IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 45, NO.

3, MARCH 1998 295

Noise in Gm-C Filters

Giorgos Efthivoulidis, László Tóth, and Yannis P. Tsividis, Fellow, IEEE

Abstract— Noise in Gm-C filters is discussed. Topology-

independent bounds for the signal-to-noise ratio (SNR) are
derived, assuming the total power dissipation or the total
capacitance are specified. Expressions for certain products of
noise with frequently used quantities are derived, and are shown
to be invariant under certain actions, which are sometimes taken
in order to improve the SNR. Finally, specification-dependent
bounds for SNR are derived.
(a)
I. INTRODUCTION

I N RECENT YEARS, there has been a great interest in low-

power low-voltage integrated circuits for portable telecom-
munications, computer, and consumer products [1]. The low-
power requirement makes necessary high impedance levels;
this, in conjunction with chip area restrictions, limits the
allowable total capacitance value and results in high noise
in integrated filters [2]–[5]. Coupled with the small signal
swings in low-voltage circuits, this problem results in limited (b)
signal-to-noise ratio (SNR). The problem is compounded by Fig. 1. A transconductor in a Gm-C filter. (a) Definition of Hi` (f ): (b)
the fact that high- active filters are inherently more noisy Definition of H`o (f ):
in comparison to passive ones. For example, the output noise
mean squared value in active resonators is at least times The derivations in this paper benefit from related approaches
that of resonators with the same transfer function and used in a previously published work on noise in active
total capacitance [6]. Thus, in contrast to what is observed (and in the related MOSFET-C) filters [27]. In general, we
in passive filters, in active filters the SNR decreases as the have attempted to maintain parallels between the development
specifications for the transfer function become more tight. in the two papers, so that we can make evident what is similar
The most popular integrated filter types are based on the between the two classes of filters, in terms of noise perfor-
transconductor-C, or “Gm-C” technique [7]–[9]. These include mance. We have opted, however, for a change in terminology
the most common implementations of gyrator-C filters. Many as it regards the term “dynamic range,” which is meant to be
integrated implementations of Gm-C filters, along with de- the ratio of the maximum signal to the minimum acceptable
scriptions of their mass-volume applications, are discussed in signal. This term is used with more than one meaning in the lit-
the papers included in [10]. The noise in such filters suffers erature. In most of the linear filter literature, it means the same
from all of the limitations discussed above, and has been the as “maximum SNR,” since the minimum acceptable signal is
subject of many papers [6], [9], [11]–[26]. In this paper, we assumed to be equal to the noise. This definition for the min-
contribute to the growing body of literature on linear Gm-C imum level is, however, inappropriate for some applications.
filter noise with several new results. In Section II, a topology- Also, the maximum acceptable signal is sometimes assumed
independent bound for SNR in Gm-C filters with specified to have a flat spectrum, whereas in other settings it is defined
power dissipation is developed. In Section III, the effects of in terms of a sinusoidal signal. To avoid any ambiguity, all
preamplification, parallel processing, and impedance scaling discussions in this paper will be in terms of SNR and its max-
are examined, and some products of noise with frequently imum value, with the signal always assumed to be a sinusoid.
specified quantities are shown to be invariant under these
actions. A bound for SNR in Gm-C filters with specified II. BOUND FOR SNR IN Gm-C FILTERS
total capacitance is developed in Section IV. Specification- WITH LIMITED POWER DISSIPATION
dependent bounds are given in Section V. In this section, we give expressions for the power dissipa-
Manuscript received June 24, 1996; revised November 5, 1996. This paper tion, the output noise, and the SNR in Gm-C filters, and we
was recommended by Associate Editor N. M. Nguyen. derive an upper bound for the SNR. The approach used is
G. Efthivoulidis is with the Department of Electrical and Computer Engi- similar to that in [27].
neering, National Technical University of Athens, Zographou 15773, Athens,
Greece. Fig. 1(a) shows a Gm-C filter with transfer function
L. Tóth is with the Department of Electromagnetic Theory, Technical A transconductor with transconductance is shown separated
University of Budapest, 1111 Budapest XI, Hungary. from the rest of the filter. All signals in Fig. 1 are assumed
Y. P. Tsividis is with the Department of Electrical Engineering, Columbia
University, New York, NY 10027 USA. to be sinusoidal with frequency and they are represented
Publisher Item Identifier S 1057-7130(98)00768-X. by their phasors (using peak values for the magnitudes of the
1057–7130/98$10.00  1998 IEEE

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 296 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 45, NO. 3, MARCH 1998

A practical transconductor (e.g., operating in class A) may

dissipate several times more power than the above processing
power.

B. Output Noise
The noise due to the transconductor is modeled by a
current source connected at the output of the transconduc-
tor, as shown in Fig. 1(b), with (one-sided) spectral density
where is the Boltzmann constant, is the absolute
(a)
temperature of the circuit, and is the noise factor of the
transconductors [6], which typically has value larger than 1
(e.g., 2 or 3).1 The spectral density of the output noise due to
the transconductors is
(2)

The output noise mean-squared value is

(3)

In cases where the above integral is infinite, the region of

(b)
integration can be reduced to a finite region, which includes
Fig. 2. (a) Currents supplied by voltage sources to transconductor G` : (b) the passband and transition bands of the filter.
Plot of voltage vi` (t) and currents i`+ (t) and i`0 (t):

C. SNR
latter). We denote by the transfer function from to
the input of Let a current source be inserted, for later The SNR is defined as the ratio of the mean squared value
use, at the output of as shown in Fig. 1(b). We denote by of the output signal, to the mean-squared value of the noise:
the transfer function from to the output of the filter.
The transfer functions and are used below, (4)
in the expressions for power dissipation and noise spectral We assume that the noise is independent of the signal.
density, respectively. Therefore, the SNR is maximum when the input signal is
maximum.
A. Power Dissipation The input voltage of transconductors must be limited in
The current from the output of transconductor in order for the filter to operate in the linear region. Let be
Fig. 1(a) is the maximum allowed peak differential voltage at the input of
the transconductors. Usually is some fraction of i.e.,
(5)
where is the phase of In order to deal with
where is a number smaller than 2 (e.g., Then the
the minimum possible power, we assume first that all power
linearity constraint is expressed as
dissipated in the transconductors is due to their output stages,
and that the latter dissipate power only when a signal is present (6)
(class B operation). Fig. 2(a) shows the transconductor and In practice, it is common that a single maximum input ampli-
the supply voltages While the output current is positive, tude is specified, which will guarantee linearity for any input
it is pushed by the voltage source When it becomes frequency. Therefore, the input signal amplitude must satisfy
negative, it is pulled by the voltage source The positive the inequality
and negative parts of denoted by and
(7)
respectively, are shown in Fig. 2(b).
The instantaneous power supplied by is where
The peak value of is and its average
value is the peak value divided by Thus the average power (8)
supplied by to the circuit through transconductor is
The same average power is supplied is the maximum amplification from the input of the filter to
by the voltage source Therefore, in the presence of the inputs of the transconductors.
signal, the total “processing” power, which is dissipated in Another limitation for especially for low-power inte-
the transconductors since capacitors do not dissipate power, is grated filters, is imposed by the requirement that the power
1 This factor is responsible for the fact that Gm-C filters are more noisy
(1) than equivalent MOSFET-C filters with the same total capacitance or power
dissipation [20], [21].

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 EFTHIVOULIDIS et al.: NOISE IN GM-C FILTERS 297

dissipation must be limited. Let be the maximum accept-

able total power dissipation. In order for this limit not to be
exceeded, the processing power in (1) must remain well below
it, to allow for extra power dissipated in the transconductor
circuits, especially if the latter operate in class A. Thus, the
processing power cannot be more than a fraction of
where has a value smaller than unity (e.g.,
We should emphasize that should not be confused with the
“efficiency factor,” which is commonly defined for the case of Fig. 3. A Gm-C filter example.
resistive loads. The following inequality must then be satisfied:
for all (9) of and all we have
Since is proportional to by (1), the above equation
gives an upper bound for We will assume that is
restricted to be below the smaller of the two bounds imposed or, using the value for given above,
by (1) and (9), and by (7).
If the linearity constraint restricts power dissipation to a
value lower than the acceptable, impedance scaling can be
used (if chip area is not already too large) to increase the SNR
Rearranging terms, this gives
of the filter, up to the point that the power dissipation has its
maximum allowed value.
The maximum SNR (over allowed input signal amplitude
and frequency) is
and taking the limit of both sides as (11) follows.

(10) E. Performance Bounds

Applying the Cauchy–Schwartz inequality [30] on vectors
where is the maximum value of that guarantees
and and using (11), we obtain
both linearity and power dissipation constraints.

D. Relationship Between and

We will now show that transfer functions (12)
and satisfy
From (1) and (8), we get
(11)

Proof: Although the above result could be derived from

general considerations involving adjoint networks [28], [29], (13)
a more straightforward approach will be taken here. Fig. 1(a)
shows a circuit with transfer function i.e., when the
From (2), (12), and (13), it follows that
input signal is the output signal is
with denoting the real
part of the complex number We change the frequency from
to while keeping the same, and we add for each (14)
transconductor a current source as shown in Fig. 1(b), The right-hand side (RHS) of (14) is a topology-independent
with frequency We will seek phasors such that the lower bound for the product A lower bound
phasors of the voltages at all nodes remain the same, as they for the product can be derived by eliminating
were at frequency Indeed, assume that this is possible. Since in (14), using (5) and (7). This gives
all voltages remain the same, the transconductor output current
phasors are the same, while the current phasors through the (15)
capacitors are multiplied by and our
assumption will be valid if just the extra current needed is Notice that as given by (1), is proportional to as is
supplied by Therefore, with the right-hand side of (14). On the other hand, the RHS of (15)
is proportional to Therefore, for small values of , the
RHS of (15) is not a tight bound for However,
when we are considering the maximum SNR of the filter,
the phasor of the output voltage remains Using we assume that the input signal has the maximum allowed
superposition to express the output as a result of the application amplitude.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 298 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 45, NO. 3, MARCH 1998

Multiplying both sides by and using

and (4), we conclude that

(22)
When and are specified, the above inequality
gives an upper bound for that cannot be exceeded for
any Gm-C filter design. When and are specified,
(22) gives a lower bound for the power dissipation. When
all quantities in (22) are specified and inequality (22) doesn’t
HP f S f
Fig. 4. Plot of (a) i G ( ) G ( ) for the circuit in Fig. 3 and (b) of the hold, then there is no hope to design a Gm-C filter achieving
RHS of (14), normalized with the maximum value of the latter. the given specifications.
In [27], an upper bound for SNR in active filters, similar
to that in (22), is given. It is noted that the two bounds cannot
Example 1: Consider the second-order Gm-C filter shown
be directly compared to derive conclusions for the whole
in Fig. 3 [31]. The transfer function of this filter is
classes of active and Gm-C filters, since the power dissi-
pation in the result in [27] is the power dissipated only in the
(16) resistors and not the total power dissipation. Moreover, the two
inequalities give bounds for and not actual values for it.
with
F. Special Cases
(17) We consider now filters derived from a low-pass prototype.
We denote by the low-pass prototype transfer function,
Assume that the following design parameters are specified: with a cutoff normalized frequency of 1.
For low-pass filters with bandwidth i.e., with
(18) it is
(19) (23)

and let and the bound on the RHS of (22) is inversely proportional
We select pF, so that the total capacitance of the to
filter is pF. From (17) and (18), it follows that For high-pass filters with cutoff frequency i.e., with
it is
It can be seen that We use the maximum allowed
(24)
value as amplitude of the input signal, i.e., The
functions and can be calculated from (1) and and the bound on the RHS of (22) is inversely proportional
(2). The product is plotted in Fig. 4, together to
with the RHS of (14), normalized with the maximum value For bandpass filters with center frequency and quality
of the latter. factor i.e., with it is [27]
The root-mean-square (rms) value of the output noise can
be calculated from (3) and is found to be mV.
The SNR, given by (4), is dB.
With the values given above, the maximum power dissi-
pation is found to be mW. If the maximum
power dissipation is specified, impedance scaling can be used
(25)
to increase or decrease it.
Equation (15) is used now to derive a bound for and the bound on the RHS of (22) is inversely proportional
From (9) and (15) it follows that to and

(20) G. Second-Order Bandpass Case

For the second-order bandpass transfer function, given by
Integrating both sides of (20) over and using (3), we obtain (16), the low-pass prototype transfer function is

(21) (26)

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 EFTHIVOULIDIS et al.: NOISE IN GM-C FILTERS 299

and TABLE I
EFFECTS OF PREAMPLIFICATION, PARALLEL
(27) PROCESSING, AND IMPEDANCE SCALING

From (16) or (26), it follows that

(28)
Using (25), (27), and (28), then (22) becomes

, and are independent of the parameters

(29) These products may be useful in deriving bounds for be-
cause they are the most likely to have a topology-independent
The RHS of this relation has the same functional form as lower bound, from which a topology-independent bound for
results derived for the dynamic range elsewhere [20], [26]. can be derived. In contrast to this, it is meaningless,
Example 2: With the parameters given in Example 1, the for example, to try to find a bound for the product
bound on the RHS of (29) is 67.9 dB. without any assumption for the maximum signal swings, since
preamplification can be used to make this product arbitrary
III. NOISE PRODUCT INVARIANTS small, as can be deduced from Table I.
Equation (14) gives a topology-independent bound for In the next section, we derive a topology-independent bound
the product in Gm-C filters. In [27] a topology- for the product and a corresponding bound for
independent bound for the product is given, where in state-space Gm-C filters with limited total capacitance. A
and are the power dissipation and output noise spectral bound for the product and a corresponding bound
density, respectively, due to resistors, in active filters. In for in state-space active filters with limited total
this section, the above products are shown to be invariant under capacitance, are given in [32].
the following actions: preamplification, parallel processing,
and impedance scaling. IV. BOUND FOR SNR IN STATE-SPACE GM-C
FILTERS WITH LIMITED TOTAL CAPACITANCE
A. Preamplification Fig. 5 shows a summing integrator in a state-space Gm-C
We multiply the input signal by a factor pass it through filter. The whole filter consists of such integrators. Index in
the filter, and multiply the output of the filter by the factor Fig. 5 runs over The input of the filter (which
The elements performing these functions (e.g., an amplifier and we will also denote by for convenience) is connected to
an attenuator) are assumed to have negligible noise. The output the th integrator through the transconductor The output
signal remains the same, while the output noise is divided by of th integrator is connected to the th integrator through
The internal voltages and currents are multiplied by the transconductor All transconductors have one of their
Therefore, the power dissipation in the transconductors and inputs grounded. Only their nongrounded input is shown in
resistors is multiplied by and respectively. Fig. 5. The sign of the transconductance is or whenever
the shown input is connected to the noninverting or inverting
B. Parallel Processing input of the transconductor, respectively. Capacitor inte-
grates the sum of the output currents of the transconductors.
We pass the input signal through identical filters and
The current source is included for later use.
we take the average of their outputs. The output signal
The circuit is described by the following state-space equa-
remains the same, while the output noise is divided by
tions:
The total capacitance and power dissipation in the resistors
and transconductors are multiplied be (30)

C. Impedance Scaling or, in the frequency domain,

We multiply all capacitances and (trans)conductances by (31)
The internal voltages remain the same, while the internal
currents are multiplied by The output voltage signal remains Index in the above equations runs over
the same. The total capacitance is multiplied by The power The output of the filter is taken from the output of one
dissipation is also multiplied by The output noise is divided of the integrators. We denote by the transfer function
by since the spectral densities of all equivalent noise voltage from the input of the filter to the output of the th summing
sources are divided by and all voltage transfer functions integrator. We also denote by the transfer function
remain the same. The effects are the same as for parallel from the current source shown in Fig. 5 to the output of the
processing, but is not restricted to integer values and can filter. Notice that in this section, we use the same symbols as in
be less than Section II for different quantities. All symbols that are redefined
Table I summarizes the effects of the above three actions. in this section should not be confused with the corresponding
From the table it is seen that the products , , symbols in Section II.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 300 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 45, NO. 3, MARCH 1998

Rearranging terms and taking the limit as (36)

follows. As was the case with (11), an alternative proof could
be given using adjoint networks [28], [29].
Applying the Cauchy–Schwartz inequality [30] on vectors
and and using (36), we
obtain

(37)
Fig. 5. A summing integrator in a state-space Gm-C filter.
Using (31), (34), (35), and the inequality ,
The total capacitance of the filter is
we obtain the following inequality:
(32)

The spectral density of the output noise due to the transcon-

ductors is (38)
(33) From (32), (35), and (38), we get

where
(34) (39)

Index in (33) runs over while index in (34) From (37), (39), and (33), it follows that
runs over The output noise mean-squared value,
SNR, and maximum SNR are defined as in (3), (4), and (10),
respectively, where is the maximum value of that
(40)
guarantees linearity. Denoting by the maximum allowed
voltage swings at all internal nodes, the linearity constraint is Integrating both sides of (40) over and using (3), we obtain
expressed by (6), or by (7) with
(35) (41)

Index in (35) runs over and by Multiplying both sides by and using (4), (8), and
convention. we conclude that
We will now prove that transfer functions ,
and satisfy
(36)

Proof: We change the frequency of the input signal from (42)

to while keeping the same, and we add for
The above inequality also gives a lower bound for the total
each capacitor a current source as shown in Fig. 5,
capacitance when is specified, or it can be used in order
with frequency We will seek phasors such that the
to rule out combinations of specifications that do not satisfy
phasors remain the same, as they where at frequency the inequality.
Indeed, assume that this is possible. Then the current phasor
The assumption that one input of the transconductors must
through the capacitor will be increased by and
be grounded, made at the beginning of this section, is not
the assumption will be valid if this extra current is supplied
necessary. If we denote by the maximum allowed voltage
by Therefore, with swings at all nodes (with respect to the ground) and all
differential inputs of transconductors, then it can be easily
the phasor of the output voltage remains Using shown that (42) still holds.
superposition to express the output as a result of the application
of and all we have A. Special Cases
We consider again filters derived from a low-pass prototype.
For low-pass filters with bandwidth i.e., with
or, using the value for given above, it is

(43)

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 EFTHIVOULIDIS et al.: NOISE IN GM-C FILTERS 301

illustrate this more clearly, one can use lower bounds for these
integrals, dependent only on the specifications for the transfer
function magnitude, Using these bounds,
specification-dependent bounds for can be derived from
(22) and (42).
Lower bounds for the integral in (22) were derived in [27]
for low-pass and bandpass filters. In a similar manner, it can
be shown that for high-pass filters, with constrained by
if (47a)
if (47b)
we have
Fig. 6. Plot of (a) Ct Hi SG (f ) for the circuit in Fig. 3 and (b) of the RHS
2
of (40), normalized with the maximum value of the latter. (48)

Bounds for the integral in (42) can be derived in a similar

For high-pass filters with cutoff frequency i.e., with
manner as in [27]. The results are as follows.
it is
For low-pass filters, with constrained by
(44) if (49a)
if (49b)
For bandpass filters with center frequency and quality
factor i.e., with it is we have

(50)

(45) For high-pass filters, with constrained by

if (51a)
B. Second-Order Bandpass Case if (51b)
Using (27), (28), and (45), (42) then becomes
we have

(52)
(46)
For bandpass filters, with constrained by
in which the RHS has the same functional form as a result
derived for dynamic range elsewhere [20], [22]. if (53a)
Example 3: The two sides of (40), for the circuit in Fig. 3, if (53b)
are plotted in Fig. 6. With the parameters given in Example 1,
the bound on the RHS of (46) is 67.8 dB. we have
Notice that it has not been shown whether the bounds in (42)
and (46) can be achieved or how tight they are. In [19], the
optimum value for in state-space filters with limited total
capacitance is given, with a different definition for (54)
norms are used). That result has a functional form similar to
the right-hand side of (42). The bound in (42) is expressed In previous sections, the dependence of the bounds for
in terms of the transfer function, while the optimum value in on the quality factor was shown. The bounds in this section
[19] is in terms of state-space parameters. Bounds expressed in also show the dependence on the transition region.
terms of magnitude specifications are given in the next section.
VI. CONCLUSION
V. SPECIFICATION-DEPENDENT BOUNDS General, topology-independent bounds for the SNR in Gm-
The bounds for in (22) and (42) are products of two C filters, for the cases of given maximum power dissipation
terms. The first term consists of technological parameters, and of given maximum total capacitance, have been derived.
such as power dissipation, total capacitance, etc., while the It was shown that these can be converted to specification-
second depends only on the transfer function implemented. dependent bounds in several special cases. Although the
This latter term expresses, in some way, how tight the transfer bounds derived do not give the optimum values for SNR,
function is. That is, the integrals in (22) and (42) increase they do show some of the limitations of linear Gm-C fil-
as the requirements become more demanding. In order to ters. The results of preamplification, parallel processing, and

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.
 302 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 45, NO. 3, MARCH 1998

impedance scaling have been examined, and some perfor- [25] P. Shah, “Design of analogue integrated circuits for very low frequency
mance parameters which are invariant under such actions have signal processing,” Ph.D. thesis, Technical Univ. Denmark, 1993.
[26] W. B. Kuhn, F. W. Stephenson, and A. Elshabini-Riad, “Dynamic range
been identified. of high-Q OTA-C and enhanced-Q LC RF bandpass filters,” in Proc.
Midwest Symp. Circuit Theory, 1995.
[27] L. Tóth, G. Efthivoulidis, V. Gopinathan, and Y. P. Tsividis, “General
ACKNOWLEDGMENT results for resistive noise in active RC and MOSFET-C filters,” IEEE
Trans. Circuits Syst. II, vol. 42, pp. 785–793, Dec. 1995.
The authors would like to thank Dr. V. Gopinathan and Dr. [28] S. W. Director and R. A. Rohrer, “Automated network design: The
G. C. Temes for useful discussions. frequency domain case,” IEEE Trans. Circuit Theory, pp. 330–337, Aug.
1969.
[29] G. C. Temes and J. W. LaPatra, Circuit Synthesis and Design. New
REFERENCES York: McGraw-Hill, 1977.
[30] A. Papoulis, Signal Analysis. New York: McGraw-Hill, 1977.
[1] R. J. van de Plassche, W. M. C. Sansen, and J. H. Huijsing, Analog [31] E. Sánchez-Sinencio, R. L. Geiger, and H. Nevarez-Lozano, “Generation
Circuit Design: Low-Power, Low-Voltage, Integrated Filters and Smart of continuous-time two-integrator loop OTA structures,” IEEE Trans.
Power, Norwell, MA: Kluwer, 1995. Circuits Syst., vol. 35, pp. 936–945, 1988.
[2] B. J. Hosticka, “Performance comparison of analog and digital circuits,” [32] G. Efthivoulidis, L. Tóth, and Y. P. Tsividis, “Further results for noise
Proc. IEEE, vol. 73, pp. 25–29, Jan. 1985. in active RC and MOSFET-C filters,” to be published.
[3] E. A. Vittoz, “Future of analog in the VLSI environment,” in Proc.
IEEE/ISCAS, 1990, pp. 1372–1375.
[4] J. O. Voorman, “Continuous-time analog integrated filters,” in Integrated
Continuous-Time Filters, Y. Tsividis and J. O. Voorman, Eds. New
York: IEEE, 1993.
[5] E. A. Vittoz, “Low-power low-voltage limitations and prospects in Giorgos Efthivoulidis was born in Greece in 1970.
analog design,” in Analog Circuit Design: Low-Power Low-Voltage, He received the Diploma in electrical and computer
Integrated Filters and Smart Power, R. J. van de Plassche, W. M. C. engineering from the National Technical University
Sansen, and J. H. Huijsing, Eds. Norwell, MA: Kluwer, 1995. of Athens, Athens, Greece, in 1993.
[6] D. Blom and J. O. Voorman, “Noise and dissipation of electronic During 1990–1992, he worked on parallel compu-
gyrators,” Philips Res. Rep., vol. 26, pp. 103–113, 1971. tation systems. In 1993, he started his Ph.D. study at
[7] K. W. Moulding and G. A. Wilson, “A fully integrated five-gyrator the National Technical University of Athens, Divi-
filter at video frequencies,” IEEE J. Solid-State Circuits, vol. SC-13, pp. sion of Computer Science, on noise in continuous-
303–307, June 1978. time filters.
[8] K. W. Moulding et al., “Gyrator video filter IC with automatic tuning,”
IEEE J. Solid-State Circuits, vol. SC-15, pp. 963–968, Dec. 1980.
[9] H. Khorramabadi and P. R. Gray, “High-frequency CMOS continuous-
time filters,” IEEE J. Solid-State Circuits, vol. SC-19, pp. 939–948, Dec.
1984.
[10] Y. P. Tsividis and J. O. Voorman, Integrated Continuous-Time Filters.
New York: IEEE, 1993. László Tóth was born in Budapest, Hungary on
[11] J. O. Voorman and D. Blom, “Noise in gyrator-capacitor filters,” Philips March 25, 1957. He received the M.Sc. and Can-
Res. Rep., vol. 26, pp. 114–133, 1971. didate’s (Ph.D.) degrees in electrical engineering
[12] A. Fettweis, “On noise performance of capacitor-gyrator filters,” Int. J. from the Technical University of Budapest and the
Circuit Theory Appl., vol. 2, pp. 181–186, 1974. Hungarian Academy of Sciences, in 1982 and 1987,
[13] C. S. Park and R. Schaumann, “Design of a 4-MHz analog integrated respectively.
CMOS transconductance-C bandpass filter,” IEEE J. Solid-State Cir- During 1982–1993, he was with the Research In-
cuits, vol. 23, pp. 987–996, Aug. 1988. stitute for Telecommunications (TKI), dealing with
[14] G. Espinosa, F. Montecchi, E. Sánchez-Sinencio, and F. Maloberti, digital filtering, analysis of switched capacitor net-
“Noise performance of OTA-C filters,” in Proc. IEEE/ISCAS, 1998, vol. works and speech compression. Since 1993, he has
3, pp. 2173–2176. been with the Technical University of Budapest.
[15] A. Brambilla, G. Espinosa, F. Montecchi, and E. Sánchez-Sinencio, During 1989–1990 and 1995–1996, he was with the MAD-VLSI Laboratory
“Noise optimization in operational transconductance amplifier filters,” and Microelectronic Circuits and Systems Laboratory, respectively, Columbia
in Proc. IEEE/ISCAS, 1989, vol. 1, pp. 118–121. University, New York, NY, dealing with noise performance of active RC
[16] P. Bowron and K. A. Mezher, “Noise analysis of continuous-time active filters, switched capacitor networks, and nonlinear circuits.
filters,” in Proc. IEEE/ISCAS, 1990, vol. 2, pp. 1181–1184.
[17] Y.-T. Wang and A. A. Abidi, “CMOS active filter design at very high
frequencies,” IEEE J. Solid-State Circuits, vol. 25, pp. 1562–1574, Dec.
1990.
[18] J. van der Plas, “MOSFET-C filter with low excess noise and accurate
automatic tuning,” IEEE J. Solid-State Circuits, vol. 26, pp. 922–929, Yannis P. Tsividis (S’71–M’74–SM’75–F’86) re-
July 1991. ceived the B.S. degree from the University of Min-
[19] G. Groenewold, “The design of high dynamic range continuous-time nesota, Minneapolis, and the M.S. and Ph.D. degrees
integratable bandpass filters,” IEEE Trans. Circuits Syst., vol. 38, pp. from the University of California, Berkeley. He
838–852, Aug. 1991. is with the Department of Electrical Engineering,
[20] , “Optimal dynamic range integrated continuous-time filters,” Columbia University, New York, NY, as Professor
Ph.D. thesis, Delft Univ. Technology, Dept. Elect. Eng., Delft, The and Director of the Microelectronic Circuits and
Netherlands, 1992. Systems Laboratory. He has worked for Motorola
[21] , “Optimal dynamic range integrators,” IEEE Trans. Circuits Syst. Semiconductor and AT&T Bell Laboratories, and
I, vol. 39, pp. 614–627, Aug. 1992. has taught at the University of California, Berkeley,
[22] , “A high-dynamic-range integrated continuous-time bandpass Massachusetts Institute of Technology, Cambridge,
filter,” IEEE J. Solid-State Circuits, vol. 27, pp. 1614–1622, Nov. 1992. and the National Technical University of Athens.
[23] A. A. Abidi, “Noise in active resonators and the available dynamic Dr. Tsividis is the recipient of the 1984 IEEE Baker Best Paper Award
range,” IEEE Trans. Circuits Syst. I, vol. 39, pp. 296–299, Apr. 1992. and the 1986 European Solid-State Circuits Conference Best Paper Award.
[24] J. P. Moree, G. Groenewold, and L. A. D. van den Broeke, “A bipolar He is co-recipient of the 1987 IEEE Circuits and Systems Society Darlington
integrated continuous-time filter with optimized dynamic range,” IEEE Best Paper Award. He has received the Great Teacher Award at Columbia
J. Solid-State Circuits, vol. 28, pp. 954–961, Sept. 1993. University.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE JUIZ DE FORA. Downloaded on November 19,2024 at 18:16:28 UTC from IEEE Xplore. Restrictions apply.