Académique Documents
Professionnel Documents
Culture Documents
9, SEPTEMBER 2006
Abstract— This paper investigates the performance of M -ary Past literature has extensively covered techniques to im-
Amplitude-Phase Shift Keying (APSK) digital modulation over prove highly spectral efficient modulations performance [4]–
typical nonlinear satellite channels. The effect of the satellite non- [7]. However, this work typically refers to uncoded square
linearity is studied, and distortion pre- and post-compensation
techniques for coded APSK are presented. Moreover, clock QAM constellations, results of lesser importance for nonlinear
timing, signal amplitude and carrier phase recovery schemes are satellite channels. Circular APSK modulations were already
discussed. For the latter, a new class of non turbo decoder-aided proposed thirty years ago in [8], where uncoded bit error rate
closed-loop phase synchronizers featuring good performance and bounds for several non band-limited APSK sets were provided;
low complexity is studied. Finally, an end-to-end coded APSK the suitability of APSK for nonlinear channels was also made
system simulator inclusive of the satellite channel model and
synchronization sub-systems is discussed and its performance explicit, but concluded that for single carrier operation over
compared to standard trellis-coded QAM concatenated with nonlinear channel APSK performs worse than PSK schemes.
Reed-Solomon codes, showing a remarkable gain in both power In the current paper we revert this conclusion.
and spectral efficiency. Coded APSK, recently selected for the
The possibility of modulator pre-compensation was briefly
new standard –DVB-S2– for digital video broadcasting and
interactive broadband satellite services [1], is shown to represent mentioned in [8]. General techniques to mitigate the nonlinear
a power- and spectral-efficient solution for satellite nonlinear distortion effects are well covered in the literature [4], [5], [7],
channels. [9]–[15]. Pre-compensation techniques counteract the ampli-
Index Terms— Turbo codes, amplitude-phase shift keying fier distortion through constellation pre-distortion at the trans-
(APSK) modulation, bit-interleaved coded modulation (BICM), mitter, while post-compensation techniques mitigate nonlinear
coded modulation, nonlinear channels, phase synchronization, distortion effects at the demodulator side through nonlinear
time synchronization, satellite communications. equalization and/or ad-hoc decoder metric computation. In
this paper we review the existing works and we discuss their
I. I NTRODUCTION applicability to APSK.
5 90
amplitude and carrier phase estimation schemes. Sect. VI gives
a brief description of the system we use as benchmark for 80
√ L−1
sT (t) = P x(k)pT (t − kTs ) (1)
by an amount D with respect to an ideal linear AWGN chan-
k=0
nel. This quantity depends on the HPA distortion and hence on
where P is the signal power, x(k) is the k-th transmitted sym- the IBO/OBO characteristic (see Fig. 1). With this degradation,
bol, drawn from a complex-valued APSK signal constellation the effective demodulator input Eb /N0 , Eb /N0 |eff , is given by:
X , with |X | = M , described in detail shortly hereafter, pT (t)
is the SRRC transmission filter impulse response, and Ts is Eb Eb
(IBO) = +OBO (IBO)+D(IBO) (in dB). (4)
the symbol duration (in seconds), corresponding to a channel N0 sat N0 eff
use. The coded modulation spectral efficiency R is the number With Eq. (4) we can find an optimum HPA operating point
of information bits divided by the modulator baud rate, i. e., that minimizes Eb /N0 |sat . This point represents the best trade-
R = r log2 M , r being the coding rate and M the modulation off between the increasing power loss (OBO) related to the
cardinality. higher IBO and the reduction of the distortion (D) due to the
The signal sT (t) passes through an amplifier (HPA) op- improved linearity experienced by a larger IBO.
erated close to the saturation point. In this region, the HPA
shows non-linear characteristics that induce phase and am- III. C ONSTELLATION O PTIMIZATION IN AWGN
plitude distortions to the transmitted signal. The amplifier is
M -APSK constellations are composed of nR concentric
modeled by a memoryless non-linearity, with an output signal
rings, each with uniformly spaced PSK points. The signal
sA (t) at time t given by:
constellation points x are complex numbers, drawn from a
sA (t) = F |sT (t)| ej(φ(sT (t))+Φ(|sT (t)|)) , (2) set X given by:
⎧
where we have implicitly defined F (A) and Φ(A) as the ⎪
⎪ r1 e
j n2π
i+θ1
i = 0, . . . , n1 − 1, (ring 1)
⎪
⎪
1
AM/AM and AM/PM characteristics of the amplifier for a ⎪
⎪
⎨r ej n2π2 i+θ2 i = 0, . . . , n2 − 1, (ring 2)
signal with instantaneous signal amplitude A. The signal X = .
2
amplitude is the instantaneous complex envelope, so that the ⎪
⎪ ..
⎪
⎪
baseband signal is decomposed as sT (t) = |sT (t)|ejφ(sT (t)) . ⎪
⎪
⎩ j n2π i+θnR
For the numerical examples in the following we will consider rnR e R i = 0, . . . , nnR − 1, (ring nR )
the AM/AM and AM/PM characteristics shown in Fig. 1. (5)
We introduce the parameter Eb /N0 |sat defined as the ratio where we have defined n , r and θ as the number of points,
between the transmitted energy per bit when the amplifier is the radius and the relative phase shift corresponding to the
driven at saturation by a continuous wave (CW) carrier and -th ring respectively. We will nickname such modulations as
the noise power spectral density at the demodulator input n1 +. . .+nnR −APSK. Fig. 2 depicts the 4+12- and 4+12+16-
[10]. Note that the sub-fix “sat” refers to HPA saturation. APSK modulations with quasi-Gray mapping. In particular, for
The signal-to-noise ratio at the demodulator input Eb /N0 |inp next generation broadband systems [1], [3], the constellation
is reduced by the output back-off (OBO, in dB) with respect to sizes of interest are |X | = 16 and |X | = 32, with nR = 2 and
the value in a system operating with a single constant-envelope nR = 3 rings respectively. In general, we consider that X is
signal at amplifier saturation: normalized in energy, i.e., E[|x|2 ]
= 1, which implies that the
n r2 = 1. Notice also
nR
radii r are normalized such that =1
Eb Eb
(IBO) = (IBO) + OBO (IBO) (in dB). (3) that the radii r are ordered, so that r1 < . . . < rnR .
N0 sat N0 inp For convenience we define the phase shifts and the ring
Additionally, due to constellation warping and satellite radii in relative terms rather than in absolute terms, as in (5).
channel induced ISI, the demodulator performance is degraded We let φ = θ − θ1 for = 1, . . . , nR be the phase shift of
2398 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 5, NO. 9, SEPTEMBER 2006
4
4+12−APSK
5+11−APSK
1+5+10−APSK
(01110) (11110) 3.5 6+10−APSK
16−PSK
16−QAM
(01010) (11010) 6+10−APSK dmin
3.08
3.06
3.02
1111 (01001) 1011 (01000) 0011 (11000) 0111 (11001) 3
2.98
r2 r1 1.5 2.96
2.94
2.92
r3
1100 (00101) 0100(10101) 0.5
0 5 10 15
(00111) (10111) E /N (dB)
b 0
1000 (00100) 0000 (10100)
Fig. 3. Mutual information for the optimized APSK signal constellations
(00010) (10010) versus QAM and PSK.
0.06
compensation parameters can be calculated off-line, in the
16 QAM following steps: 1) Generation of S blocks of W symbols
4+12 APSK ρ=2.8
0.05
5+11 APSK ρ=2.65 over which the symbol matched filter (SMF) centroids are
6+10 APSK ρ = 2.5
16 PSK computed (transmission in the absence of white noise); 2)
0.04
Computation of the error signal at the end of each block; 3)
Pre-distorted constellation point update. The latter task can be
readily achieved through an iterative least mean square (LMS)
pdf
0.03
type of algorithm illustrated by the following equations:
|x(n)
pre |(s + 1) = |x(n) (n)
pre |(s) − γr ec (s) (7)
0.02
(n) (n)
arg xpre (s + 1) = arg xpre (s) − γφ ψ(s) (8)
jθc(n) (s)
0.01 e(n)
c (s) = rc(n) (s) e (n)
− |x | (9)
1
jθc(n) (s)
rc(n) (s) e = y(k) (10)
W
0
−20 −15 −10 −5 0 5 10 k∈ln
Signal power (dB) sW +1≤k≤(s+1)W
Fig. 4. Simulated histogram of the transmitted signal envelope power for ⎧ (n) (n)
16-ary constellations. ⎪
⎨arg ec (s) − 2π if arg ec (s) > π
(n) (n)
ψ(s) = arg ec (s) + 2π if arg ec (s) < −π
⎪
⎩ (n) (n)
arg ec (s) if | arg ec (s) | ≤ π
A. Pre-compensation at the Transmitter (11)
Although results presented in Sect. III indicated a slight where the index n refers to the constellation point, l(n)
superiority of 6+10-APSK, for nonlinear transmission over an
indicates the conditioning to the constellation point n, s refers
amplifier, 4+12-APSK is preferable to 6+10-APSK because to the iteration step of the algorithm, y(k) represents the k-
the presence of more points in the outer ring allows to th SMF output complex sample, x(n) represents the APSK
maximize the HPA DC power conversion efficiency. It is better (n) (n)
complex constellation reference point, rc (s) and θc (s) are
to reduce the number of inner points, as they are transmitted
the modulus and the phase of the SMF output complex n-th
at a lower power, which corresponds a lower DC efficiency – (n)
centroid computed at step s, is xpre (s) the pre-distorted n-th
the HPA power conversion efficiency is monotonic with the
constellation point computed at step s, γr and γφ the adap-
input power drive up to its saturation point. Fig. 4 shows
tation steps for the pre-distorted constellation point modulus
the distribution of the transmitted signal envelope for 16-
and the phase respectively.
QAM, 4+12-APSK, 6+10-APSK, 5+11-APSK, and 16-PSK;
For example, the optimal 4+12-APSK pre-distortion para-
the shaping filter is a square-root raised cosine (SRRC) with
meters are ρ2 = 3.5 and Δφ = 25 deg for an IBO=3 dB
roll-off factor α = 0.35. As we observe, the 4+12-APSK enve-
and ρ2 = 3.7 and Δφ = 27 deg for a smaller IBO=2 dB
lope is more concentrated around the outer ring amplitude than
(see Fig. 5-a). As expected, the pre-distorted constellation is
16-QAM and 6+10-PSK, being remarkably close to the 16-
expanded, e. g., ρ2 > ρ2 . The new constellation points x
PSK case. This shows that the selected constellation represents
follow (5), with new radii r1 , r2 , such that F (r1 ) = r1 , and
a good trade-off between 16-QAM and 16-PSK, with error
F (r2 ) = r2 .
performance close to 16-QAM, and resilience to nonlinearity
Concerning the phase, it is possible to pre-correct for effect
close to 16-PSK. Therefore, 4+12+APSK is preferable to the
of the HPA on the phase HPA between inner and outer
rest of 16-ary modulations considered. Similar advantages
rings through a simple change in the relative phase shift by
have been observed for 32-APSK compared to 32-QAM.
φ2 = φ2 + Δφ, with Δφ = φ(r2 ) − φ(r1 ). These operations
1) Static Distortion Compensation: The simplest approach can be readily implemented in the digital modulator by sim-
for counteracting the HPA nonlinear characteristic is to modify
ply modifying the reference constellation parameters ρ , φ ,
the complex-valued constellation points at the modulator side. with no hardware complexity impact or out-of-band emission
Thanks to the multiple-ring nature of the APSK constellation, increase at the linear modulator output. The compensation
pre-compensation is easily done by a simple modification of
effort is shifted into the modulator side, allowing the use of an
the parameters ρ and φ . The known AM/AM and AM/PM optimal demodulator/decoder for AWGN channels even when
HPA characteristics are exploited in order to obtain a good the amplifier is close to saturation. The signal at the modulator
replica of the desired signal constellation geometry after the output is then
HPA, as if it had not suffered any distortion. This can be
simply obtained by artificially increasing the relative radii ρ √
L−1
1.5
but not in a modification of the AM/AM, AM/PM charac-
teristics shape. The limited gain variations are compensated
by the satellite transponder automatic level control (ALC)
1
device, thus off-line pre-compensation has a long term value.
If required, the compensated parameters can be adapted to
track larger slow variations in HPA characteristic due to aging.
0.5
2) Dynamic Distortion Compensation: The dynamic pre-
compensation approach by Sari [5] considers not only the
current constellation symbol, as done in the case of static pre-
Q
0
compensation, but also for Q symbols before and after. In gen-
eral for M-QAM modulation the possible pre-distortions are
now M 2Q+1 . Similar to the static pre-compensation described
−0.5
before, the pre-distorter complex values can be obtained off-
line minimizing the minimum squared error (MSE) between
the ideal constellation and the noiseless points measured after
−1
the demodulator symbol matched filter. This can be achieved
through an extension of the methodology described for the
static approach. In this case, the number of updating equations
−1.5
−1.5 −1 −0.5 0 0.5 1 1.5 for M-QAM is M 2Q+1 which can be reduced to M 2Q+1 /4
I
exploiting the M-QAM quadrant symmetry [5]. For APSK
(a)
the memory requirement can be further reduced to 3M Q /16.
Scattering Diagram of XDAGC - Sampled at 10 The main drawback of this technique is the amount of time
5
required to compute the pre-distortion coefficients and the
memory required which grows rapidly with the constellation
4
and memory parameters (M , Q). It should also remarked that
3
the dynamic pre-compensation is less effective than the static
one when more than one carrier is passing through the same
2 HPA. This is because the uncorrelated intermodulation noise
among carriers will dominate over the ISI thus making the
1 dynamic pre-compensation ineffective.
Quadrature
0
B. Dynamic Post-compensation at the Receiver
-1 An alternative approach to dynamic pre-distortion tech-
niques is represented by demodulator equalization. Two main
-2 approaches are revisited in the following, namely linear and
nonlinear equalization. The typical linear equalization corre-
-3
sponds to the case of a decision-directed complex AGC (also
called vector tracker) see Appendix II of [14]. It was observed
-4
in [10] that linear equalization does not provide any significant
improvement over nonlinear channels.
-5
-5 -4 -3 -2 -1 0 1 2 3 4 5 The post-compensation in the (trellis) decoder metric calcu-
In-phase
lation was proposed in [14]. In that work, the metric computer
(b)
for the TCM decoder used the distorted constellation X as
Fig. 5. 4+12 PSK signal constellation: a) Modulator output roll-off factor reference constellation, and the metric corresponding to the
0.35 with pre-compensation (ρ2 = 3.7, Δφ = 27 degrees); b) Demodulator distorted constellation point x ∈ X is given by,
SRRC filter output noiseless scatter diagram for 16-APSK in the nonlinear
channel for IBO=2 dB, roll-off factor 0.35 with pre-compensation (ρ2 = 3.7, 1
Δφ = 27 degrees). M (y|x ) ∝ exp − |y − Es x |2 (13)
N0
where ∝ indicates proportionality and the x ∈ X are the
receiver matched filter is shown in Fig. 5-b for 16-APSK with centroids of the distorted constellation at the matched filter
IBO=2 dB. For clarity, the scatter diagram at the SMF has output.
been obtained in the absence of AWGN. Despite the strong A somewhat more complex approach is adaptive nonlinear
channel nonlinearity the center of mass, corresponding to the equalization, based on a Volterra series expansion [10], where
scattered diagram, closely follows the optimum 4+12-APSK it is shown to provide considerable advantages for PSK
constellation, for which the optimum parameters are ρ2 = 2.7 modulations in terms of ISI reduction. Furthermore, it also
and φ = 0. allows to reduce the optimum operating IBO, thus improving
Measurements showed that the HPA characteristic sensitiv- the HPA utilization. However, it is expected that this advantage
ity to temperature or aging results in a limited change of gain will be more pronounced for 16-QAM than for double-ring
DE GAUDENZI et al.: PERFORMANCE ANALYSIS OF TURBO-CODED APSK MODULATIONS OVER NONLINEAR SATELLITE CHANNELS 2401
∗
where X represents the set of constellation points in the first
eτ (k) = Re y(k) y k + −y k− , (15) quadrant of the complex plane, x is one of the elements of
2 2
X and W is the phase estimator averaging time in symbols.
and y(k) are the symbol matched filter output samples. As The quadrant decision approach is somewhat fragile against
is apparent, the error detector signal is obtained using the thermal noise due to the vicinity of some of the 16-APSK
on-time interpolated sample y(k) y[(k + εk )Ts ] and the constellation points to the decision region boundaries. The
late/early samples y(k + 12 ) and z(k − 12 ) shifted by half ratio between the simulated Decision Directed (DD) 16-APSK
the symbol time. The updating step-size γτ is related to the phase standard deviation and the Modulation Constrained
equivalent one-sided loop noise bandwidth BLτ as follows, Cramer-Rao (MCR) bound [25] is about 3.55 at Es /N0 =
4BLτ Ts 10 dB, while for 16-QAM the same ratio is reduced to
γτ = , (16) 1.92. This problem can be overcome by narrowing the loop
Aτ (1 + 2BLτ Ts )
bandwidth when possible. However, in the presence of impor-
where Aτ is the slope at the equilibrium point of the average tant phase noise contributions originating from the consumer
TED characteristic (S-curve). As observed in [20], the timing receiver radio frequency front-end, the narrow loop noise
discriminator performance is invariant to carrier phase, so that bandwidth reduction will generate cycle slips. Furthermore,
timing information can be extracted without prior acquisition by doing so, the maximum residual frequency error that can
of the carrier recovery loop. As for the digital timing inter- be tolerated by the phase recovery loop is also reduced. Thus
polator, we employed the cubic interpolator described in [21] excessive narrowing the loop noise bandwidth to combat the
and [22]. phase jitter should be avoided.
In the following, we propose a general class of P -th
B. Amplitude Recovery power closed-loop NDA phase synchronizers for APSK. The
corresponding block diagram is shown in Fig. 6. Compared
The NDA Automatic Gain Control (AGC) algorithm for the to the 4Q-DD scheme an P -th order nonlinearity followed by
acquisition phase described in [14] is used, for which the the a θP phase rotation is introduced before the quadrant slicing
(k+1)-th value of the variable gain Ck+1 is derived recursively device. The proposed phase error detector algorithm has the
as:
following form:
Ck+1 = Ck − γa Ck |y(k)| − A1 , (17)
w(k) = y P (k)ejθP ,
where A1 E{|y(k)|}. For a loop adaptation step γα of
5 · 10−4 , the amplitude rms error provided by the NDA AGC, eφ (k) = Im w(k) · sign Re w(k) − j sign Im w(k)
is fully acceptable (see [14] for details). (19)
2402 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 5, NO. 9, SEPTEMBER 2006
0.25
To the decoder
0.2
Input Es/N0=10 dB E /N =15 dB
s 0
signal
(.) P 0.15
0.1
sign(R e{.})
− j sign(Im{. })
e− j φ̂ 0.05
e jθ P
S−curve
0
−0.1
Fig. 6. APSK generalized phase tracker block diagram.
−0.15
1.5
E /N =∞
Before cubic nonlin. −0.2 s 0
After cubic nonlin.
−0.25
−100 −80 −60 −40 −20 0 20 40 60 80 100
1 Phase error [deg]
0.8
0 0.6
0.4
Es/N0=15 dB
−0.5
0.2
S−curve
0
−1
−0.2
−1.5 −0.4
−1.5 −1 −0.5 0 0.5 1 1.5
E /N =10 dB
In−phase Component s 0
−0.6
−0.8
−100 −80 −60 −40 −20 0 20 40 60 80 100
Phase error [deg]
The error signal eφ (k) is then used to update the first order
phase loop phase according to φ(k) = φ(k − 1) − γφ eφ (k), Fig. 9. Simulated APSK phase error detector S-curve for Es /N0 =
∞, 15, 10 dB; 4+12-APSK C-4Q-DD scheme.
where the loop gain is related to the bandwidth BLφ by the
classical relation,
4BLφ Ts constellation, with amplitude 4r23 , while the internal points
γφ = . (20)
Aφ (1 + 2BLφ Ts ) remain on a QPSK constellation with amplitude r13 . For typical
The DD phase discriminators S-curve Aφ depends on the 4+12-APSK parameters the distance between the two rings is
actual Es /N0 [26]; the effective loop noise bandwidth depends largely increased. This process is illustrated in Fig. 7, where
in turn on the operating Es /N0 . Following [26], it is common the input 4+12-APSK signal constellation is depicted, together
practice to compute the discriminator slope Aφ |Es /N0 =+∞ and with the constellation after the cubic nonlinearity.
the corresponding loop bandwidth for Es /N0 = ∞, and refer A simple 4Q-DD (quadrant slicer type) closed-loop phase
in the simulations to BLφ Ts (∞) = BLφ Ts |Es /N0 =∞ . estimator [26] operating on the outer QPSK ring can be
The 4Q-DD scheme described in [24] is obtained by setting efficiently applied to the cubic nonlinearity output. This
P = 1 and θP = 0 in Fig. 6. Because of the simplified 4Q- scheme dubbed C-4Q-DD is obtained by setting P = 3,
DD approach the phase estimator implementation is simpler θP = 0 in Fig. 6. The main scheme drawback is related the
than a truly DD scheme for 16-APSK and decisions are thermal noise terms enhancement at low signal-to-noise ratios
insensitive to possible signal amplitude estimation errors. The due to the cubic nonlinearity. For 32-APSK a fourth power
operations performed during this estimator suggest an alterna- device transforms the constellation into an outer ring QPSK
tive scheme that keeps a similar complexity, while improving constellation with amplitude 4r33 , a middle ring composed of
its performance. Raising a 4+12-APSK constellation to the three points [4r23 (ejπ/3 , −1, e−jπ/3 )] and single inner point
3rd power (in the complex domain), the original 4+12-APSK with amplitude −2r33 . A phase shift θP = π/4 is also required
constellation transforms into a QPSK one. It is easy to see to get the right constellation orientation after the nonlinearity
that the twelve external ring points collapse into a QPSK for the slicer correct operation. Therefore for 32-APSK the
DE GAUDENZI et al.: PERFORMANCE ANALYSIS OF TURBO-CODED APSK MODULATIONS OVER NONLINEAR SATELLITE CHANNELS 2403
1
in Fig. 10: it is remarked that the narrower S-curve stable
0.8 region due to the higher constellation order. Nonetheless PED
0.6
sensitivity to phase error appears very good.
Es/N0=15 dB Fig. 11 illustrates the phase error (jitter) standard deviation
0.4
obtained by simulation for 4+12-APSK and 4+12+16-APSK
0.2 and the 4Q-DD, C-4Q-DD and F-4Q-DD phase estimators.
These results indicate that the C-4Q-DD PED provides a
S−curve
0
remarkable phase error standard deviation improvement factor
−0.2 of about 3 at Es /N0 = 10 dB compared to the 4Q-DD scheme
of [24]. At Es /N0 = 20 dB the improvement corresponds to
−0.4
Es/N0=10 dB almost one order of magnitude. As we observe, a phase jitter
−0.6 standard deviation of about 0.4 degrees can be achieved at
−0.8
Es /N0 = 10 dB for BLφ Ts (∞) = 10−4 . Performance results
E /N =∞
s 0
for 32-APSK using the F-4Q-DD PED are also remarkably
−1
−100 −80 −60 −40 −20 0 20 40 60 80 100 good. It should be recalled that 32-APSK will typically operate
Phase error [deg] at higher Es /N0 than 4+12-APSK.
In principle the proposed phase estimators can also work
Fig. 10. Simulated APSK phase error detector S-curve for Es /N0 =
∞, 15, 10 dB; 4+12+16-APSK F-4Q-DD scheme. in feed-forward mode by performing a moving average of
the complex error signal contained in the argument of the
10
1 imaginary function in the phase error expression above, and
16−APSK, C−4Q −DD
16−APSK, 4Q −DD then extracting the complex signal phase. However, similarly
32−APSK, F −4Q −DD
to what is seen in Figs. 8-10, the discriminator S-curve shape
is dependent on the actual Es /N0 . While the shape has no
major impact on the feedback configuration (apart from the
Phase error standard deviation [deg]
10
0
above mentioned loop gain changes), for the feed-forward
estimator the non-linear S-curve characteristic and its slope
dependencies on the SNR implies a biased estimator for non-
zero phase errors. This major drawback, due to low SNR
quadrant decision errors, renders impractical the exploitation
−1
10
of the feed-forward phase estimator scheme.
Information AWGN
SRRC
Constellation
bits Linear
geberator
ak filter
S/P Encoder Mapper
SRRC cos ω0t
filter HPA
K Binary n N
code π
C
Outer Inner
code πc code
CO CI
I Timing
ADC SRRC Linear
interpol a k–D
demodulator
S/P ^
e –jφ Nonlinear
1/Ts equalizer
Q Timing Phase
ADC SRRC interpol error
Early On-time detector
Timing
~ ~ error
ω0 ωs detector
Binary
ρτ(k) Demapping –1
decoder
information data bits ak , at rate Rb , enter a serial-to-parallel enters the digital vector tracker performing only amplitude
device (S/P) generating three parallel streams at rate Rb /3. estimation. A dedicated phase error detector is present at
An encoder with rate r = 3/4 generates four parallel binary the output of the digital AGC. If required, QAM frequency
symbol streams at rate Rs = Rb /3, that are mapped through estimators are applicable to our case [27]. The turbo decoder
a Gray mapping onto the APSK constellation generator. The represents the last demodulator block.
modulator I-Q streams are at symbol rate Rs = Rb /3. The
I-Q multilevel digital pulse stream is then passed to the two B. Trellis-coded APSK
baseband SRRC filters and I-Q modulated at RF. In case of We first consider a standard trellis-coded modulation (TCM)
the nonlinear channel the passband real signal then drives the scheme over APSK modulations with set-partitioning or quasi-
HPA whose model has been described in Sect. II. White noise set-partitioning binary labeling rules. The performance of the
representing the downlink satellite channel is then added. 16-QAM demodulator in the linear channel, including ampli-
The passband demodulator input signal is I-Q demodulated tude, timing and phase estimation subsystem degradations, is
to baseband using standard analog or digital techniques. The I- very close to the analytical upper bound in AWGN [28], [29].
Q streams are then asynchronously sampled at Ns = Ts /Tclock For the 16-state code the coding gain of coded 16-QAM with
samples/symbol. In practical implementations the number of respect to uncoded 8-PSK at Pe = 10−5 is about 4.4 dB. For
samples/symbol Ns depends on the characteristic of the anti- quasi-error free (QEF) performance, e.g. BER on the order of
aliasing filter contained in the I-Q demodulator front-end. 10−10 , we consider a concatenated scheme for which the outer
These samples, assumed for simplicity to be not quantized, code is the Reed Solomon code RS (188, 204) and the inner
are then passed to the receiver SRRC matched filters. The code is the TCM. Then, the inner decoder BER is in the order
SRRC filter outputs are then sub-sampled at two samples per of 2·10−4 [2]. The selected 16-state optimal trellis code for
symbol (on-time and early) by a digital interpolator [21] driven 16-QAM provides performance very close to that of the 64-
by the timing error detector unit. The output symbol stream state binary pragmatic trellis code selected for the DVB-SNG
is then split into on-time and early samples through a simple standard [2].
S/P converter. Both streams are used to derive the timing error At the receiver side, we used the NDA approach of [20]
information. The on-time one sample per symbol stream then for timing recovery, while for amplitude and phase recovery
DE GAUDENZI et al.: PERFORMANCE ANALYSIS OF TURBO-CODED APSK MODULATIONS OVER NONLINEAR SATELLITE CHANNELS 2405
IBO=3 dB, roll −off=0.35 BER AWGN and TWTA (IBO=2dB), roll −off =0.35
−1 0
10 10
16QAM
16QAM+Eq.
16QAM+Pos t−Comp.
16QAM+Pos t−Comp+Eq.
4+12−PSK −1
−2
4+12−PSK+Eq. 10
10 4+12−PSK+Pre−Comp
4+12−PSK+Pre−Comp.+Eq
−2
10
−3
10
BER
BER
−3
10
−4
10
−4
10
−5
10 4+12−PSK, NDA sync. 10it AWGN
−5
10 4+12−PSK, NDA sync. 6 it. AWGN
4+12−PSK, perfect sync. 10it AWGN
16−QAM perfect sync. 10it AWGN
4+12−PSK, NDA sync., 10 it TWTA
−6
4+12−PSK, NDA sync., 6it TWTA
−6
10 10
7 8 9 10 11 12 13 14 15 16 5 5.5 6 6.5 7 7.5 8
[Eb/N0 ] sat(dB) [Eb/N0]sat (dB)
Fig. 13. Simulated BER for TCM 16-QAM and 4+12-APSK in the nonlinear Fig. 14. Rate r =3/4 turbo coded simulated BER of 16-QAM and 16-APSK
channel: 16-state code, IBO= 3 dB, roll-off factor=0.35. in the AWGN channel and the nonlinear channel with IBO= 2 dB, roll-off
factor=0.35, clock timing and carrier phase recovery.
in power and 8 % in spectral efficiency compared to the [9] R. De Gaudenzi, A. Guillén i Fàbregas, A. Martinez, and B. Ponticelli,
conventional TC-16-QAM representing today’s baseline for “A new coded digital modulation scheme for nonlinear satellite channels
with high power and spectral efficiency,” European Space Agency
high-speed satellite links. The adoption of the more complex (ESA), ESA Technical Report ESA STR-242, Aug. 2001.
dynamic pre-distortion technique with memory Q = 5 brings [10] S. Benedetto, E. Biglieri, and V. Castellani, Digital Transmission Theory.
an extra gain of about 0.8 dB in terms of Eb /N0 |sat at the Prentice Hall, 1987.
[11] E. Biglieri, A. Gersho, R. D. Glitin, and T. L. Lim, “Adaptive
optimum operating point corresponding now to IBO = 1 dB. cancellation of nonlinear intersymbol interference for voiceband data
transmission,” IEEE J. Sel. Areas Commun., vol. 2, no. 5, pp. 765–777,
Sept. 1984.
VII. S UMMARY AND C ONCLUSIONS [12] S. Benedetto and E. Biglieri, “Nonlinear equalization of digital satellite
channels,” IEEE J. Select. Areas Commun., vol. 1, no. 1, pp. 57–62,
Extensive analysis and simulations for coded APSK modu- Jan. 1983.
lations, with particular emphasis on its applicability to satellite [13] A. Bernardini and S. D. Fina, “Analysis of different optimization criteria
for IF predistortion in digital radio links with nonlinear amplifiers,”
broadband communications have been presented in this paper. IEEE Trans. Commun., vol. 45, pp. 421–428, Apr. 1997.
The impact of typical satellite nonlinearities and techniques to [14] R. De Gaudenzi and M. Luise, “Design and analysis of an all-digital
counteract their impact on the demodulator performance has demodulator for trellis coded 16-QAM transmission over a nonlinear
satellite channel,” IEEE Trans. Commun., vol. 43, no. 2/3/4-Part I,
been analyzed. Also simple yet effective techniques for digital Feb./Mar./Apr., 1995.
demodulator synchronization recovery have been devised and [15] V. Lottici, M. Luise, and R. Reggiannini, “Adaptive nonlinear compen-
their performance analyzed. It is shown that coded APSK sation of satellite transponder nonlinearity for high-level data modula-
tions,” in Proc. 7th Intl. ESA Workshop on Digital Signal Processing
with simple digital pre-distortion techniques based on look- Techniques for Space Commun. 2001.
up tables can achieve very good performance with satellite [16] V. Lottici and M. Luise, “Carrier phase recovery for turbo-coded
HPA driven at saturation (16-APSK) or with limited back-off linear modulations,” in Proc. IEEE Int. Conference on Communications
(ICC’02), vol. 3.
(32-APSK). The simple digital algorithms for AGC, clock and [17] L. Zhang and A. Burr, “Application of turbo principle to carrier phase
phase recovery are shown to cause very small demodulator im- recovery in turbo encoded bit-interleaved coded modulation system,” in
pairment compared to theoretical performance. The proposed Proc. 3th Intl. Symposium on Turbo Codes and Applications 2003.
[18] G. Caire, G. Taricco, and E. Biglieri, “Bit-interleaved coded modula-
turbo coded APSK scheme is shown to provide remarkable tion,” IEEE Trans. Inf. Theory, vol. 44, pp. 927–947, May 1998.
advantages in terms of power saving when used over typical [19] R. De Gaudenzi and A. Guillén i Fàbregas, and A. Martinez, “Turbo-
satellite nonlinear channels. coded APSK modulations design over nonlinear satellite channels,” Int.
J. Satellite Commun, vol. 24, pp. 261-281, 2006.
For the case of carrier phase estimation, we have proposed a [20] F. M. Gardner, “A BPSK/QPSK timing-error detector for sampled
new class of closed-loop non-data-aided (NDA) synchronizers receivers,” IEEE Trans. Commun., vol. 34, no. 5, May 1986.
for APSK modulations and evaluated some of their properties [21] ——, “Interpolation in digital modems–part I: fundamentals,” IEEE
Trans. Commun., vol. 41, no. 3, Mar. 1993.
(S-curves, phase jitter). We show that the new APSK phase [22] L. Erup, F. M. Gardner, and R. Harris, “Interpolation in digital modems–
synchronizers show no phase jitter floor due to pattern noise. part II: implementation and performance,” IEEE Trans. Commun.,
We finally evaluate the overall performance of the proposed vol. 41, no. 6, 1993.
[23] C. Langlais and M. Hélard, “Phase carrier recovery for turbo codes over
turbo-coded APSK demodulator by means of an end-to-end a satellite link with the help of tentative decisions,” in Proc. 2nd Intl.
computer simulator, and we show an improvement of about Symposium on Turbo Codes and Applications 2000, p. 439.
5 dB in power and 8 % in spectral efficiency compared to [24] R. De Gaudenzi, A. Guillén i Fàbregas, A. Martinez, and B. Ponticelli,
“High power and spectral efficiency turbo coded digital modulation
conventional Trellis-Coded 16-QAM. schemes for nonlinear satellite channels,” in Proc. 7th Intl. ESA Work-
shop on Digital Signal Processing Techniques for Space Commun. 2001.
[25] F. Rice, B. Cowley, B. Moran, and M. Rice, “Cramér-rao lower bounds
R EFERENCES for qam phase and frequency estimation,” IEEE Trans. Commun.,
vol. 49, no. 9, Sept. 2001.
[1] “Digital Video broadcasting; second generation framing structure, chan- [26] R. De Gaudenzi, T. Garde, and V. Vanghi, “Performance analysis
nel coding and modulation systems for broadcasting, interactive services, of decision-directed maximum-likelihood phase estimator for M-PSK
news gathering and other broadband satellite applications (DVB-S2),” signals,” IEEE Trans. Commun., vol. 43, no. 12, Dec. 1995.
European Telecommunications Standards Institute Std. EN 302 307, [27] U. Mengali and A. N. D’Andrea, Synchronization Techniques for Digital
2005. Receivers. New York: Plenum Press, 1997.
[2] M. Cominetti and A. Morello, “Digital video broadcasting over satellite [28] S. Pietrobon, R. H. Deng, A. Lafanechere, G. Ungerboeck, and D. J. C.
(DVB-S): a system for broadcasting and contribution applications,” Intl. Jr., “TC multidimensional phase modulation,” IEEE Trans. Inf. Theory,
J. Satellite Commun., no. 18, pp. 393–410, 2000. vol. 36, no. 1, Jan. 1990.
[3] R. Rinaldo and R. D. Gaudenzi, “Capacity analysis and system opti- [29] M. Rouanne and D. J. Costello, “An algorithm for computing the
mization for the forward link of multi-beam satellite broadband systems distance spectrum of trellis codes,” IEEE J. Sel. Areas Commun., vol. 7,
exploiting adaptive coding and modulation,” in Int. J. of Satellite no. 6, Aug. 1989.
Commun. and Network., no. 22, pp. 401–423, 2004. [30] U. Wachsmann, R. F. H. Fischer, and J. B. Huber, “Multilevel codes:
[4] G. Karam and H. Sari, “Analysis of predistortion, equalization and ISI theoretical concepts and practical design rules,” IEEE Trans. Inf. Theory,
cancellation techniques in digital radio systems with nonlinear transmit vol. 45, no. 5, pp. 1361–1391, July 1999.
amplifiers,” IEEE Trans. Commun., vol. 37, no. 12, pp. 1716–1723, Dec.
1989.
[5] ——, “Data predistortion techniques using intersymbol interpolation,”
IEEE Trans. Commun., vol. 38, no. 10, pp. 1716–1723, Oct. 1990.
[6] ——, “A data predistortion technique with memory for qam radio
systems,” IEEE Trans. Commun., vol. 39, no. 2, pp. 336–344, Feb. 1991.
[7] A. N. D’Andrea, V. Lottici, and R. Reggiannini, “RF power amplifier
linearization through amplitude and phase predistortion,” IEEE Trans.
Commun., vol. 44, p. 14771484, Nov. 1996.
[8] C. M. Thomas, M. Y. Weidner, and S. H. Durrani, “Digital amplitude-
phase keying with M-ary alphabets,” IEEE Trans. Commun., vol. 22,
no. 2, pp. 168–180, Feb. 1974.
DE GAUDENZI et al.: PERFORMANCE ANALYSIS OF TURBO-CODED APSK MODULATIONS OVER NONLINEAR SATELLITE CHANNELS 2407
Riccardo De Gaudenzi was born in Italy in Albert Guillén i Fàbregas was born in Barcelona,
1960. He received his Doctor Engineer degree (cum Catalunya, Spain, in 1974. He received the Telecom-
Laude) in electronic engineering from the University munications Engineering Degree and the Electronics
of Pisa, Italy in 1985 and the PhD from the Tech- Engineering Degree from Universitat Politècnica de
nical University of Delft, The Netherlands in 1999. Catalunya, Barcelona, Catalunya, Spain, and the Po-
From 1986 to 1988 he was with the European Space litecnico di Torino, Torino, Italy, respectively, both
Agency (ESA), Stations and Communications Engi- in 1999, under the Double Degree Socrates-Erasmus
neering Department, Darmstadt (Germany) where he Scholarship Program of the European Community,
was involved in satellite telecommunication ground and the Ph.D. in Communication Systems from
systems design and testing. In particular, he followed Ecole Polytechnique Fédérale de Lausanne (EPFL),
the development of two new ESA’s satellite tracking Lausanne, Switzerland, in 2004.
systems. In 1988, he joined ESA’s Research and Technology Centre (ESTEC), From August 1998 to March 1999, he conducted his Final Research Project
Noordwijk, The Netherlands where in 2000 he has been appointed head of the at the Center for Communications and Signal Processing Research (CCSPR),
Communication Systems Section and since 2005 he is Head of the RF Payload at the New Jersey Institute of Technology (NJIT), Newark, NJ, supported with
and Systems Division. The division is responsible for the definition and a Nokia-NJIT Research Fellowship. He was with Telecom Italia Research
development of advanced satellite system, subsystems and technologies for Laboratories, TILAB (old CSELT), Torino, Italy, from November 1999 to
telecommunications, navigation and earth observation applications. In 1996 June 2000. From September 2000 to May 2001, he was with the European
he spent one year with Qualcomm Inc., San Diego USA, in the Globalstar Space Research and Technology Centre (ESTEC), of the European Space
LEO project system group under an ESA fellowship. His current interest is Agency (ESA), Noordwijk, The Netherlands. During his doctoral studies,
mainly related with efficient digital modulation and access techniques for fixed from 2001 to 2004, he has been a Research and Teaching assistant at
and mobile satellite services, synchronization topics, adaptive interference the Mobile Communications Department, Institut Eurècom, Sophia-Antipolis,
mitigation techniques and communication systems simulation techniques. France. From June 2003 to July 2004 he has been a visiting scholar at the
From 2001 to 2005 he has been serving as Associate Editor for CDMA Communications Theory Lab at EPFL. Since September 2004 he is at the
and Synchronization for IEEE Transactions on Communications. He is co- Institute for Telecommunications Research, University of South Australia,
recipient of the VTS Jack Neubauer Best System Paper Award from the IEEE Mawson Lakes, Australia, as a Research Fellow. During June-July 2005 he
Vehicular Technology Society. has held a visiting appointment at Ecole Nationale Superieure des Telecom-
munications, Paris, France. His specific research interests are in the area of
communication theory, information theory, coding theory, digital modulation
and signal processing techniques, particularly with wireless terrestrial and
satellite applications.
Dr. Guillén i Fàbregas received a pre-doctoral Research Fellowship of
the Spanish Ministry of Education to join ESTEC-ESA. He received the
Young Authors Award of the 2004 European Signal Processing Conference
EUSIPCO 2004, Vienna, Austria and the 2004 Nokia Best Doctoral Thesis
Award in Mobile Internet and 3rd Generation Mobile Solutions from the
Spanish Institution of Telecommunications Engineers. He is also a member
of the ARC Communications Research Network (ACoRN).