real-isti ommuni ation systems.
Se ondly, we show that all the multi arrier modulation formats an be
derived from su h adis rete-time model, with ageneral prototype lter
(rather than with the standard re tangular lter) and a general time
and frequen y spa ing between the information symbols; inotherword,
we apply pulseshaping te hnique to all the s hemes, so extending the
DCT-OFDM,DST-OFDM andDTT-OFDMs hemesalreadypresentin
literature.
Finally,inspiredbytheWilsonbase[87,Se tion4.2℄,thatturnsouttobe
a leverwayto design well-lo alized orthogonalGaborframeswithhigh
time-frequen ye ien ythatavoidthelimitingfa torduetothe
Balian-Lowtheorem[87℄,[88℄,wederiveapra ti almulti arriermodulationwith
ex ellent performan e on doubly-sele tive hannels.
For all the above mentioned modulation s hemes, we rst assume a
re t-angular prototype pulse, as is usually done in the literature [74,81℄, then we
designasuitableprototypepulse,welllo alizedinboththetimeandfrequen y
domains, by meansof thete hnique proposedin[89℄ or in[90℄.
w(t)
S/P
y −Q m
y Q m
kT
MODULAT OR CHANNEL DEMODULAT OR
x(t) r(t)
y −Q+1 m x i
u e −Q+1 (t)
h(t, τ ) u e −Q (t)
u e Q (t)
v e ∗ −Q (−t)
v e Q ∗ (−t) v e ∗ −Q+1 (−t)
Figure5.1: Ideal ontinuous-time lter-bank system.
transmitted symbols
{x i }
arezero-mean independent random variables (r.v.s) belonging to a given omplex onstellation. We deneσ x 2 , E
|x n | 2
. The
transmitted signal, whi h is obtained bymodulating a set of ontinuous-time
lters
{e u n (t)}
,readsx(t) = X
m∈Z
X Q n=−Q
x m n u e n (t − mT ).
(5.1)Filters
{e u n (t)}
willbe hosenequal toabasepulse (eitherare tangular pulse oranappropriatelydesignedpulse,a ordingtotheapproa hdes ribedin[85℄)withasuitablemultiplexing inthefrequen ydomain,asintheuniform
lter-bank approa h [91, Chapter 9℄. Hen e, index
n
is usually referred to as thefrequen y (or sub arrier) index. We will dene the spe tral e ien y
η
,ex-pressed insymb/s/Hz, asthe amount of ode symbols that an be loaded on
atime-frequen y region hara terized byaunitary bandwidthandtimewidth.
All the onsidered modulation formats will be designed so that they a hieve
thesame spe trale ien y,inorder to arry out afair omparison.
Aftertransmissionoveranoisydoubly-sele tive hannel,there eivedsignal
r(t)
isr(t) = Z +∞
−∞ h(t, τ ) x(t − τ) dτ + w(t)
(5.2)wherethezero-meanGaussianrandompro ess
h(t, τ )
denotesthetime-varying hannelimpulseresponse(CIR).Awide-sensestationaryun orrelateds atter-ing(WSSUS) hannel is assumed,and
h(t, τ ) = 0
ifτ < 0
orτ > τ max
,beingτ max
the maximalex ess delay ofthe hannel.τ max
givesan indi ation aboutthe hannel spreading in the time domain. A similar parameter, denoted as
ν max
(maximal Doppler spread), quanties the hannel spreading in thefre-quen y domain [92℄.
w(t)
is a omplex white Gaussian pro ess with powerspe traldensity
2N 0
.Atthe re eiver,thesignalispro essedbyabankof
2Q + 1
lters.Ther
-thsampleat the output ofthe
k
-th lter{ev k ∗ (−t)}
isy k r , y k (rT ) =
Z +∞
−∞ r(τ ) ev k ∗ (τ − rT ) , k = −Q, . . . , Q .
(5.3)In order to obtain a pra ti al implementation modelfor the mentioned
lter-bank s heme, we resort to a dis rete-time model. In parti ular, ea h impulse
responselter
e u n (t)
(and similarlyfore v k (t)
) isapproximately representedbye
u n (t) ≃ X
i∈Z
e u n (i) ⊓
t T s − i
(5.4)
where
e u n (i) , e u n (iT s )
,T s , T/N
(N
isthe oversamplingfa tor)and⊓(t)
isare tangular pulse dened as
⊓ (α) ,
( 1
if0 ≤ α < 1 0
otherwise.
(5.5)
The dis rete-time representations of the lters, a ording to (5.4) , is
su- iently a urate provided that a large enough oversampling fa tor
N
isem-ployed. Under this assumption, an equivalent dis rete-time representation of
thetrans eiverisshowninFig.5.2, where thetwo squarelters
⊓ (·)
playtherole ofdigital-to-analog and analog-to-digital onverters, respe tively.
S/P
y −Q m y −Q+1 m MODULAT OR CHANNEL DEMODULAT OR
h(t, τ ) x(t)
y Q m
r(t) u e −Q (i)
lT s
v e Q ∗ (i)
⊓ − t T s
x i
w(t) u e −Q+1 (i)
u e Q (i)
v e ∗ −Q+1 (i) v e ∗ −Q (i)
⊓ T t s
Figure5.2: Pra ti al oversampled dis rete-time lter-bank system.
The dis rete-time representation of thetransmittedsignal(5.1) is
x(t) = r 1
T s X
m∈Z
X Q n=−Q
x m n X
i
e u n (i) ⊓
t − mT T s − i
(5.6)
where the onstant
p 1/T s
is an energy normalization term. Thedis rete-time re eived samples an be obtained byusing (5.6) and (5.2) in (5.3),thus
obtaining
y k r = r 1
T s X
l∈Z
e v ∗ k (l)
Z ∞
−∞ r(t) ⊓
t − rT T s − l
dt
= X
m∈Z
X Q n=−Q
H k,n r,m x m n + n r k
withk = −Q, . . . , Q r ∈ N
(5.7)
where
H k,n r,m , X
l∈Z
e
v ∗ k (l − rN) X
i∈Z
e
u n (i − mN) h l,l−i
(5.8)with
h l,i , 1 T s
Z Z ∞
−∞
h(t + lT s , τ ) ⊓
t − τ T s + i
⊓
t T s
dτ dt
(5.9)and
n r k , r 1
T s
X
l∈Z
v e k ∗ (l − rN) Z ∞
−∞ w(t) ⊓
t T s − l
dt.
(5.10)CHANNEL
MODULATOR DEMODULATOR
x m n s(i) p(l) y k r
Figure 5.3: System model de omposed into three blo ks: the modulator, the
hanneland the demodulator.
From (5.9),thedis rete-time CIR
h l,i
hassupportover0 ≤ i ≤ L
,whereL = ⌊τ max /T s ⌋ + 1
and
⌊t⌋
meansroundt
down to the nearestinteger.Finally,weprovideanalternative systemdes ription,aimingatthesystem
separation into three dierent blo ks: the transmitter, the hannel, and the
re eiver. Repla ing (5.8) in (5.7),with a few mathemati al manipulations we
obtain the following expressionfor thestatisti s
y r k y k r = X
l∈Z
e v k ∗ (l − rN)
X
i∈Z
X
m∈Z
X Q n=−Q
x m n u e n (i − mN)
h l,l−i
+ n r k .
(5.11)Hen e we seethat our systemis omposed of thefollowing three stages,
rep-resentedinFig 5.3.
Modulator : omputes thedis rete-time signal
s(i) s(i) , X
m∈Z
X Q n=−Q
x m n u e n (i − mN)
(5.12)whi h is the sum of the output of a bank of lters whose input are
the information symbols
{x m n }
.The signals(i)
isthen forwarded to thehannel.
Channel : omputes the output dis rete-time signal
p(l)
by the onvolution ofthe inputsignals(i)
andthe hannel oe ientsh l,i
p(l) , X
i∈Z
s(i)h l,l−i .
(5.13)From(5.9),the hannel oe ient
h l,i
isgivenbythe ontributionofthe time-varying hannel impulse response and the two re tangular lters(thedigital-to-analog and analog-to-digital onverters).
Demodulator : derives the su ient statisti s
y r k
, by ltering the re eivedsignal
p(l)
bythe bank oflter{v k (l)}
y k r = X
l∈Z
e
v k ∗ (l − rN)p(l) + n r k .
(5.14)Obviously the thermal noise is generated by the hannel, but it is
on-venient for our purposetosee su hnoise asaGaussian randomvariable
n r k
introdu ed bythedemodulator.Theabovegeneral modelisvalidfor allmulti arrier s hemes,whi harebased
howeveron dierentmodulatorand demodulator stages.Hen e, itwillbe
suf- ienttoprovideforthevariouss hemes,thedierentexpressionforthesignal
s(i)
in(5.12) and for thestatistiy r k
in(5.14).5.2.1 Uniform Filter-Bank
As,alreadyanti ipated inSe tion 5.2,all multi arrier modulations hemeswe
will onsider, derivefromtheuniformlter-banks system(see[91℄, hapter9):
frequen yresponsesoflters
{e u n (i)}
andlters{ev n (i)}
areobtainedbyshiftingthefrequen y responseof thesame real lter
u(t)
, denotedasprototype lter,bya spa ing given by
F n = nF
,whereF
is the spa ing between the varioussub arriers.
Sin e
T
is the OFDM symbol period andF
the arrier separation, ea h oded symbolsx m n
an be asso iated to the point(T m , F n ) = (mT, nF )
of abidimensional grid inthe time-frequen y plane [85℄.Hen e, dening
ρ , F T
,theinverseof
ρ
anbeseenasameasureofthespe tral e ien yη
(intermsofdatasymbolsperse ondsperHertz),sin ehigher
ρ −1
valuesleadtoaredu edspa e-frequen y distan e between symbols (i.e., a redu ed distan e between
two adja ent gridpoints).