Sunteți pe pagina 1din 13

Adaptive Nonlinear Compensation of Satellite Transponder Distortion for

High-Level Data Modulations


Vincenzo Lottici, Marco Luise and Ruggero Reggiannini
Università di Pisa – Dipartimento Ingegneria Informazione
Via Diotisalvi, 2 – I-56122 Pisa, Italy - e-mail: vincenzo.lottici@iet.unipi.it

Abstract— The diffusion of wideband satellite services, especially in the field of digital TV broad-
casting and/or interactive multimedia, suggests the adoption of spectrally-efficient high-level
modulation techniques in future satellite transmission standards. It is known that such modulations
suffers from nonlinear distortion due to amplification with on-board (nonlinear) high-power ampli-
fiers, so that countermeasures are to de adopted in this respect. The aim of this contribution is to
investigate the possibility to adopt a suited mixture of adaptive data predistortion (DP) at the
transmitter and adaptive nonlinear equalization (NLE) at the receiver to counteract nonlinear dis-
tortion at best. In particular, we show that the sensitivity to nonlinear distortion of 16QAM and
64QAM is greatly reduced by clever use of both techniques, and that such an arrangement provides
non-negligible (though smaller) performance improvement when using the optimized 16APK con-
stellation recently introduced by ESA, and its extended 64-point version.

I. INTRODUCTION
The increasing demand for wideband satellite services, especially in the field of digital TV broad-
casting and/or interactive multimedia, suggests the adoption of spectrally-efficient high-level modu-
lation techniques in future satellite transmission standards. Unfortunately, such a desirable approach
has to face the issue of on-board power amplification: on the one hand, high-power amplifiers
(HPAs) have to be operated as close as possible to saturation to improve efficiency and link budget;
on the other, high-level modulations such as M-QAM or the like are more sensitive to nonlinear
distortion caused by saturated amplifier than good-old QPSK is. Many different compensation tech-
niques were studied or applied to make better use of satellite available power. Some of them are to
be used directly on the on-board HPA [1-8], others encompass signal pre-distortion in the transmit-
ter [9-10], and a few rely on post-processing at the receiver side [11-15]. Nonlinear distortion in-
duced by an HPA closely resembles (linear) inter-symbol interference on the filtered received data
signal, and consequently the theoretically optimum post-processing method is based on Maximum-
Likelihood Sequence Estimation (MLSE) [11], [12]. As is known, MLSE has a high implementa-
tion complexity, especially with high-level constellations and strictly bandlimited signals. An effec-
tive sub-optimum technique, originally developed for voiceband modems [13], then extended to
microwave radio [2] and satellite communications [14], [15], is nonlinear equalization, whose
complexity is affordable and whose performance is sufficiently good. Specifically focusing on satel-
lite broadcasting, a topic to be investigated is the criterion to split compensation of the nonlinearity
between the earth-station transmitter and the subscriber set-top receiver. The basic idea behind this
system architecture is to split the linearization task over two low-complexity devices rather than
concentrating it in a single complex unit placed at the transmitter or at the receiver sites. Further-
more, HPA linearization techniques on-board the satellite will not be taken into consideration to
pursue back-compatibility with pre-existing un-linearized satellites.
The aim of this contribution is thus to investigate the interplay between adaptive data predis-
tortion (DP) at the transmitter [9]-[10] and adaptive nonlinear equalization (NLE) at the receiver
[13]-[15]. In fact, the use of a simple DP compensation algorithm at the transmitter helps simplify-
ing the design of the NLE at the receiver, even with high-level data modulations, such as 16- or 64-
level QAM. We show that the sensitivity to nonlinear distortion of 16QAM and 64QAM is greatly
reduced by clever use of both techniques, and that such an arrangement provides non-negligible
(though smaller) performance improvement when using the optimized 16APK constellation intro-
duced in the companion paper [16], and its extended 64-point version.
In the next section we illustrate the transmission system architecture. In Section III we briefly
review the basic concepts associated with adaptive data predistortion and adaptive nonlinear equali-
zation. The combined effect of such linearization techniques is assessed in Sect.IV, and finally
some conclusions are drawn in Sect.V.

II. TRANSMISSION SYSTEM ARCHITECTURE


2.1 System model
Our reference (baseband-equivalent) transmission system is shown in Fig. 1. It encompasses linear
I-Q digital modulation with QAM or generic APSK (Amplitude and Phase Shift Keying) constella-
tions. The T -spaced data symbols ai are independent and identically-distributed. The transmit and
receive filters, have both frequency response GT ( f ) = GR ( f ) = GN ( f ) , GN ( f ) denoting a Nyquist
raised-cosine response with rolloff factor α . Prior to being fed to the earth station shaping filter,
data symbols are passed through the data predistorter block, labeled as DP. The DP re-maps the
symbols of the nominal QAM or APSK constellations so that the effect of nonlinear distortion is
minimized. DP, only affecting the constellation points is remarkably simpler than analog predistort-
ing devices like that in [4]. In the receiver, the distortion is further compensated for by the non lin-
ear equalizer, denoted by NLE, that processes the T -spaced signal samples at the matched filter
output and feeds the multi-level threshold detector. The satellite transponder HPA is modeled as a
nonlinear memoryless device, whose AM/AM and AM/PM input-output characteristics are modeled
according to Saleh [23] as follows:
2ρ 2ρ 2 π
M( ρ ) = , Φ( ρ ) = Φ0 2 , Φ0 = (1)
1+ ρ 2 1+ ρ 6

where ρ denotes the input signal amplitude, and M( ρ ) , Φ( ρ ) represent the AM/AM and AM/PM
characteristics, respectively. We also neglect the uplink noise component entering the satellite
transponder, and we only take into account the downlink noise, modeled as an AWGN process with
two-sided power spectral density (psd) N0 2 .

ai bi x(t)
PD GT ( f )
Transmitting Earth Station
Satellite
HPA

AWGN

aˆ k Threshold Nonlinear
Detector Equalizer GR ( f )
Receiving Earth Station tk = kT r(t)

Fig. 1 – Transmission System Architecture

2.2 Constellation optimization for APSK modulations


Before entering into all details about DP and NLE, we would like to briefly describe the criterion
for the optimization of a class of linear modulation schemes, the so-called M -APSK constellations
[16], which are particularly suited for a typical nonlinear satellite transmission channel. The basic
idea is to design the transmitted signal constellation taking into account the features of the HPA, in
particular trying to minimize the input signal amplitude variations. As suggested in [16], a 16-point
constellation can be designed as composed of two concentric rings of uniformly spaced points, (an
inner and an outer rings). With this particular arrangement, the new modulation shows a smaller
envelope fluctuation in comparison with 16-QAM, and therefore it is less sensitive to nonlinear
channel effects.
The optimization is carried out using the conventional criterion of maximum minimum
Euclidean distance between constellation points. Ref. [16] solves this issues via a brute-force ap-
proach. Vice-versa, we will describe here a heuristic-based approach suited to extension to higher
level modulation, such as 64-APSK. Sticking for the moment to 16-APSK, we introduce the ratio of
the radii of the external and internal rings ρ = R2 R1 . We also denote with δ11 and δ22 the minimum
distances between points in the internal and external rings only (intra-ring distances), respectively,
and with δ12 the minimum distance between internal and external points (inter-ring distance), as
outlined in Fig. 2. The M=16 points in the constellation can be partitioned into a number of differ-
ent configurations having N1 points in the inner ring, and N 2 points in the outer rings, respectively.
For a fixed relative phase ψ between the inner and outer rings points, the constellation optimization
has to obey the following basic constraints:
N1 + N 2 = M = 16 (2a)
ρ = R2 R1 > 1 (2b)

N1R12 + N 2 R22 (2c)


=1
16
where (2c) is the constant-energy condition. Given N1 and N 2 , the optimum value of ρ , and there-
fore of the radii R1 e R2 , is found by starting with a “trial” configuration such that δ11 = δ22 (equal
intra-ring distances). If δ 12 > δ 11 (i.e., if the inter-ring distance is larger than the intra-ring ones, as
in Fig. 2) the optimization is accomplished, otherwise we turn the original configuration into the
one for which δ11 = δ12 , which is the desired optimum (in fact, we will surely have δ 22 > δ 11 ).

δ12 δ22
δ11

Fig 2 – Optimization of the two-ring constellation

Browsing through all possible combination of N1 , N 2 and ψ with this criterion, the better
configurations were found to be N1 = 4, N 2 = 12 and N1 = 6, N 2 = 10, which will be designated in
the following respectively as 4-12APSK and 6-10APSK (Fig. 3). To be more specific, the respec-
tive normalized minimum distance is equal to 0.5843 and 0.5897 for
ρ = sin(π / 4) / sin(π /12) = 1 + 3 ≅ 2.732 and ρ = 2 (the minimum distance of 16QAM is 0.6325).
The first configuration is dominated by the intra-ring distance, hence no improvement can be ob-
tained playing with the relative phase ψ between the inner and the outer rings. Although the 6-
10APSK constellation is dominated by inter-ring distance, its minimum distance is substantially in-
sensitive to ψ as well. For the sake of completeness, we have also to mention that the degradation
in terms of minimum distance of the new APSK constellations with respect to the 16-ary square
(QAM) and optimum hexagonal constellations [17], are 0.69dB and 1.27dB respectively for
4-12APSK, and of 0.61dB and 1.19dB for 6-10APSK.
1.0 1.0

0.5 0.5

-1.0 -0.5 0.5 1.0 -1.0 -0.5 0.5 1.0

-0.5 -0.5

-1.0 -1.0

(a) (b)

Fig. 3 – Optimized 16-point constellations: 4-12APSK (a) and 6-10APSK (b)

The same optimization procedure can be applied to 64-point constellations with four rings of
points (of course, with some more effort). By defining the parameters ρ1 = R2 R1 , ρ2 = R3 R1 , and
ρ3 = R4 R1 as the ratios of the radii of the second, third and fourth rings with respect to the inner-
most ring, respectively, and by applying the normalized average power condition, we can find the
configurations yielding the maximum minimum distance. The better constellations turned out to be
the 4-12-20-28 APSK, 6-12-20-26 APSK, and 8-14-20-22 APSK (Fig. 4), whose degradations with
respect to the 64-ary square (QAM) and optimum hexagonal constellations [17] are 0.81dB and
1.57dB , 0.28dB and 1.03dB, and 0.27dB and 1.03dB, respectively.

III. DATA PREDISTORTION AND NON LINEAR EQUALIZATION TECHNIQUES


3.1 Adaptive data predistortion
In order to make better use of the available HPA power and to cope automatically with the time-
variant response of the HPA (due to thermal instability and/or to automatic control of the transmit-
ted power), an adaptive DP algorithm can be used at the transmitter [2], [9], [10]. The predistortion
criterion is simple: just try to obtain at the HPA output a constellation that, in the average, looks
like the ideal one you would get in the absence of distortion. And do this by apprpriately pre-
distorting the constellation points at the HPA input. Such a technique only calls for a modification
of the transmitted constellation points, without the need to resort to analog devices like those pro-
posed and analyzed in [1]-[4].
Let us assume that ρn and ϑ n denote the polar coordinates of the n -th point of the predis-
torted constellation, with 1≤ n ≤ M . Due to nonlinear distortion, each constellation point will de-
grade at the HPA output into a cluster of scattered points, whose centroid (center of mass) is moved
off its original position at the HPA input. The design is carried out iteratively so that, at the matched
filter output, the centroid of each cluster is brought back to the original “ideal” position of the de-
sired constellation. To be more specific, at the m -th step, the complex-valued difference en (m) be-
tween the estimated centroid and the desired signal point can be exploited as a convenient error
measure in a closed control loop, as follows. Define eρ n (m) = en (m) and eϑ n (m) = arg {e n (m)} as
∆ ∆

the amplitude and the phase errors, respectively. Constellation optimization is achieved by means of
the updating equations
ρ n (m + 1) = ρ n (m ) − γ ρ eρn (m ) , 1 ≤ n ≤ M (3a)

ϑn (m + 1) = ϑn (m ) − γ ϑ eϑn (m) , 1 ≤ n ≤ M (3b)

where γ ρ , γϑ are positive stepsizes, affecting the settling time of the algorithm.

1.0
1.0

0.5
0.5

-1.0 -0.5 0.5 1.0


-1.0 -0.5 0.5 1.0

-0.5 -0.5

-1.0 -1.0

(a) (b)

1.0

0.5

-1.0 -0.5 0.5 1.0

-0.5

-1.0

(c)

Fig. 4 – Optimized 64-point constellations: 4-12-20-28APSK (a), 8-14-20-22APSK (b), and 6-12-20-26APSK (c)

The DP updating rate has to be selected as a compromise between two different requirements.
On one side, it has to be slow enough to avoid loop instability caused by the intrinsic satellite round
trip delay, on the other, it has to be done frequently enough to allow for the tracking of the possible
variations of the HPA response, and for a reasonable short settling time at system start-up or after a
change in the operating point. Furthermore, the intrinsic feedback structure of the DP algorithm as-
sumes that the uplink ground station lies within the satellite footprint.
3.2 Non linear equalization techniques
A. Equalizer Architecture
As is known, a linear equalizer has a poor capability to counteract the nonlinear distortion intro-
duced by the HPA [2]. A nonlinear equalizer with memory proves far more effective, but is also
definitely more demanding in terms of implementation complexity. The latter grows indeed very
quickly with the order of nonlinearity the equalizer is intended to compensate.
For "smooth" nonlinearities such as those involved in satellite HPAs, a good tradeoff between
performance and computational complexity is attained using a nonlinear equalizer (NLE) of third
order and with a memory length of a few T -spaced samples, say up to three. In the following we
consider a third-order NLE, whose input/output equation involves only two summations, as follows
y n = ∑ c m1 x n − m1 +
m1
(4)
∑ ∑ ∑ cm1 ,m2 ,m3 xn − m1 x n − m2 x ∗n −m1
m1 m2 m 3

where xn and y n denotes the equalizer input and output samples at time nT, respectively, and the
coefficients c m1 , c m1 ,m 2 ,m 3 are set as to optimize the NLE performance according to some criterion.
The equalizer schematic is shown in Fig. 5. At step n , the shift register stores the latest N samples
of the matched filter output, i.e., the vector x n = [ xn ,L , xn − N +1 ] , where N represents the equalizer
∆ T

memory. In the following we let N = 3 . The nonlinear combiner carries out all nonlinear combina-
tions of the elements of x n up to the order three, as in (4). The various combiner outputs are finally
scaled by appropriate complex-valued coefficients (or taps), and finally they’re summed up together
to yield the current output sample y n . We note that the equalizer can be thought of as made by the
parallel of two sections: one carrying out a linear combination of the inputs x n (just as a linear
equalizer would do), another devoted to nonlinear operations on the same inputs. The outputs of
these sections are eventually summed up to yield y n .
B. Calculation of the NLE coefficients
Denote with
T
u n = u0( n ) ,L , u L( n−)1  T =  xn , xn −1 ,L , xn − N +1 , xn∗ xn2 , xn∗ xn xn −1 ,L , u n ∈ C L

(5)

the L -dimensional vector of the combiner outputs at time n (see Fig. 5), and with

c n = c0( n ) , c1( n ) ,L , cn( n−)N +1 , c−1, −1, −1, ,L T , c n ∈ C L



(6)

the vector of taps in (4), that we generalize a little by letting it depend on the time index n as well.
Then the NLE output is
yn = cTn u n (7)
xn xn−1 xn−2 xn− N+1
T T T

Nonlinear Combiner

xn xn−1 xn2 xn∗ 2


xn− ∗
N+1 xn− N +1

c0(1) c1(1) (1)


cN−1 (3)
c0,0,0 (3)
cN−1,N −1,N −1

Σ
zn
Fig. 5 – NLE Schematic

The above suggests that the taps c n can be selected in a optimal fashion according to an
MMSE criterion as follows:

copt = arg min  E


cn  {y n − an
2
} (8)

to be implemented iteratively through the gradient algorithm

c m +1 = c m − γ em u∗m (9)

where γ denotes the recursion step-size, and em = y m − am is the output error at the step m . Take
now into consideration the covariance matrices at the input and output of the combiner:
R =∆ E {x ∗ xT }
 x
 (10)
Ru =∆ E {u∗uT }

(where the time index has been dropped for simplicity). It can be shown that the eigenvalues of such
matrices {λ i} and {α i}, respectively, are related as follows [15]:
α min ≤ λ min ≤ λ max ≤ α max (11)
where (λ min ,λ max ), (α min ,αmax ) are the smallest and the largest eigenvalues of Rx and Ru , respec-
tively. Because of (11), and since the convergence rate of (9) is limited by the smallest eigenvalue,
it is expedient to resort to a multi-step recursion, with two different step-sizes, γ l and γ nl to be em-
ployed for the linear and the nonlinear sections, respectively. We note that, in view of the above
discussion, γ nl may be much smaller than γ l . In conclusion, the final version of the recursive equa-
tion for the equalizer coefficients is
c m +1 = c m − Γ em u∗m , (12)
where Γ is the matrix of step-sizes defined as

Γ = diag(γ1
l ,L,γ l ,γ nl,L,γ nl )
4 24
3 1424 3 (13)
N entries L− N entries

IV SIMULATION RESULTS
We illustrate in this section some results of computer simulations performed to evaluate the per-
formance of the (uncoded) QAM and APSK constellations introduced in the previous sections for a
nonlinear satellite channel. We assume rolloff factor α = 0.3 , and ideal symbol timing synchroniza-
tion.
The performance of the linearization algorithms is expressed using the concept of Total Deg-
radation TD , defined as the sum (in decibels) of the amplifier output backoff BO (the ratio be-
tween the HPA saturation output power and the average output power ) and the increment ∆ in the
ratio E b N 0 required to have a given BER, say 10 −3 with respect to the ideal condition of perfectly
linear HPA
TD (dB ) = BO (dB ) + ∆ (dB ) (15)
There will be a value of BO , say BOopt , where the function TD attains an absolute minimum,
which can be used as a performance metric for evaluating the distortion compensation capability of
the algorithm taken into consideration.
All results presented in the following were obtained with different trial system configurations,
namely: i) decision-directed single-tap linear equalizer performing automatic gain control and phase
recovery (AGC-DD); ii) MMSE-NLE algorithm with 3 linear and 27 nonlinear taps (MMSE), iii)
DP algorithm, and finally, iv) joint data predistortion and nonlinear equalization (DP+MMSE). Fig-
ure 6 shows curves of the total degradation as a function of the BO for the conventional square
16QAM constellation, and with a target BER of 10-3. As predicted, for each system configuration
there exist an optimum BO value than ensures the minimum TD. With respect to the trivial AGC, a
NLE slightly improves the system performance (of about 0.4 dB TD), but to gain further (1.5 dB) a
DP algorithm is needed. Finally, the combination of DP and NLE improves significantly perform-
ance over the unprotected system (AGC). The total degradation evaluated for the optimum backoff
BO = 3dB is TD = 3.3dB with respect to the linear channel.
The results for 6-10APSK and 4-8APSK are reported in Fig. 7a and 7b respectively. The suit-
ability of such constellation to the nonlinear channel is apparent: the unprotected 4-8APSK system
(AGC only) starts from a remarkable TD = 3.6dB for BO = 3.4 dB . Again, the combination of DP
and NLE further improves performance and allows to lower the TD to 2.6 dB for BO=2 dB. The 6-
10APSK doesn’t appear to be as brilliant as 4-12APSK. The latter has also the further advantage of
a 90-degree rotational invariance that simplifies mapping onto the constellation points of the output
bits of a channel coder.

10 16 QAM
-3
BER=10
(Eb/N0)lin= 10.5 dB
9

7
Total Degradation (dB)

AGC-DD
2 MMSE Nlin=3 , Nnonlin=27
Data-Predistortion
Data-Pred. + MMSE Nlin=3 , Nnonlin=27
1
1 2 3 4 5 6 7 8 9 10

BO (dB)

Fig. 6 – Optimization of the HPA operating point with 16QAM

10 10
4-12 APK 6-10 APK
-3 -3
BER=10 BER=10
9 (Eb/N0)lin= 10.95 dB 9 (Eb/N0)lin= 10.67 dB

8 8

7 7
Total Degradation (dB)

Total Degradation (dB)

6 6

5 5

4 4

3 3
AGC-DD
AGC-DD
MMSE Nlin=3 , Nnonlin=27 MMSE Nlin=3 , Nnonlin=27
2 2
Data-Predistortion Data-Predistortion
Data-Pred. + MMSE Nlin=3 , Nnonlin=27 Data-Pred. + MMSE Nlin=3 , Nnonlin=27
1 1
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

BO (dB) BO (dB)
(a) (b)
Fig. 7 – Optimization of the HPA operating point with 4-12 (a) and 6-10APSK (b)

The same kind of simulations were also carried out for higher spectral efficient APSK con-
stellations, namely, 64QAM, 4-12-20-28APSK, 6-12-20-26APSK, and 8-14-20-22APSK to deter-
mine the influence of the number of levels on system performance. The results are depicted respec-
tively in Fig. 8 and 9. Comparing Fig. 6 with Fig. 8 and Fig. 7 with 9, it is apparent that the 64-
point constellation turns out to be more sensitive to nonlinear distortion (as largely expected). But
the gain provided by the compensation algorithms is also stronger. Again, 64APSK constellations
yield better performance in the case of unprotected system when compared to 64QAM. Our better
configuration, corresponding to joint DP+NLE, provides a minimum total degradation equal to
about TD = 5dB both for 64QAM and 64APSK, contrarily to our previous results about 16-ary
modulation.

15
64 QAM
-3
14 BER=10
(Eb/N0)lin= 14.76 dB
13

12

11

10

4
AGC-DD
3 MMSE Nlin=3 , Nnonlin=27
2 Data-Predistortion
Data Pred. + MMSE Nlin=3 , Nnonlin=27
1
2 4 6 8 10 12 14
BO (dB)

Fig. 7 – Optimization of the HPA operating point with 4-12 (a) and 6-10APSK (b)

V CONCLUSIONS AND FUTURE WORK


The aim of this paper was the performance analysis of a nonlinearity compensation scheme for a
(nonlinear) satellite link with 16- or 64-ary modulations, encompassing both data predistortion at
the ground station and nonlinear equalization in the receiver. Based on the results described above,
we conclude the following:
• for unprotected 16-ary systems, optimized 4-8APSK outperforms 16QAM and 6-10APSK;
in particular, 4-8APSK bears a TD of 3.6 dB at BER=10-3 @ BO=3 dB;
• 16QAM protected by a joint PD+NLE gives better performance compared to 4-8APSK;
• 64APSK, such as 4-12-20-28APSK and 8-14-20-22APSK, outperform 64QAM in the un-
protected case, but yields the same 5 dB TD with joint PD+NLE.
Therefore, the application of 4-8APSK to satellite broadcasting is highly recommended. The adop-
tion of DP and/or NLE is a matter of cost vs. QoS considerations, but anyway the complexity in-
crease of a satellite set-top box due to the inclusion of an adaptive NLE seems quite modest. The
results shown in this paper are to be re-evaluated assuming a particular channel coding scheme,
such as pragmatic turbo coding, to yield quasi-error-free transmission.

15 15 6-12-20-26 APK
4-12-20-28 APK
-3 -3
14 BER=10 14 BER=10
(Eb/N0)lin= 15 dB (Eb/N0)lin= 15 dB
13 13

12 12

11 11

Total Degradation (dB)


10
10
9
9
8
8
7
7
6
6
5
5
4
4 AGC-DD
AGC-DD 3 MMSE Nlin=3 , Nnonlin=27
3 MMSE Nlin=3 , Nnonlin=27 Data-Predistortion
2
2 Data-Predistortion Data Pred. + MMSE Nlin=3 , Nnonlin=27
Data Pred. + MMSE Nlin=3 , Nnonlin=27 1
1 2 4 6 8 10 12 14
2 4 6 8 10 12 14
BO (dB)
BO (dB)
(a) (b)
15
8-14-20-22 APK
-3
BER=10
14
(Eb/N0)lin= 15 dB

13

12

11

10
Total Degradation (dB)

3 AGC-DD
MMSE Nlin=3 , Nnonlin=27
2 Data-Predistortion
Data Pred. + MMSE Nlin=3 , Nnonlin=27
1
2 4 6 8 10 12 14

BO (dB)
(c)

Fig. 8 – Optimization of the HPA operating point with 4-12-20-28 (a). and 8-14-20-22 (b), and 6-12-20-26APSK (c)
REFERENCES
[1] J. Namiki, “An Automatically Controlled Predistorter for Multilevel Quadrature Amplitude Modula-
tion,” IEEE Trans. Commun., vol. 31, pp. 707-712, May 1983.
[2] G. Karam and H. Sari, “Analysis of Predistortion, Equalization, and ISI Cancellation Techniques in
Digital Radio Systems with Nonlinear Transmit Amplifiers,” IEEE Trans. Commun., vol. 37, pp. 1245-
1253, Dec. 1989.
[3] M.-G. Di Benedetto and P. Mandarini, “A New Analog Predistortion Criterion with Application to High
Efficiency Digital Radio Links,” IEEE Trans. Commun., vol. 43, pp. 2966-2974, Dec. 1995.
[4] N. A. D’Andrea, V. Lottici and R. Reggiannini, “RF Power Amplifier Linearization through Amplitude
and Phase Predistortion,” IEEE Trans. Commun., vol. 44, Nov. 1996.
[5] Y. Levy, G. Karam and H. Sari, “Adaptation of a Digital Predistortion Technique Based on Intersymbol
Interpolation,” IEEE GLOBECOM ‘95, Singapore, 1995.
[6] M. G. Di Benedetto and P. Mandarini, "An Application of MMSE Predistortion to OFDM Systems",
IEEE Trans. Commun., vol. 44, n. 11, pp. 1417-1420, Nov. 1996.
[7] A. Brajal and A. Chouly, "Compensation of Nonlinear Distortions for Orthogonal Multicarrier Schemes
Using Predistortion," IEEE GLOBECOM '94, San Francisco, CA, Nov. 1994, pp. 1909-1914.
[8] H. W. Kang, Y. S. Cho and D. H. Youn, "On Compensating Nonlinear Distortions of an OFDM System
Using am Efficient Adaptive Predistorter," IEEE Trans. Commun., vol. 47, pp. 522-526, April 1999.
[9] A. A. M. Saleh and J. Salz, “Adaptive Linearization of Power Amplifiers in Digital Radio Systems,”
Bell Syst. Tech. Journal, vol. 62, pp. 1019-1033, April 1983.
[10] G. Karam and H. Sari, “A Data Predistortion Technique with Memory for QAM Radio Systems,” IEEE
Trans. Commun., vol. 39, pp. 336-344, Feb. 1991.
[11] M. F. Mesiya, P.J. McLane and L.L. Campbell, “Maximum Likelihood Sequence Estimation of Binary
Sequences Trasmitted Over Bandlimited Nonlinear Channels,” IEEE Trans. Commun., vol. 25, pp. 633-
643, July 1977.
[12] A. Gutierrez and W.E.Ryan, “The Performance of MLSD Receivers of Nonlinear Satellite Channels,”
IEEE ICC’96, vol. 57, pp. 921-925, June 1996.
[13] D. D. Falconer, “Adaptive Equalization of Channel Nonlinearities in QAM Data Trasmission,” Bell
Syst. Tech. J., vol. 57, pp. 2589-2611, Sept. 1978.
[14] S. Benedetto and E. Biglieri, “Nonlinear Equalization of Digital Satellite Channels,” IEEE J. Sel. Areas
Commun., vol. 1, pp. 57-62, Jan. 1983.
[15] A. Gutierrez and W.E. Ryan, “Performance of Adaptive Equalizer on Nonlinear Satellite Channels,”
IEEE ICC’96, pp. 488-492, June 1995.
[16] R. De Gaudenzi, A. Guillen i Fabregas, A. M. Vicente, and B. Ponticelli, “High Power and Spectral
Efficiency Coded Digital Modulation Schemes for Nonlinear Satellite Channels,” 7th International ESA
Workshop on Digital Signal Processing Techniques for Space Applications, Oct. 2001.
[17] G. F. Forney Jr., R. G. Gallager, G. R.Lang, F. M. Longstaff and S. U. Quereshi, “Efficient Modulations
for Band-Limited Channels,” IEEE J. Sel. Areas Commun., vol. 2, pp. 632-647, Sept. 1984.

S-ar putea să vă placă și