Detector characteristic for decision-directed

carrier phase recovery of 16=32-APSK

signals
W. Gappmair and J. Holzleitner
Usually implemented as second-order loops, decision-directed (DD)
feedback devices are frequently applied for carrier phase control.
Using 16-ary and 32-ary amplitude phase-shift keying (APSK), recom-
mended as modulation schemes in the new digital video broadcasting
standard for satellite communications (DVB-S2), it is shown that the DD
detector characteristic (S-curve) can be derived in closed form. In the
sequel, the slope of the S-curve at the stable equilibriumpoint, as it would
be required to design the linearised recovery loop, is computed.
Introduction: In the new digital video broadcasting (DVB) standard for
satellite communications [1], frequently denoted by DVB-S2, 16-ary and
32-ary amplitude phase-shift keying (APSK) are specied as modulation
techniques. As for any other linear scheme, decision-directed (DD)
tracking devices might be implemented in order to follow the carrier
phase [2], usually established as second-order loops. During normal
operation, the deviations from the stable equilibrium point are sufciently
small such that the loop can be described appropriately by a linearised
model. In this context, the knowledge of the detector characteristic
(S-curve) is of particular interest, which will be subsequently developed
in closed form for 16=32-APSK signals.
Equivalent baseband model: Throughout this Letter, it is assumed
that perfect symbol timing has been achieved. Furthermore, let the
independent and identically distributed 16=32-APSK symbols c
k
be
normalised to average unit energy E[jc
k
j
2
] 1, i.e. with c
k
s arranged
on two=three rings, as shown in Fig. 1; the radii R
1
, R
2
and R
3
are
established as soon as the ratios b
1
: R
2
=R
1
and b
2
: R
3
=R
1
are
given. In the sequel, the receiver samples at the output of the matched
lter are provided by
r
k
e
jy
k
c
k
n
k
1
where y
k
denotes the carrier phase. Real and imaginary parts of the
zero-mean AWGN samples n
k
are assumed to be independent, each
with the same variance of 1=2g
s
, where g
s
: E
s
=N
0
is the mean
signal-to-noise ratio (SNR) per symbol.
Fig. 1 Symbol constellation for 16=32-APSK schemes
The recovery scheme to be investigated in detail is shown in Fig. 2.
As already mentioned in the Introduction, it is usually designed as a
second-order loop, specied through noise bandwidth and damping
factor. When c
k
denotes the decision on c
k
and y

k
the estimate of y
k
, the
DD detector delivers an error signal
u
k
Im^ c
*
k
r
k
e
j
^
y
k
2
Fig. 2 Feedback recovery of carrier phase with DD detectors
Detector characteristic: The detector characteristic or S-curve is
dened as the expected value of u
k
for the open loop, i.e. S(g
s
,
y): E[u
k
jy
k
y, y

k
0]. With c
k
: a
k
jb
k
and the quadrature
symmetry of 16=32-APSK constellations taken into account, it can
be shown [3] that l
c
(g
s
, y): E[a
k
a
k
] E[b
k
b

k
] and l
s
(g
s
, y):
E[b
k
a
k
] E[a
k
b

k
]. Therefore, after some lengthy but straightfor-
ward manipulations, the S-curve develops as
Sg
s
; y 2g
c
g
s
; y sin y l
s
g
s
; y cos y 3
What remains is the evaluation of the symbol statistics l
c
() and l
s
().
To this end, it is assumed in the following without loss of generality that
the APSK constellation is rotated by p=4, which simplies the analy-
tical work considerably.
First, 16-APSK is to be investigated in detail. By inspection of Fig. 1,
it is clear that this consists of a 4-PSK and a 12-PSK ensemble,
separated by R
12
. Motivated by the fact that, for M-PSK schemes, the
S-curve is available in closed form [4], the developed relationships can
be adjusted appropriately. As a result, the partial symbol statistics of
l
c
(), i.e. l
c
() conditioned on both jc
k
j R
m
and jc
k
j R
n
, where m,
n 2{1, 2}, appears as
l
c;mn
g
s
; y
2R
m
R
n
M

M
m
=21
i0

M
n
1
l0
cos
2pi
M
m
_ _
cos
2pl
M
n
_ _
P
il;mn
g
s
; y 4
where M16, M
1
4, M
2
12 and
P
il;mn
g
s
; y
_
j
il;mn;2
y
j
il;mn;1
y
p
F
R
m
; R
12
; g
s
; j dj 5
Upper and lower limits are given by
j
il;mn;1
y l
2p
M
n

p
M
n
i
2p
M
n
y
_ _
p
p
j
il;mn;2
y l
2p
M
n

p
M
n
i
2p
M
n
y
_ _
p
p
6
where [j]
p
p
means that j has to be wrapped around p. Using the
joint probability function for complex Gaussian noise, suitably estab-
lished in polar co-ordinates [5, eqns. 4.2.100102] and denoted by
p
R,F
(S, g
s
, r, j), the angular density in (5) is, after some algebra applied
to [6, eqns. 3.322=12], immediately provided as
p
F
s; R; g
s
; j
_
R
0
p
R;F
s; g
s
; r; j dr
e
s
2
g
s
2p
1 e
g
s
R
2
2sRcos j
_

ps
2
g
s
_
cos je
s
2
g
s
cos
2
j
erfc

g
s
p
s cos j R
erfc

s
2
g
s
_
cos j
_
7
where erfc(x) :(2=
p
p)
_
x
1
e
t
2
dt symbolises the complementary error
function. Note that (7) is only valid for n 1, whereas for n 2 it has to
be replaced by
p
G
s; R; g
s
; j
_
1
R
p
R;F
s; g
s
; r; j dr
e
s
2
g
s
2p
_
e
g
s
R
2
2sRcos j

ps
2
g
s
_
cos je
s
2
g
s
cos
2
j
erfc

g
s
p
R s cos j
_
8
Following the procedure in [4], the components of l
s
() are simply
expressed through
ELECTRONICS LETTERS 7th December 2006 Vol. 42 No. 25
l
s;mn
g
s
; y
2R
m
R
n
M

M
m
=41
iM
m
=4

M
m
1
l0
sin
2pi
M
m
_ _
cos
2pl
M
n
_ _
P
il;mn
g
s
; y 9
Finally, the symbol statistics develop as
l
c
g
s
; y

m;n
l
c;mn
g
s
; yl
s
g
s
; y

m;n
l
s;mn
g
s
; y 10
Although not shown due to limited space, (3) has been veried by
extensive simulation work, conrming also that the detector is unbiased,
i.e. S(g
s
, 0) 0 irrespective of the selected value of g
s
. For the linearised
tracker model, however, knowledge of the full S-curve is of less
interestin contrast to the slope in the stable equilibrium point, dened
as the rst derivative of (3) with respect to y evaluated at y 0, i.e.
K
d
:
@
@y
Syj
y0
2l
c
g
s
; 0 l
0
s
g
s
; 0 11
where l
s
0
(g
s
, 0) :(@=@y)l
s
(g
s
, y)j
y0
. It is easily checked that the
probabilities P
il,mn
(g
s
, y) are the only terms in l
s
( ) that are a function
of y. Hence, the rst derivative of (5) with respect to y appears as
P
0
il;mn
g
s
; y
p
F
R
m
; R
12
; g
s
; j
il;mn;1
y p
F
R
m
; R
12
; g
s
; j
mn;2
y;
n 1
p
G
R
m
; R
12
; g
s
; j
il;mn;1
y p
G
R
m
; R
12
; g
s
; j
il;mn;2
y;
n 2
_

_
12
For b
1
2.5 and R
12
(R
1
R
2
)=2, Fig. 3 exemplies the evolution of
(11) as a function of g
s
E
s
=N
0
. As expected, K
d
!0 for g
s
! 0,
whereas K
d
!1 for g
s
!1. However, the computational complexity is
considerable. Fortunately, at larger SNR values, it turns out that the
contribution of l
c,mn
() and l
s,mn
(), m6n, is negligible compared to the
case with mn. In addition, (7) and (8) can be approximated by
p
(p
1
s
2
g
s
)e
s
2
gs j
2
such that (5) reduces to
P
il;mn
l
s
; y
1
2
erfc

R
2
m
g
s
_
j
il;mn;1
_ _
erfc

R
2
m
g
s
_
j
il;mn;2
_ _ _ _
13
Computing (11) with l
c,mn
() and l
s,mn
(), m6n, omitted as well as
(5) replaced by (13), the result is shown in Fig. 3. As can be seen, the
simplied analysis approaches the exact solution as soon as g
s
>
12.5 dB.
Fig. 3 Evolution of detector slope K
d
for 16=32-APSK (b
1
2.5, b
2
4.3)
For 32-APSK schemes, S-curves and slopes are achievable in the
same manner as just demonstrated with 16-APSK. Of course, M32
and m, n 2{1, 2, 3}; cardinality index M
3
16 and radius R
3
char-
acterise the third ring. Again, p
F
(R
m
, R
12
, g
s
, j) is used in (5) if n 1,
while replaced by p
F
(R
m
, R
23
, g
s
, j) p
F
(R
m
, R
12
, g
s
, j) if n 2 and
p
G
(R
m
, R
23
, g
s
, j) if n 3. For b
1
2.5, b
2
4.3, R
12
(R
1
R
2
)=2
and R
23
(R
2
R
3
)=2, Fig. 3 illustrates the evolution of the detector
slope. Applying the simplications introduced for 16-APSK, the exact
results are suitably approximated for values of g
s
>15 dB.
Conclusions: Using 16=32-APSK as modulation schemes, the S-curve
for DDrecovery of the carrier phase has been developed in closed form.
Since it is rather complex from the computational point of view,
appropriate simplications are introduced such that the exact solution
is conveniently approximated for larger SNR values. The availability of
the S-curve allows the derivation of the slope in the stable equilibrium
point, which is required to design the linearised recovery loop.
# The Institution of Engineering and Technology 2006
9 August 2006
Electronics Letters online no: 20062489
doi: 10.1049/el:20062489
W. Gappmair and J. Holzleitner (Institute of Communication Networks
and Satellite Communications, Graz University of Technology,
Austria)
E-mail: [email protected]
References
1 ETSI EN 302 307 (V1.1.1): Digital Video Broadcasting (DVB); Second
generation framing structure, channel coding and modulation systems for
Broadcasting, Interactive Services, News Gathering and other broadband
satellite applications http://www.etsi.org, 2004
2 Mengali, U., and DAndrea, A.N.: Synchronization techniques in digital
receivers (Plenum Press, New York, USA, 1997)
3 De Gaudenzi, R., Garde, T., and Vanghi, V.: Performance analysis of
decision-directed maximum-likelihood phase estimators for M-PSK
modulated signals, IEEE Trans. Commun., 1995, 43, pp. 30903100
4 Gappmair, W.: Open-loop characteristic of decision-directed maximum-
likelihood phase estimators for MPSK modulated signals, Electron.
Lett., 2003, 39, pp. 337339
5 Proakis, J.G.: Digital communications (McGraw-Hill, New York, USA,
1989)
6 Gradshteyn, I.S., and Ryzhik, I.M.: Table of integrals, series, and
products (Academic Press, London, UK, 1994)
ELECTRONICS LETTERS 7th December 2006 Vol. 42 No. 25