1.3 General Frameworks
1.3.3 Iterative Joint Dete tion/De oding S hemes
Whenwe onsider a ommuni ationsystem hara terized byan error
orre t-ing ode together with a hannel with memory, the ardinality of the
over-all system state an be very large, and thus, optimal MAP symbol dete tion
strategy at the re eiverbe omesunfeasible.In these ases,we an resort to a
suboptimaliterativejointdete tion/de odings heme,whi hexhibits
omputa-tional omplexityabsolutelylowerthanthe omplexityoftheoptimals heme,
but a performan e whi h tends to the optimal one (as veried by numeri al
results)[9℄.Inparti ular,herewedes ribetheoperationsofaserially
on ate-nateds heme, whi h is thes heme adoptedinChapters 3 and4 for dete tion
of CPMsignalinthe presen eof anouter error orre ting ode.
In an iterative on atenated joint dete tion/de oding s heme, ea h
om-ponent blo k (i.e.,the dete tor and the de oder) works separately, by
imple-mentingthe MAPsymbolstrategyoptimal forthesingleblo k,assumingthat
no other memory sour esarepresent inthe system.They employ a dete tion
(de oding)algorithmbasedontheMAPsymbolrule,whi hprovidesreliability
estimatesonthealgorithmde isions (Se tion1.3.1), sin eweneed soft-output
de isions in orderto on atenatethe two blo ks.In general, an iterative
on- atenated s heme is based on the following basi on ept: ea h omponent
blo k exploits the suggestion provided by the other omponent blo k, in
or-der to derive de isions whi h be omes more reliable with the iteration. In
detail, a serially on atenated s heme worksas follow. First of all, the
dete -torperformsan instan eof thedete tionalgorithm, operatingon the hannel
su ient statisti s
y
.Then, thesoft-de isionsitprodu es onea h symbolx n
,are forwarded to the de oder, whi h employs the dete tor a posteriori
prob-abilities, as a priori probabilities on
{x n }
while performing de oding. Thus,it also produ es soft-de isions on the
x n
symbols, whi h are in turn passedto the dete tor. The dete tor exploits su h reliability estimations as a priori
probabilities on
{x n }
and itstartsanewiterationof theserially on atenated s heme. Thejointdete tion/de odingstrategy ontinuesforaxednumberofy les;then, hard nalde isions onthe symbols
{x n }
arederived.In order to a elerate the onvergen e of theiterative dete tion/de oding
pro ess, ea h omponent blo k must re eive as input an information whi h
is not self-produ ed. With this aim,in [10,11℄ the on ept of extrinsi
infor-mation is introdu ed, whi h identies the reliabilityinformation produ ed by
a omponent blo k whi h does not depend on the information it re eived as
input. Indetail, ifwe denote by
ℓ a,n
out theLLRsdened in(1.28), produ ed bya blo k and representing the reliability measure of a MAP symbol algorithm
onthede isiononthesymbol
x n
,theextrinsi informationℓ a,n
e,out generated bysu h blo k,is given by
ℓ a,n
e,out= ℓ a,n
out− ℓ a,n
e,in.
(1.32)The FG/SPA toolintrinsi ally propagates extrinsi information, asdes ribed
by(1.30)-(1.31).
Capa ity Evaluation for CPM
signals
2.1 Introdu tion
Continuousphase modulations (CPMs)form a lass of onstant envelope
sig-naling formatsthataree ient inpowerand bandwidth[1℄.Inorder to
om-pare CPMs against linear modulations, e.g., phase-shift keying (PSK) and
quadrature-amplitude modulations(QAMs),aswellastooptimizethevarious
parameters whi h des ribe a CPM signal, namely phase response, alphabet
size,andmodulationindex,information theoreti arguments ouldbeused.In
parti ular, the availabilityof expressions for the hannel apa ity,interms of
bits per hannel use, and bandwidth o upan y of thesignal, would allow to
evaluate theamountof b/s arriedperbandunit (Hz),whi h isanimportant
omparison riterion for system designer. Unfortunately, neither losed-form
results for the apa ity of a CPM signal nor simple expressions for its
band-widthexist.
In[12℄and[13℄ananalyti alstudyoftheinformationtransferoveradditive
white Gaussian noise (AWGN) hannels for envelope onstrained waveforms
under a xed bandwidth denition is des ribed; however the results do not
seem beimmediately appli ableto theCPMs.Firstresultson CPMs apa ity
arereportedin[14℄,wheretheyarepresentedtojustifythe hooseofa ertain
outer odeinaserially- on atenated CPM(SCCPM)s heme. Theauthors do
not des ribe the method by whi h CPM apa ity is omputed and moreover,
the results they arry out do not mat h with results we will obtain in the
present work.In[15℄ multi- hannel apa ity resultsareapplied to the
de om-position [16℄ of CPM signal; unfortunately that method provides bounds too
looseon CPM apa ityto be onsidered of pra ti alinterest.
Finallyin[17,18℄and[19℄CPM apa ityis omputed byusingthemethod
by Arnold and Loeliger in 2001 [20℄ and in2006 [21℄. By this te hnique, the
information rate (IR) over a hannel with memory an be omputed by the
forward metri of a BCJR algorithm (explained in Se tion 1.3.1) applied at
theoutputofsu ha hannel. Thisworkwaspresentedinthe ontext oflinear
modulationsover hannels withmemory,but anbetransposedtoall hannels
withmemory.Hen e,representingCPMsignalsasasuitablenite-state,
nite-dimensional Markov pro ess (as done in Se tion 1.2), it is possible to derive
theinformation rateof the hannel omposedbythe on atenationof aCPM
modulator andan AWGN hannel. Moreover,in[19℄ theinformation rateof a
CPMsignalis omputedinthepresen eofaphasenoise(PN)pro ess,modeled
as a dis rete time Wiener random pro ess, whi h is a ommonly used model
intele ommuni ation s hemes.
The remainder of this hapteris organized as follows. First ofall, we
pro-vide a more exhaustive des ription of thealgorithm proposed byArnold and
Loeliger and we exploit it to derive the IR apa ity of a CPM signalover an
AWGN hannelandoveranAWGN hannelae tedbyphasenoise,modeledas
adis retetimeWienerpro ess.Then,weresorttoaphasenoise(PN)modelof
pra ti al interest, i.e.,the SATMODEphasenoisemodel,whi himpairs
satel-lite ommuni ations hara terized bylow- ost transmitter and re eiver.
SAT-MODEPN isamodelusually employedto testtheperforman eofDVB-RCS
(Digital Video Broad asting-Return Channel Satellite) systems.In parti ular,
inspe ting the phasenoise powerspe traldensity (PSD) [22℄,we see thatthe
dis rete-time phase pro ess an be well approximated by the sum of two 1st
orderauto-regressive (AR1)pro esses,andwe derive CPM'sIRinsu ha
s e-nario.Finally,we proposea novel iterative method,similarto thegeneralized
Blahut-Arimoto algorithm re ently proposed by Kav£i [23℄, to evaluate the
Markov apa ityofa CPMsignalover anAWGN hannel. Oneof thenovelty
of our approa h is that we maximize, with respe t to the input distribution,
thespe trale ien yinbps/Hzratherthanthemutualinformationinbitsper
hannel use. We address this problem by taking into a ount the bandwidth
o upan y of the CPM signal by means of the (properly modied) Carson's
rule bandwidth denition, and solving a linearly onstrained non-linear
opti-mization problem. The results show that Markov apa ity obtained with the
proposed input optimization algorithm strongly outperforms the apa ity for
independent and uniformly distributed inputs. In the numeri al results, we
also onsidera properbandwidth denitionrelatedtothespe tralpower
on- entration(SPC),although inthis asetheoptimization an beperformedby
resorting to abrute-for e approa h only.