©istockphoto.

com/eestingnef
O
ver at least the last two decades, digital pre-
distortion (DPD) has become the most com-
mon and widespread solution to cope with
the power amplifier’s (PA’s) inherent lineari-
ty-versus-efficiency tradeoff. When com-
pared with other linearization techniques, such as Cartesian
feedback or feedforward, DPD has proven able to adapt to
the always-growing demands of technology: wider band-
widths, stringent spectrum masks, and reconfigurability.
The principles of predistortion linearization (in its analog
or digital forms) are straightforward, and the linearization
subsystem precedes the PA (a nonlinear function in a digital
signal processor in the case of DPD or nonlinear device in the

Digital
Predistorters Go
Multidimensional
Pere L. Gilabert, Gabriel Montoro,
David Vegas, Nieves Ruiz, and José Angel García

Pere L. Gilabert ([email protected]) and Gabriel Montoro ([email protected]) are with the Department of Signal Theory
and Communications, Universitat Politècnica de Catalunya, Castelldefels, Spain. David Vegas ([email protected]),
Nieves Ruiz ([email protected]), and José Ángel García ([email protected]) are with the Department
of Communications Engineering, Universidad de Cantabria, Santander, Spain.
Digital Object Identifier 10.1109/MMM.2019.2898021
Date of publication: 5 April 2019

50 1527-3342/19©2019IEEE May 2019

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 case of analog predistortion and counteracts the nonlin- millimeter-wave (mm-wave) bands, and, at sub-6 GHz,
ear characteristic of the PA. Some excellent overviews on bandwidth of hundreds of MHz will be required.
DPD can be found in [1]–[4]. Let us now look at the chal- Achieving these new capabilities requires coping
lenges that DPD linearization has faced and will contin- with multiple demanding challenges that, particularly
ue to face in the near future with 5G new radio (5G-NR). for the design of radio transceivers, are related to sev-
In mobile communications, at some point after the eral factors:
evolution from 2G [global system for mobile (GSM) → 1) ensuring the linearity of signals having bandwidths
general packet radio service → enhanced data rates for of several hundreds of megahertz and peak factors
GSM evolution (EDGE)] to 3G [wideband code divi- exceeding 10 dB to ensure high transmission rates
sion multiple access (WCDMA) → high-speed packet 2) improving energy and computational efficiency, as
access (HSPA) → HSPA+] and with the advent of smart- denser deployments of base stations are expected to
phones, mobile data consumption began its unstop- scale down the need for transmitted power
pable rise. Consequently, a more efficient use of the 3) transmitting architectures with multiple anten-
radioelectric spectrum was required. For example, in nas (massive MIMO in mm-wave bands) and
2G GSM-EDGE, the maximum link spectral efficiency multiple PAs to apply beamforming techniques
was 1.92 b/s/Hz, but, only a few years later, 3G-HSPA that allow increasing capacity and decreasing
achieved 4.22 b/s/Hz [5]. However, the new waveforms radiated power
were optimized from the perspective of spectral effi- 4) simultaneous transmission and reception (full
ciency, not power efficiency. New challenges arose frequency-division duplex in sub-6 GHz bands).
from the power-efficiency perspective because we As mentioned, DPD can overcome, or at least miti-
had to go from an efficient amplification of constant- gate, the efficiency versus linearity problem in PAs.
envelope-modulation waveforms in 2G [i.e., 200-kHz However, the resulting power efficiency achieved with
bandwidth, single-carrier Gaussian minimum shift linearization techniques applied to PAs operating as
keying signals with 0 dB of peak-to-average power controlled current sources (e.g., classes A, B, and AB)
ratio (PAPR)] operating close to or in saturation with is limited. To avoid wasting excessive power resources
amplitude- and phase-modulated spread spectrum when handling high PAPR signals, the operating con-
signals in 3G (i.e., a 5-MHz bandwidth WCDMA sig- ditions of a current-source-mode PA could be forced
nal with 6 dB of PAPR). High-fidelity amplification of to follow its envelope, or switched-mode amplifying
nonconstant envelope-modulated signals with high classes could be properly introduced. Among the set
PAPR requires avoiding envelope clipping at satura- of techniques aimed at dynamic bias or load adapta-
tion, causing operation at significant power backoff tion, envelope tracking (ET) [7]–[9], Doherty [10], [11],
levels where PA efficiency figures are far below the and linear amplification with nonlinear components
maximum achievable values. (LINC) or outphasing PAs [12], [13] are the most widely
With 4G (LTE, LTE-Advanced), to satisfy the need for proposed in the literature.
higher data rates and numbers of users, spectrally effi- In addition, significant efforts have been dedicated
cient multicarrier waveforms (i.e., 20 MHz of orthogo- in recent years to designing wireless communication
nal frequency-division multiplexing–based signals) systems capable of handling multistandard or mul-
with high-density modulations [e.g., 64-quadrature- tiband signals at the same time. The advantages of
amplitude modulation (QAM)] were introduced. Addi- having a single PA process signals in multiple bands
tionally, new technologies emerged in the evolution simultaneously are the reductions in the number of
of 4G LTE, such as carrier aggregation and multiple
input/multiple output (MIMO), extending the maxi-
mum data rate up to, for example, up to 3 GB/s in LTE-
advanced when considering aggregated bandwidths
100 MHz and 4 × 2 MIMO. These technologies continue
to develop to massive speed and scale in 5G-NR.
In 5G-NR [6], the same network infrastructure will
be able to efficiently serve different types of traffic hav-
ing a very wide range of requirements, such as the huge
number of users for the Internet of Things, ultralow
latency and high reliability for mission-critical sys-
tems, and enhanced transmission rates for broad-
band mobile communications. 5G-NR is intended to
provide very high data rates everywhere. To achieve
this goal, bandwidths up to GHz will be allocated at

May 2019 51

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 DPD can overcome, or at least the antenna), are removed to avoid increasing the system
complexity and cost. As a consequence, these integrated,
mitigate, the efficiency versus multiantenna transmitters typically suffer from nonlin-
linearity problem in PAs. ear distortion because of mixing of the antenna crosstalk
and mismatch with the PA output, in addition to the non-
linear distortion caused by the PAs [15]. This is another
components and the cost of the RF subsystem [14]. For example in which DPD needs to go multidimensional to
bands separated by several hundred megahertz, the compensate for the multiple sources of unwanted dis-
implementation of a wideband DPD is not feasible, espe- tortion effects.
cially in real-time platforms. Fortunately, DPD systems Figure 1 shows a conceptual map or compact over-
for multiband signals can be significantly simplified, view of the previously mentioned applications where
assuming that the nonlinear distortions of concern multidimensional DPD is used and identifies some
are those arising close to the band of interest and that references, research groups, and companies that have
the rest can be removed by filtering. When concurrent published on these topics. This article focuses on the
multiband transmissions in PAs with dynamic-load or ­multidimensional DPD required to compensate for con-
dynamic-supply modulation are combined, the DPD current multiband transmission when using dynamic-
needs to go multidimensional. That is, multiple input/ load or dynamic-supply-modulated PAs. ET and
single output (MISO) DPD behavioral models are nec- outphasing PAs are unlikely to be deployed for ultra-
essary to compensate for all of the unwanted distortion wideband applications in 5G mm-wave bands because
effects that appear at the PA output. of the bandwidth and power/cost budget limitations
Similarly, in multiantenna systems where each trans- of these high-efficiency amplification architectures,
mit path has its own PA and antenna element, bulky com- which require DPD linearization. However, in 5G sub-
ponents, such as isolators (placed between the PA and 6-GHz macro base stations, if envelope bandwidth or

Reference: [16]–[23]

Concurrent
Multiband
Contributors on Multidimensional DPD Amplification
Chalmers University of Technology,
Sweden
KTH Royal Institute of Technology,
Sweden
Dynamic-Supply-
Ohio State University, United States and
Dynamic-Load-
Universidade de Aveiro, Portugal Modulation PAs
Universitat Politècnica de Catalunya,
Spain Reference: [24]–[31]
University of California San Diego,
United States Digital
University of Calgary, Canada Predistorter

University of College Dublin, Ireland Multidimensional

DPD
University of Waterloo, Canada

Southeast University ( ), China

Tampere University of Technology,
Finland
Tsinghua University ( ), China MIMO
Transceivers
...

Eriksson, Nokia, Qualcomm, Intel, Huawei,

Reference: [15], [32]–[37]
Powerwave, ETA Devices, Maxentric, . . .

Figure 1. The applications for multidimensional DPD models (MISO DPD) [15], [17], [26].

52 May 2019

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 slew-rate reduction techniques [25] properly combined
In ET PAs, the supply voltage of
with dimensionality reduction techniques [30] are con-
sidered to meet the low complexity requirements of the the RF PA is adjusted according
DPD implementation, the use of dynamic-supply- (e.g., to the envelope of the RF carrier.
ET or class-G PAs) or dynamic-load-modulation
techniques (e.g., outphasing or load-modulated-bal-
anced PAs) can still be seen as interesting solutions for impedance seen by the other), providing good effi-
high-efficiency amplification. ciency but not-so-good linearity, the LINC combiner
isolates the two PA outputs (allowing them to see a
DPD for PAs With Dynamic-Supply or fixed load at all times), which favors having good com-
Dynamic-Load Modulation in Concurrent bining linearity but leads to high dissipation at high
Multiband Transmissions outphasing angles.
If the PAs are assumed to operate as ideal voltage
Dynamic-Supply Versus sources, with their outputs connected to a floating
Dynamic-Load Modulation load R (usually provided by a balun), as the relative
In ET PAs, the supply voltage of the RF PA is adjusted phase between the two sources is varied between
according to the envelope of the RF carrier. Thanks to 180° and 0°, the effective loading on both PAs is var-
the dynamic supply, the RF PA (linear current-source ied between R/2 and infinity. The output power from
mode, Doherty, or even LINC PAs) can be forced to each branch PA is decreased, thereby reducing the
operate close to saturation, which increases the power dissipated power losses [12]. This is the principle of
efficiency at power backoff. Several strategies can be the concept of active-load modulation proposed by
designed (through the so-called shaping function) to Chireix, who also added two fixed compensating
shape the supply voltage signal to achieve better lin- reactances to provide some control over the efficiency-
earity and efficiency or to meet the slew-rate and band- versus-output power profile.
width restrictions of the drain modulator [7]–[9]. The concepts of dynamic-supply and active-load modu-
In outphasing PAs [12], [13], the idea introduced by lation can be appreciated based on Figure 2, which shows
H. Chireix was to use phase control of two constituent the gain and efficiency evolution versus the output power
branch PAs operated in saturation by enabling con- for different supply voltages [Figure  2(a)] and differ-
stant-output envelope signals that were summed at ent outphasing angles [Figure 2(b)] when character-
the output of the system to allow AM. The concept was izing one single or two reactively combined continuous
later reintroduced by D.C. Cox, who generalized the class-J mode gallium nitride (GaN) high-electron mobility
linear amplification with nonlinear components (i.e., transistor PAs, respectively [38]. For ET, the Nujira com-
LINC) approach. The main difference is that, although pany (acquired by Qualcomm) introduced the concept
the Chireix combiner is nonisolating (and, therefore, of isogain shaping [39], whereby the instantaneous sup-
the output signal from each PA changes the load ply voltage is designed to achieve a particular constant

30 100 30 100

25 80 20 75
Efficiency (%)
Efficiency (%)

20
Gain (dB)
Gain (dB)

60
15 10 50
40
10
20 0 25
5
0 0 –10 0
10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 50
Pout (dBm) Pout (dBm)

VDD = 5 V VDD = 15 V VDD = 25 V VDD = 32 V 0° 40° 80° 120°

VDD = 10 V VDD = 20 V VDD = 28 V 20° 60° 100° 140°
(a) (b)

Figure 2. The gain and efficiency evolution versus output power: (a) at VGS =-2.3 V and R L = 50 X for different VDD values
and (b) for different outphasing angles.

May 2019 53

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 The isogain trajectory provides in real-time platforms. Fortunately, DPD systems for mul-
tiband signals can be significantly simplified, assuming
constant gain at the price of losing that the nonlinear distortions of concern are those that
certain efficiency versus the optimum arise close to the band of interest and that the rest can be
efficiency trajectory. removed by filtering.
On the other hand, both dynamic-supply- and dynamic-
load-modulation techniques are capable of achieving
PA gain. In this case, the ET PA system achieves low high power-efficiency figures even when operated
AM/AM distortion at the price of lower efficiency than with amplitude- and phase-modulated signals pre-
when optimized. In a similar fashion, the isogain approach senting significant PAPR (i.e., good efficiency profiles
can be applied in outphasing systems [38] or alternative with backoff operation). However, the efficiency decays
dynamic-load-modulation techniques, as shown in Figure with the signal’s bandwidth. Both dynamic techniques
3, where the gain [Figure 3(a)] and efficiency [Figure 3(b)] show a tradeoff between the mean power efficiency that
evolution versus output power with a single PA is can be achieved and the instantaneous signal bandwidth
shown for different resistive loading values. The isogain handled by the PA.
trajectory provides constant gain at the price of losing cer- A Chireix outphasing PA is intrinsically narrowband.
tain efficiency versus the optimum efficiency trajectory. The frequency-dependent characteristic of the nonisolat-
In the following section, we particularize the shaping ing combiner, including the compensating reactances,
of the supply voltage in ET PAs and the outphasing angle and the reactive elements in the device model usually
in outphasing PAs when considering concurrent multi- impose strong limitations on its bandwidth. For example,
band transmissions. if designed with class-E switched-mode PAs (a highly
attractive operating class in terms of efficiency perfor-
Slow-Envelope Generation mance), the efficiency contours rotate counterclockwise
in Multiband Transmission with increasing frequency, whereas the mutual-load-
The amplification of concurrent multiband or even car- modulation trajectories offered by a passive (Fos-
rier-aggregated transmissions with high-efficiency topol- ter) combiner rotate in the expected clockwise sense [40].
ogies based on dynamic-supply modulation (e.g., ET) or In a pure outphasing operation, the constant-envelope
dynamic-load modulation (e.g., outphasing) faces several phase-modulated signals to be handled by the constitu-
challenges. On the one hand, to guarantee the desired lin- tive branches may have a bandwidth several times wider
earity levels, the DPD must be designed to account for the than the original signal to be reproduced. For wideband
difficulty of running the DPD at approximately five times or multiband signals, a broadband input matching net-
the signal instantaneous bandwidth (due to bandwidth work would also be required to avoid undesired phase-
expansion occurring during the DPD process). For bands modulation-to-AM conversion. In ET PAs, one of the
separated by several hundreds of megahertz, the imple- main challenges involves the design of efficient envelope
mentation of a wideband DPD is not feasible, especially modulators capable of supplying the power required by

28 90

25 75
Efficiency (%)

22 60
Gain (dB)

19 45
16 30
13 15
10 0
15 20 25 30 35 40 45 15 20 25 30 35 40 45
Pout (dBm) Pout (dBm)
(a) (b)

30 Ω 50 Ω 100 Ω 150 Ω 250 Ω 350 Ω 450 Ω Optimal Efficiency Isogain

Figure 3. The (a) gain and (b) efficiency evolution versus output power at VGS =-2.3 V and VDD = 28 V for different R L
values.

54 May 2019

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 the transistor at the same speed as the signal’s envelope. envelope drivers for base stations can efficiently handle
In concurrent multiband transmission, the envelope of signal bandwidths up to 40 MHz with power efficiencies
the resulting RF signal can present bandwidths that are greater than 70% [41].
several times (according to the rule of thumb, approxi- Because the efficiency of the envelope modulator
mately three times) the instantaneous bandwidth. For drops at high frequencies, to avoid dealing with high-
example, with carriers separated by hundreds of mega- speed envelope variations in multiband signals, meth-
hertz, the envelope modulator should present a slew ods to reduce the bandwidth [24] or slew rate [25] of
rate capable of efficiently amplifying the supply sig- the signal’s envelope have been proposed. In addition,
nals of gigahertz of bandwidth. Currently, commercial in the particular case of dual-band signals, two main

–80 –80
Power/Frequency (dB/Hz)

Power/Frequency (dB/Hz)
–100 –100

–120 –120

–140 –140

–160 –160

–180 –180
0.7 0.9 1.1 1.3 1.5 1.7 1.9 2.12.2 0.7 0.9 1.1 1.3 1.5 1.7 1.9 2.12.2
Frequency (GHz) Frequency (GHz)
(a)
1 1
0.8 0.8
0.6 0.6
0.4 0.4
Amplitude

Amplitude

0.2 0.2
0 0
–0.2 –0.2
RF Signal RF Signal
–0.4 –0.4
Original Envelope Original Envelope
–0.6 Slow Envelope (Peak) –0.6 Slow Envelope (Peak)
–0.8 Slow Envelope (Average) –0.8 Slow Envelope (Average)
–1 –1
1.7 1.71 1.72 1.73 1.74 1.75 1.76 1.77 1.781.79 1.8 1.7 1.71 1.72 1.73 1.74 1.75 1.76 1.77 1.781.79 1.8
Time (s) ×10–5 Time (s) ×10–5
(b)

Original Envelope Original Envelope

–80 –80
Slow Envelope (Peak) Slow Envelope (Peak)
Power/Frequency (dB/Hz)

Power/Frequency (dB/Hz)

Slow Envelope (Average) Slow Envelope (Average)

–100 –100

–120 –120

–140 –140

–160 –160

–5 –4 –3 –2 –1 0 1 2 3 4 5 –5 –4 –3 –2 –1 0 1 2 3 4 5
Frequency (GHz) Frequency (GHz)
(c) (d)

Figure 4. The (a) spectra of the RF signal, (b) slow-envelope waveforms (peak [p = 1] and average [p = 2] approaches), and
slow-envelope spectra of both the (c) dual-band (m = 2) and (d) multiband transmissions (m = 5) .

May 2019 55

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 At deep power backoff, there may be and a multiband transmission (m = 5) . The generated
slow envelope then goes through a shaping function
no benefit in terms of power-added (a comparison of different shaping functions can be
efficiency when operating in a pure found in [7]), which, in the case of multiband transmis-
outphasing mode. sions, consists of a detroughing function. Detroughing
prevents the supply signal VDD (t) from dropping to zero
volts, thus avoiding sharp amplitude nulls in the time
approaches to deal with the instantaneous dual-band domain that may increase the bandwidth requirements.
envelope of the transmitted signal have been described: Moreover, a supply is typically more efficient at higher
1) perform the sum of the modulus of the baseband voltage levels and with a limited voltage swing, whereas
signals [i.e., the peak of the instantaneous dual-band at low power levels the efficiency is not as important. Fig-
envelope, p = 1 in (1)], as proposed in [26]; or 2) perform ure 5 depicts two different shaping functions: hard and
the square root of the sum of the squared modulus of soft detroughing. Soft detroughing can be defined with a
the baseband signals [i.e., the average amplitude of the function as described in [42],
instantaneous dual-band envelope, p = 2 in (1)], as pro- 1

E[n] = c (E S–TH)6 + ^E S 6n@h6 m , (3)

6
posed in [28]. These two approaches are particular solu-
tions of the following method that can be generalized
for multiband transmissions. Therefore, to generate a where E S–TH determines the shaping curve and the
slower envelope version E S[n] of the original instanta- minimum clipping level.
neous envelope ; u[n] ;, the generalized mean or power
mean with exponent p can be extended for m bands, Multidimensional DPD
Dealing with signals presenting instantaneous band-
1
m widths of several hundreds of megahertz is a challenge.
E S[n] = c 1 / ; u i[n] ; p m , (1)
p

m i =1 This is true not only for analog-to-digital converters
(ADCs) and digital-to-analog converters (DACs) that
where the multiband complex baseband signal is need to operate with sampling rates at least five times
defined as the signal’s instantaneous bandwidth (note that, despite
m using undersampling techniques for the ADC [1], the
u[n] = / u i[n] e jX i n, (2) DAC at least has to operate at full rate). It is also true
i =1
because the field-programmable gate array (FPGA)
where X i = 2rfi /fs and fi and fs are the center fre- must operate at clock rates that, being optimistic, may
quency and the sampling frequency, respectively. be at the limit of current technology (e.g., current com-
Figure 4 shows the time-domain waveforms and spec- mercially available intellectual property for DPD are
tra of the original RF signal’s envelope and two examples limited to signal bandwidths up to 200 MHz). In mul-
of slow-envelope generation (peak, p = 1, and average, tiband signals, we can relax the clock rate constraints,
p = 2 , approaches) when considering a dual-band (m = 2) assuming that the nonlinear distortions of concern are

1 0.8
0.7
0.8
0.6
Normalized Output

0.5
Amplitude

0.6
0.4
0.4 0.3
No Shaping 0.2 No Shaping
0.2 Hard Detroughing
Hard Detroughing
Soft Detroughing 0.1
Soft Detroughing
0 0
0 0.2 0.4 0.6 0.8 1 1.7 1.72 1.74 1.76 1.78 1.8
Normalized Input Time (s) ×10–5
(a) (b)

Figure 5. The (a) detroughing shaping functions and (b) waveforms before and after detroughing.

56 May 2019

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 those that arise close to the band
of interest and, thus, that indi-
vidual DPDs will take care of each
multiband transmission.

Outphasing PA
As explained, several MISO

y (t)
DPD models have been published

ET PA
that compensate not only for the

Combiner
Chireix
intraband intermodulation dis-

y (t)
tortion (both in-band and out-of-

VDD(t)
band distortion) but also for the
crossband-intermodulation distor-

RF PA
tion (Figure 1). In addition, using

RF PA

RF PA
a significantly slower envelope

(t)
than the original instantaneous

xRF(t)

xoutph_2
(t)
multiband envelope to dynami-

LO1
LO2

xoutph_1

RF

Figure 6. A block diagram of a multidimensional DPD for both ET and outphasing PAs. MOD: modulator; EA: envelope amplifier.
EA
cally supply or dynamically load-

RF
MOD

MOD
modulate the PA will introduce

MOD
x

x
Q

Q
additional nonlinear distortion.

Q
I

I
I

LO1

LO2
E (t )

Consequently, to compensate for

DAC
DAC
DAC
DAC
the slow-envelope-dependent

ADC
ADC
DAC
DAC
DAC

distortion effects, we need a con-

[n]

[n]
[n]
[n]
current multiband DPD model

xoutph_1

xoutph_2
xoutph_2
xoutph_1
x Q [n]

Transformation
Transformation
x I[n]

that includes the information of

Outphasing

Outphasing
Q

Q
Converter
Cartesian

I
Polar-to-

the slow envelope,

x i[n] = fDPD i (u 1[n], u 2[n], f, I

u K [n], E[n], n), (4)

where x i[n] with i = 1, 2, f, K is

the ith output of the MISO DPD;
u 1[n], u 2[n], f, u K [n] are multiple
inputs corresponding to each of
+

the concurrent bands; and E[n] is

the generated slow envelope. Thus,
e–jΩ1n

ej ΩKn
e j Ω2n

the baseband signal processing

x
x

y [n]

for generating the predistorted in-

xK [n]
x2[n]
x1[n]

phase/quadrature (I/Q) signals is

E [n ]

the same for both ET and outphas-

Adaptation

ing PAs (which are considered

DPD K
DPD 2
DPD 1
Detroughing

DPD

black-box systems) and consists of

the MISO system described in (4),
with as many input signals as con-
ES [n ]

current bands and the slow-enve-

lope signal used to dynamically
Generation
Envelope

supply the PA or code the outphas-

Slow-

ing angle.
Figure 6 shows a block diagram
Multidimensional DPD

of a multidimensional DPD to com-

pensate for the intraband, crossband,
. 
. 

. 

and slow-envelope-dependent dis-

tortion effects in a concurrent mul-
tiband system when considering
either dynamic-load modulation (i.e.,
u1[n]

u2[n]

uK [n]

outphasing PA) or dynamic-supply

modulation (i.e., ET PA). In the case

May 2019 57

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 of the outphasing PA, the outphasing angle coding and (DMP) model presented in [26], which includes three
the calibration of the gain and phase mismatch of the two branches to compensate for in-band, out-of-band, and
signal paths are carried out in the outphasing transforma- crossband-intermodulation distortion and slow-enve-
tion block. Additionally, the time alignment between the RF lope distortion, respectively.
signal and the supply-modulated signal is carried out in the This particular 3D DPD model was used in a concur-
slow-envelope generation block. Finally, the time-alignment rent dual-band transmission in [26] and [30] to compen-
and amplitude normalization between the input/output sate for the unwanted nonlinear distortion of an ET PA.
I/Q signals is done in the DPD adaptation block. For example, Figure 8(a) shows the output spectra of an
ET PA used in [30] before and after 3D DPD. Using the
Experimental Results: ET and Outphasing PAs slow envelope to dynamically supply the PA is a subop-
in a Concurrent Dual-Band Transmission timal solution from the power-efficiency point of view
Let us now investigate an example of the general block when compared with using the original RF signal’s enve-
diagram in Figure 6 for the specific case of a concurrent lope. However, it is the only feasible solution, considering
dual-band transmission. Figure 7 shows the structure the drain modulator’s bandwidth limitations. Moreover,
of the proposed 3D distributed memory polynomial this approach is always more efficient than considering a
fixed supply. Figure 9 (top) sum-
marizes the results presented
in [26] for a concurrent, dual-
H1(z) band ET PA where, even using
X the slow envelope (with both
f1(.) peak and average approaches)
u1[n] .  H1(z)
Intermodulation Distortion to dynamically supply the PA,
we can double the drain effi-
f2(.) H2(z) ciency compared with the fixed
.  x1[n]
supply. In addition, the adjacent
X X + channel leakage ratio thresh-
u2[n] .  g2(.) T2(z)
old of –45 dB relative to carrier
Crossband Intermodulation Distortion is guaranteed by applying 3D
DPD linearization with fewer
. H3(z) than 100 coefficients [26].
.  f3( )
X X Let us go back to the general
Slow g3(.) T3(z) block diagram in Figure 6 to
Env. consider concurrent dual-band
ES [n ] Slow-Envelope Distortion
outphasing amplification. In
particular, we look at mixed-
Figure 7. The structure of the proposed 3D DMP DPD for band 1. or hybrid-mode outphasing

–80
–70
Power Frequency (dB/Hz)

LTE-5 MHz
Power Frequency (dB/Hz)

–90 –80

–100 LTE-10 MHz LTE-5 MHz –90 LTE-5 MHz

–100
–110
–110
–120 –120
–130 –130
–140
–140
1,895 1,910 1,925 1,940 1,955 1,970 1,985 2,000 687.5 697.5 707.5 717.5 727.5 737.5
Frequency (MHz) Frequency (MHz)
(a) (b)

Unlinearized PA Output Linearized Using 3D-DML DPD

Figure 8. The unlinearized and 3D DPD linearized output spectra of (a) an ET PA [26] and (b) a hybrid outphasing PA.

58 May 2019

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 amplification [43] because we allow some amplitude
DPD is a linearization technique
variation in the phase-modulated outphasing compo-
nents. This hybrid approach helps mitigate the bandwidth par excellence that deals with the
expansion that occurs when amplifying pure constant- inherent PA linearity-versus-power
amplitude phase-modulated components. In this way,
efficiency tradeoff.
the hybrid-outphasing PA becomes more linearizable,
even if at the price of slightly degrading power efficiency.
In fact, at deep power backoff, there may be no benefit techniques are applied to keep the number of coef-
in terms of power-added efficiency when operating in a ficients below 100. On the one hand, to select the most
pure outphasing mode. Therefore, both hybrid outphas- relevant basis of the DPD function in the forward path,
ing signal components can be defined at baseband as a greedy algorithm, such as, the orthogonal matching
pursuit [30], [44], can be used. On the other hand, fea-
x outph–1[n] = A[n] e j ({x [n] +a [n]) = x Ioutph–1 [n] + jx Qoutph–1 [n] (5) ture-extraction techniques, such as principal compo-
nent analysis [45] or the partial least squares [30], can
and be used to reduce the number of bases in the adap­tation
subsystem. Figure 9 (bottom) shows that, even with
x outph–2[n] = A[n] e j ({x [n] -a [n]) = x Ioutph–2 [n] + jx Qoutph–2 [n], (6) the slow envelope, the efficiency doubles compared
with not applying any kind of dynamic-load modula-
where half of the outphasing angle, a[n], is defined as tion. By coding the slow envelope into the outphasing
a function of the slow envelope a[n] = f (E[n]), { x[n] angle, an efficient, concurrent dual-band transmission
is the original instantaneous phase of the dual-band could be feasible using a dual-band Chireix topology
signal, and A[n] is the amplitude of the hybrid or [46], [47], with a much less demanding implementa-
mixed-mode outphasing components and, unlike the tion than the wideband counterpart, which would be
pure outphasing components, some amplitude varia- required if coding the original envelope.
tion is allowed: A[n] = (x[n] /E[n]) $ b, with 0 1 b 1 1. To validate the closed-loop DPD algorithms, a com-
Figure 8(a) shows the dual-band output spectra mon solution as a step to a future real-time implementa-
of a hybrid outphasing PA before and after 3D DPD tion is the use of a hardware-in-the-loop architecture,
linearization where, again, dimensionality reduction such as the ones shown in Figure 10. In a PC-controlled

• WCDMA at 1.75 GHz

ET Power Amplifier • LTE-10 MHz at 2.1 GHz

• Fixed Supply • >–43  <–50 (2D DPD) • >–32  <–35 (2D DPD) • 23.2
*Drain Eff. (%)
Configuration

ACLR (dBc)

NMSE (dB)

• ET (Peak Slow Envelope)

• >–25  <–45 (3D DPD) • >–16  <–35 (3D DPD) • 47.4
(EsTH = 0.33)
• >–24  <–45 (3D DPD) • >–16  <–35 (3D DPD) • 47.5
• ET (Avg. Slow Envelope)
(EsTH = 0.23) Mean Output Power: 28.1 dBm
*Only RF PA Eff.

Outphasing Power Amplifier • LTE-5 MHz at 700 MHz

• LTE-5 MHz at 725 MHz

• No Outphasing • >–24  <–36 (2D DPD) • 20.4

Efficiency (%)

• >–30  <–50 (2D DPD)

Configuration

ACLR (dBc)

NMSE (dB)

• Pure Outphasing • >–17  — • >–10  — • 44.9

• Hybrid Outphasing • >–22  <–45 (3D DPD) • >–16  <–34 (3D DPD) • 42.1
(EsTH = 0.3)
Mean Output Power: 36.6 dBm

Figure 9. The 3D DPD linearization of concurrent dual-band transmission for both ET [26] and outphasing PAs. dBc:
decibels relative to the carrier; dBm: dB-mW; NMSE: normalized mean square error; ACLR: adjacent channel leakage ratio.

May 2019 59

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 Économique et Régional (Euro-
Hardware in the Loop pea n Fu nd for Econom ic
RF Signal Generation RF/IF Signal Acquisition and Regional Development)
(DAC + Up-Conversion) (ADC)
under the Ministerio de Cien-
cia, Innovación y Universidades
projects TEC2017-83343-C4-1-R
AWG (Keysight M8190A) and TEC2017-83343-C4-2-R and
DSO (Keysight 9404A)
by the Generalitat de Catalunya
under grant 2017 SGR 813.
DUT
VSG (R&S SMW200A)
SSA (R&S FSW8)
References
[1] R. N. Braithwaite, “General pr i n -
ciples a nd desig n over v iew of
digital pre-distortion,” in Digital
Front-End in Wireless Communica-
TSW1400EVM + ADC32RF45EVM + tions and Broadcasting, F.-L. Luo, Ed.
TSW30H84EVM TSW14J56EVM Cambridge, U.K.: Cambridge Univ.
Press, 2011.
[2] J. Wood, “System-level de sig n
con siderat ion s for d ig ita l pre-
Figure 10. The hardware-in-the-loop general architecture. DSO: digital storage distortion of wireless base station
oscilloscope; DUT: device under test; IF: intermediate frequency; VSG: vector signal transmitters,” IEEE Trans. Microw.
generator; SSA: signal and spectrum analyzer; DSO: digital storage oscilloscope. Theory Techn., vol. 65, no. 5, pp.
1880–1890, May 2017. doi: 10.1109/
TMTT.2017.2659738.
system, the digital signal processing is carried out in [3] A. Katz, J. Wood, and D. Chokola, “The evolution of PA linear-
ization: From classic feedforward and feedback through analog
MATLAB. Then, instrumentation or commercial off-the-
and digital predistortion,” IEEE Microw. Mag., vol. 17, no. 2, pp.
shelf boards are used to both generate (DAC) and up- 32–40, Feb. 2016. doi: 10.1109/MMM.2015.2498079.
convert the RF signal to be amplified and, after going [4] L. Guan and A. Zhu, “Green communications: Digital predistor-
through the PA, down-convert and/or digitize (ADC) the tion for wideband RF power amplifiers,” IEEE Microw. Mag., vol.
15, no. 7, pp. 84–99, Nov.-Dec. 2014. doi: 10.1109/MMM.2014.2356037.
feedback signal.
[5] P. Rysavy, “Challenges and considerations in defining spectrum
efficiency,” in Proc. IEEE, vol. 102, no. 3, pp. 386–392, Mar. 2014.
doi: 10.1109/JPROC.2014.2301637.
Conclusions [6] K. Sundhar and L. C. Miller, 5G for Dummies, Hoboken, NJ: Wiley,
DPD is a linearization technique par excellence that 2017.
deals with the inherent PA linearity-versus-power effi- [7] Z. Wang, “Demystifying envelope tracking: Use for high-efficien-
cy power amplifiers for 4G and beyond,” IEEE Microw. Mag., vol.
ciency tradeoff. In the last decade, DPD has gone 16, no. 3, pp. 106–129, Apr. 2015. doi: 10.1109/MMM.2014.2385351.
multidimensional to address new linearity challenges [8] Z. Popovic, “Amping up the PA for 5G: Efficient GaN power am-
that have arisen due to the inclusion of techniques plifiers with dynamic supplies,” IEEE Microw. Mag., vol. 18, no. 3,
oriented toward maximizing spectral efficiency, such pp. 137–149, May 2017. doi: 10.1109/MMM.2017.2664018.
[9] G. T. Watkins and K. Mimis, “How not to rely on Moore’s law alone:
as concurrent multiband or carrier-aggregated trans- Low-complexity envelope-tracking amplifiers,” IEEE Microw. Mag.,
missions and MIMO and the adoption of high-power- vol. 19, no. 4, pp. 84–94, June 2018. doi: 10.1109/MMM.2018.2813840.
efficient architectures based on dynamic-supply or [10] R. Pengelly, C. Fager, and M. Ozen, “Doherty’s legacy: A histo-
ry of the Doherty power amplifier from 1936 to the present day,”
dynamic-load modulation of the PA. For multiband
IEEE Microw. Mag., vol. 17, no. 2, pp. 41–58, Feb. 2016. doi: 10.1109/
transmission in ET or outphasing PAs, we have shown MMM.2015.2498081.
that one strategy to cope with the bandwidth limitations [11] R. Darraji, P. Mousavi, and F. M. Ghannouchi, “Doherty goes digital:
of both amplification architectures uses a slower version Digitally enhanced Doherty power amplifiers,” IEEE Microw. Mag.,
vol. 17, no. 8, pp. 41–51, Aug. 2016. doi: 10.1109/MMM.2016.2561478.
of the original RF signal’s envelope to both dynami-
[12] T. Barton, “Not just a phase: Outphasing power amplifiers,”
cally supply and load-modulate the PA. Finally, results IEEE Microw. Mag., vol. 17, no. 2, pp. 18–31, Feb. 2016. doi: 10.1109/
considering a concurrent dual-band transmission have MMM.2015.2498078.
shown that power efficiency values can be significantly [13] Z. Popovic and J. A. Garcia, “Microwave class-E power amplifiers:
A brief review of essential concepts in high-frequency class-E PAs
enhanced in ET and outphasing PAs by using slow enve-
and related circuits,” IEEE Microw. Mag., vol. 19, no. 5, pp. 54–66,
lopes and 3D DPD to compensate for the unwanted PA July-Aug. 2018. doi: 10.1109/MMM.2018.2822202.
linear and nonlinear effects. [14] P. Roblin, C. Quindroit, N. Naraharisetti, S. Gheitanchi, and M. Fit-
ton, “Concurrent linearization: The state of the art for modeling and
linearization of multiband power amplifiers,” IEEE Microw. Mag.,
Acknowledgments vol. 14, no. 7, pp. 75–91, Nov. 2013. doi: 10.1109/MMM.2013.2281297.
This work was supported in part by the Spanish Gov- [15] K. Hausmair, P. N. Landin, U. Gustavsson, C. Fager, and T. Eriks-
ernment and the Fonds Européen de Développement son, “Digital predistortion for multi-antenna transmitters affected

60 May 2019

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.
 by antenna crosstalk,” IEEE Trans. Microw. Theory Techn., vol. 66, no. Microw. Theory Techn., vol. 60, no. 2, pp. 361–369, Feb. 2012. doi:
3, pp. 1524–1535, Mar. 2018. doi: 10.1109/TMTT.2017.2748948. 10.1109/TMTT.2011.2176956.
[16] C. Yu, W. Cao, Y. Guo, and A. Zhu, “Digital compensation for [32] S. A. Bassam, M. Helaoui, and F. M. Ghannouchi, “Crossover digi-
transmitter leakage in non-contiguous carrier aggregation appli- tal predistorter for the compensation of crosstalk and nonlinearity
cations with FPGA implementation,” IEEE Trans. Microw. Theo- in MIMO transmitters,” IEEE Trans. Microw. Theory Techn., vol. 57,
ry Techn., vol. 63, no. 12, pp. 4306–4318, Dec. 2015. doi: 10.1109/ no. 5, pp. 1119–1128, May 2009. doi: 10.1109/TMTT.2009.2017258.
TMTT.2015.2495144. [33] A. Abdelhafiz, L. Behjat, F. M. Ghannouchi, M. Helaoui, and
[17] M. Younes, A. Kwan, M. Rawat, and F. M. Ghannouchi, “Lineariza- O.  Hammi, “A high-performance complexity reduced behavioral
tion of concurrent tri-band transmitters using 3-D phase-aligned model and digital predistorter for MIMO systems with cross-
pruned Volterra model,” IEEE Trans. Microw. Theory Techn., vol. 61, talk,” IEEE Trans. Commun, vol. 64, no. 5, pp. 1996–2004, May 2016.
no. 12, pp. 4569–4578, Dec. 2013. doi: 10.1109/TMTT.2013.2287176. doi: 10.1109/TCOMM.2016.2545654.
[18] S. Amin, P. Händel, and D. Rönnow, “Digital predistortion of sin- [34] F. M. Barradas, P. M. Tomé, J. M. Gomes, T. R. Cunha, P. M. Ca-
gle and concurrent dual-band radio frequency GaN amplifiers with bral, and J. C. Pedro, “Power, linearity, and efficiency prediction
strong nonlinear memory effects,” IEEE Trans. Microw. Theory Techn., for MIMO arrays with antenna coupling,” IEEE Trans. Microw.
vol. 65, no. 7, pp. 2453–2464, July 2017. doi: 10.1109/TMTT.2016.2642948. Theory Techn., vol. 65, no. 12, pp. 5284–5297, Dec. 2017. doi: 10.1109/
[19] A. Molina, K. Rajamani, and K. Azadet, “Concurrent dual-band TMTT.2017.2766067.
digital predistortion using 2-D lookup tables with bilinear inter- [35] Q. Luo, C. Yu, and X. W. Zhu, “Digital predistortion of phased
polation and extrapolation: Direct least squares coefficient adapta- array transmitters with multi-channel time delay,” in 2018 IEEE
tion,” IEEE Trans. Microw. Theory Techn., vol. 65, no. 4, pp. 1381–1393, Topical Conf. RF/Microwave Power Amplifiers for Radio and Wireless
Apr. 2017. doi: 10.1109/TMTT.2016.2634001. Applications (PAWR), Anaheim, CA, 2018, pp. 54–57.
[20] S. A. Bassam, M. Helaoui, and F. M. Ghannouchi, “2-D digital [36] X. Liu et al. “Beam-oriented digital predistortion for 5G massive
predistortion (2-D-DPD) architecture for concurrent dual-band MIMO hybrid beamforming transmitters,” IEEE Trans. Microw.
transmitters,” IEEE Trans. Microw. Theory Techn., vol. 59, pp. 2547– Theory Techn., vol. 66, no. 7, pp. 3419–3432, July 2018. doi: 10.1109/
2553, Oct. 2011. doi: 10.1109/TMTT.2011.2163802. TMTT.2018.2830772.
[21] F. Mkadem, A. Islam, and S. Boumaiza, “Multi-band complexity- [37] M. Abdelaziz, L. Anttila, A. Brihuega, F. Tufvesson, and M. Valka-
reduced generalized-memory-polynomial power-amplifier digital ma, “Digital predistortion for hybrid MIMO transmitters,” IEEE
predistortion,” IEEE Trans. Microw. Theory Techn., vol. 64, no. 6, J.  Sel. Topics in Signal Process., vol. 12, no. 3, pp. 445–454, June
pp. 1763–1774, June 2016. doi: 10.1109/TMTT.2016.2561279. 2018. doi: 10.1109/JSTSP.2018.2824981.
[22] W. Chen et al., “Design and linearization of concurrent dual-band [38] M. N. Ruiz, A. L. Benito, J. R. Pérez-Cisneros, P. L. Gilabert, G. Mon-
Doherty power amplifier with frequency-dependent power rang- toro, and J. A. García, “Constant-gain envelope tracking in a UHF
es,” IEEE Trans. Microw. Theory Tech, vol. 59, no. 10, pp. 2537–2546, outphasing transmitter based on continuous-mode class-E GaN
2011. doi: 10.1109/TMTT.2011.2164089. HEMT PAs,” 2016 IEEE MTT-S Int. Microwave Symp. (IMS), San Fran-
[23] R. N. Braithwaite, “Digital predistortion of a power amplifier for cisco, CA, 2016, pp. 1–4.
signals comprising widely spaced carriers,” in Proc. 78th Microwave [39] G. Wimpenny, “Envelope tracking PA characterisation,” Open ET
Measurement Symp. (ARFTG), 2011, pp. 1–4. Alliance, Cambridge, U.K., White Paper, Nov. 2011.
[24] J. Jeong, D. Kimball, M. Kwak, C. Hsia, P. Draxler, and P. Asbeck, [40] P. J. Brehm and T. W. Barton, “Modeling and analysis of the fre-
“Wideband envelope tracking power amplifiers with reduced quency dependence of class-E outphasing,” in Proc. 13th Int. Conf.
bandwidth power supply waveforms and adaptive digital predis- Advanced Technologies, Systems and Services in Telecommunications
tortion techniques,” IEEE Trans. Microw. Theory Techn., vol. 57, no. (TELSIKS), 2017, pp. 170–173.
12, pp. 3307–3314, Dec. 2009. [41] H. Gheidi, P. Theilmann, J. J. Yan, T. Nakatani, and D. Kimball, “A
[25] G. Montoro, P. L. Gilabert, J. Berenguer, and E. Bertran, “Digital broadband envelope-tracking push-pull GaN power amplifier us-
predistortion of envelope tracking amplifiers driven by slew-rate ing grounded-coplanar ring Marchand balun,” in Proc. 2018 IEEE
limited envelopes,” in 2011 IEEE MTT-S Int. Microwave Symp. Dig. 19th Wireless and Microwave Technology Conf. (WAMICON), Sand
(IMS), June 2011, pp. 1–3. doi: 10.1109/MWSYM.2011.5972693. Key, FL, 2018, pp. 1–4.
[26] P. L. Gilabert and G. Montoro, “3-D distributed memory polyno- [42] A. Cidronali, G. Manes, N. Giovannelli, T. Vlasits, and R. Hernaman,
mial behavioral model for concurrent dual-band envelope track- “Efficiency and linearity enhancements with envelope shaping con-
ing power amplifier linearization,” IEEE Trans. Microw. Theory trol in dual-band envelope tracking GaAs PA,” in Proc. 6th European
Techn., vol. 63, no. 2, pp. 638–648, Feb. 2015. doi: 10.1109/TMTT.2014 Microwave Integrated Circuit Conf., Manchester, U.K., 2011, pp. 308–311.
.2387825. [43] J. Qureshi et al., “A 90-W peak power GaN outphasing ampli-
[27] H. Sarbishaei, Y Hu, B. Fehri, and S. Boumaiza, “Concurrent dual- fier with optimum input signal conditioning,” IEEE Trans. Microw.
band envelope tracking power amplifier for carrier aggregated sys- Theory Techn., vol. 57, no. 8, pp. 1925–1935, Aug. 2009. doi: 10.1109/
tems,” in 2014 IEEE MTT-S Int. Microwave Symp. (IMS2014), Tampa, TMTT.2009.2025430.
FL, 2014, pp. 1–4. [44] P. L. Gilabert, D. López-Bueno, and G. Montoro, “Spectral weight-
[28] Y. Lin, C. Quindroit, H. Jang, and P. Roblin, “3-D Fourier series ing orthogonal matching pursuit algorithm for enhanced out-of-
based digital predistortion technique for concurrent dual-band band digital predistortion linearization,” IEEE Trans. Circuits Syst.
envelope tracking with reduced envelope bandwidth,” IEEE Trans. II, Express Briefs, doi: 10.1109/TCSII.2018.2878581.
Microw. Theory Techn., vol. 63, no. 9, pp. 2764–2775, Sept. 2015. doi: [45] D. López–Bueno, Q. A. Pham, G. Montoro, and P. L. Gilabert,
10.1109/TMTT.2015.2452271. “Independent digital predistortion parameters estimation us-
[29] A. K. Kwan, M. Younes, R. Darraji, and F. M. Ghannouchi, “On ing adaptive principal component analysis,” IEEE Trans. Microw.
track for efficiency: Concurrent multiband envelope-tracking pow- Theory Techn., vol. 66, no. 12, pp. 5771–5779, Dec. 2018. doi: 10.1109/
er amplifiers,” IEEE Microw. Mag., vol. 17, no. 5, pp. 46–59, May 2016. TMTT.2018.2870420.
doi: 10.1109/MMM.2016.2525118. [46] M. N. Ruiz, R. Marante, L. Rizo, J. A. García, P. L. Gilabert, and G.
[30] Q. A. Pham, D. López-Bueno, T. Wang, G. Montoro, and P. L. Gi- Montoro, “A dual-band outphasing transmitter using broadband
labert, “Partial least squares identification of multi look-up table class E power amplifiers,” in Proc. 2014 Int. Workshop Integrated Non-
digital predistorters for concurrent dual-band envelope tracking linear Microwave and Millimetre-Wave Circuits (INMMiC), Leuven,
power amplifiers,” IEEE Trans. Microw. Theory Techn., vol. 66, no. 12, Belgium, 2014, pp. 1–3.
pp. 5143–5150, Dec. 2018. doi: 10.1109/TMTT.2018.2857819. [47] P. H. Pednekar and T. W. Barton, “Dual-band Chireix combining
[31] H. Cao, H. M. Nemati, A. Soltani Tehrani, T. Eriksson, and C. network,” in Proc. 2016 Texas Symp. Wireless and Microwave Circuits
Fager, “Digital predistortion for high efficiency power amplifier and Systems (WMCS), Waco, TX, 2016, pp. 1–4.
architectures using a dual-input modeling approach,” IEEE Trans. 

May 2019 61

Authorized licensed use limited to: University of Electronic Science and Tech of China. Downloaded on August 11,2022 at 20:11:12 UTC from IEEE Xplore. Restrictions apply.