1118 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO.

9, SEPTEMBER 1999

Nonlinear Analysis of the Colpitts

Oscillator and Applications to Design
Gian Mario Maggio, Student Member, IEEE, Oscar De Feo, and Michael Peter Kennedy, Fellow, IEEE

Abstract—This paper reports a methodological approach to the the coexistence of different attractors. The latter phenomenon
analysis and design of sinusoidal oscillators based on bifurcation can indeed be significant in applications. In practice, the
analysis. The simple Colpitts oscillator is taken as an example coexistence of attractors implies that trajectories starting close
to demonstrate this nonlinear approach for both the nearly
sinusoidal and chaotic modes of operation. In particular, it is to an attractor may suddenly jump (catastrophically) to a
shown how regular and irregular (chaotic) oscillations can be different attractor. Even more intriguing is the situation in
generated, depending on the circuit parameters. which coexisting attractors possess fractal or intermingled
Index Terms— Bifurcations, chaos, coexistence of solutions, basin of attractions. In this case, due to physical noise, the
Colpitts oscillator, continuation methods. observed signal may be the result of a random switching of
the system trajectory between two or more attractors.
On the other hand, with the aim of characterizing the
I. INTRODUCTION
statistical properties of the signal generated, it is useful to

T HE aim of this paper is to show how bifurcation analysis

can be employed as a tool for designing oscillators in
order to obtain either a nearly sinusoidal oscillation or chaotic
classify the chaotic signal depending on the mechanism which
led to chaos. In fact, the spectrum of a chaotic signal can vary
significantly, depending on the principal mechanism governing
behavior. its dynamics; for instance, whether it is the result of a classical
Bifurcation analysis has already been used in the past, Feigenbaum cascade or if it is associated with a Shil’nikov
albeit implicitly, in the design of oscillators. In fact, in most homoclinic loop.
sinusoidal oscillator configurations, the oscillation condition In this work we consider the Colpitts oscillator as a par-
coincides with the Hopf bifurcation condition: usually a su- adigm for sinusoidal oscillation, with the aim of drawing
percritical bifurcation for which an equilibrium point loses its some general conclusions about the design of this type of
stability and a stable limit cycle is generated [1]. Moreover, oscillator. The motivations for considering this circuit are
the locus of the first period-doubling bifurcation in parameter many, including the following.
space is often computed with harmonic balance methods
• The Colpitts oscillator is a single-transistor implementa-
[2]–[4], in order to bound the region of nearly sinusoidal
tion of a sinusoidal oscillator which is widely used in
behavior.
Today, with a view to developing design techniques for electronic devices and communication systems.
chaos generation, there is a definite need to characterize the • The frequency of operation can vary from a few hertz
complex behavior exhibited by oscillators for some combi- up to the microwave region (gigahertz), depending on the
nations of the circuit parameters. From this perspective, it technology.
should be noted that by moving the control parameters away • The system possesses an intrinsic nonlinearity given by
from the oscillation condition many different kinds of complex the exponential characteristic of the active device.
behavior may arise. These include chaotic behavior or even • The system is nonsymmetric and therefore generic.
more complex situations associated with coexisting attractors. • The Colpitts oscillator exhibits rich dynamical behav-
While a huge research effort has been invested in the last ior like many other third-order oscillator configurations
decade in the investigation of chaotic dynamics, the specialized analyzed in the literature [5]–[7]. In particular, for the
literature does not report much about the implications of Colpitts oscillator there is extensive numerical and ex-
perimental evidence of chaotic behavior [8]–[10].
Manuscript received November 3, 1998; revised March 23, 1999. The
work of G. M. Maggio and M. P. Kennedy was supported in part the The approach that we will follow in our analysis is in
European Union under Grant ERBFMBICT950310, and in part by the Forbairt the context of the qualitative theory of nonlinear dynamical
International Collaboration Programme under Grant IC/98/052. The work of
O. De Feo was supported in part by the Swiss National Science Foundation
systems [11], [12]. We emphasize that, in general, nonlinear
under Grant 20-47 172.96. This paper was recommended by Associate Editor dynamical systems cannot be solved analytically. In many
V. P. Villar. cases, extensive simulations have been used to study nu-
G. M. Maggio is with the Institute of Nonlineaar Science, University of
California, San Diego 92093-0402 USA. merically the behavior of these systems. However, in many
M. P. Kennedy is with the Department of Electronic and Electrical applications it is not sufficient to look at a single trajectory.
Engineering, University College Dublin, Dublin 4, Ireland. Usually a large amount of computer time is required to gain
O. De Feo is with the Department of Electrical Engineering, Swiss Federal
Institute of Technology (EPFL), Lausanne, Switzerland. a more general (but often incomplete) idea of the behavior
Publisher Item Identifier S 1057-7122(99)07227-X. of a system. Hence, the importance of qualitative analysis.
1057–7122/99$10.00  1999 IEEE
MAGGIO et al.: NONLINEAR ANALYSIS OF COLPITTS OSCILLATOR 1119

In particular, we are interested in the qualitative analysis

of time-invariant sets in the state space. Examples of these
invariant sets include steady states, periodic solutions, and
strange attractors.
Qualitative analysis is concerned mainly with the investiga-
tion of the stability of the invariant sets with respect to small
perturbations, and their dependence on the model parameters.
In particular, a bifurcation occurs when there exists a pertur-
bation such that the invariant set exhibits different qualitative
properties. Thus, the qualitative analysis of a dynamical system
can lead to the analysis of the bifurcation that the system
undergoes, as the control parameters are varied.
For the Colpitts oscillator, by making use of bifurcation
analysis and other nonsimulative techniques (in contrast with
linear analysis and numerical time-domain simulations), we
illustrate how a complete picture of the dynamical behavior of
the circuit as a function of its parameters can be obtained.
In particular, we show how regular and irregular (chaotic)
oscillations can be generated and how the parameter space
is organized in terms of local and global bifurcations.
The paper’s structure is as follows. In Section II we in-
(a)
troduce the circuit model for the Colpitts oscillator. We
illustrate its dynamical behavior in Section III. In Section IV
we introduce continuation methods (versus simulation) in
order to carry out a bifurcation analysis of the system. The
results of the bifurcation analysis for the sinusoidal and the
highly nonlinear regions of operation of the Colpitts are
presented in Sections V and VI, respectively. In Section VII
we discuss a robustness analysis of the circuit model. Finally,
design aspects are outlined in Section VIII.
(b)
Fig. 1. Circuit model: (a) Schematic of the Colpitts oscillator. (b) BJT
II. COLPITTS OSCILLATOR transistor model in common base configuration.

A. Circuit Model
controlled current source, as shown in Fig. 1(b). More-
We consider the classical configuration of the Colpitts over, the following applies.
oscillator containing a bipolar junction transistor (BJT) as the
a) (H.3a) We model the V–I characteristic of with
gain element and a resonant network consisting of an inductor
an exponential function, namely
and a pair of capacitors, as illustrated in Fig. 1(a). Note that
the bias is provided by the current source characterized by
a Norton-equivalent conductance
According to the qualitative theory in nonlinear dynamics,
we select a minimal model for the circuit. The idea here is to if (1)
consider as simple a circuit model as possible which maintains
the essential features exhibited by the real Colpitts oscillator. where is the inverse saturation current and
This requires of course an a posteriori robustness analysis mV at room temperature.
of the model in order to validate the results. In fact, as was b) (H.3b) We assume where is the
already pointed out, a rigorous (even if numerically supported) common-base forward short-circuit current gain.
bifurcation analysis of a simpler model can provide much more This corresponds to neglecting the base current.
insight than extensive simulations of a more complete circuit c) (H.3c) Parasitic dynamics of the transistor are ne-
model. glected.1
Namely, we make the following simplifying hypotheses. We emphasize that the key element in the transistor model
1) (H.1) Ideal bias circuit, i.e., the bias current on the is the nonlinear resistor (modeling the B-E junction) which
emitter (E) is provided by an ideal current source, is responsible for most of the phenomena illustrated in this
with work. The neglected elements have only a scaling effect on the
2) (H.2) Ideal linear passive and reactive elements; observed behavior. In other words, more complete models do
3) (H.3) The transistor is modeled simply by a (voltage- 1 Note that the parasitic capacitors Cbe and Cce add in parallel with C2
controlled) nonlinear resistor and a linear current- and C1 ; respectively.
1120 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 9, SEPTEMBER 1999

not affect the qualitative dynamics of the circuit, as discussed

in Section VII, only their position in the parameter space.

B. State Equations
The state equations for the schematic in Fig. 1(a) are the
following:

(2)

where is the driving-point characteristic of the nonlinear

resistor. This characteristic can be expressed in the form
and, in particular, from (1) it
Fig. 2. Sketch of the working principle of the Colpitts oscillator in the state
follows that space. The trajectory is accelerated in the forward active region of the active
device but evolves freely, according to the natural mode of the resonant circuit,
in the cut off region of operation.

III. DYNAMICAL BEHAVIOR

C. Normalization and Parameters
We now introduce a set of dimensionless state variables A. State Space Representation
and we choose the operating point of (2) to In Fig. 2, we show the (normalized) state space
be the origin of the new coordinate system. In particular, associated with the Colpitts oscillator. The origin, which is
we normalize voltages, currents, and time with respect to the only equilibrium point of the system, corresponds to the
, , and , respectively, where operating point of the circuit. Moreover, the state space can be
is the resonant frequency of the split into two distinct regions which correspond to the modes
unloaded - tank circuit. of operation of the BJT. Note that for the transistor
The state equations (2) of the Colpitts oscillator can then works in its forward active region while for it is cut off.
be rewritten in the form The orbit sketched in Fig. 2 represents a limit cycle in
state space corresponding to a nearly-sinusoidal oscillation.
We can conceive a geometrical interpretation of this class-
C mode of operation of the Colpitts oscillator. Namely, by
(3) drawing a mechanical analogy, in the forward active region
the system trajectories are accelerated by the energy provided
by the active device (BJT). On the other hand, in the cut-off
region the trajectory evolves freely according to the natural
where mode of oscillation of the - network the unloaded - tank
circuit. A more detailed description of the qualitative dynamics
of the Colpitts oscillator, referring to a piecewise-linear model,
is reported in [14].
and .
Note the remarkable simplicity of the (smooth) nonlinearity B. Periodic Solutions
of the model. Indeed, in system (3), only the first equation An example of an actual nearly-sinusoidal oscillation for
contains the nonlinear term which, in turn, depends only the Colpitts oscillator is presented in Fig. 3(a) and (b) where
on one of the state variables, namely Also, it should be projections of the state space and the time evolution are
noted that the dynamical behavior of the system (3) depends shown, respectively.
only on the following two parameters: In the sequel we will refer to this kind of orbit as the
• the loop gain of the oscillator; fundamental (nearly) sinusoidal solution of the Colpitts os-
• the quality factor of the (unloaded) tank cillator. This behavior is significantly different from the more
circuit, complicated oscillations that the Colpitts oscillator can exhibit.
while has just a scaling effect on the state variables. In fact, for higher values of the loop gain , the oscillations
The parameter represents the (open) loop gain of the can deviate significantly from nearly sinusoidal behavior.
oscillator when the phase condition of the Barkhausen criterion Fig. 3(c) and (d) shows an example of such a complicated
[13] is satisfied. We will show in Section V that a (supercrit- periodic orbit, for parameter values for which strong nonlinear
ical) Hopf bifurcation occurs at , giving birth to a effects predominate. From Fig. 3(c), we can observe that in
nearly-sinusoidal oscillation. this case the orbit in state space exhibits several oscillations
MAGGIO et al.: NONLINEAR ANALYSIS OF COLPITTS OSCILLATOR 1121

(a) (b)

(c) (d)
Fig. 3. The n-pulse families of periodic solutions for the Colpitts oscillator. Cycles belonging to different solution families differ in the number of
oscillations in the cutoff region or, equivalently, in the number of pulses in the time series x3 (t): The nearly sinusoidal cycle (1-pulse). (a) Projection
of the limit cycle onto the (x2 ; x3 ) plane. (b) The corresponding time series of x3 : A six-pulse solution. (c) Projection of the limit cycle onto the
(x2 ; x3 ) plane. (d) The corresponding time series of x3 :

in the cut-off region that manifest themselves as a system, at steady state, settles to an equilibrium point, a limit
sequence of pulses in the corresponding time series , as cycle, or a chaotic attractor.
shown in Fig. 3(d). In practice, for each parameter combination, the system is
Simulations of the circuit suggest that there exists an infinity simulated by numerical integration for a sufficiently long time
of families of periodic solutions characterized by an increasing and the transient is discarded. The resulting attractor on a
number of oscillations in the cut-off region. properly chosen Poincaré section is then analyzed by
A more detailed discussion about the solution families for evaluating the distance between successive intersections and,
the Colpitts oscillator will be presented in Section VI-A, where finally, the period is estimated.
we will discuss the bifurcations of the limit cycles associated In Fig. 4(a) different behaviors are denoted by different gray
with each periodic solution. levels. The white area indicates a period-one solution, like
the one corresponding to a nearly-sinusoidal oscillation of the
C. Simulation Results Colpitts oscillator, illustrated in Fig. 3(a) and (b). The darker
1) Bifurcation Diagram: Fig. 4(a) shows a simulated two- regions correspond to orbits of increasing period. Finally, for
parameter bifurcation diagram illustrating the dependence of the black area, no periodic behavior was detected, and so we
the dynamical behavior of the Colpitts oscillator on the two associate chaotic behavior with the corresponding region.
control parameters and . This bifurcation diagram has In particular, from Fig. 4(a) we note the presence of a
been obtained by brute force time-domain simulation of the large region of complex behavior in the parameter space
system throughout the parameter space . The resulting in which the system undergoes several bifurcations when
trajectories are analyzed2 in order to establish whether the varying either of the parameters and This is clearly
2 The algorithm to determine the dynamical behavior of the system was
visible in Fig. 4(b) which shows a one-parameter bifurca-
implemented using the C-library CHAOSLIB by A. Abel and C. Wegener tion diagram corresponding to the horizontal line segment
(E-mail: [email protected]). in Fig. 4(a). Feigenbaum period-doubling cascades,
1122 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 9, SEPTEMBER 1999

(a) (b)
Fig. 4. Bifurcation diagrams (by simulation) for the Colpitts oscillator, obtained by analyzing the intersections of the system trajectory with the Poincaré
=
section x2 1: (a) Two-parameter bifurcation diagram illustrating the dependence of the orbit period on the parameters (Q; g 3 ): The white area corresponds

!
to period-1 behavior, while darker areas correspond to more complex periodic and nonperiodic behavior. (b) One-parameter bifurcation diagram of x3
versus Q; along the path A B in Fig. 4(a) (log10 g
3 = 0:6500):

chaotic regions, and periodic windows are clearly visible in an ODE

Fig. 4(b).
2) Limitations of Simulation Methods: At this point in the
discussion we would like to emphasize the limitations of
where is the state vector and represents the parameter
simulation techniques.
set, we are interested in how the qualitative properties of the
First, it should be noted that the results shown so far depend
system invariants change when we vary the parameters. For
on the choice of the Poincaré section adopted for carrying out
example, we might be interested in changes of the stability of
the period classification. Namely, different Poincaré sections
an equilibrium point, depending on the parameters.
give rise to different diagrams, providing different kinds
For certain invariants, namely equilibria and cycles, it is
of information. It follows that simulation results may be
possible to express their existence in an algebraic implicit
only partial and sometimes misleading. Also, we remark
form of the kind
that in Fig. 4(a) a change in color might correspond to a
real bifurcation or to a jump to a coexisting attractor. It is
impossible to distinguish between these different situations by
relying only on the results of simulations. where is a suitable space that, in general, is obtained by
In the next section we will discuss how to overcome such combining the control parameters, the state variables, and the
limitations through a computer assisted bifurcation analysis of dummy3 variables [15]–[17].
the system and we will compare the results with those obtained The defining function can be very simple (for equilibria)
by brute-force simulation.

IV. BIFURCATION ANALYSIS or incredibly complex (for cycles) and can be obtained by
A. Introduction special procedures, such as collocation methods [15]. In the
case of cycles the space is constructed by extending the
In this section we illustrate the methodology and the tech-
space with variables obtained from the discretization of
niques employed for carrying out a bifurcation analysis of (our
the cycle itself.
model of) the Colpitts oscillator.
However, the main point is that given an initial solution
In particular, we consider continuation techniques, instead
(e.g., an equilibrium or a cycle), by using an appropriate con-
of simulation, to perform bifurcation analysis. Continuation
tinuation algorithm (usually based upon prediction-correction
methods allow one to translate the bifurcation analysis of
methods for locating zeroes of functions) it is possible to
equilibria and cycles into the solution of an implicit algebraic
vary one parameter and to follow the locus in the
equation which can be computed systematically. Hence, the
space satisfying the condition . In other words,
bifurcation analysis is reduced to locating the zeroes of some
this process consists of determining how the invariant moves
functions, which can be found, with the desired precision, by
and deforms with variations of one parameter. Furthermore,
using Newton-based algorithms.
it is possible to monitor, along the continuation path, a
B. Continuation Methods certain number of so-called test functions [16] whose zeroes
Continuation methods permit the analysis of the dependence correspond to bifurcations of the invariants.
of system invariants (like equilibria or cycles) on the control 3 Auxiliary variables used in the continuation process either for discretiza-
parameters. Namely, given a dynamical system, for instance tion or monitoring purposes.
MAGGIO et al.: NONLINEAR ANALYSIS OF COLPITTS OSCILLATOR 1123

There are several advantages to using continuation methods point

as opposed to simulation in systems analysis.
• The solutions can be followed in the parameter space
even if they are unstable: by contrast, simulation allows
the observation of stable solutions only. In this regard,
it should be noted that unstable periodic orbits and,
in particular saddles, are involved in many bifurcation
phenomena. Furthermore, the saddles’ invariants separate (4)
the basins of attraction of different attractors.
• There is no need to wait for transients to settle before The corresponding characteristic equation,
studying the invariants. is given by
• The results are independent of the choice of Poincaré
section. (5)
• Numerical problems associated with sensitivity to the
initial conditions are avoided. By imposing it follows that
• Continuation makes it possible to detect hysteretic phe-
nomena due to coexisting attractors. This is difficult to
achieve by simulation methods.
confirming that for the equilibrium of (3) is
characterized by a pair of purely imaginary eigenvalues with
C. Application to the Colpitts Oscillator unitary angular frequency (due to the normalization)
. Substituting into (5) it follows also that .
In this work we used the AUTO97 [18] and CONTENT [19]
We now need to check the nondegeneracy conditions for the
continuation packages for carrying out a bifurcation analysis of
Hopf bifurcation [17]. Regarding the transversality condition,
the Colpitts oscillator with respect to the parameters and .
it can be readily verified that, for the Colpitts oscillator
In particular, we distinguish between a sinusoidal region,
i.e., the region of the parameter space where the oscillation of
the Colpitts is nearly-sinusoidal, and a nonlinear region where
strongly nonlinear effects predominate and the oscillation
deviates significantly from sinusoidal behavior. On the other hand, according to [17], the first Lyapunov
The sinusoidal region includes the area in the parameter coefficient, , which determines whether the Hopf bifurcation
space close to the locus , where the is subcritical or supercritical turns out to be
Barkhausen criterion is satisfied. However, if the quality factor
is high enough, the sinusoidal region can extend to relatively (6)
large values of the loop gain This will be discussed in
more detail in Section VIII, where the parameter space will be It follows that for the Colpitts oscillator:
characterized in terms of the total harmonic distortion (THD) Hence, for the model considered, the Hopf bifurcation at
of the signal generated. is always nondegenerate and, in particular it
is always supercritical, thus giving birth to a stable limit cycle.

VI. THE NONLINEAR REGION

V. SINUSOIDAL REGION In this section we illustrate the bifurcation phenomena
In this section we consider the nearly sinusoidal mode occurring in the parameter region of the Colpitts oscillator
of operation of the Colpitts oscillator. In particular, we will where strongly nonlinear effects predominate, so that the
show that, for the circuit model considered, the sinusoidal behavior is far from sinusoidal.
oscillation observed in the Colpitts oscillator is associated with
a (supercritical) Hopf bifurcation. A. Families of Solutions and Their Bifurcations
As was previously mentioned, the Colpitts oscillator admits
different periodic solutions which are characterized by an
A. Oscillation Condition
increasing number of oscillations in the cut-off region of the
For the model of the Colpitts oscillator that we consider, BJT or, equivalently, by an increasing number of pulses in the
application of the Barkhausen criterion provides the oscillation corresponding time series (see Fig. 3). In the following,
condition . We now show that we will classify as an -pulse solution a periodic solution
actually, for this model, a Hopf bifurcation occurs for . exhibiting pulses4 (per period) in the waveform or,
Correspondingly, the equilibrium point admits a equivalently, characterized by oscillations in the cut-off
pair of pure imaginary eigenvalues 4 More precisely, the number of pulses for x (t) is defined as the number
3
To verify what we stated above, we consider the lineariza- of intersections (per period) of the waveform with the axis x3 = 0 (with
tion (i.e., the Jacobian) of the vector field (3) at the equilibrium _ 3 < 0):
x
1124 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 9, SEPTEMBER 1999

(a) (b)

(c) (d)
Fig. 5. Feigenbaum cascade of the nearly sinusoidal cycle. The parameter g 3 is fixed at the value log10 g 3 = 0:8800 while Q is decreased through the
bifurcations curves F0;1 ; F1;1 ; 1 1 1 F1;1 in Fig. 7. (a) Period-one orbit for log10 Q = 0:7500: (b) Period-two orbit for log10 Q = 0:7296: (c) Period-four
orbit for log10 Q = 0:7026: (d) Chaotic attractor for log10 Q = 0:6900:

region. According to the definition above, the fundamental homoclinic orbit for the nearly-sinusoidal solution reported
nearly-sinusoidal solution corresponds to a one-pulse solution. in Fig. 6(a).
It can be shown that the limit cycles associated with
each periodic solution will undergo several bifurcations when B. Bifurcation Diagram
varying the parameters. Namely, the following,
Even though the bifurcation diagram for the Colpitts oscilla-
• Flip (or period-doubling) bifurcations when the cycle
tor, with respect to the parameters and is quite complex,5
loses its stability and a cycle with double period appears
we will show how it can be decomposed in terms of simpler
instead.
bifurcation structures. The overall structures can be exposed
• Tangent (or fold) bifurcations when two cycles with by identifying the bifurcation structures associated with each
different stability properties collide and disappear; -pulse solution.
• Homoclinic bifurcations, where a cycle collides with 1) Basic Bifurcation Structure: Fig. 7 reports the bifurca-
an equilibrium point, giving rise to a solution which tion structure corresponding to the one-pulse solution (i.e., the
is bi-asymptotic to the equilibrium point itself. When nearly-sinusoidal oscillation).
the equilibrium has a pair of complex eigenvalues the In Section VI-B.2, we will show that the generic -pulse
homoclinic bifurcation is said to be of the Shil’nikov type. solution admits an almost identical bifurcation structure. Thus,
We emphasize that the invariants generated by the bifurca- in the following, even if Fig. 7 refers to the one-pulse solution,
tions described above exhibit similar geometrical features to we will refer to it as the bifurcation structure associated with
the -pulse solution from which they originated. In that sense, the generic -pulse solution.
they can be considered part of the same family of solutions, The notation system adopted henceforth for the single, ,
i.e., relatives of the -pulse solution. and double subscripts, will be such that the index
This concept is illustrated in the case of a period-doubling refers to the solution family pulse), while the index is used
sequence of bifurcations leading to a Feigenbaum-like strange
attractor, as shown in Fig. 5. Similarly, Fig. 6(b) shows 5 A detailed description of the bifurcation diagram is beyond the scope of
the Shil’nikov-type strange attractor associated with the this paper. It will be published elsewhere.
MAGGIO et al.: NONLINEAR ANALYSIS OF COLPITTS OSCILLATOR 1125

(a) (b)
Fig. 6. Shil’nikov chaos. (a) Shil’nikov homoclinic orbit for log10 Q = 0:1502 and log10 g 3 = 0:5000: (b) Associated chaotic attractor for
log10 Q = 0:1510 and log10 g 3 = 0:5000:

Fig. 7. Bifurcation diagram of the nearly sinusoidal (one-pulse) solution

family. The symbols F; T ; and H; denote flip, tangent, and homoclinic
(l ) (l) (g ) (g )
bifurcations, respectively. The points C1;1 ; C2;1 ; and C1 ; C2 are cusp
codimension-two bifurcation points and imply local (l) and global (g )
coexistence, respectively. The notations adopted for the single, n; and double
(q; n) subscripts, are such that the index n refers to the solution family (n Fig. 8. Sketch of the folding of the invariant manifold at a cusp bifurcation.
pulse), while the index q is used for enumeration purposes within the same Here p1 and p2 are two generic control parameters, while x is the state
kind of bifurcation. The bifurcation diagram shown can be considered also space. The surface represents how the system invariants depend upon the
as the basic bifurcation structure associated with the generic n-pulse solution parameters. Outside the cusp region there is only one invariant (for instance
family. a cycle). Inside the cusp region there are three different coexisting invariants.
The outer invariants have opposite stability properties with respect to the
middle one. For instance the outer (x1 and x3 ) are stable (attractors) and the
for enumeration purposes in the case of multiple occurrences middle one (x2 ) is a saddle. Because of the folding of the invariant manifold, a
of the same type of bifurcation. change in the parameters can give rise to different effects. Namely, by moving
the parameters from M to N along path m, the corresponding attractor is
deformed smoothly. On the other hand, hysteretic phenomena associated with
Substructure I: Homoclinic Bifurcation Structure a sudden (catastrophic) change of attractor are observed when moving the
The first substructure we discuss is that associated with the parameters along the path n.
Shil’nikov homoclinic bifurcation (curve in Fig. 7).
The bifurcations associated with a homoclinic bifurcation A cusp degeneracy occurs when three invariants (cycles)
have been completely characterized theoretically [20], [21]. collide and disappear. This degenerate situation is illustrated
Namely, the scenario includes an infinity of flip and tangent in Fig. 8, where the geometry of the corresponding cycle
bifurcations accumulating toward the homoclinic locus manifold is shown. Here, and indicate two generic
Moreover, each tangent bifurcation curve admits two control parameters, while represents some measure of the
points of cusp degeneracy. For the sake of clarity, in Fig solution. From Fig. 8 it is evident that following the path
7 we reported only the first two flip bifurcations and the solution varies smoothly, while by varying the parameters
and the first tangent bifurcation, respectively. along the path a typical hysteretic cycle is observed. We
As can be seen from Fig. 7, the latter admits two cusp remark that the presence of hysteresis implies the coexistence
degeneracy points and . of different (stable) solutions locally, near the cusp point.
1126 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 9, SEPTEMBER 1999

For the model of the Colpitts oscillator, we expect coexis-

tence of solutions and, thus, hysteretic behavior when varying
a parameter forward and backward transversely to the tangent
bifurcation curve near the cusp points. We refer to this
situation as a local coexistence (in contrast with the global
one discussed later) associated with Shil’nikov homoclinic
bifurcations.
Having identified the homoclinic scenario allows us to ex-
ploit the theory for classifying the type of attractor, depending
on the parameters values. In particular, we can use the first
flip and tangent bifurcations to bound the regions with different
dynamical behavior. The theory [21] predicts that, for instance,
the area enclosed between the curves and is char-
acterized by a (one-pulse) Feigenbaum strange attractor [see
Fig. 5(d)]. On the other hand, the region within the curve
corresponds to a (one-pulse) Shil’nikov type chaotic attractor Fig. 9. Self-similar bifurcation structure superimposed on the simulation
results [see also Fig. 4(a)] for comparison purposes. For each n-pulse periodic
[see Fig. 6(b)]. Finally, the remaining area within is char- solution (with n = 1; 2; 3 ) we have shown only the outermost bifurca-
acterized by the existence of subharmonics of the one-pulse tion curves F0;1 ; T0;n and Tn ; Fn ; associated with substructures I and II,
solution, due to the period-doubling cascade opened by respectively.

families of periodic solutions share the same basic bifurcation

structure.
Substructure II
We now discuss the other bifurcation substructure which C. Coexistence of Attractors
is visible in Fig. 7, namely, the one comprising the two Summarizing, we have identified two possible types of co-
bifurcations curves and corresponding to tangent and flip existence of solutions within the complex bifurcation diagram
bifurcations, respectively. In contrast to the previous one, this of the Colpitts oscillator, namely, the following.
structure is not implied by the theory of homoclinic systems • Local coexistence, related to cusp bistabilities occurring
and its existence should be seen as a peculiarity of this system, near the homoclinic bifurcation loci in the parameter
even though it is a relatively common structure [22], [23]. space.
We emphasize that the substructure described above orga- • Global coexistence, meaning the multistabilities due to
nizes the coexistence of (two or more) attractors belonging overlapping of the bifurcation structures associated with
to different solution families. In fact, as discussed in detail different families of periodic solutions.
in the next section, for some regions of the parameter space A few relevant examples of coexisting attractors are re-
the bifurcation structures associated with different solution ported in Fig. 10(a)–(d), corresponding to the points
families may overlap. In such regions, two or more attractors and in Fig. 9, respectively. In particular, Fig. 10(a) shows
coexist. This coexistence can be observed for example at the two coexisting limit cycles belonging to two different families
cusp degeneracy points and . The situation is similar of periodic solutions. In Fig. 10(b), a limit cycle coexisting
to that previously described for cusp bifurcations (see Fig. 8). with a chaotic attractor resulting from a Feigenbaum cascade
However, in this case, on the two floors of the cycle manifold are shown. Fig. 10(c) shows coexistence of a limit cycle and a
there exist solutions belonging to different families. Because chaotic attractor of the Shil’nikov type. Finally, in Fig. 10(d)
this coexistence is not predicted by any local theory, we refer we report the case of two coexisting chaotic attractors, of
to it as a global coexistence. Feigenbaum and Shil’nikov type, respectively.
2) General Bifurcation Diagram: As already mentioned,
we can generalize the results obtained for bifurcations of D. Comparison with Simulation
the one-pulse solution to the generic -pulse solution family. Fig. 9 shows the results of the bifurcation analysis super-
Namely, it can be shown that the -pulse solutions exhibit imposed on the simulation results, for comparison purposes.
an almost identical bifurcation structure to that shown in We emphasize the following.
Fig. 7 but, obviously, for different values of the parameters. 1) Simulation failed to distinguish between the different
Technically, the only difference is the fact that the bifurcations bifurcation structures associated with different families
of the one-pulse solution are bounded by the curve , which of solutions.
is a flip bifurcation while, for the remaining -pulse 2) Simulation did not reveal the bifurcation structures that
solutions, the corresponding bifurcation structures are bounded are visible outside the grey area, such as the Feigenbaum
by tangent bifurcation curves cascades associated with and on the right-hand
The situation is illustrated in Fig. 9. For clarity we show side and with and on the left-hand side. The
only the main bifurcation curves corresponding to the funda- fact that these solutions were not observed by simulation
mental nearly-sinusoidal solution (cf., Fig. 7), the two- and is related to their relatively small basins of attraction.
three-pulse solutions. The self-similar structure exhibited by 3) Finally, simulation is unlikely to detect coexisting at-
the bifurcation diagram reflects the fact that the different tractors.
MAGGIO et al.: NONLINEAR ANALYSIS OF COLPITTS OSCILLATOR 1127

(a) (b)

(c) (d)
Fig. 10. Coexistence of attractors for the Colpitts oscillator at the points a; b; c; and d of Fig. 9. (a) Coexistence of two different limit cycles for
log10 Q = 0:1525 and log10 g 3 = 1:5000: (b) Coexistence of limit cycle and Feigenbaum chaos for log10 Q = 0:0430 and log10 g 3 = 0:9000:
(c) Coexistence of limit cycle and Shil’nikov chaos for log10 Q = 0:3250 and log10 g 3 = 0:7200: (d) Coexistence of Feigenbaum and Shil’nikov
chaos for log10 Q = 0:3250 and log10 g 3 = 0:8000:

VII. VERIFICATION OF THE MODEL where

A robustness analysis of the circuit model has been carried
out with respect to the hypotheses (H.1)–(H.3) introduced in
Section II-A. and we have relaxed the hypothesis (H.3b) as well. We define
Most of the nonidealities neglected (on purpose) during
the modeling process introduce only quantitative differences,
but the qualitative behavior with respect to the parameters is
maintained. while and are defined as in Section II-A.
However, regarding (H.1) concerning the ideal bias circuit, The new loop gain, , for which the phase condition of the
we would like to point out the main differences related to the Barkhausen criterion is satisfied, is given (for values of
birth of the oscillation when considering a real current source sufficiently close to unity) by
which has finite output resistance. In particular, we consider
the nonzero conductance in parallel with , as shown in
Fig. 1(a). In this case, the normalized state equations [cf., (3)]
of the Colpitts oscillator become
(8)

where when
The Barkhausen criterion for this circuit configuration is
then satisfied for . In order to make a comparison with
the ideal case , we now consider the locus where the
Barkhausen criterion is satisfied in the original parameter space
for a fixed value of This situation is summarized
(7) in Fig. 11, where the Hopf bifurcation locus computed numer-
1128 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 9, SEPTEMBER 1999

Fig. 11. Oscillation conditions for the model (7) in the plane Q; g 3 ( )
for k = 05 :; F = 0 99: ; and Q0 = 0 15
: : The solid line is the
oscillation condition provided by the Barkhausen criterion, while the dashed
line corresponds to the actual Hopf bifurcation curve. The Hopf bifurcation is
supercritical on the branch H 0 and subcritical on the branch H + . At the point
DH the first Lyapunov coefficient l1 goes to zero and the Hopf bifurcation
is degenerate. Note that the Barkhausen criterion fails for low values of Q:

ically with CONTENT is also shown. As can be seen, in contrast

to the ideal case, the Hopf bifurcation locus does not coincide
entirely with the prediction of the Barkhausen criterion. In
particular, note that the Barkhausen criterion (derived from
a linear analysis) provides only necessary, but not sufficient,
conditions for oscillation.
Finally, further investigations reveal that for low values
of , the birth of the oscillation may be associated with
a subcritical Hopf bifurcation, but this will be addressed Fig. 12. Simplified bifurcation diagram for design applications. The white
regions correspond to one-pulse behavior. We can assume the oscillation to
elsewhere as it is beyond the scope of this paper. be sinusoidal below the isodistortion line at THD = 1% : Thus, the latter
separates the nearly-sinusoidal region from the nonlinear region. The gray
regions are characterized by more complex behavior. Namely, in the regions
VIII. DESIGN CONSIDERATIONS F gn and Sn the system exhibits n-pulse Feigenbaum and Shil’nikov chaos,
Fig. 12 summarizes the results of our bifurcation analysis respectively. The cusp regions Cn are characterized by coexistence of the
one-pulse solution with one or more attractors belonging to different solution
in a form that is suitable for design purposes. In particular, families. The former is more likely to be observed because of its larger basin
according to Section VI-B it is possible to identify regions in of attraction. For the same reason, the two outer regions labeled CL and CR
have been colored white. We also show experimental oscillographs (in the
the parameter space characterized by the same (qualitative)
dynamical behavior and bounded by well-defined bifurcation
( )
IL ; VC plane) for four representative points illustrating the main types of
behavior exhibited by the Colpitts oscillator.
curves.
We note that a period-one orbit can be guaranteed by setting
the control parameters, and outside the domain ,
criterion is satisfied. Conversely, for higher values of the
corresponding to the nonlinear region in Fig. 12. Referring
region of nearly sinusoidal behavior extends to relatively high
to Figs. 7 and 9, this domain is bounded by the outermost
values of , as can be seen in Fig. 12. This is not surprising
curves and on the right-hand side and by
because if the quality factor is high enough, the loop gain can
the curves and on the left-hand side.
However, if a sinusoidal oscillation is required, we need be increased without significant distortion.
to guarantee that the distortion of the signal generated does On the other hand, from the point of view of chaos gen-
not exceed a certain threshold. For this reason, we have eration we have found that, in contrast to what had been
characterized the parameter space in terms of the THD6 of hypothesized originally, homoclinic bifurcations are not the
the output signal In particular, in Fig. 12 we show the main sources of chaos in the Colpitts oscillator. In fact,
isodistortion curves for values of THD lower than 1%. We Shil’nikov chaos is present only in narrow bands around the
note that, for low values of , these curves converge toward homoclinic loci , while most of the chaotic regions that are
the line at where the Barkhausen visible in Fig. 4 are related to Feigenbaum cascades.
6 Recall that the THD of a waveform is defined as the ratio of the root-mean-
From an application point of view, in order to produce
square (rms) amplitude of the sum of the upper harmonics to the amplitude a robust chaotic oscillation Feigenbaum regions are
of the fundamental. preferable to Shil’nikov regions because the latter are
MAGGIO et al.: NONLINEAR ANALYSIS OF COLPITTS OSCILLATOR 1129

characterized by so-called Newhouse regions [24] which make [18] E. Doedel and J. Kernévez, “AUTO: Software for continuation problems
the system extremely sensitive to parameter variations. in ordinary differential equations with applications,” Applied Math.,
1986.
Also, the designer should be aware of the possible coex- [19] Y. A. Kuznetsov and V. V. Levitin, “CONTENT: A multiplatform
istence of periodic solutions for the Colpitts oscillator. This environment for continuation and bifurcation analysis of dynamical
could be a global phenomenon, as at the cusp points denoted systems,” Centrum voor Wiskunde en Informatica, Kruislaan 413, 1098
SJ Amsterdam, The Netherlands, 1997.
by in Fig. 12, or local coexistence as described in the [20] P. Glendinning and C. Sparrow, “Local and global behavior near
previous section. homoclinic orbits,” J. Stat. Phys., vol. 35, nos. 5/6, pp. 645–696, 1984.
[21] P. Gaspard, R. Kapral, and G. Nicolis “Bifurcation phenomena near
Finally, Fig. 12 shows experimentally observed Lissajous homoclinic systems: A two parameter analysis,” J. Stat. Phys., vol. 35,
figures for a practical implementation of the Colpitts oscillator pp. 697–727, 1984.
[8], illustrating the main types of behavior analyzed in this [22] J. P. Carcasses, “Determination of different configurations of fold
and flip bifurcation curves of a one or two-dimensional map,” Int. J.
paper. Bifurcation Chaos, vol. 3, no. 4, pp. 869–902, 1993.
[23] A. P. Kuznetsov, S. P. Kuznetsov, I. R. Sataev, and L. O. Chua, “Two-
parameter study of transition to chaos in Chua’s circuit: Renormalization
group, universality and scaling,” Int. J. Bifurcation Chaos, vol. 3, no.
IX. CONCLUSIONS 4, pp. 943–962, 1993.
[24] S. E. Newhouse, “The abundance of wild hyperbolic sets and nonsmooth
In this work we have shown how bifurcation analysis sets for diffeomorphisms,” Publ. Math., IHES 50, pp. 101–151, 1979.
and continuation techniques can be exploited in a systematic
way for designing both sinusoidal and chaotic oscillators. In
addition, we have emphasized that results of simulations are
often incomplete and thus should be treated with caution by
a circuit designer. We have summarized our results with a
parameter space design chart.

REFERENCES
Gian Mario Maggio (S’95) received a Five-Year Honors degree in electronic
[1] A. I. Mees and L. Chua, “The Hopf bifurcation theorem and its engineering from the Politecnico di Torino, Turin, Italy, in March 1995 and
applications to nonlinear oscillations in circuits and systems,” IEEE the Ph.D. degree in electronic engineering from University College, Dublin,
Trans. Circuits Syst., vol. 26, pp. 235–254, Apr. 1979. Ireland, in 1999. He carried out his graduation project on semiconductor
[2] M. Basso, R. Genesio, and A. Tesi, “A frequency method for predicting modeling at Trinity College Dublin (Ireland).
limit cycle bifurcations,” Nonlinear Dynamics, vol. 13, pp. 339–360, During the summer of 1995 he worked as an electronic designer in the
1997. research and development laboratory of Philips Video Monza, Milan, Italy. He
[3] R. Genesio and A. Tesi, “A harmonic balance approach for chaos also worked on harmonic balance methods at the Politecnico di Torino, Turin,
prediction: Chua’s circuit,” Int. J. Bifurcation Chaos, vol. 2, no. 1, pp. Italy. He has visited the Swiss Federal Institute of Technology, Lausanne
61–79, 1992. (EPFL), the University of Bristol, and the University of California, Berkeley.
[4] M. Gilli and G. M. Maggio, “Predicting chaos through an harmonic He has been appointed a Postdoctoral Fellow at the Institute of Nonlinear
balance technique: An application to the Time-Delayed Chua’s Circuit,” Science (INLS), the University of California, San Diego. His research interests
IEEE Trans. Circuits Syst. I, vol. 43, pp. 872–874, Oct. 1996. are in the area of analog design, solid state devices, nonlinear dynamics of
[5] L. O. Chua, “The genesis of Chua’s circuit,” Archiv Fuer Elektronik und electronic circuits (especially oscillators), and chaos generation.
Uebertragungstechnik, vol. 46, no. 4, pp. 250–257, 1992. Dr. Maggio was the recipient of a Marie Curie fellowship.
[6] E. Freire, E. Ponce, and F. Torres, “Hopf bifurcations in piecewise
linear planar dynamical systems,” in Proc. NDES’96, Seville, Spain,
June 1996, pp. 129–134.
[7] H. Kawakami and R. Lozi, Structure and Bifurcations of Dynamical
Systems. Singapore: World Scientific, 1992.
[8] M. P. Kennedy, “Chaos in the Colpitts oscillator,” IEEE Trans. Circuits
Syst. I, vol. 41, pp. 771–774, Nov. 1994.
[9] C. Wegener, G. M. Maggio, and M. P. Kennedy, “An approximate
one-dimensional model for the chaotic Colpitts oscillator,” in Proc.
NDES’96, Seville, Spain, June 1996, pp. 275–278.
[10] G. M. Maggio, C. Kennedy, and M. P. Kennedy, “Experimental mani-
festations of chaos in the Colpitts oscillator,” in Proc. ISSC’97, Derry,
Oscar De Feo received the Bachelor degree (Laude) in industrial electronics in
Ireland, June 1997, pp. 235–242.
[11] S. H. Strogatz, Nonlinear Dynamics and Chaos: With Applications in 1990 from the Maxwell High School, Milan, Italy, and the Five-Year degree
Physics, Biology, Chemistry, and Engineering. New York: Addison- (Laude) in Computer Science Engineering from the Politecnico di Milano,
Wesley, 1994. Milan, Italy, in 1995. In January 1997 he joined the Chair of Circuits and
[12] W. Jansen, CANDYS/QA: Algorithms, Programs, and User’s Manual, Systems at the Swiss Federal Institute of Technology, Lausanne, Switzerland,
Univ. Potsdam, Dec. 1995. where he is working toward the Ph.D. degree.
[13] J. Millman and A. Grabel, Microelectronics, 2nd ed. Singapore: He was with the Fondazione Eni Enrico Mattei (FEEM), Milan, Italy,
McGraw-Hill, 1988. as a Postgraduate Fellow, working on problems of sustainable development
[14] G. M. Maggio and M. P. Kennedy, “On the relationship between and environmental impact. He continued his research studies in nonlinear
continuous-time and discrete-time models of the Colpitts oscillator,” dynamics participating in the Young Scientists’ Summer Program (YSSP) at
in Proc. NDES’97, Moscow, Russia, 1997, pp. 240–245. the International Institute for Applied Systems Analysis (IIASA), Laxenburg,
[15] E. Doedel, H.B. Keller, and J.P. Kernévez, “Numerical analysis and Austria. During 1997 and 1998 he held visiting research positions at the
control of bifurcation problems (I) bifurcation in finite dimension,” Int. IIASA, at the Research Institute for Mathematics and Computer Science
J. Bifurcation Chaos, vol. 1, no. 3, pp. 493–520, 1991. (CWI) Amsterdam, The Netherlands, and at the Department of Electronic
[16] A. R. Champneys and Y. A. Kuznetsov, “Numerical detection and con- and Electrical Engineering, University College, Dublin, Ireland. Oscar’s
tinuation of codimension-2 homoclinic bifurcations,” Int. J. Bifurcation research interests are in the fields of bifurcation and nonlinear systems theory,
Chaos, vol. 4, no. 4, pp. 785–822, 1994. numerical methods for nonlinear system analysis, bioengineering and design
[17] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory. New York: of nonlinear systems for engineering.
Springer-Verlag, 1995. Mr. De Feo received the Mikhalevich Award for his research activities.
1130 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 9, SEPTEMBER 1999

Michael Peter Kennedy (S’84–M’91–SM’95–

F’98) received the B.E. degree in electronics from
the National University of Ireland in 1984, and
the M.S. and Ph.D. degrees from the University of
California, in 1987 and 1991, respectively,
He has worked as a Design Engineer with Philips
Electronics, a Postdoctoral Research Engineer at
the Electronics Research Laboratory, University of
California, Berkeley, and as a Professeur Invité at
the EPFL, Switzerland. He returned to University
College Dublin (UCD) in 1992 as a College Lecturer
in the Department of Electronic and Electrical Engineering, where he teaches
electronic circuits and computer-aided circuit analysis, and directs the
undergraduate Electronics Laboratory. He was appointed Senior Lecturer
at UCD in 1996. He has published over 100 articles in the area of nonlinear
circuits and has taught courses on nonlinear dynamics and chaos in England,
Switzerland, Italy, and Hungary. He has held visiting research positions at
the EPFL, AGH Kraków, TU Budapest, and the University of California,
Berkeley. His research interests are in the analysis, design and simulation of
nonlinear dynamical systems for applications in communications, and signal
processing.
Dr. Kennedy received the 1991 Best Paper Award from the International
Journal of Circuit Theory and Applications. He served as an Associate Editor
of the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS from 1993 to 1995,
Guest Editor of 1997 and 1999 special issues in the IEEE TRANSACTIONS ON
CIRCUIT AND SYSTEMS, and was Chairman of the IEEE Technical Committee
on Nonlinear Circuits and Systems from 1998 to 1999.