Documente Academic
Documente Profesional
Documente Cultură
Injection-Locked Frequency
Dividers
David ONeill
March 2005
Declaration
I hereby declare that, except where otherwise indicated, this document is entirely
my own work and has not been submitted in whole or in part to any other university.
Signed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Date
.............................
ii
Abstract
Injection of a periodic signal into an oscillator leads to interesting locking or
pulling phenomena. These effects investigated in the past by Adler [1] and others,
have found increasing importance in todays phase-locked loops. Recently phaselocked loops have moved toward higher and higher speeds. As this frequency of
operation has increased power consumption in the frequency divider has increased
due to the methods of division used.
Injection-locked frequency dividers (ILFD) offer a low power consumption alternative to the traditional methods of frequency division. However as these methods
have become more widely used, numerous papers have proposed contradictory models for quantifying the input locking range of the ILFD.
In this report we will seek to investigate ILFDs by modeling an ILFD and verifying the results experimentally.
iii
Acknowledgements
Firstly Id sincerely like to thank Professor Kennedy for his encouragement
and advice thought the year on the work involved in the project. Id also like
to thank David Bourke for being willing to work the long hours with me to get this
project completed. Special thanks must go to the postgraduates especially Keith
ODonoghue who was always available to help as much as he could with any queries
we had.
Thanks to Zhipeng Ye for his input toward writing the ECCTD 2005 paper.
Thanks to Gerard Hooton for all the technical support throughout the year and a
final word of thanks to Niamh OSullivan for always being extremely helpful toward
us.
iv
Contents
Abstract
ii
Acknowledgements
iii
Contents
iv
List of Figures
vii
List of Tables
1.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2
Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3
Background To Project . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4
Structure Of Report . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Constructing An ILFD
2.1
2.2
LC-VCO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.2
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.3
ILFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1
ILFD Topologies . . . . . . . . . . . . . . . . . . . . . . . . . 12
Contents
2.2.2
v
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2
Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.1
Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.2
Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.1
Existing Models . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.2
3.2.3
3.2.4
Implementation in MATLAB . . . . . . . . . . . . . . . . . . 24
4 Results
4.1
4.2
17
26
Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.2
Devils Staircases . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.3
Model Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.2
4.2.3
38
5.1
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2
Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Bibliography
40
Contents
Appendices
vi
42
MOSFET Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Photographs of Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . 43
vii
List of Figures
2.1
Oscillator Model: (a) Representation of Oscillator with NR representing the active components of the circuit. (b) Desired IV Characteristic
of NR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chosen LC-VCO Topolgy. Cross-Coupled Inverters M1 to M4 , generate the negative resistance needed to overcome the losses in the LC
tank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6
2.7
2.8
2.9
Example Driving Point Characteristic from Tektronix TDS5054 Oscilliscope. Channel 1: X-Axis 2V/Div. Channel 2: Y-Axis 2mA/Div . 12
2.2
2.3
2.4
2.5
List of Figures
viii
2.12 Injection Locked Divider Topology with load balancing for Input VCO 16
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
4.1
4.2
4.3
J
Portion of Staircase highlighted in Fig. 4.2. Note the steps at fIN
=
fd
2 fIN J
1 fIN J
2
, fd = 2 , fd = 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5
4.4
4.5
4.6
4.7
4.8
4.9
27
30
fIN J
fd
30
8
3
List of Figures
ix
5.1
5.2
List of Tables
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
Chapter 1
1.1
Introduction
1.2
Objectives
vided by a voltage controller oscillator (VCO), using the same topology as the
divider.
An automated method of testing the locking range of the divider was to be
devised and implemented.
The operation of the ILFD was to be modeled. The existing models were to
be compared to our experimental results and if they were not valid for the
observed results a new model would have to be devised to accurately predict
the observed results.
1.3
Background To Project
RF phase-locked loops are widely used in applications such as frequency synthesizers [4]. The frequency divider is one of the key components in a phase-locked loop
(PLL) and its power consumption is one of the primary concerns when designing a
PLL. Frequency dividers are commonly realized using common-mode logic (CML) [5]
or through the use of a Miller divider [6]. While both of these methods have been
realized at very high frequencies [7], the power consumption is very high. Injectionlocked frequency dividers offer a low power consumption alternative to both Miller
and CML division at high frequencies.
One of the drawbacks to the use of injection-locking as a method of frequency
division is that there is a limited input bandwidth (locking range) over which frequency division occurs. However, this is not very relevant with their use in LC-VCO
based PLLs as the LC-VCO itself has a limited tuning range.
As outlined in Section 1.2 the main objective of this project was to investigate injection-locking and attempt to match our experimental results to an existing
model.
The main contradiction in the existing models for the input locking range is that
it has been stated as being both proportional [8] and inversely proportional [9] to
the size of the inductor present in the divider.
In [9], the locking range is presented as the following, where Iin is the injected
current, Itail , the current in the tail of the VCO, RS the series resistance of the
inductor and L the inductor size:
Iin Rs
Itail 0 L
(1.1)
However the theory in [8] presents the following expression for the locking range of
an injection-locked divider.
H0 a2 Vin
(1.2)
0 = 2Q
By equating H0 to Q2 Rs , Eq. (1.2) can be reduced to
2 = 2 La2 Vin ,
where a2 is the second-order coefficient of the nonlinearity. This leads to the conclusion that larger locking ranges can be achieved by increasing the size of L and
has been mentioned previously this is clearly contradictory to what is presented in
Eq. (1.1).
This paper will attempt to resolve this conflict.
1.4
Structure Of Report
This report will begin, in Chapter 2, by discussing the process involved in designing and implementing the oscillator which formed the basis of the ILFD. Following
this the process involved in modifying this oscillator to operate as a frequency divider
will be outlined.
With the oscillator constructed Chapter 3 will discuss the process of testing and
modeling the oscillator. Chapter 4 will include the results of the project which
Chapter 5 will attempt to draw conclusions from these results and outline future
work that could be done in the area.
Chapter 2
Constructing An ILFD
This chapter outlines the steps involved in the implementation of an ILFD. It
was decided to implement an ILFD as opposed to performing SPICE simulations as
analysing injection-locking in SPICE is an extremely time consuming process. The
creation of an ILFD first requires the design on a VCO which is then modified to
act as an ILFD. These processes will now be described.
2.1
2.1.1
LC-VCO
Introduction
2.1.2
Theory
A simple electronic oscillator produces a periodic voltage output. An ideal oscillator could simply consist of a capacitor and inductor but in reality there will
2.1 LC-VCO
C
IIN
C
IIN
Figure 2.2: VCO with negative resistance (R) added to overcome positive resistance R.
The negative resistance is achieved through the use of an active circuit. This
is more accurately described as a non-linear resistor NR as shown in Figure 2.3(a).
The non-linear resistor has a characteristic which has a negative resistance for a
portion of its IV characteristic. This can be achieved through the use of a CMOS
cross-coupled pair which approximates the ideal negative resistance characteristic.
This characteristic will be similar to that shown in Figure 2.3(b). As can be seen
in the characteristic, for a portion of the voltage swing applied across NR the slope
is negative, hence if NR only acts in this region, the circuit shown in Figure 2.3(a)
will act like the circuit described in Figure 2.2.
2.1 LC-VCO
6
f (VC )
IIN
IL
L
VIN
VC
NR
VDD
VDD
g0
(a)
(b)
2.1 LC-VCO
VDD
M1
M2
C
IBIAS
M3
M5
V+
M4
M6
2.1 LC-VCO
VDD
L
2
L
2
C
IBIAS
V
V+
M3
M5
M4
M6
Figure 2.5: Alternative LC-VCO Topolgy. Consists of only NMOS Transistors. Note that inductance L is created by using two inductors of
inductance L2 .
2.1.3
With the choice of topology having been made the next step was to choose the
component values so as to be able to implement the design. At this stage of the
project PSpice was utilized to see how the VCO reacted to changes in component
values and transistor model parameters.
As was mentioned briefly in Section 2.1.2 it is necessary to create a negative
resistance using the transistors M1 to M4 to create a non-linear resistance. In order
to quantify the negative resistance created by the transistors present in our design
we used PSpice to create a driving point (DP) characteristic. This is shown in
Figure 2.6.
The DP characteristic is a plot of the current in the non-linear resistor versus the
2.1 LC-VCO
R<
1
L
and R < g0 .
g0
C
(2.1)
1 + g0 R
.
LC
(2.2)
Reading the slope of the curve in Figure 2.6 around V=0, we can calculate
g0 . Initially we believed that this would be large enough to satisfy the oscillation
criterion as outlined in Eq. (2.1). Next we choose values of L and C such that the
circuit would be capable of oscillating at a frequency of 1.75 MHz. The choice of
this frequency was not arbitrary. At this frequency it is possible to build the circuit
on breadboard, without unnecessary concern regarding the layout of the circuit for
RF reasons. To calculate the components values required Eq (2.2) is simplified to
2.1 LC-VCO
10
So at this stage using CD4007N chips the cross-coupled CMOS LC was implemented. As described briefly here and in more detail in David Bourkes report,
problems were encountered due to the lossy nature of the components used in the
oscillator. The losses involved were larger than what we first estimated. This meant
that g0 was not large enough to over come the losses, so to overcome the lossy nature of the components and breadboard, it was necessary to increase the W
ratio in
L
the transistors used in the cross-coupled pair. This is performed on the breadboard
by connecting multiple PMOS or NMOS transistors in parallel. This has the same
ratio in PSpice.
effect as increasing the W
L
To model the circuit to a reasonable degree of accuracy in PSpice it was necessary
to find some of the model parameters for the CD4007N devices. These parameters
were obtained by using the parameters found in [12] and [10]. These parameters are
repeated in Appendix A. Having input these values into PSpice, and increased the
W
ratio, an increased slope around V=0 was observed both experimentally and in
L
simulation.
ratio.
Figure 2.7 shows the increase in the slope with the increased transistor W
L
As can be seen and measured from the curve the slope g0 is now steeper and thus
the negative resistance is large enough to overcome the lossy components.
Figure 2.7: Comparison between the PSpice Driving Point Characteristics. Solid Line: Original Characteristic. Dashed Line: Increased W
L
Ratio Characteristic.
2.1 LC-VCO
11
V1
V2
V
V
VIN
NR
V
OU T
R
R
(a)
(b)
V1 V2
R
2.2 ILFD
12
Plotting the output of the op-amp, VOU T , against the voltage across the terminals
of the nonlinear resistor, V2 , yields the driving point characteristic of the particular
nonlinear resistor. A sense resistor of 100 is adequate for most characterisations.
2.2
ILFD
At this stage a VCO had been designed and implemented as outlined in 2.1.3.
The next step is to manipulate this VCO design so it is capable of being used as
an ILFD. This is described in this section. Again I was responsible for much of
the design and simualation work, while David Bourke solved many of the problems
ecountered in implementing the circuit.
2.2.1
ILFD Topologies
There are a number of different ILFD topologies available. Two of them are
outlined in this section. Initially we chose the topology shown in Figure 2.10. This
topology was chosen due to its similarity to the LC-VCO created previously. With
2.2 ILFD
13
the work characterising the LC-VCO topolgy having been done previously, the creation of an ILFD simply required the alteration of the circuit to allow an injection
signal be applied at some point in the circuit.
VDD
M1
M2
C
IBIAS
RS
M3
M4
R
M5
M6
VIN J
CIN J
2.2 ILFD
14
VDD
M1
M2
VIN J
M7
C
IBIAS
M3
M5
V+
M4
M6
2.2.2
Implementation
2.2 ILFD
15
2.2 ILFD
16
VDD
M1
M2
C
IBIAS
VDIV
M3
VDIV+
M4
R
M5
M6
C1
M5
C1
V+
Figure 2.12: Injection Locked Divider Topology with load balancing for
Input VCO.
17
Chapter 3
3.1
Testing
In this section the process for automatically testing the ILFD is outlined. In [15]
a process for automatically obtaining the Devils Staircase structure of an ILFD is
described. That structure is useful to us as it shows the input locking range for the
oscillator and thus can be used to compare our experimental results to the existing
models for injection-locking.
3.1.1
Basics
The experimental setup used is shown in Fig. 3.1. The driving voltage VIN J and
the driving frequency fIN J are produced by a precision frequency generator (Agilent
33220A) connected to a IEEE-488 bus. We estimate the frequency of the driven
oscillator (fd ) through the use of a precision counter (Agilent 53131A) which is also
under IEEE-488 control.
Once the circuit under test has been connected to the signal generator and
counter, we set the frequency of the undriven circuit to the desired natural frequency f0 . We then specify the frequency range, step size and amplitude range of
the forcing signal. The software (Agilent VEE Pro 7.0) then initializes the instruments and begins to take measurements.
Should one wish to examine a section of the staircase at a higher resolution, the
step-size can be decreased (to increase input resolution) and the resolution of the
3.1 Testing
18
FUNCTION
GENERATOR
NONLINEAR
OSCILLATOR
FREQUENCY
COUNTER
IEEE 488
3.1.2
David Bourke did the majority of work on creating the Agilent interface to automate the testing procedure and thus it is discussed in far more detail in his report.
However Figure 3.2 shows a screenshot of the Agilent VEE interface reproduced
from David Bourkes report.
3.2 Modeling
19
To obtain the data displayed in Section 4.1 the user must input the chosen voltage
sweep values and the chosen frequency sweep values. Figure 3.3 shows in detail the
area where the user inputs the values.
Figure 3.3: Agilent VEE Interface for selecting voltage and frequency
sweep. Voltage to be swept from 0.25V to 6V with a step-size of 0.25V.
Frequency is swept from 1 kHz to 10 MHz with a step-size of 1 kHz.
When Agilent creates the text files containing the output data it is necessary
to remove the error values. This can be done easily in MATLAB by parsing the
array and removing the error. The error value was then replace by an interpolated
value. The number of errors was extremely small (50 in 250,000 measurements) so
the impact of them is minimal.
When the array has been fixed, the values are then simply plotted. Using the
following command plots such as the one that can be seen in Figure 3.4 are created.
The significance of this plot will be discussed later in the report.
3.2
3.2.1
Modeling
Existing Models
Initially attempts we made to fit our experimental results to the models which
had been previously outlined in [8] and [9]. Ultimately these attempts failed. This
3.2 Modeling
20
3.2.2
At this stage we attempted to create a model for the ILFD we had constructed.
The model was required to accurately predict the size of the locking range based
on the circuit parameters, (e.g. Value of L). One of the post-graduate students
3.2 Modeling
21
in the department, Zhipeng Ye, was also researching the area of injection-locking,
so we worked with him toward modeling our ILFD. Zhipeng and I concentrated on
modeling the ILFD while David Bourke utilized the testing routine he had created
to extract the values we required from the circuit.
The model of the ILFD is shown in Figure 3.5. From this model, we can develop
the following equations:
f (VC )
IL
L
VC
NR
dVC
= IL f (VC )
dt
dIL
L
= IL R VC .
dt
(3.1)
In Eq. (3.1), f (VC ) is the driving point characteristic of the non-linear resistor.
This characteristic can be extracted both experimentally and in PSpice. This is
described in Section 3.2.3. From Section 2.1.3 we expect that the characteristic
curves will have a cubic form. We assume the curves are of the form:
a
2
VDD
(3.2)
From observing the behaviour of the ILFD we know that when the ILFD has a
signal injected into it the parameter a is altered. To model this change, we expand
a around VGS as follows:
3.2 Modeling
22
a=a
(VGS
d
a
+ vGS ) = a
(VGS ) +
vGS
dVGS VGS
(3.3)
Now substituting Eq. (3.3) into Eq. (3.2) and combining with Eq. (3.1) we get the
following differential equations:
dVC
A + daVin 3
= IL (A + daVin )VC +
VC
2
dt
VDD
(3.4)
dIL
= IL R VC
dt
(3.5)
For simplicity a
(VGS ) is denoted as A while
d
a
dVGS
VGS
represents the injected sinusoid and should be of the form V sin(t) Solving (3.4)
and (3.5) with MATLAB allows us to model the operation of the oscillator. The
results from this modeling are shown in Section 4.2.
3.2.3
Eq. (3.4) and Eq. (3.5) require us to extract some parameters from the circuit in
order to have a working model. In Section 2.1.3 a driving point characteristic was
obtained for a single bias voltage on the current source. However when an injected
signal is applied this bias point will vary, thus it was necessary to obtain a number
of different characteristics for a number of different bias voltages. The results of this
are shown in Figure 3.6.
In MATLAB polynomials were fitted to these curves in order to extract a number
of different values for the coefficient a for different bias voltages. By plotting the
values of a extracted and assuming a quadratic fit in the region of interest we can
extract da as shown in Figure 3.7
At this stage it was desirable to also get the PSpice simulation to agree with
what was observed experimentally. To achieve this a MOSFET model was extracted
experimentally from the CD4007N devices. This model is shown below.
.model PMOS level=3 L=10u W=70.8u VTO=-0.78V LAMBDA=0 Kp=55.5u
.model NMOS level=3 L=10u W=35.4u VTO=1.1V LAMBDA=0 Kn=111u
3.2 Modeling
23
da
dVGS
indicates
3.2 Modeling
24
When PSpice simulations were preformed using this model, the driving point
characteristics extracted were almost identical to those observed in the experiment.
The values for the model parameters were also extracted from PSpice and these were
used to predict the locking range also.
3.2.4
Implementation in MATLAB
At this stage the model was implemented in MATLAB. The code used for the
simulation is shown here. Essentially MATLAB is used to solve the differential
equations Eq. (3.4) and Eq. (3.5). So for the case of an injected sinusoid of amplitude 0.75V and frequency 2.66 MHz the results of injection into our divider can be
obtained using the following code:
function dy=dlfun(t,y)
dy=zeros(2,1);
dy(1)=(y(2)-(-0.0016-5.07e-4*0.75*sin(2.66e6*2*pi*t))*y(1)+
(-0.0016-5.07e-4*0.75*sin(2.66e6*2*pi*t))*y(1)^3/50)/(131e-12);
dy(2)=(-y(2)*10-y(1))/(76.8e-6);
In the above equations the value of A was -0.0016, L was 76.8 H, C was 131 pF
while da was extracted as 5.07e-4. To utilize the m file shown above it is initiated
as follows:
[t,y]=ode45(@odea_exp, [0 5*1e-5], [2 0]);
[0 5*1e-5] indicates the time step that the m file should be run over. odea_exp
is the name given to the m file. The code then outputs a time vector t and a vector
containing the output voltages y. As shown in Section 4.2 when these are plotted it
can be found whether or not the ILFD is in lock for a particular input sinusoid.
To find the locking range the user has to perform an iterative process. The
simplest way to do this to pick an injection signal near the natural frequency of
the oscillator and run the simulation over a reasonably long time. The output is
plotted using a simple MATLAB plot command plot(t,y(:,1)). This results in
an output like is shown in Figure 3.8(a). For Figure 3.8(a), the oscillator is out of
lock. This is indicated by the slight perturbations on the output. In this case the
injection signal chosen is outside the input locking range. Now the user must now
3.2 Modeling
25
choose a frequency closer to the natural frequency. This is shown in Figure 3.8(b).
In this case the new frequency is in lock. The user should now, perhaps using a
midpoint method, iteratively pick new points to find the edge of the locking range.
(a)
(b)
Figure 3.8: Finding the Locking Range: (a) Injected Signal Outside Locking Range. (b) Injected Signal Inside Locking Range.
26
Chapter 4
Results
4.1
4.1.1
Experimental Results
Introduction
The experimental results obtained from the ILFD are detailed in this section.
All the results were taken from an ILFD with a natural frequency of oscillation of
1.745 MHz unless otherwise noted. This was designed as discussed in Chapter 2.
4.1.2
Devils Staircases
The results for the classical injection-locked oscillator are shown in Tables 4.1
and 4.2. With the amplitude of the driving oscillator set to 0 V, we set the natural
frequency, f0 , of the undriven oscillator to 1.745 MHz. We then cycle through
sinusoids with frequencies from 1 kHz to 10 MHz and amplitude 0.25 Vpp to 4 Vpp .
The results of the experiment described above are shown in the surface plot in
Fig. 4.1. On the surface plot it is possible to see the flat regions of locking where
fIN J
is constant, corresponding to a step of the staircase. These regions get wider
fd
for larger input amplitudes. To see the devils staircase structure of the oscillator,
a cross section of the plot is shown in Fig. 4.2. As observed in [15] it is possible to
zoom in on specific regions of this curve and see a self-similar staircase structure.
This can be seen in Fig 4.3.
In Fig. 4.2 the structure reveals that the ILFD is capable of locking over quite
large ranges when dividing by two, four and six. The ILFD also locks and divides by
one, three and five, but the locking ranges are not as large for these division ratios.
A cross-section of Fig. 4.4 is shown in Fig. 4.5 to illustrate the staircase structure
for the direct injection topology. This is similar to the structure shown in Fig. 4.2 but
27
Figure 4.1: Locking behaviour for the classical injection locked oscillator
topology. Locking is characterised by a flat region on the plot. Note that
the widths of the locking regions increase with VIN J .
Figure 4.2: Devils Staircase with VIN J = 3V. Note the steps at fd =
fIN J fIN J fIN J
, 4 , 6 .
2
28
Figure 4.3: Portion of Staircase highlighted in Fig. 4.2. Note the steps
J
J
J
at fIN
= 25 , fIN
= 21 , fIN
= 32 .
fd
fd
fd
the locking ranges vary slightly. This is not immediately apparent, but Tables 4.1
and 4.2 show the differences in the locking ranges of the two topologies.
Fig. 4.6 shows the self-similar repeating staircase for the direct injection locked
oscillator.
Ratio
Classical (MHz)
Direct (MHz)
0.0055
0.0009
0.1080
0.0415
0.0135
0.0060
0.1600
0.0395
0.0090
0.0135
0.0950
0.0300
29
Ratio
Classical (MHz)
Direct (MHz)
0.0240
0.0185
0.3470
0.2310
0.0375
0.0440
0.4950
0.2120
0.0270
0.0830
0.3410
0.3430
Figure 4.4: Locking behaviour for the direct injection locked oscillator
topology.
30
Figure 4.5: Devils Staircase with VIN J = 3V. Note the steps at fd =
fIN J fIN J fIN J
, 4 , 6 .
2
Figure 4.6: Portion of Staircase highlighted in Fig. 4.5. Note the step at
fIN J
= 83 .
fd
4.1.3
31
Input Amplitude
Firstly the variation in the input locking range as the amplitude of the injected
sinusoid changes was obtained. The results of this are shown in Table 4.3. As can
be clearly seen there is quite a linear increase in the bandwidth for larger input
amplitudes. This agrees with the existing models (Eq. (1.2) and Eq. (1.1)) and also
fits in with the model outlined in Section 3.2.2
Amplitude (Vpp )
Bandwidth (MHz)
1.00
0.10
1.50
0.15
2.00
0.21
Table 4.3: Comparison of input locking ranges with various input amplitudes.
Inductor Size
In the previously outlined theories of [8] and [9], there has been a contradiction
regarding the dependence of locking range on the size of the inductor present in the
ILFD. To obtain the results of this section the capacitance and the inductor present
in the circuit were changed while the frequency of operation of the divider was kept
constant. This resulted in the values of input locking range as shown in Table 4.4.
As can be clearly seen in the table, the size of the locking range was seen to
increase with for large inductor sizes. This agrees with the theory outlined in [8].
The disagreement of these measurements with the model outlined in [9] can be
explained by the fact that in [9] the author tests the variation of locking range with
inductor size on three ILFDs with different frequencies of operation. We believe that
a fairer comparison is made when the frequency of operation is kept constant while
the inductor size is varied.
32
Bandwidth (MHz)
17
0.03
63
0.13
77
0.15
Table 4.4: Comparison of input locking ranges with various inductor sizes.
Current Source Bias Voltage
The final parameter that was varied while measuring the locking range was the
bias voltage on the current source in the tail of the divider. This change can be
visualised as moving the point around which the slope da is taken. So increasing
the bias voltage lowers the slope of da by moving it to either steeper or shallower
points on the curve shown in Figure 3.7.
As can be seen in Table 4.5 increasing the bias voltage, which leads to a smaller
da results in a smaller locking range. On the other hand decreasing the bias voltage,
which leads to a larger da results in a large locking range.
Bias Voltage (V)
Bandwidth (MHz)
4.18
0.21
5.42
0.15
8.08
0.06
Table 4.5: Comparison of input locking ranges with various bias voltages.
Its worth noting that myself and Zhipeng attempted to increase the locking range
by cascoding the current source in the tail and injecting on one of the transistors
while biasing the other with a constant voltage. This had the desired results in
simulation but it was not possible to implement on the breadboard due to time
constraints.
4.2
4.2.1
33
Model Simulations
Introduction
In this section the model introduced in Section 3.2 along with that provided by [8]
are used to predict the input locking range of the divider for different parameters.
For the parametric model outlined in Section 3.2 the parameters used are the those
that were outlined in Section 3.2.4.
4.2.2
As was outlined in Section 3.2.4 visually observing the simulation output allows
one to tell whether the circuit is in or out of lock. To show that this can be observed
experimentally also Figures 4.7 and 4.8 have been included. The comparisons show
that the simulation and the experiment give visually comparable outputs for the
oscillation voltage both in lock and out of lock.
34
4.2.3
Lees Model
As mentioned in Section 3.2 before attempting to devise a new model for injectionlocking, we first attempted to fit our results to the existing models. As we had already shown the locking range was proportional to the inductor size we concentrated
on the model shown in Eq. (1.2). The results for these comparisons are shown in
Table 4.6 and 4.7.
As can be seen from the two tables, for the low input injection amplitudes the
model was a good match but for larger input injection amplitudes the model did
not match the experimental results. Following the further analysis and modeling
35
82
0.045
0.04
68
0.032
0.03
22
0.009
0.01
15
0.007
0.01
82
0.228
0.09
68
0.195
0.08
22
0.053
0.02
15
0.042
0.02
performed it appears likely that the correlation between the results is purely coincidental as the Lee model appears to only predict locking in situations where the
injected signal is directly summed with the oscillation signal.
Parametric Model
In this section our model was utilised both with the parameters found experimentally and those obtained from PSpice. When PSpice is set up with the models
outlined in Section 3.2.3, the driving point characteristics and thus the parameters
obtained are almost identical to those extracted experimentally. A plot comparing
the driving point characteristics is shown in Figure 4.9.
Tables 4.8, 4.9 and 4.10 show the results of the comparison between the experimentally obtained locking ranges and the locking ranges obtained from the model.
In the tables, Model [P] refers to the results from the simulations based on parameters extracted from PSpice while Model [E] refers to results based on parameters
extracted from the experimental setup.
As can be clearly seen from the Tables, the model proposed in Section 3.2 provides
an excellent model for the locking range of the classical injection locked divider
36
VIN J
Method
Range (M Hz)
1.00
Experiment
0.10
Model [E]
0.10
Model [P]
0.10
Experiment
0.15
Model [E]
0.16
Model [P]
0.16
Experiment
0.21
Model [E]
0.20
Model [P]
0.20
1.50
2.00
Method
Range (M Hz)
17
Experiment
0.03
Model [E]
0.03
Model [P]
0.03
Experiment
0.13
Model [E]
0.12
Model [P]
0.12
Experiment
0.15
Model [E]
0.16
Model [P]
0.16
63
77
Method
Range (M Hz)
4.18
Experiment
0.21
Model [E]
0.21
Model [P]
0.21
Experiment
0.15
Model [E]
0.16
Model [P]
0.16
Experiment
0.06
Model [E]
0.07
Model [P]
0.07
5.42
8.08
37
38
Chapter 5
5.1
Conclusions
All the objectives as were stated in Section 1.2 were successfully carried out over
the duration of the project. An implementation of an ILFD being driven by a VCO
was completed. This implementation demonstrated the concept of injection-locking
and acted as a frequency scaled model of the GSM receive specification.
An automated method of testing an ILFD was devised. This technique was used
to extract the values for the locking range depending on different circuit parameters.
These results were used to show the inadequacy of the existing models for dealing
with the ILFD topology described in this report. The results obtained for the
variation in locking range with inductor size disagrees fundamentally with the theory
outlined in [9] but agrees with that found in [8].
A method of modeling to a high degree of accuracy, the locking range of the
ILFD was devised. This model predicted the locking range successfully for changes
in all the circuit parameters.
Based on the work completed, two papers have been submitted to the European
Conference on Circuit Theory and Design. The first is entitled Accurate Modeling and Experimental Validation of an Injection-Locked Frequency Divider while
the second is entitled The Devils Staircase As A Method of Comparing InjectionLocked Frequency Divider Topologies. Both of these papers have been included as
appendices to this report and can be found in Appendix C and Appendix D.
5.2
Future Work
One focus of future work that could be performed in the area would be to expand
upon the MATLAB model for injection-locking such that it would be possible for it
39
to automatically calculate the locking range based on the input parameters without
the need for constant intervention from the user.
A high speed IC layout could be investigated, with the focus of that investigation
being an analysis of the validity of the model at higher frequencies.
Based on the results of Section 4.1.3 extensive analysis could be performed on the
impact of a change in the current source bias voltage on the operation of the circuit.
Further to this analysis could be done on a circuit implementation of Zhipeng Yes
suggestion of a cascode current source to increase input locking range.
Finally since the intention of an injection-locked frequency divider is to reduce the
power consumption of the divider in a PLL an analysis of this power consumption
versus the power consumption of a traditional method of division at a particular
frequency could be undertaken.
40
Bibliography
[1] R. Adler, A study of locking phenomena in oscillators, Proc. IEEE, vol. 61,
pp. 13801385, Oct. 1973
[2] A. E. Siegman, Lasers, Mill Valley, CA: University Science Books, 1986.
[3] V. Manassewitsch, Frequency Synthesizers, 3rd ed. New York: Wiley, 1987.
[4] J. Craninckx and M. Steyaert, Wireless CMOS Frequency Synthesiser Design,
London, U.K.: Kluwer, 1998.
[5] J. Craninckx and M. Steyaert, A 1.75 GHz/3 V dual-modulus divide-by-128/129
prescaler in 0.7 m CMOS, in Proc. ESSCIRC, Sept. 1995, pp. 254257.
[6] J. Lee and B. Razavi, A 40 GHz frequency divider in 0.18 m CMOS technology, in Symp. VLSI Circuits Dig. Tech. Papers, June 2003, pp. 259262.
[7] H.-D. Wohlmuth and D. Kehrer, A high sensitivity static 2:1 frequency divider
up to 27 GHz in 120 nm CMOS in Proc. Eur. Solid-State Circuits Conf., Firenze,
Italy, Sept. 2002, pp. 823826.
[8] H.R. Rategh and T.H. Lee, Superharmonic injectionlocked frequency dividers, IEEE J. Solid-State Circuits, vol. 34, pp. 813821, June 1999.
[9] M. Tiebout, A CMOS Direct InjectionLocked Oscillator Topology as HighFrequency Low-Power Frequency Divider, IEEE J. of Solid-State Circuits,
vol. 39, July 2004, pp. 11701174.
[10] J. Buckley, A Design Methodology for CMOS Oscillators, Masters Thesis,
Dept. of Microelectronic Engineering, UCC, Sept. 2004
[11] A. Hajimiri and T.H. Lee, Design Issues in CMOS Differential LC Oscillators,
IEEE Journal of Solid-State Circuits, vol. 34, no. 5, pp. 717724, May 1999.
[12] K. ODonoghue Final Year Project Report, Dept. of Microelectronic Engineering, UCC, March 2004.
Bibliography
41
[13] S.-M. Wu and W.-L. Chen, A 5.8-Ghz CMOS VCO with Injection-Locked
Frequency Divider for IEEE 802.11a Application, ECCTD, Cracow, Poland,
Sept. 2003
[14] T.H. Lee, H. Samavati and H.R. Rategh, 5-GHz CMOS Wireless LANs,,
IEEE Trans. on Microwave Theory and Techniques, vol. 50, Jan. 2002
[15] M.P. Kennedy, K.R. Krieg and L.O. Chua, The Devils Staircase: The Electrical Engineers Fractal, IEEE Trans. on Circuits Syst, vol. 36, Aug. 1989
42
Appendices
MOSFET Parameters
NMOS
PMOS
level=1
level=1
vto=0.7
vto=-0.8
gamma=0.45
gamma=0.4
phi=0.9
phi=0.8
nsub=9e+14
nsub=5e+14
ld=0.08e-6
ld=0.09e-6
uo=350
uo=100
lambda=0.01
lambda=0.02
tox=9e-9
tox=9e-9
pb=0.9
pb=0.9
cj=0.56e-3
cj=0.94e-3
cjsw=0.35e-11
cjsw=0.32e-11
mj=0.45
mj=0.5
mjsw=0.2
mjsw=0.3
cgdo=0.4e-9
cgdo=0.3e-9
js=1e-8
js=0.5e-8
B Photographs of Circuit
43
Photographs of Circuit
B Photographs of Circuit
44
45
46
.
47
.
48
.
49
.
50