Sie sind auf Seite 1von 4

Optimizing the Design of an

Injection-Locked Frequency Divider


by Means of Nonlinear Analysis
Mohammad M. Ghahramani

, Saeid Daneshgar

, Michael Peter Kennedy

, and Oscar De Feo

Department of Microelectronic Engineering


University College Cork, Ireland
Email: saeidd@ue.ucc.ie

Tyndall National Institute, Lee Maltings, Cork, Ireland


AbstractInjection-locked frequency dividers (ILFDs) are ver-
satile analog circuit blocks used, for example, within phase-locked
loops (PLLs). With respect to their digital counterparts, they have
the advantages of a low power consumption and division ratios
greater than two. The price for these advantages is believed to
be a limited locking range. Here we show that this is not the
case; indeed, by combining nonlinear systems theory (bifurcation
analysis) with optimization techniques, we have signicantly
increased the locking range of a classical (LC oscillator-based)
injection-locked frequency divider, predicting a locking range
that is about twenty times greater than what has been reported
in the literature to date. The wider locking range predicted by
the theory has been conrmed by SPICE simulations.
I. INTRODUCTION
High frequency PLLs are widely used in modern telecom-
munications applications, such as mobile telephony [1]. The
frequency divider is one of the key components in a PLL and
its power consumption is one of the primary concerns of the
circuit design community. Frequency dividers are commonly
realized using common-mode logic (CML) [2] or through the
use of a Miller divider [3]. Whilst both of these methods
have been realized at very high frequencies [4], the power
consumption has also been very high. At high frequencies, an
energy-efcient alternative to both Miller and CML dividers
is offered by injection-locked frequency dividers [5].
It is largely believed that one of the key drawbacks in
using injection-locked frequency dividers is a limited input
bandwidth over which frequency division occurs (locking
range
1
). The literature reports widespread values of divide-
by-two locking ranges, e.g., from 12% in [6] to 29% in [7],
both of which have been obtained for similar frequencies of
oscillation, power supply voltages, power consumptions, and
by using very similar ILFD topologies. Such different locking
ranges for very similar circuits reveals that: I) the locking
range of ILFDs is not necessarily small; II) there is a need for
methods to optimize the locking range of ILFDs with respect
to their design parameters. Therefore, the question to address
is: How do changes in circuit parameters affect the width of
the locking range?
1
Dened as L
R
= 2
f
M
fm
f
M
+fm
where fm and f
M
are the minimal and
maximal frequencies for which the exact frequency division (locking) occurs.
Here, we propose to address this problem by resorting to
well-established theories from nonlinear dynamical systems
which have already proven effective in the analysis of electri-
cal oscillators [8], [9].
In dynamical systems theory, the correct tool to address the
phase locking phenomenon is bifurcation theory [10] and, in
particular, the Arnold tongue scenario [11].
Concentrating on a specic (classical) topology of ILFD,
in this work, we rst combine normal form theory [10] and
numerical continuation techniques [12] for detecting and trac-
ing the borders (fold bifurcations of limit cycles) of the Arnold
tongue scenario. Then, we combine the numerical continuation
technique with an optimization method for increasing the
locking range of the ILFD under consideration.
The rest of the paper is organized as follows. In the follow-
ing section, we briey describe the simplied model of the
ILFD topology considered and state the equations governing
the circuit behavior [13]. In Sec. III, we reveal the normalized
model and highlight the minimal set of effective design pa-
rameters. In Sec. IV, we demonstrate how the dynamics of the
circuit depend on the design parameters highlighted in Sec. III.
In Sec. V, we rst determine the Arnold tongue corresponding
to the divide-by-4 region of the considered ILFD for a typical
set of design parameters; then, by combining numerical contin-
uation and optimization, we provide a signicantly increased
locking range. In Sec. VI, we verify our qualitative predictions
using brute-force SPICE simulations, and nally, in Sec. VII,
we draw some conclusions and suggest further contributions
that will result from our approach.
II. THE CONSIDERED CIRCUIT
In this paper, we focus on a popular ILFD topology [14] shown
schematically in Fig. 1(a). This topology is equivalent to the
circuit shown in Fig. 1(b) [13], where the behavior of the four
transistors network together with the (modulated) tail resistor
has been collected in a (voltage-controlled) nonlinear resistor
(R
NL
). As qualitatively sketched in Fig. 1(c), the nonlinear
resistor has an N-shaped V-I characteristic whose dependence
upon the control voltage is suggested by the arrow.
1-4244-1342-7/07/$25.00 2007 IEEE 571
= 35
= 1.1
= 0
= 111
= 10 L
W
Vt
K

= 70
= 0.8
= 0
= 55.5
= 10 L
W
Vt
K

NL
R
v
in
v
in v
C
i
L
v
in
I
V
(c)
(b)
(a)
DD
V 9V
L
R
C
t

+ +

V
I
R =1k4(1+ )
R 1k L 1.7m
V
C 15p
in
Fig. 1. ILFD oscillator: (a) Circuit schematic with the component values
considered in the SPICE simulation; (b) Equivalent circuit; (c) Qualitative
(V, I) characteristic of the controlled nonlinear resistor and its dependence
upon the control voltage V
in
.
This N-shaped nonlinearity can be approximated reasonably
well by the following cubic equation [13]:
I = f (V, V
in
) = a (V
in
) V
_
1
_
V
V
DD
_
2
_
, (1)
where V
DD
is the supply voltage, a is the slope at the zero
crossing point of the V-I curve in Fig. 1(c), and, to simplify
the analysis, it is assumed that the modulation (by the injected
signal) of the tail resistor (R
t
) is mainly manifested as a
variation of the parameter a.
From Fig. 1(b), it is easy to obtain the circuits state
equations as follows:
_

_
C
dv
C
dt
= i
L
f (v
C
, v
in
) ,
L
di
L
dt
= Ri
L
v
C
,
(2)
where the nonlinear function f (v
C
, v
in
) is as dened in (1).
III. MODEL NORMALIZATION: THE MINIMAL SET OF
DESIGN PARAMETERS
By means of the following linear transformation of the state
variables and time:
x =
v
c
V
DD
, y =
Ri
L
V
DD
, =
t

LC
, (3)
the differential equations in (2) can be re-written as:
_
_
_
x = Q
_
y +G(1 +m
f
sin()) x
_
1 x
2
_
,
y =
1
Q
(x y) ,
(4)
where the operator stands for
d
d
and the new parameters are
dened as follows:
Q =
1
R
_
L
C
, G = Ra (0) ,
m
f
=
a (v
in
)
v
in

vin=0
a (0)
V
in
, = 2f
in

LC,
(5)
having assumed that
v
in
(t) = V
in
sin(2f
in
t) . (6)
In (5), m
f
is the relative variation of the parameter a for
small variations of the input (injected) signal v
in
, i.e., it
accounts for the strength of the input signal. Correspondingly,
accounts for the (relative) frequency of the input signal.
On the other hand, the normalization highlights that indeed
there are only two design parameters for the ILFD topology
considered, namely Q and G. Q accounts for the quality factor
of the resonant network, and G is a parameter that depends
on the value of a (determined by the nonlinearity), and the
resistance (R) of the resonator, i.e., it accounts for the gain of
the oscillator at the origin.
IV. ANALYSIS OF THE UNFORCED OSCILLATOR
Now that the effective design parameters of the ILFD (G and
Q) have been highlighted, we should verify how the dynamics
of the unforced circuit (free-running oscillator) depend upon
their values. This will also determine the range of G and Q
values that will yield a periodic solution of the system (i.e.,
self-sustained oscillations in the ILFD).
By suitably combining normal form theory [10] with contin-
uation techniques [12], we computed the bifurcation diagram
of the unforced ILFD with respect to the parameters G and
Q; this result is reported in Fig. 2.
The bifurcation diagram shows how the dynamics of the
free-running oscillator depend upon the values of G and Q.
Furthermore, the bifurcation diagram also shows that there are
six possible regions of operation. By moving around BT in
the clockwise direction, starting in the region at low values of
G and Q, initially the system has only one stable equilibrium
(the origin); crossing the curve H

(supercritical Hopf), the


origin loses stability and a stable limit cycle (C
0
) appears
around it; continuing and crossing the curve P (a pitchfork
of equilibria, subcritical above BT and subcritical below it)
the origin becomes a saddle and two symmetric repellors
E

appear inside C
0
; these two equilibria become stable
crossing the curve H
+
(subcritical Hopf) where a pair of
symmetric unstable limit cycles (C

) appears around them;


on h (symmetric homoclinic) the two cycles C

merge into
a big limit cycle (C
1
, which is internal to C
0
) by forming an
eight-shaped homoclinic trajectory to the origin; on F (fold
of limit cycles) C
0
and C
1
collide and disappear, leaving only
the three equilibria in the system (E

are the stable ones);


nally the unfolding concludes by crossing P again, where
572
0 0.3 0.6 0.9 1.2 1.5
0
1
2
3
4
5
G
Q
BT
F
h
P
H

H
+
design area
Fig. 2. Zoom nearby point BT of the bifurcation diagram of the the
unforced ILFD oscillator with respect to the design parameters G and Q.
The codimension-2 point BT is a Bogdanov-Takens degeneracy; bifurcation
theory predicts 5 curves emerging from this point: a supercritical Hopf (H

);
a pitchfork of equilibria (P); a subcritical Hopf (H
+
); a homoclinc (h); and
a fold of limit cycles (F). The ILFD circuit oscillates regularly only in the
gray shaded region.
E

shrink to the origin and disappear (cf. [10], page 426,


Fig. 9.10).
Note that simple self-sustained oscillations exist only in the
gray shaded area. Hence, the design of the ILFD should be
conned to this region by appropriately constraining the values
of the parameters.
V. ANALYSIS OF THE FORCED OSCILLATOR
Since the values of G and Q affect the dynamics of the system,
we now study the variations in locking range with respect to
these parameters.
Considering a periodically forced oscillator, it is known
from dynamical systems theory that, in the parameter plane of
the input frequency (for us ) and input strength (for us m
f
),
the regions where the oscillator locks to the input are bordered
by precise bifurcation points; namely, fold bifurcations of
limit cycles. The continuum of these bifurcation points is
known as the Arnold tongues scenario [11] due to the tongue
shaped (cuspidal) form of the locking regions. Because of this
theoretical prediction, we have been able to detect and trace
these borders by means of numerical continuation techniques.
An example of a 1:4 Arnold tongue (i.e., the locking region
in the vs. m
f
plane where the ILFD divides the frequency
of the injected signal by 4) is shown in Fig. 3(a). This locking
region has been obtained for design parameters Q = 5.00 and
G = 0.10, which corresponds to a set of circuit parameters
already considered in the literature [13].
Once one of these borders has been found, we may optimize
the width of the locking region, i.e., the locking range, with
respect to the design parameters G and Q by combining
continuation techniques and optimization methods.
1.5 2.5 3.5 4.5
0
0.25
0.5
0.75
1
1:4

m
f
(b)
1.5 2.5 3.5 4.5
0
0.25
0.5
0.75
1
1:4

m
f
(a)
Fig. 3. Examples of 1:4 locking ranges (Arnold tongues): (a) Design
parameters at values classically considered in the literature (Q = 5.00,
G = 0.10); (b) Design parameters at values suggested by our combined
continuation-optimization analysis (Q = 10.00, G = 0.50).
An example of a very large locking region (which we
obtained by using a gradient descent method) is shown in
Fig. 3(b). The region shown corresponds to design parameter
values Q = 10.00 and G = 0.50 (rounded values close to the
optimum). This Arnold tongue shows a locking range width of
approximately 40% at m
f
= 0.5. For the same value of m
f
,
this locking range is about 20 times wider compared with what
was achieved without using a systematic design methodology
(cf. Fig. 3(a)).
VI. SPICE VALIDATION
Clearly, the results obtained in the previous section depend
heavily on the modelling hypotheses assumed in the normal-
ization process. In practice, the real nonlinearity seen by the
LC tank circuit is not an ideal cubic and its slope at the origin
is not linearly modulated by the injected signal.
573
1:4

m
f
1.5 2.5 3.5 4.5
0
0.25
0.5
0.75
1
Fig. 4. Locking range determined by brute-force SPICE simulations of the
circuit shown in Fig. 1(a) assuming v
in
= V
in
sin (2f
in
t).
Nevertheless, to assess to which extent the quality of our
predictions are jeopardized by our modelling assumptions, we
performed brute-force SPICE simulations of the circuit at the
parameter values indicated in Fig. 1(a).
The results, shown in Fig. 4, clearly show a very wide 1:4
locking range (approximately 40% at m
f
= 0.5) centered at
= 2.0, i.e.f
in
= 2.0 MHz. This locking range is more 20
times wider than the 1:4 locking range previously reported for
this topology [13], highlighting the validity of our predictions.
Note that m
f
= 0.5 is the maximum achievable with the
parameter values indicated in Fig. 1(a). Higher values of m
f
are possible, but these require a larger value of inductor.
VII. CONCLUDING REMARKS
In this paper, we have taken a fresh approach to a classical
engineering problem by utilizing well-established theories
from nonlinear dynamical systems.
First, by means of a normalization of the state equations we
determined the effective design parameters of a conventional
ILFD topology. The proper design area for the ILFD con-
sidered was then determined by referring to the bifurcation
diagram of the system with respect to the effective design
parameters.
Then, by combining numerical continuation with operation
research, we optimized the design parameters with respect to
the locking range, keeping them constrained within the proper
design area. To the best of our knowledge, the improved lock-
ing range is signicantly larger than what has been reported
previously for this circuit topology [13].
Finally, the predictions of our mathematical analysis have
been conrmed by SPICE simulations.
Concluding, it worth mentioning that, whilst the locking
regions predicted by the theory (fold bifurcations of limit
cycles bordering the Arnold tongues) have been detected and
traced via a purely numerical method (i.e., continuation), the
general approach still is strongly theoretical, and should not
be confused with a batch of simulations. Indeed, by resorting
to regular perturbation techniques, such as in [15][18], one
could proceed in a similar, although analytical, way providing
(approximate) analytical expressions for the locking regions.
This is the subject of ongoing research which we will report
at later date.
ACKNOWLEDGMENTS
This work has been supported in part by Science Foundation
Ireland under Grant 06/RFP/ENE009.
MMG is grateful to Yu. A. Kuznetsov and Bart Sautois for
their help and assistance in using MATCONT.
REFERENCES
[1] C. S. Vaucher, Architectures for RF Frequency Synthesizers. Boston:
Kluwer Academic Publishers, 2002.
[2] J. Craninckx and M. S. J. Steyaert, A 5 GHz, 1 mW CMOS voltage
controlled differential injection-locked frequency divider, IEEE Journal
of Solid-State Circuits, vol. 31, pp. 890897, 1996.
[3] J. Lee and B. Razavi, A 40-GHz frequency divider in 0.18-m CMOS
technology, IEEE Journal of Solid-State Circuits, vol. 39, pp. 594601,
2004.
[4] H. D. Wohlmuth and D. Kehrer, A high sensitivity static 2:1 frequency
divider up to 27GHz in 120nm CMOS, in Proceedings of the 28th
European Solid-State Circuits Conference (ESSCIRC), Florence, Italy,
September 2002.
[5] K. Yamamoto and M. Fujishima, 55GHz CMOS frequency divider with
3.2GHz locking range, in Proceeding of the 30th European Solid-State
Circuits Conference (ESSCIRC), September 2004.
[6] H. Rategh and T. H. Lee, Superharmonic injection-locked frequency
dividers, IEEE Journal of Solid-State Circuits, vol. 34, pp. 813821,
1999.
[7] H. R. Rategh, H. Samavati, and T. H. Lee, A 5 GHz, 1 mW CMOS
voltage controlled differential injection locked frequency divider, in
Proceedings of the IEEE Custom Integrated Circuits, San Diego, CA,
USA, May 1999, pp. 517520.
[8] G. M. Maggio, O. De Feo, and M. P. Kennedy, Nonlinear analysis of
the colpitts oscillator and applications to design, IEEE Transaction on
Circuits and SystemsI, vol. 46, pp. 11181130, 1999.
[9] , A general method to predict the amplitude of oscillation in nearly-
sinusoidal oscillators, IEEE Transaction on Circuits and SystemsI,
vol. 51, pp. 15861595, 2004.
[10] Y. A. Kuznetsov, Elements of Applied Bifurcation Theory, 3rd ed. New
York: Springer-Verlag, 2004.
[11] E. Ott, Chaos in Dynamical Systems. New York: Cambridge University
Press, 1993.
[12] A. Dhooge, W. Govaerts, and Y. A. Kuznetsov, MATCONT: A
MATLAB package for numerical bifurcation analysis of ODEs, ACM
Transactions on Mathematical Software, vol. 29, pp. 141164, 2003.
[13] D. ONeill, D. Bourke, Z. Ye, and M. P. Kennedy, Accurate modeling
and experimental validation of an injection-locked frequency divider, in
Proceedings of the European Conference on Circuit Theory and Design,
vol. 3, Cork, Ireland, August 2005, pp. 409412.
[14] S.-M. Wu, R.-Y. Liu, and W.-L. Chen, A 5.8-GHz high efcient, low
power, low phase noise CMOS VCO for IEEE 802.11a, in Proceedings
of the 3rd IEEE International Workshop on System-on-Chip for Real-
Time Applications (IWSOC), Calgary, Alberta, Canada, July 2003.
[15] C. Piccardi, Bifurcations of limit cycles in periodically forced nonlinear
systems: The harmonic balance approach, IEEE Transaction on Circuits
and SystemsI, vol. 41, pp. 315320, 1994.
[16] M. Basso, R. Genesio, and A. Tesi, A frequency method for predicting
limit cycle bifurcations, Nonlinear Dynamics, vol. 13, pp. 339360,
1997.
[17] F. Bonani and M. Gilli, Analysis of stability and bifurcations of limit
cycles in Chuas circuit through the harmonic-balance approach, IEEE
Transaction on Circuits and SystemsI, vol. 46, pp. 881890, 1999.
[18] A. Buonomo and A. Lo Schiavo, Modelling and analysis of differen-
tial VCOs, International Journal of Circuit Theory and Applications,
vol. 32, pp. 117131, 2005.
574

Das könnte Ihnen auch gefallen