Circuits For 5g RF Frontend Modules

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

International Journal of Circuits for 5G RF front-end modules

Microwave and Wireless


Technologies
Florinel Balteanu

cambridge.org/mrf Skyworks Solutions Inc., 5260 California Avenue, Irvine, CA 92617, USA

Abstract
Worldwide adoption of fourth-generation wireless (4G) long-term evolution (LTE) smart-
phones and the actual transition to fifth-generation wireless(5G) is the main driving engine
Industrial and Engineering for semiconductor industry. 5G is expected to reach high data rate speeds (1 Gbps) and
Paper low latency (<1 ms). 5G requires more RF bandwidth and therefore an increase in the number
of components such as RF switches, acoustic filters, and power amplifiers integrated in few RF
Cite this article: Balteanu F (2023). Circuits for
5G RF front-end modules. International Journal x of front-end modules. Also, there is an increase in the number of RF radio transmitters and
Microwave and Wireless Technologies 15, 909– 924. receivers operating at the same time. This paper presents new architecture elements for 5G
https://doi.org/10.1017/S1759078722001295 RF front-end modules. Circuit details and measurements are presented to reduce the RF
noise, improve the efficiency, and help the coexistence of multiple radio transmitters. These
Received: 19 July 2022
Revised: 27 October 2022 circuits will help extend the 5G applications with radar-like sensing applications, remote
Accepted: 28 October 2022 medicine, and autonomous driving.

Key words:
Envelope tracking; fifth generation wireless Introduction
(5G); load modulation; power amplifier; silicon
on insulator (SOI) The continuous need for high data rate in smartphones and mobile devices and the arrival of
new applications is driving the transition from 3G/4G long-term evolution (LTE) and adop-
Author for correspondence:
Florinel Balteanu,
tion of 5G LTE [1]. To provide the increased data rate 5G is using sub-6 GHz bands [2–5]
E-mail: [email protected] and millimeter-wave (mmWave) spectrum [6]. 5G together with other RF technologies such
as Bluetooth, WiFi6 [7], licensed assisted access (LAA), enhanced licensed assisted access
(eLAA), and ultra-wideband (UWB) and sensing and computation techniques will enable
new applications; for example, remote medicine (eHealth), vehicle-to-vehicle communica-
tions (V2X), factory automation, self-driving vehicles, and internet of things (IoT). 5G is
a revolutionary technology step that will make possible new applications which require
high data rate, ultra-reliable and low latency communications (URLLC) [8, 9], and massive
connectivity to a computing cloud. With the adoption of 5G the smartphone device will
become a fast and reliable conduit between smart sensors, actuators, and cloud computing.
5G smartphones need to carry over the legacy voice (2G/3G) and need to integrate sub-6
GHz bands as well as add WiFi6 and 4 × 4 multiple input, multiple output (MIMO), to pro-
vide high data rate and seamless transition to 5G. These additions and the requirement for
one/two stock keeping units (SKUs), add a lot of pressure to keep a balance between
increased functionality and additional cost and size associated. The 5G new radio (NR)
requirements are specified for extended bands compared with 3G/4G as presented in
Fig. 1 for under 6 GHz spectrum.
The new 5G bands provide more bandwidth (BW) but also come with more technical chal-
lenges regarding the integration and RF interferences. Recently the 3rd Generation Partnership
Project (3GPP) standard committee extended the 5G millimeter-wave bands up to 71 GHz
[10] with up to 2 GHz modulation BW and downlink (DL) carrier aggregation (CA) as pre-
sented in Table 1.
With the introduction of the mmWave bands for 5G, the frequency bands are divided into
two frequency ranges:

• Frequency range 1 (FR1) includes all existing and new bands and corresponds to 450 MHz–
6 GHz; sub-6 GHz bands. These bands have been used for initial 5G deployment and carry
the most 5G traffic right now.
© The Author(s), 2023. Published by • Frequency range 2 (FR2) includes new bands and corresponds to mmWave bands 24.25–71
Cambridge University Press in association with GHz. Just several smartphone models carry mmWave modules due to size, power consump-
the European Microwave Association. This is tion, and extra cost.
an Open Access article, distributed under the
terms of the Creative Commons Attribution
licence (http://creativecommons.org/licenses/ A typical RF front-end module (RFFE) [1–5] for 5G mobile devices (Fig. 2) has 6–8 antennas
by/4.0/), which permits unrestricted re-use, for under 6 GHz space (5G, WiFi) and an 8/16 channel mmWave module.
distribution and reproduction, provided the The main goals for 5G NR are to achieve low latency and date rates of 1 GB/s. The capacity
original article is properly cited.
of a wireless system is determined by the Shannon’s formula as:


k  
e n Sk
C = Bw log2 1 + . (1)
k=1
Nx + Ik

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


910 Florinel Balteanu

Fig. 1. 4G/5G and WiFi6 spectrum – FR1.

To achieve higher capacity these are the techniques which are Mobile Communication (2G GSM) and data (3G/4G/5G)
incorporated in 5G: increases the linearity requirements for multiple Tx/Rx path
operating at the same time.
• Increase channel bandwidth Bw; e.g. 100 MHz LTE. • Intra-coexistence with the actual 3G/4G bands in 5G frequency
• Increase spatial multiplexing level k through MIMO. division duplexing (FDD) and time division duplex (TDD)
• Increase the transmit power (Tx) and Sk; as 26 dBm for high re-farmed bands [1, 2]. 5G NR, as name suggests, provides
power user equipment (HPUE) in 5G. more data capacity, and allows subscribers to use the 5G net-
• Decrease noise Nx and improve receiver (Rx) sensitivity. works. There are also existing 3G and 4G radios and therefore
• Reduce in-band interference Ik on link k, especially in multiple we must have design considerations to avoid interferences and
uplink (UL) Tx such as CA and MIMO. to provide handover mechanism so that the user equipment
• Higher order quadrature modulation (QAM) such as 256 QAM (UE) can access any service available based on their device cap-
for UL; this requires an error vector magnitude (EVM) lower abilities and subscription. 5G non-standalone (NSA) provides
than 3/5% [2, 3]. capacity where 5G NR sites can connect to existing 4G Core
• Increase signal Sk through use of envelope tracking (ET) – en factor. (4GC). 5G standalone (SA) provides capacity where 5G NR
sites can connect only to 5G Core (5GC). From RF circuit per-
Low latency is required for real-time applications such as self- spective the challenges are reflected in higher requirements for
driving and autonomous cars, telemedicine, factory automation linearity and noise performance.
as well as smart machines communicating in real time. There • Maintain at least −38 dBc adjacent channel power ratio (ACPR)
are significant challenges for 5G RFFE such as: or adjacent channel leakage ratio (ACLR); both terms are used
interchangeably.
• 5G mobile devices need 6–8 antennas which must be reached
from different 4G/5G LTE radios which must coexist with mul- Low latency is the prerequisite for real-time applications such as
tiple WiFi6 radios, Bluetooth, GPS, and UWB. self-driving and autonomous cars as well as smart machines com-
• 5G introduced dedicated bands for sub-6 GHz such as n77/n78 municating in real time. Reliable high data rate and low latency
(3.3–3.4 GHz), n79 (4.4–4.5 GHz), and eLAA bands B46 and are important in connecting smart devices in 5G. With the tran-
B47 (5.15–5.92 GHz) [1, 5]. sition from 4G to 5G one of the key enablers is the subcarrier spa-
• Wider channel BW up to 100 MHz for FR1. cing (SS) which is no longer fixed at 15 kHz and is flexible to
• Very wide channel BW up to 2 GHz for FR2 [10]. cover different latency requirements and frequency ranges, SS is
• 5G NR introduces high peak to average power ratio (PAPR) from 15 to 60 kHz for lower frequency bands and from 60 to
waveforms; PAPR is 10.5 dB for UL and 64/256 QAM modula- 240 kHz for higher frequency bands. This will allow the use of
tion BW [1]. short mini slots which will be useful for low latency applications
• The need for dual-sim operation for voice under second gener- as well as LTE 5G NR coexistence and beamforming. The theor-
ation of cellular networks, based on the Global System for etical data rate (DR) [2] is given by the formula:
Table 1. 5G millimeter-wave spectrum – FR2 DR = ns · m · (ncc · nsc · rb) · nss · nsl · ovh · tddov , (2)
5G mmWave Band duplex UL/DL low UL/DL high
bands type (MHz) (MHz) where ns represents the number of bits per symbol (8 bits for 256
QAM), m represents the number of MIMO data streams, ncc
n257 TDD 26 500 29 500 represents number of component carriers for CA, nsb represents
n258 TDD 24 250 27 500 number of sub-carriers, rb represents number of resource blocks
(RBs), nss represents number symbols per slot, nsl represents num-
n259 TDD 39 500 43 500
ber of slots, ovh (in percentage) is the overhead required for con-
n260 TDD 37 000 40 000 trol and coding, and tddov represents TDD duty cycle. Using this
n261 TDD 27 500 28 350 formula for 5 × 20 MHz CA streams, 4 × 4 MIMO, and 256 QAM,
the DR (Fig. 3) is
n262 TDD 47 200 48 200
n263 TDD 57 000 71 000 DR = 8 · 4 · (5 · 12 · 100) · 7 · 2000 · 75% · 60% = 1.2 Gb/s. (3)

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 911

Fig. 2. 5G RF front-end structure for smartphones.

Fig. 3. 5G downlink/uplink data rate.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


912 Florinel Balteanu

Fig. 4. UWB/WiFi6 and 5G MHB/UHB MIMO LTE structure.

This DL data rate can be obtained using a 4 × 4 MIMO struc- interferences through conductive and radiated paths.
ture with the new 5G mid-high (MHB) bands/ultra-high (UHB) Intermodulation products are determined by all the RF Tx as
bands (n41, n77, n78, and n79). These bands can accommodate well as all the other digital clock-related activity in a front-end
100 MHz channel modulation BW and 64/256 QAM as presented module (FEM) such as charge pumps, digital mobile industry
in Fig. 4. Data speeds of 2.5 GB/s have been demonstrated recently interface (MIPI) serial parallel interface (SPI) clocks, etc. One of
using multiple carriers in FDD and TDD bands [11]. the main design challenges for 5G FEM is how to manage all
For the majority of smartphone use, the DL is asymmetric these RF interferences. Also due to different Rx/Tx configurations
compared with UL, but to accommodate a high DL (1 Gbps) as between DL and UL and due to high power Tx capabilities for
in 5G, assuming 7–10% UL rate for acknowledge/sync signals, base-stations (40 dBm), the 4G/5G RF UL transmission is the
5G needs UL speeds around 100 Mb/s. Due to increases in the weak link. This becomes more an issue for higher BW and the
use of higher 5G carrier frequencies for 5G such as bands n41, new 5G MHB/UHB bands.
n42, n77, n78, and n79 and higher path loss for these frequencies, There are two basic deployment scenarios for 4G transition to
5G has adopted HPUE which increases the RF power delivered at 5G networks: 5G SA deployment and NSA deployment. For NSA
the antenna to 26 versus 23 dBm for 3/4G. This will allow 19% deployment the UE supports dual connectivity (DC) for 4G LTE
increase in cell coverage radius (42% increase in the base-station and 5G NR. DC will combine the coverage advantage of existing
coverage area) as shown in Fig. 5. 4G LTE networks with the higher DR throughput and latency
All these requirements for 5G determine that more than one advantages of 5G NR. NSA enables 5G NR in smartphones
Tx in UL together with several Rx in DL as well as multiple radios with a smooth evolution from 4G and will be the key to
for WiFi/WiFi6 (2.4/5 GHz), Bluetooth (2.4 GHz), GPS (1.17/1.5 mmWave mobility. Sub-6 GHz anchor will be needed for roaming
GHz), UWB (6–8 GHz), and smart narrow field communications and handovers as shown in Fig. 6. This anchor band is usually a
(13.56 kHz) might be active. All these radios will create RF low band (LB) (450–900 MHz) FDD band due to lower propaga-
tion loss [12]. The path loss (PL) from a 15 m height baseband
antenna to UE [1, 12] is determined by

PLdB = 10 log10 R4 + 21 log10 f + K, (4)

where R is distance and f frequency of operation. PL is dependent


on 1/R 4, and for different propagation scenarios for outdoor–
indoor, the loss is dependent on 1/f 3.
The DC framework allows FR2 network procedures for beam
management that are faster than the SA handover for FR2.
Especially for FR2, each UE directionally broadcasts a sounding
reference signal (SRS) that continuously sweeps the angular
space. Each 5G FR2 cell monitors the strength of the received
SRS, building a table based on the channel quality of each receiv-
ing direction. The optimal FR2 cell and beam configuration are
selected for each UE considering the absolute maximum
signal-to-noise plus interference ratio. Using FDD bands for
Fig. 5. Path loss for 3GPP cellular frequencies. FR2 handover will reduce the signaling and calibration time.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 913

Fig. 6. Cellular structure for 4G/5G with dual connectivity (DC).

For FDD there is no guard time between Tx and Rx which limits proliferation and the transition to 5G the weight in terms of cost
the capacity and there is no discontinuous transmission required and size is determined by the number and type of acoustic filters,
for UL and DL. Also, the base-stations need to be synchronized such as surface acoustic wave (SAW) and bulk acoustic wave (BAW).
with respect to the UL and DL transmission times. If neighboring The acoustic filters are placed with other components such as PAs,
cells use different UL and DL assignments and share the same chan- SOI switches, and low noise amplifiers (LNAs) into an FEM with
nel, then interference may occur between cells. The best latency in acoustic duplexers/filters (FEMiD). With the increase of the number
this process is obtained using LB FDD bands for DC [9, 11]. of antennas and the requirements such as each antenna to be
reached by multiple PAs and FEMs, the RF path loss to the anten-
nas, on the mobile device, becomes higher for 5G as presented in
5G front-end module architectures
Fig. 7. The RF loss can be 4–5 dB, and this RF loss is higher than
The smartphone market is very dynamic and the move to 5G the typical loss for 3G/4G (3 dB). Unfortunately, this RF loss must
requires coverage for more than 50 LTE bands from 500 MHz to be compensated with higher RF power delivered by the PA and
6 GHz [13]. These bands cover the entire world with very few increased SOI RF switch size to maintain the linearity.
mobile SKUs. A lot of research has been done for single die Sub 3 GHz bands provide primary LTE cellular coverage, and
power amplifier (PA) in different technologies such as silicon ger- the new 5G bands provide the increased capacity using MIMO
manium (SiGe), complementary metal-oxide semiconductor and CA. The typical FEM structure is presented in Fig. 8.
(CMOS), and silicon on insulator (SOI) [14] but still gallium arsen- The legacy 2G/3G module is directly connected to the battery
ide (GaAs) is the technology mainly used. With the band but the 4G/5G modules are connected through a power

Fig. 7. 5G RF path loss from PA output to the antenna output.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


914 Florinel Balteanu

Fig. 8. LTE 4G/5G RF front-end module structure.

management integrated circuit used also for ET [15]. The TDD • Interference from sub-harmonic mixing for CA case when the
FEM can use also the LNAs on the same module; due to isolation higher UL frequency signal is a multiple of the lower frequency
this is not a solution for FDD systems that have both UL and DL (Band 7 and Band 27/CA case) and desense the LB Rx.
operating at the same time. This is one aspect of the FEM archi- • Interference from the harmonic of lower frequency UL signals to
tecture which must be considered from the beginning together the higher frequency DL when the harmonic of UL lands into UL
with other mitigations due to multiple intermodulation products Rx frequency band. For example, when a UE is transmitting on
for multiple RF transmitters and receivers operating at the same Band 3 4G/LTE and receiving on 5G NR bands n77/n78.
time. Intermodulation products are determined by all the RF Tx Second harmonic of Band 3 will land into 5G NR Rx for bands
paths, including WiFi [16], and all other clock-related activity n77/n78. Another CA case for desense is when 3rd harmonic
as well (charge pumps, MIPI SPI clocks, etc.). There are few (H3) from LB lands in RX high band (HB) as presented in Fig. 9.
types of interference due to simultaneous UL and DL over differ- • Intermodulation distortion (IMD) products between different
ent bands in CA configurations which will degrade the Rx sensi- Tx frequencies and/or MIPI SPI and charge pump clock
tivity (desense): frequencies.

Fig. 9. FDD FEM low band to high band desense in smartphones due to third harmonic.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 915

capacitance C [17, 21] can be obtained by the formulas:


R 0.685
L = 0.732 , C= , (5)
v vR

Lo and Co provide the series resonance, and for different


bands, the structure might have 2Fo and 3Fo traps. As presented
in [22, 23] this structure can be used with a differential trans-
former/balun. In [24] a HPUE PA with an ET has been presented.
Using the same approach, a high resistivity CMOS 0.18 μm PA
with a push–pull class E output stage and integrated fast error
Fig. 10. Power amplifier class E output match. amplifier is presented in Fig. 11.
The PA operates in class E [21] and can use two or more
push–pull structures to provide HPUE power capability (26
To avoid any desense due to third harmonic, at least 90 dB isola- dBm) at the antenna for high 5G bands and operates with good
tion is required; this requires FEMiD shielding and special routing efficiency at lower RF power levels. At lower power levels, few
on the smartphone board. PA cores can be turned off. The ET structure is using an AC com-
biner which doesn’t require a DC tracking loop [17]. This is well
suited for high modulation BW where the large AC coupling cap-
Power amplifier
acitor and the DC tracking loop will slow down the ET operation.
The RF PA is one of the core components for the FEM due to One of the issues using ET [14, 17] is the noise generated in the
efficiency and linearity requirements, especially with the transi- adjacent Tx channels especially for multiple Tx transmitters oper-
tion to 5G. With the adoption of 256 QAM for 5G UL the ating at the same time. To accommodate different Tx power levels
PA linearity expressed through EVM is started to be challenging and multiple Tx operating simultaneously, the ET can be operated
to be met (<3.5%), although 5G allows 4–6 dB maximum in two modes:
power reduction (MPR) for 64/256 QAM 5G cyclic prefix–
orthogonal frequency division (CP-OFDM) signals. Several (1) For lower modulation BW with less LTE RBs, the ET is done
techniques have been used to meet the efficiency and linearity through the PA current bias (Itrck) as presented in Fig. 11
requirements, the most extensively used and researched being and the error amplifier is turned off.
Doherty and ET PAs [17–19]. The adoption of both techniques (2) For higher RBs and multiple PA operating at the same time,
has been possible by advances in digital signal processing the ET fast amplifier is turned on. Based on the signal modu-
and technology scaling such as 7/3 nm FinFET. Doherty techni- lation BW, different low frequency filters are selected on the
ques provide high efficiency but have limitation in terms of Vdc_trck ports. This helps with the RF noise due to low fre-
broadband operation, operation in back-off mode, and load quency up conversion noise.
mismatch [20]. (3) Both Itrck and ET fast amplifiers are activated; assuming the
In mobile applications, such as smartphones, ET is used with envelope delays are well controlled.
a broadband PA with class E output match for low (LB),
middle (MB), high (HB), and ultra-high (UHB) bands. The Bias control for polar loop transmitters was introduced and
class E PA (Fig. 10) with parallel components has been introduced described in [25] and can be extended to ET systems. For high
in [21]. The optimum series feed inductance L and parallel modulation BW, such as in 3G/4G/5G, the polar approach is

Fig. 11. LTE 4G/5G push–pull power amplifier.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


916 Florinel Balteanu

For HPUE, which is a requirement for mid-high 5G bands,


two identical DC voltages (Vdc_dc1 and Vdc_dc2) are provided
to the ET combiners which are located on the same FEMiD as
presented in Fig. 13. This structure allows to achieve higher
power at the PA output and accommodate the losses due to RF
switches and acoustic filters. These losses are higher for 5G due
to multiple antennas routing and higher linearity requirements
for SOI RF switches.
The PA bias (Fig. 14) needs to provide the required current for
the PA with low impedance to cancel the second-order input inter-
cept point (IIP2) terms which might be upconverted and degrades
the EVM and the ACPR/ACLR. The low impedance bias mitigates
the current imbalances which produce ACPR/EVM degradation.
Also, the feedback bias amplifier (Fig. 15) must provide
enough BW for the bias current. Due to these considerations a
simple current feedback structure with source follower is used.
To decrease the output impedance for quite large BW, a low Rs
impedance is required but this increases the current consumption.
From this perspective a programmable Rs is used.
The error amplifier which is part of ET is a current feedback
Fig. 12. Power amplifier and ET operation. amplifier with two stages and is presented in Fig. 16. The second
stage is biased from Vdd_MLS where the voltage is set based on
the modulation PAPR and the PA power level. This is determined
limited by the delay alignment for AM and PM paths. ET system is with a comparator which senses the error amplifier output current
a hybrid approach between polar and linear modulation, where the through Isense pin. The Vctrl pin is used to compensate the cur-
delay alignment is critical but not as critical such as in polar systems. rent offset with a low-frequency feedback loop [26].
Assuming the delay is calibrated and manageable, Fig. 12 presents When the PA is operated in the ET mode the voltage supply
the static PA and ET operation under different bias (ET and PA applied to the PA Vdd = Vdc_trck is following the envelope
gate bias) and therefore load line (Rload) conditions. Under ET signal (instantaneous power level). The peak voltage applied is
operation the PA operates in back off mode, in class AB, B and A determined by the maximum power which must be delivered
based on instantaneous RF power. When Itrck current is used under ET for different peak to average power ratio (PAPR)
and controlled, based on the instantaneous RF power the dynamic waveforms [17]:
operating point is changing from load line 1 to load line 2. The
operations in zones 3 and 4 are controlled by the Vdc_trck 1 and  
Vdc trck peak
2 voltages applied by the ET system based on a look-up table PAPR = 20 log Vdc trck rms . (6)
[17]. Itrck can be generated using a similar look-up table, but this
requires modem/transceiver resources. For the practical implemen- With the adoption of 5G there is an increase to 10.5 dB of the
tation Itrk is generated from Vdc_trck. PAPR for all the waveforms. This determines the requirements for

Fig. 13. 5G power amplifier and ET.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 917

Fig. 14. Differential PA bias structure.

Fig. 15. Differential PA bias structure detail.

Fig. 16. Current feedback error amplifier schematic.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


918 Florinel Balteanu

For 5G FEMs, due to power increase for HPUE and increased


PAPR for 5G NR waveforms, the number of series in FETs must
be increased and determined by the formula:

2
VTx 2(nVDS peak )2
Pmax = max
= , (7)
2Z0 Z0

where VDS_peak is the peak NFET drain source voltage. For off
state the peak RF voltage across FET drain-source for each tran-
sistor assuming equal voltage division is:

|VDS peak | = 2(Vth − Vneg ), (8)

Fig. 17. Servo amplifier detail.


where Vth is the threshold FET voltage.
As presented in [27] the maximum breakdown voltage is
increased slew rate for the error amplifier and this is the main rea- determined by:
son for using a current feedback amplifier.
To maintain a constant current and avoid any peaking and 
instability the error amplifier uses two current mode servo ampli- Cds
BVmax = + 0.5 BVFET . (9)
fiers Srv1 and Srv2 in the final stage as presented in Fig. 17. Using Cgnd
a current mode structure, the servo amplifiers provide a high
operation BW and low propagation delay.
The insertion loss (IL) and the input intercept point (IIP3)
[27, 28] for a series-shunt switch are defined by the equations:
RF FEM SOI switches for 5G
The RF switches are integrated as separate dies to increase isola-  
tion and are fabricated in 0.18 μm SOI technology. The SOI
2
Isat (Ron + 2Zo )4
IIP3 = 10 log + 30, (10)
switch structure is presented in Fig. 18. 4Ron Zo2

Fig. 18. FEM RF SOI switch schematic.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 919

Unfortunately for 5G the RF path losses are higher and the


power requirements at the antenna are increased and therefore
the switch performance requirements are more challenging.
With two or more RF transmitters operating at the same time
there are higher linearity requirements for antenna SOI switches
as well as antenna tuning elements. For example, assuming sensi-
tivity of a typical LTE 5 MHz (25RBs) as −101.5 dBm and 4.5 dB
margin, the linearity requirement IIP3 to avoid jamming assum-
ing Tx1 and Tx2 (Fig. 19) is given by:

PTx2 + 2PTx1 − PIMD 26 + 2 × 23 − (−106)


IIP3 = =
2 2
= 89 dBm. (13)

Tx2 assumes 26 dBm which is required by the 3GPP with


Fig. 19. Cellular intermodulation distortions and cellular/WiFi desense. release #17 [10].
Assuming an external blocker at −30 dBm and a Tx2 UL signal
at 23 dBm the switch linearity IIP2 is given by:
 2  2
Ron 2pCoff (Ron + Zo )
IL = 10 log 1+ + , (11) IIP2 = Pblk + PTx2 − PIMD = −30 + 26 − (−106)
2Zo 2
= +102 dBm. (14)
where Ron is the on-state channel resistance of the switch and Vpos
is the positive voltage To speed up the RF SOI switch transition, for a time response
less than 1 μs (as required in 5G for low latency), a seven-stage
1 ring oscillator with an activated clock doubler (for RF switching)
Ron = n w . (12)
mCox (V pos − Vth ) is used. In order to generate Vpos and Vneg for the switch control
l [28], a negative/positive charge pump is used and presented in

Fig. 20. RF SOI switch positive/negative voltage generation with clock doubler.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


920 Florinel Balteanu

Fig. 21. SAW acoustic filter structure.

Fig. 20. This schematic has a ring oscillator structure with seven precise steps in fabrication and are more costly. In Fig. 23, the
stages. For normal operation, the charge pump clock frequency measured TC-SAW acoustic filter characteristic for band n12
is reduced using several dividers. Low clock frequency reduces with 51–61 dB Tx rejection in Rx band is presented.
the clock feedthrough into the RF PA in the Tx chain as well as With the adoption of 5G for FDD LBs and the use of high
Rx LNAs. For Tx applications there is a limit of the clock fre- modulation BWs, such as 64/256 QAM, the Tx and Rx filter flat-
quency reduction, due to effective reduction for Vneg and there- ness is crucial. There are many efforts in this area to reduce the
fore a reduction for maximum RF power handling as expressed spurious resonances in band caused by traverse modes [29],
by equations (7) and (8). which produce different group delays based on the frequency of
operation in Tx/Rx bands. The traverse mode suppression
improves the ET delay calibration precision and reduces the
RF 4G/5G acoustic filters
time of ET calibration due to reduced number of frequency points
The transition from 3G/4G to 5G, with requirements for different for calibration.
RF transmitters’ coexistence, drives the use of several acoustic fil-
ters. For lower frequencies, up to 2 GHz, the filter requirements
Implementations and measurements
have been handled using SAW filters (Fig. 21). With the evolution
to 5G and more HB and UHB FEMs, there is a high demand for FEMs for smartphones end up into very high-volume products and
BAW filters (Fig. 22). SAW filters can be used up to 2 GHz and therefore all the costs and hardware integration tests associated with
must be temperature compensated (TC-SAW) to operate across the functionality must be considered. One of the RF Tx aspects
a wide temperature range and to accommodate reduced guard is the calibration. Calibration also requires frequency equalization
bands, which is the case for 5G LB FDD filters. due to Tx and duplexer response for different modulation BW
BAW acoustic filters present quality factors of 2000 and 3000, from 1.4 to 100 MHz. An ET calibration method where a baseband
which are higher compared with SAW filters, but require more envelope signal is aligned using a triangular low intermediate

Fig. 22. BAW acoustic filter structure.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 921

Fig. 23. TC-SAW acoustic filter response.

frequency (IF)-modulated RF signal has been presented in [28] and be extended to mmWave FR2 measurements. Also, by using a
used in current 4G/5G mobile devices. The delay is adjusted until baseband envelope signal with several peaks this can be aligned
the baseband-received signals have the same peak values. These after detection and the ET delay calibrated. For higher modulation
RF peaks require an observation receiver to be processed. The obser- BW the ACLR is strongly depended on the delay mismatch as
vation receiver is a prerequisite of actual 4G/5G transceivers but the presented in Fig. 25, where the PA is set at lower power levels
time for testing and calibration is increased. This contribution intro- to avoid ACLR degradation due to heavy non-linearities.
duces a new calibration method where instead of the observation As general rule, for 20 MHz modulation BW the delay should
receiver a fast peak/rms detector is used, as presented in Fig. 24. be calibrated below ±0.5 ns to have good efficiency and at least
The peak detectors are very precise for PA power measure- −38 dBc ACLR for high power levels.
ments (±0.2 dB) and the response BW can be adjusted from 1 For one implementation, the PA, ET amplifier, and ET con-
GHz to lower frequencies such as 2 MHz. This allows for power trol and error amplifiers have been implemented using 0.18 μm
measurements considering different modulation BWs and can CMOS with high resistivity technology with through silicon

Fig. 24. FEM calibration for RF power and ET delay.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


922 Florinel Balteanu

Fig. 25. ACLR versus ET delay mismatch.

Fig. 26. Rx noise measurements for low-frequency 5G FDD bands – n13 and n12.

vias. Together with RF SOI switches and RF acoustic filters, all Table 2. CMOS 0.18 μm and GaAs PA and ET measurement results for LTE 10 MHz,
are integrated on the same substrate in a multi-chip-module. 64 QAM, band n12
The worldwide transition of smartphones to 5G and the world- RxBN at
wide adoption for billions of people are one of main driving Gain Efficiency ACLR1 30 MHz
engine behind semiconductor industry. From this perspective Pout CMOS/GaAs CMOS/GaAs CMOS/GaAs CMOS/GaAs
the costs associated with the smartphone hardware as well as (dBm) (dB) (%) (dBc) (dBm/Hz)
fabrication and field calibration must be considered. One of
25 26.5/27 36/38 −42/−44 −125/−125
the important aspects for 5G Tx and Rx integration is the RF
noise and noise measurements in production. The ET filters 26 26/27 38/40 −41/−43 −124/−124
(Fig. 11) as well as charge pump noise mitigation have been 27 25/26 40/43 −40/−42 −120/−121
implemented to provide enough margin from the design
28 24/25 42/46 −37/−39 −117/−120
phase for the FEM noise in Rx band. In this way there is no

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


International Journal of Microwave and Wireless Technologies 923

Fig. 27. Power amplifier GaAs die for low bands 600–1200 MHz.

which exhibit very good efficiency for ACLR1 = −38 dBc, with-
out digital predistortion, for 64 QAM. These measurements are
done at the PA output. For 5G LBs, the power required at the
antenna is 23 dBm and, with 5 dB MPR for 64 QAM
CP-OFDM, the PA can be used in a LB FEM.
Similar implementation has been done using a GaAs PA
(Fig. 27) with an integrated passive device (IPD) balun. In
this case the ET bias and control circuitry is on a separate
0.18 μm CMOS die. The efficiency observed was better for
GaAs versus CMOS, with almost 3–4% for the same ACLR
and noise levels. For both implementations there is a penalty
in terms of size using an IPD balun. At LBs (400–1200
MHz), using a balun helps with the second harmonic suppres-
sion for coexistence with other 5G transmitters and mitigates
Fig. 28. DP6T insertion loss and IIP3 measurements for 50 Ohm load. the low breakdown voltage for CMOS devices [24, 30]. For
size constrained applications, which is the case for 5G smart-
phones, the use of IPD baluns becomes practical for higher
need to check thoroughly during FEM production. RF noise FR1 frequencies, such as n41, n77, n78, and n79 or FR2
mitigation is especially challenging for low frequency 5G mmWave frequencies.
bands where the duplex space is 30 MHz as in FDD bands In Fig. 28, the measurements for a DP6T SOI switch used in
n12 (699 MHz–716 MHz-UL/729 MHz–746 MHz-DL) and the FEM are shown. The IIP3 is lower than 89 dBm as derived
n13 (777 MHz–787 MHz-UL/746 MHz–756 MHz-DL). The in equation (14). In this case the RFFE must operate under
Rx noise figure (NF) will degrade with 0.5 dB for a TX noise MPR or be used with different antenna route.
level lower than −120 dBm/Hz in Rx band. The FDD acoustic
filter has a significant influence on the Rx band noise perform-
Conclusion
ance as the typical duplexer isolation between Tx and Rx is
between 51 and 61 dB. Assuming the noise floor to be approx- Smartphone industry is a very large volume industry where hard-
imatively −174 dBm/Hz, for just 0.1 dB NF degradation, the ware changes and improvements take long time to be tested for
noise after duplexer needs to be below −184 dBm/Hz. The functionality and reliability before being deployed. 5G FEM archi-
noise measurements are shown in Fig. 26. For band n12 the tectures have been presented and a fully integrated 0.18 μm LB
measurements show low noise of −124 dBm/Hz at 30 MHz FDD CMOS PA and tracker was packaged and tested into FEM
duplex space using an ET low-frequency filter with Lf = 39 nH together with SOI switches, duplexers, and filters. Same tests
and Cf = 1 nF (Fig. 10); Pout = 26 dBm. With the new TDD 5G have been performed using a GaAs PA. Elements of this architec-
bands [3] in the so-called C-band (n77, n78, and n79), there ture with CMOS PA and ET collocated, used for 4G, have been
will be more BW available for high data rate but still the low- presented initially in 2014. This paper presents FEM circuit
frequency bands will be used as 4G/5G LTE communication improvements for noise reduction and increased linearity through
anchors. For example, right now the majority of 5G deploy- ET. Elements of this architecture with CMOS PA and ET collo-
ments use primarily 3.5 GHz TDD bands using NSA mode cated, used for 4G, initially have been demonstrated at the 2014
but will transition to SA. Table 2 presents the measurements Mobile World Congress, Barcelona, and have been improved for
for LTE 5 MHz, 710 MHz, Band n12 (30 MHz LTE BW), HPUE and 5G. The next step for 5G deployment, as has been

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press


924 Florinel Balteanu

demonstrated at the 2022 Mobile World Congress, Barcelona, is to 18. Doherty WH (1936) A new high efficiency power amplifier for modulated
move toward SA mode using the new 5G CA core architecture waves. Proceedings of the Institute of Radio Engineers 24, 1163–1182.
using FDD LB bands operating in 700 MHz (n12, n13, n28) 19. Kahn R (1952) Single-sideband transmission by envelope elimination and
restoration. Proceedings of the Institute of Radio Engineers 40, 803–806.
along with supplemental DL MB band 1400 MHz (n75) and
20. Zenteno E, Isaksson M and Händel P (2015) Output impedance mis-
UHB 3500 MHz (n78). Also, the use of FDD LB for 5G will
match effects on the linearity performance of digitally predistorted
enable better URLLC services due to lower latency. This will power amplifiers. IEEE Transactions on Microwave Theory and
power enhance consumer experience with faster peak speeds Techniques 63, 754–765.
and higher throughput and allow for industry growth in areas 21. Grebennikov A and Jaeger HH (2002) Class E with parallel circuit – a
such as industrial IoT and cloud services. new challenge for high-efficiency RF and microwave power amplifiers.
The novel circuits and calibration methods in this paper will IEEE MTT-S International Microwave Symposium Digest 3, 1627–1630.
make 5G RFFE integration possible for next generation mobile 22. Kim WY, Son HS, Kim JH, Jang JY, Oh IY and Park CS (2013) A fully
5G applications using FR1 and FR2. integrated triple-band CMOS class-E power amplifier with a power cell
resizing technique and a multi-tap transformer. IEEE Microwave and
Supplementary material. The supplementary material for this article can Wireless Components Letters 23, 659–661.
be found at https://doi.org/10.1017/S1759078722001295. 23. Dinc T, Kalia S, Akhtar S and Haroun B, Cook B and Sankaran S
(2022) High-efficiency class-E power amplifiers for mmWave radar sen-
Conflict of interest. None. sors: design and implementation. IEEE Journal of Solid-State Circuits 57,
1291–1299.
24. Balteanu F, Drogi S, Hardik M, Choi Y, Khesbak S and Agarwal B
References (2019) New architecture elements for 5G RF front end modules. IEEE
1. Chen W, Gaal P, Montojo J and Zisimopoulos H (2021) Fundamentals Asia-Pacific Microwave Conference (APMC), Singapore, pp. 321–323.
of 5G Communications: Connectivity for Enhanced Mobile Broadband and 25. Sowlati T, Rozenblit D, Pullela R, Damgaard M, McCarthy E, Koh D,
Beyond. New York: Mc Graw Hill. Ripley D, Balteanu F and Gheorghe I (2004) Quad-band GSM/GPRS/
2. Pehlke D, Brunel D, Walsh K and Kovacic S (2017) Sub-6GHz 5G for EDGE polar loop transmitter. IEEE Journal of Solid-State Circuits 39,
UE, optimizing cell edge user experience. IWPC 4G/5G Multi-Band, 2179–2189.
Multi-Mode User Equipment, Austin, October 2017. 26. Drogi S, Balteanu F and Pehlke D (2022) Envelope tracking systems for
3. Pehlke D (2020) TDD in US and emerging requirements and capabilities power amplifiers. US patent 11, 387, 797.
for NR. IWPC 5G User Equipment Evolution, September 2020. 27. Zhu Y, Klimashov O, Roy A, Blin G, Whitefield D and Bartle D (2015)
4. Gupta A and Kumar R (2015) A survey of 5G network: architecture and High voltage SOI stacked switch with varying periphery FETs. Asia-Pacific
emerging technologies. IEEE Access 3, 1206–1232. Microwave Conference (APMC), pp. 1–3.
5. GTI (2020) Sub-6GHz 5G device. White Paper v5.0, November 2020. 28. Balteanu F, Zhu Y and DiCarlo P (2021) Envelope alignment calibration
6. Rappaport T, Sun S, Mayzus R, Zhao Y, Azar Y, Wang K, Wong GN, in radio frequency systems. US patent 11, 165, 514.
Schulz JK, Samimi M and Cutierrez F (2013) Millimeter wave commu- 29. Liu Y, Liu J, Wang Y and Lam CS (2019) A novel structure to suppress
nications for 5G cellular: it will work!. IEEE Access 1, 335–349. transverse modes in radio frequency TC-SAW resonators and filters. IEEE
7. Wi-Fi Alliance (2018) Introduces Wi-Fi 6. Available at http://www.wi-fi. Microwave and Wireless Components Letters 29, 249–251.
org, October 2018. 30. Aoki I, Kee S, Magoon R, Aparicio R, Bohn F, Zachan J, Hatcher G,
8. Le TK, Salim U and Kaltenberger F (2021) An overview of physical layer McClymont D, Hajimiri A (2008) A fully-integrated quad-band GSM/
design for ultra-reliable low-latency communications in 3GPP releases 15, GPRS CMOS power amplifier. IEEE Journal of Solid-State Circuits 43,
16, and 17. IEEE Access 9, 433–444. 2747–2758.
9. Esswie AA and Pedersen KI (2020) On the ultra-reliable and low-latency
communications in flexible TDD/FDD 5G networks. IEEE 17th Annual Florinel Balteanu received the M.S. degree in
Consumer Communications & Networking Conference (CCNC), pp. 1–6. electrical engineering from Polytechnic
10. 3GPP TR 38.808 V17.0.0 (2021) Study on supporting NR from 52.6 GHz Institute, Bucharest, Romania, in 1983, and the
to 71 GHz. Tech. Spec. Group Radio Access Network, Rel. 17. Ph.D. degree in electrical engineering from
11. Microwave Journal (2022) MWC Barcelona 2022 highlights, March 2022. Transylvania University, Brasov, Romania, in
12. 3GPP Technical Report TR 101 112 V3.2.0 (1998) Selection procedures 1995. From 1983 to 1992, he was with the
for the choice of radio transmission technologies of the UMTS, April 1998. Institute of Nuclear Research, Pitesti, Romania,
13. Balteanu F, Serge Drogi S, Modi H, Choi Y, Khesbak S and Agarwal B working on electronic instrumentation for
(2019) New architecture elements for 5G RF front end modules. 2019 nuclear power reactors. In 1993, he joined the
IEEE Asia-Pacific Microwave Conference (APMC), Singapore, pp. 321–323. Department of Electronics, University of Pitesti, Romania, as an Assistant
14. Lie DYC, Tsay J, Hall T, Nukala T and Lopez J (2016) High-efficiency Professor doing research in analog circuits. From 1992 to 1993, he was a
silicon RF power amplifier design – current status and future outlook. Fulbright Visiting Scholar at the Center for Reliable Computing, Stanford
In IEEE International Symposium on Radio-Frequency Integration University, Stanford, CA, USA. In 1996, he joined Philsar Semiconductor,
Technology (RFIT) Digest, pp. 1–3. Ottawa, ON, Canada, as a Senior ASIC Designer, a company later acquired by
15. Paek JS, Kim D, Bang JS, Baek J, Choi J, Nomiyama T, Han J, Choo Y, Conexant Systems, Inc. (now known as Skyworks Solutions, Inc.). His work
Youn Y, Park E, Lee S, Kim IH, Lee J, Cho TB and Kang I (2019) An has been focused on the design of radio circuits for GPS receivers, Bluetooth
88%-efficiency supply modulator achieving 1.08μs/V fast transition and transceivers, and GSM/CDMA cellular transceivers. He is presently a Technical
100MHz envelope-tracking bandwidth for 5G new radio RF power amp- Director with Skyworks Solutions Inc., Irvine, CA, USA and is involved in design-
lifier. IEEE International Solid-State Circuits Conference (ISCC), ing circuits for envelope tracking and front-end modules in CMOS and SOI
pp. 238–240. CMOS. He presented short courses and tutorials as an Invited Speaker at IEEE
16. Maldonado R, Karstensen A, Pocovi G, Esswie AA, Rosa C, Alanen O, Microwave Theory and Techniques Society (MTT-S) International Microwave
Kasslin M and Kolding T (2021) Comparing Wi-Fi 6 and 5G downlink Symposium (IMS). European Microwave Week (EuMW), European Solid-State
performance for industrial IoT. IEEE Access 9, 86928–86937. Circuits Conference (ESSCIRC), and Asia-Pacific Microwave Conference
17. Grebennikov A (2019) Radio Frequency and Microwave Power Amplifiers. (APMC). He holds 84 US patents, with several more pending. He is the author
London, United Kingdom: IET The Institution of Engineering and of the chapter “Envelope Tracking Techniques” in the IET book “Radio
Technology. Frequency and Microwave Power Amplifiers, Vol. 2: Efficiency and Linearity”.

https://doi.org/10.1017/S1759078722001295 Published online by Cambridge University Press

You might also like