Excerpted from "The Designer's Guide to High-Purity Oscillators" by Hegazi, Rael, and Abidi.

For more information, go to www.designers-guide.org/Books.

Basics of LC

1 Oscillators

1 Introduction
Phase noise has been one of the most interesting yet poorly understood topics
in circuit design. The challenge of predicting the amount of phase noise in a
given circuit has been approached from a variety of angles, from using a heu-
ristic model without formal proof [1] to simulation techniques that lead to
accurate results [2][3]. Completing this book involved hours of computer sim-
ulation, pages of algebra, translation of antiquated notation, and utilizing
archeological-type efforts to unearth significant but otherwise forgotten
papers.
With regard to this work, three major steps lead to the solution of this prob-
lem. The first is Lessons equation. All works on phase noise must reference
Lessons equation because it is simple, intuitive, and has withstood the test of
time [1]. The second innovation was in the development of Cadence, Inc.s
SpectreRF simulation tool [2]. This tool accurately predicts phase noise and
served as a test bench to validate all derived equations. Finally, Huang [4]
showed it was possible to write out equations for phase noise explicitly. This
work takes inspiration and elements from all these works and forms a model
that is as intuitive as Leeson, as accurate as SpectreRF, and as rigorous as
Huang.

2 The Mathematical Oscillator

The mathematical model of an ideal voltage-controlled oscillator starting at
t = 0 is described by the following expression:
 Chapter 1 Basics of LC Oscillators

t
v ( t ) = A ( t ) sin o t + K v v c ( u ) du ,





(1)
0

where vc is the control voltage of the oscillator as a function of time. The cen-
ter frequency of oscillation is o and the instantaneous frequency of oscilla-
tion is given by:
( t ) = o + Kv vc ( t ) , (2)
where Kv is the oscillator sensitivity and typically given in rad/sec/volt.
As can be seen from (1), the small signal model of an oscillator in frequency/
phase domain, with voltage as an input and phase as an output, functions as an
integrator. The frequency of oscillation is directly proportional to the control
voltage and the oscillator phase is the time-integral of frequency. Being a self-
timed system, the oscillator lacks the ability to correct for its own phase.
Imagine an oscillator running at some frequency with a constant bias applied
to its control voltage line. Any disturbance on the control line will result in
instantaneous frequency shift that integrates over the time the disturbance
lasts. The resulting phase error will last indefinitely and can never be recov-
ered even though the disturbance lasted for a short amount of time as shown
in Figure 1.

FIGURE 1 Phase jitter accumulation.

Time Grid
To

Control Voltage
Time

2
 3 Additive White Noise in LC Oscillators

3 Additive White Noise in LC Oscillators

An oscillation is fully characterized by its amplitude and phase. When white
noise is added to an oscillation, noise corrupts both the amplitude and phase
of oscillation. Assume a noise signal n(t) added to an oscillation v(t). It is cus-
tomary to model noise as an infinite number of uncorrelated sinusoids sepa-
rated by 1 Hz each. The sum of the oscillation signal at o and a noise signal
at frequency o + n is given by:
s ( t ) = A o sin ( o t ) + a n sin ( ( o + n )t + n ) , (3)
where an is the noise amplitude and n is a random phase. This equation can
re-written as
a a
s ( t ) = A o sin ( o t ) + ----n- sin ( ( o + n )t + n ) + ----n- sin ( ( o n )t n ) (4)
2 2
a a
+ ----n- sin ( ( o + n )t + n ) ----n- sin ( ( o n )t n ) .
2 2
The first three terms constitute an amplitude modulated carrier with the mod-
ulating tones at n. The last two terms together with the carrier, approximate a
narrow band phase modulation signal. This means that a single sideband noise
component added to the oscillator modulates both the amplitude and phase of
the oscillation. The power of amplitude modulation sidebands is equal to the
power of phase modulation sidebands.
In Chapter 2, we will rigorously define phase noise. For now, any noise that
modulates the phase of oscillation is phase noise. Any noise that modulates
the amplitude is considered amplitude noise and is unimportant in most prac-
tical cases, except when it later converts to phase or frequency noise.

4 The Linear Oscillator

4.1 Warning
In this section, we develop a misleading analysis of oscillators based on linear
system theory. Despite looking reasonable, we will show later why it is not
accurate or even correct.

3
 Chapter 1 Basics of LC Oscillators

4.2 Linear System Theory Applied to Oscillators

Oscillators are fundamentally nonlinear. In fact oscillators nonlinearity is the
reason for their stable amplitude. Yet linear models are often used to
describe oscillatory behavior. This is acceptable when oscillation start-up
conditions are pursued because oscillation at start-up is a small signal. How-
ever, the periodically stable frequency can be far different from the small sig-
nal linear prediction. Furthermore, a linear oscillator model cannot predict
the oscillation amplitude. In fact, the assumption of linearity, leads to an un-
determined amplitude. This is because in a linear system, if the input doubles,
the output doubles. In an oscillator, this leads to an amplitude that is arbitrary
[5].
So what is a linear model good for?
1. It can yields a startup condition for oscillation, and
2. it gives a rough estimate of the frequency of oscillation.

FIGURE 2 Basic LC oscillator.

R L C

gm

Consider the LC oscillator shown in Figure 2. If the oscillator loop is cut at

any point, the gain around the loop is given by:
gm 2
G ( j ) = ---------------------------------- .
(5)
1 j
--- + jC -------
R L

4
 4 The Linear Oscillator

For a sustained oscillation, Barkhausen criterion mandates that the gain

around the loop is exactly unity and the phase shift around the loop is pre-
cisely 360 degrees. This leads to the following:
gm R = 1 , (6)

1
o C = ---------- . (7)
o L
The oscillator shown in Figure 2 can be modeled as a positive feedback sys-
tem. In Figure 3, the oscillator is constructed using an amplifier and a phase
shift network. The amplifier provides no phase shift. The modes of oscilla-
tions for this system occur at the natural frequencies of the phase shift net-
work. At these frequencies, the phase shift of this network is a multiple of 360
degrees. If there are multiple frequencies at which Barkhausen criterion is
met, then the oscillator can have multiple modes of oscillation. The mode
with the highest gain is most likely to prevail but multiple modes of oscilla-
tion can coexist.

FIGURE 3 Feedback amplifier model of the oscillator.

vn

gm Oscillator
Output

G(j)

Another way to model an oscillator is a single port model. A lossless LC tank

is an oscillator with its frequency of oscillation that can be computed from (7).
Loss in the tank damps the oscillation with a time constant equal to 1/RC.
Adding a negative resistance element replenishes any current that flows
through the lossy element to sustain oscillation. If the oscillation is to grow
then the energy supplied by the negative resistance element must equal the
energy lost per cycle. For the oscillator in Figure 2, the two transistors are
arranged such that they provide a negative resistance of 2/gm. The differen-

5
 Chapter 1 Basics of LC Oscillators

tial resistance of the tank is 2R. Therefore, for a sustained oscillation, the neg-
ative resistance should be equal to the positive resistance at all times. This
leads back to (6). Redrawn in Figure 4, this circuit is known as the voltage-
biased oscillator.

FIGURE 4 Basic LC oscillator (redrawn).

5 Linear Oscillator Noise Analysis

Imagine an oscillator constructed using a parallel LC tank and a transconduc-
tor in a positive feedback loop. The frequency of oscillation is of course given
by (7). Note that the resonant frequency of the LC tank is the same as the
oscillation frequency because both the resistor and the transconductor do not
provide any extra phase shift as they carry no reactive current. The reader can
readily prove that if the resistive loss is modeled in series with the inductor
rather than it parallel, the oscillation frequency will be different from that
given by (7).
Now lets consider the noise. Noise can come from two sources in this sys-
tem: the resistor and the transconductor. Resistor noise is modeled by a white
thermal noise current whose density is given by:
4 kT
i 2( f ) = ---------
n -. (8)
R
Noise in the transconductor is also modeled as a white noise current whose
density is given by:

6
 5 Linear Oscillator Noise Analysis

i 2( f ) = 4 kT g , (9)
n m
where is the noise figure of the transconductor element.
The combined noise can be referred to the input of the transconductor as a
white noise voltage given by:
2 4kT 4kT
v n( f ) = ------------ + ----------- . (10)
gm 2
g R m
Substituting from (6), the input referred noise voltage at the transconductor
input is given by:
2
v n( f ) = 4kTR ( 1 + ) = 4kTFR , (11)
where F = 1 + is the noise figure of the entire oscillator.
The tank impedance at a frequency away from the resonance frequency
can be approximated by
R
Z ( ) = --------------------------
-, (12)

1 + j2Q -------
o
where Q is the tank quality factor.
Using basic feedback theory, it is trivial to prove that the closed-loop transfer
function from the noise input to the oscillator output is given by:
o 2
= --------------
2
H ( ) - . (13)
2Q
In noise analysis, it is customary to represent noise by a sine wave in a 1 Hz
bandwidth. Lets consider a noise component at a frequency d away from
the carrier (i.e. oscillation fundamental tone). Noise power at the output of the
oscillator can be deduced using (11) and (13),
o 2
N out ( ) = 4kTFR --------------
- . (14)
2Q

7
 Chapter 1 Basics of LC Oscillators

As shown earlier, additive noise shows as half amplitude noise and half phase
noise. The noise-to-carrier ratio is obtained by dividing the output phase
noise by the carrier power. We should also consider noise in the lower side
band at o + . The single side band noise to carrier ratio is obtained by add-
ing noise power at + and resulting in the following expression:

4kTFR o 2 2 kTFR 2 1
L ( ) = ----------------- --------------
- --- = -------------- -----o- ---------- . (15)
P o 2Q 2 P o Q 2

This is the renowned Leesons equation [1]. In the original paper it was given
as a heuristic equation without formal proof. The preceding analysis is not
part of the original paper that was based on measurements and observations.
Many other researchers and design engineers derived, over the years, proofs
similar to the one we derived here and ended up with one version or another
of (15).
Whats wrong with the above analysis? First, it is linear and time invariant.
Therefore, no frequency translations of noise can occur. This means that low
frequency noise, such as flicker noise, cannot create phase noise under the
assumptions of this model. The only type of noise that can create phase noise
in this model is noise originating around the oscillation frequency. Moreover,
it has to have an equivalent amount of amplitude noise because it is in
essence, additive noise. In any LC oscillator, this is not true. As we will show
later, some elements contribute pure phase noise and no amplitude noise.
Finally, linear analysis cannot predict the amplitude of oscillation. The ampli-
tude limiting mechanism is fundamentally nonlinear and cannot be captured
in the context of a linear time-invariant analysis.

6 How Is This Book Different?

In the following chapters we will show in detail why the derivation in
Section 4 is wrong. We will show how to use circuit theory to derive an accu-
rate model for phase noise in electrical oscillators. We will describe what we
call a mechanistic model that captures the dominant nonlinearities in an oscil-
lator and provides a closed form expression for phase noise. No fudge factors
utilized!

8
 References

Concepts from nonlinear circuit theory are sometimes utilized yet we tried to
keep that to the minimum necessary.
By doing so, this book provides deep insight into the operation of oscillators
and provides simple procedures for designing high-purity oscillators. We are
answering the seldom tackled questions: why does the oscillator behave that
way? and how is an optimal oscillator designed?

References
[1] D. B. Leeson, A simple model of feedback oscillator noise spectrum, Pro-
ceedings of the IEEE, vol. 54, pp. 329-330, 1966.
[2] K. S. Kundert, Introduction to RF simulation and its application, IEEE J.
Solid-State Circuits, pp. 1298-319, 1999. Also available from www.designers-
guide.com/Analysis.
[3] A. Hajimiri and T. H. Lee, A general theory of phase noise in electrical oscilla-
tors, IEEE Journal of Solid-State Circuits, vol. 33, pp. 179-94, 1998.
[4] Q. Huang, On the exact design of RF oscillators, Proceedings of the IEEE
1998 Custom Integrated Circuits Conference, 1998.
[5] Thomas Lee, Radio Frequency Integrated Circuits, Cambridge University
Press, 1998.

Excerpted from "The Designer's Guide to High-Purity Oscillators" by Hegazi, Rael, and Abidi.
For more information, go to www.designers-guide.com/Books. 9