Class Notes, 31415 RF-Communication Circuits

Chapter VI

OSCILLATORS

Jens Vidkjær

NB234
ii
Contents

VI Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
VI-1 Oscillator Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Prototype Feedback Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Negative Conductance Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Frequency Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Amplitude Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Oscillator Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
VI-2 Oscillator Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
LC Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Simplified Analysis of Colpitts Family of Oscillators . . . . . . . . . . . . 18
Case 1 in load resistor placement . . . . . . . . . . . . . . . . . . . 19
Case 2 in load resistor placement . . . . . . . . . . . . . . . . . . . 21
Case 3 in load resistor placement . . . . . . . . . . . . . . . . . . . 22
Arbitrary resistor placements . . . . . . . . . . . . . . . . . . . . . . . 24
Example VI-2-1 ( Seiler type LC oscillator ) . . . . . . . . . . . 25
Example VI-2-2 ( Pierce type LC oscillator ) . . . . . . . . . . . 29
VI-3 Crystal Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Crystal Equivalent Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Loading and Driving Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Crystal Oscillator Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Example VI-3-1 ( IC-connected crystal oscillators ) . . . . . . . 46
APPENDIX VI-A Oscillator Amplitude Stability . . . . . . . . . . . . . . . . . . . . 49
Example VI-A-1 ( amplitude stability ) . . . . . . . . . . . . . . . 54
Example VI-A-2 ( amplitude stability with current bias ) . . . 57
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
References and Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

J.Vidkjær iii
iv J.Vidkjær
 1

VI Oscillators

Most communication system and measurement equipment include oscillators and they
are often key components that directly influence system specifications and tolerances. There-
fore, oscillator design objectives commonly encompass one or more of the following question
in addition to the basic requirement of producing periodic waveshapes.

- Frequency sensitivity and drift in short and long terms

- Waveshapes and Harmonic distortion
- Noise
- Frequency control and adjustment range

To address the fundamental properties in an instructive manner, we shall start

considering some prototype oscillator circuits, where the different aspects of oscillator
performance may be separately considered. The prototype oscillators are simple because they
include transformer-coupled feed-back paths. In practical circuits transformers are often
avoided and the feed-back paths may be more difficult to identify.

It is not possible here to cover more than a few of the oscillator circuits that are in
practical use. Below we consider and exemplify one very common family of oscillator circuits
in some depth. Other oscillator types may be treated equivalently using the approaches that
are demonstrated here, so the reader should be able to conduct investigations of other circuits
on his own as necessary.

One topic is given more attention here than in most other introductory texts, namely
the mechanisms of amplitude control. This is a topic that must be addressed in every oscillator
design, but since it is strongly nonlinear, it is often considered as a problem to be dealt with
experimentally. Below, however, we benefit from the presentation of nonlinear amplifiers in
the foregoing chapter and include amplitude control in the design examples. Also the more
advanced problem of amplitude stability is considered but, due to the lengthy calculations, in
an appendix.

J.Vidkjær
2

VI-1 Oscillator Basics

Fig.1 Basic oscillator components and the corresponding signal flow-graph.

Oscillator circuits for sinusoidal oscillations include functions that may be illustrated
by the block-diagram of Fig.1. There is a level dependent - i.e. a nonlinear - amplification
A(Vin), which stabilizes the amplitude of oscillation, and a frequency selective positive
feedback circuit b(ω), which determines the frequency of oscillations. Besides the voltage from
the feedback loop around the amplifier, an additional input voltage Vn that represents either
amplifier noise or a fast decaying signal is inserted. It is required to start oscillations,
especially in analytical models or simulations. The voltage transfer function, which is
illustrated by the flow-graph, implies

(1)

Self sustained oscillations are possible at a frequency ω0 and an amplifier input Vin0, where
the transfer function above becomes singular. This properties, which implies a loop gain equal
to one, is known as Barkhausen’s condition,

(2)

The criterion must be applied in a strictly complex sense so the equation constrains real parts
or phasor lengths if, simultaneously, the imaginary parts or the phasor angles are forced to
zero. To ensure start of oscillation and subsequently amplitude stabilization, the loop gain
must start being greater than one with no amplitude and then decrease monotonously until
Barkhausen’s criterion is met at the desired amplitude. A characteristic like the one in Fig.2
has this property.

Fig.2 Nonlinear amplifier characteristic for oscillator amplitude stabilization.

J.Vidkjær
VI-1 Oscillator Basics 3

Separation into a nonlinear amplifier circuit and a frequency selective feedback

network may not be as distinct in practice as suggested by the initial presentation above.
Designing oscillators it is important, however, to identify the two functions and assure that
they are working properly under influence of each other including the ability to start oscilla-
ting at the required frequency. More detailed accounts on such basic questions may be found
in ref´s [1],[2].

Prototype Feedback Oscillators

Fig.3 Oscillator circuit with class AB or C amplification ( class depends on circuit

parameters ). The transistor equivalent in (c) assumes no saturation and relates
fundamental frequency components.

We shall start our discussion of oscillators by considering simple circuits where the
frequency is determined by a parallel resonance circuit, sometimes referred to as the tank-
circuit. The tuning inductance is primary winding in a transformer, where feedback is taken
from the secondary winding. Candidates for an amplifier are the parallel-tuned large-signal
amplifiers with current or resistor biasing or the differential amplifier. They have both
decreasing gain with increasing input drive voltage as it was discussed in section 5-3. An
oscillator with a single transistor is shown in Fig.3(a), and although a circuit with a separate
feed-back winding may be impractical in practice, it is easily understood and it will be a
vehicle in the present introductory discussion. Fig.3(c) shows the large signal equivalent circuit
for the transistor. Numerical methods were employed in Chap.5 - Eqs.(5-101) through (5-107)
-to find the large-signal transconductance Gm, which is a function of the drive level Vb1. It
was assumed that the transistor operates free of saturation, and that the Q-factor is high
enough to ensure a sinusoidal base-emitter voltage around the mean voltage Vb0,

(3)

Investigation of Barkhausen’s condition in an oscillator circuit may be done as shown

by the equivalent circuit in Fig.4, where we - hypothetically - have opened the loop inside the

J.Vidkjær
4 Oscillators

Fig.4 Equivalent circuit for calculating open loop gain. Rlp represents possible losses in
inductor L. Oscillations are established by connecting the two ports as indicated by
the dashed line.

transistor. Connecting input and output ports in this circuit establishes the oscillator with Vb1
= V´b1, a condition that also must apply to the open-loop equivalent circuit, where the
condition now reads,

(4)

To get correct sign in the condition, the transformer winding is turned to cancel the sense of
Gm in common emitter configuration. The total parallel resistance of Zp is combined from the
external load, RLP, a possible inductor loss, Rlp, and the transistor input resistance Rπ seen
through the transformer, which has winding ratio N. We get

(5)

where QL is the Q-factor of the inductor at the resonance frequency ω0. The impedance of the
parallel tuned circuit is expressed

(6)

Frequency dependency is here kept in the function β(ω), which was introduced in Chap.II,
Eq.II(6). The function, which becomes zero at the resonance frequency, should not be
confused with the transistor current gain βf. The Q-factor Ql is the loaded Q, which includes
all resistances in the resonance circuit. The resonance frequency expression may be considered
as the imaginary part of Barkhausen’s requirement since at resonance Zp gets no imaginary
part. The condition becomes

J.Vidkjær
VI-1 Oscillator Basics 5

(7)

Incorporating the relationship between the effective fundamental frequency transconductance

and input resistance of the transistor,

(8)

we get from the real-part requirement

(9)

The last, approximate, expression assumes that the transformer winding ratio over the
transistor current gain βf is small. This will be the common situation with N less than one and
current gain significantly greater than one. Observe that the effect of the assumption is to
disregard the loading of the resonance circuit by the transistor input resistance.

Had we insisted in a strict identification of the two basic oscillator blocks from Fig.1
in the present circuit, it would be a natural choice to let the center frequency gain represent
the level dependent amplifier, and let the feedback through the secondary winding be the
frequency selective block, i.e.

(10)

We may think in terms of this separation when basic stability and noise properties of oscilla-
tors are discussed in subsequent paragraphs.

Instead of employing a limiting amplifier with one transistor in the prototype

oscillator, the differential stage could also be used. An example is shown in Figure 5-57.Inser-
ted in (b) is the large signal equivalent circuit of a differential pair from Chap.5, Fig.5-54,
now with terminal grounded according to the diagram of the complete oscillator. The total
collector current in transistor Q2 is includes a DC component of approximately half the tail
current IET and the differential current ΔIc, where, cf. Eqs.(5-140) and (5-141),

J.Vidkjær
6 Oscillators

Fig.5 Oscillator circuit with differential amplifier. The equivalent circuit in (b) assumes
no saturation and relates fundamental frequency components.

(11)

Compared to the single transistor coupling in Fig.3, the controlled current source has changed
direction. This is also the case in the equivalent circuit of Fig.6, which illustrates the loop-gain
in Barkhausen’s criterion. To compensate the new current source orientation, the sense of the
transformer must also be changed in comparison with the single transistor case from Fig.4.
Excepts for details in the functions that govern the fundamental frequency transconductances
Gm and Gmd respectively, the equivalent circuits are similar, so with respect to the fulfillment
of Barkhausen’s criterion we need not consider the differential prototype oscillator separately,
but simply use Gmd in Eq.(9) instead of Gm.

Fig.6 Equivalent circuit for calculating the loop gain in the differential oscillator from
Fig.5. Rlp represents still possible losses in inductor L.

J.Vidkjær
VI-1 Oscillator Basics 7

Fig.7 Redrawing of Fig.4 without Rlp and Rπ1 (a), illustrating the negative conductance
point of view in oscillator design. Negative conductance is found by the setup in (b).

Negative Conductance Criteria

Instead of Barkhausen’s criterion, the conditions for oscillation may sometimes be

formulated from a negative conductance or negative resistance point of view1. In an un-
damped resonance circuit, oscillation will continue forever at the resonance frequency once
they have been started. If power is taken from the circuit by parallel loading through a
passive, i.e. positive, conductance, continued oscillation requires continuous supply of power.
To maintain the conditions for an undamped resonator, the model of the circuitry that delivers
power must contain a conductance in parallel to the load of same size but with negative sign.
We may illustrate the method by reformulating the oscillation condition for our prototype
oscillator to negative conductance form.

The diagram in Fig.7 holds a simplified version of the oscillator equivalent circuit in
Fig.4. Inductor losses and the input resistance of the transistor are disregarded in the new
circuit, which is redrawn to emphasis its separation into a passive load, YL, paralleled by an
active network with device and feed-back components, YD. The separating cut is somewhat
arbitrary provided that the ohmic parts of the device and the load are at different sides. In
Fig.7 we have chosen to let the device admittance be real and frequency independent, a
decision that also separates the amplitude control function from the frequency selective
network. The parallel admittance of the device may be found as sketched in Fig.7(b), i.e.

(12)

It follows from the initial resonance circuit discussion that the admittances - impedances as
well - sum to zero at stationary oscillations, i.e.

1) Formerly, that sort of considerations were used reversely to avoid oscillations in

setting up two-port stability criteria in sec.III-1,p.11.

J.Vidkjær
8 Oscillators

(13)

Taking real and imaginary parts in the equation above leads to the same requirements as did
Barkhausen’s criterion in Eq.(7). We get

(14)

At microwave frequencies oscillators may be build around diodes that have negative incremen-
tal output conductances where the negative conductance approach is appropriate. In most cases
there are no advantage of using the method on feedback oscillator couplings, which are our
main concern below. Having demonstrated equivalency between the two points of view we
shall therefore leave the negative conductance subject. Interested readers may consult
refs.[3] or [4] for further discussions and references.

Frequency Stability

Parameter variations due to temperature variations or parameter spreading in produc-

tion clearly changes the conditions for oscillation. Using logarithmic differentiation, variations
of the tuning capacitance or inductance changes the oscillating frequency of the prototype
circuit according to

(15)

To keep the oscillating frequency stable under thermal variations we should clearly use tuning
elements with small thermal coefficients. Crystals, which are considered in details later, are
examples showing extremely small temperature variations. Using conventional components,
a simple mean for stabilizing is to chose inductors and capacitors that have relative thermal
coefficients of equal size but opposite signs.

The amplifier circuit in the idealization from Fig.1 contributes nothing to the
frequency stability of the oscillator. Practical amplifiers, however, include phase-shifts that
must be accounted for when Barkhausen´s condition is enforced. Suppose we have an
amplifier with a small phase deviation Δϕ, i.e.

(16)

If the feedback follows the impedance of a parallel resonance circuit, the frequency of
oscillation will differ from the resonance frequency ω0 to compensate the extra phase

J.Vidkjær
VI-1 Oscillator Basics 9

Fig.8 Impedance of a parallel tuned circuit. The steepness of the phase characteristic
determines the oscillator frequency sensitivity Δω/Δϕ

deviation as indicated by Fig.8. The phase of a parallel tuned circuit is, cf. Eq.(6),

(17)

Thereby, the phase and frequency deviations around resonance become

(18)

Barkhausen´s condition requires zero phase-shift around the loop, so to compensate a small
amplifier phase-shift, the oscillation frequency must deviate Δω from ω0, where

(19)

This results shows clearly that a high Q-factor in the resonance circuit provides a low
influence on the oscillator frequency from the amplifying circuit. The latter may be build from
conventional electronic components that are more prone to temperature variations and
parameter spreading than the often highly specialized resonator components, which determine
frequency and Q. A so-called frequency stability factor Sf are sometimes introduced, where

(20)

In the prototype circuit, the Sf concept is a simple alternative to the Q-factor and seemingly
superfluous. There are, however, other oscillator configurations where the association to a Q-
factor is not straight-forward. In such cases the frequency stability factor is a meaningful
quantity by itself or, alternatively, the derivative of the open-loop gain phase with respect to
angular frequency may by used to calculate a Q-factor that indicates frequency stability.

J.Vidkjær
10 Oscillators

Amplitude Stability

Oscillators with amplitude control of a type that involves the DC bias in the amplifier
to adjust gain, may exhibit an amplitude instability problem known as squegging. This is
illuminated by Fig.9 that shows simulations on the oscillator circuit in Fig.3, where the DC
bias current through the nonlinear driven transistor depends on the signal level. In contrast,
the bias current of the differential prototype oscillator from Fig.5 is kept fixed, so this type
of problems should not be seen, if the amplifier is well designed. However, this and many
other oscillator circuits may still exhibit amplitude instabilities if the amplitude control is not
well planned, so transistors are forced into unintentional saturation or secondary breakdown,
which may provide unexpected bias variations.

Like many other analog large-signal nonlinear circuit problems, it is difficult to

conduct simple analytical investigations of the oscillator amplitude stability and to get results
that are suitable in design. Even simple cases become rather lengthy and involved so we refer
to Appendix VI-A, which discusses amplitude stability in the single transistor prototype
oscillator. In brief, it is demonstrated that to avoid squegging, the time constant in the bias
adjustment - in turns the gain regulation - should be of comparable length to the time constant
by which oscillation may build up or decay across the resonance circuit, i.e. the logarithmic
decrement that was introduced by Eq.II(30) in section II-1.

Fig.9 Prototype oscillator outputs, (a) stable amplitude, τE/τQ=1, and (b) instability for
τE/τQ=4.5 (details in Appendix VI-A).

J.Vidkjær
VI-1 Oscillator Basics 11

Oscillator Noise

Fig.10 Spectrum analyzer output from an oscillator. (Δf) is called the single sideband
noise function. The widening of the spectrum around the center frequency is caused
by noise.

Noise in oscillators manifests itself by an output spectrum that is not just a single line
but gives spectrum that widens apart from the center frequency as sketched in Fig.10. We
shall consider a simplified, partly empirical oscillator noise model, the so-called Leeson model
[5], in this section. The model explains major noise properties and parameter dependencies
in experimental observations. The prototype oscillators from above are used to make the
simplified model plausible, but due to the nonlinear amplifier operation in an oscillator, the
detailed theoretical treatment of oscillator noise is complex and far beyond our scope. To go
further, the interested reader should consult ref´s [6] or [7],sec.10.4.

Fig.11 Oscillator block-scheme for noise description.

The oscillator noise model is obtained in two steps. It is assumed that the noise can
be ascribed to the amplifier and represented by an equivalent input noise. First the relationship
between input and output noise spectra are calculated. Second, the nature and shape of the
actual noise source spectra are incorporated.

Fig.11 repeats out initial block scheme representation of an oscillator. Applied to the
prototype oscillator, the subdivision from Eq.(10) implies that the frequency selective feed-
back network holds lossless components L and C only. All noise sources are represented by
the input noise voltage Vn,in. When the oscillator is running, the output contains both the
desired constant amplitude sinusoidal voltage Vocosω0t and noise components Vn,out, which

J.Vidkjær
12 Oscillators

originates from Vn,in. We suppose initially that all noise voltages are so small that they have
no influence on the amplitude control and, furthermore, that the oscillations have settled at
their final amplitude. Barkhausen’s requirement expressed through Eqs.(7) to (9) is

(21)

Invoking this condition, and using the b-function from Eq.(6), the noise voltage transfer
function becomes

(22)

where the frequency dependency is kept in β(ω). Going to noise spectra, the power spectral
densities at the input and output of the oscillator, when normalized with respect to the same
impedance levels, are related by the absolute square of the voltage transfer function. If Sn,in(ω)
is the input noise spectrum and Sn,out(ω) the corresponding output spectrum, we have

(23)

Around the resonance frequency ω0 we may replace the Qlβ product by its narrowband
approximation, which was introduced in section II-3. Thereby, we get expressions for the
spectra in terms of the frequency deviation Δω from resonance,

(24)

Here W3dB is the usual 3dB bandwidth of the resonance circuit. We compare to only half of
that because the bandwidth is defined as the distance between the 3dB point above and below

Fig.12 Asymptotic transfer characteristic between input and output noise spectra in the
oscillator block-scheme from Fig.11.

J.Vidkjær
VI-1 Oscillator Basics 13

ω0 with Δω being either positive or negative. A break-point characteristic corresponding to

the transfer function is shown in Fig.12. It goes towards infinity when the frequency is
approaching the oscillator frequency but this was expectable as Barkhausen´s condition implies
infinite gain at resonance frequency if the oscillator is considered as an feedback amplifier.

Oscillators noise is commonly separated into amplitude noise, vna(t), and phase noise
φn(t). The total oscillator output voltage is expressed

(25)

The last approximation assumes a phase-noise much smaller than one radian, so its cosine
function approximates one and the sine function approximate the phase-noise itself. Compared
with an additive noise description

(26)

it is seen, that the last line in Eq.(25) is equivalent to an in-phase and quadrature resolution
of the additive noise around the oscillator frequency. Starting from common additive gaussian
noise sources, the noise power divides evenly between the in-phase and the quadrature terms
in such a resolution. Dealing with oscillator noise, it is custom to disregard the amplitude term
since the oscillator circuit is designed to keep the amplitude constant and, furthermore, the
output commonly undergoes further limiting before it is applied in critical application. We
shall follow this simplifying approach below.

Suppose the noise in the oscillator amplifier is white thermal noise. In equivalency
to small-signal two-port noise characterization, the portion of the input noise spectrum that
accounts for phase noise may be written in the form

(27)

where Feff is an effective noise figure that describes the amplifier under large signal sinusoidal
excitation, k is Boltzman’s constant, and To is the reference temperature for noise in Kelvin.
Since this spectrum is flat, the result is directly transferred to the so-called single-sideband
phase noise spectrum (Δf). It is a spectrum like Fig.10, which is read directly from a
spectrum analyzer using 1Hz resolution bandwidth. It gives the power spectral density relative
to the sinusoidal output power Po, which is proportional to ½Vo2. Δf denotes the difference
in frequency from the oscillator signal. The simple noise model assumes a spectrum that is
symmetric around the oscillator frequency. To fully include the power in one hertz bandwidth

J.Vidkjær
14 Oscillators

at frequency distance |Δf| from the oscillator frequency we must include twice the value of
(Δf), where

(28)

In dB scale, 10 log (Δf) is referred to as dBc, decibels over carrier. Clearly this spectrum is
a scaled version of the transfer characteristic from Eq.(24) since the input spectrum is flat.
However, it was discussed in the flicker noise paragraph of section IV-1, Eq.IV(21), that the
1/f noise in electron devices are transferred to the frequency range around a large sinusoidal
driving signal. We have the same driving situation by the last term in Eq.(25), so including
flicker noise, the input spectrum gets a 1/f term and the output may be written,

(29)

where the characteristic frequency fa is a flicker noise parameter. The expression is the final
form of the Leeson oscillator model, which gets two asymptotic appearances dependent on
whether fa is smaller or larger than half the 3dB bandwidth of the resonance circuit in the
oscillator. These two cases are summarized in Fig.13 (a) and (b) respectively.

Fig.13 Asymptotic single sideband oscillator noise characteristics where in (a) resonator half-
bandwidth exceeds the flicker noise bandwidth and vice versa in (b).

J.Vidkjær
 15

VI-2 Oscillator Circuits

The number of oscillator circuits that are used in practice is overwhelming, and it is
impossible here to give a broad coverage. Therefore, the treatments below serves the purposes
of both presenting a few common oscillator configurations and of presenting analytical
techniques that also may be useful with other oscillator circuits. Furthermore, we shall point
out a few important design aspects.

LC Oscillators

Fig.14 LC oscillator example. The equivalent circuit in (b) includes the large signal transis-
tor model in Fig.15. Resistors RA, RB, and RC may represent losses and/or loads.

Circuits intended for sinusoidal outputs up to a few GHz in frequency may be build
using inductors and capacitors as tuning elements - the so called LC-oscillators. Compared
with the prototype circuits in the forgoing section, the circuits we consider here are simpler
in the sense that the required feed-back is inherent in the circuit topology. However, the
requirements for oscillations are not identified as easily as it could be done in the transformer
feed-back circuits.

One family of practical oscillator circuits, which we shall call the Colpit’s family, is
based on the circuit topology in Fig.14(a), where capacitors CA, CB, and inductor LC are

Fig.15 Transistor equivalent circuit. Under large signal conditions, the components apply
to the fundamental frequency and they depend on the signal levels at the input and
output ports.

J.Vidkjær
16 Oscillators

supposed to be the dominant tuning elements. The transistor symbol in the diagram represents
the transistor including biasing arrangements under large signal operation. At this level of
approximation the transistor may be either bipolar or FET types. In the foregoing analysis of
the prototype oscillator we used the simple, ohmic large signal transistor model in Fig.3(c).
With caution, this model may be extended to incorporate capacitive effects, so the transistor
is described by the large signal model in Fig.15. Like the simpler model, the new components
Cπ1, Cμ1, Co1, and Ro1 apply to the fundamental frequency and depend on the voltages across
the transistor, so the quantities are not deductible in details from the small-signal data that
commonly are provided in data-sheets. However, it is seen in the equivalent circuit of
Fig.14(b), which is cast in a form that facilitates open loop gain calculations, that all elements
in the transistor are shunted by external components. Resistors RA, RB, and RC might
represent possible capacitor and inductor losses, external loads, or both. External components
are expected to play the dominant role, since it is a common design objective to keep the
oscillator insensitive to transistor parameter variations. For that purpose even crude approxi-
mations to the large signal transistor parameters suffices. Note finally that with exception of
possible biasing contributions we may often disregard the transistor resistors Rπ1 in FET and
Ro1 in bipolar transistor calculations.

Fig.16 Circuit structure and signal flow-graph for calculating the open loop gain in the
oscillator equivalent circuit from Fig.14(b).

Like the previous oscillator prototype analysis, the condition for oscillation is that the
open loop gain in the circuit equals one in size and zero in phase. The structure of the circuit
in Fig.14(b) has a form that is given by Fig.16. In general terms the open loop voltage gain,
Aol , and the criterion for oscillation are expressed,

(30)

(31)

Inserting components

(32)

and taking real and imaginary parts in Eq.(31), the condition splits into the following two
conditions that must be simultaneously met,

J.Vidkjær
VI-2 Oscillator Circuits 17

(33)

(34)

In details, the admittances to be inserted in the two requests are

(35)

Inserting these expressions, the two requests constitutes a system of equations that determines
the frequency of oscillation ω0 and the large signal transconductance Gm, which is necessary
to maintain oscillations. Besides these two significant quantities, we shall often require the Q-
factor - the loaded Q - of the open-loop gain since it determines the important properties of
frequency stability and noise in the oscillator. A direct way of conducting the calculation is
to find the derivative of phase with respect to angular frequency in loop gain and then,
cf.Eq.(20), get the Q-factor from the relationship

(36)

Finally, the loading of the transistor might be required to check whether or not a device is
adequate in a given application. According to Fig.16 and Eq.(30), the load impedance seen
from the port of the tranconductance current generator is given through

(37)

(38)

The oscillator circuit topology in Fig.14(a) contains no ground node and the corre-
sponding criteria for oscillations are independent of which one of the transistor terminal that
is set to - or taken as - the signal ground. Distinctions between the different ways of ground-
ing is made from practical reasons since the literature often refer to circuits named after their

J.Vidkjær
18 Oscillators

Fig.17 Colpitt’s (a), Pierce (b), and Seiler (c) oscillators. They are deduced from Fig.14 by
grounding base, emitter, or collector - gate, source, or drain - respectively.

inventors. The tree possibilities with the present topology are summarized in Fig.17. They are
known as Colpitts, Pierce and Seiler oscillators. Considering the figures it should still be kept
in mind that the transistor symbol represents a transistor under large signal operation, either
bipolar’s or FET’s. It includes the type of bias network that is required to reduce transcon-
ductance and in turn stabilize the oscillation amplitude. Note also that the output indications
in the figures are suggestions as any non-grounded node in the circuit has an oscillating
voltage component. Which one to use depends on the application since the different nodes
differ with respect to distortion, noise, and output power capabilities if they are loaded.

Simplified Analysis of Colpitts Family of Oscillators

To illuminate and to make practical use of the conditions from Eqs.(33) to (38), we
consider a few simplified situations where it is assumed that the transistor feed-back capaci-
tance is negligible, so the susceptances are in all cases given by

(39)

Assume for a moment that there were no resistive losses in the circuit, i.e.

(40)

In this hypothetical situation, the real part condition from Eq.(33) provides

(41)

J.Vidkjær
VI-2 Oscillator Circuits 19

As seen, the frequency of oscillation ω0 becomes the resonance frequency ω00 of an ideal
resonance circuit made of inductor LC and the series connection of capacitances Ca and Cb.
The corresponding imaginary part condition implies

(42)

which means that once the oscillations have started, there are no need for an active element
to compensate resistive losses, since they are absent in this particular case. Below we shall
include losses and find the corresponding transconductance requirements. Since the analysis
becomes involved with more than one resistive component, we assume one dominant resistive
component in the initial calculations. Afterwards, the results may be slightly adjusted to
correct for less significant losses.

Case 1 in load resistor placement

If the only ohmic component resides across the transistor output terminals the
equivalent diagram becomes the one shown in Fig.18. The conductances are

(43)

The real-part condition and, consequently, the frequency of oscillation remains unaltered from
the results in Eq.(41). The imaginary part condition now provides

(44)

i.e. is an expression for the size of the transconductance that the transistor must approach
when oscillations build up to the desired amplitude. Including the oscillation frequency from
Eq.(41), we get

(45)

Fig.18 Equivalent circuit for the oscillator loop gain with simplifying assumptions, no
transistor feed-back and resistive loading across the transistor output terminals only.

J.Vidkjær
20 Oscillators

To find the Q-factor of the open-loop gain, direct insertion of Eq.(43) into Eq.(30) gives

(46)

(47)

At the oscillation frequency ω=ω0 this angle should vanish and direct substitution from
Eq.(41) reveals that the expression in brackets becomes zero. The derivative with respect to
angular frequency at ω0 now develops

(48)

and the corresponding Q-factor becomes

(49)

Here we have demonstrated a direct approach to the Q-factor calculation. It may seem
cumbersome, but it has the virtue that we don’t need to identify and refer all ohmic losses to
a particular resonance circuit. In the present simplified case with only one resistor we may,
however, easily check the result as we have already seen, that the oscillating frequency is set

Fig.19 Parallel to series to parallel conversion used for estimating the Q-factor in the
oscillator from Fig.18.

by resonance between inductor LC and the series connection of Ca and Cb. Since Ra appear
across Ca, its equivalent series resistance Rser in Fig.19(b) is determined by this capacitor, but
it is also the total series loss in resonance with the series connected capacitors. By parallel-to-
series transformations according to Eq.(III-59), we get

J.Vidkjær
VI-2 Oscillator Circuits 21

(50)

i.e. the same result as before. Clearly we would also get this result - which is included here
for future reference - by transforming to parallel form like Fig.19(c), where

(51)

The voltage gain around the transistor and the transistor load are calculated by Eqs.(37), (38),
and (45) to yield

(52)

Case 2 in load resistor placement

The second simplified case we consider has resistive loading across the transistor
input terminalsas seen in Fig.20. Here the simplified analysis is based on the conductances,

(53)

Again, the real-part condition and the frequency of oscillation remain unaltered from the
results in Eq.(41). The imaginary part condition provides

Fig.20 Equivalent circuit for the oscillator loop gain with simplifying assumptions, no
transistor feed-back and resistive loading across the transistor input terminals only.

J.Vidkjær
22 Oscillators

(54)

Compared with Fig.19(a), a-terms and b-terms have changed roles, so in analogy to Eq.(50),
the Q-factor with resistive losses across the transistor input port becomes

(55)

Finally, the voltage gain and the load at the frequency of oscillation from Eqs.(37) and (38)
simplify to

(56)

(57)

In contrast to the previous case, gain and load impedance now get small imaginary compo-
nents as a consequence of the reactive separation between the transistor transconductance and
the ohmic load. However, the higher Rb, the higher Q, and the less significant become the
imaginary components.

Case 3 in load resistor placement

The final simplified case to consider is the one where the load and losses connect
across the inductance LC as indicated by Fig.21. There are no components from the transistor
equivalent circuit in this position, so the conductances are given by

(58)

In this case, the real part and the imaginary part conditions from Eqs.(33) and (34) are no
longer independent, since

(59)

The last equation is similar to

J.Vidkjær
VI-2 Oscillator Circuits 23

Fig.21 Equivalent circuit for the oscillator loop gain with simplifying assumptions, no
transistor feed-back and resistive loading across the inductor only.

(60)

which - when inserted in the real part condition - provides

(61)

Here we have used the resonance frequency ω00 defined in Eq.(41) and the quality factor QL
for the inductor at resonance ,

(62)

It is seen that the larger QL, the closer comes the oscillator frequency to the resonance
frequency ω00. Since the only resistive loss is provided by RC, the inductor Q-factor QL
becomes also the Q-factor for the oscillator. Inserting the oscillator frequency into Eq.(60)
gives the necessary transconductance,

(63)

The voltage gain and the transistor load become

(64)

(65)

J.Vidkjær
24 Oscillators

Fig.22 Oscillator equivalent circuit. Parallel resistance Rp accounts for all resistive losses
in the circuit.

Arbitrary resistor placements

A unified approach to the three cases above can be achieved through Fig.22, where
resonance circuit and capacitive transformer properties of the oscillator are emphasized. The
transformation ratio is, cf. sec.II-6,

(66)

With high Q-factors, the reactive currents in LC and the series connection of Ca and Cb at
resonance are large compared to the loss currents through Rp or, equivalently the transistor
load, Ztr0, which may be expressed

(67)

Accordingly, the capacitive current Ica is nearly equal to Icb, and the voltage ratio across the
two capacitors becomes inversely proportional to the capacitances,

(68)

To maintain oscillations the loop gain must be one so voltage Vb1 across Cb1 in the diagram
equals the control voltage for transconductance Gm. We get

(69)

where the last expression sets the requirement to Gm. With Rp set to RC it is seen that the Gm
we have just derived corresponds to the value in Eq.(63) as expected, since the diagram in
Fig.22 is a reorganization of the diagram in Fig.21. More noticeably is the observation that

J.Vidkjær
VI-2 Oscillator Circuits 25

if we replace Rp with the equivalent parallel resistor Rpar from Eq.(51) for losses that are
concentrated across the transistor output port, we get

(70)

which is the same as the Gm request in Eq.(45). Had we made a similar derivation from
Eq.(69) in the case where ohmic losses are concentrated across the transistor input terminal
like in Fig.20, we would have obtained the transconductance value in Eq.(54). Thus we
estimate Gm correctly from different loss components taken one by one if they are transformed
to the parallel equivalent resistance Rpar. This fact suggests that the joint effect upon Gm of
more losses from different locations in the circuit may be calculated from a parallel connection
of their equivalent parallel resistances. We may even use series-to-parallel transformation to
convert all loss contributions to one of the special case that were considered if it suits our
need for making other design calculation. Such approaches are much simpler than an attempt
to solve the corresponding real and imaginary part condition simultaneously. We shall
demonstrate them in the following two examples.

Example VI-2-1 ( Seiler type LC oscillator )

Fig.23 Seiler oscillator. With exception of current bias details, part (a) shows the circuit
diagram while part (b) shows the functional diagram that is used in the design.

The oscillator circuit in Fig.23(a) is a Seiler circuit where the transistor is current
biased, presently without showing details on how this is made in practice. The tuning capacitor
across the transistor output port is made of a series connection, CA1 and CA2, which trans-
forms the required load resistance RL to a larger equivalent transistor load resistance, Ra, as
indicated by the functional diagram in Fig.23(b). This way we get more suitable component
values and better utilization of the battery supply than a direct connection of the load to the
transistor would provide. Resistors Rb1 and Rb2 are base bias resistors and Ccp1 is a decoup-
ling capacitor. The main requirements to the circuit are

J.Vidkjær
26 Oscillators

- frequency, f0 = 300 MHz,

- output power, Pout = 2 mW in RL=50Ω,
- loaded Q-factor, Ql > 20,
- gain reduction at nominal output, Gm:gm ∼ 1:2.5

It is furthermore assumed that the transistor requires Vce > Vce,min=0.5V to prevent saturation
effects and that it has a forward current gain βf = 50, equivalently αf = 0.98.

We shall assume that the load Ra is the only resistive component in the oscillator
equivalent circuit, so the design corresponds to the first case that was considered above in
conjunction with the equivalent circuit in Fig.18. To find the equivalent load resistor Ra
assume that the transistor output voltage swing Va1 is half the battery voltage minus two times
the minimum voltage Vce,min, where the first count comes from transistor Q1, the second count
leaves room for a transistor in the bias current generator. With Va1 fixed, the corresponding
load resistor is calculated from the output power requirement,

(71)

(72)

According to Eq.(5-100) and Figure 5-44 in chapter 5, a 1:2.5 reduction in gain when
the oscillations grow up requires the final drive level xmax, which is normalized with respect
to the thermal voltage Vt=kT/q∼25mV, where

(73)

To find the capacitances Ca and Cb, Eqs.(50) and (52) provide

(74)

(75)

The tuning condition follows from Eq.(41), which gives

J.Vidkjær
VI-2 Oscillator Circuits 27

(76)

Impedance transformation from RL to Ra implies the transformation ratio

(77)

so the two capacitances that connect to Ca become

(78)

To keep the oscillations running, the final large signal conductance is given by Eq.(45),

(79)

The DC bias current IEE is now calculated from the current gain reduction of 1:2.5. We get

(80)

The final step in the design is to fix the base biasing and to implement the IEE current source.
In agreement with the initial emitter oscillator amplitude calculation in Eq.(71), the mean
emitter voltage Ve0 is given by

(81)

When oscillations grow up and the transistor is forced into class C operation, the mean base
emitter voltage will be slightly reduced from its initial DC value, which commonly is in the
range of Vbe0=0.65V. We ignore this adjustment here since the effect is small with current
biasing, so the base DC-bias voltage should be Vb0=Vbe0+Ve0=2.4V. Choosing the common
DC current Ibr through Rb1 and Rb2 to be approximately ten times the base dc current, which
is Ib0∼IEE/βf=0.04mA, i.e. Ibr=0.4mA, the bias resistors become

J.Vidkjær
28 Oscillators

(82)

The decoupling capacitor Ccp1 should be chosen so its reactance at the oscillation frequency
is small compared to both the parallel connection of Rb1 and Rb2 and to the reactance of LC.
The last requirement is the strongest since Rb1 Rb2∼1.9kΩ while ω0LC=Ra/Q=20Ω. Following
the results from Eq.(76), we chose

(83)

Current biasing may ideally be implemented by a current mirror as sketched in Fig.24(a). If

transistors Q2 and Q3 are identical, Rb3 should be adjusted so that its current equals the bias
current IEE. Less ideally, current biasing may be provided simply by a large resistor as show
in Fig.24(b). With the DC currents and voltages from above, the resister should be

(84)

Although this value is significantly larger than the equivalent load resistor Ra, which REE is
connected across, the power loss at the oscillation frequency in REE cannot be ignored
compared to the output requirement. If this simpler solution is chosen, the design should be
adjusted to compensate for the additional loss.

Fig.24 Current bias methods for the Seiler oscillator using, (a), a current mirror or, (b), a
large resistor REE.

Example VI-2-1 end

J.Vidkjær
VI-2 Oscillator Circuits 29

Example VI-2-2 ( Pierce type LC oscillator )

Fig.25 Pierce oscillator. The transistor is driven into class C operation to provide a self-
limiting gain.

Fig.25 shows the complete diagram for a Pierce oscillator build around a bipolar
transistor. The emitter is biased through RE, capacitor Ccp1 is a decoupling capacitor, to
provide self-limiting gain. The oscillating frequency is determined by LC, CA, and CB. Ccp2,
Ccp3 are coupling capacitors, and Lrfc is a RF choke, i.e. it has little DC resistance for bias
but provides a large impedance at the oscillating frequency. Finally, resistors Rb1 and Rb2
provide the base biasing.

The Pierce oscillator above should be designed to meet the following requirements,
- frequency, f0 = 10 MHz,
- output Va1= 1.6 Vp in RA = 600 Ω,
- 2nd and higher harmonic output components below -36dBc each,
- approximate gain reduction at nominal output Gm/gm ∼ 1:3.
Furthermore, it is assumed that inductor LC has a Q-factor of 150 or more and that the
transistor has current gain β=60 or, correspondingly, αf=0.98. Vce should be kept above
Vce,min = .6 V to prevent distortion by saturation effects in the transistor.

We know from the forgoing chapter on nonlinear amplifiers, that the class C operated
amplifier stage in Fig.25, which includes a bypassed emitter resistor, cannot be solved
analytically from the set of specifications at hand. Therefore, the design strategy is to start
with a few initial assumption, which must be verified in the course of the work. The first one
concerns the driving level Vb1 at the transistor input. In the limit of RE→∞ it is seen in Figure
5-44 - using A→∞ - that a relative input level reduction of 1:3 requires a relative input level
of x≈6, i.e.

(85)

J.Vidkjær
30 Oscillators

According to the discussion of the bipolar parallel tuned amplifier on page 19 in chapter 5,
we see from Figure 5-16 that for a given argument x - the second order modified Bessel
function is greater than third and any higher order functions. Since the harmonic components
in the collector current over the fundamental frequency component is the ratio of the cor-
responding ordered modified Bessel function over the 1st order function, the 2nd harmonic
component introduces the largest distortion and is the one to be used in the design. Using
Figure 5-16 we get

(86)

The fundamental frequency output voltage component corresponds to resonance in the

resonance circuit containing LC and the series connection of Ca and Cb. The collector load
impedance is a downscaled version of this circuit which at resonance is Ztr0. The correspond-
ing impedance at n times the resonance frequency is

(87)

where Ql is the total loaded Q-factor for the resonance circuit. Imposing the distortion
requirements for 2nd order harmonic components, we get

(88)

This Q-factor is considerably smaller than the Q-factor that represents losses in inductor LC,
the internal losses, and load resistor RA is supposed to contribute most significantly to the
total loss in the circuit. Other contributions to the internal losses come from the transistor
input impedance and biasing networks, so we design with an internal Q-factor that is some-
what lower than QL, say Qint=120, and verify that we stay within specifications afterwards.
As RA is still the most significant loss, it is natural to use our first special case as foundation
for the design, i.e. use the circuit in Fig.18 and the pertinent equations. However, we should
first modify the load to a resistor RA’ that incorporates the internal losses. The latter are
represented by a resistor Rint across RA. Following the terminology that was introduced in
section II-4, the Q-factor corresponding to RA may be referred to as the external Q-factor,
Qext. Due to the parallel connections and the proportionality between Q-factors and parallel
resistors we have

J.Vidkjær
VI-2 Oscillator Circuits 31

(89)

The equivalent intrinsic loss resistance to be parallelled across RA has the size

(90)

Using Eqs.(50) and (52) we get

(91)

The corresponding inductor is given by

(92)

When the circuit oscillates with the assumed output voltage swing of 1.6 Vp, the correspond-
ing fundamental frequency collector current component is

(93)

With a relative driving level of x=6, the corresponding collector and emitter DC currents with
ongoing oscillations are

(94)

J.Vidkjær
32 Oscillators

The difference between the supply voltage and the DC voltage across RE must leave room for
the output swing and the minimum collector-emitter voltage. In order to meet the assumption
of RE→∞, we choose RE as large as possible and get

(95)

The large signal fundamental frequency transconductance and transistor input impedance
become

(96)

The gain reduction factor of three leds to the initial assumption of a nearly constant transistor
bias current. We may check the assumption from the small-signal transconductance gm, the
corresponding DC-current Ic00, and the design constant A, which guided our decision on basis
of Figure 5-44. We get

(97)

Compared to the figure and to Eq.(94) it is seen, that the results are consistent with the
assumptions. To complete the bias design we assume that the base-emitter voltage under small
signal conditions is Vb00=0.65 V. As discussed in section 5-3 on page 5.53, the voltage is
reduced by an amount ΔVb0 given through

(98)

Thus, the base should be biased to

(99)

The DC base-current under oscillations is Ic0/β=1.997/60=0.033 mA. We let the current

through the bias resistors be more than ten times as large, namely Ibr= 0.5mA. Then the
current through Rb2 is approximately the same as the current through Rb1, and we get

J.Vidkjær
VI-2 Oscillator Circuits 33

(100)

The parallel connection of Rb1, Rb2, and the large signal input resistance Rπ1 from Eq.(96)
contributes to Q-factor. Since they are placed across capacitor Cb, we get

(101)

This is a high Q-factor compared to the assumed value of Qint = 120, so the transistor input
and biasing resistors contribute nothing to the losses compared with the inductor loss. The
parallel loss Rint comes from tuning inductor LC has QL=150, and a corresponding parallel
resistance,

(102)

so within the design above, there is room for a loss introduced by the RF choke Lrcf. Its
parallel resistance may go down to

(103)

To complete the design, coupling capacitor Ccp2 should not contribute to the tuning so it must
be large compared Cser from Eq.(92), say Ccp2=100nF. The reactance of coupling capacitor
Ccp3 should be small compared to the load resistor RA, i.e.

(104)

Finally Ccp1 must bypass RE. Since the time constant of τE=RECcp1 directly influence the gain
regulation in Gm, the capacitance should not be chosen too high in order to avoid the squeg-
ging type of amplitude instability. According to the discussion in Appendix VI-A, a safe
choise is to set the time constant τE equal to the logarithmic decrement of the resonance
circuit. Thereby we get

J.Vidkjær
34 Oscillators

(105)

Capacitor Ccp1 controls the time constant in the gain control of the transistor under large
signal operations in a manner that can be foreseen. However, the two coupling capacitors Ccp2
and Ccp3 might also contributes to this control, but it is difficult to work out the details
analytically. The values that were calculated formerly did not incorporate any stability con-
cerns, so we must be prepared to adjust the capacitor values when the circuit is build in
practice or simulated to check performance. The latter is done in Fig.26, and there are no
indications of amplitude instabilities in this design.

Fig.26 Build-up of oscillations in the Pierce oscillator from Fig.25 using the component
values that were calculated in the example.

Example VI-2-2 end

J.Vidkjær
 35

VI-3 Crystal Oscillators

Quartz is a piezo-electrical material where mechanical deformations cause voltage

differences across distinct surfaces and visa versa. A small sheet of quartz exhibits mechanical
resonant modes that depend on its geometry, the orientation of the cut compared to the crystal
axes, and the type of vibration engaged like compression, torsion, or shear oscillations. With
two electrodes evaporated on the crystal, we get a component that typically has an impedance
pattern as sketched in Fig.27. It shows pronounced resonances at the fundamental frequency
and odd numbered overtones. Even numbered mechanical modes make no resultant displace-
ment of the electrodes and cannot be exited electrically. Between dominant modes spurious
resonances from other type of vibrations may be observed. Mechanical resonances have
generally high Q-values and combined with excellent temperature stability and long-term aging
properties, the quartz crystal is widely used as the frequency controlling element in oscillators
that operates within close tolerances in the frequency range of .1 to approximately 350 MHz.

Fig.27 Impedance magnitude of quartz crystal with strong resonances at the fundamental
mode and odd overtones. In between spurious resonances from other types of
vibrations may be observed.

Crystal Equivalent Circuits

From an electrical point of view, the quartz crystal may be represented by the
equivalent circuit in Fig.28(a), where the capacitor C0 is the capacitance between the elec-
trodes if there were no mechanical resonances. Each series resonant circuit Ci, Li, ri, represent
a mechanical mode that may be excited by the piezo-electric effect. Commonly crystals are

Fig.28 Electrical equivalent circuit of a quarts crystal. (a) includes all modes and (b)
suffices for narrowband calculations with a particular mode. (c) is a redrawing of
(b) emphasizing its parallel resonance.

J.Vidkjær
36 Oscillators

optimized with respect to a particular mode that is recommended for circuit design. In the
frequency range around the selected mode, it suffices to consider an equivalent circuit
composed of only the electrode capacitance and the series circuit associated with the mode
in question. This circuit is shown in Fig.28(b) and has the impedance,

(106)

The last expression introduces the series resonance frequency ω0 and the corresponding Q-
factor of the series circuit,

(107)

Setting the numerator and the denominator in Eq.(106) equal to zero and solving for s, we get
the impedance zeros and poles,

(108)

Practical data justify the approximations, which imply large Q-factors. Extremely high Q´s are
obtainable with crystals - up to 100000 - so the small losses get no influence on the imaginary
parts of poles and zeros. Moreover, the electrode capacitance is much greater than any
resonator capacitance and we may use the estimation

(109)

in the last expression, which determines a parallel resonance frequency - sometimes also called
the antiresonance frequency - ωp, where

(110)

The series resonance is responsible for the zero of the impedance function. In conjunction with
the electrode capacitance Co it also provides the pole that implies the parallel resonance,
which may be viewed upon like a parallel resonant circuit with capacitive transformation as
shown in Fig.28(c).

J.Vidkjær
VI-3 Crystal Oscillators 37

Fig.29 Impedance in magnitude, phase and in real, imaginary forms for the crystal example
in Eq.(111). Note the logarithmic impedance scales and the sign indications for the
imaginary part of the impedance.

(111)

Due to the existence of both series and parallel resonances there may be two ways
of utilizing a crystal in an oscillator, either as a low impedance devices in series resonance
around ω0 or as a high, inductive impedance device in parallel resonance up to ωp. The
possibilities are readily seen in Fig.29, which shows a calculation of the impedance function
from Eq.(106) based on the crystal data in Eq.(111). With practical data, the series and parallel
resonances are closely spaced in frequency. In this case, however, the distance is still large
compared to the real part of poles and zeros, so the two resonances are clearly distinguishable.
The separation may be expressed through the following figure of merit M,

J.Vidkjær
38 Oscillators

Fig.30 Pole-zero pattern for a single resonant mode in the crystal. The encircled region is
shown in (b) for series and (c) for parallel resonances and indicate phasor assump-
tions in narrowband approximations.

(112)

where the distance ωp-ω0 is given through Eq.(110).

The two types of resonances in Fig.29 show both the steep phase characteristic that
is required for good frequency stability. We chose between them in design by letting the rest
of the circuit meet oscillation criteria with either high or low impedance levels. Crystals
operated in fundamental mode have usually high M values. Therefore, the impedance may be
narrowband approximated around either ω0 or ωp using the method that was applied to LC
tuned circuits in section II-3. With known poles and zeros the crystal impedance is expressed

(113)

The pole-zero pattern for the crystal is shown in Fig.30(a) while Fig.30(b) and (c) indicate the
geometry to be used in the narrowband approximation. Notice that the figures are not drawn
to scale as the pole and zero distances to the imaginary axis should be much smaller compared
to their mutual spacings. With the assumptions in the figures we get

J.Vidkjær
VI-3 Crystal Oscillators 39

(114)
It is seen that the two narrowband approximations in (a) and (b) are of the same form as the
approximations that were derived for simple resonance circuits in section II-32. The two
resonance frequencies ω0 and ωp are so close that they are taken equal unless we are consider-
ing the deviation Δω from the either the series or the parallel resonance frequency.

The series resonance in Eq.(114)a is independent of the electrode capacitance C0. In

parallel resonance the electrode capacitance determines the equivalent parallel resistance,
which is the numerator in Eq.(114)b,

(115)

where the numerical example corresponds to the crystal data in (111). This resistance value
is recognized as the peak impedance in Fig.29. The final expressions in Eq.(114) may also be
represented by narrowband equivalent circuits given by Fig.30. In series mode the crystal is
represented by a small resistance in series with a resonant reactance, in parallel mode by a
large resistor in parallel to a resonant susceptance.

Fig.31 Narrowband equivalents of the crystal in series (a) and parallel (c) resonance
modes.

2) The validity of the narrowband approximation for the equivalent circuit in Fig.28b
is implicitly the reason why all parallel resonators in Fig.28a excepts the one selected
may be disregarded. They contribute only canceling pole-zero pairs if they are
incorporated in Eq.(113).

J.Vidkjær
40 Oscillators

Fig.32 Impedance magnitude and phase of a low-M crystal before and after inductive
compensation. Three cases are shown, 1) Lp set exactly to, 2) 10% over, and 3)
10% below the ideal value.

(116)

Overtone crystals have commonly smaller relative distances between the series and
parallel resonances - the higher the mode number, the smaller the distance. This is exemplified
by ninth overtone crystal data in (115), where the figure of merit M from Eq.(112) even falls
below one, so the zero and pole separation along the imaginary axis is smaller than the
distance to the imaginary axis. With M approaching zero the contributions from all resonance
pole-zero pairs to the crystal impedance through Eq.(113) tend to cancel each other, so all
what will be left is the shunting impedance of the electrode capacitance. With exception of
modest deviations closely around f0, the shunting effect of C0 dominates the patterns in
Fig.32(a), which show simulations based on data in the example. According to Eq.(112), M
may also be interpreted as the ratio of the electrode impedance over the series circuit imped-
ance at ω0. To get a distinct frequency response that may serve for oscillator design with a
low M value crystal, the effect of the electrode capacitance must be compensated. One
obvious way to do so is to neutralize C0 by a parallel inductance Lp as shown in Fig.33. If
the tuning is exact,

(117)

J.Vidkjær
VI-3 Crystal Oscillators 41

Fig.33 Compensation of electrode capacitance by a parallel inductance in a low-M crystal.

only the series resonant circuit remains and may be used as an equivalent circuit around ω0.
However, a less precise compensation still works if the impedance of C0 paralleled by Lp is
large compared to ri. Then the series resonance pattern - distinct magnitude minimum and
steep rise in phase - prevails as seen in the characteristics of Fig.32(b), where all examples
coincide around f0.

Loading and Driving Crystals

To benefit from crystals in oscillator design, it is important to keep the Q-factor high.
Compared to the Q-factor of the crystal itself, which in the present context often is referred
to as the unloaded Q, some reductions are unavoidable when the crystal is connected to the
remaining part of the oscillator circuit, which contains transistor impedances, biasing networks,
and the external load. If the losses transform to a series resistor rL in a crystal that is used in
series mode, Fig.34(a), the resultant Q-factor is the one we get by adding the external resistor
to the internal resistor in Eq.(107), i.e.

(118)

Correspondingly, if the external losses may be represented by a parallel resistance RL to a

crystal that is operated in parallel mode, the resultant Q-factor is the one we get from parallel-
ling the external resistance to the parallel resistance Rp, Eq.(115),

(119)

Fig.34 Resistive reduction of the Q-factors in series (a) and parallel (b) resonance modes.

Reactive loading of the crystals may change both resonance frequencies and Q-
factors. In some situations this is an unavoidable imperfection but adding a capacitance in

J.Vidkjær
42 Oscillators

Fig.35 Equivalent representation of a capacitance in parallel with a crystal.

series or parallel to a crystal may also be a mean for external adjustments of the resonance
frequencies. The simplest reactive loading situation is the one where a capacitor Cp is
connected in parallel to the crystal because Cp simply adds to the electrode capacitance C0.
All we have to do is to replace C0 in the initial developments from Eqs.(106) to (110) with
C0+Cp. Thus, the new parallel resonance frequency ωpp becomes

(120)

As seen, ωpp is lowered from its unloaded value in Eq.(110) towards the series resonance
frequency ω0, which is approached in the theoretical limit of Cp → ∞. In this context, the
frequency interval fp - fo is called the pulling range of the crystal. The range cannot be fully
exploited as the figure of merit in Eq.(112), with the effect of Cp included, must stay well
above 1 to keep the parallel resonance defined. Electrode capacitance C0 had no influence on
the real part of the poles and zeros. This will neither be the case when Cp is added, so the Q-
factor remains unaffected of a parallel capacitor. However, the resultant parallel resistance is
changed and given by Eq.(115) using C0+Cp instead of C0. The crystal and capacitance in
parallel connection gets the equivalent representation in Fig.35, where

(121)

Another view upon the parallel connection of a crystal and a capacitor is illuminated
by Fig.36. It takes outset in the common way of employing a parallel resonant crystals as
replacements for the coil in oscillator circuits since the crystal is inductive between fp and fo.
Starting from a LC oscillator designed to operate at frequency ωpp with a tuning capacitance

Fig.36 Series resistance evaluation for a crystal used in parallel mode as a coil replace-
ment.

J.Vidkjær
VI-3 Crystal Oscillators 43

equal to Cp and susceptance Bp, the canceling negative susceptance -Bp from the inductance
may instead be provided by the crystal to get improved stability3. Although the crystal is
utilized in its parallel mode, the resonance circuit in the oscillator may be of either series or
parallel form. In the first case, losses are represented by parallel resistance Rpp from Eq.(121).
With series circuits it may be more convenient to consider series components as indicated in
Fig.36(c). Using the series-to-parallel conversions, the equivalent series resistance is given by

(122)

If a capacitance Cs is placed in series with the crystal that is not inductively compen-
sated, Fig.37(a), we get the impedance function

(123)

Compared with the impedance Zc of the crystal itself from Eq.(106), it is seen that the
denominator will leave the poles unaffected. The new zeros become

(124)

There are no changes in the real part of the zeros due to the series capacitance Cs so Q is
unchanged. The imaginary part, which determines the series resonance frequency, is enlarged,
but it is seen that the pulling range is again the interval between frequencies fp and fo,

(125)

3) Crystals may be optimized and specified for parallel usage, given the resonance
frequency fpp and the required parallel capacitance. The latter is commonly called the
load capacitance of the crystal in data sheets.

J.Vidkjær
44 Oscillators

Fig.37 Equivalent circuits for a capacitor and a crystal in series connection. The component
values are defined in Eqs.(126) and (128).

With a series capacitor it is possible to control the series resonance frequency where the
crystal has minimum impedance. Exactly how low it is may be crucial for setting up the gain
requirement in an oscillator. To find the series resistance we use the following rewriting,

(126)

Direct substitution of these relations into the last part of Eq.(123) yields

(127)

This expression has exactly the same structure as the crystal impedance Zc in Eq.(106), so the
series connection may be represented equivalent circuits having the same topology as the
crystal without the series capacitor, Fig.37(b) and (c), but clearly with different component
values. In particular the new series resistor at resonance becomes

(128)

Crystal Oscillator Circuits

Like the situation with LC oscillators, the number of circuits that are proposed for
crystal oscillators is overwhelming, so again we can only consider a few types and examples.
For a more complete presentation, which also includes temperature compensation techniques
to meet tight tolerances, you should consult the specialized literature, for instant ref. [8].

J.Vidkjær
VI-3 Crystal Oscillators 45

Fig.38 Colpitts (a), Pierce (b), and Seiler (c) oscillators where crystals have replaced the
inductors of the LC circuits. The optional series capacitor may be inserted to tune
frequency and/or transform impedance levels.

To get an idea on how crystals are used in oscillators, the examples in Fig.38 show
the basic configuration that was introduced for LC oscillators in Fig.17, but now inductors are
replaced by crystals. Here the crystals are used as high impedance, inductive devices that must
tune to the capacitances at a frequency between the parallel and the series resonance of the
crystal. To adjust frequency or impedance level or both, we may insert a capacitor Cs in series
with the crystal as indicated in the figures. In these cases the crystal equivalent from Fig.37
and the related expressions may be used for the design.

Closely around series resonance, the crystal has a low impedance and may, therefore,
be used to make a highly frequency selective closing of the feed-back loop in the oscillator.
This is demonstrated in the examples of Fig.39, where the function of the crystal is to
establish connection to the transistor output in some of the LC oscillator circuits we consid-
ered above. At a first glance we may suppose the LC resonance circuit is superfluous in these
examples. However, we saw initially that a crystal has a fundamental mode and several odd-
numbered overtones, each presenting an impedance minimum. The LC circuit, which should
resonate around the desired oscillator frequency, gets the role of selecting the correct mode
of the crystal.

Fig.39 Colpitts (a) and Seiler (b) oscillators using crystals in low-impedance series modes
to close the feed-back loop directly.

J.Vidkjær
46 Oscillators

Fig.40 Colpitts (a), Pierce (b), and Seiler (c) oscillators where crystals are used in low
impedance series mode to make highly frequency selective component connections.

Other ways of using a crystal in low impedance mode around the series resistance are
demonstrated by Fig.40. The function of the circuit example in Fig.40(a) is that the crystal
makes a frequency selective grounding of the base/gate terminal. In a narrow frequency band
this establish the necessary gain in a Colpitts oscillator, where it is supposed that the active
element operates in grounded base or gate configuration. In the two other examples, the crystal
connects the inductor and compensates its inductance to provide oscillations close to the series
resonance of the crystal. This technique is often referred to as impedance inversion since we
use the crystal as a low impedance device in conjunction with an external inductor to establish
a high inductive impedance. The method is often used in conjunction with overtone crystals
since - according to the discussion in the example of Fig.32 - high impedance parallel modes
may give difficulties in practical use.

Example VI-3-1 ( IC-connected crystal oscillators )

A demonstration of crystal oscillator structures is given in Fig.41. It shows a block

diagram of an IC circuit, which contains all active circuits that are required in a narrowband
FM receiver. The circuit is intended for double conversion so it uses two intermediate
frequencies and two local oscillators are required. They are denoted 1st LO and 2nd LO
respectively where the first one uses impedance inversion of the type in Fig.40(c) and last one
is of the type in Fig.38(c) without the series capacitance.

J.Vidkjær
VI-3 Crystal Oscillators 47

These devices can be used as stand±alone VHF receivers or as the lower

IF of a triple conversion system. Applications include cordless telephones,
short range data links, walkie±talkies, low cost land mobile, amateur radio
The MC13135/MC13136 are the second generation of single chip, dual receivers, baby monitors and scanners.
conversion FM communications receivers developed by Motorola. Major
improvements in signal handling, RSSI and first oscillator operation have Complete Dual Conversion FM Receiver ± Antenna to Audio Output
been made. In addition, recovered audio distortion and audio drive have Input Frequency Range ± 200 MHz
improved. Using Motorola's MOSAICΕ 1.5 process, these receivers offer Voltage Buffered RSSI with 70 dB of Usable Range
low noise, high gain and stability over a wide operating voltage range.
Low Voltage Operation ± 2.0 to 6.0 Vdc (2 Cell NiCad Supply)
Both the MC13135 and MC13136 include a Colpitts oscillator, VCO tuning
diode, low noise first and second mixer and LO, high gain limiting IF, and Low Current Drain ± 3.5 mA Typ
RSSI. The MC13135 is designed for use with an LC quadrature detector and Low Impedance Audio Output < 25 Ω
has an uncommitted op amp that can be used either for an RSSI buffer or as VHF Colpitts First LO for Crystal or VCO Operation
a data comparator. The MC13136 can be used with either a ceramic
discriminator or an LC quad coil and the op amp is internally connected for a Isolated Tuning Diode
voltage buffered RSSI output. Buffered First LO Output to Drive CMOS PLL Synthesizer

MC13135 MC13136

MC13136
VCC Figure 20.
1.0 μH 1st LO 24
2200 p Varicap
+ 1
1.0 23
27 p
1.0 k 39 MHz 2
0.001
Xtal 5.0 p 22
62 pF RF Input
Buffered LO 3
21 0.2 μH 150 pF 50 Ω Source
Output VCC1
0.01 5.0 k 0.01
4
0.1 20
5
Ceramic
120 p 2nd LO Filter
50 p VCC2
6 10.7 MHz
19
10.245 MHz 7 0.1
Xtal
18 360
Ceramic 8
Filter
455 kHz
9
AF
17 1.0 k
Recovered
Demod Audio
10 0.15
Limiter
0.1 16 10 k
0.1 11

15 RSSI
Output
12 14
270 p
0.1 13
2.7 k muRata
455 kHz
Resonator
CDB455C34

Figure 20a. Single Channel Narrowband FM Receiver at 49.7 MHz

Fig.41 Extracts from Motorola’s online IC selection guide. Capacitances are in μF unless
it is specified otherwise

Example VI-3-1 end

J.Vidkjær
48 Oscillators

J.Vidkjær
 49

APPENDIX VI-A Oscillator Amplitude Stability

To present and investigate the amplitude stability problem in the sinusoidal prototype
oscillator from Fig.3 we consider the block-diagram in Fig.42(a). Here, the transistor is charac-
terized by the relationships between DC and fundamental frequency input voltages and the
corresponding collector or emitter current components. The relations were formerly expressed
by Eqs.(5-40) to (5-46). In summary the expressions that are needed below read,

(129)

The hatted quantities Î1 and Î0 are the modified Bessel functions of order one and zero
respectively and αf=1-1/βf is the common base current gain, which is close to but slightly
below one. The block representing the tuned circuit in the diagram relates fundamental
frequency amplitudes. According to sect.II-1, the voltage amplitude in response to a current
amplitude at resonance is expressed in Laplacian terms by a single time constant

(130)

The DC biasing relationship is expressed through the impedance in series with the emitter,
which is also a single time constant relationship

(131)

While τQ represents the maximum rate of amplitude variations across the resonance circuit,
τE sets the rate by which the oscillator circuit attempt to control this amplitude. To succeed,
the two time constant must be comparable in size. If the amplitude control is too slow the
amplitude will vary periodically in time as illustrated in Fig.9(b). If, on the other hand, τE is

Fig.42 Block-diagram representation of oscillator amplitude control. To conduct a dynamic

simulation, the two generic blocks in (a) must be replaced by the integration blocks
in (b) and (c).

J.Vidkjær
50 Oscillators

too short the emitter decoupling becomes insufficient and the circuit may either start oscillat-
ing with low amplitude at a wrong frequency or not start at all. As will be discussed below,
a good design choice is to set the two time constants equal, like it was done to get the smooth
amplitude start-up in Fig.9(a).

The time constant expressions in Eqs.(130),(131) are s-domain versions of the time-
domain relationships that are shown in Fig.42(b) and (c), which are necessary to conduct
calculations on how the oscillator amplitude evolves in time. With two integrators, the
problem to be solved is a second order differential equation. It is nonlinear due to the
nonlinear relationships in the transistor description from Eq.(129) and no simple solution
exists. We shall, however, assume that such a solution with amplitude Vb1 has build up to
formal stationarity where Barkhausen’s criterion applies,

(132)

If this solution is stable it is supposed that the oscillator amplitude remain stable when it is
disturbed by small perturbations δVb1. The question of amplitude stability is discussed from
that point of view.

Linearizing around a given solution to find the current variation, which corresponds
to an amplitude perturbation ΔIc1= GΔ11δVb1, may be represented by the block-scheme in
Fig.43. Here, linearized transconductances G11 through G00 depend on the actual large signal
voltages Vb1 and Vb0, and the block-scheme or the flow-graph provide directly

(133)

Fig.43 Linearization of the block-scheme from Fig.42 and the corresponding signal flow-
graph. Both describe how small amplitude variations develop in oscillations .

J.Vidkjær
VI-A - Appendix - Oscillator Amplitude Stability 51

Introducing the determinant

(134)

a reorganization including collection of terms in equal powers in s gives

(135)

It is supposed that the response to an enforced amplitude pulse δVb1 should be stable and
approach zero if the underlying large-signal oscillation is stable. The last condition is fulfilled
if the transfer function is stable, since in that case we have,

(136)

Stability requires that the poles of the transfer function are in the left half-plane only, a
condition that is met if all three coefficients in the second order denominator polynomial have
equal and - following the second order term - positive signs.

To illuminate the role of the constant term C in the denominator of Eq.(135), which
must be positive, we consider small stationary perturbations of the driving signal in the
oscillator amplifier if no feedback for oscillation was established. The corresponding block-
scheme and signal flow-graph are given in Fig.44, from which we derive the transfer function,

Fig.44 Linearized block-scheme and corresponding signal flow-graph for the oscillator
amplifier without the upper oscillation feedback loop that is present in Fig.43.

J.Vidkjær
52 Oscillators

Fig.45 Stationary, large signal loop-gain response. Barkhausen’s criterion is met at the unit
slope line.

(137)

This ratio must be less than one in a self-limiting amplifier. It is the crossing slope of the
stationary loop response curve at the stabilized amplitude, Vb1,osc, where the curve moves
from the greater than one loop gain region to the smaller then one region as sketched in
Fig.45. The condition is expressed

(138)

The left hand side of this inequality is recognized as the constant denominator term. We have
seen that it is positive if the amplifier is adequate for oscillator applications.

Investigating the first order coefficient B in the s term of the denominator from
Eq.(135), it is useful to introduce the time constant ratio

(139)

The condition to be fulfilled is written

(140)

a requirement that sets an upper bound on the time constant ratio

(141)

J.Vidkjær
VI-A - Appendix - Oscillator Amplitude Stability 53

To be more specific in calculations, the partial derivatives of the current components

and some of their interrelationships are required. From Eq.(129) we get directly

(142)

Comparisons between Eq.(142) and Eq.(5-95) show, that the effective fundamental frequency
transconductance in oscillations is related to G10 through,

(143)

Therefore, fulfillment of Barkhausen´s condition links derivatives and circuit components by

(144)

Using the modified Bessel function identities, [9] 9.6.27,

(145)

the complete set of partial derivatives is expressed

(146)

Finally, the determinant of large signal transconductances becomes

(147)

This quantity is positive for all positive arguments x.

J.Vidkjær
54 Oscillators

Positiveness of the determinant ΔG implies that the time constant condition in

Eq.(141) always will be met choosing Kτ=1 if the inequality in Eq.(138) applies. Then the
underlying requirement from Eq.(140) reads

(148)

Thus, setting the two time constants equal to each other gives a stable amplitude if the
amplifier is self-limiting so all polynomials coefficients in the denominator from Eq.(135) are
positive.

To gain more insight on circuit parameters, it is necessary to make the simplifying

assumption, that the amplifier is driven strongly in class C. The larger argument x, the smaller
relative difference between the modified Bessel functions, and we may estimate

(149)

One consequence of these assumptions is, that the determinant in of the derivatives looses
significance, so the denominator coefficient of Eq.(135) are approximated,

(150)

In the simplified version, the stability limit corresponding to Eq.(141) becomes

(151)

Substituting for the time constants from Eqs.(130),(131) sets directly a bound on the decouplig
capacitor CE in terms of the tuning capacitor C

(152)

so to keep the oscillator free of squegging, the decoupling capacitor must be carefully selected
and not just chosen from an "infinitely large" way of thinking.

Example VI-A-1 ( amplitude stability )

By this example we shall employ the self-limiting class-C amplifier design from
Example 5-3-1 in a 10 MHz prototype oscillator. Formerly the collector bias current was
supplied separately by a RF-choke, Lchk, but the tuning inductor may be used for that purpose
as well, as it is done by the circuit in Fig.46. Summarizing previous results, the amplifier has

J.Vidkjær
VI-A - Appendix - Oscillator Amplitude Stability 55

Fig.46 10 MHz prototype oscillator based on the class-C amplifier from Example 5-3-1.
Output waveforms were shown in Fig.9 for two settings of CE.
initial DC emitter current, transconductance, and voltage gain

(153)

With maximum input signal, Vb1=125 mVrms, a gain reduction by a factor of three was
required and the DC current, the transconductance, and the voltage gain change to

(154)

To build an oscillator around this amplifier we must select the winding ratio of the
feedback transformer so that Barkhausen’s criterion is met. It gives

(155)

Using the results from Eqs.(5-95) and (5-100), the partial derivatives for the transistor current
components evaluate to

(156)

J.Vidkjær
56 Oscillators

(157)

Now the stability criterion from Eq.(141) gives an upper bound for stable time-constant ratios,

(158)

This limit can be verified by simulations on both the amplitude control block-scheme in Fig.42
or by complete circuit simulations based on the diagram in Fig.46. Examples of the latter type
were shown previously in Fig.9, where the lower, unstable case were a little above the limit,
Kτ=4.5, in order to clearly demonstrate stationary oscillations. Deviating more severely from
the stability limit gives unstable output waveshapes that in the audio frequency range sounds
like the frogs squegging from which this type of instability was named. Fig.47 shows an
example where Kτ=10.

The time constants τQ was calculated in Example 5-3-1, and the maximum limit
becomes

(159)

Fig.47 Oscillator output for Kτ=10 showing the characteristic ’squegging’ waveform that
has named this type of instability.

J.Vidkjær
VI-A - Appendix - Oscillator Amplitude Stability 57

This result should be compared to the maximum value from the simplified analysis,

(160)

As seen, we are on the safe side, but it is remarkable, that this simple solution to a complicat-
ed problem works. Going behind the result, it is observed that the correctly evaluated deriva-
tives in Eq.(156) are in reasonable agreement with the simplifying assumptions from Eq.(150).

Example VI-A-1 end

Example VI-A-2 ( amplitude stability with current bias )

Fig.48 Simple feedback oscillator with current bias in common emitter configuration (a).
Current biasing is often established by a current mirror (b).
Current biasing is an ideal way to get the self-limiting amplification, which is
necessary for stabilizing the oscillator amplitude. The basic amplifier properties was discussed
earlier in section 5-3, and Fig.48 shows its inclusion in the transformer feed-back oscillator
that was discussed in the previous example. The large signal control of amplitude is now
governed by the describing function block diagram in Fig.49, and its linearized counterpart
is given by Fig.50. Compared to the similar schemes in the resistor biased case, Fig.42 and
Fig.43, we get the linearized current biased scheme below as the limit of RE approaching
infinity in Fig.43.

Since the amplitude stability analysis formerly was based on the linearization, stability
criteria for the present current biased oscillator are obtained from the resistor biased criteria
as the limit case of RE going to infinity. To do this, it is convenient to introduce a new time
constant τEE. Instead of τE from Eq.(131) we use

J.Vidkjær
58 Oscillators

Fig.49 Block-diagram representation of oscillator amplitude control in a current biased

oscillator like Fig.48.

(161)

This time constant may be related to the current bias, since by Eqs.(5-99) and (142) we have

(162)

In the limit of RE→∞, Eq.(135) reduces to

(163)

Fig.50 Linearization of the block-scheme from Fig.49 and the corresponding signal flow-
graph.

J.Vidkjær
VI-A - Appendix - Oscillator Amplitude Stability 59

To ensure stability, the B coefficient in the denominator must still be positive, and now the
criterion reads

(164)

Under large drive conditions, where the approximations in Eq.(149) apply, we get the same
upper bound on the decoupling capacitor CE as we had before in Eq.(152), i.e.

(165)

To demonstrate the usefulness of the criteria for current bias we consider the
oscillator circuit in Fig.48. It is organized to resemble the resistor biased oscillator from
Example VI-1-1 as close as possible, i.e. the normalized drive level xmax is maintained and
the current bias IEE is set equal to the former DC current from Eq.(154). Under such condi-
tions the large signal transconductance Gm remains unchanged and Barkhausen’s condition is
fulfilled to provide the same output voltage that we had before with unchanged load Rp and
winding ration N. Thereby, the stability bound becomes

Fig.51 Outputs from the oscillator circuit in Fig.48. The upper and lower curves are at the
stable and unstable side of the stability criterion with Kτ=0.1675 and 0.1875 respec-
tively.

J.Vidkjær
60 Oscillators

(166)

The two simulations below demonstrates the sharpness of the criterion by showing how the
oscillations build up with Kτ = .1675 and .1875 respectively. In the first case the amplitude
eventually stabilizes while in the second, spurious oscillations are sustained.

Example VI-A-2 end

J.Vidkjær
VI Problems 61

Problems

P.VI-1

Fig.52

The oscillator in Fig.52 should be designed for maximum output at 2MHz without
transistor saturation, i.e. keeping Vce≥0. The supply voltage is VCC=6V, the load
resistor is RL=800Ω. The inductance at the primary side of the transformer is
L=2.5μH and a Q-factor of QL=50. The amplifier should be designed to show a gain
reduction of 1:2.5 when oscillations build-up from initial small-signal to the required
amplitude. Assume that transistor βf is so high that the transistor input impedance
may be ignored and that the base emitter voltage is 0.6V.

- Calculate the bias current IET and the small-signal gain gmd.
- Find the transformation ratio N and capacitor C.
- Estimate the total harmonic distortion of the output voltage.

P.VI-2

Fig.53 shows a Colpitts oscillator circuit with a bipolar transistor. At the oscillating
frequency, the transistor base and the collector supply are AC-grounded through the
decoupling capacitors Ccp1 and Ccp2. Ccp3 is a coupling capacitor and resistors Rb1,
Rb2, and RE are bias resistors. Inductor Lrcf is a bias choke that may be considered
infinitely large in design. The tuning components are CA, CB, and LC. They are all
assumed lossless.

J.Vidkjær
62 Oscillators

Fig.53 Colpitts Oscillator Circuit

The circuit should be designed to meet the following requirements,

- frequency, f0 = 20 MHz,
- output voltage Vo = 1.5 Vrms in RL= 200Ω,
- loaded Q-factor, Ql > 25,
- gain reduction at nominal output, Gm:gm ∼ 1:3.5

It is supposed that the transistor requires Vce > Vce,min=0.5V to prevent saturation
effects, that it has a forward current gain βf = 80, and that its base-emitter voltage
in small-signal operation is 0.80V.

Calculate all component values of the circuit in Fig.53. Assume in the initial steps
of the design that the transistor is current biased and find the bias current IE. After-
wards, RE should be chosen as high as possible to approximate the current bias
conditions.

P.VI-3
Consider a crystal operated in series resonance with resonance frequency f0. Show
that the frequency where the impedance becomes real differs from f0 by

(167)

if the electrode capacitance C0 is accounted for in details. Explain why the frequency
difference is insignificant.

P.VI-4

Fig.54 shows a Pierce crystal oscillator. It is supposed that the crystal is the 10 MHz
crystal from Eq.(111) with a series capacitance of Cser=22 pF. The crystal and the
series capacitor replaces the inductance LC in the oscillator design from Example
VI-2-2 on page 29. All other components are kept unchanged.

J.Vidkjær
VI Problems 63

Fig.54
Calculate the output voltage, the oscillator frequency, and the loaded Q-factor of the
oscillator in Fig.54.

P.VI-5

Fig.55 shows a Pierce crystal oscillator. It is supposed that the crystal is the 10 MHz
crystal from Eq.(111) used in series, impedance inverting mode like Fig.40(b) with
inductance LC. The rest of the circuit is the oscillator design from Example VI-2-2
on page 29, and all components values are kept unchanged from the example.

Fig.55
Calculate the output voltage, the oscillator frequency, and the loaded Q-factor of the
oscillator in Fig.55.

J.Vidkjær
64 Oscillators

J.Vidkjær
 65

References and Further Reading

[1] N.M.Nguyen, R.G.Meyer,"Start-up and Frequency Stability in High-Frequency

Oscillators", IEEE Journ.Solid-State Circuits, Vol.SC-27, pp.810-820, May 1992.
Included in [2].

[2] B.Razavi,ed., Monolitic Phase-Locked Loops and Clock Recovery Circuits, Theory
and Design, IEEE Press 1996, Selection of 65 papers including approximately 10 on
oscillator design and properties.

[3] R.S.Elliott, An Introduction to Guided Waves and Microwave Circuits, Prentice-Hall

1993.

[4] G.Gonzalez, "Microwave Transistor Amplifiers" 2nd ed. Prentice-Hall 1997.

[5] D.B.Leeson, "A simple Model of Feedback Oscillator Noise Spectrum", Proc.IEEE,
Vol.54,Feb.1966,pp.329-333. Included in [2].

[6] E.Hafner, "The Effect of Noise in Oscillators", Proc.IEEE, vol.54, pp.179-198,

Feb.1966. Included in [2].

[7] A.van der Ziel, Noise in Solid State Devices and Circuits, Wiley 1986.

[8] M.E.Frerking, Crystal Oscillator Design and Temperature Compensation, Van-

Nostrand 1978.

[9] M.Abromowitz,I.A.Stegun, Handbook of Mathematical Functions, Dover,NY, 1965

J.Vidkjær
66 Oscillators

J.Vidkjær
 67

Index
Amplitude Noise . . . . . . . . . . . . . . . 13
Antiresonance . . . . . . . . . . . . . . . . . 36
Barkhausen’s criterion . . . . ....... 2
Class C Amplifier
BJT oscillator example . 25, 29, 54,
57
Colpitt’s Oscillator . . . . . . . . . . . . . . 18
Crystal
antiresonance . . . . . . . . . . . . . 36
equivalent circuits . . . . . . . . . . 35
figure of merit M . . . . . . . . . . . 37
parallel resonance . . . . . . . 36, 41
poles and zeros . . . . . . . . . 36, 43
series resonance . . . . . . . . . 36, 44
Crystal Oscillators . . . . . . . . . . . . . . 35
Frequency Stability Factor . . . . . . . . 9
Impedance Inversion . . . . . . . . . . . . 46
LC Oscillators . . . . . . . . . . . . . . . . . 15
Leeson’s oscillator noise model . . . . . 11
Oscillators . . . . . . . . . . . . . . . . . . . . 1
amplitude stability . . . . . . . 10, 49
Barkhausen’s criterion . . . . . . . 2
crystal . . . . . . . . . . . . . . . . . . 35
frequency stability . . . . . . . . . . 8
LC couplings . . . . . . . . . . . . . 15
negative conductance criterion . . 7
noise . . . . . . . . . . . . . . . . . . . 11
Parallel Resonance . . . . . . . . . . 15, 35
Phase Noise . . . . . . . . . . . . . . . . . . 13
Pierce Oscillator . . . . . . . . . . . . . . . 18
LC tuned . . . . . . . . . . . . . . . . 29
Seiler Oscillator . . . . . . . . . . . . . . . . 18
LC tuned . . . . . . . . . . . . . . . . 25
Tank-Circuit . . . . . . . . . . . . . . . . . . 3

J.Vidkjær