Published in IET Circuits, Devices & Systems

Received on 19th August 2009
Revised on 11th April 2010
doi: 10.1049/iet-cds.2010.0030
ISSN 1751-858X
Design of non-balanced cross-coupled
oscillators with no matching requirements
A.S. Elwakil
Department of Electrical and Computer Engineering, University of Sharjah, P.O. Box 27272, Sharjah, Emirates
E-mail: elwakil@ieee.org
Abstract: Using two-port network transmission parameters we derive the general characteristic equation of a
cross-coupled circuit topology which involves two active devices and four or six impedances. The derived
equations are generic and apply both to BJT or MOS transistors and even any other active device without any
matching constraints. Application to realising novel non-balanced non-matched cross-coupled oscillators is
demonstrated. Spice simulations of a MOS oscillator in a 0.25 m technology are given as well as experimental
results from an oscillator employing discrete Bipolar transistors.
1 Introduction
Cross-coupled oscillators are very important and widely used
in the realisation of transceivers. A huge amount of literature
pertaining to the design and analysis of these oscillators both
in their sinusoidal oscillation mode and their relaxation
oscillation mode can be found [1–7]. Two typical widely
used cross-coupled oscillator structures are shown in Figs. 1a
and b employing MOS transistors. The same structures
may well be implemented using Bipolar transistors. In both
structures, the design requirements usually imposed are:
1. the two transistor should be matched in order to realise an
effective differential negative resistance r = −1/g
m
where g
m
is the small signal transconductance for each individual
transistor;
2. the load impedances of both transistors should be similar
in order to obtain a symmetrical (balanced) structure.
It is not however clear whether these design requirements
are essential or not in order to obtain an oscillator.
In this work, we derive using two port network
transmission parameters the characteristic equation of the
general four-impedance cross-coupled structure shown in
Fig. 1c and the six-impedance structure shown in Fig. 1d
without assuming that the employed transistors have to be
matched. The derived equation is valid both for BJT and
MOS transistors and indeed for any other active device.
We then find the oscillation start-up condition and
oscillation frequency for the two classical cases shown in
Fig. 1 and for other cases in which the two load
impedances (Z
2
and Z
3
) are not similar; that is the cross-
coupled oscillator in this case is not balanced. Of course,
oscillators whether sinusoidal or relaxation are highly non-
linear circuits and their accurate modelling and analysis
implies using non-linear dynamics techniques [8–10].
However, linearised small-signal models are usually the
start point for the design phase of any oscillator in order to
derive the oscillation start-up condition (Hopf bifurcation
condition) and also estimate the expected oscillation
frequency as a function of the circuit parameters. By virtue
of using transmission parameters, which are particularly
suited for networks in cascade, we show how the effects of
parasitic impedances can be easily incorporated into the
derived characteristic equation. Two novel design examples
of cross-coupled oscillators where (i) the transistors are not
matched and are even of non-identical types and (ii) the
loads are not balanced are given. Spice simulation using a
0.25 m MOS technology file is used to verify one example
while the other is verified experimentally using discrete
components.
2 Background
Fig. 2a shows the typical two-port network input and
output variables for which the transmission matrix (a) is
IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373 365
doi: 10.1049/iet-cds.2010.0030 & The Institution of Engineering and Technology 2010
www.ietdl.org
defined as
V
1
I
1
_ _
=
a
11
a
12
a
21
a
22
_ _
V
2
−I
2
_ _
(1)
where V
1
and I
1
are the input port voltage and current, while
V
2
and I
2
are the output port voltage and current, respectively.
Fig. 2b shows a basic small signal equivalent circuit of an
NPN BJT transistor operating in the forward active mode.
The model comprises two terminal resistors (r
b
, r
e
) (ideally
r
e
= 0) and a dependent current-controlled current source
bI
1
with output resistance r
o
. b is the forward active
current gain (assumed constant) and r
b
is related to the
BJT small-signal transconductance g
m
as r
b
= b/g
m
. It can
be shown that the (a) matrix for this equivalent circuit is [11]
a
11
a
12
a
21
a
22
_ _
=
−1
br
o
−r
e
r
b
+r
e
br
e
r
o
+r
b
(r
o
+r
e
)
1 r
o
+r
e
_ _
(2)
The above (a) may be simplified considering that br
o
≫r
e
Figure 1 Cross-coupled oscillator structures
a and b Two classical balanced-load cross-coupled oscillators
c Generic structure with four impedances
d Generic structure with six impedances
366 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0030
www.ietdl.org
and (1/br
o
) 0 and then becomes
a
11
a
12
a
21
a
22
_ _
=
−1
A
− r
e
1 +
1
A
_ _
+
1
g
m
_ _
0
−1
b








(3)
where A = g
m
r
o
. If A is sufficiently large such that
(1/A) 0, a further simplification yields
a
11
a
12
a
21
a
22
_ _
=
0 − r
e
+
1
g
m
_ _
0
−1
b








(4)
In a similar manner, and recalling the small signal equivalent
circuit of a MOS transistor operating in the saturation mode
(Fig. 2c), it can be shown that the (a) matrix for this circuit is
a
11
a
12
a
21
a
22
_ _
=
−1
g
m
r
ds
1 r
s
(1 +g
m
r
ds
) +r
ds
0 0
_ _

−1
g
m
r
ds
− r
s
+
1
g
m
_ _
0 0




(5)
where g
m
is the small signal transconductance, r
ds
is the drain
to source resistance and r
s
is the source resistance (ideally
r
s
= 0). It is worth noting that under the ideal conditions
(r
e
, r
o
, b) = (0, 1, 1) for the BJT and (r
s
, r
ds
) = (0, 1)
for the MOS, both transistors can be described by the ideal
transmission matrix
a
11
a
12
a
21
a
22
_ _
=
0 −1/g
m
0 0
_ _
(6)
where the transconductance g
m
is the only design parameter.
However, more complex transistor models can also be
accommodated for. Consider for example the MOS
transistor small signal model of Fig. 2d where an
impedance Z (typically a parasitic C
gd
capacitor) is
connected from gate to drain. The transmission matrix in
this case can be shown to be
a
11
a
12
a
21
a
22
_ _
=
1
r
ds
(1 −g
m
Z) +r
s
×
r
ds
(1 +g
m
r
s
) +Z r
ds
(1 +g
m
r
s
)Z
1 +g
m
r
ds
r
ds
(1 +g
m
r
s
)
_ _
(7)
Further, and since transmission parameters are well suited
for networks in cascade, we may additionally incorporate
the effects of C
gs
and C
ds
(treating each as an individual
two-port network) via matrix multiplication with (7) in the
Figure 2 Two-port network representations
a Two-port network variables
b Basic small signal equivalent model of a BJT transistor
c Basic small signal equivalent model of a MOS transistor
d MOS transistor model with a gate-to-drain impedance Z
IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373 367
doi: 10.1049/iet-cds.2010.0030 & The Institution of Engineering and Technology 2010
www.ietdl.org
order of the cascade yielding
1 0
sC
gs
1
_ _
·
a
11
a
12
a
21
a
22
_ _
·
1 0
sC
ds
1
_ _
=
a
11
+sC
ds
a
12
a
12
a
21
+sC
gs
(a
11
+sC
ds
a
12
) +sC
ds
a
22
a
22
+sC
gs
a
12
_ _
(8)
It is thus clear that the complexity of the transistor model
automatically reflects in the elements of the matrix. As
such, a two-port network analysis of the cross-coupled
structure of Fig. 1c would render a general characteristic
equation which
1. is independent of any particular active device model;
2. can automatically absorb any level of complexity in the
active device model without needing to re-analyse the
oscillator to derive a new characteristic equation every time
the model is changed;
3. can be used to study the possibility of having non-
matched and non-balanced oscillators; which is the target
of this work.
It is also worth noting that the (a) matrix elements can be
experimentally found via a network analyser.
3 Four-impedance cross-coupled
structure
Consider the two-port network representation of Fig. 1c,
shown in Fig. 3a. Assuming the two transistors are not
similar (This is more general than assuming they are not
matched.) implies they are, respectively, represented by the
distinct transmission matrices (a) and (b); as defined by (1).
Noting from Fig. 3a that V
1
= V
4
, V
2
= V
3
and
I
1
+I
2
+I
3
+I
4
= 0, it can then be shown that a
characteristic equation of Fig. 3a exists only if Z
1
1
and is given by (Z
1
is therefore usually replaced by a DC
biasing current source.)
1 +Z
3
1
Z
4
+
b
11
b
12
+
a
22
a
12
_ _
+Z
2
1
Z
4
+
1
a
12
+
b
11
b
22
b
12
−b
21
_ _
1 +Z
3
1
Z
4
+
1
b
12
+
a
11
a
22
a
12
−a
21
_ _
+Z
2
1
Z
4
+
b
22
b
12
+
a
11
a
12
_ _
= −
b
21
+
b
11
(1 −b
22
)
b
12

1 −a
22
a
12
a
21
+
a
11
(1 −a
22
)
a
12

1 −b
22
b
12
(9)
Figure 3 Two-port network models of
a Four-impedance cross-coupled oscillator structure
b Six-impedance structure
368 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0030
www.ietdl.org
If the transistors are matched and identically biased; that is
(a) ¼ (b) the above equation reduces to
1 +(Z
2
+Z
3
)
1
Z
4
+
1 +a
11
+a
22
+a
11
a
22
2a
12

a
21
2
_ _
= 0
(10)
Without loss of generality, we consider the transmission
matrix of (6) which applies both for the BJT and MOS
transistors under ideal conditions. This does not undermine
the generality of our analysis since we may instead use the
more complex matrices of (2), (5), (7) or (8).
Substituting with (6) in (9) yields
1 +
Z
2
+Z
3
Z
4

g
ma
g
mb
g
ma
+g
mb
(Z
2
+Z
3
) = 0 (11)
where g
ma
and g
mb
are the transconductances of the two
transistors. Interestingly, the characteristic equation (11) is
dependent on the effective transconductance of the two
transistors g
meff
= g
ma
g
mb
/(g
ma
+g
mb
). Now, if the two
transistors are matched and have balanced DC biasing then
g
ma
= g
mb
= g
m
and (11) simplifies to
1 +(Z
2
+Z
3
)
1
Z
4

g
m
2
_ _
= 0 (12)
Obviously the choice of g
ma
= g
mb
= g
m
may simplify the
design but it is not a necessary choice; that is there is no
reason why the two transistors have to be strictly matched.
Reverting back to (11) four different cases of cross-coupled
oscillators can be designed. In particular:
Case A: matched devices (g
ma
= g
mb
= g
m
) with balanced
loads (Z
2
= Z
3
).
Case B: matched devices with non-balanced loads (Z
2
=Z
3
).
Case C: non-matched devices (g
ma
=g
mb
) with balanced
loads.
Case D: non-matched devices with non-balanced loads.
In what follows we give possible design examples of each
case. However, before doing so it is also worth
commenting on the six-impedance cross-coupled structure
shown in Fig. 1d. Here, two more impedances Z
5
and Z
6
are added to complete all possible inter-node connections
of the two-port representation of Fig. 3b. Because of using
transmission parameters, one does not need to re-analyse
this six-impedance structure. In particular, the characteristic
equation (9), derived for the four-impedance case, will
remain valid taking into consideration that the (a) and (b)
matrices have to be modified as
(a)
new
=
1 0
Z
5
1
_ _
· (a)
old
=
a
11
a
12
a
21
+a
11
Z
5
a
22
+a
12
Z
5
_ _
(13)
(b)
new
=
1 0
Z
6
1
_ _
· (b)
old
=
b
11
b
12
b
21
+b
11
Z
6
b
22
+b
12
Z
6
_ _
(14)
4 Design Examples
4.1 Cross-coupled oscillators with
balanced loads
Fig. 1a and b, which are conventional cross-coupled oscillators,
represent examples of balanced load oscillators whichmay either
belong to case A or case C above. Fig. 1a has Z
2
= Z
3
= Z =
(sL +r)/(1 +rCs +LCs
2
) andZ
4
= 1. Hence using (11) one
obtains the characteristic equation
1 −2g
meff
Z = 0 (15)
or equivalently
s
2
+
r
L
−2
g
meff
C
_ _
s +
1
LC
(1 −2rg
meff
) = 0 (16)
Figure 4 Examples of non-balanced cross-coupled
oscillators
a Second-order oscillator
b Third-order oscillator
IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373 369
doi: 10.1049/iet-cds.2010.0030 & The Institution of Engineering and Technology 2010
www.ietdl.org
The oscillation start-up condition and oscillation frequency are
thus given by
g
meff
=
1
2
rC
L
and
v
0
=
1
....
LC

.............
1 −
r
2
C
L
_ _
¸
¸
¸
_
=
1
....
LC
√ |
r0
(17)
In the special case of two matched and symmetrically biased
transistors (g
ma
= g
mb
= g
m
), the above start-up condition
simply reduces to g
m
= rC/L maintaining the same
oscillation frequency; as widely known.
For Fig. 1b we have Z
2
= Z
3
= Z = sL +r and
Z
4
= 1/sC. Using (11) one obtains the characteristic equation
1 +2Z
1
Z
4
−g
meff
_ _
= 0 (18)
or equivalently
s
2
+
r
L

g
meff
C
_ _
s +
1
LC
1
2
−rg
meff
_ _
= 0 (19)
The oscillation start-up condition and oscillation
frequency are thus
g
meff
=
rC
L
and
v
0
=
1
....
LC

.............
1
2

r
2
C
L
_ _
¸
¸
¸
_
=
1
.....
2LC
√ |
r0
(20)
Again, if the two transistors are matched and
symmetrically biased the above start-up condition
reduces to g
m
= 2rC/L maintaining the same oscillation
frequency.
Table 1 All possible non-balanced second-order cross-coupled oscillators (R
eff
¼ rR/(r +R), C
eff
¼ C
2
C
3
/(C
2
+C
3
) and
R
t
¼ r +R)
Case Z
2
Z
3
Z
4
g
meff
v
0
Remarks
1 1/sC
2
1/sC
3
r +sL rC
eff
L
1
.....
LC
eff

...........
1 −
r
2
C
eff
L
_
2 1/sC
2
1/sC
3
R//(r +sL) rC
eff
L
+
1
R
1
.....
LC
eff

..............
1 −
r
2
C
eff
L
_ _
_
3 1/sC r +sL R 1
R
eff
1
...
LC

4 1/sC
2
R +1/sC
3
r +sL 1
R
eff
+
L
R
t
C
eff
1
.....
LC
eff

............
r
2
−L/C
eff
R
2
−L/C
eff
_
5 1/sC R r +sL 1
R
eff
+
L
R
t
C
1
...
LC

..........
r
2
−L/C
R
2
−L/C
_
6 r +sL 1/sC R 1
R
eff
1
...
LC

mirror of case 3
7 R +1/sC
2
1/sC
3
r +sL 1
R
eff
+
L
R
t
C
eff
1
.....
LC
eff

............
r
2
−L/C
eff
R
2
−L/C
eff
_
mirror of case 4
8 R 1/sC r +sL 1
R
eff
+
L
R
t
C
1
...
LC

..........
r
2
−L/C
R
2
−L/C
_
mirror of case 5
9 R r +sL 1/sC CR
t
L
1
...
LC

.........
1 −
R
2
t
C
L
_
circuit in Fig. 4a
10 r +sL R 1/sC CR
t
L
1
...
LC

.........
1 −
R
2
t
C
L
_
mirror of Fig. 4a
370 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0030
www.ietdl.org
4.2 Cross-coupled oscillators with
non-balanced loads
Non-balanced load oscillators imply that Z
2
=Z
3
. Two
novel examples are given in Fig. 4. Consider the circuit of
Fig. 4a where Z
2
= R, Z
3
= sL +r and Z
4
= 1/sC. Using
(11) we obtain the characteristic equation
s
2
+
R
t
C −g
meff
L
LC
_ _
s +
1 −g
meff
R
t
LC
= 0 (21)
where R
t
= R +r. The start-up condition and oscillation
frequency are thus g
meff
= R
t
C/L and v
0
=
(1
_ ....
LC

)
...............
1 −(R
2
t
C/L)
_
.
For the second example of Fig. 4b, we have Z
2
= sL +r,
Z
3
= R/(1 +RCs) and Z
4
= 1/sC. Using (11) then yields
the third-order characteristic equation
s
3
+
1
RC
2
+
r
LC

g
meff
C
2
_ _
s
2
+
2 +(r/R) −g
meff
r
LC

g
meff
RC
2
_ _
s
+
1 −g
meff
(R+r)
LRC
2
=0 (22)
from which the start-up condition at r 0 is g
meff
= RC/L
while the oscillation frequency is v
0
= 1/
....
LC

. It is clearly
not necessary in any of the circuits of Fig. 4 to have two
matched transistors.
The examples given in Fig. 4 are not the only possible non-
balanced cross-coupled oscillators. In particular, using (11), a
search for all possible Z
2
, Z
3
and Z
4
which yield a positive
start-up value for g
meff
and positive v
0
can be automated
through a Matlab code. We have written this code in search
for all valid cases which yield a second-order characteristic
equation while using a maximum of two resistors, two
capacitors or two inductors. The obtained results are given
in Table 1. A similar code could be written in search for
third-order cases or to impose other constraints.
4.3 Experimental and simulation results
The oscillator of Fig. 4a was experimentally constructed using
two different Bipolar transistors of types Q2N2222 and
BC109. This is to ensure that no matching or even similarity
is there. The components used where C = 50 pF,
L = 120 mH (with internal resistance r = 1.5 V),
R = 500 V and V
CC
= 5 V. Oscillations started at a biasing
current I = 85 mA, which was found to be split unequally
between the two transistors such that I
a
= 29 mA and
I
b
= 56 mA. Consulting (21), the ideal start-up condition
would be g
meff
≃ 0.2 mA/V while oscillations actually started
at g
meff
≃ 0.7 mA/V. The waveform observed at the L–C
connection node is shown in Fig. 5 with a measured
oscillation frequency f
0
≃ 1.5 MHz which is close enough to
the 1.9 MHz indicated by (21) given the discrete component
tolerances and capacitance of the measuring probe.
For the third-order oscillator of Fig. 4b, Spice simulations
using 0.25 m CMOS process parameters were used to verify its
function. We also choose to demonstrate this oscillator works
with non-matched transistors by setting the aspect ratios as
(W/L)
a
= (0.5 m/2 m), and (W/L)
b
= (1 m/1 m). The other
components were selected to be L = 1 mH, r = 15 V,
R = 1 kV and C = 1 pF. The biasing current is I = 10 mA
and the power supply is V
CC
= 1.5 V. Fig. 6a shows the
waveforms at the drain of M
a
and also across the
floating capacitor (drain voltage of M
a
– drain voltage of M
b
)
while Fig. 6b shows the currents in the two transistors; which
are clearly not symmetrical. The oscillation frequency is
158 MHz which is very close to the 1/
....
LC

frequency of
159.15 MHz.
4.4 Parasitic effects
To demonstrate how parasitic effects can be incorporated into
the analysis of cross-coupled oscillators using the derived
two-port network equations, let us consider the general
oscillator structure of Fig. 1c. For simplicity, we will
consider the case of two matched and identically biased
MOS transistors with r
ds
1 and r
s
0. Then using
(7) and (8) with Z = 1/sC
gd
, the transmission
matrix modelling the MOS transistor including the C
gs
,
C
ds
and C
gd
parasitics with g
m
as the only design parameter
becomes
a
11
a
12
a
21
a
22
_ _
=
1
1 −(g
m
/sC
gd
)
×
1 +
C
ds
C
gd
1
sC
gd
g
m
+s C
gs
+C
ds
+
C
gs
C
ds
C
gd
_ _
1 +
C
ds
C
gd












(23)
Figure 5 Experimental waveform measured from the
constructed oscillator of Fig. 4a using two non-similar
Bipolar transistors
IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373 371
doi: 10.1049/iet-cds.2010.0030 & The Institution of Engineering and Technology 2010
www.ietdl.org
Substituting with the above matrix coefficients in (9) yields
the characteristic equation
(Z
2
+Z
3
)
2(sC
gd
−g
m
)
Z
4
+q
2
C
2
gd
s
2
−q
1
g
m
C
gd
s +g
2
m
_ _
+2(sC
gd
−g
m
) = 0 (24)
where q
1
= 5 +(2C
ds
/C
gd
) and q
2
= 4 +3q
1
+q
2
1
+ [(q
1

1)C
gs
/C
gd
]. Note that (24) reduces to (12) at C
gd
= 0. This
example emphasises the generality of the characteristic (9).
5 Conclusion
Although many researchers have targeted the analysis and
design of cross-coupled oscillators, theoretical analysis using
two-port network techniques was lacking. This work has
addressed this issue in favour of general design equations not
restricted to matched devices. It is important however to
remember that two-port network analysis is a linear system
analysis technique and hence the oscillation conditions
derived are necessary but not sufficient conditions. In
particular, some oscillators may latchup and never oscillate
Figure 6 Spice simulations of the non-balanced oscillator of Fig. 4b
a Output waveform at the drain of M
a
and across the floating capacitor C
b Drain currents in the two transistors
372 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0030
www.ietdl.org
[12] and in this case one has to revert to non-linear analysis in
order to explain the reason for the latchup and fix it if possible
[13]. Detailed analysis of the latch-up behaviour in cross-
coupled oscillators using higher-order non-linear models can
be found in [14].
6 References
[1] CRANINCKX J., STEYAERT M.: ‘A 1.8-GHz low-phase-noise
CMOS VCO using optimized hollow spiral inductors’, IEEE
J. Solid-State Circuits, 1997, 32, pp. 736–744
[2] JIA L., MA J., YEOK., DOI M.: ‘9.3-10.4 GHz band cross-coupled
complementary oscillator with lowphase noise performance’,
IEEE Trans. Microw. Theory, 2004, 52, pp. 1273–1278
[3] ZHAN J., MAURICE K., DUSTER J., KORNEGAY K.: ‘Analysis and
design of negative impedance LC oscillators using bipolar
transistors’, IEEE Trans. Circuits Syst., 2003, 52,
pp. 1461–1464
[4] OZOGUZ S., ELWAKIL A.S., ERGUN S.: ‘Cross-coupled chaotic
oscillators and application to random bit generation’,
Proc. IEE Circuits Devices Syst., 2006, 153, pp. 506–510
[5] YANG C.L., CHIANG Y.C.: ‘Low phase-noise and low-power
CMOS VCO constructed in current-reused configuration’,
IEEE Microw. Wirel. Compon. Lett., 2008, 18, pp. 136–138
[6] SHEKHAR S., WALLING J., ANIRUDDHAN S., ALLSTOT D.: ‘CMOS VCO
and LNA using tuned-input tuned-output circuits’, IEEE
J. Solid-State Circuits, 2008, 43, pp. 1177–1186
[7] HSU M., CHIU C., CHEN S.: ‘Low power design of CMOS 5-
GHz voltage- controlled oscillation from narrowband to
wideband switching capacitor module’, Proc. IET Circuits
Devices Syst., 2009, 43, pp. 99–105
[8] BUONOMO A., LO SCHIAVO A.: ‘Finding the tuning curve of a
CMOS-LC VCO’, IEEE Trans. Circuits Syst., 2008, 55,
pp. 887–891
[9] BUONOMO A.: ‘Nonlinear analysis of voltage-controlled
oscillators: a systematic approach’, IEEE Trans. Circuits
Syst., 2008, 55, pp. 1659–1670
[10] DJURHUUS T., KROZER V., VIDKJAER J., JOHANSEN T.: ‘Nonlinear
analysis of a cross-coupled quadrature harmonic oscillator’,
IEEE Trans. Circuits Syst., 2005, 52, pp. 2276–2285
[11] ELWAKIL A.S.: ‘On the two-port network classification of
Colpitts oscillators’, Proc. IET Circuits Devices Syst., 2009,
3, (5), pp. 223–232
[12] ELWAKIL A.S., AHMED W.M.: ‘On the necessary and sufficient
conditions for latch-up in sinusoidal oscillators’,
Int. J. Electron., 2002, 89, pp. 197–206
[13] ELWAKIL A.S.: ‘Explaining and eliminating latchup in
a classical Wien oscillator via nonlinear design’, J. Analog
Integr. Circuits Signal Process., 2006, 48, pp. 239–245
[14] ELWAKIL A.S., SALAMA K.N.: ‘Higher dimensional models of
cross-coupled oscillators and application to design’,
J. Circuits Syst. Computers, (Special issue on ‘Advances in
Oscillator Analysis and Design’) 2010, 19, (4), pp. 787–799
IET Circuits Devices Syst., 2010, Vol. 4, Iss. 5, pp. 365–373 373
doi: 10.1049/iet-cds.2010.0030 & The Institution of Engineering and Technology 2010
www.ietdl.org

org Figure 1 Cross-coupled oscillator structures a and b Two classical balanced-load cross-coupled oscillators c Generic structure with four impedances d Generic structure with six impedances defined as V1 I1 = a11 a21 a12 a22 V2 −I2 (1) bI1 with output resistance ro . pp. 2010. Fig.www. The model comprises two terminal resistors (rb .0030 . It can be shown that the (a) matrix for this equivalent circuit is [11] a11 a21 a12 a22 = −1 bro − re rb + re 1 where V1 and I1 are the input port voltage and current.1049/iet-cds. 365– 373 doi: 10.. Iss. 2b shows a basic small signal equivalent circuit of an NPN BJT transistor operating in the forward active mode. re ) (ideally re = 0) and a dependent current-controlled current source 366 & The Institution of Engineering and Technology 2010 bre ro + rb (ro + re ) ro + re (2) The above (a) may be simplified considering that bro ≫ re IET Circuits Devices Syst.ietdl. 4. Vol. respectively. while V2 and I2 are the output port voltage and current.2010. 5. b is the forward active current gain (assumed constant) and rb is related to the BJT small-signal transconductance gm as rb = b/gm .

1) for the MOS. It is worth noting that under the ideal conditions IET Circuits Devices Syst. ro . However.0030 & The Institution of Engineering and Technology 2010 . a further simplification yields ⎛ a11 a21 a12 a22 ⎜0 =⎜ ⎝ 0 ⎞ 1 − re + gm ⎟ ⎟ ⎠ −1 b where the transconductance gm is the only design parameter. it can be shown that the (a) matrix for this circuit is a11 a21 a12 a22 −1 1 rs (1 + gm rds ) + rds gm rds 0 0 ⎛ ⎞ −1 1 − rs + ≃ ⎝ gm rds gm ⎠ 0 0 = = (5) where gm is the small signal transconductance.ietdl. rds is the drain to source resistance and rs is the source resistance (ideally rs = 0).. we may additionally incorporate the effects of Cgs and Cds (treating each as an individual two-port network) via matrix multiplication with (7) in the 367 (4) In a similar manner. b) = (0. 1. If A is sufficiently large such that (1/A) 0. 2010. 1) for the BJT and (rs .2010. rds ) = (0. 4. and since transmission parameters are well suited for networks in cascade. pp.www. Consider for example the MOS transistor small signal model of Fig. 2c).1049/iet-cds. and recalling the small signal equivalent circuit of a MOS transistor operating in the saturation mode (Fig. The transmission matrix in this case can be shown to be a11 a21 a12 a22 1 rds (1 − gm Z) + rs × rds (1 + gm rs ) + Z 1 + gm rds rds (1 + gm rs )Z rds (1 + gm rs ) (7) Further. both transistors can be described by the ideal transmission matrix (3) a11 a21 a12 a22 = 0 −1/gm 0 0 a11 a21 a12 a22 (6) where A = gm ro . Iss. more complex transistor models can also be accommodated for. 365– 373 doi: 10. Vol.org Figure 2 Two-port network representations a b c d Two-port network variables Basic small signal equivalent model of a BJT transistor Basic small signal equivalent model of a MOS transistor MOS transistor model with a gate-to-drain impedance Z and (1/bro ) 0 and then becomes −1 ⎜ A =⎜ ⎝ 0 ⎛ − re 1 + ⎞ 1 1 + A gm ⎟ ⎟ ⎠ −1 b (re . 2d where an impedance Z (typically a parasitic Cgd capacitor) is connected from gate to drain. 5.

5. 2.) 1 + Z3 1 + Z3 1 b a 1 1 b b + 11 + 22 + Z2 + + 11 22 − b21 b12 Z4 b12 a12 Z4 a12 1 1 a11 a22 1 b a + + − a21 + Z2 + 22 + 11 a12 Z4 b12 Z4 b12 a12 b21 + It is thus clear that the complexity of the transistor model automatically reflects in the elements of the matrix. Noting from Fig. 1c would render a general characteristic equation which 1. as defined by (1). pp.2010. 3a..) implies they are. 2010. respectively.ietdl.0030 . 3a exists only if Z1 and is given by (Z1 is therefore usually replaced by a DC biasing current source. Assuming the two transistors are not similar (This is more general than assuming they are not matched. can be used to study the possibility of having nonmatched and non-balanced oscillators. represented by the distinct transmission matrices (a) and (b). V2 = V3 and I1 + I2 + I3 + I4 = 0. 1 0 sCds 1 1 0 sCgs 1 = · a11 a12 a21 a22 · a11 + sCds a12 a12 a21 + sCgs (a11 + sCds a12 ) + sCds a22 a22 + sCgs a12 (8) 3 Four-impedance cross-coupled structure Consider the two-port network representation of Fig. 3a that V1 = V4 . 365– 373 doi: 10. 4. a two-port network analysis of the cross-coupled structure of Fig.1049/iet-cds. Vol. can automatically absorb any level of complexity in the active device model without needing to re-analyse the oscillator to derive a new characteristic equation every time the model is changed. shown in Fig. 1c. it can then be shown that a 1 characteristic equation of Fig. 3.org order of the cascade yielding It is also worth noting that the (a) matrix elements can be experimentally found via a network analyser. b11 (1 − b22 ) 1 − a22 − a12 b12 =− a11 (1 − a22 ) 1 − b22 a21 + − b12 a12 (9) Figure 3 Two-port network models of a Four-impedance cross-coupled oscillator structure b Six-impedance structure 368 & The Institution of Engineering and Technology 2010 IET Circuits Devices Syst. is independent of any particular active device model. Iss. As such.www. which is the target of this work.

Iss.0030 Figure 4 Examples oscillators of non-balanced cross-coupled a Second-order oscillator b Third-order oscillator 369 & The Institution of Engineering and Technology 2010 . 2010.1 Cross-coupled oscillators with balanced loads Fig. Case C: non-matched devices (gma = gmb ) with balanced loads. Because of using transmission parameters. the characteristic equation (11) is dependent on the effective transconductance of the two transistors gmeff = gma gmb /(gma + gmb ). Vol. Fig. derived for the four-impedance case. if the two transistors are matched and have balanced DC biasing then gma = gmb = gm and (11) simplifies to 1 + (Z2 + Z3 ) 1 g − m 2 Z4 =0 (12) Obviously the choice of gma = gmb = gm may simplify the design but it is not a necessary choice. which are conventional cross-coupled oscillators.ietdl. 1a and b. Interestingly. 1a has Z2 = Z3 = Z = (sL + r)/(1 + rCs + LCs2 ) and Z4 = 1. 365– 373 doi: 10. In particular: Case A: matched devices (gma = gmb = gm ) with balanced loads (Z2 = Z3 ).www. Case D: non-matched devices with non-balanced loads. we consider the transmission matrix of (6) which applies both for the BJT and MOS transistors under ideal conditions. Case B: matched devices with non-balanced loads (Z2 = Z3 ). Hence using (11) one obtains the characteristic equation 1 − 2gmeff Z = 0 or equivalently s2 + r g 1 − 2 meff s + (1 − 2rgmeff ) = 0 C L LC (16) (15) where gma and gmb are the transconductances of the two transistors. Now. two more impedances Z5 and Z6 are added to complete all possible inter-node connections of the two-port representation of Fig. However. Here. (7) or (8). one does not need to re-analyse this six-impedance structure. 4. (5). This does not undermine the generality of our analysis since we may instead use the more complex matrices of (2). In particular. In what follows we give possible design examples of each case.1049/iet-cds. 5. that is (a) ¼ (b) the above equation reduces to 1 1 + a11 + a22 + a11 a22 a21 + − 1 + (Z2 + Z3 ) 2a12 2 Z4 =0 (10) Without loss of generality. will remain valid taking into consideration that the (a) and (b) IET Circuits Devices Syst.2010. the characteristic equation (9). pp. 1d. Reverting back to (11) four different cases of cross-coupled oscillators can be designed.. 3b. that is there is no reason why the two transistors have to be strictly matched. before doing so it is also worth commenting on the six-impedance cross-coupled structure shown in Fig. Substituting with (6) in (9) yields 1+ Z2 + Z3 g g − ma mb (Z2 + Z3 ) = 0 Z4 gma + gmb (11) matrices have to be modified as (a)new = = (b)new = = 1 Z5 0 1 · (a)old a12 a22 + a12 Z5 (13) a11 a21 + a11 Z5 1 Z6 0 1 · (b)old b12 b22 + b12 Z6 (14) b11 b21 + b11 Z6 4 Design Examples 4.org If the transistors are matched and identically biased. represent examples of balanced load oscillators which may either belong to case A or case C above.

if the two transistors are matched and symmetrically biased the above start-up condition reduces to gm = 2rC/L maintaining the same oscillation frequency. Table 1 All possible non-balanced second-order cross-coupled oscillators (Reff ¼ rR/(r + R).org The oscillation start-up condition and oscillation frequency are thus given by gmeff = 1 rC 2L 1 LC and 1− r2C L =√ 1 |r LC or equivalently s2 + r gmeff 1 1 s+ − − rgmeff C L LC 2 condition and =0 (19) v0 = √ 0 (17) The oscillation start-up frequency are thus gmeff = rC L and 1 LC oscillation In the special case of two matched and symmetrically biased transistors (gma = gmb = gm ). Using (11) one obtains the characteristic equation 1 + 2Z 1 − gmeff Z4 =0 (18) v0 = √ 1 r2C − 2 L =√ 1 |r 2LC 0 (20) Again. Vol.ietdl. 2010. the above start-up condition simply reduces to gm = rC/L maintaining the same oscillation frequency. Ceff ¼ C2C3/(C2 + C3) and Rt ¼ r + R) Case 1 Z2 1/sC2 1/sC2 Z3 1/sC3 1/sC3 r + sL R + 1/sC3 Z4 r + sL R//(r + sL) gmeff rCeff L rCeff 1 + L R 1 Reff 1 Reff + 5 1/sC R r + sL L Rt Ceff √ √ √ v0 1 r 2 Ceff 1− LCeff L Remarks 2 1 r 2 Ceff 1− L LCeff 1 √ LC 1 r 2 − L/Ceff LCeff R2 − L/Ceff 3 1/sC R r + sL 4 1/sC2 1 L Reff + Rt C 1 r 2 − L/C √ LC R2 − L/C 1 √ LC √ 1 r 2 − L/Ceff LCeff R2 − L/Ceff mirror of case 3 6 r + sL R + 1/sC2 1/sC R r + sL 1 Reff 1 L Reff + Rt Ceff 7 1/sC3 mirror of case 4 8 R 1/sC r + sL 1 L Reff + Rt C 1 r 2 − L/C √ LC R2 − L/C 1 R2 C √ 1− t L LC 1 R2 C √ 1− t L LC mirror of case 5 9 R r + sL r + sL R 1/sC CRt L CRt L circuit in Fig. Iss. 4.0030 . 1b we have Z2 = Z3 = Z = sL + r and Z4 = 1/sC. 365– 373 doi: 10. as widely known. 4a 370 & The Institution of Engineering and Technology 2010 IET Circuits Devices Syst. 4a 10 1/sC mirror of Fig.2010. 5.www.1049/iet-cds.. For Fig. pp.

5 V). the ideal start-up condition would be gmeff ≃ 0. Then using MOS transistors with rds (7) and (8) with Z = 1/sCgd . IET Circuits Devices Syst. 4. Z3 and Z4 which yield a positive start-up value for gmeff and positive v0 can be automated through a Matlab code. 5. The examples given in Fig.5 MHz which is close enough to the 1.7 mA/V. 5 with a measured oscillation frequency f0 ≃ 1. The oscillation frequency is √ 158 MHz which is very close to the 1/ LC frequency of 159. The obtained results are given in Table 1.ietdl.5 V . Iss.www. and (W /L)b = (1 m/1 m). Oscillations started at a biasing current I = 85 mA. r = 15 V.0030 ⎜ ⎜ ×⎜ ⎜ ⎝ g + s C + C + Cgs Cds m gs ds Cgd Cds Cgd ⎞ 1 sCgd ⎟ ⎟ ⎟ Cds ⎟ ⎠ 1+ Cgd (23) 371 & The Institution of Engineering and Technology 2010 . Using (11) we obtain the characteristic equation s2 + Rt C − gmeff L 1 − gmeff Rt =0 s+ LC LC (21) where Rt = R + r. The waveform observed at the L–C connection node is shown in Fig. Spice simulations using 0. Consider the circuit of Fig. 1c. A similar code could be written in search for third-order cases or to impose other constraints.25 m CMOS process parameters were used to verify its function. Using (11) then yields the third-order characteristic equation s3 + + 1 r g 2 + (r/R) − gmeff r gmeff − meff s2 + − s + 2 2 C RC 2 RC LC LC 1 − gmeff (R + r) =0 LRC 2 (22) Figure 5 Experimental waveform measured from the constructed oscillator of Fig. 4. R = 1 kV and C = 1 pF. 4. 6a shows the waveforms at the drain of Ma and also across the floating capacitor (drain voltage of Ma – drain voltage of Mb) while Fig.3 Experimental and simulation results The oscillator of Fig. Vol.2 mA/V while oscillations actually started at gmeff ≃ 0. We have written this code in search for all valid cases which yield a second-order characteristic equation while using a maximum of two resistors.2010.5 m/2 m). 4a using two non-similar Bipolar transistors from which the start-up condition at r 0 is gmeff = RC/L √ while the oscillation frequency is v0 = 1/ LC . 4a was experimentally constructed using two different Bipolar transistors of types Q2N2222 and BC109.9 MHz indicated by (21) given the discrete component tolerances and capacitance of the measuring probe. a search for all possible Z2 . 4b. R = 500 V and VCC = 5 V.15 MHz. 4a where Z2 = R.. For simplicity. Z3 = sL + r and Z4 = 1/sC. using (11). In particular. pp. For the third-order oscillator of Fig. Consulting (21). We also choose to demonstrate this oscillator works with non-matched transistors by setting the aspect ratios as (W /L)a = (0. 4b. The start-up condition and oscillation and v0 = frequency are thus gmeff = Rt C/L √ 2 (1 LC ) 1 − (Rt C/L). 2010. 4 to have two matched transistors. the transmission matrix modelling the MOS transistor including the Cgs . which was found to be split unequally between the two transistors such that Ia = 29 mA and Ib = 56 mA.4 Parasitic effects To demonstrate how parasitic effects can be incorporated into the analysis of cross-coupled oscillators using the derived two-port network equations.org 4.2 Cross-coupled oscillators with non-balanced loads Non-balanced load oscillators imply that Z2 = Z3 . L = 120 mH (with internal resistance r = 1. two capacitors or two inductors. This is to ensure that no matching or even similarity is there. The biasing current is I = 10 mA and the power supply is VCC = 1. Cds and Cgd parasitics with gm as the only design parameter becomes a11 a12 a21 a22 = 1 1 − (gm /sCgd ) ⎛ 1+ 4. 4 are not the only possible nonbalanced cross-coupled oscillators. Two novel examples are given in Fig. 365– 373 doi: 10.1049/iet-cds. 6b shows the currents in the two transistors. which are clearly not symmetrical. we will consider the case of two matched and identically biased 1 and rs 0. we have Z2 = sL + r. For the second example of Fig. The other components were selected to be L = 1 mH. It is clearly not necessary in any of the circuits of Fig. let us consider the general oscillator structure of Fig. Fig. The components used where C = 50 pF. Z3 = R/(1 + RCs) and Z4 = 1/sC.

some oscillators may latchup and never oscillate IET Circuits Devices Syst. Although many researchers have targeted the analysis and design of cross-coupled oscillators. Note that (24) reduces to (12) at Cgd = 0.0030 372 & The Institution of Engineering and Technology 2010 . It is important however to remember that two-port network analysis is a linear system analysis technique and hence the oscillation conditions derived are necessary but not sufficient conditions.www.2010. 5. theoretical analysis using two-port network techniques was lacking.. 4b a Output waveform at the drain of Ma and across the floating capacitor C b Drain currents in the two transistors Substituting with the above matrix coefficients in (9) yields the characteristic equation (Z2 + Z3 ) 2(sCgd − gm ) Z4 2 2 + q2 Cgd s2 − q1 gm Cgd s + gm 5 Conclusion + 2(sCgd − gm ) = 0 (24) 2 where q1 = 5 + (2Cds /Cgd ) and q2 = 4 + 3q1 + q1 + [(q1 − 1)Cgs /Cgd ]. 365– 373 doi: 10.ietdl.1049/iet-cds. 2010. 4. This work has addressed this issue in favour of general design equations not restricted to matched devices. In particular. Iss. Vol. pp. This example emphasises the generality of the characteristic (9).org Figure 6 Spice simulations of the non-balanced oscillator of Fig.

Circuits Syst. 3.: ‘On the two-port network classification of Colpitts oscillators’.3-10. pp. ERGUN S..: ‘On the necessary and sufficient conditions for latch-up in sinusoidal oscillators’. IEEE Trans. 52. YEO K. Wirel. Circuits Signal Process.. 365– 373 doi: 10.. CHEN S. Proc. 43. Computers. 2276–2285 [11] ELWAKIL A.: ‘Finding the tuning curve of a CMOS-LC VCO’.8-GHz low-phase-noise CMOS VCO using optimized hollow spiral inductors’. 2006.: ‘A 1. Circuits Syst.: ‘Higher dimensional models of cross-coupled oscillators and application to design’. Compon. J. IEEE Trans. WALLING J.. 887– 891 [9] BUONOMO A. Solid-State Circuits.M. 19.C. 2008. (4)..: ‘Analysis and design of negative impedance LC oscillators using bipolar transistors’... 153. 52. Theory. Analog Integr. Proc.N. Electron. CHIU C.controlled oscillation from narrowband to wideband switching capacitor module’. 48. pp.... SALAMA K.: ‘Explaining and eliminating latchup in a classical Wien oscillator via nonlinear design’. 2004. 43. 1997. pp. 736– 744 [2] JIA L.ietdl. IEEE Trans.S.S.: ‘Low power design of CMOS 5GHz voltage. 2008. pp. pp. (Special issue on ‘Advances in Oscillator Analysis and Design’) 2010.L. KROZER V.0030 373 & The Institution of Engineering and Technology 2010 . pp. 55. 4. 2005. 223 – 232 [12] ELWAKIL A.. Circuits Syst.4 GHz band cross-coupled complementary oscillator with low phase noise performance’. (5). 55. 239 – 245 [14] ELWAKIL A. Microw. 2002.. 1461 – 1464 [4] OZOGUZ S.. ALLSTOT D. pp. Detailed analysis of the latch-up behaviour in crosscoupled oscillators using higher-order non-linear models can be found in [14]. IEEE Microw.. IET Circuits Devices Syst. LO SCHIAVO A.S. Circuits Syst. ELWAKIL A. 52.. MA J. J. Solid-State Circuits. Lett. 2009.S. 18. Proc. 2003.. 197– 206 [13] ELWAKIL A..org [12] and in this case one has to revert to non-linear analysis in order to explain the reason for the latchup and fix it if possible [13]... 1273–1278 [3] ZHAN J.. pp. AHMED W. VIDKJAER J. IEEE Trans. [7] HSU M. DOI M. pp. 2010. 787– 799 6 References [1] CRANINCKX J. 5. 136 – 138 [6] SHEKHAR S. 2008. J. Iss.: ‘Nonlinear analysis of voltage-controlled oscillators: a systematic approach’. 89. IEEE J. 99– 105 [8] BUONOMO A.: ‘Cross-coupled chaotic oscillators and application to random bit generation’... 2006. pp. IEEE Trans. 2008.2010. STEYAERT M..: ‘9...... 32. IET Circuits Devices Syst.www.. 1177– 1186 IET Circuits Devices Syst. pp. 2009. pp. Int.1049/iet-cds. pp. pp.S. IEEE J. 506 – 510 [5] YANG C.: ‘Nonlinear analysis of a cross-coupled quadrature harmonic oscillator’. Circuits Syst. DUSTER J. ANIRUDDHAN S. CHIANG Y.. Vol. JOHANSEN T. pp. IEE Circuits Devices Syst. 1659 – 1670 [10] DJURHUUS T.: ‘Low phase-noise and low-power CMOS VCO constructed in current-reused configuration’..: ‘CMOS VCO and LNA using tuned-input tuned-output circuits’.. MAURICE K. KORNEGAY K.