You are on page 1of 7

Journal of Electronic Testing (2019) 35:359–365

https://doi.org/10.1007/s10836-019-05801-3

Saleh Model and Digital Predistortion for Power Amplifiers


in Wireless Communications Using the Third-Order Intercept Point
Haider Al-kanan 1 & Xianzhen Yang 1 & Fu Li 1

Received: 12 February 2019 / Accepted: 2 May 2019 / Published online: 31 May 2019
# Springer Science+Business Media, LLC, part of Springer Nature 2019

Abstract
This paper presents a novel approach for estimating the Saleh model’s parameters using the gain and third-order intercept point
(IP3) of a power amplifier. The IP3 is a widely used technical parameter for describing the nonlinearity of the power amplifier in
wireless communications. The proposed approach minimizes an objective function of error between the third-order Taylor series
and the Saleh Amplitude-to-Amplitude (AM/AM) model. This approach of the parameters estimation is also used for identifying
the Digital Predistortion (DPD) to compensate for the memoryless AM/AM nonlinear distortion of power amplifiers in wireless
communications. Finally, the Saleh model accuracy and DPD linearization are validated based on the WCDMA measurement
signal.

Keywords Power amplifier . Nonlinear model . Digital predistortion . Third-order intercept point

1 Introduction behavioral technique. The behavioral model is the most effi-


cient and low complexity approach for characterizing the non-
Power Amplifiers (PAs) are major nonlinear components in linearity and DPD of the power amplifier. Various memoryless
wireless communications. In fact, the nonlinear distortion in behavioral models have been proposed in the literature of the
PAs is critical in the communications systems. This is because memoryless nonlinear AM/AM modeling such as the Taylor
the nonlinearity can degrade the signal-to-noise ratio, bit-er- model, Rapp model, Cann model, and Saleh model [6–8].
ror-rate, and affect the reliability of wireless communications The Saleh AM/AM model is a lower complexity function
[2, 5, 10, 11]. Modeling the nonlinear distortion in power compared to the state-of-the-art behavioral models, since it
amplifiers has become an important design aspect for two requires fewer parameters to quantify the nonlinearity of the
main reasons. First, the nonlinear model is used to predict power amplifier. Another advantage of the Saleh model is that
the spectrum regrowth for the in-band and adjacent channel it can be simply inverted when modeling the DPD for PA
bands in the transmitted signals in order to meet the 3GPP linearization. Thus, the Saleh model has become a popular
wireless standard and control the spectrum interference. memoryless model in power amplifiers for modern wireless
Second, most linearization approaches for power amplifiers communications [1, 3, 4, 6, 8, 11].
are derived from a specific nonlinear model. The parameters of the Saleh model are typically calculated
The linearization of the power amplifier based on direct using the Least-Squares (LS) method. The primary drawback
nonlinear inversion using Digital Predistortion (DPD) is an of the LS method is that it requires many data points for the
extensively used approach for mitigating the nonlinear distor- input and output amplitudes in the measurement of the power
tion in wireless transmitters. The nonlinearity characteristics amplifier, in addition to high computations of large matrices,
of the PAs can be modeled using either a circuit approach or a as described in the literature of the Saleh model identification
[1, 3, 4, 7, 8]. Hence, this paper presents a simplified approach
to support the RF designer defining a simple nonlinear behav-
Responsible Editor: S. Mir
ior of the power amplifier. This faster approach does not re-
quire extensive measurements or complex mathematical com-
* Haider Al-kanan
msc.haider@gmail.com
putations in the model identification. New mathematical ex-
pressions are derived in this paper for estimating the parame-
1
ters of the Saleh model from gain and IP3 of the solid-state
Department of Electrical and Computer Engineering, Portland State
power amplifier. The presented approach can reduce the
University, Portland, OR 97207-0751, USA
360 J Electron Test (2019) 35:359–365

complexity of the model identification for both the power where x1d is the input amplitude at the 1 dB compression
amplifier and DPD. The identification of the DPD becomes point, g is the small signal gain. By substituting (2) into (3),
straightforward because it uses the design specifications of the and substituting α into g yields
PA, as described in this paper. This paper is organized as α:x1d
follows. Section 2 describes the characteristics of the Saleh 20 logj j ¼ 20 logjα:x1d j−1 ð4Þ
1 þ β:x21d
model in the AM/AM conversion. Section 3 describes the
model’s identification approach based on gain and IP3 of the Equation (4) can be re-written as
power amplifier. Section 4 presents the inversion of the Saleh
α:x1d α:x1d
model for modeling the digital predistortion. Section 5 in- 20 logj j ¼ 20 logj 1 = j ð5Þ
cludes the experimental validation and results. Finally, a con- 1 þ β:x21d 10 20
clusion is presented in Section 6.
Finally, the input amplitude at the 1 dB compression point
is
2 Nonlinearity of Power Amplifiers 1
x1d ≈ pffiffiffiffiffiffi ð6Þ

Power amplifiers have two main operating regions in the AM/
AM characteristics: the back-off region in the low input am- Power amplifiers produce high nonlinear distortion when
plitude, and the compression region (near the 1 dB compres- the input amplitude reaches this point. Therefore, the 1 dB
sion point) in the high input amplitude. The operating regions compression point is often specified as a design target for
of power amplifier in back-off and compression specify the the power amplifier operating range in the practical applica-
fitting accuracy of the behavioral model. For instance, some tions of wireless communications [10].
behavioral models show better model accuracy for power am-
plifiers of strong nonlinearity. Other models have better accu-
racy in small signals or weak nonlinearity power amplifiers.
The IP3 and 1 dB compression points are important measures 3 Identification Approach for the Saleh Model
of the power amplifier’s characteristics and operating regions.
Therefore, we use these parameters consistently in our pro- The third-order intercept point is a figure-of-merit in power
posed derivations. amplifier linearity as defined by the design specifications. The
IP3 is defined as the input power level at which the output
2.1 Memoryless Saleh Model power at the fundamental frequency intercepts with the output
power at the third-order intermodulation (IM3) frequency as in
The Saleh AM/AM model is an empirical model for Fig. 1 [10]. The higher the IP3 value, the better the power
memoryless nonlinearity of the power amplifier. The gain amplifier linearity specifications. The scope of this section is
and saturation level of the model are specified by the values to derive the relationship between the IP3 and Saleh model
of the parameters α and β which are positive and real numbers parameters based on the 3rd-Order Taylor series. Only odd-
[1]. The baseband AM/AM conversion H[.] for the PA is
given by:

ys ðtÞ ¼ H½xðtÞ ð1Þ


α:xðtÞ
H½xðtÞ ¼ ð2Þ
1 þ β:x2 ðtÞ

where x(t) and ys(t) are the magnitudes of the PA input and
output signals, respectively. The parameter α is a small signal
gain and the parameter β adjusts the shape of the compression
region [8]. The value of the 1 dB compression point can be
related to the parameter β as we demonstrate in this section.
The 1 dB compression point is defined as the power level at
which the small signal gain drops by 1 dB. This is simply can
be written:

20 logjys ðx1d Þj ¼ 20 logjg:x1d j−1 ð3Þ Fig. 1 Graph illustrates the third-order intercept point (IP3) in the AM/
AM characteristics of the power amplifier
J Electron Test (2019) 35:359–365 361

order coefficients are considered in (7), because the even- Section 2.1. The total square error e2T represents the integral
order coefficients quantify the even-order intermodulation dis- of (11) over the defined input amplitude range as
p1ffiffiffi
8β  2
tortion which are multiple frequencies of the fundamental fre-
quency and they can be easily filtered out [13, 14]. In addition, αx ð12Þ
e2T ¼ ∫ c 1 x þ c3 x3 − dx
the Saleh model is an odd function, and it is compatible with 0 1 þ βx2
the odd-order Taylor series expansion.
The integral calculation of (12) is:
yT ¼ c1 x þ c3 x þ :::::: þ cn x
3 n
ð7Þ   pffiffiffi pffiffiffi pffiffiffi
ð9 α2 þ 36c1 αÞ tan−1 1=2 2 −2α2 = 2−9c1 α 2
e2T ¼ pffiffiffiffi þ
where x and yT are the power amplifier input and output mag- 18β β
pffiffiffi   pffiffiffi pffiffiffi  
nitude, respectively. cn is the n-th coefficient of the odd-order 640 2c3 α tan−1 1=2 2 −23 2=96 þ c1 c3
pffiffiffiffiffiffi þ ð13Þ
Taylor series. Most of the nonlinear distortion in power ampli- 320β2 2β
fiers is quantified by IM3, because the power of IM3 distortion c23 c21
is relatively higher than any of the other higher odd-order inter- 3 pffiffiffiffiffiffi
þ pffiffiffiffiffiffi
7168β 2β 48β 2β
modulation distortions. In addition, the IM3 distortion is close
to the fundamental frequency in the frequency domain [13]. The minimum square error can be calculated such that the
The IM3 distortion is a major challenge in power amplifiers gradient of (13) with respect to the Taylor coefficients (c1, c3)
because it is difficult to eliminate. The c1 and c3 coefficients approaches zero.
can be calculated as derived in [13, 14]:
 
  ∂e2T ∂e2T
∇ e2T ¼ ; ¼ ð0; 0Þ ð14Þ
c1 ¼ 10ð 20Þ ∂c1 ∂c3
G
ð8Þ
−2 ð −IP103 þ 3G
20 Þ The gradient calculation in (14) results the following two
c3 ¼ 10 ð9Þ
3 equations:
where G is the power gain in dB and IP3 is the output third-
order intercept point in dBw. Here we propose to minimize the pffiffiffi   pffiffiffi pffiffiffi  pffiffiffi
2c1 −1280 α tan−1 1=2 2 −23 2=96 2 c3
square error function between the Taylor model and Saleh p ffiffiffi
ffi þ pffiffiffiffi ¼ 0
640 β2 β 7168 β3 β
model bounded by the PA input amplitude range {0 x1d} as
written in (10). ð15Þ
pffiffiffi −1  pffiffiffi pffiffiffi
c3 c1 þ 48 α 2 tan 1=2 2 −2 2
( pffiffiffiffiffiffi þ pffiffiffiffiffiffi ¼0
minimize ðyT −yS Þ2 320 β2 2β 24 β 2β
C 1 C3 ð10Þ
subject to amplitude 0 ≤ x≤ x1d ð16Þ

The square error function (yT - yS)2 can be written as The instantaneous solution of (15) and (16) is:
 2
  αx c1
n α¼ pffiffiffi  pffiffiffi
e ð x Þ ¼ c1 x þ c 3 x − ffio ð11Þ
2 3
ð17Þ
1 þ βx2 x∈ 0pffiffiffi 1

1760−3660 2 tan−1 1=2 2
 pffiffiffi  pffiffiffi 
−3c3 183 2 tan−1 1=2 2 −88
The variable x is specified by the power amplifier input
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi β¼  pffiffiffi  pffiffiffi  ð18Þ
range {0, x 1 d } where x 1 d = ð1=8βÞ as derived in 56c1 129 2 tan−1 1=2 2 −62

Fig. 2 Simplified block diagram


of the transmitter system
depicting the parameters of both
the power amplifier and digital
predistortion
362 J Electron Test (2019) 35:359–365

Fig. 3 Measurement set-up for Power Supply


power amplifier experiment

Signal Analyzer

Signal Generator
RF Amplifier

By substituting (8) into (17) and simplifying both (17) and that the parameter β specifies the nonlinearity in the power
(18) we obtain the following: amplifier due to the IM3. In addition, the parameter β shapes
the curve of the Saleh model and adjusts the compression
region as we have demonstrated in Section 2.1.
10ð 20Þ
G

pffiffiffi ≈1:003  10ð 20Þ


G
α¼ pffiffiffi  ð19Þ
−1
1760−3660 2 tan 1=2 2
 pffiffiffi  pffiffiffi  4 Identification of Digital Predistortion
−3 c3 183 2 tan−1 1=2 2 −88 c3
β¼  pffiffiffi  pffiffiffi  ≈−1:137 ð20Þ
−1
56 c1 129 2 tan 1=2 2 −62 c1 Digital Predistortion (DPD) is one of the most efficient
ways in linearization power amplifiers, and it is an often-
By substituting (8) and (9) into (20) we obtain: used technique for eliminating the nonlinear distortion. The
−IP3 G DPD is basically a nonlinear function that has inverse char-
β ¼ 0:758  10ð 10 þ 10Þ ð21Þ acteristics in the AM/AM conversion of the power amplifi-
er. The DPD is connected in cascade to the input of the
power amplifier as shown in Fig. 2. The combined charac-
Equations (19) and (21) illustrate that the parameter α is
teristics of the power amplifier AM/AM conversion (gain
directly proportional to the power amplifier gain, and the pa-
compression) and the DPD nonlinearity (gain expansion)
rameter β is a function of both gain and IP3. This also implies
result in a linear signal amplification. However, an exact
inverse model of the power amplifier is difficult to obtain

Fig. 4 Amplitude-to-Amplitude (AM/AM) results of the power amplifier Fig. 5 Square residuals between the PA measured and modeled output
measured as well as the 3rd-Order Taylor and Saleh model amplitude
J Electron Test (2019) 35:359–365 363

Table 1 Accuracy evaluation of


the Saleh model and the 3rd-order Power amplifier operating range Saleh model Taylor model
Taylor model, both with respect to
the measured results for different Total square MSE (dB) Total square MSE (dB)
operating ranges of the power error (V2) error (V2)
amplifier
(0) V to the 1 dB compression point. 5.3757*10−4 −47.61 5.2889*10−4 −47.68
(0) V to saturation of the Taylor model. 0.0022 −43.77 0.0031 −42.26
(0) V to saturation level of the PA. 0.112 −26.46 39.05 −1.07

in practice, because of several reasons such as noise and sffiffiffiffiffiffiffi


circuit parasitic effects. The DPD model in this work is α2
expressed in (22) and calculated by solving the quadratic 0< u< ð25Þ

equation (23) [9, 12].
Finally, by substituting the parameters α and β from (19)
xðtÞ ¼ H−1 ½uðtÞ ð22Þ and (21) into (24) we obtain the following:
where u(t) and x(t) are the DPD input and output signals, qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
−IP3
respectively. The inverse function can be calculated by −1 1− 1−3:012 u2 10ð 10 Þ
H ½u ¼ −IP3 G ð26Þ
swapping the input and output variables in H[.] and re- 1:512 u 10ð 10 þ 20Þ
arranging them as
The DPD AM/AM model in (26) is a simple expression of
u β x2 −α x þ u ¼ 0 ð23Þ only two parameters (G and IP3) for the specific power am-
plifier device. Thus, the DPD system design becomes straight-
forward. However, the DPD input amplitude (u) is limited by
The inversion of the Saleh model can be easily obtained by the condition in (27) to obtain a real amplitude value at the
solving (23) for x. DPD output as demonstrated in (25).

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
α− α2 −4 β u2
0:331  10ð 10 Þ
−1 IP3
H ½u ¼ ð24Þ 0< u< ð27Þ
2βu

The signal magnitude u should satisfy (25) in order to ob-


tain real magnitude on the DPD output.
5 Experimental Validation

An experiment is implemented to verify the results of our


theoretical expressions. The experiment consists of the mea-
surement set-up shown in Fig. 3, which includes the Keysight
E4438C signal generator, the Tektronix RSA 6120A spectrum
analyzer, and the commercial class-A power amplifier ZFL-
1000LN from Mini-Circuit. The measured parameters of the
power amplifier are as follows: gain (G = 22.5 dB) and (IP3 =
12.86 dBm). The calculated Taylor coefficient c1≈ 13.3 as
from (8).
The AM/AM measured results for the power amplifier are
obtained using the two-tone test at a center frequency 1 GHz
and 50 KHz tone spacing. The input power of the two-tone
was swept and the output power was measured to obtain the
AM/AM curve of the PA. The Saleh model identification
using this approach and the calculated 3rd-Order Taylor model
Fig. 6 Power spectral density of the power amplifier input, and output are both depicted in Fig. 4, in addition to the AM/AM mea-
signals, with and without digital predistortion sured results. The compared modeling results show that both
364 J Electron Test (2019) 35:359–365

Fig. 7 Constellation diagram of 16-QAM signal at the output of the power amplifier. (a) without digital predistortion, and (b) with digital predistortion

the Saleh model and the 3rd-Order Taylor model accurately model as well as the 3rd-Order Taylor model. The results in
reflect the results obtained from the PA’s measurements for the Table 1 are calculated for different amplitude ranges: from (0)
amplitude range from (0) V to the 1 dB compression point V to the 1 dB compression point, from (0) V to the saturation
(which occurs at 31 mV input amplitude). However, the 3rd- level of the Taylor model, and finally from (0) V to the satu-
Order Taylor model decreases monotonically after the maxi- ration region of the power amplifier.
mum value because the third-order coefficient (c3) is a high The DPD model using this approach is evaluated using
negative number which causes a high gain compression effect WCDMA signal of 3.84 MHz bandwidth. The linearization
at high input amplitude. However, the Saleh model gain de- performance of the DPD in the Power Spectral Density (PSD)
creases slowly after the 1 dB compression point because the is presented in Fig. 6 which shows a reduction in the spectral
output amplitude is attenuated slowly by the denominator regrowth of the PA nonlinear distortion. The improvement in
(1 + βx2) of the Saleh function in (2). Adjacent Channel Power Ratio (ACPR) is 13.01 dBc for the
The square error between the measured and modeled AM/ power amplifier with DPD. In addition, the DPD evaluation in
AM results is calculated using (28). The residual error be- the time domain is presented using the 16-QAM signal. The
tween the measured results and Taylor series as well as the signal constellation diagram in Fig. 7 shows the effect of the
Saleh model are depicted in Fig. 5. This shows a significant PA nonlinearity on the locations of the complex symbols in (a)
lower residual error in the Saleh model compared to the 3rd- as well as the DPD correction in (b). Finally, the results of both
Order Taylor model. the ACPR and Normalized Mean Square Error (NMSE) be-
tween the signals of the PA input, and PA output with DPD are
depicted in Table 2. The ACPR is calculated for both the upper
K
e2T ¼ ∑ ðymod ðnÞ−ymes ðnÞÞ2 ð28Þ and lower adjacent channels (± 2 MHz) of the WCDMA
n¼1 signal.
where ymod(n) is the modeled output amplitude, ymes(n) is the
measured output amplitude, e2T is the total residual square
error, and K is the number of samples. Table 1 depicts the total 6 Conclusion
squares error ðe2T ) and the Mean Square Error (MSE) between
the measured amplitude and the output amplitude of the Saleh This paper presented a new method of estimating the Saleh
behavioral model for power amplifiers. We derived simple
expressions for the relationship between the Saleh coefficients
Table 2 ACPR and NMSE results of the power amplifier’s linearization
approach and the power amplifier parameters: gain and third-order in-
tercept point. The IP3 is a popular technical specification of
Case ACPR (dBc) NMSE (dB) the power amplifier’s nonlinearity and it can be obtained from
± 2 MHz
a technical data sheet or easily measured using a two-tone test.
PA Input −36.05 / -36.55 * The results of the Saleh model fit well with the memoryless
PA Output without DPD −20.04 / -21.85 * AM/AM conversion of the power amplifier and the 3rd-Order
PA Output with DPD −33.05 / -33.97 −13.27 Taylor model in the amplitude range up to the 1 dB compres-
sion point. In addition, the results of the Saleh model showed a
J Electron Test (2019) 35:359–365 365

better modeling accuracy than the 3rd-Order Taylor model in 9. Mohammad S, Seyed S (2012) Performance of a predistorter based
on Saleh model for OFDM systems in HPA nonlinearity. In
the AM/AM region after the 1 dB compression point of the
Proceedings 14th Interenation Conference on Advanced
power amplifier. Communication Techniques (ICACT), pp. 148–152
A digital predistortion using the inversion of the Saleh 10. Steve C (2002) Advanced techniques in RF power amplifiers de-
model was derived as a function of both the gain and IP3 sign. Artech house, Nonvood
parameters of the power amplifier. The DPD model in this 11. Terry H, Ge J, Geng S, Wang G (2007) A nonlinearity predistortion
technique for HPA with memory effects in OFDM systems.
work compensated for the nonlinear distortion in power am- Nonlinear Anal. Real World Appl 8(1):249–256. https://doi.org/
plifier as demonstrated in both the frequency and time domain 10.1016/j.nonrwa.2005.08.007
results. The DPD model results showed a significant reduction 12. Tien N, James Y, Charles L, Hien T, Diana J (2003) Modeling of
of the spectrum regrowth in the power spectral density of the HPA and HPA linearization through a predistorter: global broad-
WCDMA signal. The power amplifier linearization using this casting service applications. IEEE Trans Broadcast 49(2):132–141.
https://doi.org/10.1109/TBC.2003.813650
approach showed 13.01 dB improvement in ACPR when the 13. Xiao L, Fu L (2013) RF power amplifier’s nonlinear modelling with
DPD model was applied to the power amplifier. In addition, a memory effect. Int J Electron Lett 1(1):44–49. https://doi.org/10.
linearity improvement in the 16-QAM constellation diagram 1080/21681724.2013.805382
was clearly observed using the DPD approach in this paper. 14. Xiao L, Chunming L, Yang X, Fu L (2012) Obtaining polynomial
coefficients from intercept points of RF power amplifiers. Electron
Lett 48(19):1238–1240. https://doi.org/10.1049/el.2012.1734

Acknowledgments The authors would like to acknowledge Dr. J. Publisher’s Note Springer Nature remains neutral with regard to
Veerman from the Math department, and Chenyang Li from the electrical jurisdictional claims in published maps and institutional affiliations.
and computer engineering department at Portland State University for
their insight suggestions in the equations’ derivation.

Haider Al-kanan received the B.Sc. degree in electrical engineering from


References University of Basrah, Iraq in 2004, the M.Sc. degree in communication
systems engineering from University of Portsmouth, UK in 2008. He is
1. Adel S (1981) Frequency-independent and frequency-dependent currently working toward the Ph.D. degree in electrical and computer
nonlinear models of TWT amplifiers, in IEEE Transactions on engineering from Portland State University, OR, USA. His doctoral re-
Communincations, vol. 29, no. 11, pp. 1715-1720. https://doi.org/ search and interests include signal processing techniques for high power
10.1109/TCOM.1981.1094911 efficiency and linearity in power amplifiers for wireless communications.
2. Claude S (1948) A mathematical theory of communication. Bell He is a recipient number of awards for professional achievements.
Syst Tech J 27:379–423 623–656. https://doi.org/10.1002/j.1538-
7305.1948.tb01338.x Xianzhen Yang received the B.S. degree in electrical engineering from
3. Haider A, Fu L, Felice T (2017) Extended Saleh model for behav- Beijing Jiaotong University, China in 2014, and the M.S. degree in elec-
ioral modeling of envelope tracking power amplifiers. In trical and computer engineering from Portland State University, Portland,
Proceedings of IEEE 18th Wireless and Microwave Techniques. OR, USA in 2018. He is currently pursuing the Ph.D. degree in electrical
(WAMICON), Cocoa Beach, FL, pp. 1–4. https://doi.org/10.1109/ and computer engineering at Portland State University. His research in-
WAMICON.2017.7930244 terests include the wireless communication and signal processing.
4. Haider A, Felice T, Fu L (2018) Hystersis nonlinearity modeling
and linearization approach for envelope tracking power amplifiers Fu Li received the B.S. and M.S. degrees in physics from Sichuan
in wireless systems. Microlectron Jl 82:101–107. https://doi.org/10. University, Chengdu, China, in 1982 and 1985, respectively, and the
1016/j.mejo.2018.10.006 Ph.D. degree in electrical engineering from the University of Rhode
5. John W (2014) Behavioral modeling and linearization of RF aower Island, Kingstone, RI, USA, in 1990. Since 1990, he has been with
amplifiers. Artech house, Norwood Portland State University, Portland, OR, USA, where he is currently a
6. José P, Stephen M (2005) A comparative overview of microwave Full Professor of electrical and computer engineering. He has authored
and wireless power-amplifier behavioral modeling approaches. over 100 papers in the refereed journals and international conferences.
IEEE Trans Microw Theory and Tech 53(4):1150–1163. https:// His current research interests include signal, image, video processing,
doi.org/10.1109/TMTT.2005.845723 wireless, network, and multimedia communications. Dr. Li was an
7. Magnus I, David W, Daniel R (2006) A comparative analysis of Associate Editor of the IEEE Transactions on signal processing from
behavioral models for RF power amplifiers. IEEE Trans Microw 1993 to 1996 and a column Associate Editor of the IEEE signal process-
Theory Tech 54(1):348–359. https://doi.org/10.1109/TMTT.2005. ing magazine in 2002. He was the general chair of the ninth IEEE-SP
860500. Workshop on Statistical Signal and Array Processing (SSAP-1998).
8. Mairtin O, Serban M, Yiming L (2009) New modified saleh models
for memoryless nonlinear power amplifier behavioural modelling.
IEEE Commun Lett 13(6):399–401. https://doi.org/10.1109/
LCOMM.2009.090222

You might also like