A C Sherihan, P P Shameena, Shamna K K Ayisha, International Journal of Advance Research, Ideas and Innovations in

Technology.

ISSN: 2454-132X
Impact factor: 4.295
(Volume 4, Issue 1)
Available online at www.ijariit.com

A Survey on Closed-Loop Digtal Predistortion (DPD) Used to

Linearize RF Power Amplifier
Sherihan A C Shameena P P Ayisha Shamna K K
Sherihan20@gmail.com Shamzz89@gmail.com ayishamna@gmail.com
Cochin College of Engineering Cochin College of Engineering Cochin College of Engineering
and Technology, Valanchery, and Technology, Valanchery, and Technology, Valanchery,

ABSTRACT
Digital predistortion (DPD) used to linearize an RF power waveform including distortion caused by nonlinearities
amplifier (PA) can achieve wide bandwidth distortion within the transmit path.
cancellation. Predistortion technique is one of the most The transmit path includes a DPD module, digital-to analog
effective linearization technique. Model based DPDs also converter (DAC), modulator, and PA. DPD has a nonlinear
called polynomial based DPDs, are mostly used for the response intended to compensate for the subsequent PA
linearization of PA. For memory less system, the nonlinearity so that the transmit path appears linear. The
polynomial model can be used to model the predistorter. DPD module is made adaptive using coefficients selected by
For memory system, the models such as Volterra series an estimator that attempts to minimize the residual distortion
model, generalized memory polynomial model, measured by the observation path.
Hammerstein model and Wiener model all are used to
model the predistorter. DPD architectures can be classified
into two types Indirect Learning Architecture (ILA) and
Direct Learning Architecture (DLA).In this paper a survey
on different model digital predistortion used to linearize
RF power amplifier in different works.
Keywords: Amplifier Distortion, Digital Predistortion,
Direct Learning Architecture (DLA), Memory less System,
Power Amplifier (PA).
1. INTRODUCTION
Fig-1 Digital Transmitter with DPD
A Transmitter within a wireless base station converts a
digital baseband input signal into an RF output signal. The Digital Pre distortion (DPD) is one of the most effective
final active stage of the transmitter, the power amplifier techniques to mitigate the distortions caused by power
(PA), is often designed to operate near saturation to amplifier (PA) nonlinearity and memory effects. As the
maximize efficiency. However, the PA exhibits a nonlinear input signal bandwidth increases, the required bandwidth on
behavior under such conditions, which must be linearized. the DPD feedback channel becomes even larger, i.e.,
The transmitter, shown in Fig. 1, comprises a transmit path normally five times the signal bandwidth. However, the
and observation path. The transmit path performs the desired DPD feedback bandwidth is often restricted by the non ideal
conversion of the digital baseband signal into the electronic components, e.g., the anti-aliasing filter and
amplified RF output signal .Part of the RF output associated circuits, which therefore introduce bandwidth
mismatch between the PA model basis functions and the
signal is coupled into the observation path, which converts it
feedback signal, and thus degrade the linearization
into a digital baseband output signal . This observation
performances of the DPD. DPD structure, a feedback
signal is used to monitor the quality of the transmitted channel is required to down convert, filter, and capture the
© 2018, www.IJARIIT.com All Rights Reserved Page | 599
 A C Sherihan, P P Shameena, Shamna K K Ayisha, International Journal of Advance Research, Ideas and Innovations in
Technology.

PA output signal. Conventionally, the bandwidth of the PA the error signal to produce the update for DPD parameters.
output signal reaches about five times that of the input Comparing with the indirect learning architecture, the use of
signal due to the PA spectral regrowth. the PA model may lead to additional computation
requirements. But the direct approach enjoys benefits such
2. RELATED WORKS as avoiding parameter bias caused by measurement noise
The DPD coefficients are tuned using a closed loop and more robust to spurs and distortions in feedback data.
estimator, which compares the input and output signals, and, 2.3 Digital Predistortion Architecture Using Constrained
respectively. This places the DPD module within the Feedback Bandwidth for Wideband Power Amplifiers
estimation loop, as shown in Fig. 1. Many researchers use an
open-loop estimator where the PA nonlinearity (or its Ying Liu et al [2] proposed a general DPD architecture for
inverse) is estimated based on the pre distorted signal and wideband PA systems with constrained feedback bandwidth.
the output signal. The DPD coefficients are derived By using linear operations to cancel the bandwidth
subsequently from the estimate. This places the DPD mismatch between the proposed model and the PA feedback
module outside of the estimation loop, Different models signal, the full-band PA model parameters can be estimated
DPD used in linearization. with bandwidth-limited observations. This estimated PA
model is subsequently used with the PA input signal to
2.1 Digital Baseband Predistorter Constructed extract the DPD function by applying the direct learning
Using Memory Polynomials algorithms.DPD architecture reduces the feedback
Lei Ding and Tong Zhou et al [7] proposed PA model and bandwidth to less than two times that of the input signal,
building a corresponding predistorter, directly on the while it maintains its linearization performance, as in the
predistorter structure .A memory polynomial model for the full-band case.
predistorter and implement it using an indirect learning DPD architecture with constrained feedback bandwidth is
architecture. Linearization performance is demonstrated on a proposed to effectively compensate for the distortions
three-carrier WCDMA signal. Memoryless PA (i.e., the caused by wideband PA nonlinearities and memory effects.
current output depends only on the current input), The effect of feedback bandwidth restriction on the
memoryless predistortion is sufficient. Higher power conventional MP model identification is analyzed, a band-
amplifiers such as those used in wireless base stations limited PA model extraction method is proposed to mitigate
exhibit memory effects. The cause of memory effects can be the effect of feedback bandwidth restriction on a three-box
electrical or electro thermal .Nonlinear PA with memory, its PA model, where cross terms are included along with the
inverse must also be a nonlinear system with memory. A conventional MPs. Direct learning algorithm is extended to
memory polynomial is a good model for the PA, predistorter an iterative form to cover the proposed PA model and derive
parameters are easy to extract, involving only linear least the DPD function. Due to no restriction being placed on the
squares. The effectiveness of predistortion is demonstrated DPD function derivation, this general DPD architecture can
on a W–H system, a memory polynomial nonlinearity, a compensate for the distortion over the transmission
perturbed Wiener (full Volterra) system, and a parallel bandwidth without either using a high-performance RF filter
Wiener model. to suppress the sidelobe or combining multiple narrowband
2.2 Digital Predistortion using Direct Learning observations.
With Reduced Bandwidth Feedback Microwave cavity filters are inserted to the feedback path to
Lei Ding et al [3] proposed model, it is capable of achieving restrict the bandwidth. The modeling and linearization
near full-rate DPD performance and linearization bandwidth performances clearly proposed DPD architecture can
with significantly reduced feedback bandwidth. mitigate the bandwidth constrains of the conventional
Measurement results on a Doherty system and provide excellent results even when the DPD
PA achieved more than 20 dB corrections over 200 MHz feedback bandwidth is restricted to less than two times the
bandwidth for a 2-carrier WCDMA signal spanning 40 MHz input signal bandwidth. Linear operations to the proposed
with only 81.92 MHz feedback bandwidth. The direct PA model basis functions and the band-limited feedback
learning architecture directly constructs a perverse, i.e., signal, the full-band PA model parameters can be extracted
DPD, of a PA by minimizing the error between the original from band-limited feedback observations.
input x(n) and the PA output z(n) as shown in Fig. 2. 2.4 Digital Predistorters Using Principal
Component Analysis
Gabriel Montoro1 et al [4] proposed apply order reduction
in wide-band digital predistortion (DPD) linearizers using
the principal component analysis (PCA) technique. PCA is a
well-known technique suitable for converting a basis of
observed and eventually correlated data into a basis of
uncorrelated data. This property of eliminating redundancies
can be used for order reduction in linear regression
problems. This approach relaxes the computational load of
Fig-2 Direct Learning Architecture the subsystem responsible for assisting the real-time FPGA
To generate the gradient for DPD parameter adaptation, a device in the task of updating the DPD parameters, such as
PA model is typically required and estimated from PA input for example, a soft-core microprocessor (e.g. Xilinx
and output data. For each iteration in the training process, Microblaze) or any other microprocessor device.
the PA model is first updated and then used together with

© 2018, www.IJARIIT.com All Rights Reserved Page | 600

 A C Sherihan, P P Shameena, Shamna K K Ayisha, International Journal of Advance Research, Ideas and Innovations in
Technology.

the DPD coefficients are biased due to the errors.

Conventional compensation techniques for these errors (I/Q
imbalance, dc offset, and/or nonlinearity) using post-
compensators based on separate modeling a modified least
squares (MLS) extraction method is proposed.
2.7 Closed-Loop Digital Predistortion (DPD) Using an
Observation Path with Limited Bandwidth
R. Neil Braithwaite et al [1] proposed system having (DPD)
used to linearize an RF power amplifier (PA) can achieve
Fig-3 Direct Learning Approach wide bandwidth distortion cancellation using measurements
Applying order reduction techniques also improves the obtained from a narrow bandwidth observation path. The
conditioning of the basis waveforms used. DPD module creates a correction signal using a set of
nonlinear basis waveforms, weighted by adjustable
2.5 Digital Predistortion of Power Amplifier based on coefficients. The coefficients are optimized using a closed-
Compound Memory Polynomial loop estimator. Using superposition that the basis waveform
Zhuohui He et al [5] proposed a compound memory sets used in the DPD module and closed-loop estimator can
polynomial (CMP) model to enhance the accuracy. The differ by a linear transformation, which includes filtering.
CMP model is constructed by adding the term concerning By filtering the waveforms presented to the estimator,
the difference of input to the MP model. coefficients are optimized for distortion cancellation in a
specific part of the spectrum corresponding to the observed
bandwidth. The narrow bandwidth coefficient estimate
provides wide bandwidth distortion cancellation when used
by a DPD module having the unfiltered basis waveform set.
Success of the approach relies on further filtering within the
estimator to attenuate (notch) the input signal bandwidth,
thereby reducing biases in the closed-loop estimation. The
closed-loop estimator is forgiving with respect to in-band
errors within the transmitter and made robust when the in-
(a) band notch attenuation, observation bandwidth, and model
order of the estimator are selected properly. As the input
signal bandwidth increases, reducing in-band errors (or
EVM) in the transmitter becomes more challenging, making
the closed-loop estimator the better choice. Signals
presented to the coefficient estimator are filtered: to fit
within the bandwidth of the observation path and to
attenuate the in-band errors. Such filtering reduces
sensitivity to linear impairments within the transmitter. If
adequate in-band suppression is not possible, the model
order of the estimation is reduced to avoid degradations
outside of the observation bandwidth.
2.8 Digital Predistortion to Power Amplifiers Used in
Third Generation Systems

(b) B. Abdulrahman et al [8] proposed a new method of

Fig-4 Block Diagram of (a) MP Model adaptation, called the slope-dependent is then Introduced
(b) CMP Model and compared with the direct method concerning their time
of convergence and the residual error after convergence. The
2.6 Least Squares Extraction for Volterra Series Digital Slope-dependent Method is a simplified stochastic gradient
Predistorter in the Presence of Feedback Measurement adaptation algorithm. The motivation for this choice is its
Errors simplicity in implementation especially when dealing with
You-Jiang Liu et al [6] proposed a generalized analysis for high rate systems such as W-CDMA. It is based on the
the Volterra-series DPD system is presented in the presence slope of the P.A.
of feedback measurement errors. The DPD coefficients are 3. CONCLUSION
biased due to these errors. A modified least squares (MLS)
method is then proposed for DPD coefficients extraction, This review presents a detailed survey of different model
which can eliminate the detrimental effect of feedback digital predistortion used to linearize RF power amplifier.
measurement errors without using a post-compensator. Polynomial based DPDs, are mostly used for the
linearization of PA. All Digital predistortion easy to
The proposed MLS method has the advantage of being free implement, doesn’t have delay, doesn’t consume large
of behavioral modeling for the feedback path or the post- power, and cancelling distortions.
compensator. The feedback measurement errors have an
important effect on the Volterra-series DPD. A generalized
analysis based on the indirect learning architecture and least
squares (LS) extraction algorithm is presented to show that
© 2018, www.IJARIIT.com All Rights Reserved Page | 601
 A C Sherihan, P P Shameena, Shamna K K Ayisha, International Journal of Advance Research, Ideas and Innovations in
Technology.

4. REFERENCES
[1]R. Neil Braithwaite,“Closed-Loop Digital Predistortion
(DPD) Using an Observation Path With Limited
Bandwidth,” in IEEE MTT-S Int. Microw. Symp. Dig.,
Vol.63, no.2, pp.726-736, Feb 2015
[2] Y. Liu, W. Pan, S. Shao, and Y. Tang, “A new digital
predistortion using indirect learning with constrained
feedback bandwidth for wideband power amplifiers,” in
IEEE MTT-S Int. Microw. Symp. Dig., Tampa, FL, USA,
Jun. 1–6, 2014, pp. 1–3.
[3] L. Ding, F. Mujica, and Z. Yang, “Digital predistortion
using direct learning with reduced bandwidth feedback,” in
IEEE MTT-S Int. Microw. Symp. Dig., Seattle, WA, USA,
Jun. 2–7, 2013, pp. 1–3.
[4] P. L. Gilabert, G. Montoro, D. López, N. Bartzoudis, E.
Bertran, M. Payaró, and A. Hourtane, “Order reduction of
wideband digital predistorters using principal component
analysis,” in IEEE MTT-S Int. Microw. Symp. Dig., Seattle,
WA, USA, Jun. 2–7, 2013, pp. 1–4
[5] Zhuohui He,Wu Ye,and Suili Feng,”Digital predistortion
of power amplifier based on Compound memory polynomial
model,”IEICE Electronic express,Vol.10,No.21,1-5
[6] Y.-J. Liu, W. Chen, J. Zhou, F. M. Ghannouchi, and Y.-
N. Liu, “Modified least squares extraction for Volterra-
series digital predistorter in the presence of feedback
measurement errors,” IEEE Trans. Microw. Theory Techno.
vol. 60, no. 11, pp. 3559–3570, Nov. 2012.
[7] L. Ding et al., “A robust digital baseband predistorter
constructed using memory polynomials,” IEEE Trans.
Commun., vol. 52, no. 1, pp. 159–165, Jan. 2004.
[8] B. Abdulrahman, G.Baudoin,”Applying Digital
Predistortion to Power Amplifiers Used in Third Generation
Systems,” ESIEE, Signal Processing and
Telecommunications Department, BP-99, 93162, Noisy-Le-
Grand CEDEX.

© 2018, www.IJARIIT.com All Rights Reserved Page | 602