2.4 IR of CPMs over Channel Ae ted by Phase Noise
2.4.2 SATMODE Phase Noise Model
We now onsidera more realisti PN model, i.e., theSATMODE phasenoise
whi hae tsthesatellite ommuni ationsystemswithlow- ostequipment[22℄.
Itspowerspe traldensityisknownanditis representedinTable.2.1. Evenif
inprin iple thatmask refersto the phasor
e jθ(t)
,it hasbeen assumedthat itapproximately orresponds to the PSD of thephase pro ess
θ(t)
.In order toFrequen y [Hz℄ PSD [dB/Hz℄
10 1
-2310 2
-4810 3
-6810 4
-6810 5
-8010 6
-100Table2.1: SATMODEPSD of the ontinuous-time phasepro ess
θ(t)
.ompute the IR for CPM modulation s hemes over SATMODE hannel and
to derive some mat hed dete tionalgorithms, we needto derive anite-order
statisti al des ription of the PN. Sin e we want to deal with a dis rete-time
pro ess, we sample the phase pro ess with a sampling period
T S = 10 −7
inorderto avoid aliasing;hen eweobtain adis rete-timepro ess withthesame
-100 -90 -80 -70 -60 -50 -40 -30 -20
10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7
PSD [dB/Hz]
f [Hz]
SATMODE PSD mask
Figure 2.2:SATMODEPSD of the ontinuous-time phase pro ess
θ(t)
.PSD maskof the ontinuous-time pro ess.
ByanalyzingtheSATMODEPSD,representedinFig2.2,wenotethatthe
mask anbeseenasthesumoftwodierent slopesand hen ethePN pro ess
an be well approximated as the sum of two 1st order auto-regressive (AR1)
Gaussian pro esses, instead of just one AR1. In detail, one AR1 des ribes
the PN behaviour at low frequen ies (slow pro ess hara terized by higher
power)whiletheotheronerepresentsthePNathigherfrequen ies(fastpro ess
hara terized bylowerpower).
In general aGaussian AR1 pro ess isdened as
u k = a u k−1 + v a,k
(2.25)where
v a,k
are zero-mean independent and identi ally distributed Gaussian random variables, with varian eσ 2 v
and wherea
is a real value su h that|a| < 1
(toensure stability). The varian e ofu k
isthenrepresentedbyσ 2 a , E
|u k | 2
= σ 2 v
1 − a 2
(2.26)and itsPSD is
S u (f ) = σ 2 a (i − a 2 )
1 + a 2 − 2a cos(2πf T S ) , f ∈
− 1 2T S , 1
2T S
.
(2.27)Thus, inour asewe have
θ k , u a,k + u b,k
(2.28)where
u a,k = a u a,k−1 + v a,k u b,k = b u b,k−1 + v b,k .
(2.29)
Hen e, our aim isto nd the four parameters
a, b, σ a 2 , σ 2 b
su h that thetargetPSD of the dis rete-timephase pro ess(Fig2.2) iswell approximated by
S θ (f ) = σ 2 a (i − a 2 )
1 + a 2 − 2a cos(2πf T S ) + σ b 2 (i − b 2 )
1 + b 2 − 2b cos(2πf T S ) ,
(2.30)f ∈
− 1 2T S , 1
2T S
.
InFig2.3wereporttheSATMODEPSD maskalongwiththePSDsresulting
from the double-AR1 approximations with three dierent set of parameters,
reported in Table 2.2. The approximations denoted as UB (upper bound)
and LB (lower bound) represent respe tively an overestimated and an
un-derestimated approximation of the target PSD, while the IA (intermediate
approximation) isaPSD representation fallingbetween thetwoextreme
rep-resentations. Fromnow on,thePN random pro ess will be always generated
a ording to the IA double-AR1 approximations.Moreover, sin e thePSD of
the oversampled dis rete-time phase pro ess we generate by su h parameters
is equal to that of the ontinuous-time pro ess up to
1/T S
, we will denotesu h PN generation method as the ontinuous-time (CT) generation, in
on-trast withthe symbol-time generation (ST) we will des ribe in thefollowing.
a σ 2 a b σ 2 b
LB
1 − 10 −7 1.40 0.980 8 · 10 −3
IA
1 − 10 −7 2.45 0.985 16.2 · 10 −3
UB
1 − 10 −6 1.00 0.987 2 · 10 −2
Table2.2:Parameters oftheoversampled ontinuous-time phasenoise
double-AR1 approximation.
-100 -90 -80 -70 -60 -50 -40 -30 -20
10 1 10 2 10 3 10 4 10 5 10 6
PSD [dB/Hz]
f [Hz]
SATMODE PSD mask Intermediate Approximation Lower Bound
Upper Bound
Figure 2.3: SATMODE PSD mask and PSDs obtained from various
double-AR1 approximations.
-140 -120 -100 -80 -60 -40 -20 0 20
10 -1 10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7
PSD [dB/Hz]
f [Hz]
PSD of slow AR1 component PSD of fast AR1 component double-AR1 PSD
Figure 2.4:PSDs ofslowand fastAR1 obtained withIAparameters.
In order to better understand the ee ts of the two PN omponents, we
re-porttwo gures. Inthe rst one,Fig 2.4, we seethe PSD of ea h singleAR1
omponents evaluated by repla ing in (2.27) the IA parameters of Table 2.2:
ifone AR1 (denoted by
a
) determinestheresulting phase noise PSD at lowerfrequen ies, the other one (denoted by
b
) determines the phase noise PSD athigher frequen ies. Fig 2.5 shows the time-domain snapshot of the two
om-ponentsofthe ontinuous-time phasenoise
θ(t)
,i.e.,theslowoneandthefastone, generated following the double-AR1approximations withIA parameters
of Table2.2. The behaviourof thetwo PN omponentsis lear:thefast
om-ponent is ae ted by very qui k and low power u tuations, while the slow
omponent is hara terized byhighvarian e and slowu tuations. Notethat
in pra ti e it is very di ult to tra k the faster omponent, therefore it is
-3 -2 -1 0 1 2 3
0 1 2 3 4 5 6 7
θ (t)
t [s]
Slow AR1 component Fast AR1 component
Figure2.5:Time snapshotofthetwo phasenoiseAR1 omponents, generated
a ordingto theIA parameters.
themajor reasonofperforman edegradation.Wedis ussinthefollowing,the
ee t of phasenoise on digital ommuni ations. Given the CPMsignal (1.1),
theuseful re eived signalis
r(t) = s(t; x)e jθ(t) .
We donot areaboutthermalnoise sin eitsstatisti sarenotae tedbyPN.
Atthe
k
-thsymbolintervaltheoutput ofthebankof mat hed ltersise −jπ n
Z (n+1)T nT
s(t; x)¯ s ∗ (t − nT ; x n , ω n )e −jθ(t) dt
(2.31)wheretheCPMwaveform
¯ s ∗ (t −nT ; x n , ω n )
andthephasestateπ n
aredenedin(1.17) and(1.5)respe tively.Duetothepresen eofthePN
θ(t)
,theoutputof the mat hed lter
s ¯ ∗ (−t; x n , ω n )
is no more a su ient statisti but it isa orrupted version of the su ient statisti
y n (x n , σ n )
dened in (2.12). Inother words, the phase noise auses a sort of inter-symbol interferen e (ISI)
whi h,inprin iple, anhave anextremely harmfulee t whena re eiverthat
negle ts itisemployed.
However,we veriedthatthisisnotthe aseinall s enariosofinterest(as
shown later). Hen e, we negle t its ee t and we also provide a dierent PN
generation,denotedassymbol-time(ST)generation,bywhi hjustonesample
isgenerated atea hsymbolinterval,withouttakingintoa ountthemat hed
ltering distortionpresentsinthe ontinuous-time (CT)generation. Thenthe
mat hed ltersoutput(2.31) an beapproximated as
e −jπ n
Z (n+1)T nT
s(t; x)¯ s ∗ (t − nT ; x n , ω n )e jθ(t) dt ≃ y n (x n , σ n )e jψ n
(2.32)where
ψ n = 1 T
Z (n+1)T nT
θ(t)dt.
(2.33)Thus the PN pro essat symboltimeisapproximately awindowed and
down-sampled version of the original PN pro ess. In order to obtain a statisti al
hara terization of
ψ n
asdone for the ontinuous-time pro essθ(t)
,we derivetheauto orrelation fun tion(ACF)of
ψ n
R ψ (m) = E {ψ n ψ n−m } .
(2.34)In Fig 2.6 the auto orrelation fun tion (2.34) for a signal rate of
256
kBaudis arried out by numeri al simulation: we have generated various
θ(t)
se-quen es employing IA parameters, derived the orresponding
ψ n
sequen esfollowing (2.33) when the symbol rate is
R = 1/T = 256
kBaud and thenomputed
R ψ (m)
byaveraging overall simulatedψ n
sequen es.WenotefromFig 2.6 that the symbol-time phase noise pro ess
ψ n
an again beapproxi-mated as the sum of two independent AR1 pro esses, a slowly varying (i.e.,
highly orrelated) omponent anda rapidly varyingand low varian e
ompo-nent.Thesetofparameters
a, b, σ 2 a , σ b 2
bywhi hwederivedtheACFinFig2.62.134 2.136 2.138 2.14 2.142 2.144 2.146 2.148 2.15 2.152
0 10 20 30 40 50 60
R φ (m)
m
ACF at 256 kBaud
ACF of the double AR1 approximation
Figure 2.6: Auto orrelation fun tion of the ontinuous-time PN pro ess
θ(t)
and ofthe symbol-timePN pro ess
ψ n
at256
kBaud.are listed in Table 2.3, together with the same parameters obtained for a set
of sele ted baudrates. As mentioned, themajor reason of performan e
degra-Baudrate [KBaud℄
a σ 2 a b σ 2 b
64 1 − 1.76 · 10 −5 2.163 0.263 8.51 · 10 −3 128 1 − 8.9 · 10 −6 2.143 0.447 1.150 · 10 −2 256 1 − 4.45 · 10 −6 2.136 0.604 1.399 · 10 −2 512 1 − 2.15 · 10 −6 2.140 0.758 1.558 · 10 −2 1024 1 − 1.05 · 10 −6 2.126 0.871 1.611 · 10 −2 2048 1 − 4.7 · 10 −7 2.144 0.942 1.609 · 10 −2
Table2.3: Parameters ofthesymbol-time phasenoise double-AR1
approxima-tion, provided for dierent baudrate values.
dation is the faster omponent, whi h is very di ult to tra k. For example,
from Fig.2.6 itis lear that su ha omponent at
256
KBaud has astandarddeviation of about
4
degrees and the orresponding samples at symbol-time arealmost independent.Taking advantage of the double-AR1 PN des ription, it is interesting to
investigate the IR of a hannel ae ted by SATMODE phase noise, in order
to quantify the mutual information rate loss due to su h a PN. In parti ular
we evaluate IR a ording to the Arnold and Loeliger method, for dierent
hannel representations and dierent dete tion algorithms (ne essary for the
forward metri omputation). In parti ular, we onsider two possible hannel
generation models:
the real hannel, whi h works in a pra ti ally ontinuous-time setting
(i.e.,samplingtime
10 −7
s),byaddinganoversampledversionofthePN pro essθ(t)
generated a ording to theCT double-AR1approximation, with IA set of parameters; insu h a ase an ISI ee t on the su ientstatisti smayo urs;
the auxiliary hannel, whi h introdu es a dis rete-time PN pro ess
ψ n
onstant over ea hsymbolperiod and following the STdouble-AR1
ap-proximation.
The dete tion algorithms we propose are all based on MAP symbol
dete -tion, but ingeneral they annotbe mat hed to the hannelgenerationmodel.
Moreover they areall derived following a Bayesian approa h, i.e., we assume
a proper probabilisti model for the phase noise
{θ n }
, and we exploit it forderiving algorithmsfor MAP symboldete tion. Indetailwe onsider:
DP-BCJR,alreadyanti ipatedinSe tion2.4.1anddetailedinSe tion2.4.3;
inthe algorithm derivation we onsider a rst order auto-regressive PN
modelor a WienerPN model.
Double-DP-BCJR, whi h is a generalization (provided inSe tion 2.4.3)
ofthe DP-BCJR;inthealgorithmderivationwe onsidertworst order
auto-regressive omponentsfor the PN modeland not justone;
Improved-DP-BCJR (Se tion 2.4.3), whi h assumes both the two AR1
omponents, but onsiders all the sample of one omponent (the faster
one)independent to ea h others.
We notethatwhenthereal hannelgenerated bytheCTdouble-AR1appro
x-imationis onsidered,thereisnoonedete torthat anbe onsideredmat hed
to the hannel sin e all algorithms are based on a phase dis retization
te h-nique (even if the number of levels is very large) and all algorithms onsider
justone PN samplefor ea h symbolinterval (and not a ontinuous-time
gen-eration). Howeverit isinteresting toevaluate theIRfor thereal hannelwith
theDouble-DP-BCJR inorder to quantify theee t of theISIdistortion. On
theotherhand,when we take into a ount theST hannelgeneration, ex ept
for the phase dis retization (whi h auses justa little mismat h between the
dete tor and the hannel model), the double-DP-BCJR is the optimum one
in the MAP sense. Moreover, it is interesting also to evaluate the IR by a
DP-BCJR operatingovera CT orST hannel generation, sin e we derive the
maximumperforman ewe ana hieve,whenemploying adete tionalgorithm
arried out by starting from a single rst order PN model (Wiener or AR1),
overa hannel ae tedby SATMODEPN.Finally,thethird typeof dete tor
is parti ularly suitable to ompute the IR for double-AR1 PN generation at
low baudrate values. Looking to the double-AR1 parameters in Table 2.3 at
64
kBaud, for example,we see that thefast omponents exhibits ab
valuesolose tozerothatinthedete torderivation, we an onsiderits orresponding
PN samples as almost independent. In su h a way, in the dete tion
deriva-tion, we take into a ount both the two AR1 omponent but we also a hieve
a redu tioninthedete tor omplexity.