Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

WeIF1-17

Investigation on a Desirable DPD Architecture and Trapping


Characteristics for GaN Power Amplifier Linearization
P. Song, Z. Mokhti, Q. Mu
Wolfspeed, Morgan Hill, CA, USA
{philip.song, zul.mokhti, qianli.mu}@wolfspeed.com

Abstract— A 2-level digital predistortion (DPD), a simplified made from understanding the GaN trapping characteristics and
version of a piecewise DPD, is designed and used to verify that a making it favourable for piecewise DPD.
piecewise DPD is effective in linearizing typical GaN devices in To verify that piecewise DPD is effective for GaN and to
industry that exhibit low power distortions due to trapping. A investigate if certain GaN device characteristics can work better
further investigation is taken to understand the device
characteristics that are favored by a piecewise DPD by
with piecewise DPD, a 2-level DPD test bench is designed
characterizing various devices. using MATLAB, which is a simplified version of piecewise
2021 IEEE/MTT-S International Microwave Symposium - IMS 2021 | 978-1-6654-0307-8/21/$31.00 ©2021 IEEE | DOI: 10.1109/IMS19712.2021.9574877

Keywords— piecewise DPD, 2-level DPD, GaN power DPD. Two groups of GaN devices that have different trapping
amplifiers, trapping, memory effect, GMP. characteristics typically encountered in the marketplace are
tested in a Doherty circuit using both 2-level and GMP DPDs,
I. INTRODUCTION and results are compared.
Following the advancements in communication systems
such as mMIMO (massive multiple-input multiple-output) and II. PIECEWISE DPD ADVANTAGES FOR GAN
active phased array, power amplifiers (PAs) are moving Most DPD focuses on correcting the high-power
towards the direction of smaller physical size with higher compression as it is the primary source for distortion in power
integration and a larger number of elements where each element amplifiers. For GaN power amplifiers, trapping causes an
can work for wider signal bandwidth. Such system-level additional non-linearity which is more prominent in low power
development is particularly important for modern commercial regions, and it is difficult for a DPD to correct both the low
communication infrastructure systems, like 5G, which enables power and high power non linearities in a single model. A
wider data bandwidth and larger coverage capacity. piecewise DPD can be effective in correcting devices with such
Power amplifiers based on GaN technology serve as good distortions because it can separately model the non-linearities
candidates for such applications due to its advantages on power in each power region.
density and efficiency. However, trapping effect in GaN high One of the challenges in piecewise DPD is handling
electron mobility transistors (HEMTs) causes challenges to memory effects in different power segments. However, for GaN
incumbent linearization techniques. Trapping causes dynamic HEMTs with long trapping release time constant, this may not
gain collapses during PA operation under modulated signals, be a big concern. The trap release is much slower than typical
which makes it difficult to fit the gain responses to memory IQ sampling rates, and such a long-term memory effect
polynomials for digital pre-distortion (DPD). Such dynamic correction is not effective with a DPD [4]. Therefore, it is
gain collapse behaviour could lead to poor EVM performance, possible, for such GaN PA, to simplify the piecewise DPD
especially when operating under high PAPR signal conditions, structure by excluding the memory effect correction and relying
like ETM2a signal. This dynamic gain collapse behaviour in on other system level solutions to handle the long-term memory
GaN HEMT is related to trapping consistency and trapping effects.
capture and release time constants.
In this paper, the discussion is focused on trapping release, III. PIECEWISE DPD VERIFICATION TESTS
or recovery, time constant. For long trapping release time A. Test setup
constant, more than 100ms, trapping impact on gain collapse is
relatively stable. Piecewise DPD can be more effective on A simplified version of the piecewise DPD, which we call
correcting such gain response than a regular memory 2-level DPD, is designed to divide the power into two ranges
polynomial DPD is. Piecewise DPD is a proven DPD technique and extract DPD coefficients separately for each region as there
but may not be widely used because popular devices such as is no commercially available piecewise DPD. A test bench
LDMOS (laterally diffused metal oxide semiconductor) and equipped with the 2-level DPD is used to experimentally
GaAs (gallium arsenide) are not impacted by trapping from confirm that a piecewise DPD can be effective for a GaN power
linearity aspect and do not have such gain response that needs amplifier and to study the device characteristics that a
a piecewise DPD. piecewise DPD favours.
A piecewise DPD has been tried on GaN PAs [1, 3], as the The tests were done using an ETM2a waveform, which is a
authors noted that the stronger nonlinearity from GaN requires 3GPP-defined test waveform that has alternating 2 or 3 low
a better linearization technique. The references focus on the power data symbols and a high-power sync symbol as in Fig. 1.
piecewise DPD technique itself. Further improvements can be The impact on the low power symbols by trapping is of great

978-1-6654-0307-8/21/$31.00 © 2021 IEEE 531 2021 IEEE/MTT-S International Microwave Symposium


Authorized licensed use limited to: Amrita School of Engineering. Downloaded on June 24,2024 at 10:47:57 UTC from IEEE Xplore. Restrictions apply.
interest, and EVMs are of primary interest, rather than ACLR B. Test Results
which is dominated by high power symbols. Fig. 3 compares the linearity of the first and second groups
of the devices when using the 2-level DPD and GMP
(generalized memory polynomial) DPD. The plots show the
rms EVM for each of the 140 symbols, which is equivalent to
10mS. The average and max rms symbol EVM for different
cases are compared and summarized in Table 1.
Group 2, which has a larger static gain drop and a smaller
dynamic gain variation, shows significant improvement using
the 2-level DPD over the GMP DPD. No noticeable
improvement is seen for Group 1 devices from using 2-level
DPD against GMP DPD in terms of corrected EVMs.
Fig. 1. ETM2a test waveform over one frame (10mS or 140 symbols)

Two groups of devices, 5 for each, are selected which have


clearly different trap characteristics for this investigation and
tested in a 2-way symmetric Doherty at 2.6GHz. Group 1 has a
smaller static gain drop and a larger dynamic gain variation.
Devices from this group have ~10ms trap release time constant
at room temperature from PIV based transient current
measurement. The static gain drop is the average gain drop over
the period of time observed, and the dynamic gain variation is
the instantaneous gain variations around the average gain drop,
indicating gain stability vs time. Trap release time constant is
defined as time t that gives max ( ) in the Id recovery curve.
(a)
Group 2 has a larger static gain drop and a smaller dynamic gain
variation. They have ~200ms trap release time constant at room
temperature.
The static gain drops and the trap release time constants for
Group 1 and Group 2 devices are illustrated in Fig. 2.

(b)

(a) Fig. 3. Corrected EVM for 2-level DPD vs GMP DPD: (a) on Group 1 devices
(smaller static gain drop and larger dynamic gain variation); (b) on Group 2
devices (larger static gain drop and smaller dynamic gain variation)

Table 1. EVM comparisons between 2-level DPD vs GMP DPD on Group 1


and 2 devices
GMP DPD 2-level DPD
Max rms Avg rms Max rms Avg rms
symbol symbol symbol symbol
EVM EVM EVM EVM
Group 1 7.9% 4.9% 7.6% 5.1%
Group 2 8.7% 5.6% 5.0% 2.7%
(b)
Fig. 2. static gain drops and the trap release time constants: (a) Static gain drops C. Analysis
of Group 1 and 2 devices; (b) Quiescent drain current recovery and release time
constant τ of Group 1 and 2 devices when pulsed to (-8V,100V) for 50us. The test results indicate that piecewise DPD may not offer
an advantage over GMP if the dynamic gain variation is large,

532
Authorized licensed use limited to: Amrita School of Engineering. Downloaded on June 24,2024 at 10:47:57 UTC from IEEE Xplore. Restrictions apply.
i.e. if the gain drop consistency is not good. An advanced DPD the gain in a low power burst drops more when the preceding
such as piecewise DPD can correct a more serious non-linear high power burst has a higher amplitude. It indicates that not
distortion but cannot do much if the distortion is not consistent only the trap release speed but also the static gain drop if it is
over time or is not predictable. too big can cause a significant enough dynamic gain variation.
The gain drop versus time plots for Group 1 and Group 2
devices are shown in Fig. 4, which helps to understand why a IV. SUMMARY
piecewise DPD is not effective for a device that has a larger Piecewise DPD is suitable for devices that have
dynamic gain variation. The gain variations shown in the plots nonlinearities over a large power range such as GaN because it
are due to the long-term memory effect related to trapping, and can model each power segment separately. To prove the
most sample based DPD is not effective in correcting long term effectiveness of piecewise DPD, a simplified version, 2-level
memory effect. DPD, is designed and tested. The data shows that 2-level DPD
The best way forward to minimize such gain variation is in can achieve linearized average EVM around 2.5% vs. 5% when
the device technology. Fig. 4b for Group 2 devices show less GMP is used.
variation and as a result, better linearization with the 2-level But not both groups of GaN devices that we tested showed
DPD compared to Group 1 devices which have a larger good improvements with 2-level DPD. Group1 devices have
variation. Note that if it were not for 2-level DPD, Group 2 faster trap release time constant, which leads to larger dynamic
would not demonstrate better linearization due to more severe gain variation under large PAPR signals, like ETM2a. The
static gain drop, but the static gain drop is well handled by 2- longer trap release time constant of Group 2 devices helps to
level DPD even if it is large. reduce dynamic gain variation. Therefore, 2-level DPD is more
effective on improving its linearity.
A piecewise DPD can linearize PAs that have large static
gain drop but not ones with large dynamic gain variations. On
the other hand, designing GaN devices to have slower trap
release time constants can improve linearization when
piecewise DPD is used. However, trap release time constant is
not the only parameter that needs to get attention to when
reducing trapping impacts from GaN HEMTs on PA’s linearity.

ACKNOWLEDGMENT
The authors gratefully acknowledge the encouragement and
support of B.Noori, J.Fisher, J.Wood, and H.Jang and
(a)
appreciate technical discussions.

REFERENCES
[1] G. Baudoin, “On segmented predistortion for linearization of RF power
amplifiers”, Radioengineering, vol. 29, No. 1, pp. 21-36, April, 2020.
[2] J. Wood, Behavioral modeling and linearization of RF power amplifiers,
United States, Artech House Publishers, 2014, ISBN:978-1-60807-120-
3.
[3] D. Lopez-Bueno, T. Wang, P. L. Gilabert, and G. Montoro, “Digital
predistortion linearization strategies for power amplifiers under
wideband 4G/5G burst-like waveform operation”, IMS2015 student
design competition winner.
[4] P. M. Tome, F. M. Barradas, T. R. Cunha, and J. C. Pedro, “Hybrid
analog/digital linearization of GaN HEMT-based power amplifiers”,
IEEE Transactions on microwave theory and techniques, vol. 67, No. 1,
pp. 288-294, January 2019.
(b)
Fig. 4. Dynamic gain variation (or gain drop consistency): (a) Group 1 devices;
(b) Group 2 devices.

The larger dynamic gain variation of Group 1 devices is fast


trap release time constant related. The release action is more
clearly seen in low power bursts, which are 140uS or 210uS
long, corresponding to 2 or 3 symbol lengths, in ETM2a
waveform. Since the gain recovers faster, it rises more during
the low power bursts resulting in larger gain variation.
However, there are other factors that affect the dynamic
gain variation. Although Group 2 in Fig. 4b shows less gain
variation related to the trap release speed, another type of
variation is related to the amplitude of the high-power bursts;

533
Authorized licensed use limited to: Amrita School of Engineering. Downloaded on June 24,2024 at 10:47:57 UTC from IEEE Xplore. Restrictions apply.

You might also like