5G RF Front End Design NI-SEPT-2018-EBOOK-v3 PDF

E-BOOK
5G RF Front End Design

and Measurement
SEPTEMBER 2018
S P O N S O R E D B Y
Table of Contents
3

Introduction: 5G RF Front End Design and Measurement
Patrick Hindle
Microwave Journal Editor
4 Design of a Single Chip Front-End Module for 28 GHz 5G

Stuart Glynn, Robert Smith, Liam Devlin, Andy Dearn and Graham Pearson
Plextek RFI Ltd., U.K.
9

5G Fixed Wireless Access Array and RF Front-End Trade-Offs
Bror Peterson and David Schnaufer
Qorvo, Greensboro, N.C.
17 Basics of Power Amplifier and Front End Module

Measurements
National Instruments
43 5G New Radio: Introduction to the Physical Layer

National Instruments
2
Introduction
5G RF Front End Design and Measurement

With 3GPP release 15 complete, the initial 5G standards are in place and everyone
is off to the races to design RF front ends for sub 6 GHz and mmWave applications. CCS
Insight is estimating almost 60 million 5G connections worldwide in 2020 and believes
1 billion will be reached by mid-2023. But with so many semiconductor process options and
radio architectures available for these applications, what are the important design tradeoffs
and measurements needed to be successful. This eBook takes a look at a few design
examples and the tradeoffs done to optimize those designs. The eBook also covers critical
RF measurements needed to verify the design and how to set them up.
The first article is about the design of a single chip RF front end module for 28
GHz 5G applications. Designers at Plextek RFI in the UK go through the design tradeoffs
comparing simulations and measurements to achieve their design goals for this project.
Written by experts from Qorvo, the next article looks at the trade offs for mmWave beam
forming arrays and which technologies are best suited for various components to maximize
power and efficiency.
The third article is written by experts from National Instruments and covers the
basics of power amplifier and front end module measurements stepping through why
each important measurement is needed. It also covers how to setup the measurements
with practical examples. The final article covers the physical layer details of the 5G NR
specification. It includes waveforms, subcarrier and symbol information, reference signals,
MIMO and mmWave specifications.
This eBook provides designers with the basics of measurement and design
techniques being used to achieve high performance designs for 5G RF front ends. These
articles are written by industry experts from leading companies as educational pieces
with design examples and measurement methods to provide the latest information about
5G RF design.
Pat Hindle, Microwave Journal Editor
3
Design of a Single Chip
Front-End Module for 28 GHz 5G
Stuart Glynn, Robert Smith, Liam Devlin, Andy Dearn and Graham Pearson
Plextek RFI Ltd., U.K.
W
ith the roll-out of mmWave This article describes the design, realiza-
5G expected to commence tion and evaluation of an FEM MMIC for
soon, the research and de- the 28 GHz 5G band (27.5 to 28.35 GHz)
velopment activities of the which satisfies all of these requirements. The
industry’s key players are now well advanced part was developed by Plextek RFI and de-
and reached the point where custom com- signed on WIN Semiconductors’ PE-15 pro-
ponents have been specified, designed and cess which is a 4 V, 0.15 µm, enhancement
evaluated. An essential component required mode GaAs PHEMT process. It is realised in
to enable future mmWave 5G systems is the a compact, low-cost 5 mm x 5 mm plastic
front-end module (FEM) providing the final overmolded SMT compatible QFN package
stages of amplification in a transmitter and making it suitable for high volume, low cost
the earliest stages of amplification in a re- manufacture. It covers 27 to 29 GHz and so
ceiver together with a transmit/receive (Tx/ encompasses the full 28 GHz 5G band.
Rx) switch to allow time division duplex (TDD)
operation. The FEM must demonstrate high DESIGN GOALS
linearity in transmit mode and low noise fig- The design of the FEM’s transmit path
ure in receive mode. As mmWave 5G sys- focused on achieving high efficiency when
tems are likely to require user terminals con- operating at back-off to provide linear am-
taining multiple FEMs as part of a phased ar- plification, as will be required by 5G sys-
ray or switched antenna beam architecture, tems. A target power added efficiency (PAE)
they must also be highly efficient, compact of 6 percent at back-off was specified, with
and low cost. Ease of control and monitoring third-order intermodulation (IMD3) levels
is also highly desirable. below ‐35 dBc (around 7 dB backed-off from
WWW.MWJOURNAL.COM/ARTICLES/30052
4
PA_ON PAE and Delivered Power Contours
PA_Vg12 PA_Vg3 PA_Vd12 PA_Vd3 Vref Vdet
Zload_at_mPdel_vs_PAE
PA Enable 6.5 + j8.0
Circuit Power
Detector 1.0
Pdel_dBm_contours_at_1_Pavs_Rho
2.0
PAE_contours_at_1_Pavs_Rho
0.5
PA_RFin
SPDT 5.0
PA
mPdel_vs_PAE_Rho
Antenna 0.2
0.5 1.0 2.0
LNA –0.2
LNA_RFout
–5.0
–0.5
–2.0
LNA SPDT –1.0
Enable Circuit Control Circuit
Reference Z0 = 5
LNA_Vsense LNA_Vd LNA_Vg LNA_ON VD_SW Vctrl1 Vctrl2
m2
indep(m2) = 14.208
 Fig. 1 Block diagram of FEM MMIC for 28 GHz 5G. plot_vs(ThirdOrdIMD_I, Pload_dBm) = –39.245
the 1 dB compression point). The optimize PAE at this operating point Third-Order IMD (dBc)
–10
RF output power at the 1 dB com- the PA was biased in deep class AB.
–20
pression point (P1dB) was specified
at 20 dBm. For the receive path, a DESIGN TRADE-OFFS –30 m2
–40
noise figure of below 4 dB (includ- The design was initiated by run-
–50
ing switch losses) was required with ning device level simulations on
–60
very low current consumption—a candidate unit cell transistors. This
–70
target of 15 mA maximum from the work forms a solid foundation for 2 4 6 8 10 12 14 16 18 20 22 24
+4 V supply was specified. the subsequent detailed PA design Output Power, Both Tones (dBm)
A block diagram depicting the as it yields key information such as (a)
functionality of the FEM MMIC is device sizes, bias points, imped-

PAE and Delivered Power Contours
shown in Figure 1. The transmit sig- ance targets, required number of
nal path runs from left to right in the PA stages and drive ratios. Zload_at_mPdel_vs_PAE
9.5 + j4.5
top half of the diagram; the input is A significant part of this work
Pdel_dBm_contours_at_1_Pavs_Rho
at the pin labelled “PA_RFin.” The was aimed at identifying how to

PAE_contours_at_1_Pavs_Rho
1.0
2.0
signal is amplified by a three-stage maximize PAE at back-off. Gener-
0.5
mPdel_vs_PAE_Rho
PA and then routed to the antenna ally speaking, this is achieved by 5.0
via an RF power detector and a reducing the device quiescent bias 0.2
single pole double throw (SPDT) current density. However, the extent 0.5 1.0 2.0
switch. The on-chip directional to which this approach can be ad- –0.2
power detector allows monitoring opted is limited by the gain and lin-
of the transmitted RF output pow- earity, both of which degrade as the –0.5 –5.0
er and incorporates temperature current density is reduced. There is –2.0
compensation. The compensated a clear trade-off of PAE (at back-off) –1.0
detector output is given by the dif- with gain and linearity. Reference Z0 = 5
ference between the voltages “Vref” The main linearity metric of inter-
and “Vdet.” A fast switching enable est was the IMD3 level at back-off m2
circuit “PA Enable Circuit” is includ- which had to be below ‐35 dBc. It was indep(m2) = 15.398
plot_vs(ThirdOrdIMD_I, Pload_dBm) = –29.875
ed on-chip and is controlled by the found that the IMD3 performance at
(active low) logic signal “PA_ON.” reduced bias currents became par- Third-Order IMD (dBc)
This is used to rapidly power up and ticularly sensitive to the fundamen- –10
power down the PA when switching tal load condition, as illustrated in –20
m2
between Tx and Rx mode such that Figure 2. Figure 2a shows load-pull –30
it draws only 0.1 mA when not in simulation results for an 8 × 50 µm
–40
use, maximizing the overall system device biased toward deep class AB
–50
efficiency. at 75 mA/mm at 4 V and highlights
The PA will typically operate the optimum load for PAE at P1dB. It –60
2 4 6 8 10 12 14 16 18 20 22 24
backed-off from compression to also shows the corresponding simu- Output Power, Both Tones (dBm)
preserve modulation fidelity of the lated IMD3 performance at back-off (b)
transmitted signal. The design ap- for this load and indicates that there
proach was to optimize the perfor- is about 4 dB margin on the speci-  Fig. 2 Load for optimum PAE at
P1dB and corresponding IMD3 at back-
mance of the PA when operating 7 fication of ‐35 dBc. The simulated off (a) and load for optimum power at
dB back-off from P1dB. In order to PAE at the same back-off was about P1dB and corresponding IMD3 at back-
off (b).
5
15 percent which is for the device only and excludes any a fast switching enable circuit such that the LNA draws
output losses. Figure 2b shows similar information for the as little as 0.1 mA when not in use. A key part of the
load condition that is best for power at P1dB. The IMD3 LNA design process was to produce a design which had
performance at the same relative back-off is considerably low current consumption but good noise figure and ad-
worse, more than 5 dB outside the specification, PAE at equate linearity.
back-off was similar at around 15.7 percent. Selection of appropriate transistor sizes was an im-
Other points on the Smith Chart were also evaluated portant first step. Multiple short fingers were used to re-
for performance at P1dB and at back-off, but the load duce the gate resistance of the transistors and improve
condition, highlighted in Figure 2a, was found to be the noise figure. Series inductive feedback was added
the best overall and was selected for the output stage to both stages to shift the impedance required for opti-
design. This also offered scope for further bias current mum noise figure closer to that required for a conjugate
reduction, into deep class AB, to trade some IMD3 per- match and optimum gain.
formance for PAE performance while still maintaining The first stage of the LNA was optimized for noise fig-
adequate gain. A bias current of 52 mA/mm was ulti- ure but still had to produce enough gain to adequately
mately chosen and an 8 × 50 µm device selected as reduce the impact of the second stage noise figure. The
a suitable unit cell for the output stage, allowing the noise figure of the second stage is not as critical, and this
power specification to be met. It was also determined stage was designed with higher gain than the first. The
that three stages would be required to meet the overall resulting LNA design requires just 10 mA of DC supply
transmit gain specification. current from its +4 V supply. The gate bias voltage is ap-
The design of the complete three-stage PA pro- plied at pin “LNA_Vg” and the +4 V drain bias is applied
gressed with selection of the optimum transistor sizes at “LNA_Vd.” The “LNA_Vsense” pin is provided to allow
for the driver and pre-driver stage. Again careful trade- for bias current monitoring. Monitoring the bias current al-
offs were considered; larger transistor sizes improve the lows control of the gate voltage to compensate for chang-
overall linearity but reduce the PAE. With the size and es in environmental conditions, for example a change in
bias of all transistors selected the detailed design of temperature. When correctly biased this monitoring pin is
the matching and biasing circuitry could proceed. The at 3.9 V. The use of an enhancement mode process meant
layout was considered from an early stage of the de- that only positive supply voltages were required, making
sign process to ensure a practical implementation was the MMIC very convenient for system integration.
possible without incurring unacceptable parasitics. A Careful EM simulation was essential to ensure good
common gate bias line was used for stages one and RF performance from the various blocks. A step by step
two (applied at pin PA_Vg12) and a separate bias line approach was adopted adding a part of the circuit to
for stage three (PA_Vg3). This allowed the possibility of the EM simulation at a time with the rest of the block
separately optimizing the two voltages for potential lin- still simulated using Process Design Kit (PDK) models.
earity or PAE improvements to the PA. The drain sup- As the IC was destined for packaging in an overmould-
plies were similarly applied through two separate pins, ed plastic package, the presence of the moulding com-
although these were connected on the PCB; the +4 V pound on top of the IC also needed to be accounted
drain supply is applied at “PA_Vd12” and “PA_Vd3.” for in the EM simulation.
The SPDT switch is a series-shunt design incorpo-
rating multiple transistors in both the series and shunt EVALUATION AND TEST
arms for improved linearity.1 The off-state capacitance A photograph of the FEM die is shown in Fig-
of the transistors limits the inherent isolation of the off- ure 3. The FEM MMIC die measures 3.38 mm ×
state device at high frequencies; at 28 GHz the switch 1.99 mm. Its pad/pin positions are similar to those
transistors will have an isolation of just a few dB.2 Re- shown in the block diagram although it incorpo-
ducing the transistor size to improve the inherent isola- rates a number of GND pads in order to make it
tion increases the on-state insertion loss and degrades fully RF-on-wafer (RFOW) testable. It was designed
its linearity and so was not an option. The approach to be packaged in a low-cost plastic overmolded
taken was to include on-chip inductive compensation
to improve the off-state isolation. Care was taken to en-
sure low insertion loss in the on-state to enable a high
output power from the transmit path and a low noise
figure from the receive path. The switch is controlled
by a single bit, “Vctrl1,” which is set to 4 V for Tx mode
or 0 V for Rx mode. Single bit control is facilitated by
the “SPDT Control Circuit,” which is essentially a one to
two line decoder. The combined supply current drawn
by both the control circuit and the SPDT itself is just 1
mA from the +4 V applied at “VD_SW.”
The input to the receive path is at the “Antenna” pin
which is routed to the input of a two-stage LNA by the
SPDT. The output of the receive path is at the pin la-
beled “LNA_RFout.” As with the PA, the LNA also has
 Fig. 3 Die photograph of FEM MMIC for 28 GHz 5G.
6
5 mm × 5 mm QFN. In addition to accounting for the Evaluation results for the packaged FEM MMIC
effects of the moulding compound, the RF transition mounted on the PCB and referenced to the package’s
from IC to PCB needed to be carefully designed. A cus- RF pins. Throughout the evaluation, a commercially
tom lead-frame was designed to facilitate this and the available multi-channel DAC and ADC IC was used to
RF ports of the package are all implemented as ground- control and monitor the FEM. The FEM does not re-
signal-ground interfaces. quire any negative voltages as it was designed on an
Following fabrication, several of the die were tested enhancement mode process. A comparison of the mea-
RFOW which confirmed that the first pass design had sured to simulated S-parameters of the Tx path of a typ-
been successful prior to packaging. The RFOW results ical FEM are shown in Figure 5. The measured data and
are not presented here but all measurements were simulated results match reasonably well. In this mode,
made on a packaged assembled IC mounted on a rep- the LNA is powered down, the SPDT control bit “Vctrl1”
resentative evaluation PCB. is toggled high and the PA biased to around 70 mA
The evaluation PCB was designed using a low cost total quiescent current from +4 V. Small signal gain (S21)
laminate PCB material suitable for mass volume produc- is 17.1 dB ± 0.4 dB from 27 to 29 GHz. The input return
tion. Samples of the packaged FEMs were assembled on loss (S11) is better than 18 dB across the band. The out-
to the evaluation PCBs; all of the measured performance put is matched for best PAE at back-off rather than best
is calibrated to the package pins on the evaluation PCB S22 but the measured S22 (not shown) is 8 dB or better
and include the effects of the IC to PCB transition. A TRL across the band.
calibration tile was designed to allow the calibration of The output referred third-order intercept point (OIP3)
the measured performance to the reference planes of of the Tx path was evaluated with a tone spacing of
the package. A photograph of one of the evaluation 100 MHz to reflect the wide channel bandwidths antici-
PCBs next to a TRL calibration PCB is shown in Figure 4. pated in 5G systems. Figure 6 is a plot of the measured
OIP3 of a typical FEM with the wanted
32 output tone powers ranging from 1 to
28 11 dBm per tone. It can be seen that
24
the OIP3 is around +28 dBm across the
OIP3 (dBm)
Measured
20 Simulated
16 5G band and shows very little variation
12 with tone power over a 10 dB dynamic
8 range. A plot comparing the measured
 Fig. 4 Photograph of packaged FEM 4
0 to simulated OIP3 versus frequency is
evaluation PCB and TRL calibration PCB.
26 27 28 29 30 shown in Figure 7 and demonstrates
Frequency (GHz) good agreement.
Although 5G systems will require lin-
S11 _Meas, S21_Meas, S11 _Sim, S21 _Sim (dB)
 Fig. 7 Comparison of measured and ear amplification to preserve modula-

simulated OIP3 vs. frequency.
20 tion fidelity, the output referred P1dB
15 and PAE were also measured to provide
10
5
25 50 a figure of merit for comparative purpos-
es. The measured performance is shown
Tx Power (dBm)
0 20 40
Tx PAE (%)
–5
15
P1dB_meas
30
in Figure 8 and shows a P1dB around
–10 PAE at P1dB_meas 20.2 dBm, which rises to 21 dBm at sat-
–15 10 20
–20 uration. The PAE of the FEM Tx path is
–25 5 10 around 20 percent, falling slightly at the
–30
20 23 26 29 32 35
0 0 top of the band.
26 27 28 29 30
Frequency (GHz)
As mentioned above, the FEM is de-
Frequency (GHz) signed for optimum performance (OIP3
and PAE) when operated at around 7 dB
 Fig. 5 Measured to simulated small-  Fig. 8 Measured P1dB and PAE of backed-off from P1dB, specifically with
signal performance of the Tx path of the the Tx path vs. frequency.
FEM. the IMD3 at a level of below ‐35 dBc
relative to the wanted products during a
25
two-tone test with 100 MHz tone spac-
Tx Power (dBm) / PAE (%)
Power at IMD3 = 35 dBc_meas

32 Power at IMD3 = 35 dBc_sim
28 20
PAE at IMD3 = 35 dBc_meas ing. This operating point is close to that
24 11 dBm
PAE at IMD3 = 35 dBc_sim envisaged in the 5G system for which
OIP3 (dBm)
20 9 dBm 15
16 7 dBm the FEM was designed.
5 dBm
12
3 dBm
10 Figure 9 shows a plot of the mea-
8
4
1 dBm 5 sured and simulated PAE and total RF
0 output power when operating at an
26 27 28 29 30 0 IMD3 point of ‐35 dBc. The measured
26 27 28 29 30
Frequency (GHz)
Frequency (GHz) PAE is 6.5 percent, which is good and
largely due to the PA being designed to
 Fig. 6 FEM Tx OIP3 vs. frequency  Fig. 9 Measured and simulated operate in deep class AB. The total RF
vs. output tone power (100 MHz tone Tx power and PAE operating at ~7 dB
spacing). backed-off.
7
output power is around 13.5 dBm, which equates to an
OIP3 level of +28 dBm. 1000
The on-chip Tx power detector characteristic pro-
Tx Vref-Vdet (mV)
vides a DC voltage that allows monitoring of the RF
output power. The temperature compensated detec- 100
tor output “Vref-Vdet” is plotted in mV on a logarithmic
scale against output power in dBm over a 15 dB dy-
namic range in Figure 10. On this scale the characteris-
10
tic is linear making power monitoring easier. 4 6 8 10 12 14 16 18 20 22 24
When the Rx path of the FEM is selected the PA is Pout (dBm)
powered down, “Vctrl1” is set to 0 V and the LNA biased
to around 10 mA from +4 V with 3.9 V observed on the
“LNA_Vsense” pin. Figure 11 is a plot comparing the  Fig. 10 FEM Tx on-chip power
detector measured characteristics at 28
measured and simulated gain and noise figure (NF). The GHz.
measured small signal gain is around 13.5 dB with a gain
flatness of just ±0.3 dB across the band. The Rx path has
an excellent noise figure of typically 3.3 dB from 27 to 16
Gain & Noise Figure (dB)

14
29 GHz with good agreement between simulated and
12
measured performance. 10
NF
Gain
The Rx path also demonstrates impressive linearity for 8 Gain (Sim.)
NF (Sim.)
the modest power consumption (just 40 mW: 10 mA at 6
4 V). Key parameters such as P1dB and OIP3 are around 4
6.2 and 21 dBm, respectively, across the band. Figure 2
12 is a plot of the measured P1dB and OIP3 versus fre- 0
26 27 28 29 30
quency. Frequency (GHz)
CONCLUSION
The FEM MMIC described here will potentially play
 Fig. 11 Rx path measured and
simulated gain and NF.
a key role in future 28 GHz, 5G systems. The part has
been shown to address all the requirements for inte-
gration into mmWave phased-array or beam switched 30
terminals and offers excellent Tx linearity and efficiency 25
together with outstanding Rx noise figure. The key per-
OIP3 (dBm)
20
formance specifications for both transmit and receive OIP3
15
paths were met, ensuring that the part is highly suit- P1dB
able for mm-Wave 5G applications. The IC also includes 10
useful features such as a Tx power detector, Tx and Rx 5
enable circuits, an SPDT decoder circuit and Rx bias 0
monitoring. Realized on a state of the art 0.15 µm en- 26 27 28 29 30
hancement mode GaAs PHEMT process the part is ex- Frequency (GHz)
tremely easy to control and monitor using widely avail-
able multi-channel ADC and DAC ICs. In addition, the  Fig. 12 Rx path measured P1dB and
OIP3.
part is conveniently housed in a compact and low cost
5 mm × 5 mm plastic overmolded QFN SMT package.n
References
1. L. Devlin, “The Design of Integrated Switches and Phase Shift-
ers,” Proceedings of the IEEE Tutorial Colloquium on Design of
RFICs and MMICs, November 24, 1999, pp. 2/1-14.
2. S. Glynn and L. Devlin, “The Design of a Dual-Band PA for
mmWave 5G Applications,” Proceedings of the RF and Micro-
wave Society (ARMMS) Conference, November 13, 2017.
8
5G Fixed Wireless Access Array
and RF Front-End Trade-Offs
Bror Peterson and David Schnaufer
Qorvo, Greensboro, N.C.
T
he vision of next-generation 5G a radio access technology (RAT) that takes
networks is to deliver an order-of- advantage of new wideband frequency allo-
magnitude improvement in ca- cations, both sub-6 GHz and above 24 GHz,
pacity, coverage and connectivity to achieve the huge peak throughputs and
compared to existing 4G networks, all at low latencies proposed by the International
substantially lower cost per bit to carriers Mobile Telecommunications vision for 2020
and consumers. The many use cases and and beyond.1
services enabled by 5G technology and Mobile network operators are capitalizing
networks are shown in Figure 1. In this first on the improvements introduced by NR RAT,
phase of 5G new radio (NR) standardiza- particularly in the mmWave bands, to deliver
tion, the primary focus has been on defining gigabit fixed wireless access (FWA) services
to houses, apartments and businesses, in
Device-to-Device a fraction of the time and cost of tradi-
Communications Automobile-to-Automobile
Densification Communications tional cable and fiber to the home installa-
tions. Carriers are also using FWA as the
Smart Grid testbed toward a truly mobile broadband
experience. Not surprisingly, Verizon, AT&T
Smart Home
Enhanced
and other carriers are aggressively trialing
Mission Critical
Mobile Broadband Services FWA, with the goal of full commercialization
Fixed Wireless in 2019.
Access
In this article, we analyze the architecture,
Massive Internet
of Things
Critical/Emergency semiconductor technology and RF front-
Broadcast on Services
Mobile Device end (RFFE) design needed to deliver these
new mmWave FWA services. We discuss the
Augmented link budget requirements and walk through
Reality & Virtual Reality an example of suburban deployment. We
IoT Smart Cities address the traits and trade-offs of hybrid
Machine-to-Machine beamforming versus all-digital beamform-
ing for the base transceiver station (BTS) and
 Fig. 1 5G use cases.
WWW.MWJOURNAL.COM/ARTICLES/29707
9
carrier aggregation supporting up to 1.2 GHz of instan-
taneous bandwidth. Customer premise equipment (CPE)
will support peak rates over 2 Gbps and come in several
form factors: all outdoor, split-mount and all indoor desk-
top and dongle-type units. Mobile-handset form factors
will follow.
Global mmWave spectrum availability is shown in
Figure 2. In the U.S., most trials are in the old block A
LMDS band between 27.5 and 28.35 GHz, but the plan-
of-record of carriers is to deploy nationwide in the wider
39 GHz band, which is licensed on a larger economic
 Fig. 2 Global 5G bands above 24 GHz. area basis. These candidate bands have been assigned
by 3GPP and, except for 28 GHz, are being harmonized
globally by the International Telecommunications Union.2
Active Antenna System FWA describes a wireless connection between a cen-
tralized sectorized BTS and numerous fixed or nomadic
users (see Figure 3). Systems are being designed to le-
verage existing tower sites and support a low-cost, self-
install CPE build-out. Both are critical to keeping initial
deployment investment low while the business case
Customer Premise
Equipment for FWA is validated. Early deployments will be mostly
Mobile
Equipment Customer Premise outdoor-to-outdoor and use professional roof-level in-
Equipment stallations that maximize range, ensure initial customer
satisfaction and allow time for BTS and CPE equipment
Edge
Central Data to reach the needed cost and performance targets.
Center
Data Center Large coverage is essential to the success of the FWA
business case. To illustrate this, consider a suburban de-
 Fig. 3 End–to–end FWA network. ployment with 800 homes/km2, as shown in Figure 4. For
BTS inter-site distance (ISD) of 500 m, we need at least
20 sectors, each covering 35 houses from nine cell sites.
Assuming 33 percent of the customers sign up for 1 Gbps
service and a 5x network oversubscription ratio, an aver-
age aggregate BTS capacity of 3 Gbps/sector is needed.
• Random Dallas Suburb This capacity is achieved with a 400 MHz bandwidth, as-
- 800 Houses/km2 suming an average spectral efficiency of 2 bps/Hz and
- 500 m ISD
- 9 Cell Sites four layers of spatial multiplexing. If customers pay $100
- 23 Sectors per month, the annual revenue will be $280,000/km2/year.
- ~35 Houses/Sector
Of course, without accounting for recurring costs, it is not
• Capacity Per Sector
- 35 Houses/Sector clear FWA is a good business, but we can conclude that
- 5x Oversubscription as ISD increases, the business case improves. To that end,
- 1 Gbps Service
- Capacity ~5 Gbps carriers are driving equipment vendors to build BTS and
• Random Dallas •Suburb
Random Dallas• Suburb
BTS Parameters
CPE equipment that operate up to regulatory limits to
- 800 Houses/km2- 800 Houses/km - 2Capacity ~5 Gbps maximize coverage and profitability.
- 500 m ISD - 500 m ISD - 400 MHz BW
- 9 Cell Sites - 9 Cell Sites - 16-QAM w/LDPC: 3 bps/Hz In the U.S., the Federal Communications Commission
- 23 Sectors - 23 Sectors - 4 Spatial Streams/Layers has defined very high effective isotropic radiated pow-
- ~35 Houses/Sector
- ~35 Houses/Sector
• Business Case er (EIRP) limits for the 28 and 39 GHz bands,3 shown in
• Capacity Per Sector
• Capacity Per Sector - 35% Take Rate
- 35 Houses/Sector
Table 1. The challenge becomes building systems that
- 35 Houses/Sector - $100/Month for 1 Gbps SLA
- 5x Oversubscription
- 5x Oversubscription - $14k/Sector/Year meet these targets within the cost, size, weight and pow-
- 1 Gbps Service - 1 Gbps Service- $177k/km2/Year er budgets expected by carriers. Selecting the proper
- Capacity ~5 Gbps - Capacity ~5 Gbps
front-end architecture and RF semiconductor technology
• BTS Parameters• BTS Parameters
are key to getting
 Fig.- 4 FWA in
Capacity - Capacity environment.
~5aGbps
suburban
- 400 MHz BW - 400 MHz BW
~5 Gbps
there. TABLE 1
- 16-QAM w/LDPC: 3 bps/Hz
analyze the semiconductor
- 16-QAM w/LDPC: 3 bps/Hz technology and RFFE com-
- 4 Spatial Streams/Layers
- 4 Spatial Streams/Layers FWA Link Budget FCC POWER LIMITS FOR 28 AND
ponents that enable each. Finally, we discuss the de- 39 GHz BANDS
• Business Case • Business Case The standards
sign of- 35%
a GaN-on-SiC front-end
Take Rate - 35% Take Rate module (FEM) designed
for-5G
1$100/Month for 1 Gbps SLA community has Equipment Class Power (EIRP)
specifically for the
- $100/Month FWA
Gbps SLA market.
- $14k/Sector/Year
- $14k/Sector/Year been busy defining
2
- $177k/km2/Year- $177k/km /Year Base Station 75 dBm/100 MHz
FWA DEPLOYMENT the performance
A clear advantage of using mmWave is the availability requirements and Mobile Station 43 dBm
of underutilized contiguous spectrum at low cost. These evaluating use
bands allow wide component carrier bandwidths up to cases over a broad Transportable 55 dBm
400 MHz and commercial BTSs are being designed with range of mmWave Station
10
frequencies. The urban-macro scenario is the best rep- To avoid overdesign of the cost-sensitive CPE equip-
resentation of a typical FWA deployment: having large ment and shift the burden toward the BTS, the link de-
ISD of 300 to 500 m and providing large path-loss bud- sign begins at the CPE receiver and works backward to
gets that overcome many of the propagation challeng- arrive at the BTS transmitter requirements. In lieu of the
es at mmWave frequencies. To understand the needed conventional G/T (the ratio of antenna gain to system
link budget, consider a statistical path-loss simulation noise temperature) figure-of-merit (FOM), we define a
using detailed large-scale channel models that account more convenient G/NF FOM: the peak antenna gain (in-
for non-line-of-site conditions and outdoor-to-indoor cluding beamforming gain) normalized by the NF of the
penetration, like those defined by 3GPP.4 Figure 5 shows receiver. Figure 6 illustrates the required EIRP for the
the result for a 500 m ISD urban-macro environment range of receive G/NF to overcome a targeted path loss
performed by equipment vendors and operators. For delivering an edge-of-coverage throughput of 1 Gbps,
this simulation, 28 GHz channel models were used with assuming the modulation spectral efficiency is effectively
80 percent of the randomly dropped users falling in- 2 bps/Hz and demodulation SNR is 8 dB. From the
doors and 20 percent outdoors. Of the indoor users, 50 graph, the BTS EIRP for a range of CPE receiver’s G/NF
percent were subject to high penetration-loss models can be determined. For example, 65 dBm BTS EIRP will
and 50 percent lower loss. Long-term, carriers desire be needed to sustain a 1 Gbps link at 165 dB of path
at least 80 percent of their potential users to be self- loss when the CPE receiver G/NF is ≥ 21 dBi.
installable to minimize more expensive professional Next, we consider the impact of receiver NF by plot-
roof-level installations. The distribution curve shows the ting the minimum number of array elements needed to
maximum system path loss to be 165 dB. achieve G/NF of 21 dB (see Figure 7). We also plot the
Closing the link depends on many variables, includ- total low noise amplifier (LNA) power consumption. By
ing transmit EIRP, receive antenna gain, receiver noise adjusting the axis range, we can overlap the two and
figure (NF) and minimum edge-of-coverage throughput. see the impact NF has on array size, complexity and
power. For this example, each LNA consumes 40 mW,
100
which is typical for phased arrays. The NFs of RFFEs,
Probability Pathloss is Less Than
500 m ISD ~333 m Cell Range

90
including the T/R switch losses, are shown for 130 nm
80 Pro Install Self Install CPE NEC
70 CATT SiGe BiCMOS, 90 nm GaAs PHEMT and 150 nm GaN
HEMT at 30 GHz. The compound semiconductor tech-
Abscissa (%)
Qualcomm
60
ZTE
50 Huawei
nology provides ≥ 1.5 dB advantage, translating to a
40 Samsung 30 percent savings in array size, power and, ultimately,
Ericsson CPE cost.
30
Intel
20 China To explore architecture trades that are key to tech-
10 Telecom nology selection and design of the RFFE components,
0
180 160 140 120 100
we start by understanding the antenna scanning re-
Path Loss (dB)
quirements. We highlight the circuit density and pack-
aging impact for integrated, dual-polarization receive/
 Fig. 5 Statistical path loss simulation for urban-macro transmit arrays. Finally, we investigate all-digital beam-
environment with 500 m ISD. forming and hybrid RF beamforming architectures and
the requirements for each.
75
Transmit EIRP (dBm)
1D or 2D Scanning
160 dB
70 The number of active channels in the array depends
165 dB
on many things. Let’s start by first understanding the azi-
170 dB
65 muth and elevation scanning requirements and whether
two-dimensional beamforming is required for a typical
60 FWA deployment or if a lower complexity, one-dimen-
5 10 15 20 25 30 35
sional (azimuth only) beamforming array is sufficient.
Receive G/NF (dB) This decision impacts the power amplifier (PA). Figure 8
 Fig. 6 Transmit EIRP and receive G/NF vs. path-loss for
1 Gbps edge-of-coverage throughput.
15-25 m
256
30 (a)
# of Array Elements
Total LNA Pdc (W)
192
20
128
SiGe
1.5 dB 10
64
GaAs/GaN
0 0
2 3 4 5 6 7 8
Noise Figure (dB) (b)
 Fig. 7 Array size vs. front-end NF and power consumption  Fig. 8 Array complexity depends on the scanning range
for G/NF = 21 dB. needed for the deployment: suburban (a) or urban (b).
11
and far users. The nominal half-power beamwidth can
be approximated as 102°/NANT and the array gain by
10log10(NANT ) + 5 dBi. With passively combined an-
tennas, the elevation beam pattern is focused and the
1:4 Splitter
fixed antenna gain increases, as shown in Table 2. For
the suburban FWA deployment, a 13 to 26 degree
beamwidth is sufficient, with the passively combined
column array from four to eight elements. In the urban
scenario, however, the elevation scanning requirements
• Nx Fewer Components • Nx More Components
• Nx Larger PA • Nx Smaller PAs are greater, and systems will be limited to one or two
• Higher Feed Losses • Lower Feed Losses passive elements.
• Fixed Elevation Pattern • Elevation Beam Steering
(a) (b) Figure 9b illustrates the per-element active array.
Both the per-element and column-fed array architec-
 Fig. 9 Column-fed (a) and per-element (b) active arrays. tures have the same antenna gain, but the column-fed
array has a fixed elevation beam pattern. The per-el-
TABLE 2 ement array supports wider scan angles but needs 4x
as many PAs, phase shifters and variable gain compo-
APPROXIMATE PERFORMANCE FOR CORPORATELY
FED ELEMENTS nents for an antenna with four elements. To achieve
the same EIRP, the PA driving a column-fed array with
Column Array Size Beamwidth (°) Gain (dB) four antennas will need to provide at least 4x the out-
Single Element 102 5 put power, which can easily change the semiconductor
2-Element 51 8 selection. It is reasonable to assume a suburban BTS
will use antennas with 6 to 9 dB higher passive antenna
4-Element 26 11
gain compared to an urban deployment. As a result, the
8-Element 13 14 phased array needs far fewer active channels to achieve
the same EIRP, significantly reducing active component
shows two FWA deployment scenarios. In the suburban count and integration complexity.
deployment, the tower heights range from 15 to 25 m
and the cell radius is 500 to 1000 m, with an average Array Front-End Density
house height of 10 m. Just as with traditional macro Early mmWave FWA BTS designs used separate, single-
cellular systems, there is no need for fully adaptive el- polarization transmit and receive antenna arrays, which al-
evation scanning. The elevation beam can be focused lowed significantly more board area for components. These
down by corporately feeding several passive antenna designs avoided the additional insertion loss and linearity
elements, as shown in Figure 9a. This vertically stacked challenges of a T/R switch. However, a major architecture
column of radiating elements is designed to minimize trend is integrated T/R, dual-polarization arrays (see Figure
radiation above the houses and fill in any nulls along 10), which is driving RFFE density. The key reason is spa-
the ground. Further, the gain pattern is designed to tial correlation. Adaptive beamforming performance de-
increase at relatively the same rate as the path loss. pends on the ability to calibrate the receive and transmit
This provides more uniform coverage for both near arrays relative to one another. As such, it is important to
T Array R Array T/R Array Dual-Polarization T/R Array
Isolation
10 cm > 40 dB
1:N Splitter 1:N Combiner

1:N Combiner/Splitter
2x the Circuit Density 1:N Combiner/Splitter
4x the Circuit Density
Transitioning From Separate Arrays Integrated T/R Integrated T/R and

Dual Polarization
 Fig. 10 FWA antenna arrays are evolving from separate T and R arrays to integrated T/R arrays with dual polarization.
12
RF-DAC
14-bit 4.5 Gbps DVG A Hybrid IQ Mixer 9 W GaN PA Circulator
IRF VGA Driver
BPF
Column-Antenna
Corporate Feed
RF-DAC IQ Mixer
Column-Antenna
14-bit 4.5 Gbps DVG A Hybrid Driver 9 W GaN PA Circulator
IRF BPF VGA
Column-Antenna
Column-Antenna
Corporate Feed
DUC DAC 90°
0°
JES D204B
RF-ADC
14-bit 3 Gbps Gain Block DVG A LNA + IQ Mixer
AAF BPF
DDC ADC 90°
 Fig. 11 Array design using digital beamforming and commercial, off-the-shelf components.
active channels, there is a valid concern that the power
Tx Total/Channel = 13 W Rx Total/Channel = 4 W dissipation and cost of such a system would be pro-
Other: 0.5 Down-Converter/LNA: 0.8 hibitive. Therefore, vendors are exploring hybrid beam-
Final PA: RF-DAC: 1 RF-ADC: formed architectures,5 which allows flexibility between
8.8 2.2 the number of baseband channels and the number of
DVGA: 0.5
active RF channels. This approach better balances ana-
VGA: 1.2 log beamforming gain and baseband processing. The
DVGA: following sections analyze the two architectures and
Driver: 1 0.9 discuss the RFFE approaches needed for each.
Digital Beamforming
(a) (b) Gain Block: 0.2
Assuming large elevation scanning is not required for
suburban FWA and a well-designed, column antenna
 Fig. 12 Power dissipation of the transmit (a) and receive
provides gain of up to 14 dBi, we start with a mmWave
(b) chains.
integrate the transmit and receive channels for both po- BTS transceiver design targeting an EIRP of 65 dBm and
larizations, so the array shares a common set of antenna compute the power consumption using off-the-shelf
elements and RF paths. The net result is a requirement point-to-point microwave radio components that have
for the RFFE to have 4x the circuit density of earlier sys- been available for years, including a high-power, 28 GHz
tems. GaN balanced amplifier. The multi-slat array and trans-
At mmWave frequencies, the lattice spacing be- ceiver are shown in Figure 11. Assuming circulator and
tween phased-array elements becomes small, e.g., feed-losses of 1.5 dB, the power at the antenna port is
3.75 mm at 39 GHz. To minimize feed loss, it is impor- 27 dBm. From the following equations, achieving 65
tant to locate the front-end components close to the dBm EIRP requires 16 transceivers that, combined, pro-
radiating elements. Therefore, it is necessary to shrink vide 12 dB of digital beamforming gain:
the RFFE footprint and integrate multiple functions, ei-
ther monolithically on the die or within the package, EIRP = GBF ( dB) + GANT ( dBi) +
using a multi-chip module. Tiling all these functions in PAVE _ TOTAL ( dBm)
a small area requires either very small PAs, requiring a
many-fold increase in array size, or using high-power EIRP = 10log10 (NCOLUMNS ) +
density technologies like GaN. Further, it is critical to 10log10 (NPAs ) + GANT +
use a semiconductor technology that can withstand PAVE/CHANNEL ( dBm)
high junction temperatures. The reliability of SiGe de-
grades rapidly above 150°C, but GaN on SiC is rated The power consumption for each transceiver is
to 225°C. This 75°C advantage in junction temperature shown in Figure 12. The total power dissipation (PDISS)
has a large impact on the thermal design, especially for at 80 percent transmit duty cycle for all 16 slats will be
outdoor, passively-cooled phased arrays. 220 W per polarization, and a dual-polarized system will
require 440 W. For all outdoor tower-top electronics,
ALL-DIGITAL VS. HYBRID ARRAYS where passive cooling is required, it is challenging to
It was natural for BTS vendors to first explore extend- thermally manage more than 300 W from the RF subsys-
ing the current, sub-6 GHz, all-digital beamforming, tem, suggesting an all-digital beamforming architecture
massive MIMO platforms to mmWave. This preserves using today’s off-the-shelf components is impractical.
the basic architecture and the advanced signal pro- However, new GaN FEMs are on the horizon to help
cessing algorithms for beamformed spatial multiplex- address this. As shown in Figure 13, the GaN PAs inte-
ing. However, due to the dramatic increase in channel grated in the FEM apply the tried-and-true Doherty ef-
bandwidths offered by mmWave and the need for many ficiency-boosting technique to mmWave. With Doherty
13
Hybrid Beamforming
The basic block diagram for a hybrid beamforming
active array is shown in Figure 14. Here, N baseband
channels are driving RF analog beamformers, which di-
vide the signal M-ways and provide discrete phase and
amplitude control. FEMs drive each M-element subar-
Corporate Feed
ray panel. The number of baseband paths and subarray
Transceiver
panels is determined by the minimum number of spatial
streams or beams that are needed. The number of beam-
former branches and elements in each subarray panel is
a function of the targeted EIRP and G/NF. While a pop-
ular design ratio is to have one baseband path for every
16 to 64 active elements, it really depends on the de-
(a)
ployment scenario. For example, with a hot-spot small
cell (or on the CPE terminal side), a 1:16 ratio single
Power-Added Efficiency (%)
40
36
panel is appropriate. A macro BTS would have two to
32 four subarray panels with 64 active elements, where
28 each panel is dual-polarized, totaling four to eight base-
24 band paths and 256 to 512 active elements. The digital
20
16 and analog beamforming work together, to maximize
12 coverage or independently, to provide spatially sepa-
24 26 28 30 32 34 36 38 40 42 rated beams to multiple users.
Output Power (dBm) There is an important trade unfolding, whether SiGe
(b)
front-ends can provide sufficient output power and ef-
 Fig. 13 Integrated FEM with symmetric GaN Doherty PA and ficiency to avoid the need for higher performance III-V
switch-LNA (a) and PA performance from 27.5 to 29.5 GHz (b). technology like GaAs or GaN. With good packaging
and integration, both approaches can meet the tight
PAs, digital pre-distortion (DPD) is needed; however, antenna lattice-spacing requirements.
the adjacent channel power ratio (ACPR) requirements
defined for mmWave bands are significantly more re- FRONT-END SEMICONDUCTOR CHOICES
laxed, enabling a much “lighter” DPD solution. The The technology choice for the RFFE depends on the
estimated power dissipation of a 40 dBm PSAT, sym- EIRP and G/NF requirements of the system. Both are a
metric, multi-stage Doherty PA can be reduced more function of beamforming gain, which is a function of the
than 50 percent. In the above system, this improvement array size. To illustrate this, Figure 15 shows the aver-
alone drops the total PDISS below 300 W. Combined age PA power (PAVE) per channel needed as a function
with power savings from next-generation RF-sampling of array size and antenna gain for a uniform rectangular
digital-to-analog and analog-to-digital converters, array delivering 65 dBm EIRP. The graph is overlaid with
advancement in mmWave CMOS transceivers and in- an indication of the power ranges best suited for each
creased levels of small-signal integration, it will not be semiconductor technology. The limits were set based
long before we see more all-digital beamforming solu- on benchmarks of each technology, avoiding exotic
tions being deployed. power-combining or methods that degrade component
RF Beamformer Front-Ends
Digital Processing Mixed Signal IF-RF Conversion
LO
D/A Subarray
DUC Panel 1
Digital Beamformer
DUC A/D
N:Number of 1:M/N
Baseband Channels
M:N
DUC D/A
DUC A/D
Subarray
Panel N
CMOS
SiGe-BiCMOS
GaAs-/GaN
 Fig. 14 Active array using hybrid beamforming.
14
reliability or efficiency. As array size gets large (more
Average Tx Power per Element (dBm)
50 35
45
EIRP = 65 dBm than 512 active elements), the power per element be-
comes small enough to allow SiGe, which can be in-
Antenna Array Gain (dBi)

40 f = 28 GHz 30
35 y/2 = 5.4 mm tegrated into the core beamformer RFIC. In contrast,
GaN emax = 90%
30 4πemaxDarray2
25 by using GaN for the front-end, the same EIRP can be
25 Array Gain ≈
y2
achieved with 8 to 16x fewer channels.
20 20
15
GaAs System Power Dissipation
10 15 For an array delivering 64 dBm EIRP, Figure 16
5 SiGe shows an analysis of the total PDISS of the beamformer
0 10 plus the front-end as a function of the number of ac-
32
64
96
128
256
512
1024
tive elements in each subarray panel. The PDISS is shown
Number of Active Elements
for several error vector magnitude (EVM) levels, since
 Fig. 15 Optimum RFFE technology vs. array size. the EVM determines the power back-off and efficiency
achieved by the front-end. We assume each beamform-
110 40.0 er branch consumes 190 mW, which is the typical power
Element Gain = 8 dBi EVM = 8% consumption of core beamformers in the market.6 The
EVM = 6% 35.0
100
GaN
EVM = 4% system on the far right of the figure represents an all-
Power Dissipated (W)
Pave/Channel
Pave/Channel (dBm)
30.0
90 SiGe solution with 512 elements, with an output power
25.0
80 per element of 2 dBm and consuming approximately
20.0 100 W. Moving left, the number of elements decreases,
70
2-Stage
15.0 the PAVE per channel increases and PDISS is optimized to
60 GaAs 10.0 a point where beamforming gain starts to roll off sharp-
50 3-Stage SiGe 5.0
ly, and the PDISS to maintain the EIRP rapidly increas-
es. The small steps in the dissipation curves represent
40 0
where the front-end transitions from a single stage to
16
40
64
88
112
136
160
184
208
232
256
280
304
328
352
376
400
424
448
472
496
512
two-stage and three-stage designs to provide sufficient

Number of Active Channels
gain. As stages are added, the efficiency drops with the
 Fig. 16 System power dissipation vs. array size and EVM increase in power dissipation.
for 64 dBm EIRP. Designing to optimize system PDISS without regard-
ing complexity or cost, an array of about 128 elements
with a two-stage, 14 dBm output PA (24 dBm P1dB) is
TABLE 3 the best choice. However, if we strive to optimize cost,
RELATIVE COST OF ALL SiGe AND SiGe complexity and yield for a PDISS budget of under 100
BEAMFORMER WITH GaN FEM W, the optimum selection is the range of 48 to 64 ac-
Parameter Units All SiGe GaN +SiGe tive channels using a three-stage GaN PA with an aver-
Average Output Power age output power of 20 to 23 dBm, depending on the
dBm 2 20 EVM target. The trends shown in Figure 16 are less a
per Channel
Power Dissipation per
function of PA efficiency and more a function of beam-
mW 190 1329 former inefficiency. In other words, the choice to in-
Channel
crease array size 8x to allow an all-SiGe solution comes
Antenna Element Gain dBi 8 8
with a penalty, given that the input signal is divided
Number of Active many more ways and requires linearly biased, power
512 64
Channels
consuming devices to amplify the signal back up.
EIRP dBmi 64 64
Cost Analysis
Total Power Dissipation W 97 97 The cost of phased arrays include the RF compo-
Beamformer Die Area nents, printed circuit board material and the antennas
mm2 2.3 2.3
per Channel themselves. Using compound semiconductor front-
Front-End Die Area per ends allows an immediate 8x reduction in array size
mm2 1.2 5.2 with no increase in PDISS. Even with lower-cost printed
Channel
antenna technology, this is a large saving in expen-
Total SiGe Die Area mm2 1752 144
sive antenna-quality substrate material. Consider-
Total GaN Die Area mm2 0 334 ing component cost, the current die cost per mm2 of
150 nm GaN on SiC fabricated on 4-inch wafers is only
Die Cost Units Notes
4.5x the cost of 8-inch 130 nm SiGe. As 6-inch GaN pro-
All SiGe System Die
1752 $/x duction lines shift into high volume, the cost of GaN rela-
Cost tive to SiGe drops to 3x. A summary of the assumptions
GaN + SiGe System
1647 $/x
4-inch GaN = and a cost comparison of the relative raw die cost of the
Die Cost (4-inch GaN) 4.5x two technologies is shown in Table 3. Using a high-pow-
GaN + SiGe System 6-inch GaN er density compound semiconductor like GaN on 6-inch
1146 $/x
Die Cost (6-inch GaN) = 3x wafers can save up to 35 percent in the raw die cost rela-
15
tive to an all-SiGe architecture. Even though the cost of the GaN MMICs contains a three-stage linear PA, three-
silicon technologies is lower per device, the cost of the stage LNA and a low-loss, high-linearity SPDT switch.
complete system is significantly higher. The FEM covers 37.1 to 40.5 GHz and provides 23 dBm
average output power, which supports 256-QAM EVM
GaN FRONT-END MODULES levels, with 24 dB transmit gain. In receive mode, the
To validate the concept of a GaN FEM for mmWave NF is 4.1 dB, and receive gain is 16 dB. The package
FWA arrays, Qorvo set out to design the highest power, size is 4.5 mm × 6.0 mm × 1.8 mm.7-8
lowest NF FEM for the 37 to 40 GHz band. To support the
trend to integrated transmit/receive arrays, the front-end SUMMARY
includes a PA, integrated T/R switch and a low NF LNA. FWA is rapidly approaching commercialization. This is
The module was designed with sufficient gain to be driv- due to the abundance of low-cost spectrum, early regu-
en by core beamformer RFICs, which have a typical drive latory and standards work and the opportunity for op-
level of 2 dBm. The FEM’s PAVE of 23 dBm was selected erators to quickly tap a new market. The remaining chal-
from an analysis similar to that shown in Figure 16, and the lenge is the availability of equipment capable of closing
PSAT was determined by analyzing the needed headroom the link at a reasonable cost. Both hybrid beamforming
to support a back-off linearity of ≥ 33 dBc ACPR, EVM ≤ and all-digital beamforming architectures are being ex-
4 percent and a 400 MHz orthogonal frequency-division plored. These architectures capitalize on the respective
multiple access (OFDMA) waveform. strengths of commercial semiconductor processes. The
A key design decision was determining if GaAs or use of GaN front-ends in either approach provides op-
GaN or a combination of both were needed. The die erators and manufacturers a pathway to achieving high
size for a GaAs PA would not allow the FEM to meet the EIRP targets while minimizing cost, complexity, size and
tight 3.75 mm lattice spacing at 39 GHz. The equivalent power dissipation. To prove the feasibility, Qorvo has
output power GaN PA is 4x smaller with no sacrifice in developed a 39 GHz FEM based on a highly integrated
gain and a slight benefit in efficiency. Considering the GaN-on-SiC T/R MMIC and is developing similar FEMs
LNA, the 90 nm GaAs PHEMT process was favored due for other millimeter wave frequency bands proposed for
to its slightly superior NF. However, the net improve- 5G systems.■
ment was only a few tenths of a dB once the additional
bond wires and 50 Ω matching networks were consid- References
1. International Telecommunications Union, ITU-R Radiocommunications
ered. The trade-off analysis concluded it was better Sector of ITU, “IMT Vision–Framework and Overall Objectives of the Fu-
to stay with a monolithic GaN design that allowed co- ture Development of IMT for 2020 and Beyond,” August 2015, www.itu.
matching of the PA, LNA and T/R switch. Such a design int/dms_pubrec/itu-r/rec/m/R-REC-M.2083-0-201509-I!!PDF-E.pdf.
2. International Telecommunications Union, Resolution 238 (WRC-15), “Stud-
was lower risk, easier to assemble and test, and the ies on Frequency-Related Matters for International Mobile Telecommuni-
MMIC was as compact as possible. The system thermal cations Identification Including Possible Additional Allocations to the Mo-
analysis indicated that the higher junction temperature bile Services on a Primary Basis in Portion(s) of the Frequency Range 24.25
and 86 GHz for Future Development of IMT-2020 and Beyond,” 2015,
offered by GaN-on-SiC was critical for passively-cooled www.itu.int/dms_pub/itu-r/oth/0c/0a/R0C0A00000C0014PDFE.pdf.
arrays. 3. Federal Communicationws Commission, “Use of Spectrum Bands Above
24 GHz for Mobile Radio Services, In the Matter of GN Docket No. 14-
As shown in Figure 17, the 39 GHz FEM integrates 177, IB Docket No. 15-256, RM-11664, WT Docket No. 10-112, IB Docket
two of the multi-function GaN MMICs into an air-cavity, No. 97-95,” July 2016, apps.fcc.gov/edocs_public/attachmatch/FCC-16-
embedded heat-slug, surface-mount package, sized 89A1.pdf.
4. 3GPP TR 38.901, “Study on Channel Model for Frequencies from 0.5
to meet the array element spacing at 39 GHz. Each of to 100 GHz,” September 2017, www.3gpp.org/ftp//Specs/archive/38_
series/38.901/38901-e20.zip.
5. A. F. Molisch et al., “Hybrid Beamforming for Massive MIMO: A Survey,”
LNA IEEE Communications Magazine, Vol. 55, No. 9, 2017, pp. 134–141.
Rx1 SW 6. B. Sadhu et al., “7.2 A 28GHz 32-Element Phased-Array Transceiver IC with
PA ANT1 Concurrent Dual Polarized Beams and 1.4 Degree Beam-steering Resolu-
Tx1 tion for 5G Communication,” 2017 IEEE International Solid-State Circuits
Conference (ISSCC), San Francisco, Calif, 2017, pp. 128–129.
1875 µm
6 mm
LNA 7. B. Kim and V. Z. Q. Li, “39 GHz GaN Front-End MMIC for 5G Applica-
Rx2 SW
tions,” 2017 IEEE Compound Semiconductor Integrated Circuit Sympo-
PA ANT2 sium (CSICS), Miami, Fla., 2017, pp. 1–4.
Tx2 8. “QPF4005 37-40.5 GaN Dual Channel FEM Datasheet,” www.qorvo.com/
products/d/da006271.
2700 µm 4.5 mm
(a) (b)
(c)
 Fig. 17 Integrated 39 GHz GaN front-end MMIC –

intentionally blurred (a), dual-channel FEM (b) and package (c).
16
WWHHITITEEPA
PAPPEERR
Basics
Basicsof
ofPower
PowerAmplifier
Amplifierand
and
Front End
Front EndModule
ModuleMeasurements
Measurements
CCOONNTTEN
ENTTSS
Gain
Gainand
andOutput
OutputPower
Power
Calibrating
CalibratingPower
PowerMeasurements
Measurementswith
withaaPower
PowerMeter
Meter
Measuring
MeasuringGain
Gainwith
withaaVector
VectorNetwork
NetworkAnalyzer
Analyzer
Return
ReturnLoss
Lossand
andReverse
ReverseIsolation
Isolation
Noise
NoiseFigure
Figure
Noise
NoiseUnit
UnitConversion
Conversion
Noise
NoiseFigure
FigureMeasurements
Measurements
Y-Factor
Y-FactorMethod
MethodUsing
UsingaaCalibrated
CalibratedNoise
NoiseSource
Source
Harmonics
Harmonics
Intermodulation
IntermodulationDistortion
Distortion
Theory
TheoryofofIntermodulation
IntermodulationDistortion
Distortion
IMD
IMDMeasurement
MeasurementConfiguration
Configuration
Efficiency
Efficiency
Drain
DrainEfficiency
Efficiency
Power
PowerAdded
AddedEfficiency
Efficiency
ni.com/rf
ni.com/rf
17
3 Basics of Power Amplifier and Front End Module Measurements
Gain and Output Power

Two important characteristics of an RF PA are gain and output power. Gain describes
the relationship between the input and output power of the device. In general, a PA exhibits
relatively constant gain across a wide range of input power levels with the gain dropping
as the output power approaches the device’s saturation region. This effect is known as gain
compression.
INPUT VERSUS OUTPUT POWER
Pout Ideal Behavior
1 dB
1 dB Compression
Point
Actual Response
-8 dBm
Gain = Pout – Pin = 18 dB
+10 dBm Pin
Figure 2. Input versus Output Power in a Typical PA
One of the most common methods to characterize a PA’s usable maximum output power
is with the 1 dB compression point metric. The 1 dB compression point, shown in Figure 2,
describes the operating point at which a PA delivers gain that is exactly 1 dB less than the
gain it would otherwise deliver in its linear operating region. For example, if a PA delivers
18 dB of gain in its linear region of operation, the 1 dB compression point is defined at the
output power at which the PA delivers exactly 17 dB of gain.
When testing the 1 dB compression point, you can use either a power-calibrated vector
network analyzer (VNA) or a combination of an RF signal generator and an RF signal analyzer.
The RF signal generator and signal analyzer combination provides the fastest method to
measure the 1 dB compression point, and you can use either a continuous wave (CW) signal
generator or a vector signal generator (VSG) to perform this measurement.
You can measure gain as a function of input power by sweeping the power level of the
signal generator and measuring the output power of the PA with the RF signal analyzer. One
optimization technique to consider for production test is configuring the VSG to generate
a ramp waveform instead of a series of CW tones at different power levels, as shown in
Figure 3. By acquiring the ramp signal with the vector signal analyzer (VSA), you can easily
correlate the input power to output power to determine gain versus input power. This ramp
signal method is substantially faster than configuring the signal generator for multiple steps,
and can save valuable test time.
ni.com/rf
18
MEASURING THE COMPRESSION POINT WITH A RAMP SIGNAL
VECTOR SIGNAL GENERATOR
DAC
VECTOR SIGNAL ANALYZER (SIMPLIFIED)
PA ADC
DAC
Power
Power
Gain
Time Time
Figure 3. You can measure the 1 dB compression point faster by stimulating a PA with a ramp signal.
Fast Power Level Servoing Using the NI Vector Signal Transceiver

A unique technology of the NI PA test solution is FPGA-based power level servoing using the
NI Vector Signal Transceiver (VST). Power level servoing is traditionally a time-consuming process.
However, you can achieve the fastest possible power level convergence by performing the control
loop entirely on the instrument FPGA. If you decouple the power level servoing algorithm from the
embedded controller and perform it on an FPGA, the test software can exploit dramatic measurement
parallelism. This results in significant reductions in test time and test cost. For more information about
fast power measurements using the NI VST, visit FPGA Servoing for Power Amplifier Test.
PXI SYSTEM
PXI CONTROLLER
CPU Memory
Data
Transfer
VECTOR SIGNAL
TRANSCEIVER
Digital I/O
FPGA
RFout PA
RFin
ni.com/rf
19
One important technique to improve the power accuracy of gain and power measurements
is to use a small attenuator, or pad, between the instrumentation and the PA under test. The
power uncertainty due to mismatch error is significantly reduced by using an in-line fixed
attenuator on both the PA input and output, as shown in Figure 4.
IMPROVING MISMATCH UNCERTAINTY
VSG PA VSA
Figure 4. Attenuators between the instrument and the PA improve mismatch uncertainty.
Calibrating Power Measurements with a Power Meter

You can measure the output power of a PA using either a power meter or a VSA. Historically,
the power meter was the most accurate method of measuring power with absolute power
measurements , providing accuracies to within ±0.2 dB. However, today’s modern VSAs
are equipped with tools such as an onboard calibration standard that greatly improve their
accuracy when measuring absolute power. Using only onboard calibration, a VSA, such as the
NI PXIe-5668R, can measure power to within ±0.4 dB, and can achieve even better power
accuracy when referenced to a calibration standard, such as a power meter.
Although power meters can generally measure RF power more accurately than a VSA, there are
several advantages to measuring the DUT’s output power and gain with a VSA. In addition to the
simplicity of doing many measurements with a single instrument, VSAs can generally measure
power faster than power meters, and, because of this, many engineers rely on the VSA to
measure power along with the 1 dB compression point in automated RF test applications.
An important step when measuring power and gain is to calibrate the system setup using
a power meter. To complete this calibration step, you first connect a power meter to the
reference plane at the input of the DUT, as shown in Figure 5. Using the power meter, you
measure the output power of the signal generator plus any attenuators and cabling over
ni.com/rf
20
a range of frequencies. Once you complete this step, you have characterized the signal
generator to within the power accuracy of the power meter.
SYSTEM CALIBRATION
Step 1: Calibrate the signal generator with the power meter.
VSG Power Meter
Step 2: Calibrate the signal analyzer with the signal generator.
VSG VSA
Figure 5. System calibration is a two-step process that uses a power meter

to calibrate both the signal generator and analyzer.
After you calibrate the signal generator setup, you then connect the signal analyzer setup,
which includes both instruments and any cables, attenuators, and so on, directly to the
signal generator setup. Using the calibrated response of the signal generator, and assuming
the measurements made with the power meter are correct, you can then determine the
measurement offset of the signal analyzer setup. By executing these calibration steps, you can
more accurately measure output power and gain by referencing the result to the power meter.
Measuring Gain with a Vector Network Analyzer

Although the most common, and often fastest, way to measure PA gain in automated testing
scenarios is with a VSG and VSA, you can measure PA gain with a VNA as well. To measure
PA gain using a two-port VNA, connect port 1 of the VNA to the PA input and connect port 2
of the VNA to the PA output. Then measure the S21 response, which equals the PA gain.
One of the key considerations when measuring PA gain with a VNA is to ensure that the
output power of the PA does not saturate or damage the VNA receiver. In this scenario,
the exact amount of external attenuation can significantly affect the accuracy of the S21
measurement. Although many VNAs have a maximum safe input power level that is typically
on the order of 1 W (+30 dBm), measurement accuracy typically degrades when operating
the instrument close to the maximum power level, especially because VNAs typically have a
much narrower programmable attenuator range than a VSA does.
An accurate PA measurement using a VNA requires careful attention to the power levels
present at the input of port 2. As a general rule of thumb, ensure that the source power of
the PA and the input power of the VNA port 2 are relatively similar. Thus, if you expect the PA
to produce 20 dB of gain, you should connect a 20 dB attenuator between the PA output and
the VNA port 2, as shown in Figure 6.
ni.com/rf
21
AVOIDING PORT SATURATION
VNA
Port 1 PA
Port 2
Figure 6. Use an attenuator when measuring PA gain

with a VNA to avoid saturating Port 2.
One important nuance of using an attenuator between the output of the PA and the VNA port 2
are the implications on the calibration reference plane. Whether you calibrate the VNA using
the short-open-load-thru (SOLT) method or with an automatic calibration kit, you should aim
to establish a reference plane as close to the DUT as possible.
In the case of using an external attenuator, you should calibrate the measurement system
with the attenuator and any associated cables and fixturing in the path, as shown in Figure 7.
As a result of calibrating the measurement system with the attenuator in the signal path,
subsequent VNA S21 measurements display the gain directly. For more information about
VNA calibration, visit Introduction to Network Analyzer Measurements available on ni.com.
UNDERSTANDING THE REFERENCE PLANE
Calibration Calibration
Reference Plane Reference Plane
VNA
Port 1
Port 2
Figure 7. The VNA calibration reference plane must extend

beyond the external attenuator.
Return Loss and Reverse Isolation

Although measurements such as gain do not technically require a VNA to perform the
measurement, return loss and isolation measurements do require full network analysis.
The instrumentation setup for return loss and reverse isolation can vary depending on
whether you are attempting to characterize the small signal or large signal behavior of the PA.
The small signal is the signal within the linear region of operation, and the large signal is the
signal in the nonlinear region of operation. When measuring the small signal behavior, you
can accurately measure S11 (input return loss) and S22 (output return loss) using a VNA.
In some instances, measuring the output return loss might require slight modifications to
the test configuration, shown in Figure 8. The required attenuation used between the PA
output and the VNA port 2 can be relatively high, especially for high-gain PAs. In this scenario,
the combination of high PA gain and relatively low return loss results in an extremely
low-powered reflection signal, measured by port 2 of the VNA. As a result, accurate S22
measurements on a high-gain PA often require the use of an attenuator that produces less
loss than the amplifier produces gain. In these instances, it is actually common to use one
ni.com/rf
22
Fast S-Parameter Measurements in Production Test with STS

The NI Semiconductor Test System (STS) is a fully automated production test system that applies
an innovative approach to S-parameter measurements in production test. This system combines port
modules with the NI Vector Signal Transceiver (VST). In addition to switching and pre-selection, the
port modules contain directional couplers that effectively turn the VST into a VNA. As a result, you
can quickly perform S-parameter measurements in a production test environment without the cost of
additional instrumentation. S-parameter measurements are calibrated using the multiport calibration
module, which enables you to calibrate up to 48 RF ports automatically. For more information on the
NI STS, visit ni.com/semiconductor-test-system.
attenuator value for S11, S12, and S21 measurements and another attenuator value for
S22 measurements.
MEASURING S-PARAMETERS
S11 S12 S22
VNA
Port 1 PA
Port 2
Figure 8. You can measure reverse isolation and return

loss with a VNA.
When testing PAs in large signal conditions, the test configuration is substantially more
complex. In large signal conditions, a substantial portion of the output energy is converted
to harmonics and not captured by a traditional VNA. As a result, complete characterization
ni.com/rf
23
of the large-signal performance of a PA requires a large-signal network analyzer (LSNA), or

load-pull test bench, as illustrated in Figure 9. Because S12 and S21 measurements are more
difficult in large-signal conditions, one approach is to measure S21 performance as a function
of input and/or output impedance. In this scenario, a programmable tuner is placed at either
the input or the output of the DUT.
BASIC LOAD-PULL TEST CONFIGURATION
S21
Signal
PA
Generator
Signal
Analyzer
Figure 9. Block Diagram of Basic Load-Pull Test Configuration
Although this method does not enable you to measure the input impedance (S11) or output
impedance (S22) directly, it does enable you to estimate the input/output impedances that
result in the best PA performance or efficiency through trial and error. Note that the classic
configuration involves using a CW signal generator to source power and a power meter to
measure it. Today, it is now possible to measure large-signal performance using modulated
signals sourced and measured with VSGs and VSAs.
Noise Figure
Although gain and output power are some of the most critical measurements of power
amplifiers, noise figure remains the most crucial measurements of low noise amplifiers
(LNAs). Because the LNA is typically the first component in a receiver, the noise figure and
gain of the LNA drives the noise figure of the receiver.
There are a number of ways to quantitatively describe both noise figure and noise factor.
One of the earliest definitions is the one proposed by Harold Friis in the 1940s. In Friis’s
definition, noise factor (the linear equivalent of noise figure) is the degradation of a particular
signal’s signal-to-noise ratio (SNR) as it passes through a particular component. Noise factor
and noise figure are inherently unitless ratios, and while noise factor expresses this ratio in
liner terms, noise figure expresses this ratio in logarithmic terms.
SNRin
F = SNR
out
(1)
Equation 1. Noise Factor as a Function of SNR
Using Equation 1, if a signal had an SNR of 100 dB at the input of an LNA with a noise figure
of 5 dB, the SNR at the output would be 100 – 5 dB = 95 dB. As Figure 10 illustrates, a “black
box” component with a noise figure of XdB would degrade the SNR by XdB.
ni.com/rf
24
INTRINSIC NOISE POWER OVER THERMAL NOISE
Signal Power Signal Power
SNR =
Black Box SNR = XdB - NF
XdB
Noise Power Thermal Noise Noise Power
Figure 10. Noise figure is the addition of a component‘s intrinsic noise power
on top of thermal noise.
Another view of noise figure is that it describes the noise power, in dB, that a particular
passive or active component adds on top of room temperature’s thermal noise of -174 dBm/Hz.
This definition closely mirrors the widely accepted IEEE definition of noise factor, which is
defined in Equation 2.
Nadded + kTo BG
F= kTo BG (2)
where k represents Boltzman’s constant
T0 represents room temperature
B represents bandwidth
G is the gain of the DUT
Equation 2. Formal Definition of Noise Factor
In Equation 2, kTo is simplified to the thermal noise at room temperature, or -174 dBm/Hz.
Thus, noise factor is a component’s noise power added on top of signal power.
For example, in a scenario where an antenna is connected to an LNA, the noise power at the
input of the LNA is -174 dBm/Hz. At the LNA’s output, the noise power is -174 dBm/Hz plus
the noise figure of the LNA. In this scenario, a noise figure of 5 dB would yield an output
noise power of -169 dBm/Hz. Note that in this case, you can simply add 5 dB to -174 dBm/Hz
because we are describing noise figure in logarithmic terms.
Noise Unit Conversion

Before we describe noise figure measurements in detail, it is useful to first define some
units and terms commonly used to describe noise measurements. Some of the most
common metrics include noise figure, noise factor, and noise temperature.
Noise figure (NF) describes the noise power a component adds on top of thermal noise in
dB, and noise factor (F) describes the noise power a component adds on top of thermal noise
in linear terms. You can convert NF to F and vice versa using Equations 3 and 4.
ni.com/rf
25
NF = 10 log10 (F) (3)

NFdB
F = 10 10 (4)
Equations 3 and 4. Conversion of Noise Factor to Noise Figure and Vice Versa
A related expression of noise power is noise temperature. Because noise power is directly
proportional to the temperature of the device in Kelvin, the noise temperature (Te) is the
equivalent temperature of a device that produces a certain amount of noise power. It is
important to recognize that the equivalent noise temperature of a device is a theoretical value that
merely describes the theoretical temperature at which a passive device produces a particular
noise power level. You can relate noise temperature to noise factor using Equations 5 and 6.
Te = T0 (F –1) (5)
Equation 5. Noise Temperature as a Function of Noise Factor
Te
F= T + 1
0
(6)
Equation 6. Noise Factor as a Function of Noise Temperature and Vice Versa
In Equations 5 and 6, T0 is an expression that commonly refers to room temperature, or 290 K.

Given these equations, a component with a noise factor of 4, or a noise figure of 6.02 dB,
would have an equivalent temperature of 290 K (4 – 1) = 870 K. Given this calculation, the
inherent thermal noise of a component that has been heated to 870 K is exactly 6.02 dB
higher than a component at room temperature, which has a temperature of 290 K. Thus,
having an equivalent temperature of 870 dB is the same as having a noise factor of 4 and
a noise figure of 6.02 dB.
The Friis formula for the noise factor of a cascaded RF system is a final key formula that is
useful for noise factor measurements. This equation is important because when measuring
the noise figure of a component, you must consider that any measurement includes both the
noise contribution of the DUT and the noise contribution of the instrument itself. When using
the Friis formula, consider the cascaded RF system shown in Figure 11.
CASCADED RF SYSTEM
Figure 11. Each component can be described

by both its gain and noise figure.
Using the Friis formula, shown in Equation 7, you can calculate the noise factor (F)
of the system.
ni.com/rf
26
F2 – 1 F –1 Fn – 1
FSystem = F1 + + 3 + ... +
G1 G1G2 G1G2G3...Gn-1
Equation 7. The Friis Formula for Noise Factor of a Cascaded System
Note that the Friis formula requires that both noise and gain are expressed in linear terms
and not logarithmic terms. Also, note that when the first component of the system has a
high gain, like an LNA, the noise figure of the system is dominated by the first component.
Thus, for typical noise figure measurements you can generally omit all but the first two terms
of Equation 7 and use the simplified version in Equation 8.
F2 – 1
F12 = F1 + G1
Equation 8. Noise Factor of a Cascaded Two-Stage System
Likewise, you can relate the noise temperature of the cascade using a similar relationship.
By substituting the equation for noise temperature from noise factor, you can derive that the
noise temperature of the first component in a cascaded system is equal to the noise figure
of the system minus the noise contribution from the second element, as shown in Equation 9.
T2
T12 = T1 + G
1
Equation 9. Noise Temperature of a Cascaded Two-Stage System
Noise Figure Measurements

Although there are several methods for measuring noise figure, the two most common methods
are the cold source method, also known as the gain method, and the Y-factor method. The basic
principle of the gain method is to terminate the input of the DUT and then measure the output
noise of the DUT using a signal analyzer, as shown in Figure 12. In this scenario, output noise
power is the intrinsic noise of the DUT that is amplified by the gain of the DUT.
COLD SOURCE NOISE FIGURE MEASUREMENT TECHNIQUE
LNA MODULE RF SIGNAL ANALYZER
LNA
Figure 12. Terminate the input of the DUT when using the cold source method.
The cold source method is generally most effective in high-gain LNAs since signal analyzers can
measure noise power more accurately for signals that are significantly above their inherent noise
floors. One of the drawbacks of the cold source method is that it is most susceptible to voltage
standing wave ratio (VSWR) uncertainty. In addition, traditional methods to improve VSWR, such as
ni.com/rf
27
the use of external attenuators, degrade the instrument’s ability to measure low-power signals. As
a result, the cold source measurement technique is more accurate when you can compensate for
VSWR. In fact, you can occasionally use a network analyzer to measure noise figure, assuming
that its noise floor is low enough, because it is able to reduce uncertainty due to VSWR.
Y-Factor Method Using a Calibrated Noise Source

A second, and perhaps more common, noise figure measurement is the Y-factor method. This
method involves introducing a calibrated noise source to an LNA, or a PA, and measuring noise
power both when the noise source is turned on and when the noise source is turned off. The
premise of the Y-factor method is actually straightforward if you consider that both the DUT
and the signal analyzer are part of a two-part cascaded RF system, as shown in Figure 13.
Y-FACTOR NOISE FIGURE MEASUREMENT TECHNIQUE
NOISE SOURCE
f1 f2
RFin RFout
LNA
g1 g2
Figure 13. Connecting an LNA to a signal analyzer produces a cascaded RF system.
With a noise source, typically either an LNA or a demodulator, connected to the input of
the DUT, you can model the test setup as a two-stage system. In this case, the noise figure
of the system includes the noise figure of the first component, an LNA, plus the noise
contribution of the RF signal analyzer. The Y-factor approach is designed to measure the noise
factor of the DUT (F1) by first solving for both the noise factor of the system (F12) and the
gain of the DUT (G1). Thus, the process of measuring the noise figure of an RF component
using the Y-factor method involves the following two steps:
1. Measure the noise figure of the signal analyzer.

2. Measure the noise figure of the system with the DUT in place.
One of the essential elements of the Y-factor test setup is the calibrated noise source.
A calibrated noise source is extremely useful when measuring noise figure because it
is able to provide a noise-like signal into the DUT at a relatively low power level, with a
calibrated on/off ratio.
Noise sources have two settings, on and off, and their characteristic specification is their
excess noise ratio (ENR). ENR is defined by Equation 10, where TsON and TsOFF represent the
equivalent temperature, noise power, in each setting. For practical measurement purposes,
you can generally assume that TsOFF = T0 = 290 K. The ENR of a noise source is typically
printed on the source directly or offered through a specification document, and typical ENR
values range from 5 dB to 30 dB depending on the application.
(
ENRdB = 10 log
T ON – T OFF S S
(
T0
Equation 10. ENR is essentially the power ratio of noise on to noise off.
ni.com/rf
28
Step 1: Characterize the Noise Figure of the Signal Analyzer

The first step in measuring the noise figure using the Y-factor method is to measure the noise
figure of the signal analyzer without the DUT connected. Note that typical noise sources
require a 28 VDC supply that is usually provided via the 28 VDC port of the RF signal analyzer,
as shown in Figure 14.
CALIBRATION STEP FOR Y-FACTOR TECHNIQUE
NOISE SOURCE
RF SIGNAL ANALYZER
f2
g2
28
V DC
Figure 14. Connect the noise source directly to the signal analyzer
to measure the signal analyzer’s inherent noise figure.
With the setup shown in Figure 14, the Y-factor is the measured ratio of the noise source’s
output noise power when turned on (Non) versus when it is turned off (Noff). Thus, measuring
the Y-factor consists of two power measurements, Non and Noff. Note that the ratio of Non and
Noff must be expressed in linear terms, with noise power in watts. This ratio is illustrated in
Equation 11.
Non
Y= N
off
Equation 11. The Y-factor is the ratio of Non to Noff.
You can measure Non and Noff with an RF signal analyzer using a channel power measurement.
Because the accuracy of a noise figure measurement with an RF signal analyzer depends on
the noise figure of the instrument itself, it is important to minimize the instrument’s noise
figure by taking the following steps:
1. Turn on the instrument’s pre-amplifier (if available).
2. Set the reference level as low as possible, typically to less than -50 dBm.
3. Manually set the instrument’s attenuation to 0 dB.
Note that for high-gain DUTs, the VSWR benefit of having the instrument’s attenuation
set higher than 0 dB may outweigh the noise-floor improvement of removing attenuation.
Although by using the Y-factor method some uncertainty due to VSWR is theoretically
removed, a small amount of error due to VSWR is present given that the signal analyzer
sees different mismatch during the calibration step and the measurement step.
With the settings above, you can measure the RF signal analyzer’s noise power using the
power-in-band measurement. The power-in-band measurement provides a more accurate
ni.com/rf
29
approach to measuring noise power than merely measuring noise floor with a marker. If you
are measuring power in dBm, simply convert dBm to watts using Equation 12.
PdBm –30
Pw = 10 10
Equation 12. Power in Watts as a Function of dBm
Because the power-in-band measurement integrates the noise power over a large number
of frequency bins, the bandwidth of the measurement significantly affects the measured
power result. For example, -90 dBm in a bandwidth of 1 MHz is equivalent to a measurement
of -100 dBm in a bandwidth of 100 kHz. For that reason, it is often useful to express noise
power in terms of dBm/Hz, as shown in Equation 13.
PdBm/Hz = PdBm – 10 x log (Measurement Bandwidth)
Equation 13. Converting Measured Power to dBm/Hz
Note that although expressing noise power in dBm/Hz is useful because of the insight it
provides about the noise floor of the signal analyzer, the measurement bandwidth typically
does not actually affect the Y-factor ratio unless it is wider than the bandwidth of the noise
signal itself. Assuming you use the same measurement bandwidth to measure both Non and
Noff, the bandwidth units cancel each other out. The general rule of thumb is to ensure your
measurement bandwidth is narrower than the output bandwidth of the noise source and
equal to or narrower than the bandwidth of the signal the DUT is designed to amplify. Once
you have determined the Y-factor based on the power measurements described previously,
noise figure is merely a function of ENR and Y-factor, as illustrated in Equation 14.
NFSA = ENRdB – 10 log10 (Y – 1)
Equation 14. Noise Figure as a Function of ENR and Y-Factor
Alternatively, you can also solve for the noise figure and noise factor by expressing noise in
terms of noise temperature. Assuming that T0 = 290 K (room temperature) when the noise
source is off, the noise temperature in the on state of the noise source is a function of ENR.
Using Equations 15 and 16, you can first solve for the noise temperature of the noise source
based on its ENR, and then use that value in conjunction with the measured Y-factor to
determine the noise temperature of the signal analyzer.
ENR
T ON = T0 (10
source
10 )10 + T OFF
source
T ON – (YSA x 290)
source
TSA = YSA – 1
Equations 15 and 16. You can use the Y-factor to determine

the noise temperature of the signal analyzer.
ni.com/rf
30
Step 2: Insert the DUT

Once you’ve solved for the noise figure/factor/temperature of the RF signal analyzer by
connecting the noise source directly to the signal analyzer, the next step is to measure the
noise figure of the system with the DUT in place. To do so, connect the output of the noise
source to the input of the DUT, as shown in Figure 15.
MEASUREMENT STEP FOR Y-FACTOR TECHNIQUE
NOISE SOURCE
f1 f2
RFin RFout
LNA
g1 g2
Figure 15. Insert the DUT to measure the noise figure of the RF system.
With the DUT inserted between the noise source and the signal analyzer, terms such as F12, G12,
and T12 refer to the noise factor, gain, and temperature, respectively, of the entire system. Similar
to the calibration step, you must next calculate the Y-factor of the entire system. In this step, you
measure the Y-factor of the system or cascade, with the end result of calculating Y12.
N ON
Y12 =
12
N OFF
12
Equation 17. The Y-factor of the system is the ratio of measured noise with the DUT inserted.
Again, you can calculate either the noise figure or the noise temperature of the system using
either Equation 18 or 19, respectively.
NF12 = ENRdB – 10 log10 (Y12 – 1)
Equation 18. Calculating Noise Figure in dB
ni.com/rf
31
T ON – (Y12 x 290)
source
T12 = Y12 – 1
Equation 19. Calculating Noise Temperature in Kelvin
Once the noise figure (NF12) or the noise temperature (T12) of the entire system is known,
it is possible to determine the noise figure of the DUT by applying the Friis formula.
Step 3: Calculate the Noise Figure

Once you have measured the noise figure, or noise factor, of both the signal analyzer alone
and the measurement system with the DUT in place, you are almost ready to solve for the
noise figure of the DUT. The last remaining step before you can do this is to calculate the gain
of the DUT, as shown in Equation 20.
N12ON − N12OFF
GDUT =
N 2ON − N 2OFF
Equation 20. Calculating the Gain Based on
All Four Noise Power Measurements
With the system noise figure (F12) and DUT gain (G1) both known, you can solve for the noise
figure of the DUT using the Friis formula, as shown in Equation 21. Note that the Friis formula
expresses noise factor in linear terms, so you must convert any units of gain or noise figure
to linear terms.
F2 – 1
F1 = F12 –
G1
Equation 21. Calculating the Noise Factor

of the DUT Using the Measured Results
Alternatively, if you’ve kept all the measurements in terms of noise temperature, you can
solve for the noise temperature of the DUT using Equation 22.
T2
T1 = T12 –
G1
Equation 22. Calculating the Noise Temperature

of the DUT Using the Measured Results
Again, the gain must be expressed in linear terms. Once the equivalent noise temperature
of the DUT (T1) is known, you can convert it to noise figure using Equation 23.
T1
F1 = 1 –
T0
Equation 23. Converting Noise Temperature

to Noise Factor Assuming T0 = 290 K
ni.com/rf
32
The Y-factor method for measuring noise figure is an accurate method to measure the noise
figure of an LNA or even a PA. Although it requires some mental exercise to think in terms
of noise figure, noise factor, and noise temperature, you can easily measure noise figure
accurately with basic knowledge.
Harmonics
A second key attribute of active PAs and FEMs is their harmonic behavior. Harmonic behavior
is caused by nonlinear operation and results in output power at frequencies that are a multiple
of the transmit frequency. Because many wireless standards prescribe strict specifications
for out-of-band emissions, engineers measure harmonics to assess whether a PA or FEM
violates these emissions requirements.
The precise method of measuring harmonic power often varies according to the intended
use of the PA. For devices such as a general purpose PA, harmonics measurements involve
stimulating the DUT with a CW signal and measuring the power of the resulting tone at
various frequencies. By contrast, when testing wireless handsets or base station PAs, harmonics
measurements often require a modulated stimulus. In addition, measuring harmonic power
often requires special attention to the bandwidth characteristics of the signal.
Harmonics Using a CW Stimulus

Harmonic measurements using a CW stimulus require a signal generator and a signal
analyzer. On the stimulus side, you configure the signal generator to produce a CW tone
at the desired output power and frequency. With the signal generator producing a stimulus,
the signal analyzer measures the output power at multiples of the input frequency. Common
harmonic measurements are the third and fifth harmonics, which are measured at 3X and 5X
the stimulus frequency, respectively.
With an RF signal analyzer, you can use one of several measurement techniques to measure
the output power of the harmonic. One straightforward approach is to tune the analyzer
to the expected frequency of the harmonic and perform a peak search to find the harmonic.
For example, when measuring the third harmonic of a PA when generating a 1 GHz tone, the
third harmonic occurs at exactly 3 GHz.
ni.com/rf
33
A second method to measure harmonic power is to use the signal analyzer’s zero span mode
and perform the measurements in the time domain. Configured in zero span mode, the signal
analyzer effectively performs a series of power-in-band measurements and displays the results
as a function of time. With this mode, you can use some of the built-in averaging capabilities
of the signal analyzer by measuring the power over a gated window in the time domain.
Harmonics Using a Modulated Stimulus

In practice, many PAs are designed to amplify modulated signals, and the harmonic
performance of these PAs requires a modulated stimulus. Similar to the CW case, you
introduce a stimulus at a known power to the input of the PA, generally at a power level close
to the device’s saturation point.
When measuring the output harmonic power, engineers often use a range of methods according
to constraints such as measurement time and required accuracy. In practice, wireless
standards such as 3GPP LTE and IEEE 802.11ac do not specifically prescribe a harmonics
requirement. Rather, they specify maximum spurious emissions requirements over a range
of frequencies. For example, 3GPP LTE dictates that an LTE transmitter may not emit power
exceeding -30 dBm within a bandwidth of 1 MHz at frequencies over 1 GHz. In this case,
validating that the PA does not cause a transmitter to exceed this limit requires engineers to
measure the emissions in a 1 MHz bandwidth at the harmonic frequencies.
In practice, engineers use a range of methods to ensure that a PA does not violate the spurious
emissions requirements. In an R&D or characterization lab, it is common for engineers
to measure spurious emissions directly with a spectrum signal analyzer or a vector signal
analyzer. However, in a manufacturing environment where test time is critical, it is more
common to measure the power of the harmonic directly and use statistical correlation to predict
whether or not the PA violates the spurious emissions requirement.
Measuring the harmonics of modulated signals requires careful attention to the

measurement bandwidth. This is because the required measurement bandwidth of the
harmonic changes as a function of the harmonic. For example, when testing the output
harmonics of a PA that requires N MHz measurement bandwidth, the measurement
bandwidth of the third harmonic must be 3 * N MHz, and the measurement bandwidth of the
fifth harmonic must be 5 * N MHz. For example, Figure 16 illustrates that the bandwidth of a
harmonic increases along with the order of the harmonic.
ni.com/rf
34
BANDWIDTH OF HARMONIC
Fundamental Second Harmonic Third Harmonic

3.84 MHz 7.68 MHz 11.52 MHz
Power
Frequency
Figure 16. Bandwidth increases with the harmonic order.
Given the wide bandwidth requirements of the harmonics of modern communications

signals, engineers can measure harmonics in either the time or frequency domain, depending
on the instantaneous bandwidth of a signal analyzer. Time domain harmonics measurements
using the zero-span mode of the signal analyzer are preferable but are not always realistic.
For example, accurately measuring the third harmonic of a 160 MHz 802.11ac signal requires
480 MHz of instantaneous bandwidth. In this case, it is important to either generate a non-
bursted stimulus or carefully configure the signal analyzer’s power trigger to ensure that each
acquisition is measuring an equivalent snapshot of the bursted signal.
Note that the specifications of cellular standards such as GSM, UMTS, and LTE provide
specific guidance on the maximum spurious emissions of a transmission rather than the
harmonic power itself. As a result, many engineers characterize wireless PAs according to the
spurious emissions limit in addition to the actual harmonic.
Intermodulation Distortion
Another important metric in PA linearity is intermodulation distortion (IMD). Although the
IMD metric is an extremely useful tool to describe the linearity of all PAs, this metric is most
commonly used on general purpose power amplifiers for which an adjacent channel power
measurement is not relevant.
Theory of Intermodulation Distortion

In order to understand IMD, it is worthwhile to review the theory surrounding multi-tone
signals through a nonlinear system. Although a single tone stimulus creates harmonic
behavior at each multiple of the tone’s frequency, the nonlinear products resulting from
a multi-tone signal occur at a much broader range of frequencies.
As shown in Figure 17, the second order distortion products at the output of a PA occur at
frequencies that are every multiple of the input signal frequency. These distortion products
at f2 - f1, 2f1, f1 + f2, and 2f2 include the second harmonics of each of the input tones
ni.com/rf
35
in addition to two additional distortion products at both the sum and difference of the
frequencies of the input tones.
THEORY OF INTERMODULATION DISTORTION
f1 f2 2f2-f1
f1+f2
Intercept Point
Power
of the Third 2f1 2f2

Order (IM3)
2f1+f2 f1+2f2
f2-f1 2f1-f2 3f1 3f2

Third Order
Distortion
Product
Frequency
Figure 17. Theory of Intermodulation Distortion
Third order distortion products describe the interaction between the first order fundamental
tones and each second order distortion product. In fact, working through the mathematics,
you can visualize how two specific third order distortion products occur at a frequency
that is relatively close to the fundamental tones. In a practical application, where the PA
is transmitting a modulated signal, the effect of third order distortion occurs as in-band
distortion that is adjacent to the band of interest.
The IMD measurement describes the ratio of the power difference between the fundamental
tones and the adjacent third order distortion products, and is expressed in dB. One important
characteristic of the IMD measurement is that the power ratio between first order and third
order distortion products is entirely dependent on the absolute power level of the tones.
ni.com/rf
36
In many devices, the ratio of first order tones and third order distortion products is often quite
high in the linear region of operation. However, as the input power of the fundamental tones
increases, the third order distortion products increase as well. In fact, IMD products should
increase by 3 dB for every 1 dB increase in the power of the fundamental tones.
In theory, as third order distortion products increase in power at a faster rate than the
fundamental tones, the two types of signal are eventually equal in power level as illustrated
in Figure 18. The intercept point is the point at which the fundamental tones and third distortion
products are equal in theoretical power. This point is also known as the third order intercept
(TOI) or intercept point of the third order (IP3).
IMD and TOI Measurements Using PXI Signal Analyzers

Intermodulation distortion (IMD) and third order intercept (TOI) are built-in measurements of the NI-RFSA
Soft Front Panel (SFP). When performing these measurements, you can center the signal analyzer
on the two fundamental tones, ensuring that the third order distortion products are visible above the
noise floor. Select detect tones in the NI-RFSA SFP to produce the measurement results. The NI-
RFSA SFP automatically accounts for power differences in the fundamental tones along with power
differences in the third order distortion products to present the correct measurement result. For more
information on PXI RF signal analyzers, please visit ni.com/rf/test.
ni.com/rf
37
RELATIONSHIP OF OUTPUT POWER AND IMD
2f1-f2 f1 f2 2f2-f1 TOI
1 dB Gain Max Power

Power
Initial
Intermodulation
Distortion
Frequency
Figure 18. Every 1 dB increase in fundamental signal power produces a 3 dB increase in third order distortion products.
In practice, IP3/TOI is calculated rather than measured. Given the 3:1 ratio between the
power increase of the first and third order products, you can calculate IP3 using Equation 24.
IMD
IP3 = + PFundamental
2
Equation 24. Converting IMD to IP3
TOI is an extremely useful metric of PA performance because the IMD ratio is dependent on
power level. The TOI measurement combines an element of IMD performance with absolute
power level and presents the performance as a single number.
IMD Measurement Configuration

As the theory of IMD measurements suggests, performing this measurement requires a
two-tone stimulus. In most applications, the preferred approach to configuring a two-tone
stimulus is to use RF signal generators connected to an RF power combiner, as shown in
Figure 19.
ni.com/rf
38
IMD MEASUREMENT CONFIGURATION
TxEnable LTE PA MODULE
SMU DCin
RFout
Signal RFin
PA VSA
Generator
Signal
Generator
Figure 19. IMD measurements require two signal generators connected to a power combiner.
Because IMD is a common measurement, many RF signal analyzers have built-in

measurement functions to measure IMD/TOI. In fact, the NI-RFSA SFP automatically detects
both fundamental tones and third order distortion products and calculates the IMD ratio.
When testing high-performance PAs, it is important to generate the cleanest two-tone signal
possible. In some cases, a combiner alone does not provide sufficient isolation between the
two signal generators to offer a two-tone signal that is sufficiently clean. In these instances,
energy from one source can leak into the other source, creating small intermodulation
products that are introduced to the DUT by the test instruments.
One way to improve the isolation is to choose a combiner with a high port-to-port isolation.
Generally, purely resistive combiners feature only between 6 dB and 12 dB of isolation,
depending on the resistor topology. A good rule of thumb is that roughly 40 dB of isolation
is required to measure IP3 numbers above +25 dBm. In the event that the combiner’s
isolation is insufficient, you can improve the port-to-port isolation of the combiner using either
attenuators, isolators, or even an amplifier.
Assuming the source power is sufficiently high, a method to improve isolation is to introduce
a pad, or attenuator, between each source and the power combiner, as shown in Figure 20.
The attenuator provides additional isolation for signals traveling in the reverse direction.
Additional measures to increase the isolation using either directional couplers or isolators can
provide up to 50 dB of isolation if used at both ports. However, couplers are often limited to a
single octave and are thus not suitable for broadband applications.
ni.com/rf
39
IMPROVING SOURCE ISOLATION WITH ATTENUATORS
TxEnable LTE PA MODULE
SMU DCin
Signal RFin RFout

PA VSA
Generator
Signal
Generator
Figure 20. Attenuators can improve the quality of the stimulus signal by isolating each of the signal generators.
Amplifiers with ample reverse isolation are an excellent option when you require a high-
power stimulus. In addition to providing isolation between ports, an amplifier can offer gain to
the stimulus to enable the generation of a high-power two-tone stimulus.
Efficiency
Efficiency describes the ability of the PA to convert DC energy into RF energy. The two most
commonly used metrics of PA efficiency are drain efficiency and PAE. Each measurement
involves a signal generator, signal analyzer, and power supply or source measure unit (SMU),
as shown in Figure 21.
TEST CONFIGURATION POWER EFFICIENCY
SMU
VSG DUT VSA
Figure 21. Configuration for Power Efficiency and

Power Added Efficiency
The SMU is a critical instrument when measuring PA efficiency because of its ability to
measure the DC current consumption of the PA. Typically, engineers measure PA efficiency
over a range of supply voltages and use the SMU to measure current consumption at each
voltage supply.
Drain Efficiency
Drain efficiency is a metric of PA efficiency and describes the percentage of the DC power that
is converted to RF energy. The term drain efficiency comes from PA implementations that use
a field-effect transistor (FET), in which the DC power is supplied to the drain of the device.
You can calculate drain efficiency by dividing the output power of the PA by the supplied DC
power. Although most RF signal analyzers display measured power in watts, you can convert
units of power from dBm to watts using Equation 25.
ni.com/rf
40
(
PWatts = 10
PdBm - 30
(
10
Equation 25. Converting Power in dB to Power in Watts
With measured output power expressed in watts, you can calculate drain energy as the
output power divided by the DC power, as shown in Equation 26.
PRF Output
Drain Efficiency (n) =
PDC Supply
Equation 26. Drain efficiency is a function of RF output power and DC supplied power.
Drain efficiency is a useful metric of PA performance, but it is less useful in PAs with lower
gain where the input power is often significant. As a result, a second key metric of PA
efficiency that factors into the supplied input power is PAE.
Power Added Efficiency

Calculating PAE is similar to calculating drain efficiency except that it characterizes the power
added by the PA instead of the power at its output. Determining PAE requires knowledge
of the input power to a PA and can be calculated using Equation 27.
PRF PRF Output

– PPRF Output – PRF–Input
PRF Input
– POutput (η)RF
Power Added Power
Power AddedAdded
Efficiency Efficiency
ηEfficiency
= PRF η=
Output η–=PRF
PRF ηPRF
POutputOutput
– –PRF–Input
PRF
InputPDrain PRF InputPDrain
InputPDrain
Efficiency Efficiency
(η) P
(η) = =Input P
= =Efficiency
RF Output RF Input
Power Added
ficiency η = PRF Output – PRF InputPDrain Efficiency (η) =
Efficiency =RF
PRF Output
Output PRF InputPDrain Efficiency (η)
P PDC Supply
DC SupplyP DC Supply
DC Supply
PDC Supply
Equation 27. Calculating PAE
In practical use, PAE depends on a range of operating conditions, including the bias voltage
and output level of the amplified signal. As shown in Figure 22, the maximum PA efficiency
occurs at or around the saturation point. In addition, Figure 22 illustrates that the PAE versus
output profile is also dependent on the PA bias voltage. In general, a higher bias voltage
tends to lead to higher efficiency and higher maximum PAE.
ni.com/rf
41
POWER ADDED EFFICIENCY FOR AN ENVELOPE TRACKING PA
0.60
0.50
0.40
PAE
0.30
0.20
0.10
0.00
0 4 8 12 16 20 24 28 32
Pout (dBm)
Figure 22. Power Added Efficiency (PAE) as a Function of Output Power for Various
VCC Levels
Because PAE is dependent on the output power, the use of signals with higher peak to
average power ratios (PAPRs) can significantly influence the PAE of the device. Modern
wireless standards such as 802.11ac and LTE are based on OFDM technology, which
inherently involves amplifying signals with high PAPR. Because efficiency is often highest
close to the compression region of a PA and because driving a PA into compression
introduces distortion, the input power of a PA must be slightly backed off when amplifying
high PAPR signals. Variations in the power level result in the PA spending a smaller
percentage of time operating near its compression point, and the unfortunate trade-off is
a lower PAE. As a result, PA designers are increasingly looking towards techniques, such
as envelope tracking, to improve overall PAE.
©2016 National Instruments. All rights reserved. National Instruments, NI, and ni.com are trademarks of National Instruments. Other product and company names listed are trademarks or trade names of their
respective companies. 18909
ni.com/rf
42
W H IT E PA P E R
5G New Radio:
Introduction to the Physical Layer
CO N T EN T S
Introduction
PHY Design Considerations
Waveforms for 5G NR
Flexible Subcarrier Spacing and Symbol Lengths
NR Reference Signals
MIMO
mmWave for 5G
Bandwidth Parts
Conclusion: Comparing LTE and 5G NR PHY
Glossary
ni.com/5g/nr
ni.com/5G
43
Introduction
To understand the three broad use cases that 5G wireless technology seeks to transform (Figure 1),
consider a typical morning office commute in a 5G-connected car just a few years down the road.
The vehicle is constantly exchanging position, behavior, and system status information with
nearby vehicles, the surrounding highway infrastructure, and traffic control centers. Doing so in a
fast and reliable manner augments the car’s awareness of its surroundings and allows the driver
to turn the steering, accelerating, and braking functions over to the car’s semiautonomous driving
system. He can now focus on the morning’s first conference call.
The driver’s team is trying to find the root cause of a turbine malfunction. He puts on his
augmented reality (AR) set, and a wireless 4K video feed of an airplane turbine overlaid with
sensor data and gauge readings fills his screen. Collaborating in real time with a group of
engineers in three different countries, the team guides a technician to isolate one of the
components and recommends a troubleshooting procedure.
A few minutes later, when his intelligent-highway exit comes up, the driver takes back control
of the car, switches over to a low-bandwidth voice-only connection, and drives into work. The
car guides him to the closest available parking spot with an electric charging station. The
parking sensor at that spot detects his car and updates the parking availability information on
the network. When he plugs in the car to charge, the charging terminal establishes a low-
data-rate connection to verify his account and process payment.
10,000X More Traffic
20 Gbps DL Peak Data Rates 100 Mbps, High Mobility

ENHANCED MOBILE
BROADBAND
5G
MASSIVE MACHINE ULTRA-RELIABLE MACHINE
TYPE COMMUNICATION TYPE COMMUNICATION
1 M devices per km2 1 ms Latency
10-Year Battery Life 99.9999% Reliability
Figure 1. 5G Use Cases and Targeted Key Performance Indicators
ni.com/5g/nr
ni.com/5G
44
Enhanced Mobile Broadband (eMBB) seeks to significantly improve the data rate, latency,
user density, capacity, and coverage of mobile broadband access to allow the live streaming
of AR/VR applications, even in more crowded environments, such as the intelligent highway
the driver uses. Ultra-Reliable Low-Latency Communication (URLLC) enables users and devices
to communicate bidirectionally with other devices while generating minimal latency and
guaranteeing high network availability. Finally, Massive Machine-Type Communication (mMTC)
makes it possible for many low-cost, low-power, long-life devices to support applications such
as embedded highway sensors, parking sensors, and smart utility meters.
The requirements of these distinct use cases pose complex technical trade-offs, which involve
delicate design decisions. To guarantee interoperability and global access to 5G, the
International Telecommunication Union (ITU), an international union of telecommunication
industry players, national and regional standards development organizations, regulators,
network operators, universities, and research institutions, must approve the standards for 5G
technologies. The ITU-R (the radiocommunication sector), the ITU-T (the standardization sector),
and the 3GPP (3rd Generation Partnership Project between groups of telecommunications
associations that standardizes cellular wireless access) are working concurrently toward a
unified 5G New Radio standard. The 3GPP expects to complete Phase 1 by June 2018 with
Release 15, which will focus on the eMBB and URLLC use cases. Around the end of 2019,
Phase 2 will add functionality to 5G to support more services, large IoT deployments, and
much higher frequency bands beyond 52.6 GHz.
For now, the standards bodies have reached fundamental decisions with the Release 14 study
phase, including the current focus on non-stand-alone operation of NR to support only the
eMBB and URLLC use cases. NR does this by relying on existing 4G infrastructure, the EPC
(enhanced packet core) and the eNodeB acting as the Master cell, and the 5G gNodeB (gNB)
providing secondary access (based on the dual connectivity principle, as Figure 2 shows).
NON–STANDALONE (DEC 2017) STANDALONE (JUN, 2018)
Evolved Packet Evolved Packet

Core (EPC) Core (EPC)
Figure 2. Non-Stand-Alone Versus Stand-Alone Operation
ni.com/5g/nr
ni.com/5G
45
Furthermore, the IEEE has a 5G track that oversees the direction of future developments that
existing and new IEEE technologies will need to support 5G targets. These include 802.11ax and
802.11ay (WLAN), 802.15 (short range technologies), and 802.22 (fixed wireless broadband).
A key point to keep in mind is that Release 15 breaks compatibility with 4G standards. This is
similar to how 4G standards (LTE) departed from 3G wireless standards (UMTS). Yet, the
designers of the NR standard planned for future 5G releases to maintain compatibility with
the initial 5G NR specification. Because designers, industry experts, and the market haven’t
defined all the possible new uses for 5G technology, the 5G physical layer needs to be
flexible. This paper presents an introductory tutorial to the 5G physical layer and its
implementation to support the key 5G target applications of eMBB and URLLC.
PHY Design Considerations

Many researchers from industry and academia are actively working on addressing the
requirements for a robust and reliable 5G implementation. The following key features have
played a defining role in the 5G NR physical layer:
■■
Supporting a wide range of operation bands, a variety of channel bandwidths within those
bands, and multiple deployment options.
■■
Serving applications with very low latency, which requires short subframes and puncturing
and bursty interference from mission-critical transmissions.
■■
Sharing the spectrum dynamically to provision uplink (UL), downlink (DL), sidelink, and
backhaul links.
■■
Implementing multiantenna technology (multiple input, multiple output or MIMO) for larger
spectral efficiency.
■■
Maintaining tight time operation and more efficient frequency use for better time division
duplex (TDD) and frequency division duplex (FDD) deployments.
■■
Having symmetrical DL and UL requirements to enable operation at millimeter wave
(mmWave) frequencies of compact, low-cost base stations.
Waveforms for 5G NR
CP-OFDM: Downlink and Uplink
Recently, researchers have been investigating many different multicarrier waveforms and
proposing them for 5G radio access. However, because orthogonal frequency division
multiplexing (OFDM) implementations lend themselves well to TDD operation and delay-
sensitive applications, and because they have demonstrated successful commercial
implementation by efficiently processing ever-larger bandwidth signals, cyclic prefix (CP)
OFDM became the preferred choice for NR. The strong benefits of CP-OFDM that make it a
great fit for 5G implementation are:
High Spectral Efficiency—This essential feature of OFDM access helps meet the extreme
data rate needs, especially for backhaul links. Also, in future cases like vehicular
communication in dense urban environments, high spectral efficiency will help address
capacity constraints when many users broadcast periodically and asynchronously.
MIMO Compatibility—Both base stations and mobile devices will take advantage of MIMO
technology to implement spatial and frequency multiplexing with Single-User MIMO and
ni.com/5g/nr
ni.com/5G
46
Multiuser MIMO (MU-MIMO). MIMO deployments also overcome high propagation losses
and extend coverage range with beamforming.
Phase Noise Resistance—As the frequency of operation (and with it the oscillator phase
noise) increases, an OFDM system can minimize intersymbol interference due to phase
noise by applying larger OFDM subcarrier spacing (SCS).
Transceiver Simplicity—OFDM transceivers offer lower implementation complexity

compared with other waveforms that designers considered for 5G deployments. Having
worked with OFDM designs for several years, the wireless industry knows that their well-
understood operation and wide commercial deployment can enable 5G devices with
powerful OFDM baseband processing at lower prices.
Channel Time- and Frequency-Selectivity Resistance—With the right selection of SCS and
frequency of operation, an OFDM system can finish a transmission between devices in an
interval shorter than the channel coherence time and enable high-mobility (high-speed) and
high-data-rate scenarios while minimizing the effects of time selectivity. Also, as Figure 3
shows, with channel estimation and equalization techniques, OFDM waveforms demonstrate
great resiliency against frequency-selective channels.
CHANNEL
f f f
GENERATED OFDM RECEIVED OFDM EQUALIZED OFDM
WAVEFORM WAVEFORM WAVEFORM
Figure 3. Representation of an OFDM Waveform’s Frequency-Selectivity Resistance
Timing Error and Intersymbol Interference Resistance—Because of the CP, a receiver can
better tolerate synchronization errors and prevent the previous OFDM symbol from smearing
into the currently received OFDM symbol. Figure 4 shows two subsequent OFDM symbols,
each with a dedicated CP. The CP at the beginning of each OFDM symbol contains a copy of
the end of the OFDM symbol. When the receiver demodulates the signal, it operates on the
symbol after the CP (FFT window). This mechanism prevents intersymbol interference
between adjacent OFDM symbols
ni.com/5g/nr
ni.com/5G
47
OFDM SYMBOL 1 OFDM SYMBOL 2
CP FFT WINDOW CP FFT WINDOW
CYCLIC PREFIX CYCLIC PREFIX
Figure 4. A Cyclic Prefix Separates OFDM Symbols
With CP-OFDM, user equipment (UE) supports the following modulation schemes:
■■
QPSK
■■
16-QAM
■■
64-QAM
■■
256-QAM
DFT-S-OFDM: Higher Efficiency Uplink

One of the main drawbacks of OFDM waveforms is their high peak-to-average power ratio
(PAPR). As a result, RF output power amplifiers on transmitters lose efficiency and can’t
minimize high-order, nonlinear effects well. For UE such as smartphones, preserving battery
life and being energy efficient is important. The RF power amplifier that transmits the signal
to the base station consumes the most power within the mobile device, so system
designers need a type of waveform that promotes high-efficiency amplifier operation while
meeting the spectral demands of 5G applications. Although single-carrier waveforms have
very low PAPR and more efficient power amplifier operation, they don’t offer high spectral
efficiency and dynamic spectrum utilization, their compatibility with MIMO systems is lower,
and they are susceptible to frequency-selective channels.
For uplink, NR allows UEs to use CP-OFDM or a hybrid format waveform called discrete Fourier
transform spread orthogonal frequency division multiplex (DFT-S-OFDM). In DFT-S-OFDM, the
transmitter modulates all subcarriers with the same data. The right side of Figure 5 shows that
the first group of subcarriers (all red) takes the same amount of bandwidth as the OFDM
symbol on the left. The DFT-S-OFDM modulator maps the same data to all subcarriers but for
a shorter duration. It then maps the next data symbol (green) to all subcarriers for another
short interval. By the end of the equivalent OFDM symbol time, the transmitter sends the
same amount of data as it sends with an OFDM waveform by mapping the data symbols to
all subcarriers simultaneously but with shorter transmission intervals. This DFT-S-OFDM
waveform combines a lower PAPR with the multipath interference resilience and flexible
subcarrier frequency allocation that OFDM provides.
ni.com/5g/nr
ni.com/5G
48
PAPR: 11–13 dB HIGHER EFFICIENCY WITH

QPSK Modulating Data Symbols LOWER PAPR: 6–9 dB
-1,1 1,1 Transmit QPSK Data Symbols Sequence
1,1 -1,-1 -1,1 1,-1
-1,-1 1,-1
Data Symbols Occupy N x Subcarrier
Time Time
OFDM Symbol Duration OFDM Symbol Duration
Frequency
Frequency
Subcarrier Spacing
Symbol Width, e.g.: 60 kHz
e.g.: 15 kHz
Downlink: CP-OFDM Uplink: CP-OFDM and DFT-S-OFDM
Figure 5. OFDM Versus DFT-S-OFDM
With DFT-S-OFDM, UE supports the following modulation schemes:

■■
Pi/2-BPSK—note that this is a new modulation scheme in NR, and it requires new IP for
implementation
■■
16-QAM
■■
64-QAM
■■
256-QAM
Flexible Subcarrier Spacing and Symbol Lengths

The 3GPP intends for NR to operate in multiple frequency bands ranging from existing cellular
bands (below 1 GHz) to wider bands between 3 GHz and 5 GHz and up to the mmWave region
of the spectrum. Figure 6 illustrates the current bands defined for NR operation above 6 GHz.
26.25 29.5 40
3GPP: ALREADY DEFINED
24.25 27.5 37
29.5
KOREA
26.5
29.5 40 43.5
JAPAN
24.25 37 40.5
27.5 42.5
CHINA
24.75 37
28.35 38.6 40
US FCC 5G
27.5 37 38.6
24.25 43.5
EUROPE
27.5 40.5
FREQUENCY (GHz)
Figure 6. 3GPP-Defined and Locally Adopted Bands for NR in the Millimeter Wave Portion of the Spectrum
ni.com/5g/nr
ni.com/5G
49
As the carrier frequency increases, so does the system phase noise. For example, in the carrier
phase noise plot of Figure 7, the difference in phase noise between a carrier at 1 GHz and 28 GHz
is about 20 dB. This phase noise increase makes it difficult for a receiver to demodulate OFDM
waveforms with the narrow, fixed SCS and symbol duration of LTE at mmWave frequencies.
CARRIER PHASE NOISE SUBCARRIER WITH PHASE NOISE EVM
0
Phase Noise Power (dBc/Hz)
-20
-40
-60
-80
-100
-120
-140
10 1001 000 10000 100000 1000000
Offset Frequency
Figure 7. Effect of Phase Noise on Error Vector Magnitude
Additionally, the Doppler shift increases with carrier

0.
frequency, as shown in Figure 8. For
example, UE traveling at a speed of 60 km/h using a carrier frequency of 28 GHz sees a Doppler
shift of close to 1500 Hz, or 10 percent of a 15 kHz SCS. Because the channel coherence time, or
the time when the system can assume that the radio channel remains constant, is approximately
inversely proportional to the Doppler shift, it decreases as mobility increases. Therefore, at higher
carrier frequencies and higher speeds, the system has less time to measure the channel and
finish a single slot transmission.
DOPPLER SHIFT CHANNEL COHERENCE TIME VS. SPEED
3500.00 0.06
Doppler Frequency (Hz)
3000 .00 0.05

2500.00
0.04
2000 .00
0.03
1500.00
0 0.02
10 0.00
0.01
500.00
0.00 0
10 20 30 40 50 60 70 80 90 100 110 120 130 10 20 30 40 50 60 70 80 90 100 110 120 130
Speed (km/h) Speed (km/h)
Doppler 1 GHz Doppler 4 GHz Doppler 28 GHz Doppler 1 GHz Doppler 4 GHz Doppler 28 GHz
Figure 8. Doppler Shift and Channel Coherence
Also, phase noise and Doppler shift define the requirements for SCS to meet specific error
vector magnitude (EVM) criteria. That means using narrow SCS causes higher EVM because of
phase noise unless system designers implement the design with a high-quality local oscillator at
a high cost. Also, when SCS is small, the system performance can suffer because of Doppler
shift in high-mobility scenarios. On the other hand, selecting a large SCS results in excessive
channel bandwidth. Furthermore, given that SCS is inversely proportional to the OFDM symbol
duration, the OFDM symbol and CP length shortens as SCS increases and makes the system
ni.com/5g/nr
ni.com/5G
50
more susceptible to delay spread. Therefore, SCS should be as small as possible while providing
enough performance in the presence of phase noise and Doppler for a desired channel
bandwidth.
In the context of cellular standards, “numerology” refers to the physical layer SCS and symbol
length. The 3GPP standardized on a flexible numerology that scales the space between
orthogonal subcarriers, starting with the 15 kHz SCS of LTE. One of the fundamental reasons for
leveraging the exhaustive work already completed for LTE numerology was the ability of NR
deployments to coexist and be time-aligned with LTE networks during the first phases of
deployment. This gives LTE users a gradual path to adoption of the new technology.
The NR numerology scales according to the following formula1,2:
15 kHz SCS: 12 Subcarrier RB –> 180 kHz
*The NR specification also includes 240 kHz SCS.
Figure 9. Flexible SCS
The standard specifies that the smallest allocatable frequency unit consists of 12 subcarriers,
designated as a physical resource block (PRB). Consequently, the smaller the SCS, the
narrower the PRB, as shown in Figure 9.
Figure 10 illustrates the NR channel PRBs and guard bands.
ni.com/5g/nr
ni.com/5G
51
Channel Bandwidth (MHz)
Transmission Bandwidth Configuration

(Number of RB)
Transmission
Bandwidth (RB)
Channel Edge
Channel Edge
Resource Block
Active Resource
Blocks
Guardband, Can Be Asymmetric
Figure 10. NR Channel Divided Into Resource Blocks
A More Scalable and Flexible Frame Structure

Along with flexible SCS, 5G NR implements a flexible frame structure that ensures 5G
forward compatibility. It also minimizes design trade-offs for supporting key features like low
latency, coexistence with LTE, variable length transmissions, and TDD and FDD operation in
licensed and unlicensed spectra.
Slot Configurations Scale With SCS

NR slots have 14 OFDM symbols. A special case for 60 kHz SCS can have an extended CP
and 12 OFDM symbols. Since OFDM symbol duration has an inversely proportional
relationship with SCS, the duration of the slots scales down as SCS increases. Figure 11
shows the standard NR slot configurations.
ni.com/5g/nr
ni.com/5G
52
SLOT
14 OFDM Symbols
15 kHz
1 ms
SLOT
14 Symbols 30 kHz
500 µs
SLOT
14 Symbols
60 kHz
250 µs
SLOT
14 Symbols
120 kHz
125 µs
Figure 11. TDD Slot-Based Scheduling
The frame structure numbers the slots and groups them into subframes of 1 ms duration.
Ten 1 ms subframes form a complete NR frame. The number of slots within a frame also
varies with the choice of numerology, for example:
■■
Using 15 kHz of SCS results in a single 1 ms slot within the subframe, amounting to
10 slots per frame
■■
Using 30 kHz of SCS results in a subframe with two 500 µs slots within the subframe,
amounting to 20 slots per frame
OFDM 1ms
Symbol
CP Symbol 0 CP Symbol 1 CP Symbol 13 15 kHz SCS
Slot 02 13 26 27 30 kHz SCS
14 OFDM Symbols
01 23 45 55 65 75 2 3 4 5 60 kHz SCS
08192 34 5 67 1 1 1 1 1 1
0 1 2 3 4 5 120 kHz SCS
Subframe
1ms
0 1 2 3 4 5 6 7 8 9 Frame
10 Subframes - 10 ms
Figure 12. NR Frame Structure
ni.com/5G ni.com/5g/nr
53
The time and frequency resource structure defines the NR resource grid in Figure 13.
Depending on SCS, the resource grid changes as the number of available subcarriers and
OFDM symbols change. That is, for each numerology and carrier, NR specifies a resource grid
with a width given by the maximum number of resource blocks per SCS multiplied by the
number of subcarriers per resource block, and a length given by the number of OFDM
symbols per subframe.
T Slot
Slot
N OFDM Symbols
Resource Block
Channel Width
Subcarriers
Resource Element
Figure 13. NR Resource Grid
To support agile and efficient use of TDD resources, NR also implements a flexible slot
structure. The system can allocate a slot as all DL, all UL, or a mix of DL and UL to service
asymmetric traffic, as Figure 14 illustrates. DL control takes place at the beginning of the slot
and UL control at the end. The system can either configure the mixed DL/UL slot statically, as
in an LTE DL/UL TDD configuration, or change the allocation of the DL/UL mix dynamically for
better efficiency and scheduling based on traffic needs.
To accomplish this, the NR standard includes the slot format indicator (SFI), a field that
informs a user whether an OFDM symbol contains DL, UL, or flexible (either DL or UL) slots.
The SFI indicates the link direction for one or many slots by indexing a row of the UE’s
preconfigured table of possible link direction assignments.
ni.com/5g/nr
ni.com/5G
54
DL DL DL DL DL DL DL DL DL DL DL DL UL
DL–HEAVY TRANSMISSION
WITH UL PART
SLOT
DL UL UL UL UL UL UL UL UL UL UL UL UL
UL–HEAVY TRANSMISSION
WITH DL Control
SLOT
Figure 14. Flexible Slot Structure for Managing TDD Resources Dynamically
Furthermore, when the system needs to work with large payloads that don’t demand the
most immediate attention, the standard allows for slot aggregation. In the case of eMBB, for
example, having aggregated slots and longer transmission times meets application
requirements while reducing TDD switching and signaling overhead.
Minislots Enable Even Further Dynamic Scheduling

The NR standard is also considering the use of “minislots” to support bursty, asynchronous
transmissions with variable start positions and durations shorter than the typical, 14-symbol
slot. A minislot represents the smallest possible scheduling unit, and it can last for 7, 4, or 2
OFDM symbols. Minislots are especially important for enabling low-latency transmissions.
Imagine the future case of a mission-critical system (URLLC) that needs to communicate its
information with minimal latency, so the standard 10 ms frame is too long. The NR
numerology allows minislots to “puncture” an existing frame without waiting for the system
to schedule it. To avoid collisions, the network detects a minislot burst and manages the
URLLC device with the highest priority. Additionally, the network can schedule minislots
ahead of time, which becomes increasingly relevant in a mmWave operation where the
transmission of a few OFDM symbols mapped across large bandwidths might be enough
to carry smaller payloads. As of December 2017, the 3GPP has not yet fully specified this
feature, and it will not be part of Release 15.3
NR Reference Signals
To increase protocol efficiency, and to keep transmissions contained within a slot or beam
without having to depend on other slots and beams, NR introduces the following four main
reference signals. Unlike the LTE standard, which is constantly exchanging reference signals
to manage the link, an NR transmitter sends these reference signals only when necessary.4
1. Demodulation Reference Signal (DMRS)—The DMRS is specific for specific UE, and
a system uses this signal to estimate the radio channel. The system can beamform the
DMRS, keep it within a scheduled resource, and transmit it only when necessary in either
DL or UL. Additionally, multiple orthogonal DMRSs can be allocated to support MIMO
transmission. The network presents users with DMRS information early on for the initial
decoding requirement that low-latency applications need, but it only occasionally presents
this information for low-speed scenarios in which the channel shows little change.
Alternatively, tracking fast changes in high-mobility scenarios might increase the rate of
transmission (called “additional DMRS”).
2. Phase Tracking Reference Signal (PTRS)—As mentioned before, the phase noise of the
transmitters increases as the frequency of operation increases. The PTRS plays a crucial
role especially at mmWave frequencies to minimize the effect of the oscillator phase noise
on system performance. One of the main problems that phase noise introduces into an
ni.com/5g/nr
ni.com/5G
55
OFDM signal appears as a common phase rotation of all the subcarriers, known as
common phase error (CPE). The NR system typically maps the PTRS information to a few
subcarriers per symbol because the phase rotation affects all subcarriers within an OFDM
symbol equally but shows low correlation from symbol to symbol. The system configures
the PTRS depending on the quality of the oscillators, carrier frequency, SCS, and
modulation and coding schemes that the transmission uses.
3. Sounding Reference Signal (SRS)—As a UL-only signal, the SRS is transmitted by the UE
to help the system obtain the channel state information (CSI) for each user. This
information describes how the NR signal propagates from the transmitter to the receiver
and represents the combined effect of scattering, fading, and power decay with distance,
for example. The system uses the SRS for resource scheduling, link adaptation, Massive
MIMO, and beam management.
4. Channel State Information Reference Signal (CSI-RS)—As a DL-only signal, the CSI-RS
the UE receives is used to estimate the channel and report channel quality information
back to the gNB. During MIMO operations, NR uses different antenna approaches based
on the carrier frequency. At lower frequencies, the system uses a modest number of
active antennas for MU-MIMO and adds FDD operations. In this case, the UE needs the
CSI-RS to calculate the CSI and report it back in the UL direction.
MIMO
With the goal of using the spectrum more efficiently and serving more users, NR plans to
take full advantage of MU-MIMO technology. MU-MIMO adds multiple access (multiuser)
capabilities to MIMO by exploiting the distributed and uncorrelated spatial location of those
multiple users. In this configuration, the gNB sends the CSI-RS to UE in the coverage area,
and based on the SRS response of each UE device, the gNB computes the spatial location of
each receiver. The streams of data destined for each receiver go through a precoding matrix
(W-Matrix), where the data symbols get combined into signals streaming to each of the
elements of the gNB’s antenna array5 (see Figure 15.)
h00
h10
X0 r0 UE
S0
ESTIMATES S0
h01
W TX h11
X1 r1
UE
S1
ESTIMATES S1
gNB COMPUTERS PRECODING

W-MATRIX
Figure 15. Representation of MU-MIMO on the DL
ni.com/5g/nr
ni.com/5G
56
The multiple data streams have their own independent and appropriate weightings that apply
different phase offsets to each stream so that the waveforms interfere constructively and
arrive in phase at each receiver. This maximizes the signal strength at each user’s location
while presenting minimum signal strength (a null) in the directions of the other receivers,
as Figure 16 shows.
gNB MU-MIMO ANTENNA ARRAY
Signal Null
Signal Null
Maximum Directivity
and Signal Strength
Maximum Directivity
and Signal Strength
SPATIALLY MULTIPLEXED UE DEVICES

Figure 16. MIMO Beamforming for Spatial Multiplexing
Consequently, the gNB talks to multiple UE devices independently and simultaneously,

effectively multiplexing them in space. As an additional benefit, in this MU-MIMO
implementation, the UE devices don’t need any knowledge of the channel or additional
processing to get their data streams.
MU-MIMO on the DL boosts the NR system’s capacity. It scales with the minimum of the
number of gNB antennas and the sum of the number of UE devices multiplied by the number
of antennas per UE device. In other words, MU-MIMO can achieve MIMO capacity gains
with gNB antenna arrays and much simpler single-antenna UE devices.
Massive MIMO for 5G

Taking the MIMO approach a step further, a Massive MIMO configuration is implemented
when a system has many times more gNB antennas than the number of UE devices per
signaling resource. The large number of gNB antennas relative to the number of UE devices
can yield huge gains in spectral efficiency. Such conditions enable the system to serve many
more devices simultaneously within the same frequency band compared with today’s 4G
systems (see Figure 17). NI, along with industry leaders such as Samsung, continues to
demonstrate the viability of Massive MIMO systems through its platform of software defined
radio and flexible software for rapid wireless prototyping.
ni.com/5g/nr
ni.com/5G
57
Figure 17. Multiantenna Array for Massive MIMO
Currently, the strongest case for Massive MIMO operation is at frequencies below 6 GHz.
Spectrum is scarce and valuable in this region. In these bands, Massive MIMO systems can
achieve significant spectral efficiency by spatially multiplexing many terminals. The system
can also achieve superior energy efficiency by exploiting large antenna array gains to lower
the amount of power that each front end must handle.
In Massive MIMO systems, each antenna has its own RF and digital baseband chain. The
gNB maintains tight phase control and processes the signals from all antennas. The system
can gain a fuller picture of the channel response on the UL and respond quickly to changes in
the channel using digital processing. Massive MIMO operates mainly in TDD, which permits
the assumption of channel reciprocity. That enables the system to estimate DL channels from
UL pilots and eliminates the need for prior knowledge of the channel.
Another advantage of future Massive MIMO systems is that they’ll provide better and more
consistent service to all UE in a coverage area. Because of an improved link budget and the
ability to place target UE precisely within the radiated beam while nulling nontarget UE
(spatial resolution), power control algorithms can achieve greater fairness among the UE.
User mobility can limit how well Massive MIMO solutions scale up in performance. For
proper channel estimation, the system needs to send UL pilots and payload in the UL
direction. The faster UE moves, the shorter the channel coherence time. For example, in
large coverage areas with fast UE, such as a car traveling at 120 km/h on a highway, the
channel’s coherence time at 2 GHz carrier frequency drops to around 1 ms. This requires the
system to recalculate the channel 1000 times per second to track the UE as it moves and
limit the multiplexing gain to a smaller number of terminals. Figure 18 shows how coherence
time scales with both carrier frequency and UE speed. Conversely, in more controlled
environments with little or no mobility such as fixed wireless access, the system can
accommodate hundreds of terminals through spatial multiplexing using narrow beams.
ni.com/5g/nr
ni.com/5G
58
CHANNEL COHERENCE INTERVAL
Uplink Pilots Matrix Computation Downlink Symbols
Frequency 3 km/h 30 km/h 120 km/h 500 km/h

2 GHz 45 ms 4.5 ms 1.125 ms 27 µs
28 GHz 3.2 ms 320 µs 80 µs 19 µs Limited Mobility
Figure 18. Effect of Mobility on Channel Coherence Time
mmWave for 5G
Industry and academic researchers consider available mmWave bands as the next frontier
to serve the data-hungry wireless applications of the future. New 5G systems operating at
28 GHz and above offer more available spectrum for larger channels, which work well for
multi-Gbps links. Although these frequencies see less spectral crowding than those below
6 GHz, they experience different propagation effects such as higher free-space path loss
and atmospheric attenuation, weak indoor penetration, and poor diffraction around objects.
To overcome these undesired effects, mmWave antenna arrays can focus their beams and take
advantage of the antenna array gain. Fortunately, the size of these antenna arrays decreases as
the frequency of operation increases, allowing a mmWave antenna array with many elements
to take up the same area that a single sub-6 GHz element occupies (Figure 19).
SINGLE-ELEMENT PATCH ANTENNA 64-ELEMENT ANTENNA ARRAY (30 GHz)
Omnidirectional Highly Directive

Radiation Pattern Radiation Pattern
60 mm
60 mm
60 mm 60 mm
Figure 19. Comparison of mmWave and Sub-6 GHz Antenna Arrays
ni.com/5g/nr
ni.com/5G
59
Analog Beam Steering to Manage Complexity

As presented above, Massive MU-MIMO systems require far more transmit RF chains than UE
devices for proper spatial multiplexing. This differs from a system comprising just one RF chain
that feeds many antennas, the phase of which is controlled analogically to focus and steer the
radiation pattern (see Figure 20). For MU-MIMO purposes, such a system can be categorized
as a single-antenna terminal that happens to have a directive, steerable antenna.
ANALOG PHASE CONTROL
Baseband Chain RF Chain
NA
PHASE SHIFTER
Figure 20. Single RF Chain With Analog Beam Steering
One of the main drawbacks of Massive MIMO systems is the high complexity and cost
of integrating and deploying a massive number of RF chains, especially at mmWave
frequencies. Researchers have proposed several hybrid (digital and analog) beamforming
alternatives6 to allow 5G gNBs to maintain a high number of antennas while keeping the
MU-MIMO implementation costs down. Figure 21 illustrates a hybrid system with a common
baseband processing stage that feeds multiple data streams to their corresponding RF chains.
These streams undergo digital beamforming signal processing before moving on to the analog
stage. At this last stage, the system applies beam steering with analog phase shifters, which
focus the beam toward a specific direction. This results in spatially multiplexed RF streams
contained within a directional beam.
ni.com/5g/nr
ni.com/5G
60
ANALOG PHASE CONTROL
RF Chain
NA
Baseband Processing
W-Matrix
ND
RF Chain
NA
PHASE SHIFTER
Figure 21. Hybrid Digital and Analog Beamforming
Finally, recall that the channel coherence time decreases significantly at mmWave frequencies,
which places tough restrictions on mobility applications. Researchers continue to investigate
new ways to improve UE mobility at mmWave frequencies, but most likely the first 5G
mmWave deployments will serve fixed wireless access applications such as backhaul and
sidelink.
Gaining Access and Managing Beams

Managing the large signal propagation loss at frequencies above 20 GHz is one of the main
technical challenges of operating in mmWave bands. In practical terms, this loss reduces the
possible cell coverage area and range. To compensate for it, the standard designers settled on
beamforming technology with antenna arrays as a way of focusing the RF energy toward
individual users and boosting the signal gain. However, UE can no longer rely on the mmWave
gNBs to broadcast omnidirectional signals to establish the first connection.
The NR standard implements a new procedure for UE to gain initial access to the gNB. Upon
arrival to a new cell coverage area, UE is blind to the location of the beam, ignoring the
direction in which the gNB is currently transmitting to begin the network access procedure.
ni.com/5g/nr
ni.com/5G
61
The NR initial access procedure presents an elegant solution for UE to establish

communication with the gNB. It solves the problem of finding the gNB in the dark not only for
mmWave operation but also for sub-6 GHz omnidirectional operation. This means that the initial
access procedure must work in single-beam and multibeam scenarios. It also must support NR
and LTE coexistence.
The procedure follows the steps depicted in Figure 22.
gNB UE
Synchronization Signals
Beam-Sweeping
Transmission Basic System
Information for All UEs
Random Access Channel

Beam-Sweeping
Reception
Random Access
Response and System
Information
UE-Specific
Selected Beam
Data and
Control Channels
UE-Specific
Beamforming
Figure 22. Initial Access Procedure
1. Beam-sweeping transmission—The gNB transmits the physical broadcast channel

(PBCH) in groups of four OFDM symbols called synchronization signal blocks (SS blocks)
sequentially in multiple directions, as depicted by the blue, green, and yellow beams, and
maps each one to a different spatial direction. Using the concept of beam sweeping, the
gNB transmits both the synchronization signals and the system configuration information
that the UE needs to access the network.
2. Beam-sweeping reception—The UE detects the best SS block (strongest detected beam)
by listening until it matches the beam direction of the transmitter. This allows the UE to
decode the best SS block and extract its time index. Knowing when the gNB will use that
beam direction again, the UE transmits to the gNB on the physical random access channel
(PRACH) at the right time. The gNB now knows in which direction and at what time the UE
will transmit UL information.
3. UE-specific selected beam—Once the UE and gNB establish communication on the best
beam, the gNB sends the rest of the system information that the UE needs to set up a
connection with the gNB.
4. UE-specific beamforming—At this point, the system can switch from general, wider beam
coverage to UE-specific coverage with a narrower beam using beam-refining procedures.
ni.com/5g/nr
ni.com/5G
62
Bandwidth Parts
In future 5G applications, a large range of devices and equipment will have to operate
successfully across many different bands with varied spectrum availability. An example is UE
with limited RF bandwidth needing to operate alongside a more powerful device that can fill a
whole channel using carrier aggregation and a third device that can cover the whole channel
with a single RF chain7 (Figure 23).
Wideband UE
Carrier Aggregated UE
Narrowband UE
Resource Block
NR Carrier From the Network Perspective
Figure 23. Parts to Manage the Spectrum More Efficiently
Though wide bandwidth operation has a direct effect on the data rates that users can
experience, it comes at a cost. When UE doesn’t need high data rates, wide bandwidth leads
to inefficient use of RF and baseband processing resources.
To address this, the 3GPP developed the new concept of bandwidth parts (BWPs): the network
configures certain UE with one wideband carrier and separately configures other UE with a set
of intraband contiguous component carriers using carrier aggregation. This allows for a greater
diversity of devices with varying capabilities to share the same wideband carrier. This kind of
flexible network operation that adjusts to UE’s differing RF capability does not exist in LTE.
A BWP consists of a group of contiguous PRBs. Each BWP has an associated SCS and CP
(numerology). As a result, the system can use the BWP to reconfigure UE with a certain
numerology. UE starts out with a default active BWP during the initial access until the system
configures the UE’s BWPs explicitly during or after connection establishment. Figure 25 shows
that the network is allocating two BWPs (BWP 1 and 2) to one UE device while reserving a
third, full-channel, overlapping BWP (BWP 3) for potential use by another higher bandwidth
UE device or application.
ni.com/5g/nr
ni.com/5G
63
BWP 3
BWP 1 BWP 2
Numerology 1 Numerology 2
Figure 24. Bandwidth Parts
The system can configure DL and UL BWPs for each serving cell separately and independently.
In Release 15, only one BWP in DL and one in UL are active at any point in time, but the UE
can have up to four configured BWPs.
To summarize, NR will have the flexibility to serve many different use cases effectively by using
BWPs, for example:
PRB
Supporting UE that has

narrow RF capabilities
and reducing energy
BWP
consumption when a
device doesn’t require
full bandwidth operation
BWP 3
Supporting multiple
numerologies and allowing
BWP 1 BWP 2
operation in noncontiguous
spectrum
Numerology 1 Numerology 2
Enabling forward
compatibility with devices
TBD Future
and applications that the BWP
Application
market will define in
the future
Figure 25. NR Serving Many Use Cases With BWPs
ni.com/5g/nr
ni.com/5G
64
Conclusion: Comparing LTE and 5G NR PHY

Now compare the fundamental technical features of 5G NR with those of current LTE
implementations.
Better Spectrum Utilization

Wideband 5G carriers are planned to occupy up to 98 percent of the channel, avoiding large
guard bands between carriers. This helps reduce channel overhead and allows for faster load
balancing than LTE aggregated carriers can implement. For example, Figure 27 compares five
20 MHz aggregated LTE carriers versus a proposed single 98 MHz 5G NR carrier.
LTE 5x20 MHz
100 MHz
18 MHz 18 MHz 18 MHz 18 MHz 18 MHz
5G 100 MHz
100 MHz
Up to 98 MHz
Figure 26. Improved Channel Utilization With Wideband 5G Carriers
Flexible Numerology and Frame Structure

LTE uses fixed 15 kHz SCS with a maximum of 1200 subcarriers in a 20 MHz channel. In
contrast, NR allows for greater spectrum utilization with channels of various sizes, variable
SCS and slot length, and a maximum of 3300 subcarriers per channel.
Enhanced Efficiency With Leaner Signaling

Unlike LTE, which transmits cell-specific reference signals four times per millisecond,
synchronizes every 5 ms, and broadcasts every 10 ms, 5G has no cell-specific reference
signals and synchronizes and broadcasts every 20 ms. This enables greater base station
power savings.
ni.com/5g/nr
ni.com/5G
65
LTE
Synchronization Broadcast LTE Reference Signals
20 ms
5G NR
Figure 27. Signaling Efficiency in LTE Versus 5G NR
Manage TDD Resources Dynamically

LTE has a fixed, static TDD structure that allocates slots to either DL, UL, or synchronization
and control signals. That is, within a radio frame, LTE TDD switches multiple times between
DL and UL transmission and vice versa. On the other hand, within a slot, 5G can change
dynamically between DL and UL to handle traffic demands in either direction.
Download Guard Uplink LTE
Slot 0 Slot 1 Slot 2
Download UL DL DL UL 5G
Slot 0 Slot 1 Slot 2
Figure 28. NR Manages TDD Resources More Dynamically
Operation at mmWave Frequencies With Wider Channels

Today’s licensed LTE networks are limited to operating at a maximum frequency of around
3800 MHz. The 5G NR networks will take advantage of both existing cellular bands and wide
channels in newly licensed spectrum around 30 and 40 GHz.
The specification of higher bandwidth channels and multiple numerology options will enable
NR systems to operate in sub-6 GHz bands and mmWave bands with appropriate handling of
multipath delay spread, channel coherence time, and phase noise. Furthermore, NR will
support existing and new services with even higher data rates and address different latency
and mobility requirements by changing the transmission turnaround time using variable SCS
and by allocating wide channels around and above 28 GHz. NR will use the latest
ni.com/5g/nr
ni.com/5G
66
developments in Massive MIMO and beamforming technology to maximize spectral

efficiency and guarantee better service for a larger number of users.
Additionally, considering the commercial practicalities of deploying different UE with different

RF capabilities, the new BWP concept in NR will lead to more energy-efficient UE operation
and superior spectrum management.
In conclusion, 5G wireless technology promises to deliver an abundance of reliable, data-rich,

and highly connected applications for more of the world’s population. Although deploying an
infrastructure that can support this and creating the next generation of 5G devices present
significant design and test challenges, NI’s platform-based approach to designing, prototyping,
and testing wireless technologies will be key in making 5G a reality within the next decade.
Glossary
3GPP Third Generation Partnership Project
AR augmented reality
BCH broadcast channel
BPSK binary phase-shift keying
BS base station
BWP bandwidth parts
CP cyclic prefix
CP-OFDM cyclic prefix orthogonal frequency division multiplexing
CSI-RS Channel State Information Reference Signal
DFT discrete Fourier transform
DFT-SOFDM discrete Fourier transform spread orthogonal frequency division multiplexing
DL downlink
DMRS demodulation reference signal
eMBB Enhanced Mobile Broadband
EPC Evolved Packet Core
EVM error vector magnitude
FDD frequency division duplex
FDMA frequency division multiple access
FFT fast Fourier transform
GP guard period
ni.com/5g/nr
ni.com/5G
67
gNB g node b, a 5G base station
mMTC Massive Machine-Type Communication
mmWave millimeter wave
MIMO multiple input, multiple output
MU-MIMO Multiuser MIMO
NR New Radio
OFDM orthogonal frequency division multiplexing
PA power amplifier
PAPR peak-to-average power ratio
PBCH physical broadcast channel
PRACH Physical Random Access Channel
PRB physical resource block
PTRS phase-tracking reference signal
QAM quadrature amplitude modulation
QPSK quadrature (quaternary) phase-shift keying
RAN radio access network
SCS subcarrier spacing
SRS Sounding Reference Signal
TDD time division duplex(ing)
TDM time division multiplexing
UE user equipment
UL uplink (reverse link)
URLLC Ultra-Reliable Low-Latency Communication
VR virtual reality
ni.com/5g/nr
ni.com/5G
68
Endnotes
1
3GPP TS 38.101-1 V15.0.0 (2017-12) Technical Specification Radio Access Network; NR; User Equipment (UE) radio transmission
and reception.
2
3GPP TS 38.211 V15.0.0 (2017-12) Technical Specification Radio Access Network; NR; Physical channels and modulation.
3
3GPP TR 38.912 V14.1.0 Technical Specification Group Radio Access Network; Study on New Radio (NR) access technology
(Release 14.)
4
3GPP TS 38.214 V15.0.0 (2017-12) Technical Specification Radio Access Network; NR; Physical layer procedures for data.
5
Q. H. Spencer, C. B. Peel, A. L. Swindlehurst, and M. Haardt, “An introduction to the multi-user MIMO downlink,” in IEEE
Communications Magazine, vol. 42, no. 10, pp. 60–-67, Oct. 2004.
6
S. Han, C. l. I, Z. Xu, and C. Rowell, “Large-scale antenna systems with hybrid analog and digital beamforming for millimeter wave
5G,” in IEEE Communications Magazine, vol. 53, no. 1, pp. 186–194, January 2015.
7
3GPP TS 38.213 V15.0.0 (2017-12) Technical Specification Radio Access Network; NR; Physical layer procedures for control.
©2018 National Instruments. All rights reserved. National Instruments, NI, and ni.com are trademarks of National Instruments. Other product and company names listed are trademarks
or trade names of their respective companies. 32178
ni.com/5g/nr
ni.com/5G
69
TODAY’S
SMART DEVICES
DEMAND SMARTER
SEMICONDUCTOR
TEST SYSTEMS
NI’s Semiconductor Test System (STS) meets the test needs of today
and tomorrow by providing engineers with an open, software-centric
platform based on PXI instrumentation, TestStand test management
software, and LabVIEW code module development. With lab-grade
measurement quality and production-grade performance, the STS
accelerates RF and mixed-signal tests from the characterization
bench to the production floor, which lowers test time and cost.
Prepare for the future at ni.com/sts.
©2017 National Instruments. All rights reserved. LabVIEW, National Instruments, NI, ni.com, and NI TestStand are trademarks of National Instruments. Other product and company names listed are trademarks or trade names of their respective companies. 29944

5G RF Front End Design NI-SEPT-2018-EBOOK-v3 PDF

Uploaded by

Copyright:

Available Formats

5G RF Front End Design NI-SEPT-2018-EBOOK-v3 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

5G RF Front End Design NI-SEPT-2018-EBOOK-v3 PDF

Uploaded by

Copyright:

Available Formats

E-BOOK

5G RF Front End Design

4 Design of a Single Chip Front-End Module for 28 GHz 5G

17 Basics of Power Amplifier and Front End Module

43 5G New Radio: Introduction to the Physical Layer

5G RF Front End Design and Measurement

Pat Hindle, Microwave Journal Editor

functionality of the FEM MMIC is device sizes, bias points, imped-

at the pin labelled “PA_RFin.” The was aimed at identifying how to

 Fig. 7 Comparison of measured and ear amplification to preserve modula-

Power at IMD3 = 35 dBc_meas

Gain & Noise Figure (dB)

500 m ISD ~333 m Cell Range

Total LNA Pdc (W)

T Array R Array T/R Array Dual-Polarization T/R Array

1:N Splitter 1:N Combiner

Transitioning From Separate Arrays Integrated T/R Integrated T/R and

 Fig. 14 Active array using hybrid beamforming.

Antenna Array Gain (dBi)

two-stage and three-stage designs to provide sufficient

 Fig. 17 Integrated 39 GHz GaN front-end MMIC –

Gain and Output Power

INPUT VERSUS OUTPUT POWER

Pout Ideal Behavior

Gain = Pout – Pin = 18 dB

+10 dBm Pin

Figure 2. Input versus Output Power in a Typical PA

MEASURING THE COMPRESSION POINT WITH A RAMP SIGNAL

VECTOR SIGNAL GENERATOR

VECTOR SIGNAL ANALYZER (SIMPLIFIED)

Fast Power Level Servoing Using the NI Vector Signal Transceiver

IMPROVING MISMATCH UNCERTAINTY

Calibrating Power Measurements with a Power Meter

Step 1: Calibrate the signal generator with the power meter.

VSG Power Meter

Step 2: Calibrate the signal analyzer with the signal generator.

Figure 5. System calibration is a two-step process that uses a power meter

Measuring Gain with a Vector Network Analyzer

AVOIDING PORT SATURATION

Figure 6. Use an attenuator when measuring PA gain

UNDERSTANDING THE REFERENCE PLANE

Figure 7. The VNA calibration reference plane must extend

Return Loss and Reverse Isolation

Fast S-Parameter Measurements in Production Test with STS

S11 S12 S22

Figure 8. You can measure reverse isolation and return

of the large-signal performance of a PA requires a large-signal network analyzer (LSNA), or

BASIC LOAD-PULL TEST CONFIGURATION

Figure 9. Block Diagram of Basic Load-Pull Test Configuration

INTRINSIC NOISE POWER OVER THERMAL NOISE

Signal Power Signal Power

Noise Power Thermal Noise Noise Power

Equation 2. Formal Definition of Noise Factor

Noise Unit Conversion

NF = 10 log10 (F) (3)

Equation 5. Noise Temperature as a Function of Noise Factor

In Equations 5 and 6, T0 is an expression that commonly refers to room temperature, or 290 K.

Figure 11. Each component can be described

Equation 7. The Friis Formula for Noise Factor of a Cascaded System