OPTIMIZATION OF CMOS 0.18 M LOW NOISE AMPLIFIER USING NSGA-II FOR UWB

International Journal of VLSI design & Communication Systems (VLSICS) Vol.5, No.5, October 2014
DOI : 10.5121/vlsic.2014.5505 59
OPTIMIZATION OF CMOS 0.18 M LOW NOISE
AMPLIFIER USING NSGA-II FOR UWB
APPLICATIONS
V. P. Bhale* and U. D. Dalal
VLSI Design Laboratory, Department of Electronics Engineering
Sardar Vallabhbhai National Institute of Technology (SVNIT),
Surat-395 007,Gujrat, India
ABSTRACT
A design and optimization of 3-5 GHz single ended Radio Frequency (RF) Low Noise Amplifier (LNA) for
ultra-wide-band (UWB) applications using standard UMC 0.18 µm CMOS technology is reported.
Designing of RF circuit components is a challenging job, since even after performing lengthy calculations
and finding parameter values it is less guarantee that the design performs as expected. In view of this the
optimization tool; Elitist Non-Dominated Sorting Genetic Algorithm (NSGA-II); has been employed to get
the optimized starting values of components in the proposed LNA design. The obtained NSGA-II
parameters were simulated using Cadence Spectre- RF simulator. The designed Low Noise Amplifier
achieves a power gain of 22 dB and a minimum Noise Figure of 3 dB is achieved. It dissipates 12.5 mW of
power out of 1.8 V supply.
KEYWORDS
LNA, NSGA-II Algorithm, Noise Figure, Power Gain, return losses
1. INTRODUCTION
Recent years have experienced explosive growth in the Radio Frequency /microwave
semiconductor industry owing to the proliferation of a host of applications. Single-chip Bluetooth
devices are already available and similar integration is likely to be achieved in cellular telephones
and wireless networking in near future. Radio Frequency components are the basic building
blocks of transceivers operating in GHz frequency range. Designing of RF circuit component
needs lot of effort. After performing lengthy calculations and finding the parameter values it is
not guaranteed that the circuit performs as expected. In Radio Frequency Integrated Circuits
(RFIC), Low Noise Amplifiers are considered as black magic box because of their uncertain
response with higher frequencies. Due to mismatch of input impedance and output impedance
maximum power transformation is not possible. For Designing of the tank circuit, input and
output impedance circuit, we need to design a passive filter with optimized component values.
Optimization of the component value is a time consuming job. In view of this a CAD tool using
Non-Dominated Sorting Genetic Algorithm (NSGA-II) has been employed. The design goal is
formulated as an objective function. Some approximations and estimations on the design
parameters are made in order to satisfy the requirement of the genetic algorithm. Many
applications of genetic algorithm and optimization of LNA parameter by binary coded genetic
algorithm is reported in [2, 3]. In this paper the design and optimization of single ended LNA
using real coded genetic algorithm is presented. The rest of the paper is organized as: section-II
gives brief introduction of Non-Dominated Sorting Genetic Algorithm. In section-III analysis and

60
design problem of low noise amplifier is presented, in section-IV design objective and constrains
optimization of LNA is mentioned. In section-V simulation result and discussion is presented.
Scope and limitation of the NSGA-II has been discussed in section-VI and finally section-VII
concludes the paper.
2. ELISTIST NON-DOMINATED SORTING GENETIC ALGORITHM
(NSGA-II)
Genetic algorithms [3, 4] are search techniques used in computing to find true or approximate
solutions to search or optimization problems. It is based on the concepts of natural selection,
reproduction and mutation and has been used extensively in optimization problems. It can be
classified in two sets depending on type of coding of the members; one is binary coded and the
second is real coded [3]. In the past few years GA has undergone lots of developments developing
its features, processing time, etc., some such developments are Multi Objective Genetic
Algorithm (MOGA) [5,6], Elitist Non Dominated Sorting Genetic Algorithm [3].
3. ANALYSIS AND DESIGN OF LOW NOISE AMPLIFIER
The early part of this section is based on the literature survey and concludes with own design. As
one of the essential components, Low Noise Amplifiers (LNAs) for wireless applications have
attracted significant research interest and various approaches to the design of narrow band LNAs
(operating below 3 GHz) and wideband LNAs (operating above 3 GHz) have been proposed
previously [7-15] and as shown in Fig. 1(a-d). Distributed amplifiers [7] can provide very large
bandwidth because of their unique gain-bandwidth tradeoff. However, large power consumption
and chip area make them unsuitable for typical low –power, low cost wireless applications.
Figure 1. Various LNA Topologies (a) Distribute amplifier, (b) common gate, (c) inductive degeneration,
(d) resistive feedback

61
Common-gate amplifiers [8, 9] exhibit excellent wide band input matching, but suffer from a
relatively large noise figure (NF). Narrow-band LNAs like an inductively degenerated common-
source amplifier can also be converted into a wideband one by adding a wideband input matching
network [10]. However, the insertion loss of the passive input matching degrades the NF rapidly
with frequency. Resistive-feedback amplifiers [11, 12-14] have very good wideband input
matching characteristics. However, low NF and low power consumption can be hardly achieved
simultaneously across a large frequency range. In [15], the noise cancellation technique is used to
relax this trade-off in resistive feedback amplifiers.
(a) (b)
Figure 2. (a) Inductive source degeneration, (b) Small signal equivalent of 2(a)
A typical LNA must fulfill several challenging requirements. The LNA must provide a good input
matching over a band more than 500 MHz. A high gain is also preferred to amplify the weak
signals at the receiver and to overcome the noise effects from the subsequent stages. In addition,
the noise figure of the LNA must be low (typically < 3 dB) since it plays a major role in defining
the receiver's sensitivity. Moreover, the LNA also has to be power efficient and physically small
to save power and reduce the cost, respectively. An inductively degenerated LNA configuration is
proposed, as shown in Figure 2 (a). Inductive degeneration improves the linearity of the amplifier.
The input impedance can be derived from the small signal analysis [16] of Figure 2 (b). By
looking into the input side of Figure 2 (b), input impedance Zin can be:
s
gs
m
gs
ggsin L
C
g
sC
RLLsZ )(
1
)( ++++=
(1)
Where Ls, Lg are the source and gate inductances, respectively; Rg is the transistor gate resistance,
Cgs is the transistor gate-to-source capacitance; gm is the transistor trans-conductance. The
inductor parasitic resistance is ignored here. Input match requires that at the resonance frequency
of the circuit, the impedance of the input stage is purely real and should be equal to 50 Ω. It
follows that:
50)( =+ s
gs
m
g L
C
g
R
(2)
Where ω0 is the resonance frequency (rad/s), and

62
0
1
)(
0
0 =++
gs
gs
Cj
LLj
ω
ω
(3)
The noise factor (F) is defined as [7, 8]:
20
)(1
T
sm
s
g
s
l
Rg
R
R
R
R
F
ω
ω
α
γ
+++=
(4)
)(log10 10 FNF ⋅= Where unity frequency:
gs
m
T
C
g
=ω ,
0d
m
g
g
≡α , Rs : source resistance,
Rl : series resistance of inductors, Rg : gate resistance, γ: the thermal noise coefficient, ω0 : the
resonance frequency, gm: transistor trans-conductance. For a source inductively degenerated
LNA in Fig.2, we could put a lower bound on the trans-conductance of the input transistor to
ensure that the final designed LNA can provide a reasonable gain [9]








−





==
01
2
0
10
0 1
1
CLw
Ljw
Ljw
ZGA
s
eqmV
(5)
In order to formulate a geometric programming problem, we have to do some transformation and
introduce a new variable to satisfy the requirement of geometric programming on the objective
and constraints. Inequality constraints, and the objective function must be in the form of
polynomial, equality constrains must be in the form of monomial. Here noise figure and gain are
formulated for low noise amplifier. For low noise amplifiers, Objective function of Noise Figure
can be formulated as:
( )2
min 02.050 optn GRFF −××+=
(6)
Where,
5
)1(21 2
0
min
×
−××××+
=
Wt
CW
F
γδ
mG
Rn
×
=
α
γ
γ
δ
α
×
−×
××=
5
)1( 2
0
C
CgsWGopt
The objective function of the Gain of LNA can be formulated as:
load
loadt
RW
RW
Gain
××
×
=
20 (7)
And the objective function of power consumption has been formulated as:
DCDDC VIP ×=
(8)
Where,
gthChannelLen
VWidthC
I odoxn
D
×
×××
=
2
2
µ
In above equation, second order effects on drain current have been neglected to reduce the
complexity of the program. As the schematic diagram of the LNA is depicted in Figure 3, an

63
input impedance matching circuit is needed to match the source impedance to transistor input
impedance. Objective function for Input impedance can be formulated from following equation.
s
gs
m
gs
ggsin L
C
g
sC
RLLsZ )(
1
)( ++++=
(9)
Ls and Lg is used to tune the input impedance (Zin) to 50 Ω at 3 GHz frequencies. Cgs and Rg
Model the impedance looking into the gate of MOSFET. Here we can include pad capacitance
and bond wire inductance also [10].
4. DESIGN OBJECTIVES AND CONSTRAINT OPTIMIZATION FOR
NSGA-II
The simulations have been done for LNA with following constraints. Based on the technology
parameters following have been defined as constant:
VVddVVth
mLnH
RLRsmfFCox
8.1,47.0
,5.0,18.0)(,5.2,1)/(
,50)(,50)(,632.8)2/(
==
====Ω
=Ω=Ω=
λµγβ
µ
Design constraints for the component of circuit were taken as follows:
(1) mW µ10515 ≤≤ (2) nHLs 501.0 ≤≤
(3) nHLg 501.0 ≤≤ (4) pFC 5.01.0 ≤≤
Simulations have been done with these parameters using NSGA-II. The NSGA parameters that
were given are as follows: Mutation probability= 0.23671, Population size=100, Crossover
Probability=0.99431, Number of generations=50.
5. SIMULATION RESULTS AND DISCUSSIONS
5.1. Simulation results using NSGA-II
The above mentioned optimization technique is implemented for Low Noise Amplifier design to
optimize gain and noise figure. Direct equations have been used for gain and noise figure which
were calculated for cascode LNA with inductive source degeneration. Program structure for LNA
optimization using NSGA-II is shown in Figure 3 below. The “func-con. h” header file has been
modified and the fitness functions and constraint functions have been replaced with the equations
as discussed in section 3. Also, “LNA_equations.h” file has been added, which actually evaluates
all these equations.
Figure 3. LNA Optimization Program Structure
LNA Equations header file
NSGA-II Optimization Core
Input file
handler
Output file handler

64
After NSGA-II optimizer generates output files, another program convert those into design
parameters and we can plot those values using GNUPLOT software as plotted in Figure 4(a-f).
The trans-conductance, gm, of the device is very much important to achieve good gain for LNA.
However, this gm depends on the transistor width. Also, gm/ID ratio needs to be maintained to
achieve a required LNA gain. Thus, the width of the transistor needs to be chosen to give enough
gain and minimum noise figure. From Figure 4 (a), thus, ‘W’ is chosen to be 105 µm. The
overdrive voltage is one of the important parameters for transistor operating point analysis. The
transistor should operate in the desired operating region for maximum gain. Thus, plots in Figure
4 (b) are plotted using NSGA-II to guess the initial overdrive voltage required for the transistor to
be in the desired operating region.
The optimized value of food is chosen to be 0.35 V from Figure 4 (b) since at this overdrive
voltage maximum gain with low noise figure can be achieved simultaneously. The role of a
source degenerated inductor (Ls) here is to match the input impedance to 50 Ω with low noise.
From Figure 4 (c), it can be seen that the noise is increasing with Lsource .Thus the optimum value
of source at a minimum noise figure is chosen to be 0.27 NH.
(a)
(b)

65
(c)
(d)
(e)

66
(f)
Figure 4. Simulation results using NSGA-II (a) Noise vs. Gain vs. Width of the transistor, (b) Noise vs.
Gain vs. Overdrive voltage, (c) Noise vs. Lsource (d) Power dissipation vs. Lsource , (e) Gain vs. Lsource &
(f) Power vs. Noise vs. Gain
The effect of variation of a source inductor in power consumption can be seen from Figure 4 (d),
where it can be seen that with the increase in value of source degeneration inductor; power
dissipation reduces. But, for RF design the inductor should not be bulky. There always exists a
trade-off among power dissipation, gain and noise figure for a Low Noise Amplifier design. The
same can be seen in Figure 4 (e). While, from Figure 4 (f), it can be seen that the gain reduces
with the increase in value of the source degenerated inductance. Thus, Lsource = 0.2 nH has been
chosen as further optimized value which gives the best compromise between gain and noise
figure trade-off.
5.2. Simulation of LNA using Specter RF
The initial start up values given after running NSGA-II algorithm has been used for the designing
of CMOS Low Noise Amplifier for 0.18 µm technology. The design is finally simulated using the
Cadence Specter design tool. Figure 5 shows the proposed current reused LNA design. The S-
parameter analysis is performed to obtain the gain and noise figure parameters. Also parallel LC
tank circuit at the output tunes to the resonating frequency of 4 GHz. Capacitors at the input and
output are dc blocking capacitor. The source terminal inductor (Ls) needs to be properly design to
have a 50 Ω input matching. The dc blocking capacitors has been chosen to be of 103 fF each.
While Lg is calculated from Eq. (13), once the optimum value of Ls is chosen to be 0.2 nH.
Usually the value of a lead is picked up from the available literature as references and accordingly
the parallel tuning capacitor needs to be tuned to operate at 4 GHz frequency. The value of bias
resistor, Rbias, is chosen to be large to minimize the noise entering into the design.

67
Figure 5. Proposed current reused two stage LNA
5.2.1. Parametric analysis
The parametric analysis is used for automating the generation of multiple simulations to test the
effect of a source degeneration inductor (Ls) variation on the small signal gain (S21). Ls is sweep
from 50 pH to 200 pH as shown in Figure 6. It is observed that for 3-5 GHz band gain flatness is
good for Ls as 200 pH, so, Ls is chosen to be 200 pH.
Figure 6. Parametric Analysis for Small Signal Gain

68
Figure 7. Overall S-parameter response of Single stage LNA
5.2.2. S-Parameter analysis
The simulated S-parameter results are as shown in Figure 7. The simulated results show that the
LNA has a maximum flat gain of + 22 dB from 3 to 5 GHz band. Also, Figure 8 shows a smith
chart analysis plots for input impedance matching. From smith chart it can be seen that the real
part of the impedance is matched to 47 Ω. Thus minimum input and output return loss of < -10
dB each can be observed for 3 to 5 GHz. The input and output return losses have to be further
improved by proper matching. The LNA also attends a high reverse isolation (S12) of < - 40 dB
for 3 to 5 GHz.
Figure 8. Input Impedance Matching using smith chart

69
5.2.3. Stability analysis
In the presence of feedback paths from the output to the input, the circuit might become unstable
for certain combinations of source and load impedances. An LNA design that is normally stable
might oscillate at the extremes of the manufacturing or voltage variations, and perhaps at
unexpectedly high or low frequencies. The Stern stability factor characterizes circuit stability as
in equation below:
Kf= 1+|βf |2
-|S11|2
-|S22|2
/ 2|S21||S12| (10)
Figure 9. Stability Analysis
Where, βf = S11S22-S21S12 , If Kf >1 and βf < 1, then the circuit is unconditionally stable. The
stability evaluation for the S parameters over a wide frequency range has been done to ensure that
the Kf remains greater than one for all frequencies. As the coupling (S12) decreases, that is as the
reverse isolation increases, stability improves. One can use the techniques such as resistive
loading and neutralization to improve stability for an LNA [11]. Stability analysis shows that Kf
>1 and βf < 1 across the frequency band of interest, stating that designed LNA is un-conditionally
stable as shown in Figure 9.
5.2.4 Linearity analysis
Linearity is also important parameter for LNA design along with gain. The linearity limits the
actual power that can be drawn to the load by the LNA. The linearity is simulated using periodic
steady-state (PSS) analysis. Figure 7 shows the linearity analysis with P1dB compression point of
-6.46719 dBm at 4 GHz.

70
Figure 10. P1dB compression point
5.2.5. Noise analysis
The overall noise figure (NF) and minimum noise figure (NFmin) of 2.31 dB and 2.1 dB
respectively can be seen from Figure 11 for cascade common source with current re-use feedback.
Table-1 shows the performance summary of the proposed low noise amplifier and its comparison
with the previously reported LNA designs. At the time of designing parasitic play important role
in obtained results but here we are not considering them because our first motivation is to take
initial guess of the values. The design parameters obtained from multi objective genetic algorithm
is comparable with the result obtained in Cadence Spectre tool.
Figure 11. Noise Figure plots for two stage LNA

71
Table 1 Comparison of Wideband LNAs: Published and the optimized LNA design parameters
Proposed LNA designs Previously Reported LNA Designs
Targeted parameters
Optimized two stage
LNA design with
current re-used using
NSGA-II [15][16] [17] [18]
Technology 0.18µm CMOS 0.18µm CMOS
0.65 µm
CMOS
0.18 µm
CMOS
0.18 µm
CMOS
Frequency of
Operation 3-5 GHz 3-5 GHz 0.2-5.2 GHz 1-5 GHz 3-5 GHz
Supply
Voltage 1.8 V 1.8 V 1.2 V 1.8 V 1.8 V
Gain
20 dB
22 dB 13-15.6 dB 11-13.7 dB 8.6-9.5 dB
S11 < -10 dB < -10 dB *** <-10 dB <-10.3 dB
S22 < -10 dB < -10 dB *** *** -11.8 dB
S12 < -10 dB -40 dB *** *** ***
NF < 4 dB 3.5 dB < 3.5 5-6.5 dB 2.7 dB
NFmin < 3 dB 3 dB *** *** ***
P1dB < -10 dBm -6.46719 *** *** ***
Power < 15 mW 12.5 mW *** 9 mW 15 mW
Note: *** Not mentioned
6. SCOPE AND LIMITATION OF DESIGNED TOOL
Designed optimization tool gives only approximate values of the design parameters. This tool is
designed only for final stages of RFIC design process. Parasitic effects associated with passive
on-chip components like capacitors, inductors and resistors have not been taken into account for
shorter simulation time of the circuit.
7. CONCLUSION
This paper shows that the optimization of RF Circuits is possible with real coded genetic
algorithm. It is found that real coded Multi-Objective Genetic Algorithm has many advantages
over binary coded genetic algorithm. Non-Dominated Sorting Genetic Algorithm is used for
optimization tool, which is giving comparative results with design software simulation like
Cadence Spectre tool. In this paper it is shown that the Low Noise Amplifier can be designed for
the noise figure of 3.5 dB and power gain of 22 dB. The two stage LNA topology with current
reused technique is designed and optimized for 3-5 GHz UWB applications. The simulated S11
and S22 parameters are well below – 10 dB is obtained with the simple matching network for the
desired band of 3-5 GHz. The designed LNA dissipates 22 mW of power out of 1.8 V supply.
Also the proposed LNA design is found to be unconditionally stable and operates linearly
throughout the desired band. A comparison between the present results and the results of
previously reported LNA shows that the reported LNA design reaches the state-of-the-art LNA
designs for UWB application. In future; the NSGA-II optimization tool can be used to extend for
3.1-10.6 GHz LNA design for ultra-wide-band wireless RF system. Thus the design tool is useful
in finding circuit element values quickly reducing the RF circuit designer time.

72
REFERENCES
[1] Federal communications commission (FCC), first order and report,”2002
[2] Min Chu, “Elitist Non dominated Sorting Genetic Algorithm Based RF IC Optimizer”, IEEE
Transaction Vol. 52, No.3.
[3] K. Deb.,“Multi-objective optimization using evolutionary algorithms”, Wiley,2003. Pg.-248-
[4] D.E. Goldberg, “Genetic Algorithms in Search, Optimization, and Machine Learning”, Addison-
Wesley Longman Publishing, Co., Inc. Boston, MA, USA. 1989.
[5] N. Chaiyaratana, A. M .S. Zalzala, “Recent Development in Evolutionary and Genetic Algorithms:
Theory and Applications,” no. 446, IEE, 1997.
[6] Deb, K., “Single and Multi-Objective Optimization Using Evolutionary Algorithms”,Kan GAL
Report No. 2004002, February, 2004.
[7] B. M. Ballweber, R. Gupta, and D. J. Allstot, “A full integrated 0.5-5.5 GHz CMOS distributed
amplifier”, IEEE Journal of Solid State Circuit, vol.35, no.2 pp.231-239, Feb. 2000.
[8] C.F. Liao, S-I Liu, “A broad band noise cancelling CMOS LNA for 3.1-10.6 GHz wireless receivers”,
IJSSCC, 42(2):329-339, Feb. 2007.
[9] R. Gharpure, “A broadband low noise front end amplifier for ultra wideband in 0.13 µm CMOS”,
IEEE Journal of Solid State Circuits, vol 40, no.9 pp.1983-1986, Sep.2005.
[10] W. H. Chen, G. Liu, B. Zdravko, and A. M. Niknejad, “A highly linear broadband CMOS LNA
employing noise and distortion cancellation”, in IEEE RFIC Symp. Dig.papers, pp.61-64, 2007
[11] A Bevilacqua and A.M. Niknejad, “An ultra wide band CMOS LNA for 3.1-10.6 GHz wireless
receivers”, In ISSCC, Dig.Tech. Papers, pp. 382-533, 2004.
[12] C. W. Kim, M. S. Kang, P. T. Anh, H. T. Kim, and S. G. Lee, “An ultra-wide band CMOS low noise
amplifier for 3-5 GHz UWB System”, IEEE J. Solid State Circuits, vol. 40, no.2, pp.544- 547, Feb
2005.
[13] R. Ramzan, S. Anderson, J. Dabrowski, and C. Svensson, “A 1.4V 25mW inductor less wideband
LNA in 0.13µm CMOS”, in ISSCC Dig. Tech. papers, pp.424-425, 2007.
[14] M. Vidojkovic, M. Sanduleanu, J. V. D. Tang, P. Baltus, and A.V. Roermund, “A 1.2V, inductorless,
broadband LNA in 90 nm CMOS Low Power”,In IEEE RFIC Symp .Dig. Tech Papers, pp. 53-56,
2007.
[15] F. Bruccoleri, E. A. M.Klumperink, and B. Nauta, “Noise Cancelling wideband CMOS LNAs”, in
ISSCC Dig. Tech. papers, pp.406-407, 2002.
[16] Ming ShenTian, Jan. H. Mikklensen, Olek Jensen Torben Larsen, “Design and Implementation of a 1-
5 GHz low noise amplifier in 0.18µm CMOS,” Analog Integrated Circuits and Signal Processing,
pp.41-48, 2011
[17] Ji-Hai Duan, Xiao-ting Han, Sheng Li, “A Wideband CMOS LNA for 3-5GHz UWB Systems”, IEEE
Conference, 2009.
[18] Andrea Bevilacqua, Christoph Sandner, et.al.,“A Fully integrated Distributed CMOS LNA for 3- 5
GHz Ultra wideband wireless receivers”, IEEE Microwave and Wireless Components Letters,2006
AUTHORS
Vishakha P. Bhale received B.E. (Electronics and Telecommunication) and M.E. (Digital
Electronics) in 2002 and 2006 respectively. Presently she is pursuing Ph.D. in RF circuit
designs for wireless communication application using VLSI design. Her area of interest
is analogue RF circuit designs, Low power VLSI designs, Wireless communication
system designs for portable applications.
Dr. Upena Dalal received the B.E. in Electronics from SVRCET, Surat in 1991 and
obtained M.E. (Electronics & Communications) from DDIT, Nadiad with Gold Medal.
Presently she is working as Associate Prof. in Electronics Engineering Department, Sardar
Vallabhbhai National Institute of Technology (SV NIT), Surat,Gujrat, India. She ha s 20
years of academic experience. She has published 70+ conference and journal papers at
national and international level. Her book on “Wireless Communication” is published by
Oxford University Press in July 2009. One more book edited by her and Dr Y P Kosta
titled “Wi-MAX New Developments” is published by Intech, Vienna, Austria. She has organized 8 National
level training programs and two international conferences. She is a LM of several technical technical bodies.

OPTIMIZATION OF CMOS 0.18 M LOW NOISE AMPLIFIER USING NSGA-II FOR UWB

More Related Content

What's hot (20)

Similar to OPTIMIZATION OF CMOS 0.18 M LOW NOISE AMPLIFIER USING NSGA-II FOR UWB (20)

Recently uploaded (20)