288 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 1, JANUARY 2002
Concurrent Multiband Low-Noise
Amplifiers—Theory, Design,
and Applications
Hossein Hashemi, Student Member, IEEE, and Ali Hajimiri, Member, IEEE
Invited Paper
Abstract—The concept of concurrent multiband low-noise-am-
plifiers (LNAs) is introduced. A systematic way to design concur-
rent multiband integrated LNAs in general is developed. Applica-
tions of concurrent multiband LNAs in concurrent multiband re-
ceivers together with receiver architecture are discussed. Exper-
imental results of a dual-band LNA implemented in a 0.35-
m
CMOS technology as a demonstration of the concept and theory
is presented.
Index Terms—Amplifier noise, land mobile radio cellular sys-
tems, low-noise amplifier, radio communication, radio receivers.
I. INTRODUCTION
S
TANDARD receiver architectures, such as superhetero-
dyne and direct conversion, accomplish high selectivity
and sensitivity by narrow-band operation at a single input
frequency [1]. These modes of operation limit the system’s
available bandwidth and robustness to channel variations and
thus its functionality. On the other hand, wide-band modes of
operation are more sensitive to out-of-band unwanted signals
(blockers) due to transistor nonlinearity. These out-of-band
blockers can severely degrade receiver’s sensitivity.
Thediverserangeofmodernwirelessapplicationsnecessitates
communication systems with more bandwidth and flexibility.
More recently, dual-band transceivers have been introduced to
increase the functionality of such communication systems by
switching between two different bands to receive one band at
a time [2]–[5]. While switching between bands improves the
receiver’s versatility(e.g.,inmultibandcellularphones), itis not
sufficient in the case of a multifunctionality transceiver where
more than one band needs to be received simultaneously (e.g.,
a multiband cellular phone with a global positioning system,
global position system (GPS), receiver and a Bluetooth inter-
face). Using conventional receiver architectures, simultaneous
operation at different frequency bands can only be achieved by
building multiple independent signal paths with an inevitable
increase in the cost, footprint, and power dissipation. Although
there have been efforts to minimize the number of additional
Manuscript received May 28, 2001. This work was supported by Conexant
Systems and by NSF-ERC.
The authors are with the Caltech High-Speed IntegratedCircuits Group, Elec-
trical Engineering Department, California Institute of Technology, Pasadena,
CA 91125-9300 USA (e-mail: hashemi@caltech.edu; hajimiri@caltech.edu).
Publisher Item Identifier S 0018-9480(02)00843-8.
components used for the second band of operation (e.g., for
addingGPS to a CDMA phone[6]), none ofthese effortsattempt
simultaneous reception of more than one band.
In this work, a new concurrent dual-band receiver architec-
ture is introduced that is capable of simultaneous operation at
two-different frequencies without dissipating twice as much
power or a significant increase in cost and footprint [7]. This
concurrent operation can be used to extend the available band-
width, provide new functionality, and/or add diversity to battle
channel fading. The concurrent operation is realized through
an elaborate frequency conversion scheme in conjunction
with a novel concurrent dual-band low-noise amplifier (LNA).
These new concurrent multiband LNAs provide simultaneous
narrow-band input matching and gain at multiple frequency
bands, while maintaining low noise.
Section II reviews the current advances of single-band LNAs
from technological and architectural points of view. Section III
briefly describes one such receiver architecture demonstrating
the central role of the concurrent LNAs in the receiver. The
general design methodology of concurrent multiband LNAs is
discussed in Section IV. Experimental results of a concurrent
dual-band CMOS LNA will be presented in Section V.
II. A R
EVIEW OF SINGLE-BAND LNA DESIGN ISSUES
Being the first active element in the receiver chain, the noise
figure (NF) of an LNA plays a significant role in the overall
NF of the receiver, which controls its sensitivity and output
signal-to-noise ratio (SNR) [8]. Before exploring the design
details of concurrent multiband LNAs, it is helpful to review
some of the existing technological and topological choices for
single-band LNAs.
A. Technology
The bipolar junction transistor was the first solid-state active
device to provide practical gain and NF at microwave frequen-
cies [9]. In the seventies, breakthroughs in the development
of field-effect transistors (FETs) (e.g., GaAs MESFET) led
to higher gain and lower NF than bipolar transistors for the
frequencies in the range of several gigahertz [10]. Currently,
advanced FETs and bipolar transistors still compete for lower
NF and higher gain at frequencies in excess of 100 GHz.
0018–9480/02$17.00 © 2002 IEEE
HASHEMI AND HAJIMIRI: CONCURRENT MULTIBAND LOW-NOISE AMPLIFIERS 289
Examples are the high electron-mobility transistors (HEMTs),
such as pseudomorphic high electron-mobility transistors
(pHEMTs) [11], metamorphic high electron-mobility tran-
sistors (MHEMTs) [12], as well as heterojunction bipolar
transistors (HBTs) [13], [14], built using a variety of semicon-
ductor materials (e.g., GaAs, InP, Si, SiGe).
Traditionally, very low-noise amplifiers at high frequencies
have been made using transistors with high electron mobility
and high saturation velocity on high-resistivitysubstrates for the
following principal reasons.
1) Higher carrier mobility and peak drift velocity result in
a higher transistor transconductance and shorter carrier
transit time [10] for a given current, thus allowing for the
reduction of the dc current for the same transconductance
(gain) in transistors which lowers the input-referred noise
and, hence, the NF. This gives compound semiconduc-
tors a significant advantage over silicon, as for instance,
the electron mobility and the peak drift velocity are typ-
ically six and two times larger, respectively, for GaAs
when compared to silicon [10].
2) Higher carrier mobility also results in lower parasitic
drain and source series resistors. The parasitic source
resistance can be a major contributor to the overall
NF of certain LNAs, such as those used for satellite
communications.
3) Due to mostly technological limitations, the series input
resistance of silicon-based transistors is usually higher
than those of compound semiconductors. In particular,
the lower resistance of the metal gate of GaAs MESFETs
compared to higher resistance of the poly-silicon gate in
MOSFETs and thin bases in bipolar transistors, result in
a lower NF for GaAs transistors.
4) The loss properties of on-chip passive components can
have a significant effect on the noise and gain perfor-
mance of the LNAs. High-resistivity substrates minimize
the substrate loss components. As the loss and noise
are closely related through the fluctuation-dissipation
theorem of statistical physics [15], [16], the energy loss
reduction translates to a lower NF for the amplifier.
Despite the above mentioned limitations of silicon technolo-
gies, several silicon LNAs have been reported. Meyer
et al.
reported one of the early LNAs made on a low-resistivity (i.e.,
lossy) silicon substrate using bipolar junction transistors for
commercial cellular applications [17], where very low NF is not
needed. Recently, a large number of efforts have been reported
to use the advanced digital CMOS processes for single-chip
implementation of the complete radio transceiver [18], [19].
Significant progress in CMOS LNA design has been made
during the last several years where more recent results, such
as [20], demonstrate significant improvements over the earlier
works [21]–[23] and show that CMOS LNAs can be a worthy
competitor for compound semiconductor implementations in
many portable applications.
B. Topology
Although several different topologies have been pro-
posed to implement LNAs, we will only focus on two most
Fig. 1. Commonly used single-band CMOS LNAs. (a) Common-gate. (b)
Common-source with inductive degeneration.
common single-stage
1
LNAs in CMOS processes, namely,
the common-gate topology [22] and inductively degenerated
common-source stage [23], [24], shown in Fig. 1.
The common-gate configuration uses the resistive part
looking into the source of the transistor to match the input to a
well-defined source impedance (e.g., 50
). This impedance is
in the case of a MOSFET, where and
are transconductances of the top-gate and back-gate transistors,
respectively. However, it can be shown that the NF is lower
bounded to 2.2 dB for a perfectly matched long-channel CMOS
transistor [22] unless a transformer is used at the input [25].
In a common-source LNA, inductive degeneration is used to
generate the real part needed to match the LNA input to the pre-
ceding antenna or filter. Strutt and Van der Ziel first noticed that
inductive degeneration can enhance the output SNR [26]. The
ideal lossless inductive feedback moves the source impedance
for optimum NF toward the optimum power match with a minor
increase in the minimum NF [27]. Unfortunately, in silicon im-
plementations, the loss associated with inductors will degrade
the NF. It should be mentioned that in these cases cascode con-
figuration can be used to enhance the stability and reverse-iso-
lation of the amplifier.
While the problem of achieving the lowest noise in an ampli-
fier has been solved for a general case through a mathematical
treatment [28], this general approach still does not provide the
necessary insights into the design.
An alternative approach is to use Smith charts to find the op-
timum impedance for noise and power matching at the input
of the amplifier for given active device [29]–[31]. Although the
Smith chart is a very convenient tool for seeing how close we are
to the minimum NF and the maximum gain of a given device,
it does not show the effect of individual noise sources on the
total NF. This is particularly important for a concurrent multi-
band LNA, since different noise sources behave differently at
different frequencies.
Unlike bipolar transistors whose dc current sets the transcon-
ductance and minimum noise-figure, MOSFETs offer extra de-
grees of freedom in choosing the devicewidth and length. These
extra degrees of freedom can be used to improve the NF and the
gain of the amplifier. Recently, some work has been done to cal-
culate and minimize the NF of a single-band common-source
1
This discussion is also valid for the first stage of multistage LNAs.
290 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 1, JANUARY 2002
Fig. 2. Evolution process of two parallel receivers to a concurrent dual-band receiver.
CMOS LNA with inductive degeneration using a more system-
atic approach [23]. In the next section, we will introduce the
concept of a concurrent receiver and a new architecture to im-
plement it. In Section IV, we present a general approach for
the design of concurrent multiband LNAs which are important
building blocks in concurrent receivers.
III. C
ONCURRENT RECEIVER ARCHITECTURES
In this section, we will develop a concurrent dual-band re-
ceiver architecture that can be fully integrated. The objective is
to devise a receiver that can simultaneously receive signals at
two different frequency bands with maximum reuse of power
and building blocks.
Fig. 2 shows the conceptual evolutionof a dual-band receiver,
starting with two totally independent heterodyne receiving
paths, and leading to an efficient concurrent dual-band receiver.
The first gain stage in a concurrent dual-band receiver is its
LNA. Traditional single-band LNAs use a single or cascode
transistor stage to provide wide-band transconductance and
combine it with proper passive resonant circuitry at the input
and output as discussed briefly in the previous section. This
approach shapes the frequency response, ensures stability, and
achieves gain and matching at the single band of interest [31].
A very important observation is that the transconductance of
the transistor is inherently wide-band and can be used to provide
gain and matching at other frequencies without any penalty in
the power dissipation. This observation leads to a compact and
efficient front-end for a concurrent dual-band receiver which
consists of a dual-band antenna [32]–[34], followed by a mono-
lithic dual-band filter [35] and a concurrent dual-band LNA
that provides simultaneous gain and matching at two bands, as
shown at the bottom of Fig. 2. A detailed approach to the de-
sign of such a multiband LNA will be described in the subse-
quent sections. It should be noted that the concurrent dual-band
Fig. 3. An architecture for concurrent dual-band receiver.
receiver does not need any dual-band switch [36] or diplexer
[37], because simultaneous reception at both bands is desired.
Then a dual-band down-conversion scheme is needed to trans-
late different information carrying signals to baseband with as
few local oscillators (LOs) and external filters as possible, while
maintaining isolation between the two bands. This can be done
in many different ways, for instance, Fig. 3 shows a simplified
block diagram of one such receiver.
The frequency of the first LO that appears after the LNA and
performs the first down conversion determines the image fre-
quency(ies) and plays an important role in the performance of
the system. For a nonconcurrent receiver, it has been proposed
to choose the first LO frequency halfway between the two fre-
quency bands and select the band of interest by choosing the
appropriate sideband produced by an image-separation mixer
[2]. Although this method is sufficient for the nonconcurrent ap-
proaches, it will suffer from some serious shortcomings if used
for a concurrent receiver, where the LNA amplifies the signal in
both of the desired bands. This is because one band is the image
of the other and there is no attenuation of the image by either the
antenna or the filter. The situation is exacerbated by the LNA
HASHEMI AND HAJIMIRI: CONCURRENT MULTIBAND LOW-NOISE AMPLIFIERS 291
Fig. 4. Frequency-domain signal evolution in the concurrent dual-band
receiver of Fig. 3.
gain in the image band. In this scenario, one is solely relying
on the image rejection of the single sideband receiver, which is
limited by the phase and amplitude mismatch of the quadrature
LOs and the signal paths [38], [39], and is insufficient in a con-
current receiver.
An alternative approach that does not suffer from the above
problemand,infact,significantlyimprovestheimagerejectionis
touse anoffsetLO asshownin Fig. 4. The LO frequencyis offset
from the midpoint of the two bands of interest (
and )in
such a way that the image of the first band at
falls at the notch
of the front-end transfer function at
. The attenuation at
is determined by the compounded attenuation of the dual-band
antenna, filter, and LNA. Similarly, the image of the second
band at
will fall outside the passband of the front-end at
and will be attenuated, accordingly. Using a quadrature first LO
makes the stage fit to act as the first half of any single-sideband
image-reject architecture, such as that proposed by Weaver[40].
Sincethe receiver has to demodulate twobands concurrently and
independently, two separate paths must be used eventually. Each
pathcomprisesthesecondhalfoftheimagerejectarchitecture,as
shown in Fig. 3, which provides further image rejection (Fig. 4).
This architecture eliminates an extra antenna, a front-end filter,
an LNA, and a pair of high-frequency mixers, which in turn
results in power, footprint, and area savings. At the same time,
large image rejection in excess of that of the single-sideband
receiver is achieved through diligent frequency planning and
proper usage of stop-band attenuation.
IV. C
ONCURRENT MULTIBAND LNA
In a single-band LNA, passive networks are used to shape
the response of the wide-band transconductance of the active
device in the frequency domain to achieve gain and matching
at the frequency of interest. This concept can be generalized to
multiple frequency bands noting that the intrinsic transconduc-
tance of the active device is inherently wide-band and can be
used at multiple frequencies simultaneously.
Fig. 5. General model for a single-stage amplifier in common-source
configuration.
It is crucial to note the fundamental differences between the
concurrent and the existing nonconcurrent approaches. In con-
ventional dual-band LNAs, either one of the two single-band
LNAs is selected according to the instantaneous band of oper-
ation [41], [42], or two (three) single-band LNAs are designed
to work in parallel using two (three) separate input matching
circuits and two (three) separate resonant loads [2], [43]. The
former approach is nonconcurrent, while the latter consumes
twice (three times) as much power if used in a concurrent set-
ting. The other existing approach is to use a wide-band amplifier
in the front-end [44]. Unfortunately, in a wide-band LNA,strong
unwanted blockers are amplified together with the desired fre-
quency bands and significantly degrade the receiver sensitivity.
In this section, we present an analytical approach to the design
of a general class of integrated concurrent multiband LNAs. The
concurrent LNA is proposed as a solution to the aforementioned
problems in a concurrent receiver.
A. General Amplifier in Common-Source Configuration
In this section, we use a general model for an amplifier in
the common-source configuration to obtain an equivalent circuit
for the input impedance and a general expression for the gain
at multiple frequencies. This equivalent circuit will be used to
achieve simultaneous power and noise matching in a concurrent
multiband LNA. Fig. 5 shows a transistor
2
with arbitrary gate
impedance
, gate–source impedance , source impedance
, gate–drain impedance , and load impedance . The
impedances shown in Fig. 5 also include the transistor’s inherent
passive components (e.g.,
, ). General expressions for
input impedance and voltage-gain of this amplifier are found in
Appendix A.
B. Input Matching
The input of the LNA is either fed directly by the antenna
or is connected to the antenna through a bandpass filter, a
diplexer/duplexer, or both. In any case, the impedance looking
into the input of the LNA should be power matched (i.e., com-
plex conjugate matched)
3
to the impedance of the preceding
stage for maximum signal power transfer. Additionally, it is
2
While the general active device discussed here is a MOS transistor, a similar
analysis applies to other active devices (e.g., BJT, MESFET.)
3
In large-signal devices, power match does not necessarily correspond to the
complex conjugate matching. However,since the LNAdesign is based on small-
signal principals, we can use two terms, synonymously.
292 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 50, NO. 1, JANUARY 2002
Fig. 6. General representation of any (a) noisy two-port and (b) its equivalent circuit.
essential to provide the correct impedance to the preceding
stage to satisfy its nominal specifications (e.g., bandpass filter
characteristics, such as rolloff, etc., depend on filter loading).
The expressions in Appendix A can be further simplified if
we assume that
is much larger than the other impedances.
This assumption neglects the effect of the transistor’s intrinsic
and its associated Miller effect. Then, the input impedance
expression of (21) simplifies to
(1)
This expression will be used in the following section to design
multiband input matching networks.
Theoretically, the input impedance of any stable amplifier
with a nonzero real part can be perfectly matched to any arbi-
trary source impedance (with a positive real part) for a single
frequency using lossless passive components at the input of the
amplifier [45]. Equation (1) can be used to generalize this power
match concept to multiple frequencies. It can be used to gen-
erate numerous topologies to achieve simultaneous impedance
matching at multiple distinct frequencies.
In an LNA, it is also necessary to achieve a noise match at the
input for the frequency(ies) of interest to minimize the NF. In
the following section on noise matching, we will demonstrate
that one way to minimize the NF of the amplifier of Fig. 5 is
by designing the passive network so that it satisfies
at multiple frequencies of interest. However, this can
only be achieved using lossless passive components. Therefore,
in practice, one should minimize
, to its smallest
real part,
. Having satisfied the above condition, the input
impedance will be
(2)
Theoretically, a large number of passive topologies for
and can provide input impedance matching at multiple
frequency bands. One particular example which is of great
practical value is when
is just the intrinsic gate–source
capacitance,
and, hence, has to be an inductor as in
the single-band common-source LNA in [26], [24], [23]. For
negligible passive loss (
) and a real-value impedance
, (e.g., 50 for most practical cases), the source
inductor is given by
(3)
This will result in a passive network for
that will minimize
for all the frequencies of interest. One example
of such a design can be found in Section V.
The optimum source inductance depends on
and, hence,
process parameters, as (3) suggests. Ignoring
, in a deep
short-channel CMOS biased in the velocity saturation region,
is approximately given by
(4)
where
is electron mobility in the channel, is the critical
field,
is the saturation velocity, and is transistor’s channel
length. Therefore, for a given deep submicrometer CMOS tech-
nology with constant channel length where carriers are velocity
saturated, the value of
is almost fixed and is independent of
the bias current and the device size. For a bipolar transistor,
has a current dependency. However, if junction capacitors are
negligible for a transistor biased with a high collector current,
this current dependency is small and again the value of
is
independent of bias current. In a long-channel CMOS,
de-
pends on the bias current and the device width and so will
.
C. Noise Matching
An important design parameter in receiver design, which is
the measure of receiver noise, is the noise factor
(also known
as NF, when expressed in decibels). The definition of the noise
factor of any transducer (e.g., LNA, mixer, filter, etc.) given by
[46] is
(5)
where
is the total noise power per unit bandwidth avail-
able at the output port
4
at a corresponding output frequency
when the noise temperature of its input termination is a standard
290 K at all frequencies and
is that portion of en-
gendered at the input frequency by the input termination at the
standard noise temperature
5
290 K.
Anynoisy two-port network can be represented by a noiseless
two-port network with input equivalent voltage
and current
4
Actually, noises can be referred to any other node, e.g., input node, in the
circuit.
5
While the NF is a useful parameter in practice, it is an incomplete measure
of an LNA’s performance, as it is desirable to have a low NF while maintaining
a high gain. For example, feedback can be used to reduce
F
as close to unity as
possible, at the price of lowering the gain in the process [28], [47]. Cascading
multiple stages of such feedback amplifiers to recover the original gain will
result in a noise factor larger than or equal to the noise factor of the original
amplifierwithout feedback [48]. A more accurate measure of an amplifiers noise
performance, called the noise measure
M
is defined in [28] to take the effects
of both gain and NF into account.
