] and where @~ (1 + k) exp (-lctio T/2). The circuit of Fig. 3 simulates an inductor only at signal frequencies a satisfying UT<< 1. Hence, in (27) the substitution z-l = e-jwTs 1 - j~l may be made. Carrying out this
substitution, and assuming also that 8, e, c#, and UT are all

data recursive filters using state variable techniques: in Proc.


ISCAS, Apr. 1977, pp. 525-529. [2] , MOS sampled data recursive filters using switched capacitor integrators IEEE J. Solid-State Circuits, vol. SC-12, pp. 600608, Dec. 1977. [3J D. J. Allstot, R. W. Brodersen, and P. R. Gray, Fully integrated, high order NMOS sampled-data ladder filter: in LSSCCDig., Feb.

much smaller than unity, the first-order approximation Q(z) VI (z)

= ()
1-8 l-e

2C1
juT

(28)

1978, Pp. 82-83. [4] G. M. Jacobs, D. J. Allstot, R. W. Brodersen, and P. R. Gray,


Design considerations for MOS switched capacitor ladder filters, in Proc. ISCAS, May 1978, pp. 324-329. [5] , Design techniques for MOS switched capacitor ladder filters; IEEE Tram. Circuits Syst., vol. CAS-25, pp. 1014-1020, Dec. 1978.

results. Hence, only the value of the equivalent inductor changes by a factor (1 - e)/(1 - 6). The Q is not affected, and no other parasitic elements appear.

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-15, NO. 3, JUNE 1980


[6] R. W. Brodersen, P. R. Gray, and D. A. Hodges, MOS switched capacitor filters; Proc. IEEE, vol. 67, pp. 61-75, Jan. 1979. [7] B. J. Hosticka and G. S. Moschytz, Switched-capacitor simulation of grounded inductors and gyratorsj Electron. Lett., vol. 14, pp. 788-790, NOV.23, 1978. [8] G. C. Temes and M, Jahanbegloo, Switched-capacitor circuits which are bilinearly equivalent to a floating inductor or FDNR, Electron. Lett., vol. 15, pp. 87-88, Feb. 1, 1979. Gabor C. Temes (SM66-F73) received the Dipl. Ing. degree from the Technical University of Budapest, Budapest, Hungary, in 1952, the Dipl. Phys. degree from Eotvos University, Budapest, Hungary, in 1954, and the Ph.D. degree in electrical engineering from the University of Ottawa, Ont., Canada, in 1961. He was a member of the faculty of the Technical University of Budapest from 1952 to 1956, and was employed by Measurement Engineering Ltd., Arnprior, Ont., Canada, from

361

1957 to 1959. From 1959 to 1964 he was with Northern Electric R & D Laboratories, Ottawa, Ont., Canada. From 1964 to 1966 he was a research group leader at Stanford University, Stanford, CA; from 1966 to 1969 he was a Corporate Consultant for Ampex Corporation, Redwood City, CA. Since 1969 he has been a Professor at the University of California, Los Angeles. Between 1975 and 1979, he was also Chairman of the Department of Electrical Sciences and Engineering. He has published about 90 technical papers. He was coeditor (with S. K. Mitra) and coauthor of Modern Filter Theory and Design (New York: Wiley, 197 3), coauthor of Introduction to Circuit Synthesis and Design (New York: McGraw-Hill, 1977), and a contributor to several other edited volumes. Dr. Temes is an Associate Editor of the Journal of the Franklin Zn$titute, a former Editor of the IEEE TRANSACTIONS ONCIRCUITTHEORY, a former Vice President of the IEEE Ckcuits and Systems Society, and a former Chairman of its Best Paper Awards and Nominations Committees. He was a member of the Fellow Award Committee as well as the CANDE Committee of the IEEE Ckcuits and Systems Society. He was a cowinner (with H. J. Orchard) of the 1968 Darlington Award of the IEEE Ckcuits and Systems Society.

A Phototransistor

Optical Isolator Noise Model

KOY B. COOK, JR., SENIOR MEMBER, IEEE, WEN-CHIN YEH, AND VICTOR W. RUWE, SENIOR MEMBER, IEEE

AbstractA noise model for the phototransistor opticsl isolator is presented and used to predict optical isolator noise performance. Results are demonstrated to agree with existing experimental data on phototransistor opticsl isolator noise. The model presented includes burst noise, flicker noise, and shot noise.

1. lNTRODLJCTION

HE results of experimental investigations of the noise performance of several commercially available phototransistor optical isolators have been presented in prior publications [1] , [2] . It is the purpose of this paper to present a circuit model which will accurately predict phototransistor optical isolator noise performance and which will thus be useful for computer simulations for low noise circuit design. An optical isolator is composed of a phototransistor or photodiode detector optically coupled to a light-emitting diode (LED) as shown in Fig. 1. The LED not only provides the bias current to establish the transistor Q-point, but it also provides the input signal to the phototransistor. The equations which describe the operation of the optical isolator are also shown in Fig. 1 where
ID = dc LED bias current, Ic = dc phototransistor collector current, IPH = dc photocurrent

1PH

= kOID

lPH AIC

= k(AID) = hfek(AIO)

lC = FEkOTD

Fig. 1. Optical isolator schematic and describing equations.

k. = static current transfer ratio, k = dynamic current transfer ratio. The static and dynamic current transfer ratios lco and k, respectively, are, in general, dependent on LED current [2]. II. THE MODEL A noise model for the phototransistor optical isolator is presented in Fig. 2. The right-half is the electrical noise model of the phototransistor under ideal LED illumination (one which exhibits only shot noise in its optical fluctuations) and the lefthrdf is the electrical noise model for the LED. The hybrid-n transistor equivalent circuit has been used and rm, Cm, r., Cp, and gm have the conventional hybrid-n definitions. In the phototransistor, ~& and ~~ are shot noise current generators [3], The contribution of flicker noise in the phototransistor is represented by two flicker noise current generators ~~1 and $00.75 @ 1980 IEEE

in the phototransistor absorption from the LED,

due to photon

Manuscript received May 23, 1979; revised February 8, 1980. K. B. Cook, Jr., and W.-C. Yeh are with the Department of Electrical Engineering, Auburn University, Auburn, AL 36830. V. W. Ruwe is with the U.S. Army MICOM, Redstone Arsenal, Huntsville, AL 35809.

0018-9200/80/0600-0361

You might also like