328

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

High-Frequency Harmonic Distortion in Feedback

Amplifiers: Analysis and Applications
Gaetano Palumbo, Senior Member, IEEE and Salvatore Pennisi, Member, IEEE

AbstractAn approach for the evaluation of high-frequency

harmonic-distortion factors in feedback systems is proposed and
the results obtained are applied to feedback amplifiers. Under
the assumption that transistors are not driven out of their linear
operating regions, small-signal analysis and conventional algebra
are exploited to derive understandable and compact expressions
highly improving the comprehension of harmonic-distortion
generation. The impact of the frequency compensation utilized
(namely, dominant-pole or Miller technique) on linearity performance is evaluated and the high-frequency distortion properties of
closed-loop single-stage and two-stage amplifiers are analyzed and
compared. The accuracy of the analysis, also in view of the given
applications, is confirmed through extensive simulations with
Spectre on idealized models as well as on CMOS transistor-level
circuits. Despite the approximated nature of the analytical models,
predicted data are found in very close agreement with simulations
in nearly all the frequency range of interest.

(a)

Index TermsAnalog integrated circuits, feedback amplifiers,

harmonic distortion, nonlinear distortion.

(b)
Fig. 1. Nonlinear amplifier in (a) open-loop; (b) closed-loop configuration.

I. INTRODUCTION

HE effects of nonlinearities in circuits and systems are

customarily evaluated by means of the harmonic-distortion analysis [1]. Let us consider a nonlinear open-loop amin Fig. 1(a). When nonlinplifier, schematized by block
earities are weak, the amplifiers transfer characteristic exhibits
only gradual slope changes. Hence, harmonic distortion can be
easily calculated with the series expansion of the nonlinear dc
input-output characteristic, which can be usually well approximated by the first three power terms

The second and third harmonic-distortion factors versus either input or output fundamental amplitudes (remember that
) are then
(4)
(5)

are the amplifier input and output incremental

where and
variables. If we now apply a pure sinusoidal incremental input
, the output signal becomes
tone,

where the gain compression, which arises in term and is due

to coefficient , has been neglected [2]. Observe that the last
formulation of (4)(5) can be used to compare the linearity of
amplifiers having the same output fundamental amplitude.
Alternatively, one can represent the input signal by 1
and the output signal, through (1), becomes

(2)

(6)

(1)

where the terms

up to the third order are given by

(3a)
(3b)
(3c)
(3d)

Manuscript received November 21, 2001; revised September 21, 2002. This
paper was recommended by Associate Editor P. V. A. Mohan.
The authors are with the Dipartimento Elettrico Elettronico e Sistemistico (DEES), Universita di Catania, I-95125 Catania, Italy (e-mail:
gpalumbo@dees.unict.it; spennisi@dees.unict.it).
Digital Object Identifier 10.1109/TCSI.2003.808835

As we will show in the following, this last representation is

useful to simplify the characterization of nonlinear systems in
the frequency domain using conventional algebra [3].
The ability of feedback to improve the closed-loop linearity
performance is widely recognized. The classical theory of
feedback amplifiers asserts that negative feedback reduces
harmonic distortion by an amount related to the loop gain [4],
and
[5]. Indeed, considering the same nonlinear amplifier
assuming a fraction of the output signal fed back to the input
(see Fig. 1(b), the above harmonic-distortion factors referred
to the input signal amplitude are approximately reduced by
and
, respectively. This means a reduction
1The

exact representation is

1057-7122/03$17.00 2003 IEEE

x = RefX e

g.

PALUMBO AND PENNISI: HIGH-FREQUENCY HARMONIC DISTORTION IN FEEDBACK AMPLIFIERS

329

by a factor (
) of the harmonic-distortion terms referred
to the output signal magnitude. A more accurate analysis yields
the following result [6]:
(7)

(8)
Fig. 2. Amplifier with feedback in the frequency domain.

showing that even if coefficient is negligible, third-order harmonic distortion is still generated via coefficient . We can take
on
by considering an equivinto account the effect of
alent coefficient
(9)
This theory has been extended by the authors to the case
where also a nonlinear feedback network is employed [7]. However, in all these developments, both the amplifier and the feedback network were assumed to be frequency independent. This
hypothesis is clearly only a simplification. Indeed, a real amplifier must be frequency compensated to ensure closed-loop
stability, while the feedback network can include reactive (usually capacitive) components. Therefore, the previous expressions can be used with reasonable accuracy only under the hypothesis of low-frequency input signals.
The exact evaluation of harmonic distortion of a dynamic
system requires complex calculation involving Volterra series
or even Wiener series [8], [9]. An effort to perform the analysis
from a circuit-level point of view was given by Sansen et al.
[10][12], that however requires a huge algebraic manipulation.
In this paper, a simplified approach for the evaluation of the
second and third harmonic-distortion factors as a function of
frequency, for a feedback circuit involving nonlinear amplifiers
(and linear feedback), is carried out in Section II. The analysis
provides simple closed-form equations that are a direct extension of (7) and (8). These results are applied to single-stage
and two-stage amplifiers in Sections III and IV, respectively.
Moreover, they are used to evaluate the impact on linearity
of the kind of frequency compensation utilized. Section V
reports some validation results and Section VI summarizes the
authors conclusion.
II. PROPOSED ANALYSIS
The assumption of low-distortion conditions means, in practice, that the amplifier output is not saturated and transistors do
not leave their active region of operation. In addition, let us assume that no slew-rate limitations occur. Otherwise, the output
would not be able to adequately follow the input signal resulting
in additional high-frequency distortion. Under these conditions
and as already mentioned, we can use linearization around the
dc operating point to achieve reasonably accurate results.
Consider again the nonlinear closed-loop amplifier in
and the feedback factor are now frequency
Fig. 1(b) where
be characterized by the
dependent. Specifically, let block

frequency-dependent coefficients
,
and
and denote as
the transfer function of linear feedback
block, as illustrated in Fig. 2.
. Due
Apply now a sinusoidal input tone
to the nonlinear block in the direct path the output signal will
include harmonic components that can be described in terms
of the source signal through closed-loop nonlinear coefficients,
,
and

(10)
To derive the above closed-loop coefficients, we will follow first
an intuitive approach that is explained in the next subsection. To
our opinion, compared to the analytical approach which is reported in Section II-B, the intuitive one gives additional details
for the comprehension of the high-frequency nonlinear generation mechanisms. A further generalization of the analysis and
results is carried out in Section II-C.
A. Intuitive Approach
is responsible for the linear closed-loop
Coefficient
behavior and is customarily evaluated. It equals the for, divided by 1 plus the
ward-path transfer function,
loop-gain transfer function
(11)
and
The above equation implies computation of
at the frequency of the input tone (i.e., the fundamental
frequency).
we have to
To evaluate the second-order coefficient
follow a simple, but not trivial, reasoning. The second harmonic
component at the output is produced by the nonlinear block
when a signal at the fundamental frequency is presented to its
input. Now observe that the second harmonics is proportional to
. If the circuit is perfectly linear (i.e.,
),
in module would be equal to
. Therefore, the
nonlinear block produces a second harmonic component with
. This
amplitude equal to
can be viewed as a spurious signal injected at the output of the
nonlinear block, as depicted in Fig. 3, in which the phase variation is neglected for simplicity. The signal is subsequently processed by the feedback loop and appears at the output terminal

330

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

where only dominant terms were considered. After equating the

terms with the same exponential factor in (15) and (17), we get
(18a)
(18b)

(18c)
,
, and
Solving the above system for
exactly yields (11) and (12), while (13) slightly modifies to

Fig. 3. Equivalent representation to evaluate closed-loop second harmonic

distortion in the frequency domain.

decreased by the loop gain but evaluated at the frequency of the

harmonic considered, i.e., 2 . From the above discussion we
get
(19)

(12)
A similar procedure can be applied to derive coefficient
. Neglecting the contribute of
we get

.
where the approximation presumes
Using the definition of the harmonic-distortion factors given
in (4)(5) we get

(13)
could be included approximately by substiThe effect of
with the equivalent coefficient
defined
tuting
in (9).

(20)

B. Symbolic Approach
With the representation given in (10), we have separate information regarding the module and the phase of the coeffi. A compact mathematical representation can be
cients
obtained by considering the following input signal [3]
(21)

(14)
and the output signal will be of the form
(15)
The error signal, , is the difference of the source signal and
the output signal times the value of the feedback factor evaluated
at the appropriate frequency
(16)
then, it is processed by the nonlinear block whose output signal
is

(17)

These results are coherent with (7)(8), valid in the case of frequency-independent loop gain or, that is the same for low-frequency input signals (more precisely, for frequencies lower than
the dominant pole of the loop gain). In the present case, distortion of a feedback network in terms of the output signal is reduced of a quantity still equal to the return ratio but evaluated at
the considered harmonic frequency.
C. Further Generalization
It is useful for our purposes to extend the above results to a
more general model in which we put the nonlinear block between two linear blocks in the forward path, as shown in Fig. 4.
To obtain distortion factors of the system in Fig. 4, we can
follow the same intuitive procedure described in Section II-A.
to
Let us first evaluate the nonlinear coefficients that relate
. Being
, the firstorder coefficient is
(22)

PALUMBO AND PENNISI: HIGH-FREQUENCY HARMONIC DISTORTION IN FEEDBACK AMPLIFIERS

331

Fig. 4. General model of closed-loop nonlinear amplifier for evaluation of harmonic distortion in the frequency domain.

Fig. 5.

Equivalent representation to evaluate the second harmonic distortion for system in Fig. 4.

To obtain the second-order coefficient, it is convenient to refer

to Fig. 5, which shows the second-order component injected at
the output of the nonlinear block. Hence, we get
(23)
A similar procedure can be applied to the third-order coefficient, yielding
Fig. 6. Small-signal model of a single-stage amplifier.

(24)
Hence, the harmonic-distortion factors are expressed by

(25)

(26)
The obtained results can be usefully applied to investigate
the high-frequency distortion properties of closed-loop singlestage and two-stage amplifiers and to highlight the effects of the
chosen frequency compensation technique on linearity.

III. SINGLE-STAGE AMPLIFIERS

We shall first study the harmonic distortion of single-stage
amplifiers. These architectures are frequently employed in IC
applications for their high-frequency performance. Indeed, a
single-stage amplifier exhibits only an output high-resistance
node. Moreover, when a high gain is required, this output node
often exploits cascoding, allowing a voltage gain similar to that
of two-stage amplifiers to be achieved. Of course, these amplifiers are used in closed-loop configurations and, due to the internal structure, output dominant-pole compensation is invariably utilized.
The small-signal model of a (open-loop) single-stage ampliis the input transconfier is illustrated in Fig. 6, in which
ductance, is the output resistance and is the load capacitor
also providing compensation.
Generally, there are two sources of harmonic distortion in
such amplifiers: the nonlinear VI conversion accomplished by
the input transconductance and the nonlinear IV conversion of
the output resistance.
Let us first analyze the effect on linearity of the nonlinear
output resistance. To this end, we express the input signal as in

332

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

(12). Moreover, for our purposes, it is better to characterize the

nonlinear resistor in terms of its (nonlinear) conductance
(27)
and
are nonlinear coefficients normalwhere
. The addiized to the linear part of the output conductance
tional subscript out was used to avoid confusion with the nonlinear coefficients of the input transconductance that will be
and
used herein after. Nonlinear coefficients
cause harmonic-distortion components to appear in the output
voltage which, taking only the first three terms, is expressed according to

Fig. 7. Model of a single-stage amplifier with a nonlinear output resistance.

The amplifier is in closed loop with a feedback factor f .

(28)

(37)

(29)

This scheme is equivalent to the one analyzed in Fig. 4 by

properly updating the block transfer functions. After applying
(25) and (26) we get the following equations in which
and
(expressions in terms of
only are reported for compactness)

Then, the current through the capacitor is

Using (29) and the current through the nonlinear resistor

found by substituting (28) in (27) in the KCL at the output node
(30)
and equating all the harmonic components with the same fre,
quency, we can derive the expression of coefficients
and
(31)

(38)

(32)

(33)
where only dominant terms were included. The above coefficients could be used to directly find the closed-loop harmonicdistortion coefficients using (20) and (21). However, we prefer
to perform a further decomposition leading to a system model
in which linear and nonlinear blocks are isolated. After normaland given that
izing the second and third coefficient to
, we get
(34)

(35)

(39)
Distortion due to the nonlinear output conductance is effective at low frequencies. Indeed, so long as the loop gain is high,
(the error signal) is small and distortion is mainly
signal
due to nonlinearities arising in the output resistance which is
operated under large-signal conditions. For increasing frequencies, the compensation capacitor shunts the output impedance
to ground, thereby reducing the weight of nonlinearities due to
the output resistance. Moreover, signal increases (due to the
reduction in the loop gain) and the nonlinear effects of the input
transconductance become more pronounced. Thus, at high frequencies, the amplifier is more adequately modeled by the block
diagram in Fig. 8, which includes normalized nonlinear coeffiand
and ascients of the input transconductance
sumes the output resistance to be linear.
Again, by comparing this scheme with that in Fig. 4 and utilizing (25) and (26) we get

Hence, the single-stage amplifier in feedback can be schematized by the block diagram in Fig. 7, where the blocks inside the
dashed area represent the linear and nonlinear contributes of the
RC output node, with the nonlinear coefficients given by
(36)

(40)

PALUMBO AND PENNISI: HIGH-FREQUENCY HARMONIC DISTORTION IN FEEDBACK AMPLIFIERS

Fig. 8. Model of a single-stage amplifier with a nonlinear input

transconductance.

333

As a final analysis step, we consider the two distortion mechanisms together in the same block scheme as depicted in Fig. 10.
The exact resolution of this nonlinear system can be avoided
by considering that the two distortion mechanisms are dominant over different frequency ranges, as previously stated. Consequently, expressions of complete distortion factors
and
which provide asymptotic approximation can be found by
combining (38) with (40) and (39) with (41), as shown in (42)
and (43) at the bottom of the page.
These relationships have simply been obtained by algewith
braically adding, before taking their modules,
and
with
. It is confirmed that at low
frequency second- and third-order distortion components due to
the input transconductance are negligible since they are divided
and
, respectively. These contributions can
by
become dominant at high frequency, as they start to increase at
the frequency of the dominant pole.
IV. TWO-STAGE AMPLIFIERS

Fig. 9. Typical plots of second harmonic-distortion factors due to the nonlinear

and the input transconductance,
.
output resistance,

HD

HD

(41)
Both the above distortion factors increase for frequencies
higher than the amplifier pole. As a consequence, their effects
can be significant at high frequencies.
To qualitatively compare the effects on output distortion due
to the output resistance and the input transconductance, let us
consider the plots in Fig. 9. They illustrate the typical behavior
of second harmonic-distortion factors due to the nonlinear
and due to the input transconductance,
output resistance,
. The frequency determining which contribution is
and
and is close to
dominant is located between
if
approaches
. Similar plots can also be
deduced for the third harmonic-distortion factors.

In this section, we shall study the frequency behavior of

harmonic distortion in two-stage amplifiers and specifically
in two-stage transconductance amplifiers (OTAs) widely used
in CMOS applications. Fig. 11 shows the small-signal circuit
and
of a typical such architecture, it includes capacitors
that are alternatively used for dominant-pole or Miller
frequency compensation [13]. Of course, Miller compensation
is preferred thanks to its inherent advantages in terms of
lower required capacitance, frequency behavior (due to the
pole-splitting property) and process tracking features. This
makes Miller compensation, when applicable, the technique
of choice. Nevertheless, we believe instructive the comparison
of the effects of both compensation techniques, which will
be analyzed in Sections IV-A and IV-B, respectively. In the
followings and unless otherwise specified, we assume that the
main contribution to harmonic distortion is due to the output
stage. Indeed, so long as the gain of the second stage is high,
is small, while the second stage is operated under
signal
large-signal conditions. We will show that this hypothesis is
verified when using dominant-pole compensation, but leads
to slight inaccurate results for Miller compensation at high
frequency.
A. Dominant-Pole Compensation
The analysis carried out in Section II can be straightforwardly
applied to two-stage amplifiers compensated with the domi-

(42)

(43)

334

Fig. 10.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

Model of a single-stage amplifier with both nonlinear input transconductance and output resistance.

Fig. 11. Small-signal circuit of a two-stage amplifier. Capacitors C and

C are alternatively used to provide dominant-pole and Miller compensation,
respectively.

(a)

nant-pole technique. We can use the same model in Fig. 4 and

and
assume
(44)
is the voltage gain of the first stage (subscripts and
where
are used to identify parameters of the first and second stage,
respectively).
In our model, we assume also that the second stage is char, equal to
and
acterized by a linear voltage gain,
and
, are independent of
the nonlinear coefficients,
frequency
(45)
From (25) and (26), being
, we get directly

and

(46)

(b)
Fig. 12. Models of a Miller-compensated two-stage amplifier with a nonlinear
second stage: Norton equivalent input stage a), Thvenin equivalent input stage.

of interest in feedback amplifiers, as for higher frequencies the

loop gain is lower than one and we loose any beneficial effect
from feedback.
An important observation concerns the distortion caused by
the first amplifier stage. Nonlinear contributions of the input
stage are reduced by the loop gain at low frequencies and by
the compensation capacitor at high frequencies (compensation
tends to shunt to ground the output of the first stage). Therefore,
assuming the output stage as a principal source of nonlinearity
is very well justified both at low and high frequencies.
B. Miller Compensation

(47)
Second- and third-order harmonic-distortion factors start
to linearly increase (from their low-frequency values) at a
and
, respectively. Moreover,
frequency equal to
and
they become constant at frequencies equal to
, respectively. At
they begin to decrease. Note
represents the upper limit for the frequency range
that

Let us evaluate the harmonic distortion of a two-stage amplifier employing Miller compensation. At this purpose, to simplify calculations and to focus only on the parameters of interest, we rearrange the model in Fig. 11, first as in Fig. 12(a)
and then as in Fig. 12(b). In the last model, voltage-controlled
voltage-sources are utilized to model both gain stages. Again,
the (open-loop) output voltage is expressed in terms of voltage
by (45).
A block representation of the circuit in Fig. 12(b) can be
obtained by adopting different techniques. One of these is the
Rosenstark approach [13], [14] which requires computation of

PALUMBO AND PENNISI: HIGH-FREQUENCY HARMONIC DISTORTION IN FEEDBACK AMPLIFIERS

From (25) to (26), and given that

second and third harmonic-distortion factors 2

335

we get the

(54)
Fig. 13.

Equivalent block diagram of the amplifier in Fig. 12(b).

the return ratio, , the asymptotic gain,

mission term,

and the direct trans-

(48a)
(48b)
(48c)

The loop gain of the equivalent system coincides with the

and the
return ratio. Moreover, the feedback factor is
. The resulting block diagram
gain of the direct path is
is
is depicted in Fig. 13, in which the nonlinear amplifier
characterized by the same nonlinear coefficients in (45). The
diagram in Fig. 13 is similar to that in Fig. 4. Specifically, by
and
comparison we see that

(55)
To better compare the above results with those obtained in the
case of dominant-pole compensation we must express (54)(55)
and
in terms of , that is now equal to
equal to

(56)

(57)
(49)
(50)
Then, from (22)(24) we get the equivalent nonlinear coeffi,
and
which relate to
cients
in Fig. 13, as shown in(51)(53) at the bottom of the page.
The closed-loop Miller-compensated amplifier can then be
modeled with the scheme depicted in Fig. 14(a), where the amplifier studied above is closed in a loop with feedback factor .
Note that to further simplify the scheme, Fig. 14(b) includes
, with its nonlinear coefficients
the new nonlinear block
,
and
defined above. Moreover,
for conformity with the notation used in the previous subsecas equal to
tion, we define the gain of the first block
.

Starting from their low-frequency values, second- and

third-order harmonic-distortion factors linearly increase at
and
,
a frequency equal to
respectively. Compared to dominant-pole compensation, we see
that the frequency band where distortion factors remain equal to
their low-frequency values is greater in the Miller-compensated
as illustrated in Fig. 15.
amplifier by a factor equal to
This effect is due to the local negative feedback operated by
around the nonlinear gain stage.
capacitor
Equations (56) and (57) also predict that second- and thirdorder distortion factors become constant at frequencies equal
, and
, respectively. At
, they begin
to
2It is worth noting that the exact relationship derived in [10] includes another
high-frequency zero due to the load capacitance. Equation (54) does not presents
this zero since the adopted amplifier model has a zero output impedance.

(51)
(52)
(53)

336

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

(a)

(b)
Fig. 14.

Models of two-stage Miller-compensated closed-loop amplifier with nonlinear second stage. Coefficients a

j!) are defined in (51)(53).

for a two-stage amplifier compensated with Miller technique are

expressed by

(58)

(59)
Fig. 15. Typical behavior of second-order distortion factors in two-stage
amplifiers with dominant-pole compensation (curve a) and Miller compensation
(curve b). Distortion caused by the second stage only is assumed in both cases.

to decrease. This behavior was already found appropriate in

two-stage amplifiers compensated with the dominant-pole approach. In contrast, when using Miller compensation, it is unrealistic. Indeed, the local feedback operated by the Miller capacto decrease with frequency. At
itor causes coefficients
high frequencies, distortion of the first stage becomes dominant
and a nonlinear model of the first stage should then be included
to accurately predict harmonic distortion. To take into account
both effects, we can follow the same approach used in Section III
for single-stage amplifiers. This is developed in the Appendix
for completeness and reveals that including in our model nonlinearities of the first stage causes distortion to increase at frequencies around the gain-bandwidth product approximately of 20 dB
per decade. Then, a simple and rough approximation for the distortion factors including both nonlinear effects is achieved by reand
moving the poles in (56) and (57). As a result,

V. VALIDATION
The proposed analysis was first validated through extensive
simulations with Spectre on the ideal circuit models in Figs. 6
and 11 and using nonlinear controlled generators to simulate
the nonlinear circuit elements. All schemes were simulated in
open- and closed-loop conditions and under different degrees of
nonlinearity. Moreover, the sources of distortion were first simulated separately and then altogether. The error between simulated and calculated data was always within the simulator relative tolerance for a wide range of frequencies, even above the
gain-bandwidth product of the loop.
The same amplifier architectures were subsequently simulated at the transistor level, using the schematic circuits illustrated in Figs. 16 and 17 and implemented in a 0.35- m CMOS
process.
Supply voltages, bias currents, transistors dimensions, and
compensation capacitors of the two schemes were set according
to Tables I and II. Note that rather large quiescent currents were
used to avoid slew-rate limitations in the frequency range of
interest.

PALUMBO AND PENNISI: HIGH-FREQUENCY HARMONIC DISTORTION IN FEEDBACK AMPLIFIERS

337

TABLE II
DESIGN PARAMETERS OF CIRCUIT IN FIG. 17

TABLE III
PERFORMANCE PARAMETERS OF CIRCUIT IN FIG. 16
Fig. 16.

CMOS single-stage (cascode) amplifier used for the simulations.

Fig. 17. CMOS two-stage OTA used for the simulations. Capacitors C and
C are alternatively used to provide dominant-pole and Miller compensation,
respectively.

TABLE IV
PERFORMANCE PARAMETERS OF CIRCUIT IN FIG. 17

TABLE I
DESIGN PARAMETERS OF CIRCUIT IN FIG. 16

The dc voltage gain, dominant pole and other parameters of

the amplifiers are summarized in Tables III and IV.
Preliminary Fourier analyses, with a weak and low-fre, 10 Hz) and with the
quency input signal (500 V
amplifiers in open-loop configuration, were made to identify
the low-frequency values of the open-loop nonlinear coefficients. Specifically, for the cascode amplifier, parameters
and
were obtained by evaluating the output
and
were obtained by evaluating
voltage, while
the short-circuit output current. These values are reported in the
last four rows of the open-loop section in Table III. For the
two-stage amplifier, we evaluated the open-loop low-frequency

distortion factors of the second stage only. The last two rows of
the open-loop section in Table IV report the values of these
parameters (note that they are not normalized).
The low-frequency open-loop parameters were used to
evaluate the frequency behavior of harmonic-distortion factors
for the amplifiers in closed-loop configuration with a feedback
factor, , equal to 1/11. To be more precise, the amplifiers were
used in noninverting configuration, feedback was accomplished
by an ideal voltage-controlled voltage source and a 100-mV
input signal was applied. Fig. 18 shows predicted (in solid
and
lines) and simulated (marked points) values of
for the single-stage amplifier. An excellent agreement
is found between calculated and simulated data within the

338

Fig. 18.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

Harmonic-distortion factors of circuit in Fig. 16 closed in feedback.

Fig. 20. Third harmonic-distortion factors of circuit in Fig. 17 closed in
feedback. Curves (a) and (b) refer to dominant-pole and Miller compensation,
respectively.

exactly anticipated. As expected, the largest errors are found in

the third harmonic-distortion factors in Fig. 20 and are caused
by the extensive approximations adopted in modeling the amplifier distortion.
Although outside the frequency band of interest, the beand
in the Miller OTA for frequencies
haviors of
require a final comment. The monotonic increase
above
and
is not predicted by our models. The
of both
fact that curves (b) in Figs. 19 and 20 exhibit this behavior
is caused by the artificial elimination of the poles in (53) and
(54), respectively.
VI. CONCLUSION

Fig. 19. Second harmonic-distortion factors of circuit in Fig. 17 closed in

feedback. Curves (a) and (b) refer to dominant-pole and Miller compensation,
respectively.

whole range of simulated frequencies (up to 5

). Note
and caused
also that the negative peaking exhibited by
by the interaction of the two different distortion mechanisms
discussed in Section III, is precisely anticipated. It is simply
and
have opposite sign, as
due to the fact that
can be deduced from Table III.
The plots in Figs. 19 and 20 show, respectively, factors
and
for the two-stage amplifier employing the two types
of compensation. The curves labeled with (a) refer to the dominant-pole compensation, while those labeled with (b) refer to
the Miller compensation. Even if Fig. 19 (Fig. 20) does not show
(
), they are the same inthe low-frequency values of
dependently of the two compensation approaches. In contrast,
the frequency behaviors differ substantially in the two cases but
they are still predicted with excellent accuracy, at least up to
. Specifically, the better linearity of the Miller OTA is

A simplified frequency-domain analysis of the harmonic

distortion in feedback systems was presented in this paper.
The approach presumes low-distortion conditions and, by exploiting small-signal analysis, avoids the reference to complex
mathematical tools like the Volterra series. Given the system dc
loop gain and dominant pole and the low-frequency open-loop
second and third harmonic-distortion factors, the frequency
behaviors of the related closed-loop distortion factors are
analytically determined. An intuitive method of analysis was
also illustrated to further corroborate formal results. All these
derivations gave a simple but effective means to deeply understand the mechanism of harmonics generation in nonlinear
feedback systems.
As a useful application, the frequency behavior of distortion
in single-stage and two-stage CMOS amplifiers was studied and
simple equations were found. Moreover, the effect of frequency
compensation was investigated. It was analytically shown that
the Miller compensation is better than the dominant-pole one in
order to enhance the linearity performance of two-stage amplifiers. Based on the comparison between computer simulations
and calculations, it was seen that the models produce highly

PALUMBO AND PENNISI: HIGH-FREQUENCY HARMONIC DISTORTION IN FEEDBACK AMPLIFIERS

339

From (25)(26) and using (51), we get

Fig. 21. Model of a two-stage Miller-compensated amplifier with nonlinear

input stage.

(60)

(61)

Fig. 22. Calculated values of (62) under different conditions. Setting

parameter a
0: curve (a). Setting the ratio a =a equal to 0.5, 1 and
2 V : curves (b), (c) and (d), respectively. Setting parameter a
= 0 and
removing the two poles: curve (e).

accurate results, at least up to the half of the gain-bandwidth

product, thus covering nearly all the frequency range of interest.
It was also demonstrated that the principal mechanisms influencing the frequency behavior of the distortion factors in the analyzed architectures were correctly identified, thus minimizing
the modeling effort. The extension of the analysis to other more
sophisticated amplifier topologies also including nonlinear capacitors is currently under study.
APPENDIX
COMPLETE EVALUATION OF THE MILLER OTA
DISTORTION FACTORS
The Miller OTA, in which both the first and the second stage
are nonlinear can be studied in a similar manner as made in
Section III for a single-stage amplifier. First each contribution
is separately evaluated and then the effects combined by using
superposition. This method gives reasonably accurate results
when the two distortion mechanisms are dominant over different
frequency ranges. In our case distortion due to the input stage is
effective at high frequencies, whilst distortion due to the output
stage is dominant at low frequencies.
The effects of nonlinearities in the second stage have been
already studied, leading to (56)(57). Let us now consider the
amplifier under study with a nonlinear input stage and a linear
output stage, as modeled in Fig. 21.

Superimposing the effects of the two nonlinear stages (i.e.,

both input and output stage) we obtain the overall expressions
(62)(63), shown at the top of the next page.
These equations modify the frequency behavior predicted by
(56) and (57). Indeed, the value of both harmonic-distortion fac. This is qualtors now increases for frequencies around
. Curve (a) shows
itatively illustrated in Fig. 22 for factor
the typical behavior obtained through (62) setting
or, which is the same, through (56). The curve shows that the
. Curves
second harmonic distortion decreases for
(b), (c), and (d) are obtained through (62), for increasing (and
. Curve (e)
reasonably acceptable) values of the ratio
and after removing the two poles or,
is obtained for
now increases
which is the same, through (58). Factor
, roughly apwith a slope of 20 dB per decade around
proximating the behavior of curves (b)(d).
ACKNOWLEDGMENT
The authors wish to thank Dr. G. Giustolisi for the valuable
help in performing computer simulations.
REFERENCES
[1] K. Simons, The decibel relationship between amplifier distortion products, Proc. IEEE, vol. 58, pp. 10711086, July 1970.
[2] R. Meyer and A. Wong, Blocking and desensitization in RF amplifier,
IEEE J. Solid-State Circuits, vol. 30, pp. 944946, Aug. 1995.
[3] F. OpEynde and W. Sansen, Analog Interfaces for Digital Signal Processing Systems. Norwell, MA: Kluwer , 1993.
[4] P. Gray and R. Meyer, Analysis and Design of Analog Integrated Circuits, 3rd ed. New York: Wiley, 1993.
[5] K. Laker and W. Sansen, Design of Analog Integrated Circuits and Systems. New York: Mc Graw-Hill, 1994.
[6] D. Pederson and K. Mayaram, Analog Integrated Circuits for Communication: Principle, Simulation and Design. Norwell, MA: Kluwer,
1991.
[7] G. Palumbo and S. Pennisi, Harmonic distortion in nonlinear amplifier
with nonlinear feedback, Int. J. Circuit Theory Applicat., vol. 26, pp.
293299, 1998.
[8] E. Bedrosian and S. Rice, The output properties of Volterra systems
(Nonlinear systems with memory) driven by harmonic and Gaussian inputs, Proc. IEEE, vol. 59, pp. 16881707, Dec. 1971.

340

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 50, NO. 3, MARCH 2003

(62)

(63)

[9] R. Meyer, M. Shensa, and R. Eschenbach, Cross modulation and intermodulation in amplifiers at high frequency, IEEE J. Solid-State Circuits, vol. SC-7, pp. 1623, Feb. 1972.
[10] P. Wambacq and W. Sansen, Distortion Analysis of Analog Integrated
Circuits. Norwell, MA: Kluwer, 1998.
[11] P. Wambacq, G. Gielen, P. Kinget, and W. Sansen, High-frequency distortion analysis of analog integrated circuits, IEEE Trans. Circuits Syst.
II, vol. 46, pp. 335344, Mar. 1999.
[12] W. Sansen, Distortion in elementary transistor circuits, IEEE Trans.
Circuits Syst. II, vol. 46, pp. 315324, Mar. 1999.
[13] G. -. Palumbo and S. Pennisi, Feedback Ampilfiers: Theory and Design. Norwell, MA: Kluwer , 2002.
[14] S. Rosenstark, A simplified method of feedback amplifier analysis,
IEEE Trans. Educ., vol. E-17, pp. 192198, Nov. 1974.

Gaetano Palumbo (M91SM98) was born in

Catania, Italy, in 1964. He received the laurea degree
in electrical engineering and the Ph.D. degree from
University of Catania, Catania, Italy, in 1988, and
1993, respectively.
In 1994, he joined the Dipartimento Elettrico
Elettronico e Sistemistico (DEES), University of
Catania as Assistant Professor, where he became
Associate Professor in 1998, and in 2000, a Full
Professor. His primary research interest has been
analog circuits with particular emphasis on feedback
circuits, compensation techniques, current-mode approach, and low-voltage
circuits. In recent years, his research has also embraced digital circuits with
emphasis on bipolar and MOS current-mode digital circuits, adiabatic circuits,
and high-performance building blocks focused on achieving optimum speed
within the constraint of low-power operation. In all these fields, he has
collaborated with STMicroelectronics, Catania. He is coauthor of CMOS
Current Amplifiers (Norwell, MA: Kluwer, 1999), and Feedback Amplifiers:
Theory and Design (Norwell, MA: Kluwer, 2002), and is a contributor to the
Wiley Encyclopedia of Electrical and Electronics Engineering, and the author
or coauthor of more than 180 scientific papers in international journals (over
70) and in conferences.
Dr. Palumbo was awarded a grant from AEI of Catania in 1989. From
June 1999 to the end of 2001, he served as Associated Editor of the IEEE
TRANSACTIONS ON CIRCUITS AND SYSTEMSPART IFUNDAMENTAL
THEORY AND APPLICATIONS for Analog Circuits and Filters.

Salvatore Pennisi (M01) was born in Catania, Italy,

in 1965. He received the laurea degree in electronics
engineering and the Ph.D. degree in electrical engineering, from the University of Catania, Catania,
Italy, in 1992, and 1997, respectively.
In 1996, he joined the Dipartimento Elettrico
Elettronico e Sistemistico (DEES) at the University
of Catania, as a researcher, where, since 2002, he
is an Associate Professor. His primary research
interests include circuit theory and analog design
with emphasis on current-mode techniques. More
recently, his research activities have involved low- voltage/low-power circuits,
multi-stage amplifiers (with related optimized frequency compensation) and
IF CMOS blocks.
He is the author or coauthor of more than 80 publications in international
journals and conferences, as well as of the books CMOS Current Amplifiers
(Norwell, MA: Kluwer, 1999) and Feedback Amplifiers-Theory and Design
(Norwell, MA: Kluwer, 2002) and has written an entry in the Wiley Encyclopedia of Electrical and Electronics Engineering.