Piecewise-Polynomial and Cascade Models of Predistorter
for Linearization of Power Amplifier
E. B. Solovyeva
Saint Petersburg Electrotechnical University “LETI”, St. Petersburg, Russia
Received in final form February 27, 2012
Abstract—To combat non-linear signal distortions in a power amplifier we suggest using predistorter
with cascade structure in which first and second nodes have piecewise-polynomial and polynomial
models. On example of linearizing the Winner–Hammerstein amplifier model we demonstrate that
cascade structure of predistorter improves precision of amplifier’s linearization. To simplify
predistorter’s synthesis the degree of polynomial model used in first node should be moderate, while
precision should be improved by higher degree of second node’s polynomial.
DOI: 10.3103/S0735272712080055
In digital communications systems power amplifier is an important node, which in case of high energetic
characteristics (high efficiency) becomes a source of non-linear distortions. Approximately reciprocal
dependency between efficiency and linearity of a power amplifier is well known. Non-linearity in power
amplifier causes extensions of converted signals’ spectra which leads to inter-channel interference [1, 2].
Among different ways of power amplifier’s linearization synthesis of adaptive digital predistorters
(DPD) based on operand approach using input and output signals is widely used which does not require
significant hardware and financial expenditures. After connecting DPD power amplifier’s output signal
should not contain non-linear distortions. DPD aims to introduce sign non-linear predistortions, which
would allow to compensate power amplifier’s non-linear distortions (linearize power amplifier). The
majority of DPD are designed using multidimensional polynomials that represent truncated volterra series
[3–6].
To improve precision of power amplifier’s linearization in this paper we suggest synthesizing DPD using
piecewise-polynomial and polynomial models. Let’s consider properties of the mentioned models.
— In case of polynomial approximation of non-linear operand precision as a rule is achieved due to
application of essentially non-linear approximating model. When using piecewise-polynomial
approximation one can generate a set of sub-models with lower degrees than in the case of polynomial
operand’s model.
Piecewise-polynomial model is built on impact subsets considering the following properties [7]:
– output signal in a sub-region depends on impact in this sub-region;
– criteria for sub-regions’ boundaries selection are different;
– degrees of non-linearity and memory length of sub-models are independent in different sub-regions.
The mentioned properties allow adjusting piecewise-polynomial model of DPD to power amplifier’s
characteristic.
— Cascade model of DPD describes cascade connection of piecewise-polynomial and polynomial
models. In this case polynomial model is used to combat errors caused by transient processes in output signal
of piecewise model at sub-regions’ boundaries, as well as to smoothen response of piecewise model to input
sub-signals with first kind discontinuities.
PIECEWISE MODEL AND CASCADE STRUCTURE OF PREDISTORTER
During synthesis of piecewise model K sub-regions (zones) of DPD impact set xn ( ) are considered [7].
Boundaries (radia) of zones are specified by values l
i
, i K = - 12 1 , ,...,( ), where i denotes zone’s number,
l l l
1 2 1
< < <
-
K
K
. In Fig. 1 example of choosing three sub-regions is depicted.
Independent non-linear operand effective in different zones are introduced according to the following
conditions:
375
ISSN 0735-2727, Radioelectronics and Communications Systems, 2012, Vol. 55, No. 8, pp. 375–380. © Allerton Press, Inc., 2012.
Original Russian Text © E.B. Solovyeva, 2012, published in Izv. Vyssh. Uchebn. Zaved., Radioelektron., 2012, Vol. 55, No. 8, pp. 49–55.