Sunteți pe pagina 1din 16

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO.

12, DECEMBER 2014 1


Consumption Factor and Power-Efciency Factor:
A Theory for Evaluating the Energy Efciency of
Cascaded Communication Systems
James N. Murdock, Member, IEEE and Theodore S. Rappaport, Fellow, IEEE
AbstractThis paper presents a new theory, called the con-
sumption factor theory, to analyze and compare energy efcient
design choices for wireless communication networks. The ap-
proach presented here provides new methods for analyzing and
comparing the power efciency of communication systems, thus
enabling a quantitative analysis and design approach for green
engineering of communication systems. The consumption factor
(CF) theory includes the ability to analyze and compare cascaded
circuits, as well as the impact of propagation path loss on the total
energy used for a wireless link. In this paper, we show several
examples how the consumption factor theory allows engineers to
compare and determine the most energy efcient architectures
or designs of communication systems. One of the key concepts
of the consumption factor theory is the power efciency factor,
which has implications for selecting network architectures or
particular cascaded components. For example, the question of
whether a relay should be used between a source and sink
depends critically on the ratio of the source transmitter power-
efciency factor to the relay transmitter power-efciency factor.
The consumption factor theory presented here has implications
for the minimum energy consumption per bit required to
achieve error-free communication, and may be used to extend
Shannons fundamental limit theory in a general way. This work
includes compact, extensible expressions for energy and power
consumption per bit of a general communication system, and
many practical examples and applications of this theory.
Index TermsPower Consumption, Energy Efciency, Power
Efciency, Millimeter-wave, Wireless, Cascaded circuits, Capac-
ity, Relay channel.
I. INTRODUCTION
C
OMMUNICATION systems today, including both wire-
line and wireless technologies, consume a tremendous
amount of power. For example, the Italian telecom operator
Telecom Italia used nearly 2 Tera-Watt-hours (TWh) in 2006
to operate its network infrastructure, representing 1% of Italys
total energy usage [1]. Nearly 10% of the UKs energy usage
is related to communications and computing technologies [1],
while approximately 2% of the USs energy expenditure is
dedicated to internet-enabled devices [2]. In Japan, nearly 120
W of power are used per customer in the cellular network
Manuscript received: April 15, 2012, revised: October 12, 2012. Portions
of this work appeared in the 2012 IEEE Global Communications Conference
(Globecom).
J. N. Murdock is Texas Instruments, Dallas, TX (e-mail:
james.murdock@ti.com). This work was done while James was a student at
The University of Texas at Austin.
T. S. Rappaport is with NYU WIRELESS at New York University and
NYU-Poly,715 Broadway, Room 702, New York, NY 10003 USA (e-mail:
tsr@nyu.edu).
Digital Object Identier 10.1109/JSAC.2014.141204.
[1]. Similar power/customer ratios are expected to hold for
many large infrastructure-based communication systems. [2]
estimates that 1000 homes accessing the Internet at 1 Giga-
bit-per-second (Gbps) would require 1 Giga-Watt of power.
All of these examples indicate that energy efciency of
communication systems is an important topic. Given the trend
toward increasing data rates and data trafc, energy efcient
communications will soon be one of the most important chal-
lenges for technological development, yet a theory that allows
an engineer to easily compare and analyze, in a quantitative
fashion, the most energy efcient designs has been allusive.
Past researchers have explored analytical and simulation
methods to compare and analyze the power efciencies
of various wireless networks (see, for example, works in
[3][4][5][6][7][8]). In [3], researchers explored the energy
efciency in an acoustic submarine channel and illustrated
how the choice of signaling, when matched to the channel,
could approach Shannons limit. In [4], researchers considered
a position-based network routing algorithm that could be
optimized locally at each user, in an effort to reduce overall
power consumption of the network, but were unable to derive
convenient and extensible expressions for power efciency
that could be generalized to any network. In [5], energy
consumption was compared to the obtainable data rate of end-
users, and an analysis technique was used to determine energy
efciency through the use of distributed repeaters. In [6], a
novel bandwidth allocation scheme was devised to optimize
the power consumed in the network while maximizing data
rate, but the analysis was not extensible to a cascaded system
of components, nor could it be easily generalized. [7] illus-
trates how cumbersome and complicated the eld of energy
conservation can be in ad-hoc networks, at both the link and
network layers (e.g. the individual wireless link, as well as the
network topology, where both have a strong impact on energy
utilization). In fact, a recent book, Green Engineering [8],
illustrates the importance, yet immense difculty, in providing
an easy, generalized, standardized method for analyzing and
comparing power efciency in a communications network.
Despite the extensive body of literature aimed at energy
efcient communication systems, we believe this paper is the
rst to present a generalized analysis that allows engineers
to provide a standard gure of merit to compare the power
efciency (or energy efciency) of different cascaded circuit
or system implementations over a wide array of problem
domains. The analysis method presented here is general,
in that it may be applied to power efcient circuit design,
0733-8716/14/$31.00 c 2014 IEEE
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
transmitter and receiver design, and also to various network
architectures such as relay systems (the relay problem in [5]
is validated using the CF theory in this paper).
This work has been motivated by the need to have a
compact, repeatable, extensible analysis method for comparing
the power efciency of communication systems and network
designs. In particular, as cellular communication networks
evolve, the base station coverage regions will continue to
shrink in size, meaning that there will be a massive increase
in the number of base stations or access points, and relays
are likely to complement the base stations over time [9].
To accommodate the demand for increased data rates to
mobile users, we envisage future millimeter-wave (mm-wave)
communication systems that are much wider in bandwidth
than todays cellular and Wi-Fi networks. These future systems
will use highly directional steerable antennas and channel
bandwidths of many hundreds of MHz thereby supporting
many Gigabits per second data rates to each mobile device
[9] [10][11][12]. As such systems evolve, small-scale fading
in the channel will become much less of a concern, and more
attention will need to be placed on the power efcient design
of handsets and light weight base stations and repeaters that
use wideband channels and multi-element phased arrays with
RF ampliers. The theory presented here aims to aid in the
design of these wideband wireless networks and devices. As
shown in this paper, the CF framework gives communication
engineers a methodology to analyze, compare and tradeoff
circuit and system design decisions, as well as network archi-
tectures (e.g. whether to use relays or small cells, and how
to trade off antenna gain, bandwidth, and power efciency in
future wireless systems) [9][10][11][12].
In this paper, we provide fundamental insight into the
required power consumption for communication systems, and
create an-easy-to-use theory, which we call the consumption
factor (CF) theory, for analyzing and comparing any cascaded
communication network for power efciency. In Section II
we present the consumption factor framework for a homo-
dyne transmitter [12]. Section III generalizes the concept
of power-efciency analysis, which is fundamental to the
consumption factor framework, for any cascaded communi-
cation system. Section IV provides numerical examples of the
power-efciency factor used in the consumption factor theory.
Section V presents a general treatment of the consumption
factor, based on the power-efciency analysis of the preceding
sections. Section VI demonstrates a key characteristic of the
power-efciency factor i.e. that gains of components that
are closest to the sink of a communication system reduce
the impact of the efciencies of preceding components. In
Section VII, we use the consumption factor framework to
develop fundamental understandings of the energy price of
a bit of information. We use our analysis to demonstrate how
the consumption factor theory may be applied to designing
energy efcient networks, for example by helping to determine
the best route to send a bit of information in a multi-hop
setting to achieve the lowest energy consumption per bit.
Section VIII provides conclusions. The key contribution of this
paper is a powerful and compact representation of the power
consumption and energy consumption per bit of a general
communication system. The representation takes the gains
Fig. 1. Block diagram of a homodyne transmitter used to demonstrate the
power-efciency factor and consumption factor (CF).
and efciencies of individual signal-path components (such
as ampliers and mixers) into account. A second key result is
that, in order to align the goals of lower energy per bit and
higher data rates, it is advantageous to design communication
systems that require as little signal power as possible, so
low, in fact, that ancillary power drain (e.g. for cooling, user
interfaces, etc.) dominates signal power levels. While this may
seem intuitive, the CF theory proves this, and provides a
tangible, objective way of comparing various designs while
showing the degrees to which communication systems must
reduce ancillary power drain, but must also seek means of
reducing required signal levels even more dramatically than
the ancillary power drain. By making every bit as energy
efcient as possible, we show it is possible to greatly expand
the number of bits that can be delivered for a given amount
of energy. Means of achieving this goal include the use very
short link distances (such as femtocells) at millimeter-wave
frequencies for future massively broadband wireless systems.
Earlier, less developed versions of the consumption factor were
presented in [13].
II. CONSUMPTION FACTOR FOR A HOMODYNE
TRANSMITTER
We dene the consumption factor (CF) for a communication
system as the maximum ratio of data rate to power consumed,
or equivalently as the maximum number of bits that may
be transmitted through a communication system for every
Joule of expended energy. A study of the consumption factor
requires a careful analysis of both the power consumption
and data rate capabilities of a communication system. In this
section, we will provide a simple analysis for a homodyne
transmitter as illustrated in Figure 1, to motivate the theory
presented here. We consider a homodyne transmitter because
this topology is attractive for many massively-broadband sys-
tems due to its low cost and low complexity [12]. We will
generalize our analysis in Section III to be applicable to a
general cascaded communication system.
The homodyne transmitter in Figure 1 is comprised of
components that directly handle the signal, such as the mixer
and power amplier, in addition to components that interact
indirectly with the signal, such as the oscillator. Components
that interact directly with the signal are designated on the
signal path, while components that are not in the path of the
signal are designated off the signal path.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 3
A key component of the consumption factor framework is
understanding that the efciency of each signal path compo-
nent may be used to relate the ancillary or wasted power
of each component to the total signal power delivered by that
component. For example, the efciency of the power amplier
is used to nd the partial power required to bias the amplier
(which is not used, or wasted, in terms of providing signal path
power) as a component of the total signal power delivered to
the load. We dene the efciency of the power amplier,
PA
,
and of the mixer,
MIX
,:

PA
=
P
PA
RF
P
PA
RF
+P
PA
NONRF
(1)

MIX
=
P
MIX
RF
P
MIX
RF
+P
MIX
NONRF
(2)
where P
PA
RF
is the signal power delivered by the power
amplier to the matched load, and P
MIX
RF
is the signal power
delivered by the mixer to the power amplier. P
PA
NONRF
and
P
MIX
NONRF
are the power levels used by the power amplier,
and mixer, respectively, that do not directly contribute to
delivered signal power. Using (1) (2), we nd:
P
PA
NONRF
= P
PA
RF
_
1

PA
1
_
(3)
P
MIX
NONRF
= P
MIX
RF
_
1

MIX
1
_
(4)
The second key step in the consumption factor analysis results
from the realization that the signal powers delivered by each
component in the cascade, P
PA
RF
, and P
MIX
RF
, may be related
to the total power delivered by the communication system,
through the gains of each signal path component. As shown
in Section III, this formulation for a cascaded systems power
efciency is reminiscent of Friis classic noise gure analysis
technique for cascaded systems [24]. Using (1)-(4), we now
nd the delivered RF power to the matched load, P
RADIO
RF
,
in terms of the signal power from the baseband signal source
and the various gains stages as:
P
RADIO
RF
= P
BB
SIG
G
MIX
G
PA
(5)
P
PA
NONRF
=P
PA
RF
_
1

PA
1
_
=P
RADIO
RF
_
1

PA
1
_
(6)
P
MIX
NONRF
=P
MIX
RF
_
1

MIX
1
_
=
P
RADIO
RF
G
PA
_
1

MIX
1
_
(7)
where P
BB
SIG
is the signal power delivered by the baseband
components to the mixer, and G
MIX
and G
PA
, are the power
gains of the mixer and power amplier, respectively. Equation
(5) simply states that the power delivered to the matched load
is equal to the power delivered by the baseband components
multiplied by the gain of the mixer and of the power amplier.
Note that we have implicitly assumed an impedance matched
environment. Impedance mismatches may be accounted for by
including a mismatch factor less than one in the gain of each
component.
The total power consumption of the homodyne transmitter
may be written as:
P
RADIO
consumed
= P
RADIO
RF
+P
PA
NONRF
+P
MIX
NONRF
+ P
BB
+P
OSC
(8)
where P
BB
is the power consumed by the baseband com-
ponents and P
OSC
is the power consumed by the oscillator.
The term P
RADIO
RF
is the total signal power in the homodyne
transmitter delivered to the load. Using equations (5) through
(7) in (8), we re-write the total homodyne power consumption
as:
P
RADIO
consumed
= P
RADIO
RF
_
1+
_
1

PA
1
_
+
1
G
PA
_
1

MIX
1
__
+ P
BB
+P
OSC
(9)
P
RADIO
consumed
=
P
RADIO
RF
_
1 +
_
1
PA
1
_
+
1
G
PA
_
1
MIX
1
__
1
+ P
BB
+P
OSC
(10)
From (10), the factor
_
1 +
_
1
PA
1
_
+
1
G
PA
_
1
MIX
1
__
1
plays a role in the transmitter power consumption analogous
to that of efciency. In other words, this factor may be
considered the aggregate efciency of the cascade of the
mixer and power amplier. In Section III we will generalize
this result and dene this factor as the power efciency factor
for an arbitrary cascaded system (where the cascade may
be either a cascade of components or circuits, or may even
include the propagation channel).
Now that we have formulated a compact representation
of the power consumption of a homodyne transmitter, we
must determine the maximum data rate that the transmitter
can deliver in order to formulate the consumption factor of
the transmitter. To do this, we assume that the transmitter is
communicating through a channel with gain G
channel
to a
receiver of gain G
RX
having noise gure F with bandwidth
B. We assume also that the transmitter matched load is
replaced by an antenna with gain G
ANT
TX
. The signal power
used by the receiver in the detection process,
P
RX
, is given by:
P
RX
= P
RADIO
RF
G
ANT
TX
G
channel
G
ANT
RX
G
RX
(11)
where G
ANT
RX
is the gain of the receiver antenna, and G
RX
is the gain of the receiver excluding the antenna. We will
assume an AWGN (Additive White Gaussian Noise) channel,
for which the received noise power at the detector, P
noise
, is:
P
noise
= KTFB G
RX
(12)
where K is Boltzmanns constant (1.38x10
23
J/K) and T is
the system temperature in Kelvin. The SNR at the receiver
detector is therefore:
SNR =
P
RADIO
RF
G
ANT
TX
G
channel
G
ANT
RX
G
RX
KTFB G
RX
(13)
The SNR is related to the minimum acceptable SNR at the
output of the receiver, SNR
min
, as dictated by the modulation
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
Fig. 2. An example of the use of the power-efciency factor to nd the
consumption factor of two different cascades of a baseband amp, mixer, and
RF amp.
and signaling scheme, through a particular operating margin
M
SNR
:
SNR = M
SNR
SNR
min
(14)
The minimum power consumption occurs when M
SNR
is
equal to 0 dB (i.e. M
SNR
= 1). Solving for P
RADIO
RF
, we
nd:
P
RADIO
RF,min
=
SNR
min
KTFB
G
ANT
TX
G
channel
G
ANT
RX
(15)
where we now denote P
RADIO
RF
as P
RADIO
RF,min
to indicate that
this power level corresponds to the minimum acceptable SNR
at the receiver. The minimum power consumption for the
transmitter is found using (10) and (15) as:
P
RADIO
consumed,min
=
SNRminKTFB
G
ANT
TX
G
channel
G
ANT
RX
_
1 +
_
1
PA
1
_
+
1
G
PA
_
1
MIX
1
__
1
+ P
BB
+P
OSC
(16)
The maximum data rate R
max
at the receiver is given in
terms of the SNR and the bandwidth according to Shannons
capacity formula if the modulation and signaling scheme are
not specied. If these are specied, then we nd the maximum
data rate in terms of the spectral efciency of the modulation
and signaling scheme
sig
(bps/Hz):
R
max
= Blog
2
(1 +SNR) , General Channel
R
max
= B
sig
, Specific ModulationScheme (17)
The consumption factor, CF, for the homodyne transmitter is
then found by taking the ratio of (17) to (16):
CF =
R
max
P
RADIO
consumed,min
(18)
CF =
Blog
2
(1 +SNR)
_
SNR
min
KTFB
G
ANT
TX
G
channel
G
ANT
RX

1+

PA
1

+
1
G
PA

MIX
1

1
+P
BB
+P
OSC
_
(19)
We will assume a standard log-distance channel gain model:
G
channel
= PG
o
+ 10 log
10
_
d
o
d
_
[dB] (20)
Fig. 3. Higher values of power consumption off of the signal path com-
ponents result in higher values of SNR needed to maximize the consumption
factor (CF).
where PG
o
is the close-in free-space path gain (usually a
large negative number in dB) received at a close-in reference
distance d
o
, d is the link distance ( d >d
o
), and is the
path loss exponent [10][12][19][20]. Two examples for CF
using equation (18) and (19) are shown in Figures 2 and 3.
Figure 2 shows how the consumption factor of a 60 GHz
wireless communication system varies as the efciency of the
power amplier or the mixer are changed, and indicates that
the efciency of the power amplier is much more important
in terms of maximizing the overall system efciency than the
mixers efciency. The key lesson from this example is that the
efciencies of the devices that handle the highest signal power
levels should be maximized in order to have the most dramatic
effect in maximizing the consumption factor. Figure 3 shows
the impact of changing the minimum required SNR at the
receiver. Note that we have assumed an SNR margin of 0 dB.
The gure indicates that higher levels of power consumption
by non-signal-path devices such as the oscillator result in
higher levels of SNR to maximize the consumption factor. The
gure also indicates an optimum value of SNR to maximize the
consumption factor. This optimum value depends critically on
the amount of power consumed by devices off the signal path.
Note that in these gures, we have assumed the efciency and
gain of the mixer are equal. This assumption will be explained
in Section III, where we will nd that the gain and efciency
of an attenuating device are equal (similar to Friis noise gure
analysis). Note that we have used a logarithmic scale in Figure
3 to allow for easy comparison between the different curves.
III. GENERAL CASCADED COMMUNICATION SYSTEM
We will now generalize the consumption factor to provide
a framework for analyzing a general cascaded communication
system.
The consumption factor is dened [18] as the maximum
ratio of data rate to total power consumption for a commu-
nication system. To determine the consumption factor, we
must rst determine a compact representation of the power
consumption of a general cascaded communication system.
Consider a general cascaded communication system as shown
in Figure 4 in which information is generated at a source, and
sent as a signal down a signal path to a sink. Signal path
components such as ampliers and mixers are responsible
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 5
Source
. . .
Sink
. . . . . . . . . .
Signal Path Devices
Non-Signal Path Devices
1 2 N
1 k M
Fig. 4. A general communication system composed of components on and
off the signal path.
for transmitting the information signal to the sink. Non-
signal path components include voltage regulation circuitry,
displays or cooling components that do not participate directly
in the signal path, but do consume power. The total power
consumption of the cascaded communication system in Figure
4 (ignoring the source and sink) may be written as:
P
consumed
= P
sig
+
N

k=1
P
nonsig
k
+
M

k=1
P
nonpath
k
(21)
where P
sig
is the sum of all signal powers of each component
in the cascade, P
nonsig
k
is the signal power used by the k
th
signal path component but not delivered as signal power to
the next signal-path component, and P
nonpath
k
is the power
used by the k
th
component off the signal path. To evaluate
(21), we must consider each component on the signal path
separately. The efciency of the i
th
signal path component
may be written as:

i
=
P
sigi
P
sigi
+P
nonsigi
(22)
Where P
sigi
is the total signal power delivered by the i
th
stage
to the (i + 1)
th
stage, and P
nonsigi
is the signal power used
by the i
th
stage component but not delivered as signal power.
This is a very general representation of efciency that may be
applied to any communication system component. A similar
measure of efciency, the PUE (Power Usage Effectiveness),
is already used to measure the performance of data centers,
and is the total power used for information technology divided
by the total power consumption of a data center[14].
Let us consider (22) applied to an attenuating stage, such as
a wireless channel or attenuator. Fundamentally, an attenuator
should consume only the signal power delivered to it by
the preceding stage (i.e. the consumption factor theory treats
attenuators as passive components that do not take power from
a power supply). The signal power delivered by an attenuator
to the next stage is a fraction of the signal power delivered to
the attenuator. Therefore, if the i
th
stage is an attenuator, then
the efciency of an attenuator,
atten
, as given by (22) is:
P
sigi
= G
atten
P
sigi1
(23)
P
nonsigi
= (1 G
atten
) P
sigi1
(24)

atten
=
G
atten
P
sigi1
G
atten
P
sigi1
+ (1 G
atten
) P
sigi1
= G
atten
(25)
where G
atten
is the gain of the attenuator, and is less than
one. Thus, we have shown that
atten
= G
atten
for a passive
device or channel.
The total power consumed by the i
th
stage on the signal
path may be written:
P
consumedi
= P
nonsigi
+P
addedsigi
(26)
where P
addedsigi
is the total signal power added by the
i
th
component, which is the difference in the signal power
delivered to the (i + 1)
th
component and the signal power
delivered to the i
th
component. We can sum all the signal
powers added by the components on the signal path (from
left to right in Figure 4) to nd:
N

i=1
P
addedsigi
= P
sigN
P
sigsource
(27)
where P
sigsource
is the signal power provided by the source,
and P
sigN
is the signal power delivered by the N
th
(and last
stage) signal-path component. Adding (27) to the signal power
from the source, we nd that the total signal power in the
communication system is equal to the signal power delivered
to the sink (in other words, the signal power delivered by the
last stage is equal to the sum of all signal powers delivered
by each component in the cascade):
P
sig
= P
sigN
(28)
From (22) the total wasted power of the k
th
stage (i.e. power
consumed but not delivered to the next signal path stage) may
be related to the efciency and total delivered signal power
by that stage:
P
nonsig
k
= P
sig
k
_
1

k
1
_
(29)
Also, the signal power delivered by the k
th
stage may be
related to the total power delivered to the sink by dividing by
the gains of all stages after the k
th
stage, (i.e. to the right of
the k
th
) thus yielding:
P
sigN=
P
sig
k
N

i=k+1
G
i
(30a)
P
nonsig
k
=
P
sigN
N

i=k+1
G
i
_
1

k
1
_
(30b)
where G
i
is the gain of the i
th
stage. We can therefore
compute the total power consumed by the communication
system as the power consumed by the source which is assumed
to equal the signal power delivered by the source, and the three
additional terms that represent the power consumed by the in-
path cascaded components, and the power dissipated by the
non-signal path components:
P
consumed
= P
sigsource
+
N

i=1
P
consumedi
+
M

k=1
P
nonpath
k
(31a)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
P
consumed
= P
sigsource
+
N

i=1
P
addedsigi
+
N

k=1
P
nonsig
k
+
M

k=1
P
nonpath
k
(31b)
= P
sig
N

1 +
N

k=1
1
N

i=k+1
G
i
_
1

k
1
_

+
M

k=1
P
nonpath
k
(31c)
In certain circumstances, such as when comparing two dif-
ferent smart phones or other devices that have substantial
power consumed by displays or computer processors, it is
useful to incorporate the impact of the power efciencies of
the non-path components. To do this, we may simply add these
components to the end of the signal-path cascade in Figure 4
and assume unity gain. For example, write the total power
consumption of the k
th
non-path component P
nonpath
k
in terms of its usefully dissipated power P
u
k
(power that
directly contributes to its intended functionality) and its power
efciency
nonpath
k
(the ratio of usefully dissipated power
to its total power consumption):
P
nonpath
k
=
P
u
k

nonpath
k
(32)
We may then re-write (31c) as (33). For simplicity, we now
carry on the development of the CF analysis with the power
consumption expression given by (31c) rather than (33), as
we wish to isolate the impact of the efciencies of non-path
components (noting that such analysis may be done by simply
appending the power efciencies of non-path components as
described above). We see from (31c) and (35) that the on-path
cascade components may be conveniently represented in the
total power consumption of the cascade as:
P
consumed
=
P
sigN
H
+P
nonpath
(34)
where P
nonpath
is the total power used by devices off the
signal path, and in (34), we introduce the system power-
efciency factor H of all cascaded components dened as:
H =

1+
N

k=1
1
N

i=k+1
G
i
_
1

k
1
_

1
(35)
Where H ranges between 0 and 1, and we call H the power-
efciency factor of the entire signal-path cascade. Note H
1
ranges from 1 to innity (just like Friis Noise Figure). Equa-
tion (35) is a very general expression relating the gains and
power efciencies of the individual components on the signal
path to the signal-path efciency of the overall communication
system. An implication of this is that the efciencies of devices
that handle the most power are most important in terms of
the power-efciency factor of the entire cascade, as these will
be the components in (35) whose efciencies are divided by
the smallest numbers. As shown subsequently, the presence of
attenuators, such as a wireless channel, makes it such that the
power efciencies of stages that handle the most power just
prior to the large attenuator, such as a power amplier, have
the largest impact on overall system power efciency.
Note that we have dened the power efciency of a signal-
path component (22) in terms of the total power it delivers,
i.e. using (26) in (22) we have:

i
=
P
sigi
P
sigi
+P
nonsigi
=
P
sigi1
+P
addedsigi
P
sigi1
+P
addedsigi
+P
nonsigi
(36a)

i
=
P
sigi1
+P
addedsigi
P
sigi1
+P
consumedi
(36b)
This representation of power efciency is useful as it is
applicable to both passive components, which do not add
signal power, and active components that add signal power.
For an attenuator, P
addedsigi
is zero (i.e. for the CF theory
we assume that attenuators can not add additional signal
power), while P
nonsigi
is the signal power removed by the
attenuator. One important application of (36) and (25) is when
the attenuator is a wireless channel. From (25), the power-
efciency factor of a wireless channel H
channel
is given by
its gain G
channel
:
H
channel
= G
channel
(37)
The power-efciency factor is a powerful and general means
of determining the power consumption of a communication
system. For example, consider two cascaded sub-systems
whose Hs have already been characterized, where sub-system
#2, with power-efciency factor H
subsystem2
and gain
G
subsystem2
follows sub-system #1 with power-efciency
factor H
subsystem1
. We can show from (35) that the power-
efciency factor of the entire cascade, H
cascaded system
, may
be written much like the classic noise gure theory (Eqn.
38) where the rst sub-system is composed of components
1 through M-1, and the second subsystem is composed of
components M through N. Of course, M may be any integer
from 1 through M, so (38d) is a completely general result.
Note from (38) that the power-efciency factor of
the second stage H
subsystem2
is an upper bound for
H
cascaded system
. This is easily seen by inverting (38d) and
examining the limiting case in which the rst stage has an
optimal power-efciency factor of 1 as in (39). Consider also
the case of a single component [9]. Using (22) and (35), we
see:

i
= H
i
(40)
Consider now the case in which a wireless channel exists be-
tween a transmitter and receiver. The overall power-efciency
factor of the entire transmitter-receiver pair, H
link
is given by:
H
1
link
= H
1
RX
+
1
G
RX
_
1
G
channel
1
_
+
1
G
RX
G
channel
_
H
1
TX
1
_
(41)
where H
RX
is the power-efciency factor of the receiver, H
TX
is the power-efciency factor of the transmitter, G
RX
is the
gain of the receiver, and G
channel
is the channel gain (which
is less than 1) which is equal to the power-efciency factor of
the channel. Note from (41) that if the receiver gain is much
smaller than the expected channel path loss, the cascaded
power-efciency factor will be very small and on the order
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 7
P
consumed
= P
sig
N

1 +
1
P
sigN
M

k=1
P
u
k

nonpath
k
+
N

k=1
1
N

i=k+1
G
i
_
1

k
1
_

(33)
H
1
cascadedsystem
=
_
1+
_
1

N
1
_
+
1
G
N
_
1

N1
1
_
+. . .+
1
G
N
. . . G
M+1
_
1

M
1
_
+
1
G
N
. . . G
M
_
1

M1
1
_
+ . . .
1
N

i=1
G
i
_
1

1
1
_

1
(38a)
H
1
subsystem2
= 1+
_
1

N
1
_
+
1
G
N
_
1

N1
1
_
+. . . +
1
G
N
. . . G
M+1
_
1

M
1
_
(38b)
H
1
subsystem1
= 1 +
_
1

M1
1
_
+. . .
1
M1

i=1
G
i
_
1

1
1
_
(38c)
H
1
cascadedsystem
= H
1
subsystem2
+
1
G
subsystem2
_
H
1
subsystem1
1
_
(38d)
H
cascadedsystem
=
H
subsystem1
H
subsystem2
H
subsystem1
+
H
subsystem2
G
subsystem2
(1 H
subsystem1
)
(39a)
lim
H
subsystem1
1
H
cascadedsystem
=H
subsystem2
(39b)
of the product of the channel gain with the receiver gain. In
this case, we nd that the overall power-efciency factor is
approximated by:
H
link
G
RX
G
channel
H
TX
(42) (42)
This is an important result of this analysis. In particular, it
indicates that in order to achieve a very power-efcient link,
it is desirable to have a high gain receiver and a highly
efcient transmitter. This can be understood by realizing that
a higher gain receiver reduces the output power requirements
at the transmitter. Eqn. (42) indicates the great importance
of the transmitter efciency. Note, however, that the receiver
efciency is still important, as from (39b) it is clear that the
receivers efciency is an upper bound on the efciency of the
overall link.
IV. NUMERICAL EXAMPLES
To better illustrate the use of the consumption factor theory,
and the use of the power-efciency factor, consider a simple
scenario of a cascade of a baseband amplier, followed by
a mixer, followed by an RF amplier. We will consider two
different examples of this cascade scenario, where different
components are used, in order to compare the power ef-
ciencies due to the particular specications of components.
Assume that for both cascade examples, the RF amplier is a
commercially available MAX2265 power amplier by Maxim
technology with 37 % efciency[15]. In both cases, the mixer
is an ADEX-10L mixer by Mini-Circuits with a maximum
conversion loss of 8.8 dB[16]. In the rst case, the baseband
amplier (the component furthest to the left in Figure 4 if
in a transmitter, and furthest to the right if in a receiver)
is an ERA-1+ by Mini-circuits, and in the second case the
baseband amplier is an ERA-4+ [17], also by Mini-Circuits.
The maximum efciencies of these parts are estimated by
taking the ratio of their maximum output signal power to their
dissipated DC power. As the mixer is a passive component,
its gain and efciency are equal. Table 1 summarizes the
efciencies and gains of each component in the cascade. Using
(35), the power-efciency factor of the rst scenario is
H
scenario 1
=
1
1
0.37
+
1
16.17
_
1
0.36
1
_
+
1
0.3616.17
_
1
0.1165
1
_
= 0.2398,
whereas the power-efciency factor of the second scenario is
H
scenario 2
=
1
1
0.37
+
1
16.17
_
1
0.36
1
_
+
1
0.3616.17
_
1
0.1836
1
_
= 0.2813.
Therefore, we see that the second scenario offers a superior
efciency compared to the rst scenario, due to the better
efciency of the baseband amplier, but falls far short of
the ideal power efciency factor of unity. Using different
components and architectures, it is possible to characterize and
compare, in a quantitative manner, the power-efciency factor
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
TABLE I
AN EXAMPLE OF THE USE OF THE POWER-EFFICIENCY FACTOR TO COMPARE TWO CASCADES OF A BASEBAND AMP, MIXER, AND RF AMP.
Component Gain Efciency
Example 1
MAX2265 RF Amp 24.5 dB (voltage gain of 16.17) 37%
ADEX-10L Mixer -8.8 dB 36%
ERA-1+BB Amp 10.9 dB 11.65 %
Example 2
MAX2265 RF Amp 24.5 dB 37%
ADEX-10L Mixer -8.8 dB 36%
ERA-1+BB Amp 13.4 dB 18.36%
and consumption factor (see subsequent sections) of cascaded
components.
As a second example, consider the cascade of a transmitter
power amplier communicating through a free-space channel
with a low-noise amplier at the receiver. Let us assume that
the cascade, in the rst case, uses the same RF power amplier
as in the previous example (MAX2265), while the LNA is
a Maxim Semiconductor MAX2643 with a gain of 16.7 dB
(6.68 absolute voltage gain) [18]. We will assume this LNA
has 100% efciency for purposes of illustrating the impact
of the PAs efciency and the channel (i.e. here we ignore the
LNAs efciency, although this can easily be done as explained
above). For a carrier frequency of 900 MHz, now consider
the cascade for a second case where the MAX2265 RF power
amplier is replaced with a hypothetical RF amplier device
having 45% power efciency (a slight improvement). Assume
the link is a 100m free space radio channel with gain of -71.5
dB. Since the propagation channel loss greatly exceeds the
LNA gain, (42) applies, where H
TX
is the efciency of the RF
amplier, so that in the rst case using the MAX2265 amplier
(37% efciency), the power efciency factor of the cascaded
system is 173.5e-9, while in the second case (using an RF
Power amplier with 45% efciency), the power efciency
factor is 211.02e-9. The second case has an improved power-
efciency factor commensurate with the power efciency
improvement of the RF amplier stage in the receiver. These
simple examples demonstrate how the power-efciency factor
may be used to compare and quantify the power efciencies
of different cascaded systems, and demonstrate the importance
of using higher efciency RF ampliers for improved power
efciency throughout a transmitter-receiver link.
V. CONSUMPTION FACTOR
We now dene the consumption factor, CF, and operating
consumption factor (operating CF) for a general communi-
cation system such as that in Figure 4, where CF is dened
as:
CF =
_
R
P
consumed
_
max
=
R
max
P
consumed,min
(43)
operating CF =
R
P
consumed
(44)
where R is the data rate (in bits-per-second or bps), and R
max
is the maximum data rate supported by the communication
system. Further analysis based on only maximizing R or
minimizing P
consumed
is also pertinent to system optimization
in terms of consumed power and carried data rate. For a very
general communication system in an AWGN channel, R
max
may be written using Shannons information theory according
to the operational SNR and bandwidth, B:
R
max
= Channel Capacity = Blog
2
(1 +SNR) (45)
Or, for frequency selective channels [3]:
R
max
=
_
B
0
log
2
_
1 +
P
r
(f)
N (f)
_
df
=
_
B
0
log
2
_
1 +
|H (f)|
2
P
t
(f)
N (f)
_
df (46)
where P
r
(f), P
t
(f), and N (f) are the power spectral densi-
ties of the received power, the transmitted power, and the noise
power at the detector, respectively. H (f) is the frequency
response of the channel and any blocks that precede the de-
tector. Note that equations (45) and (46) make no assumptions
about the signaling, modulation, or coding schemes used by
the communication system. To support a particular spectral
efciency
sig
(bps/Hz), there is a minimum SNR required
for the case of an AWGN channel:
SNR
M
SNR
= SNR
min
= 2
sig
1 (47)
The operating SNR of the system, as well as the operating
margin of the operating SNR ( denoted by M
SNR
which
represents the operating margin above the minimumSNR
min
)
may be used to nd the consumption factor and operating
consumption factor expressed in terms of the systems power-
efciency factor H:
CF =
Blog
2
(1 +SNR)
P
nonpath
+
_
SNR
MSNR
_

Pnoise
H
(48a)
CF =
Blog
2
(1 +M
SNR
(2
sig
1))
P
nonpath
+ (2
sig
1 )
Pnoise
H
, (48b)
and (49) where we have made use of (34) and the fact that
the signal-power available to the sink, P
sigN
is related to the
noise power available to the sink, P
noise
and the SNR at the
sink:
P
sigN
= P
noise
SNR = KTFBG
RX
SNR (50)
And where the right hand equality in (50) holds for an AWGN
channel. K is Boltzmans constant (1.38x10
23
J/K), T is the
system temperature (degrees K), F is the receiver noise factor,
and B is the system bandwidth.
There is an important implication of the consumption factor
that relates to the selected cell size and capacity of future
wireless broadband cellular networks. To see this, consider
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 9
Operating CF =
Blog
2
_
SNR
MSNR
+ 1
_
P
nonpath
+SNR
Pnoise
H
(49a)
Operating CF =
B
sig
P
nonpath
+M
SNR
(2
sig
1)
Pnoise
H
(49b)
two limiting cases illuminated by the consumption factor
theory. In the rst case, we assume that the signal path
power consumption is the dominant power drain for a link,
as opposed to the non-path power. This may be the case,
for example, in a macrocell system in which a base-station
is communicating to the edge of the macrocell, and the RF
channel requires more power to be used in the RF amplier
to complete the link than the power used to power other
functions. In this case, the consumption factor equation (48a)
is approximated by:
CF H
Blog
2
(1 +SNR)
_
SNR
MSNR
_
P
noise
. (51)
For an AWGN channel, we nd that the consumption factor
is relatively insensitive to bandwidth if the signal-path power
dominates the non-path power:
CF H
log
2
(1 +SNR)
_
SNR
MSNR
_
KTFG
RX
. (52)
Equation (52) indicates that for such a link we can increase
data rate by increasing bandwidth, but that unless the signal
path components are made much more efcient (i.e. the system
power-efciency factor is made closer to 1), then as data rate
increase we will require approximately the same energy per
bit. In other words, if transmission power is the dominant
cause of energy expenditure, then there is little that can
be done to drive down the energy-price per bit through an
increase in bandwidth. There are two problems that arise: A)
efciency improvements in inexpensive IC components are
becoming harder to achieve due to performance issues when
supply voltages are scaled below 1 volt, which is approx-
imately the supply voltage used by many present-day high
efciency devices, and B) with the exponential growth in data
trafc that is occurring today, unless the energy cost per bit can
be reduced exponentially, we face an un-tenable requirement
for increased power consumption by communication systems.
The upshot of (52) is that for conventional cellular systems, all
signal-path devices, and particularly the RF power amplier
the precedes the lossy channel, and other components that
precede lossy attenuators, must be made as power efcient as
possible, thus suggesting that modulation/signaling schemes
should be chosen to support as efcient an RF amplier as
possible.
Consider the second limiting case of equation (48a), in
which the non-path power dominates the signal power. In
this case, we are assuming that items such as processors,
displays, and other non-signal path components (typical of
smart-phones and tablets) dominate the power drain. We nd
from (48a) that in this case:
CF
Blog
2
(1 +SNR)
P
nonpath
. (53)
Eqn. (53) indicates that wider-band systems are preferable
on an energy-per-bit basis provided that signal-power can be
made lower than the total power used by components off the
signal path. This situation is clearly preferable to the rst case
as it indicates that by increasing channel bandwidth (say, by
moving to millimeter-wave spectrum bands where there is a
tremendous amount of spectrum [3][12][13]), we also achieve
an improvement in the consumption factor, i.e. a reduction in
the energy cost per bit. Interestingly, this indicates that the
goals of massive data rates (through larger) bandwidths and
smaller cell sizes combined together can be used to achieve
a net reduction in the energy cost per bit. As an increase in
bandwidth also enables an increase in data rate, this limiting
case allows us to simultaneously increase both data rate and
consumption factor: i.e. our goals for more data and more
efcient power utilization in delivering this higher speed data
are aligned. This is not to say that we should increase non-
path power to the point that equation (53) holds. Rather, we
would desire to decrease the required signal path power to
the point where (53) holds. If the non-path power can be
reduced, but the signal power can be reduced even faster, then
we arrive at the ideal situation of improving power efciency
with a move to higher bandwidths and greater processing
and display capabilities in mobile devices. To achieve this
goal, it is likely that link distances will need to be reduced
as bandwidths are increased. The goal of making the signal
power as low as possible so that the non-path power dominates
may at rst be counter-intuitive. However, realize that in
order to have as many bits as possible owing through a
communication system it is advantageous to make each bit
as cheap as possible. In order for (53) to apply, we require:
P
nonpath
>
_
SNR
M
SNR
_

P
noise
H
(54)
Recall the form of the power-efciency factor of a wireless
link given by (41). We will model the channel gain as:
G
channel
=
k
d

(55)
Where d is the link distance, is the path loss exponent,
(which equals 2 for free space), and k is a constant. Using
(55) in (41) and (54), we nd (56). And by isolating distance,
we nd
d

<
P
NP
M
SNR
P
noise
SNR
G
RX
H
TX
k
kH
TX
H
RX
(G
RX
H
RX
)

P
NP
M
SNR
P
noise
SNR
G
RX
H
TX
k
kH
TX
H
RX
(G
RX
) (57)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
P
nonpath
>
_
SNR
M
SNR
_
P
noise
_
H
1
RX
+
1
G
RX
_
d

k
1
_
+
d

G
RX
k
_
H
1
TX
1
_
_
(56)
and when further simplifying, we see
d

< G
RX
H
TX
k
_
P
NP
M
SNR
P
noise
SNR

1
H
RX
_
(58)
and, nally, solving for distance, we see that
d <
_
G
RX
H
TX
k
_
P
NP
M
SNR
P
noise
SNR

1
H
RX
__1

(59)
If (59) is satised, then increasing bandwidth will have the
double benet of enabling both increased data rates and
higher consumption factors i.e. lower energy consumption
per bit. One caveat to (59) is that an increase in bandwidth
clearly also requires a smaller radio link distance in order for
(59) to apply. To ensure that this is the case, we may introduce
a scaling factor > 1 to ensure that by increasing bandwidth
within a given bound, we do not violate (59):
d <
1

_
G
RX
H
TX
k
_
P
NP
M
SNR
P
noise
SNR

1
H
RX
__1

(60)
Re-writing (60) in terms of bit rate, we nd that for a given
operating SNR:
R = Blog
2
_
SNR
M
SNR
+ 1
_
(61)
where
M
SNR
SNR
=
1
2
R
B
1
(62)
hence both data rates and consumption factors increase when
d <
1

_
G
RX
H
TX
k
_
P
NP
KTFB
1
2
R
B
1

1
H
RX
__1

(63)
In addition to characterizing the power consumption, power-
efciency factor, and CF of a transmitter-receiver pair, the
consumption factor framework may be applied to an individual
transmitter or receiver. To do this for a transmitter, simply
replace the transmitter antenna and channel with a matched
dummy load as was done in Section II for the homodyne
receiver. Similarly, to analyze the case of a receiver, simply
apply a passive matched source to the receiver input.
The preceding analysis uses a distant-dependent large-scale
spatial channel model that represents the channel path loss as
a function of distance between the transmitter and receiver,
as expressed by the path loss exponent (see equation (20)).
As channel bandwidths increase to several hundreds of MHz
at millimeter-wave bands, recent propagation measurements
show that small scale fading is almost neglible, and large-scale
fading is less variable with directional antennas that nd the
best pointing directions at both the transmitter and receiver
[10][19][20] thus validating (20) as a reasonable rst-order
assumption.
An interesting extension of the theory presented here, which
is beyond the scope of this paper, would consider more
sophisticated channel models that include fading or variability
due to transients in beam switching, or the power efciencies
and power consumption tradeoffs for various antenna array
hardware or beam steering processing needed to implement
future mm-wave cellular networks. For example, antennas that
exploit multipath or beam combining, and can be beamsteered
towards the strongest reections will be used in future wireless
networks [11][21][22][23]. In [22][23], it was shown that
the direction of arrival of multipath energy for a steerable
antenna can be found by measuring the cross correlation of
narrowband (e.g. CW) fading signals, thus suggesting that fu-
ture broadband millimeter-wave devices might simultaneously
use narrow band pilot tones that can be detected by closely
spaced low gain omni-directional antennas on the handset or
base station, while the communication trafc is simultaneously
carried using high gain (narrow beam) steerable directional
antennas[11][10]. The CF theory can be easily extended to
analyze the power tradeoff for this additional antenna com-
plexity (and many others). This is readily seen by considering
the homodyne transmitter example, where equations (1), (3)
and (11) may be used to represent the power consumption and
power efciency of a transmitting antenna that is actually a
combined phased array antenna with multiple RF power ampli-
er stages. The power drain caused by signal processing would
be represented in the efciency and power consumption of the
off-path components (e.g., the signal processing components)
as represented in equations (21) and (22), or (33) and (34) in
the general result. As should be clear, by quantifying the addi-
tional power consumption and power efciencies of different
types of processing requirements and hardware requirements,
the CF theory allows for a quantitative comparison of a wide
range of circuit and system implementations. We now illustrate
some numerical examples to highlight the use of this analysis
method, and show how to apply the CF analysis to network
architectures (e.g. relay systems) subsequently in the paper.
VI. CF AND POWER-EFFICIENCY FACTOR EXAMPLE
To illustrate some pertinent effects of the Consumption
Factor and power-efciency factor of a communication system,
rst consider how the efciencies of the individual blocks in a
communication system impact the power-efciency factor H.
For simplicity, we will consider H
TX
, the power-efciency
factor of a transmitter. First assume the transmitter is com-
posed of a cascade of N stages, each with an absolute power
gain of G. Further, assume that we may vary the efciency of
the i
th
stage in the transmitter from 0 to 100%. The rest of
the stages are assumed 100% efcient. For this case, H
TX
is
given by:
H
TX
=
1
1 +G
iN
_
1

TX
i
1
_ (64)
(64) is plotted in Figure 5 for a seven-stage transmitter, with
each block having an absolute power gain of 2. The gure
makes it clear that stages closer to the output of the transmitter,
which would be closer to the sink if the transmitter were
used with a receiver, have the largest impact on H
TX
. Further,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 11
Fig. 5. If all the blocks in the communication system have positive gain,
then the efciencies of the blocks closest to the sink will have the most
impact on the overall systems power-efciency factor. In other words, it is
most important to maximize the efciencies of components that handle the
highest power levels.
equations (48) (49) make clear that as H
TX
increases from
0 to 1, the Consumption Factor also increases.
VII. ENERGY PER BIT
Shannons limit describes the minimum energy-per-bit-per-
noise spectral density required to achieve arbitrarily low prob-
ability of bit error through proper coding scheme selection:
E
b
N
o
= ln (2) (65)
This limit is generally found by using Shannons capacity
theorem, and allowing the code used to occupy an innite
bandwidth [24].
As shown in (48) the CF is given as the maximum ratio of
data rate to power for a communication system, and may be
written as:
CF =
Blog
2
(1 +SNR)
P
NP
+SNR
min

Pnoise
H
(66)
Let us take the limit of (66) as bandwidth approaches innity,
assuming AWGN. This is equivalent to allowing our systems
coding scheme to spread out innitely in bandwidth, driving
our SNR down to the minimum acceptable to still achieve ar-
bitrarily low error. First, recall that the SNR may be written in
terms of the energy-per-bit E
b
, the time required to transmit a
single bit T
b
, the noise spectral density N
o
, and the bandwidth
of the system B [24]:
SNR =
E
b
T
b
N
o
B
(67)
In the limit as B approaches innity, the SNR approaches the
minimum acceptable SNR. Therefore we have (68). Where
E
bc,min
is the minimum energy per bit that must be consumed
by the communication system, and E
b
is the minimum energy-
per-bit that must be present in the signal carried by the
communication system and delivered to the receivers detector.
Note that E
bc,min
and E
b
are not equal, as E
bc,min
is the
amount of energy consumed/expended by the communication
system per bit (including the operation of ancillatory functions
such as powering non-signal path components like oscillators),
while E
b
is the amount of energy per bit in the signal
itself. Note that the denominator is no longer a function of
bandwidth. We may therefore apply Shannons theorem [24]
to nd:
1
E
bc,min
=
E
b
N
o
T
b
ln (2)

1
P
NP
+
E
b
T
b
H
(69)
E
bc,min
=
ln (2) N
o
E
b
P
NP
C
+
N
o
ln(2)
H
=
P
NP
C
+
ln (2) N
o
H
(70)
where we have made the substitution C =
1
T
b
, i.e. that in
the limit, the bit-rate approaches the channel capacity C.
Eqn. (70) should be interpreted as the minimum energy that
must be expended/consumed by the communication system
per bit (as opposed to the energy per bit in the signal) over
the noise spectral density in order to obtain arbitrarily low
error rate. This interpretation should not be confused with
the interpretation of the original Shannon limit, which relates
to the bit energy per noise spectral density within the signal
that ows through the communication system. Note that if the
system is 100 % efcient on the signal-path, and no power is
used off the signal path, then (70) degenerates to Shannons
limit, indicating that in effect the communication system and
the signal it carriers have become identical. Equation (70)
indicates the importance of the power-efciency factor of
a communication system in determining the true, practical
energy cost of a single bit. Note also from (34) that the total
power consumption to send a single bit is given by P
bc,bit
:
P
bc,bit
= P
NP
+
E
b
C
H
= CE
bc,min
. (71)
Bits delivered to the edge of a wireless cell (greater propa-
gation distance) are expected to be the most costly from an
energy perspective. It is instructive to estimate the required
power consumption per bit as a function of a cell radius for
a single user at the edge of the cell.
Recall rst that the power-efciency-factor over a wireless
link may be written with (41) as (72), where G
channel
is the
link channel gain, H
RX
is the power-efciency factor of the
receiver, H
TX
is the power-efciency factor of the transmitter,
and G
RX
is the gain of the receiver. Using (72) in (70), we
nd (73).
If we factor out the inverse of the channel gain, we nd
(74). In the limit of very small channel gains (see (42)), this
yields:
H
link
H
TX
G
RX
G
channel
(75)
E
bc,min
=
P
NP
C
+
ln (2) N
o
H
TX
G
RX
G
channel
(76)
The interpretation of (76) is that for cases in which
G
RX
G
channel
is much smaller than unity (i.e. a highly
attenuating wireless channel), the stage immediately after
the attenuation should have high gain, so that the stage
immediately before does not need to have an extremely high
output power, resulting in increased loss. Secondly, we see
the importance of power amplier efciency and the need to
overcome the loss incurred in the channel.
Equation (76) conrms that the energy cost of a single bit
does indeed increase as the channel gain decreases. Several
examples using equation (76) are shown in Figure 6 through
Figure 9. Figures 6 and 7 show an example of a 20 GHz carrier
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
12 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
lim
B
CF =
1
E
bc,min
= lim
B
_
(Blog
2
(1 +SNR
min
))
P
NP
+SNR
min

Pnoise
H
_
= lim
B

_
Blog
2
_
1 +
E
b
T
b
NoB
__
P
NP
+
E
b
T
b
NoB

NoB
H

(68)
H
link
=
H
RX
H
TX
G
RX
G
channel
H
TX
G
RX
G
channel
+ H
RX
H
TX
(1 G
channel
) + H
RX
(1 H
TX
)
(72)
E
bc,min
=
P
NP
C
+ ln (2) N
o
_
H
1
RX
+
1
G
RX
_
1
G
channel
1
_
+
1
G
RX
G
channel
_
H
1
TX
1
_
_
. (73)
E
bc,min
=
P
NP
C
+
ln (2) N
o
G
RX
G
channel
_
G
RX
G
channel
H
1
RX
+ H
1
TX
G
channel
_
(74)
system with path loss modeled according to a log-distance
break-point model.
Contrasting Figures 6 and 7 shows that highly efcient
systems can afford to use longer link distances while systems
with less efcient signal path components should use shorter
distances. The decrease in efciency is reected in the change
in H
TX
and H
RX
between the two plots. Figures 8 and 9,
where the carrier frequency has been increased to 180 GHz,
for which k is higher due to atmospheric absorbtion [12],
indicate that shorter link distances should be used for higher
carrier frequencies (k is the value of the path loss at a close-in
reference measurement distance).
With equation (74), we can determine the maximum wire-
less transmission distance d for which non-path power domi-
nates the power expenditure per bit, and hence the maximum
distance before each bit becomes progressively more energy-
expensive:
P
NP
C
>
ln (2) N
o
G
RX
G
channel
__
G
RX
H
RX
1
_
G
channel
+H
1
TX
_
(77)
G
channel
>
ln(2) N
o
C
H
TX
_
P
NP
G
RX
+N
o
Cln(2)
_
1
GRX
HRX
__ (78)
If we model the channel gain as (55), we nd:
k
d

>
ln(2) N
o
C
H
TX
_
P
NP
G
RX
+N
o
Cln(2)
_
1
GRX
HRX
__ (79)
and (80). If P
NP
<
NoCln(2)
HRX
, then (80) is unlikely to have
a positive solution due to the small value of ln(2)N
o
C. Our
interest in keeping the non-path energy per bit larger than
the signal energy per bit stems from interest in making every
bit as cheap from an energy perspective as possible while
simultaneously achieving the goal of a higher capacity. As
discussed in Section V, by forcing non-path energy to be larger
than signal energy per bit, we can achieve the simultaneous
Fig. 6. For a system with high signal path efciency and high non-path power
consumption, we see that the energy expenditure per bit is dominated by non-
path power, indicating little advantage to shortening transmission distances.
goals of lower energy per bit and higher capacities through an
increase in bandwidth.
Note that the maximum value of d that ensures that non-
path power exceeds signal power increases as the amount of
non-path power increases. Also, as the gain of the receiver in-
creases, the link distance may be extended while still achieving
a lower price-per-bit through an increase in bandwidth versus
a lower receiver gain system. As expected, as the path loss
exponent increases, the maximum value of d decreases, as
indicated by (80).
Required energy consumption per bit can be used to eval-
uate the energy requirements of multi-hop versus single-
hop communications. For example, consider the situation
illustrated in Figure 10 in which the source and sink may
communicate directly or through a relay. A similar analysis,
[4], showed the importance of path loss exponent and how
it can often be benecial to transmit through relay nodes,
especially with high path loss exponents. Work such as that by
[5] similarly attempts to determine under what circumstances
it is advantageous to use a relay from an energy perspective
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 13
d <
_
H
TX
k
ln (2) N
o
C
_
P
NP
G
RX
+N
o
Cln(2)
_
1
G
RX
H
RX
___1

(80)
Fig. 7. When signal-path components are less efcient, as illustrated here,
then shorter transmission distances start to become advantageous, as signal-
path power starts to represent a larger portion of the power expenditure per
bit.
Fig. 8. A higher carrier frequency system that provides a much higher bit
rate capacity (e.g. bandwidth) without substantially increasing non-path power
consumption may result in a net reduction in the energy price per bit.
based on the placement of the relay. But, to our knowledge
this paper presents the rst such treatment in terms of the
gain of the sink and power-efciency of the source transmitter.
Consider a three node network as shown in Figure 10. If the
path loss exponent for the network is , then the required
power to transmit over a given distance d is proportional to d

.
Here we extend the analysis of [4] to account for the gain and
efciencies of the devices participating in the network. The
results indicate when, on a per bit basis, it is advantageous to
use the relay or the direct path from source to sink.
To determine the difference in energy consumption that
would result from using a double-hop versus a single hop
as illustrated in Figure 10, we simply compare the energy
consumptions for each link. For the single link through
distance d
3
, we nd the minimum energy consumption per
bit E
bc,min,direct
:
E
bc,min,direct
=
P
NP
d3
C
d3
+
N
0
ln (2)
H
d3
(81)
H
d3
G
RX,sink
G
channel,d3
H
source
(82)
where C
d3
is the capacity of the direct link, P
NP
d3
is the
non-path power associated with the direct link, H
d3
is the
Fig. 9. Lower efciencies of signal-path components motivates the use of
shorter transmission distances.
power efciency factor of the direct link, H
source
is the power-
efciency factor of the source transmitter, G
RX,sink
is the
gain of the sink receiver, and G
channel,d3
is the channel gain
through the direct link. For the link through the relay, we nd
the minimum energy consumption per bit E
bc,min,relay
:
E
bc,min,relay
=
P
NP
d1
C
d1
+
P
NP
d2
C
d2
+
N
0
ln (2)
H
d1
+
N
0
ln (2)
H
d2
(83)
H
d1
G
RX,relay
G
channel,d1
H
source
(84)
H
d2
G
RX,sink
G
channel,d2
H
relay
(85)
where C
d1
and C
d2
are the capacities of the links through
distances d
1
and d
2
, H
d1
and H
d3
are the power-efciency
factors associated with the links through d
1
and d
2
, P
NP
d1
and P
NP
d2
are the non-path powers used by the links through
distance d
1
and d
2
. G
RX,relay
is the gain of the relay receiver,
H
relay
is the power-efciency factor of the relay transmitter,
and G
channel
d1
and G
channel
d2
are the channel gains through
link distances d
1
and d
2
, respectively. Assume that all non-
path powers are equal P
NP
d1
= P
NP
d2
= P
NP
d3
= P
NP
, and
that the capacities through each link are equal to C. The ratio
of the minimum energy consumption per bit in both cases
determines when it is advantageous to transmit through the
relay:
E
bc,min,relay
E
bc,min,direct
=
2
PNP
C
+N
0
ln (2)
_
1
H
d1
+
1
H
d2
_
PNP
C
+N
0
ln (2)
_
1
H
d3
_
=
2H
d1
H
d2
H
d3
P
NP
+N
0
CH
d3
ln (2) (H
d1
+ H
d2
)
H
d1
H
d2
H
d3
P
NP
+N
0
Cln(2) H
d1
H
d2
(86)
Whenever (86) evaluates to be less than one, then it is
advantageous from an energy perspective to use the relay.
Solving for the power-efciency factor of the direct link
needed to satisfy this condition, we nd (87) through (89).
H
d3
<
N
o
Cln(2) H
d1
H
d2
H
d1
H
d2
P
NP
+N
o
Cln(2) (H
d1
+ H
d2
)
(87)
If we model the channel gain according to equation (55), we
nd (90). As the value of k depends only on the path loss
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
14 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
G
RX,sink
G
channel,d3
H
source
<
N
o
Cln(2) H
d1
H
d2
H
d1
H
d2
P
NP
+N
o
Cln(2) (H
d1
+ H
d2
)
(88)
G
channel,d3
<
_
1
G
RX,sink
H
source
_
N
o
Cln(2) H
d1
H
d2
H
d1
H
d2
P
NP
+N
o
Cln(2) (H
d1
+ H
d2
)
(89)
k
d

3
<
_
1
G
RX,sink
H
source
_
N
o
Cln(2) H
d1
H
d2
H
d1
H
d2
P
NP
+N
o
Cln(2) (H
d1
+ H
d2
)
(90)
or
d

3
>G
RX,sink
H
source
H
d1
H
d2
P
NP
+N
o
Cln(2) (H
d1
+H
d2
)
N
o
Cln(2) H
d1
H
d2
Fig. 10. The basic consumption factor analysis determines when it is
advantageous to use a relay.
value at a reference distance from the transmitter, we shall
here let k = 1, as it gives little insight by retaining it in the
equations.
= G
RX,sink
H
source
_
P
NP
N
o
Cln(2)
+
1
H
d1
+
1
H
d2
_
(91)
See also (92) next page. To gain intuition, assume that the
non-path power is zero, in which case:
d

3
>
_
G
RX,sink
G
RX,relay
_
d

1
+
_
H
source
H
relay
_
d

2
(93)
From (93), the link over which the relay is receiving (i.e. d
1
)
is scaled according to the ratio of the overall gains of the sink
and the relay receivers (gain here is the ratio of the power that
is retransmitted as for the relay - or processed/demodulated
as for the sink - to the power that is received from the
transmitter). Antenna gain is included in this gain because
we compute path loss assuming that the path loss has been
normalized for antenna gain. Therefore, if we have a relay
with a very low gain compared to the gain of the sink receiver,
the high output power required to communicate with the relay
may outweigh the benet derived from communicating over
a shorter distance. The second key point is that the link
over which the relay acts as a transmitter (i.e. d
2
) is scaled
according to the ratio of the power-efciency factors of the
source transmitter and relay transmitter. Therefore, if the relay
has a very low efciency transmitter, any savings derived from
the shorter link distance between the relay and the sink may
be outweighed by the loss incurred due to the inefciencies of
the relay. For the case of free-space channels, we may re-write
Fig. 11. From an energy perspective, it is advantageous to use a relay
provided the relay link distances are contained within the ellipse dened by
equation (94). This assumes a free space path loss exponent of 2.
(93) as:
1 >
_
d1
d3
_
2
_
G
RX,relay
G
RX,sink
_ +
_
d2
d3
_
2
_
H
relay
Hsource
_ (94)
which is the equation for the interior of an ellipse, as illustrated
by Figure 11. It is advantageous to use the relay rather than
a direct path from source to sink if the distances d
1
, d
2
, and
d
3
are such that they are in the interior of the ellipse in Figure
11.
Figures 12 through 14 illustrate the use of equation (93) and
how the region in which it is advantageous to place a relay
changes in size as certain parameters are varied. In Figure 12,
see that this region increases in size as the relays transmitter
becomes more efcient. In Figure 13, the same impact occurs
as the relays gain increases. Figure 14 indicates that higher
path loss exponents result in a larger area in which it is
benecial to use the relay from an energy perspective. This
is because an increase in path loss incentivizes a means of
shortening link distances.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
MURDOCK and RAPPAPORT: CONSUMPTION FACTOR AND POWER-EFFICIENCY FACTOR: A THEORY FOR EVALUATING THE ENERGY EFFICIENCY... 15
d

3
>
_
G
RX,sink
H
source
P
NP
N
o
Cln(2)
+
_
G
RX,sink
G
RX,relay
_
d

1
+
_
H
source
H
relay
_
d

2
_
(92)
Fig. 12. The interior regions of each ellipse indicate where it is advantageous
to place a relay. This gure shows that as the relay becomes more efcient,
that it is advantageous to use the relay over a wider area.
Fig. 13. Impact of changing the relay gain. The areas inside the ellipses are
where it is advantageous to use a relay. We see that as relay gain increases,
the regions over which the relay is advantageous also increases.
VIII. CONCLUSIONS
We have presented a basic theory of consumption factor
(CF) and have used this framework to develop useful con-
cepts in designing future energy efcient wireless broadband
networks. Additionally, we have developed understanding of
how the efciency of a communication system impacts the
minimum required energy expenditure per bit. We demon-
strated the use of the consumption factor by showing when
it is advantageous to use a relay in a multi-hop setting.
Our analysis indicates the importance of the receiver gain
and transmitter efciency for wireless communications. The
theory presented here allows engineers to compare cascades
of various components, and show quantitatively that when
the receiver gain is high, the transmitter may use lower
power, often resulting in a net energy savings. The continued
expansion of world-wide communication systems and the
exponential increase in data trafc necessitates reducing the
energy costs per bit. A key realization from the consumption
factor analysis is that, in order to align the goals of higher
data rates and lower energy expenditure per bit, it is necessary
to reduce the signal powers used in communication systems
to a point where ancillary power consumption (e.g. power
Fig. 14. Higher values of path loss result in a larger area where it is
advantageous to use a relay. The regions inside the ellipses are where it is
advantageous to use a relay.
consumed by oscillators and cooling equipment) is higher than
on-path signal power. Ideally, such ancillary forms of power
consumption will be decreased rapidly, but on-path signal
powers should be decreased even more quickly. To achieve this
goal for wireless systems, very short link distances, such as
those in a femtocells, become advantageous, or alternatively,
much more efcient RF power ampliers if longer distances
are to be used. Our theory shows that shorter link distances
combined with massive bandwidths (e.g. at millimeter-wave
carrier frequencies) and highly directional antennas will enable
unprecedented data rates and lower energy consumption per
bit. This in turn enables continued exponential growth in total
data trafc while mitigating the dramatic increase in energy
consumption. In fact, as future mm-wave wireless systems
evolve using untapped spectrum above 5GHz [11][4][12], the
power consumption factor theory presented here may give
insight into proper beam forming and minimum power cong-
urations for future wireless devices that use high gain adaptive
antennas that sense from where most multipath energy arrives.
ACKNOWLEDGMENTS
The authors would like to acknowledge F. Gutierrez, E. Ben-
Dor, Y. Qiao, J. I. Tamir, and K. Shabaik for their useful
thoughts, discussions, and careful reviews.
REFERENCES
[1] R. Bolla, R. Bruschi, F. Davoli, and F. Cucchietti, Energy Efciency
in the Future Internet: A Survey of Existing Approaches and Trends in
Energy-Aware Fixed Network Infrastructures, IEEE Communications
Surveys and Tutorials, vol. 13, no. 2, Feb. 2011.
[2] J. Baliga, R. Ayre, K. Hinton, and R. S. Tucker, Photonic Switching
and the Energy Bottleneck, in Photonics in Switching, Aug. 2007, pp.
125-126.
[3] K. Hyuck and T. Birdsall, Channel Capacity in bits per joule, IEEE
Journal of Oceanic Engineering, vol. 11, no. 1, pp. 97-99, Jan. 1986.
[4] V. Rodoplu, T. Meng, Minimum energy mobil wireless networks,
IEEE Journal on Selected Areas in Communications, vol. 17, no. 8,
pp. 1333-1344, Aug. 1999.
[5] C. Bae and W. E. Stark, Minimum Energy-Per-Bit Multi-hop Wireless
Networks, in IEEE 46th Allerton Conference, Sept. 2008.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
16 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 12, DECEMBER 2014
[6] W. Wang, Energy-efcient bandwidth allocation in wireless networks:
algorithms, analysis, and simulations, IEEE Trans. on Wireless Com-
munications, vol. 5, pp. 1103-1114, May 2006.
[7] A. J. Goldsmith and S. B. Wicker, Design Challenges for Energy-
Constrained Ad Hoc Wireless Networks, IEEE Wireless Communica-
tions Magazine, pp. 1-22, Aug. 2002.
[8] J. Wu, S. Rangan, and H. Zhang, Green Communications. CRC Press,
2012.
[9] T. S. Rappaport and J. N. Murdock, Power Efciency and Consumption
Factor Analysis for Broadband Millimeter-Wave Cellular Networks, in
IEEE Global Communications Conference, Anaheim, California, 2012.
[10] T. S. Rappaport et al., Broadband Millimeter Wave Propagation Mea-
surements and Models using Adaptive Beam Antennas for Outdoor
Urban Cellular Communications, IEEE Trans. of Ant. Prop. , Apr. 2013.
[11] S. Rajagopal, S. Abu-Surra, Z. Pi, and F. Khan, Antenna Array Design
for Multi-Gbps mmWave Mobile Broadband Communications, in IEEE
Global Communications Conference (GLOBECOM), Houston, 2011.
[12] T. S. Rappaport, J. N. Murdock, and F. Gutierrez, State of the Art in
60 GHz Integrated Circuits and Systems for Wireless Communications,
Proceedings of the IEEE, vol. 99, no. 8, pp. 1390-1436, Aug. 2011.
[13] J. N. Murdock and T. S. Rappaport, Consumption factor: A gure
of merit for power consumption and energy efciency in broadband
wireless communications, in Proceedings of 2011 IEEE GLOBECOM
Workshops, Dec. 2011, pp. 1393-1398.
[14] D. Meisner and T. F. Wenisch, Does low-power design imply energy
efciency for data centers?, in IEEE International Syposium on Low
Power Electronics and Design (ISLPED), Aug. 2011.
[15] Mini-Circuits (2012, Apr.) MAX2265 Data Sheet. [Online]. http://
para.maxim-ic.com/en/search.mvp?fam=rf pwr amp&tree=wireless
[16] Mini-Circuits. (2012, Apr.) ADEX-10L Data Sheet. [Online].
http://www.minicircuits.com/products/fm sm level 3.shtml
[17] Mini-Circuits. (2012, Apr.) ERA-1+and ERA-4+Data Sheets. [Online].
http://www.minicircuits.com/products/ampliers smt gpw.shtml
[18] Mini-Circuits (2012, Apr. ) MAX2643 Data Sheet. [Online].
http://www.maxim-ic.com/datasheet/index.mvp/id/2224
[19] T. S. Rappaport, et. al., Cellular broadband millimeter wave propaga-
tion and angle of arrival for adaptive beam steering systems, in IEEE
Radio and Wireless Symposium (RWS), Santa Clara, CA, 2012, pp. 151-
154.
[20] Y. Azar et. al., 28 GHz Propagation Measurements for Outdoor Cellular
Communications using Steerable beam Antennas in New York City, in
2013 IEEE International Conference on Communications (ICC 2013),
to appear June 2013.
[21] J. C. Liberti and T. S. Rappaport, Analysis of CDMA cellular radio
systems empolying adaptive antennas in mulitpath environments, in
IEEE Vehicular Technology Conference, April,1996, pp. 10760-1080.
[22] G. Durgin and T. S. Rappaport, Basic relationship between multipath
angular spread and narrowband fading in wireless channels, IET
Electronics Letters, vol. 10, no. 25, pp. 2431-2432, Dec. 1998.
[23] G. Durgin and T. S. Rappaport, Effects of multipath angular spread
on the spatial cross-correlation of received voltage envelopes, in IEEE
Vehicular Technology Conference, 1999, pp. 996-1000.
[24] L. W. Couch, Digital and Analog Communication Systms, 7th ed. .
Prentice Hall, 2007.
Theodore (Ted) S. Rappaport (Fellow, IEEE) received the B.S., M.S.,
and Ph.D. degrees in electrical engineering from Purdue University, West
Lafayette, IN, in 1982, 1984, and 1987, respectively, and is an Outstanding
Electrical Engineering Alumnus from his alma mater. Currently, he holds
the David Lee/Ernst Weber Chair in Electrical Engineering at NYU-Poly,
and is also center director of the NSF I/UCRC WICAT center at NYU-
Poly, and Professor of Computer Science and Professor of Radiology at New
York University. Earlier in his career, he founded the Wireless Networking
and Communications Group (WNCG) at the University of Texas at Austin
(UT), where he also was founding NSF I/UCRC site director for the Wireless
Internet Center for Advanced Technology (WICAT). Prior to UT, he was on
the electrical and computer engineering faculty of Virginia Tech where he
founded the Mobile and Portable Radio Research Group (MPRG), one of the
worlds rst university research and teaching centers dedicated to the wireless
communications eld. In 1989, he founded TSR Technologies, Inc., a cellular
radio/PCS software radio manufacturer that he sold in 1993 to what is now
CommScope, Inc. In 1995, he founded Wireless Valley Communications Inc.,
a site-specic wireless network design and management rm that he sold in
2005 to Motorola, Inc. (NYSE: MOT). Rappaport has testied before the U.S.
Congress, has served as an international consultant for the ITU, has consulted
for over 30 major telecommunications rms, and works on many national
committees pertaining to communications research and technology policy. He
is a highly sought-after consultant and technical expert, and serves on various
boards of several high-tech companies. He has authored or coauthored over
200 technical papers, over 100 U.S. and international patents, and several
prize papers and bestselling books. In 2006, he was elected to serve on the
Board of Governors of the IEEE communications Society (ComSoc), and was
elected to the Board of Governors of the IEEE Vehicular Technology Society
(VTS) in 2008 and 2011.
James N. Murdock (Member, IEEE) received the B.S.E.E. degree from
the University of Texas at Austin, Austin, in 2008, where he specialized
in communication systems and signal processing. He obtained a Master of
Engineering degree from The University of Texas at Austin in 2011, for
which he focused on sub-THz and electromagnetic engineering, in addition
to channel modeling and scientic data archiving. Currently, he is working at
Texas Instruments, where he focuses on low power radio systems and sub-THz
radar applications. James has co-authored over ten conference and technical
magazine papers and two journal papers.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

S-ar putea să vă placă și