Power Amplifier Adaptive Linearization Using

Predistortion with Polynomial Functions

Mohammad Ghaderi*, Surinder Kumar"

ABSTRACT
Quadrature ampliltude modulation has high spectral efficiency but is highly sensitive to power amplifier
nonlinearities. It is shown that an adaptive predistorter usig polynomial amplitude and phase
predistortion functions can be used to linearize the power amplifier. Tie recursive least square algonrthm
is employed in an estimator which uses demodulated signals to estimate the required predistortion.
Computer simulation results are provided and these results show that fast convergence and high spectm
improvement can be obtained using the proposed linearization method. The results are significantly
better than those reported before for other linearization techniques.

I.INTRODUCTION
Spectrum being a scarce resource, there is considerable interest in using high level modulation such as
multilevel quadrature amplitude modulation (QAM). A 16 level QAM can provide about 3 bits/Hz
spectral efficiency. Such a modulation method is of considerable interest for the future mobile
communication systems. In QAM, the information being sent is contained in both the amplitude and
phase of Xt carier. This makes such modulation method highly sensitive to the channel nonlinearities.
The strongest nonlinearity in the channel is caused by the transmitter power amplifier.
One solution could be to use a class A power amplifier with 10 to 15 dB back-off from the saturation.
This results in very poor prime power efficiency. This is obviously not a feasible solution for the mobile
transmitter where batery power has to be conserved. Poor prime-power efficiency could also be a
problem in the cell stations and other high power QAM systems.
An effective solution is to employ power amplifier linearization using predistortion [l]-(4]. In this
linearization method the amplitude and phase nonlinearites of the power amplifier are compensated by
the predistortion of the input signal. This predistortion could be implemented at the baseband, IF or RF
frequencies. The predistorter amplitude and phase charateristics have to be apprximately inverse of the
power amplifier amplitude and phase nonlineariies. The nonlinear transfer characteristics of the
predistorter can be implemented using either a look up table [lI]-[3] or polynomial functions [4]. Because
of changes with time, temperature and operating channel, it is necessary that the predistorter be adaptive.
In this paper an adaptive polynomial type predistortion linearizer with fast convergence is presented. A
new estimation method for the polynomial coefficients is described. An impornt feature of the
implementation is the desired quadratic relation between the error and the polynomial coefficients. This
ensures a fast convergence to the global minimum. The RF signal at the output of the power amplifier is
demodulated and sampled. The coefficients estimator is in effect a post-distorter which distorts the
demodulated power amplifier output. Adaptation process ensures that the error between the post-distorter
output and the signal obtained by demodulating the amplifier input is minimum. The recursive least
square (RLS) algorithm is employed for this adaptation process. After convergence the post-distorter
generates the polynomials which correspond to the inverse nonlinearity of the amplifier. The coefficients
of these polynomial are used in the predistorter amplitude and phase polynomials. A high degree of
linearization becomes possible using this method.
The linearizer structure is further described in the next section. Section HI contains details of the

* Department of Electrical Engineering, University of Saskatchewan Saskatoon, Saskatchewan, Canada

S7N OWO.

1125
Authorized licensed use limited to: Nokia. Downloaded on July 09,2024 at 15:05:56 UTC from IEEE Xplore. Restrictions apply.
adaptation algorithm used in the estimator. The linearizer performance results are given in section IV.
The paper is concluded in the last section.

H.LINEARIZER STRUCTURE
As shown in Fig. 1, the predistortion is implemented at IF. The samples of the demodulated signals at
the input and output of the power amplifier are fed to the estimator.

Figure 1: Block schematic of a QAM system transmit chain with adaptive

predistorter
The QAM signal input to the power amplifier may be wnrtten as
(1)
v,1) r,r) cos[kot+01,t)]
=

Where rie 01 is th complex envelope of the modulated QAM signal and wo) is the carrier frequency. The
output of the power amplifier may be written as
(2)
v =(t) A(r,it)) cos[wjOt+Of,r)+D(r1{t))]
=

=r0(t)) cos[wOt+00(O)]
A(.) and Q(.) denote the amplitude and phase transfer functions of the amplifier and are of the form [5]
(3)
A(r)=ajr+a3r3+y3r5+ . . .

(4)
(D(r)=p2r2+pi4r4+pi6r6+
A detailed block diagram of the predistorter and estimator is given in Fig. 2. The coefficient estimator
generates the coefficients of the amplitude distortion polynomial g(.) and phase distortion polynomial

1126
Authorized licensed use limited to: Nokia. Downloaded on July 09,2024 at 15:05:56 UTC from IEEE Xplore. Restrictions apply.
 11Dee R=9(r) |l

r < _ ALGORITHM

PREDISTORTER DELA*

EST IMATOR
Figure 2: Schematic diagram of the adaptive linearizer

m.ADAPTAnON ALGORITHM
Different algorithm were tried. The recursive least square (RLS) algorithm was found to be suitable for
the proposed linearization method [61. The algorithm is used to miinimize te error functions er and e0.
At Ith sample these fiunctons may be wraiten as
(5)
er(t = rj(1)-g(r0(t))
(6)
es(l) = OI-8(1Q)-P(rf))
Where
(7)
g(r0(O)) aj,r(O(+a3r03(O+a5ro5(l)+
=
(8)
P(r5(O) = b2Qri2(O)+bUrA(k)+b6ri6(l)+
n=Total number of data samples used in the estimation
The coefficients in the predistorter polynomials are set using the coefficients estimated using n-I data
samples.
The cost functions are defined as a residual sum of weighted squares as
(9)
Jr(n) -§nfl-e 2(1y=
r zX?l[r,ty.g(r0(o)]2

1127
Authorized licensed use limited to: Nokia. Downloaded on July 09,2024 at 15:05:56 UTC from IEEE Xplore. Restrictions apply.
 (10)
J9(n) = I
1=1

Where x is a positive equal or less than one. The set of the coefficients which minimizes J(n) and
J0(n) are found by the algorithm.

oI\
0

0
z - I

* A
_ ~~~~~~~ Sat
C6 30 30 40
llERAnON NUM n

Figure 3: Performance of the Estimator

IV.LINEARIZATION PERFORMANCE
The power amplifier model used for the sinulation is given by (5]
(1 1)
A(r) =+2
I +r2
(12)
¢(r) = 60 -

Amplitude and phase predistortion polynomials with four coefficients were considered, so that
(13)
g(r)=a1r+a3r3+a5r5+a7r7
(14)
P(r)=b2r2+b4r4+b6r6+b8r8
The rate of convergence of the normalized ampliltude and phase of the predistorter/power amplifier
system to the desired linear system is illustrated in Fig. 3. As may be seen from this figure amplitude and
phase error of less than 2% are obtained after 10 iterations. Steepest descent (gradient) algorithm was
also tested. It was found that RLS algorithm converges much faster than the gradient algorithm.

1128
Authorized licensed use limited to: Nokia. Downloaded on July 09,2024 at 15:05:56 UTC from IEEE Xplore. Restrictions apply.
 C.- 40

.30
N.6 -

OA
0.2-

oDt
0~4 Di OA
WiUT AMULD

Figure 4: Amplitude/Phase characteristics of the Predistorter

40

30

016 2
5

m0
Logend
PHASE__
0 &VP UDJOLW ITJ5PQ
o
prH '-Tf-lA

Figure 5: AMIAM and AM/PM characteristics of a class AB

power amplifier with and without predistortion
function P(.) for a power
Figure 4 shows the predistorter amplitude function g(.) and the phasecharacteristics
and phase of a class AB
arnplifier with 0.9 dB back-off from saturation. The inamplitude be this figure, the
amplifier with and without predistortion are shown Fig. 5. AsOnly may seen
small
from
back-off of 0.9 dB is
predistorter is quite effective in linearizing the power amplifier. a
output of the power amplifier with and
required. The power spectrum density of the QAM signal at the with a raised-cosine pulse with
without predistortion are compared in Fig. 6. The input signal is 16QAM, the spectrum
25% roll off. Once again it is seen that linearization is quite effective in eliminating
spreading.

1129
Authorized licensed use limited to: Nokia. Downloaded on July 09,2024 at 15:05:56 UTC from IEEE Xplore. Restrictions apply.
 -20O

~-3O-
-70-

-80

V.CONCLUSION
~ ~ ~ ~ ~13zp-
~QUDCT Qd4zW
Loen

Figure 6: Power spectnun of the filtered 16QAM data signal

with and without linearization

Predistortion using amplitude and phase polynomial functions is an effective method of power
amplifier linearization. Recursive least square algorithm resulted in a fast convergence of the
coefficients. Using this linearizer, amplitude and phase errors could be reduced to 2% from the ideal
linear response. This resulted in 60 dB improvement in spectrum spreading.

ACKNOWLEDGEMENT
Funding for this work was provided by Natural Science and Engineering Research Council of Canada
under the University-Industry chair program.

REFERENCES
[1] A. A. M. Saleh and J. Salz, "Adaptive Linearization of Power Amplifiers in Digital Radio
systems," Bell Syst Tech. ., Vol.62, No. 4, Apr. 1983, pp.1019-1033.
[2] Y. Nagata, " Linear Amplification Technique for Digil Mobile Communications," in Proc. IEEE
Veh. Technol. Conf., 1984, pp. 159-164
[3] J. Cavers, "Amplifier linearization using a digital predistorter with fast adaptation and low memory
requirements," IEEE Trans. Veh. Technol., Vol. 39, No 4, Nov. 1990, pp. 374-382.
[4] J. Namiki, " An Automatically Controlled Predistorter for Multilevel Quadratur Amplitude
Modulation," IEEE Trans. Commum, Vol. COM-31, May 1983, PP.707-712
[5] A. A. M. Saleh, "Frequency-Independent and Frequency-Dependent Nonlinear Models of TWIT
Amplifiers," IEEE Trans. Comm., COM-29, No. 1 1, Nov. 1981, pp. 1715-20.
[6] S. S. Haykin, "Introducfion to Adaptive filters," Newyork, CoUier Macmilan, 1984.

1130
Authorized licensed use limited to: Nokia. Downloaded on July 09,2024 at 15:05:56 UTC from IEEE Xplore. Restrictions apply.