Variable Negative Gm Technique for RF LC VCO

with Very Large Tuning Range

H. Shi, J. H. Liu, G. Y. Zhang, H. L. Liao, R. Huang, Y. Y. Wang

Abstract- When the tuning range of the RF LC

VCO gets very large, the VCO start up condition, II. ANALYSIS OF VCO WITH VERY
VCO oscillation swing and phase noise performance LARGE TUNING RANGE
will vary obviously. These issues result from the fact
that the equivalent parallel resistance of the LC tank
varies a lot due to very large tuning range. In this LC tank Active circuit
_ _ _ _ _ _ _ _ _ _I
paper, we propose a variable negative Gm technique
to solve these problems. From simulation results, it is I
I I I
found that by using this technique the VCO start up I ..-" I I I
condition at different frequency bands can be I M.,*-', I~~~~~~~~~~~~~~~~~~~~~
IIL-gactiv I
independently met. The variation of the VCO I I ¶C Igtank ;
I
II
oscillation swing can be effectively reduced. The phase I

L- -1 L _ __ _ l
noise fluctuations between different bands can be
- - - - - - -

decreased close to the intrinsic level. Fig. 1 A negative Gm view of the RF LC VCO.
I. INTRODUCTION A negative Gm view of the RF LC VCO is
illustrated in Fig.l. The VCO consists of two parts: the
The design of RF LC VCO has been extensively LC tank and the active circuit. In Fig. 1 gtan, represents the
discussed in papers during the past a few years. In recent tank loss and -gactive is the effective negative conductance
researches, the VCO tends to work under lower supply of the active devices that compensates the losses in the
voltage [I]-[2], in higher frequency band [3]-[4] or with tank. For fully integrated LC VCO, gtak is dominantly
larger tuning range [5]-[6]. For example, VCO with very contributed by inductor. By using simple inductor model,
large tuning range is required for digital broadcast gt,k can be expressed as:
satellite television application (925MHz 21 75MHz)
-

[7]-[8]. In other applications such as Wireless LAN, it is

requested for the VCO to have multiple frequency bands g RL (1)
gtank co2 L)
(IEEE 802.1 la/b/g) which have very large frequency span
[9]. To realize such VCOs, some researchers used more
than one VCO to cover the required frequency range, in which co is the working frequency, RSL is the parasitic
which occupy a larger chip area [7]-[8]. Though in some series resistance of the inductor and Ls is the inductance.
papers only one VCO is realized to have a very large Assuming the VCO works in current limited region, the
tuning range [6], it lacks a thorough analysis and oscillation swing can be approximately calculated by:
optimization for the design of such VCOs. For the design
of narrow band VCO, some issues such as variations of '2bI
VCO start up condition, oscillation swing and phase noise (2)
performance are negligible, but these issues become RsL
obvious when the tuning range or frequency span of VCO
extends a lot. However, very few researches are reported
to solve these problems up to now. where Ibias is VCO bias current. To make the VCO start
In the rest of this paper, the issues for the design of oscillating, it is required that
VCO with very large tuning range or frequency span are
analyzed. A variable negative Gm technique is proposed
which can optimize the design of such VCOs. Simulation g active . amingtank (3)
results and discussions will be presented which verify our
conclusions. in which a min is a constant and is greater than 1.
From equation (1) (3), it can be seen that gta5,k is a
H. Shi, J. H. Liu, G. Y. Zhang, H. L. Liao, R. Huang and function of frequency and both the VCO oscillation swing
Y. Y. Wang are with the Institute of Microelectronics, Peking and start up condition are related to g,1,k. For RF LC VCO
University, China, E-mail: haoshi( pku.or-i.cn which has a very large tuning range or frequency span,

0-7803-9339-2/05/$20.00 ©2005 IEEE. 145

Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on September 24,2021 at 04:58:49 UTC from IEEE Xplore. Restrictions apply.
 gtank will vary a lot due to the big changes of the working tuning range. A possible way to solve these problems is
frequency. Therefore, the VCO start up condition will be adjusting the negative Gm of the active devices when gtank
obviously different at different frequency bands and the changes at different frequency bands. RF LC VCOs are
oscillation swing will also vary consequently. usually realized by using switched capacitor technique for
band selection as shown in Fig.3. The control bits of the
600 switched capacitor determine the specific frequency band.
The proposed idea is that in different frequency bands the
5000
negative Gm of the active circuits are simultaneously
40 I
changed by the switched capacitor control bits.

VDD
0I

100-

...................~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.......
.

2 4 5
freq. (CHz)

GND
L g.p
-1
Fig.2 I/gtar of the LC tank versus oscillation frequency
when sweeping the tank capacitor value. (a) (b) (c)

Fig.2 shows the simulation results of 1/ gtak of the

LC tank versus frequency. When sweeping the capacitor Fig.4 Core active circuit for RF CMOS LC VCO. (a)
value to make the oscillation frequency of the LC tank NMOS-only structure, (b) PMOS-only structure and (c)
change widely, it is found that the tank parallel resistance NMOS-PMOS structure.
at the oscillation point varies obviously.
The above problem becomes un-negligible for the Taking the VCO design in CMOS technology for
design of VCO with very large tuning range. The VCO example, the core active circuit usually uses one of the
start up condition will change obviously at different bands. topologies in Fig.4, in which g. and gmp are
If a unique -gactive is used to ensure oscillation for the transconductance of the NMOS and PMOS devices,
whole frequency range, the VCO cannot be optimized at respectively. The active circuit acts as a negative Gm to
different frequency bands. Meanwhile, the variation of the compensate the losses in the LC tank. For the topologies
oscillation swing will affect the working of the prescaler in Fig.4, the negative Gm is proportional to the device
which is the succeeding block of the VCO in frequency transconductance gm. By using the simple MOS device
synthesizer. Also, oscillation swing is an important factor model, gm can be calculated by:
that influences the VCO phase noise performance. Thus,
the VCO phase noise performance can be affected w
consequently. (4)
L
III. VARIABLE NEGATIVE GM VCO
VDD

out

x<2:0> y<2:0>
MP1 MP2 MP3 MP4

B3 B2 Bl BO
Vcont B<3:0>: <0000> <1111>
(a) (b)
GND

Fig.3 RF LC VCO with switch capacitor tuning. (a)

tuning curve, (b) switch-cap VCO topology. Fig.5 The proposed variable negative Gm VCO.

From the analysis in previous part, the key factor In equation (4), 1i and Cox are process parameters.
that leads to the variations of VCO start up condition, The designers can only control the gm0 by changing the
VCO oscillation swing and phase noise is that gta,k varies device size and bias current. A possible VCO core
with oscillation frequency obviously due to very large topology to realize such an idea is shown in Fig.5, in

146

Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on September 24,2021 at 04:58:49 UTC from IEEE Xplore. Restrictions apply.
 which the active circuit of Fig.4 (b) is used as the swing of the VCO at low frequency band versus the total
negative conductance. As shown in Fig.5, two identical bias current. For the line with the dot symbol, only one
groups of active circuits are used. They are respectively group of active circuit in Fig.5 is working and for the line
controlled by the MOS switches x<3> and y<3>, which with the triangle symbol, both groups of the active circuit
equivalently realizes programming the device size. The in Fig.5 are working.
bias current for each group can also be independently
programmed by x<2:0> and y<2:0>, respectively. Besides,
the active circuit in Fig.5 can be realized by Fig.4 (a) and
(c) by applying the same idea.

IV. RESULTS AND DISCUSSIONS

In this part, a wide band VCO using the topology in
Fig.5 is designed and simulated. The VCO features a very
large tuning range from 2.25GHz 4.5GHz which is
aimed for digital broadcast satellite television application.
To demonstrate the effects of the proposed technique,
only the representative results in high band (- 4.5GHz),
middle band ( 3.4GHz) and low band (- 2.25GHz) are lbias (mA)
shown. (a)

A IL
I0.

90
WJL~~~~~~~

PMOS
(MPl MP4) in Fig. orthedthre bands.I h

lbias (mA
(b)

simuaton thae totaibascrrn of the VCO

is0kzfrqec kffept Fig.7 (a) phase noise and (b) oscillation swing of the
identicalIth canibe seei fomte figur tha ther is ang5) VCO at low band versus the total bias current in Fig.5.
opimaldeicsizeoher the miimumae phase noise
othVCOan
The results in Fig.7 show an obvious improvement
beO achieved. iFi.forthe three feunybands, Itheopia of both the phase noise and oscillation swing of the VCO
at low band when two groups of the active are both
working. If the VCO changes to work in higher band, it
device sizes are much different, which indicate that for can be optimized by using only one group of active circuit
the design of the VCO with a very large tuning range, a and by adjusting its bias current at the same time. When
unique negative Gm can not meet the optimal design for the VCO works in lower frequency band, another group
all the bands. In Fig.6, the left-most points of the three of the active circuit provides one more way to adjust the
curves approximately indicate the minimum device sizes gm of the device compared to simply changing the bias
that meet the VCO start up condition at the three bands, current. Adjusting the gm of the device only by increasing
respectively. There are obvious differences between them, the bias current tends to lead the VCO to work in voltage
which means that the start up conditions for the three limited region which is usually not good for the VCO
bands are also much different. design. In comparison, the method in Fig.5 can effectively
From equations (1) (4), it can be found that when meet the optimum negative Gm requirement of VCO start
the VCO works in lower band, that is, when X becomes up condition and oscillation swing at different bands by
smaller, a larger gm of the MOSFET is required. Fig. 7 (a) programming both the device size and bias current.
(b) show the simulated phase noise and oscillation

147

Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on September 24,2021 at 04:58:49 UTC from IEEE Xplore. Restrictions apply.
 ones in Fig.8 (b) is quite small, which is attributed to the
2.50 .
....
optimization of the VCO by the variable negative Gm
2.25. ... A-High
band technique.
-X-Middle band
........ -o -Low band
2.00-
V. CONCLUSIONS
; 1.75.
¢ 1.50 ...............? ......................... For the design of narrow band VCO the variations of
X 1.25. VCO start up condition, oscillation swing and phase noise
0
1.00. performance are usually omitted, but these issues become
un-negligible when the tuning range or frequency span of
0.75
VCO extends a lot. The key factor that leads to these
U.U*. * ''' '':'' problems is that the equivalent LC tank parallel
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Vcon (V) conductance varies obviously due to big frequency
(a) changes. A variable negative Gm technique is proposed in
this paper to solve these problems. The proposed
technique provides two ways to optimize the negative Gm
of the VCO: programming both the device size and bias
current simultaneously. Simulation results prove that by
I using this technique the VCO start up condition at
different frequency bands can be independently optimized.
.5 The variations of the VCO oscillation swing can be
effectively reduced and the phase noise fluctuations
between different bands can be decreased close to the
intrinsic level.
z
0.
ACKNOWLEDGEMENT

Vcon (V) The authors wish to thank Intel for the sponsorship
(b) of this research project.
REFERENCE
Fig.8 (a) VCO oscillation swing and (b) phase noise [1] Ahmed H. Mostafa, Mourad N. El-Gamal and Ramez A.
versus the control voltage for all the high band, middle Rafla, "A Sub-l-V 4-GHz CMOS VCO and a 12.5GHz
band and low band. Oscillator for Low-Voltage and High-Frequency
Applications", IEEE Transactions on Circuits and
Systems-II: Analog and Digital Signal Processing, Vol.48,
Fig.8 (a) and (b) show the results of the VCO No.10, Oct. 2001
oscillation swing and phase noise versus the control [2] Ahmed H. Mostafa, Mourad N. El-Gamal, "A CMOS
voltage for all the high band, middle band and low band, VCO Architecture Suitable for Sub-I Volt High-Frequency
in which the bias current and device size are adjusted (8.7-1OGHz) RF Applications", ISLPED'O1, Aug. 2001
simultaneously. In Fig.8 (a), the oscillation swings are [3] Ren-Chieh Liu, Hong-Yeh Chang, Chi-Hsueh Wang, et al,
"A 63GHz VCO using a standard 0.25um CMOS Process",
optimized to be very close for the three bands. In Fig.8 (b), ISSCC Feb. 2004
it is found that the phase noise performances at the three [4] P-C. Huang, M-D. Tsai, H. Wang, et al, "A 114GHz VCO
bands are still much different from each other. This can be in 0.13um CMOS Technology", ISSCC Feb. 2005
explained by the following analysis. It can be calculated [5] M.A. Do, R.Y. Zhao, K.S. Yeo and J.G Ma, "New
that the phase noise will improve 20logN dB if the Wideband/Dualband CMOS LC Voltage-Controlled
Oscillator", IEE Proc.-Circuits Syst. Vol.150, No.5, Oct.
oscillation frequency becomes N times smaller (e.g. by 2003
division) when the noise sources are identical. In our [6] Kostas Manetakis, Darryl Jessie and Chiewcham
simulation, when the VCO switches oscillation bands, the Narathong, "A CMOS VCO with 48% Tuning Range for
noise sources which are mainly flicker noise and channel Modem Broadband System", IEEE 2004 Custom
thermal noise will vary oppositely with device width. If Integrated Circuits Conference, 2004
[7] Arun Jayaraman, Brian Terry, Bert Fransis, et al, "A Fully
the total noise sources are assumed the same, a big change Integrated Broadband Direct-Conversion Receiver for
of the oscillation frequency will intrinsically lead to the DBS Applications", IEEE International Solid-State
obvious variation of the phase noise. By rough calculation, Circuits Conference, 2000
the phase noise difference between the high band (- [8] Bo-Eun Kim, Se-Yeob Kim, Tae-Ju Lee, et al, "A CMOS
4.5GHz) and middle band (- 3.4GHz) caused by this Single-chip Direct Conversion Satellite Receiver for
Digital Broadcasting System", IEEE Symposium on VLSI
reason is about 2.4 dB, and this value is about 3.6 dB Circuits Digest of Technical Papers, 2002
between middle band and low band (- 2.25GHz). It can [9] Neric H.W.Fong, Jean Olivier Plouchart et al. "Design of
be seen that the above mentioned reason accounts for a Wide-Band CMOS VCO for Multiband Wireless LAN
large portion of the phase noise difference in Fig.8 (b). Applications", IEEE Journal of Solid-State Circuits,
The residual phase noise difference except for the above Vol.38, No.8, Aug. 2003

148

Authorized licensed use limited to: MANIPAL INSTITUTE OF TECHNOLOGY. Downloaded on September 24,2021 at 04:58:49 UTC from IEEE Xplore. Restrictions apply.