![PAPR (Watts/Watts) and bandwidth efficiency (bps/Hz) for a progression of uplink signal modulation types, in linear scale. [9]](https://www.researchgate.net/profile/Rida-Khan-6/publication/331416182/figure/fig2/AS:794327664037911@1566393881983/PAPR-Watts-Watts-and-bandwidth-efficiency-bps-Hz-for-a-progression-of-uplink-signal_Q320.jpg)
![Another view of (4) showing the impact of signal PAPR on the energy efficiency of a linear amplifier for four common signals used in cellular networks. [9]](https://www.researchgate.net/profile/Rida-Khan-6/publication/331416182/figure/fig3/AS:794327668248576@1566393882000/Another-view-of-4-showing-the-impact-of-signal-PAPR-on-the-energy-efficiency-of-a_Q320.jpg)
Content uploaded by Rida Khan
Author content
All content in this area was uploaded by Rida Khan on Aug 21, 2019
Content may be subject to copyright.
TGCN-18-0136
1
Abstract—With the anticipated industry-wide adoption of
5G, it is a matter of urgency to quantify and optimize the
energy efficiency of 5G communication devices, and
particularly their wireless transmitters. This paper
contributes two major results to address these requirements.
First, the energy efficiency of linear transmitters is
quantitatively evaluated as a primary performance metric
when using any communication signal waveform, particularly
those transmitters used in 5G NR (New Radio), with Long
Term Evolution (LTE) as the baseline. A generic upper-bound
methodology called Modulation-Available Energy Efficiency,
MAEE, is applied. Second, the impact of multiple device
technologies is examined against this calculated upper bound
and it is shown that implementing technology choice has a
significant influence on actual transmitter efficiency results.
Index Terms—5G, 5G NR, energy efficiency, power
amplifier, modulation, signal waveforms.
I. INTRODUCTION
NDUSTRY research firms are predicting widespread
adoption of 5G communications with projections of 1.5
billion 5G connections by 2025 and service revenues of
almost $270 Billion [1]. Technologies such as network
functions virtualization, network slicing, enhanced self-
organizing and ultra-dense heterogeneous networks, as well
as low-power and low-throughput Internet of Things (IoT)
networks are expected to adopt 5G NR. Emerging
applications such as mobile high-resolution video traffic,
virtual/augmented reality, connected vehicles, real-time
automation, autonomous robots, smart surveillance and
immersive internet, cover diverse 5G use-cases such as
healthcare, intelligent transportation systems, real-time
gaming, factory automation, smart homes and smart cities.
The heterogeneous characteristics of the expected 5G
workloads are, on one hand, very high data rates, low
This paper was submitted for review on September 12, 2018.
S. R. Biyabani is the founder & CTO of GridComm startup in CA,
USA (e-mail: sara@gridcomm.net).
R. Khan is a graduate student at Tallinn University of Technology,
Tallinn, Estonia (e-mail: Rida.Khan@ttu.ee).
M. M. Alam is a professor in the Electronics Department, Tallinn
University of Technology, Tallinn, Estonia (e-mail:
muhammad.alam@ttu.ee).
A. A. Biyabani is with Hawaz LLC, Riyadh, Saudi Arabia and VTL
LLC, Pittsburgh, PA USA (e-mail: ahmed@virtualtrafficlights.com).
E. McCune is CTO of Eridan Communications and Chair of the Energy
Efficient Communications Hardware Working Group (P1923.1) (e-mail:
emc2@wirelessandhighspeed.com)
latency, and high energy density in cellular networks and,
on the other hand, unsynchronized access from massive
numbers of IoT devices [2], some tolerant of high latency
and low reliability and requiring long battery life, and yet
others, such as video monitoring systems, requiring high
data rates and low latency.
A brief survey of current literature shows that data rate
performance, translating to spectral efficiency of
waveforms, is the primary 5G performance metric that is
optimized and standardized without considering energy
efficiency. A few authors have taken a network or traffic
optimization perspective [3, 4] while spectral efficiency has
been identified as a metric of prime concern with a
comparison of several candidate modulation formats for 5G
[5]. With the new class of low power, highly distributed,
and energy-constrained devices emerging in 5G
applications, the maximization of energy efficiency
becomes a major requirement. The importance of this
energy efficiency is highlighted in [6] for 5G ultra dense
cellular networks by examining the impact of small cells
density over the backhaul energy efficiency. Moreover,
authors in [7] have developed an energy efficiency model
for cellular networks based on the spatial distributions of
traffic load and power consumption and have analyzed the
role of interference coordination for improving energy
efficiency. One rare study [8] has pointed out the lack of
energy efficiency considerations in analysis of waveforms
in 5G transmission waveforms and has defined “aggregate
energy efficiency” as the total throughput of all downlink
users divided by the total power consumption in the entire
wireless system. But the study does not take into account
the impact of inherent device technology used for
waveform transmission.
As of this writing, international standards bodies are in
the process of defining which wireless technologies will be
termed “5G”. Some candidates for 5G are: Quadrature
Amplitude Modulation (QAM) based Orthogonal
Frequency Division Multiplexing (OFDM), Generalized
Frequency Division Multiplexing (GFDM) and Filter Bank
Multicarrier (FBMC). Considering the communications
module that will implement any of these 5G waveforms,
power amplifiers (PAs) dominate the energy consumed by a
transmitter in any given type of wireless communication.
Delving further into transmitter PAs [9], it is increasingly
difficult to maintain the energy efficiency of the transmitter
PA due to the following reasons: a) efficiency is not
constant over output power, b) high efficiency PAs are
nonlinear, and c) load variations negatively impact
SS5G: Energy Efficiency Evaluation of Linear
Transmitters for 5G NR Wireless Waveforms
Sara R. Biyabani, Senior Member, IEEE, Rida Khan, Student Member, IEEE,
Muhammad Mahtab Alam, Member, IEEE, A. Ahmed Biyabani, Earl McCune, Fellow, IEEE
I
Authors' version of the paper that appears in IEEE Transactions on Green Communications and Networking (Volume: 3 , Issue: 2 , June 2019). DOI: 10.1109/TGCN.2019.2902179. ©
2019 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this
material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in
other works.
TGCN-18-0136
2
efficiency. Various non-linear or time varying PA designs
exist that have good energy efficiency; however, for multi-
carrier signals, which constitute the potential 5G
waveforms, any circuit nonlinearity causes cross-
modulation distortions among the carriers. This results in
linear amplifier design still being adopted to avoid these
distortions. The traditional tradeoff between circuit linearity
vs. energy efficiency continues to apply.
The objective of this paper is not to offer an optimal
device or circuit design but instead to examine the
waveforms that are transmitted in 5G communications, and
to quantitatively evaluate their energy efficiency
characteristics with respect to the linear PAs used in the
wireless transmitter circuitry. Modulation-available energy
efficiency (MAEE) [10] is used as the primary metric to
evaluate, by MATLAB simulation, the energy efficiency of
several waveforms likely to be used in 5G networks: a)
QAM-based OFDM b) Offset QAM-based OFDM
(OQAM-OFDM) c) Discrete Fourier Transform Spread
OFDM (DFT-S-OFDM) d) Zero Tail Discrete Fourier
Transform Spread OFDM (ZT-DFT-S-OFDM) e) GFDM f)
FBMC [11]. MAEE is an energy efficiency metric for
individual power amplifier transmitters based on both
waveform and device technology characteristics. The peak
normalized envelope probability density function (PDF) of
each of the above waveforms is calculated as a
characteristic of that modulation and used in the evaluation
of the MAEE. The peak to average power ratio (PAPR) is
also calculated from the PDF and used as a comparison
value. MAEE is a generally applicable transmitter-level
metric, rather than the composite, system-specific aggregate
energy efficiency metric suggested in [8]. Additionally,
several device technologies for linear PAs are rated in this
paper, according to their energy efficiency performance
when using the above 5G candidate waveforms and the
corresponding results are presented.
The remainder of this paper is structured to discuss the
impact of factors that constitute the MAEE methodology,
namely, power dissipation, modulation and device
technology in Section II; details of the MAEE methodology
and its hardware validation in Section III; details of the 5G
waveforms evaluated, MAEE simulation results and
observations in Section IV and finally, conclusions in
Section V.
II. IMPACT OF POWER DISSIPATION, MODULATION, AND
DEVICE TECHNOLOGY
This section discusses three factors that impact energy
efficiency of 5G transmitters. First, the relationship
between the various power flows in a typical power
amplifier and power dissipation is examined in II-A, and
then the impact of modulation type and corresponding
PAPR is examined in II-B, and lastly, the impact of device
technology is discussed in II-C.
A. Power Dissipation
Energy efficiency in any circuit is lost through the
dissipation of power in the form of heat primarily in the
transistors. This fact forces any evaluation of energy
efficiency to be closely tied to actual hardware
implementations. Our objective is to abstract the hardware
details as much as possible so that general theoretical
constructs for maximizing energy efficiency are possible
and practical. To start, Figure 1 shows the power flows in
any circuit, including modern power amplifiers.
Fig. 1. Power flows typical in a power amplifier. Loss of
efficiency is represented by the power dissipation PD. [10]
Even though the conservation represented in Fig.1 shows
a conservation of power (watts), this is directly equivalent
to the physical Principle of Conservation of Energy (joules)
if all of the time units in the power measurements are
identical [9, 12], which usually means measuring all powers
as root-mean-square (rms) power values. Therefore, the
hardware energy efficiency (
η
) relationship can be stated
as:
(1)
where the final approximation is valid for high amplifier
gain; when POUT >> PIN. This shows that energy efficiency
is inversely proportional to the power dissipation. The goal
then is to design for minimum power dissipation, which
then maximizes energy efficiency.
For maximum energy efficiency, (1) shows that the
power dissipation must be driven close to zero. The PA
circuitry then converts most of the applied DC power into
RF signal output power. It is useful to consider the required
sizes of the power supply (providing PDC) and the heatsink
(absorbing PD) normalized to the transmitter output power.
From (1) these two relationships are:
11
D
OUT
P
P
η
= −
(2)
1
1
DC D
OUT OUT
PP
PP
η
= +=
(3)
with respect to the circuit energy efficiency
η
. These
relationships are presented in the curves shown in Fig. 2,
which represent the necessary sizing of the transmitter
power supply and heat sink, corresponding to a PA that
provides a set output signal power. Normalizing to the set
output signal power, as the circuit efficiency falls then both
the size of the power supply (3) and the size of the heatsink
(2) begin to grow rapidly due to the reciprocal relationships.
TGCN-18-0136
3
Fig. 2. Circuit energy efficiency impacts on the relative
sizes of the associated power supply and heatsink attached
to a linear amplifier.[10]
Power supply size and heatsink size both directly
impact implementation cost, which must always be
minimized for market competitiveness. The graph shows
that in order to meet this economic objective it is necessary
to achieve higher values of circuit efficiency. A minimum
acceptable value (lower bound) for circuit efficiency is 40%
[10]. Efficiency values that exceed 70% (upper bound)
provide diminishing returns, particularly due to asymptotic
power supply cost, and therefore are not economically
valuable.
B. Modulation Type and corresponding PAPR
It is critical that the signal PA must be capable of
generating the signal peak power, as shown in Fig. 3. The
design of the signal modulation therefore has a direct
impact on what is achievable for the PA efficiency. The
dashed hyperbolic curves overlaying the transistor
characteristic curves (here for a Gallium-arsenide
heterojunction bipolar transistor (GaAs HBT)) are contours
of constant power dissipation at the transistor. Achieving
high efficiency, eqn. (1) requires that the signal operation
intersect only the curves of low PD value. These preferred
curves are close to the axes. The power dissipation
contours maximize their value near the center of the load
line. According to (1), these high PD values inherently
minimize PA efficiency. These results follow directly from
Conservation of Energy and Ohm’s Law (which is
contained within Maxwell’s Equations [13]), are
consequences of physics, and therefore cannot be
manipulated by clever design. This is precisely the origin
of the well-known tradeoff between energy efficiency and
PA linearity. In other words, any linear PA can never have
both excellent circuit linearity and high energy efficiency.
Fig. 3. Plot of transistor characteristic curve, IC (A) vs.
VCE(V), (Icollector vs. Vcollector-emitter), together with the load
line. Signal Peak-to-peak waveform must fit within the
amplifier linear range (shown for GaAs HBT). [10]
Linear PA operation requires the transistor to operate as a
controlled current source, forcing all operation to exceed
the transistor knee voltage. The entire output signal,
especially the signal power peaks, must always remain
within the boundaries of linear operation between transistor
compression and transistor cut-off. The example waveform
above is an OFDM variant used for LTE uplink operation.
As seen in Fig. 3, the entire signal peak-to-peak
waveform must fit within the amplifier linear range,
emphasizing that any amplifier is a peak-power limited
device. Yet, it is the average signal power that sets the
communication distance of any signal. This makes the
signal peak to average power ratio (PAPR) an extremely
important parameter. It is known from [14] that the
minimum theoretical impact on PA energy efficiency due to
signal PAPR (in dB) is:
()
20
010PAPR
PAPR
ηη
=
(4)
which is plotted in Fig. 4 for the theoretically ideal linear
class-A amplifier, where
η
0 = 50%. Therefore the
efficiency cost, and ultimately the actual monetary cost
(Fig. 2), is strongly impacted by signal PAPR. The analysis
in [14] does not provide any information into the
assumptions used to provide this simple mathematical
result. Because the transistor power dissipation is a
nonlinear characteristic of signal envelope, the actual power
dissipation strongly depends on the probability density
function (PDF) of the actual signal envelope. This paper
takes this PDF dependence into account, improving on the
earlier analysis.
0
1
2
3
4
5
6
7
8
9
10
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Power / Output power (Normalized)
Circuit Energy Efficiency
Input Power
Power Dissipation
Power supply size
TX power
Heatsink
size
0
0.005
0.01
0.015
0.02
0.025
0.03
00.5 11.5 22.5 33.5
IC(A)
VCE (V)
LTE signal
TGCN-18-0136
4
Fig. 4. Impact of signal PAPR on the maximum available
linear PA energy efficiency (4) for a theoretically perfect
linear amplifier.
Ongoing standardizations have adopted signals with
increasing peak-to-average power ratio (PAPR). The
bandwidth efficiency of these signals does not scale well
with the PAPR [9]; in fact, for several standardized signals,
the bandwidth efficiency decreases while the PAPR
increases, as shown in Figure 5. This decrease is
particularly noticeable for the 3G signal used in the
universal mobile telephone service (UMTS) where the
spread spectrum chip code for code division multiple access
(CDMA) operation expands the signal bandwidth without
any improvement in the information data rate.
Fig. 5. PAPR (Watts/Watts) and bandwidth efficiency
(bps/Hz) for a progression of uplink signal modulation
types, in linear scale. [9]
Subsequently, as the PAPR of the signal modulation used
increases, the average power available from the amplifier,
and the energy efficiency of the same amplifier, decrease as
shown in Figure 6.
Fig. 6. Another view of (4) showing the impact of signal
PAPR on the energy efficiency of a linear amplifier for four
common signals used in cellular networks. [9]
C. Device Technology
The relatively low efficiency anticipated in the PAs
necessitates higher transmitted power and hence causes
devices to operate at higher temperatures. Apart from
tackling heat dissipation, systems designers must also
consider smarter power control algorithms [15] or alternate
circuit topologies such as phased array antennae and beam
forming [16] for directional transmission, which are more
economical in overall power usage.
In terms of transistor level design, many PA developers
use materials such as Gallium arsenide (GaAs) / Gallium
nitride) GaN [17,18] or Complementary metal oxide
semiconductor (CMOS) Silicon-on-insulator (SOI) / bulk
CMOS [19]. While GaAs technologies tend to be more
power efficient, they may have additional costs of
calibration when implementing phased arrays of PAs;
whereas CMOS PAs tend to have higher yields, lower
manufacturing costs and offer greater integration with other
system components [20]. Some designers use Silicon
germanium (SiGe) BiCMOS (integration of bipolar junction
transistor and CMOS transistor) [21] to achieve greater
efficiency and reliability.
Textbook theoretical amplifier performance is based on a
transistor that behaves as a controlled current source even
when there is no voltage across it. Such a transistor has the
characteristic curves of Fig. 7a, and the knee voltage (dash-
dot line) is zero. Such a transistor can never exist, so a
practical yet ideal transistor is shown in Fig. 7b where a
voltage is required across the transistor for it to operate as a
controlled current source, though the transition from ON
resistance to current source operation is instantaneous. The
knee voltage in this case is non-zero and is dominated by
the ON resistance. Measured data from actual transistors
are shown in Fig, 7c (for GaAs HBT) and Fig. 7d (for
Silicon CMOS). Knee voltages for the actual transistors are
higher than for the theoretical constructs, and it is vital to
take into account this actual physical behavior in order to
accurately predict the amplifier performance.
0
10
20
30
40
50
60
012345678910 11 12 13 14 15
Maximum Linear Efficiency
(%)
PAPR (dB)
0
5
10
15
20
25
GSM NADC EDGE UMTS HSPA LTE OFDM
PAP R &
BW Efficiency
Signal Modulation Type
PAPR power ratio (Watts/Watts)
Bandwidth Efficiency (bps/Hz)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0.0 0.5 1.0 1.5 2.0
Norma lized O utput (V);
Efficiency Factor
Input Voltage ( (V) normalized to P1dB)
Voltage
clipping
boundary
GSM
UMTS
HSPA
LTE
Amplifier
Response
Output
Efficiency
TGCN-18-0136
5
a b
c d
Fig. 7. Linear amplifier operation for four transistor types:
a) theoretical transistor (completely impractical), b) ideal
practical transistor, and measured values for c) GaAs HBT
and d) Silicon CMOS transistors.
According to [10], the theoretically important ratio for
efficiency prediction is the ratio of the amplifier transistor
knee voltage (Vk) to the amplifier supply voltage (Vs), Vk/Vs.
Table I provides typical Vk/Vs ratios for theoretical, ideal
(idealXSTR) and actual transistor types used in our
evaluation: GaAs metal-semiconductor field-effect
transistor (MESFET), GaAs HBT, Silicon (Si) CMOS
Cascode (two-stage amplifier consisting of a common-
emitter stage feeding into a common-base stage), and Si
MESFET.
TABLE I. Transistor technologies and corresponding knee
voltage to supply voltage ratios, Vk/Vs .
Transistor technology Vk/Vs ratio
Theoretical 0
Practical Ideal, idealXSTR
0.12
GaAs HBT 0.17
GaAs MESFET 0.31
Si CMOS 0.43
Si MESFET 0.50
III. METHODOLOGY
This section describes the methodology for obtaining an
upper bound on energy efficiency of linear PA based on the
waveform and the device technology. The MAEE metric is
described in III-A based on factors discussed in the
previous section, and the hardware validation of MAEE
predictions is presented in III-B.
A. Modulation-Available Energy Efficiency (MAEE)
Using the MAEE metric introduced in [10], results are
derived using MATLAB simulations for a number of
candidate waveforms for 5G communications, with LTE as
the baseline to compare with 5G.
MAEE is different from the metric in [8] which defines
an “aggregate energy efficiency” as the total throughput of
all downlink users divided by the total power consumption
in the entire wireless system. In contrast, MAEE is an
energy efficiency metric for individual power amplifier
transmitters based on both waveform and device
technology. MAEE is a generally applicable transmitter-
level metric, rather than the composite, system-specific
aggregate energy efficiency metric. MAEE results directly
apply to the PA efficiency parameter α (alpha) in [8].
MAEE is based upon two known, general characteristics
of any signal and of any amplifier: 1) the signal envelope
voltage probability density function (PDF) normalized to a
peak value of unity, and 2) the inverted-parabolic circuit
power dissipation characteristic of a linear amplifier
transistor with respect to the actual output waveform. It is
also important to take into account the first-order effect of
the transistor knee voltage, Vk. Given these characteristics,
the steps for evaluating MAEE are [10]:
1. Identify the envelope PDF of the proposed
modulation, and convert this to a two-sided PDF to
match with the PA peak-to-peak output signal
linearity requirement.
2. Identify the knee voltage to supply voltage ratio,
Vk/Vs, for the particular PA transistor technology and
amplifier design.
3. Calculate the transistor power dissipation profile
(pdiss_profile) for linear amplifier operation above
the knee voltage.
4. Perform an expectation calculation (pdiss) of
amplifier power dissipation1 between the two-sided
envelope PDF and the transistor power dissipation2
both scaled to fill the linear amplifier operating
region above the amplifier knee voltage.
5. Calculate MAEE using the expected power
dissipation and the PA output power by following
these steps:
a. First, normalize the expected power dissipation,
pdiss, from step 4 to obtain norm_pdiss,
b. Then, calculate the signal power (sigpwr) as a
function of signal RMS value (sigRMS) and the
maximum signal amplitude (Amax),
c. MAEE = sigpwr / (sigpwr + norm_pdiss).
[22] provides more details and the pseudo-code for
1 This is the power dissipation in the amplifier as the output sinusoidal
signal waveform magnitude varies.
2 This is the power dissipation in the transistor as its output voltage
varies (independent of waveform).
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
00.5 11.5 22.5 33.5
I_Drain (A)
V_DS (V)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
00.5 11.5 22.5 33.5
I_Drain (A)
V_DS (V)
0
0.005
0.01
0.015
0.02
0.025
0.03
00.5 11.5 22.5 33.5
IC(A)
VCE (V)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
012345
I_Drain (A)
V_DS (V)
TGCN-18-0136
6
MAEE derivation.
MAEE values are an upper bound, in that the calculation
assumes that the transistor gain is constant and uniform
between the device threshold voltage and the device knee
voltage. This assumption is not true for any real transistor.
Therefore the actual amplifier performance is expected to
be always less than the performance predicted by the
MAEE calculation. The MAEE calculation is a useful
comparison among different modulation options to identify
which one can provide the highest energy efficiency when
using linear amplifiers.
B. Hardware Validation of MAEE Predictions
The final validation test for the MAEE method is to
review reported hardware performance values and compare
them to the corresponding MAEE predictions while
keeping in mind that the MAEE assumptions provide upper
bounds or ceilings on realizable performance. Two
waveforms of interest are listed in Table II. As can be seen,
reported hardware performance is indeed close to, and yet
below, the calculated MAEE value. The closeness of the
hardware performance to the MAEE value is indicative of
the quality of the hardware design. Both of these designs
are of very high quality since their energy efficiency is
close to the upper bound predicted by MAEE.
TABLE II. Comparisons of reported hardware performance
with MAEE upper bound predictions.
Waveform
Reference
Reported
Efficiency
MAEE
LTE PA
[23] 2016
9.5%
10.1%
OFDM for
Wireless
Local Area
Network
(WLAN) PA
[24] 2016 7.5% 7.7%
IV. SIMULATIONS
This section describes the candidate 5G waveforms that
are investigated using the MAEE metric, simulation results,
observations and discussion of both MAEE and PAPR
results as well as a comparison of transistor technologies. It
starts with a description of 5G waveforms and their
simulation setup in IV-A, then a discussion of MAEE
simulation results as well as PAPR results in IV-B for the
ideal theoretical transistor, and concludes with a
comparison of MAEE simulation results for four different
device technologies and the baseline practical ideal
transistor in IV-C.
The MATLAB simulation and modeling tool is used to
encode the MAEE algorithm described in Section III-A,
and also to set up the 5G waveforms as described in Section
IV-A below.
A. 5G Waveforms
The six candidate 5G waveforms evaluated for their energy
efficiency upper bounds are:
1) OFDM,
2) Offset QAM-based OFDM (OQAM-OFDM),
3) Discrete Fourier transform spread OFDM (DFT-S-
OFDM),
4) Zero tail discrete Fourier transform spread OFDM
(ZT-DFT-S-OFDM),
5) GFDM,
6) FBMC.
A general tutorial on each of the waveforms is outside
the scope of this paper; however, references are provided
where applicable. Details of how each waveform is
generated for simulation are described next.
The first waveform is a 52 subcarrier OFDM with no
cyclic prefix (CP) and 4QAM modulated random data on
each subcarrier. The second waveform is a 1024 subcarrier
OQAM-OFDM and is generated in MATLAB following a
procedure similar to the one described in [25]. The
procedure is as follows:
At first, two bipolar four-amplitude shift keying (4-ASK)
data are generated. For the in-phase component, the phases
of even subcarriers are 0 while the phases of odd
subcarriers are π/2. For the quadrature component, the
phases of even subcarriers are 3π/2 and those of odd
subcarriers are π. An inverse fast Fourier transform (IFFT)
is applied over the in-phase component and the delayed
quadrature component to produce the time-domain samples.
The generated symbols are then passed through the squared
root raised cosine filter with a roll-off factor of 0.5 for pulse
shaping, followed by parallel-to-serial conversion.
The third waveform, DFT-S-OFDM, is generated in
MATLAB following the steps mentioned in [26], using the
localized FDMA approach. This DFT-S-OFDM waveform
contains 16QAM randomly generated symbols, 256-point
DFT for spreading, 512 subcarriers and one-fourth CP
length. The fourth waveform, ZT-DFT-S-OFDM, shares
similar parameters, except that the CP is replaced by the
addition of zero head and tail before performing the 256-
point DFT spreading [27].
The last two waveforms, GFDM and FBMC, were
provided by National Instruments using their waveform
generators.
The two-sided PDFs of all selected 5G potential
waveforms are illustrated in Fig. 8.
TGCN-18-0136
7
Fig. 8. Two-sided probability density functions for six
proposed 5G waveforms.
B. MAEE and PAPR Results and Discussion
The 5G waveforms are evaluated using the MAEE
algorithm under the ideal knee voltage to source voltage
ratio, Vk/Vs, conditions, to give the upper bound on their
energy efficiencies for the theoretical ideal transistor, with
Vk/Vs = 0. The MAEE simulation results for the 5G
waveforms, and the corresponding PAPR values are
summarized in Table III. It is observed that for all
waveforms, an increase in PAPR leads to a reduction in
MAEE.
TABLE III. Candidate 5G waveforms and corresponding
MAEE and PAPR for theoretical transistor.
Waveform
MAEE (%)
from
Simulation
PAPR (dB)
OFDM
11.41
12.06
OQAM based OFDM
11.92
11.66
DFT-Spread -OFDM
16.01
8.93
Zero Tail DFT-Spread-
OFDM
14.92
9.59
GFDM
13.28
10.66
FBMC
12.69
11.07
The data in Table III is compared against the mathematical
model (eqn. 2) to determine what differences, if any, appear
when the actual signal envelope PDF is taken into account
in simulations. The comparison is shown in Fig. 9, where
the MAEE simulated values (X) are all close to, and yet
below, the mathematical model line. The difference
between the MAEE values and the model curve is
essentially constant, validating the model. Results from the
model are slightly optimistic since both the mathematical
model and the MAEE simulated calculations are based on
the theoretical transistor of Fig. 7a. Hence, the results are an
upper bound on the performance of actual hardware that
can be built and used.
Fig. 9. Comparison of MAEE simulated values for the
signals in Table III (X markers) to the efficiency model for
PAPR variation from (2). The model (2) , which does not
explicitly consider signal envelope PDF details (as MAEE
does) is slightly optimistic, and provides the correct trend.
In summary, the 5G waveforms in order of maximum
energy efficiency, from the highest to the lowest are:
1) DFT-S(pread)-OFDM,
2) Zero Tail DFT-S(pread)-OFDM,
3) GFDM,
4) FBMC,
5) OQAM-based OFDM, and
6) OFDM.
C. Comparison of Different Transistor Device
Technologies
Fig. 10 summarizes the energy efficiency ceilings for the
simulated 5G candidate waveforms, with each waveform
simulated using four different transistor technologies: GaAs
HBT, GaAs MESFET, Si CMOS Cascode, and Si
MESFET. The ideal practical transistor, idealXSTR, (Vk/Vs
= 0.12) is also included for reference. It can be seen that
GaAs outperforms the Silicon technologies, validating the
market preference for GaAs over Silicon technologies.
Fig. 10. Waveforms energy efficiency (in %) for different
transistor technologies.
Waveforms with higher PAPR (Table III) again show
lower energy efficiency values. The results also illustrate an
inverse relationship between the transistor’s knee voltage
-1 -0.5 0 0.5 1
0
0.5
1
1.5
OFDM PDF
-1 -0.5 0 0.5 1
0
1
2
3
OQAM-OFDM PDF
-1 -0.5 0 0.5 1
0
0.5
1
1.5
2
DFT-S-OFDM PDF
-1 -0.5 0 0.5 1
0
0.5
1
1.5
2
2.5
ZT-DFT-S-OFDM PDF
-1 -0.5 0 0.5 1
0
0.5
1
1.5
2
GFDM PDF
-1 -0.5 0 0.5 1
0
0.5
1
1.5
2
FBMC PDF
OFDM
OQAM based
OFDM
DFT-Spread -
OFDM
Zero Tail DFT-
Spread-OFDM GFDM
FBMC
0
10
20
30
40
50
60
012345678910 11 12 13 14 15
Maximum Linear Efficiency
(%)
PAPR (dB)
idealXSTR GaAs HBT GaAs MESFET Si CMOS Si MESFET
OFDM 9.13 8.3 6.26 4.79 4.03
OQAM-OFDM 9.67 8.79 6.64 5.08 4.28
DFT-S-OFDM 12.86 11.71 8.87 6.8 5.73
ZT-DFT-S-OFDM 11.94 10.86 8.22 6.3 5.31
GFDM 10.65 9.69 7.32 5.61 4.72
FBMC 10.17 9.25 6.99 5.35 4.51
0
2
4
6
8
10
12
14
Energy Efficiency Ceiling (in %)
Waveform Energy Efficiency Ceiling (in %) for different
Transistor Technologies
TGCN-18-0136
8
profile and the energy efficiency ceilings. The higher
transistor knee voltage to source voltage ratio, Vk/Vs, leads
to lower energy efficiency values. (Refer to Table I for
Vk/Vs values). For example, the OFDM waveform (12.06 dB
PAPR) results in energy efficiency of 8.3% for GaAs HBT
transistor (Vk/Vs = 0.17) but the efficiency is reduced to
4.03% for Si MESFET transistor (Vk/Vs = 0.5). However,
for the DFT-S-OFDM waveform, which has a lower PAPR
(8.93 dB) than that of standard OFDM, the same transistor
types result in higher energy efficiencies of 11.71% and
5.73%, respectively.
The overall observations are:
• Device Technologies in order of energy efficiency
from the highest to the lowest are:
GaAs HBT, GaAs MESFET, Si CMOS Cascode, and
Si MESFET.
• Relative ranking of the simulated 5G waveforms within
each device technology is maintained. (Refer to IV-B).
• DFT-S-OFDM scores the highest for MAEE for all
technologies.
• OQAM-OFDM and OFDM score the lowest MAEE for
all implementing technologies.
V. CONCLUSION
Signal modulation design is a determining factor of
achievable energy efficiency in linear transmitters. Not
only is the signal PAPR important, but the actual envelope
probability density function also matters significantly. The
MAEE algorithm is designed to take these factors into
account, as well the impact of transistor knee voltage and
the resulting loss of efficiency. Using the MAEE tool
provides a realistic means to quantitatively compare
modulations for performance when implemented using
widespread linear amplifiers.
With respect to the 5G NR proposed waveforms, this
analysis shows that all of the waveforms provide very low
(<12%) linear PA efficiency. The best result is 11.71% for
DFT-S-OFDM implemented in GaAs HBT technology. The
low energy efficiency is a direct result of the envelope
statistics for each of the proposed waveforms. Because all
the 5G candidate signals are from the OFDM (multicarrier)
family, the presence of any circuit nonlinearity causes
cross-modulation among the signals’ multiple carriers. To
avoid the cross-modulation distortion the transmitter
circuitry must remain linear. Therefore, this MAEE
analysis is very applicable.
As future work, we plan to expand the concept and
methodology of MAEE to non-linear transmitter structures.
Additional devices will also be characterized as the
necessary data becomes available.
Because of the generality of the MAEE tool, it is under
active standardization effort as IEEE P1923.1 [22] and will
be widely available for future evaluation and comparison of
energy efficiency of proposed signal modulations.
ACKNOWLEDGMENT
Dr. Alam and Ms. Khan’s work has received funding
from the European Union’s Horizon 2020 research and
innovation program under grant agreement No. 668995.
This material reflects only the authors’ view and the EC
Research Executive Agency is not responsible for any use
that may be made of the information it contains.
REFERENCES
[1] J. P. Tomas, “Japan, Korea to account for 43% of 5G global
connections in 2019: study”, Aug. 21, 2018. Available:
https://www.rcrwireless.com/20180831/5g/japan-korea-account-43-
5g-global-connections-2019-study
[2] Y. Yifei, and Z. Longming. "Application Scenarios and Enabling
Technologies of 5G." China Communications 11, no. 11, pp. 69-79,
Nov. 2014.
[3] A. Zappone, L. Sanguinetti, G. Bacci, E. Jorswieck, M. Debbah.
"Energy-efficient power control: A look at 5G wireless
technologies." IEEE Transactions on Signal Processing 64, no. 7,
pp. 1668-1683, 2016.
[4] R. Cavalcante, S. Stanczak, M. Schubert, A. Eisenblaetter, U.
Tuerke. "Toward energy-efficient 5G wireless communications
technologies: Tools for decoupling the scaling of networks from the
growth of operating power." IEEE Signal Processing Magazine 31,
no. 6, pp. 24-34, 2014.
[5] P. Banelli, S. Buzzi, G. Colavolpe, A. Modenini, F. Rusek, and A.
Ugolini. "Modulation formats and waveforms for 5G networks: Who
will be the heir of OFDM?: An overview of alternative modulation
schemes for improved spectral efficiency." IEEE Signal Processing
Magazine 31, no. 6, pp. 80-93, 2014.
[6] X. Ge, S. Tu, G. Mao, C. X. Wang, and T. Han, “5G Ultra Dense
Cellular Networks,” IEEE Wireless Communications 23, no. 1, pp.
72-79, Feb. 2016.
[7] L. Xiang, X. Ge, C. X. Wang, F. Y. Li, and F. Reichert, “Energy
Efficiency Evaluation of Cellular Networks Based on Spatial
Distributions of Traffic Load and Power Consumption” IEEE
Transactions on Wireless Communications 12, no. 3, pp. 961-973,
Mar. 2013.
[8] S. Zhang, X. Xu, Y. Wu, L. Lu, Y. Chen, “A survey on 5G new
waveform: From energy efficiency aspects,” 2014 48th Asilomar
Conference on Signals, Systems and Computers pp. 1 939 – 943,
2014.
[9] E. McCune, Dynamic Power Supply Transmitters: Envelope
Tracking, Direct Polar, and Hybrid Combinations, Cambridge
University Press, 2015.
[10] E. McCune, “Linear Power Amplifier Efficiency Ceilings due to
Signal Modulation Type,” The 47th European Microwave
Conference, Oct. 2017.
[11] B. Farhang-Boroujeny, “OFDM Versus Filter Bank Multicarrier,”
IEEE Signal Process. Mag., vol. 28, no. 3, pp. 92 – 112, May 2011.
[12] E. McCune, D. Babić, R. Booth; D. Kirkpatrick, “Decade
Bandwidth Agile GaN Power Amplifier Exceeding 50%
Efficiency,” Proc. of MILCOM 2015, Tampa FL, pp. 541-546,
October 2015.
[13] S. Ramo, J. Whinnery, T. Van Duzer, Fields and Waves in
Communications Electronics, John Wiley & Sons, New York, 1965.
[14] S. Miller, R. O’Dea, “Peak Power and Bandwidth Efficient Linear
Modulation,” IEEE Trans. On Communications, Vol. 46, No. 12,
Dec. 1998.
[15] A. Mohammadian, M. Baghani, C. Tellambura, “Optimal power
allocation of GFDM secondary links with Power Amplifier
Nonlinearity and ACI,” IEEE Wireless Communications Letters,
Jul. 2018.
[16] A. Sayag, E. Cohen, "A 4 Element Phased Array Transmitter with
Efficiency Enhancement Using Beamforming for High-Bandwidth
WLAN Applications." 2018 IEEE/MTT-S International Microwave
Symposium-IMS, pp. 1199-1202, 2018.
[17] A. Alizadeh, A. Medi, "A broadband integrated class-J power
amplifier in GaAs pHEMT technology." IEEE Transactions on
Microwave Theory and Techniques 64.6, pp. 1822-1830, 2016.
[18] S. Jee, J. Lee, J. Son, S. Kim, C.H. Kim, J. Moon, B. Kim,
"Asymmetric broadband Doherty power amplifier using GaN
MMIC for femto-cell base-station." IEEE Transactions on
Microwave Theory and Techniques 63, no. 9, pp. 2802-2810, 2015.
[19] N. Rostomyan, J.A. Jayamon, P. Asbeck, "15 GHz 25 dBm
multigate-cell stacked CMOS power amplifier with 32% PAE and≥
30 dB gain for 5G applications." In Microwave Integrated Circuits
TGCN-18-0136
9
Conference (EuMIC), 2016 11th European, pp. 265-268. IEEE,
2016.
[20] S. Shakib, H-C. Park, J. Dunworth, V. Aparin, K. Entesari, "A
highly efficient and linear power amplifier for 28-GHz 5G phased
array radios in 28-nm CMOS." IEEE Journal of Solid-State Circuits
51, no. 12, pp. 3020-3036, 2016.
[21] S. Mortazavi, K-J. Koh. "A 38 GHz inverse class-F power amplifier
with 38.5% peak PAE, 16.5 dB gain, and 50 mW P sat in 0.13-µm
SiGe BiCMOS." In Radio Frequency Integrated Circuits
Symposium (RFIC), 2015 IEEE, pp. 211-214. IEEE, 2015.
[22] IEEE, “P1923.1 - Standard for computation of energy efficiency
upper bound for apparatus processing communication signal
waveforms”, [Online]. Available:
https://standards.ieee.org/project/1923_1.html
[23] Skyworks Inc., SKY66184-11 LTE power amplifier, data sheet
203259E, May 11, 2016.
[24] E. Schwartz, et.al, “A 20dBm Configurable Linear CMOS RF
Power Amplifier for Multi-Standard Transmitters,” Proc. of RFIC
2016, San Francisco, May 2016, paper RTU1D-3.
[25] J. Zhao, “DFT-based offset-QAM OFDM for optical
communications,” Opt. Express, vol. 22 no: 1, Jan. 2014.
[26] G. Berardinelli, K. I. Pedersen, T. B. Sorensen and P. Mogensen,
“Generalized DFT-Spread-OFDM as 5G Waveform,” IEEE
Communications Magazine, vol. 54 no: 11, Nov. 2016.
[27] G. Berardinelli, K. Pajukoski, E. Lahetkangas, R. Wichman, O.
Tirkkonen and P. Mogensen, “On the Potential of OFDM
Enhancements as 5G Waveforms,” IEEE 79th Vehicular
Technology Conference, May 2014.
Sara R. Biyabani is an IEEE Senior
Member and received a B.A. in
Physics and Computer Science from
Smith College, Northampton, MA,
USA and a M.S. in Electrical and
Computer Engineering from the
University of Massachusetts, Amherst,
USA.
She is a technologist with extensive industry experience
as a Computer and Performance Architect in the design,
modeling and optimization of microprocessors, ASICs,
interconnection fabrics, embedded and computer systems.
She holds a US patent for unified memory architecture
design optimized for graphics. She is currently the founder
and CTO of a startup involved in intelligently integrating
Distributed Renewable Energy Resources into Electric
Power Distribution Systems. Her technical interests include
computer architectures and platforms for compute- and
data-intensive, connected and secure applications with real-
time constraints.
Ms. Biyabani is a member of the IEEE Standards
Association and the IEEE Computer, Communications and
Power & Energy Societies. She has actively contributed to
the development of Interoperability Standards IEEE 2030-
2011, IEEE 2030.2, and IEEE 1547, and is currently the
Vice Chair of the IEEE Energy Efficient Communications
Hardware Working Group for Stds. P1923.1 and P1924.1.
Rida Khan is an IEEE Student
Member and received B.E degree in
Telecommunication Engineering from
Mehran University of Engineering and
Technology, Pakistan, and M.Sc.
degree in Electronics and
Communication Engineering from
Istanbul Technical University, Turkey,
respectively in 2013 and 2017. She is
currently pursuing Ph.D. degree in Information and
Communication Technology, at Tallinn University of
Technology, Estonia.
Ms. Khan is a student member of IEEE Standards
Association and is currently serving as the Secretary of the
Energy Efficient Communications Hardware Standards
Working Group. Her research interests include Wearable
Wireless Networks, Network Coding and Energy Efficient
Modulation techniques.
Muhammad Mahtab Alam is an
IEEE Member and received the M.Sc.
degree in electrical engineering from
Aalborg University, Denmark, in 2007,
and Ph.D. degree from the University
of Rennes1 (INRIA Research Center),
France, in 2013.
He conducted postdoctoral research
at Qatar Mobility Innovations Center
during 2014-2016. In 2016, he was elected as European
Research Area Chair in cognitive electronics project and
Associate Professor at Thomas Johann Seebeck Department
of Electronics, Tallinn University of Technology, Estonia.
In 2018, he obtained tenure professorship to chair Telia
Professorship under the cooperation framework between
Telia and Tallinn University of Technology. He has
authored and co-authored over 50 research publications.
His research interests include self-organized and self-
adaptive wireless sensor and body area networks specific to
energy efficient communication protocols and accurate
energy modeling, Internet-of-things, public safety and
critical networks, embedded systems, digital signal
processing, and software defined radio.
A. Ahmed Biyabani is a technologist
and academic. He holds a S.B. in EECS
from M.I.T., Cambridge, MA, USA and
a M.S. and Ph.D. in ECE from Carnegie
Mellon University, Pittsburgh, PA,
USA.
He previously taught computer
architecture and AI at AIMSIU, Riyadh,
Saudi Arabia and researched biosensors
as a visiting professor at EPFL, Lausanne, Switzerland and
IBEC, Barcelona, Spain. He has worked extensively in
analog and mixed-signal IC and systems design at Sharp,
Intel, JDS Uniphase, WJ Communications, Synaptics and
others.
He is currently doing product and business development for
Virtual Traffic Lights LLC, a Pittsburgh, PA, USA startup
focusing on autonomous and connected vehicles.
TGCN-18-0136
10
Earl McCune is an IEEE Fellow, and
received his BSEE/CS from UC
Berkeley, CA, USA MSEE from
Stanford, and Ph.D. from UC Davis,
CA, USA. His research interests include
RF circuits and systems including
modulation design, with an emphasis on
jointly maximizing throughput and
energy efficiency while also minimizing implementation
cost. He is a Silicon Valley serial entrepreneur, and has 93
issued US patents. He is an emeritus MTT Distinguished
Microwave Lecturer, a member of multiple IEEE
conference committees and serves as the Chair of the
Energy Efficient Communications Hardware Standards
Working Group.
