baseband transmission in noise - skkuclass.icc.skku.ac.kr/~dikim/teaching/3012/notes/eee3012p... ·...
TRANSCRIPT
Baseband Transmission
in Noise (Proakis: chapter 8.3, 10.5)
EEE3012 – Spring 2020
Di i l C i i SDigital Communication System
i SBasic Setup
S[ ]x k
XT( )s t ( )y t ( )v t
mV( )H f
thresh.[ ]x̂ k
X
( )n t ( ) NbT
( )n t ( )2
on
NS f =
the LP filter suppresses noise, but (ideally) not signal
( )
/ 2N
( )S f ( )H fis this the optimum
filter?/ 2oN
( )S f ( )H f
cont … / 2oN
Not necessarily! the optimum filter should minimize the error rate
it may pay to distort the signal if more noise can be rejected
i di h O i ilFinding the Optimum Filter
Focus on the “no ISI” case Signals are sent every Tb seconds Signals are sent every Tb seconds s1(t) is sent for a “0”
(t) i t f “1” s2(t) is sent for a “1” each pulse lasts for Tb seconds
Example: FSK( ) ù( )( ) ( )
1
2
cos0
cosc
bc
s t A tt T
s t A t
w
w w
ù= ú £ £ú= +D úûû
[ ]x k ( )s t ( )y t ( )v t V( )H f
[ ]x̂ k
cont … S[ ]
XT
( )n t ( ) oNS f =
( ) ( ) ( )mV
thresh.
bT
[ ]
( ) ( )2nS f =
The signals are thusg
( )( ) ( )1 k k k bs t t n t t t t T
y tì - + £ £ +ïï=í( )
( ) ( )2 k k k b
y ts t t n t t t t T
íï - + £ £ +ïî
sampling at givesk bt t T= +
( ) [ ] 0T N kì +ï [ ] 0E N( ) [ ]( ) [ ]
01
02
0
1b k
kb k
s T N x kV
s T N x k
+ =ïï=íï + =ïî
[ ]
( )22 2
0
2k
k
oN N
E NN H f dfs s
¥
-¥
=
= = ò2 ¥ò
cont …
the Vk are Gaussian random variables
Ziemer
depends on Pr[0] and Pr[1]
if Pr[0]=Pr[1], then
( ) ( )1k s T s Té ù= +ë û( ) ( )01 022opt b bk s T s Té ù= +ë û
cont … ( ) ( )101 022opt b bk s T s Té ù= +ë û
with this formulation
[ ] ( )P |"0" | "0"f d¥
ò[ ] ( )Pr error|"0" | "0"kVk
f v dv= ò( )01 bk s T
Qæ ö- ÷ç ( ) ( )02 01b bs T s T
Qæ ö- ÷ç ÷( )01 b
N
Qs
÷ç ÷= ç ÷ç ÷çè ø
( ) ( )02 01
2b b
N
Qs
÷ç ÷= ç ÷ç ÷çè ø
[ ]P |"1"[ ]Pr error|"1"=
the error probability is minimized by maximizing( ) ( )02 01b bs T s T
z-( ) ( )02 01b b
N
zs
=
( ) ( )02 01b bs T s Tz
-
cont …( ) ( )02 01b b
N
zs
=
( ) ( ) ( ) let ( ) ( ) ( )2 1g t s t s t= -
( ) ( )2
2 g Tg T( ) ( ) ( )0 02 01g t s t s t= - ( ) ( )02 02 2
k k
bb
N N
g Tg Tz
s s= =
now,( ) ( ) ( )1g t H f G f- é ù( ) ( ) ( )0g t H f G fé ù= ë û
2¥
ò ( ) ( )
( )
2
2
2
bj fTH f G f e df
N f df
p
z
¥
-¥
¥=ò
ò ( )2
2oN H f df
-¥ò
S h liSchwarz’s Inequality
This inequality states that2 2 2
( ) ( ) ( ) ( )2 2 2
*X f Y f df X f df Y f df¥ ¥ ¥
-¥ -¥ -¥£ò ò ò
with equality if( ) ( )X f kY f=( ) ( )f f
cont …( ) ( )*X f Y f df
¥
ò Proof:
( ) ( )2
X f Y f df¥
ò
( ) ( )
( )2
let X f Y f df
Y f dfa -¥
¥
-¥
-=
òò
( ) ( )
( ) ( ) ( )2
* *
X f Y f df
X f df X f Y f df
a
a
-¥
¥ ¥
+
+
ò
ò ò
¥
a is arbitrary
( ) ( ) ( )
( ) ( ) ( )2
2*
+
X f df X f Y f df
X f Y f df Y f df
a
a a
-¥ -¥
¥ ¥
= +
+
ò ò
ò ò( ) ( ) ( ) + X f Y f df Y f dfa a-¥ -¥
+ò òthese will now cancel
( ) ( ) ( )( ) ( )
2*
2 2 X f Y f dfX f Y f df X f dfa
¥
¥ ¥ -¥+
òò ò( ) ( ) ( )
( )2
X f Y f df X f dfY f df
a ¥-¥ -¥
-¥
+ = -ò òò
( ) ( ) ( ) ( )2 2 2
*¥ ¥ ¥
ò ò òcont …2
( ) ( ) ( ) ( )*X f Y f df X f df Y f df-¥ -¥ -¥
£ò ò ò
( ) ( ) ( )( ) ( )
2*
2 2
2
X f Y f dfX f Y f df X f dfa
¥
¥ ¥ -¥
¥+ = -ò
ò òò
( ) ( ) ( )( )
2Y f df
¥-¥ -¥
-¥
ò òò
always +ve
( ) ( )( )
2*
2X f Y f dfX f df
¥
¥-¥£
òò
( )( )
2X f df
Y f df¥ -¥
-¥
£òò
( ) ( ) ( ) ( )2 2 2
*X f Y f df X f df Y f df¥ ¥ ¥
-¥ -¥ -¥£ò ò ò
( ) ( )equality when X f kY f=
( ) ( ) ( ) ( )2 2 2
*¥ ¥ ¥
ò ò òcont … ( ) ( ) ( ) ( )*X f Y f df X f df Y f df-¥ -¥ -¥
£ò ò ò
Return to our maximization2¥
ò ( ) ( )
( )
2
2
2
bj fTH f G f e df
N
p
z
¥
-¥
¥=ò
ò let
( )2
2oN H f df
-¥ò let
( ) ( ) ( ) ( )*; bj fTH f X f G f e Y fp2= =
( ) ( )( )
2 2
222
2 2G f df H f dfG f df
N Nz
¥ ¥
¥-¥ -¥
¥£ =ò ò
òò ( )
( )2
o oN NH f df¥ -¥
-¥
òò
( )22 2 G f dfz
¥£ òcont … ( )2
o
G f dfN
z-¥
£ ò
The maximum z occurs under equality
( ) ( )X f kY f( ) ( )X f kY f=
( ) ( ) 2* bj fTH f kG f e p-=
in the time domain,,
( ) ( )*g t G f- «
( ) ( ) ( ) ( )2 1b b bh t g T t s T t s T t= - = - - -
cont … ( ) ( ) ( )2 1b bh t s T t s T t= - - -
Ziemer
this filter is said to be matched to s1(t)
the matched filter assumed no ISI and symbol width pulses
for raised cosine pulse shapes square root of the response at the transmitter square root of the response at the transmitter square root at the receiver
l iImplementation
Implement matched filters with correlators
( )bs T t-
0 bt T< <
( ) ( ) ( )y t s t n t= + ( )v t
t T
( )bv T
( ) ( ) ( )v t y t h t= * ( ) ( )bT
y s T t dt t t= - +ò
0 bt< <bt T=
( ) ( ) ( )v t y t h t= * ( ) ( )0 by s T t dt t t= +ò
( ) ( )0
( )bT
b s bs s T t d N R T t Nt t t= - + + = - +ò( ) ( ) ( )
0(0)
bT
b sv T y s d R Nt t t= = +ò0ò
a correlator
bT
òcont … ( ) ( ) ( )0
bT
bv T y s dt t t= ò
if we have rectangular pulses
( ) ( ) ( ) ( ); bs t u t h t u T t= = -
b bT T
ò ò( ) ( ) ( ) ( )0 0
b bT T
bv T y u d y dt t t t t= =ò òj t i t t d d th i tjust integrate and dump the input
b biliError Probability
Recall that
[ ]( ) ( )02 01P b bs T s T
Q Q zæ ö é ù- ÷ç ÷ ê úç[ ]( ) ( )02 01Pr error
2 2b b
N
Q Q zs
ç ÷= = ê úç ÷ç ÷ç ê úè ø ë û
( )1/ 2
2max
2
o
G f dfN
z¥
-¥
é ùê ú= ê úë û
ò ( ) ( ) ( )2 1G f S f S f= -
using Parseval’s theorem
oë û
( ) ( ) ( )1/ 2 1/ 2
22max 2 1
2 2g t dt s t s t dtN N
z¥ ¥
¥ ¥
é ù é ùé ùê ú ê ú= = -ë ûê ú ê úë û ë û
ò òo oN N-¥ -¥ê ú ê úë û ë ûò ò
( ) ( )1/ 2
22 dz¥é ùé ùê úòcont … ( ) ( )max 2 1
2
o
s t s t dtN
z-¥
é ùê ú= -ë ûê úë ûò
expanding gives
1/ 22 é ù( ) ( ) ( ) ( )1/ 2
2 2max 2 1 1 2
2 2o
s t dt s t dt s t s t dtN
z¥ ¥ ¥
-¥ -¥ -¥
é ù= + -ê úê úë ûò ò ò
2E 1Esignal correlation
1 ( ) ( )12 1 21 2
1 s t s t dtE E
r¥
-¥ò
we want as negative as possible
best performance when r12 = -1
lExample
Square full-width pulses, polar signaling the matched filter is ( )h t
A( )
( ) ( ) ( ) ( )1/ 2
2 22 1 1 2
2 2s t dt s t dt s t s t dtz¥ ¥ ¥é ù
= + -ê úò ò ò
bT
( ) ( ) ( ) ( )2 1 1 22o
s t dt s t dt s t s t dtN
z-¥ -¥ -¥
+ê úê úë ûò ò ò
b bAT E= b bAT E= 2 2b bAT E- =-22 2
b bAT E b b b b
[ ] 21 2Pr error 4 bb
EQ Q E Qz é ù é ùé ù ê ú ê ú= = =ê ú ê ú ê úê ú[ ]
2 2 bo o
Q Q QN Nê ú ê úê úë û ë û ë û
“SNR”
Optimum Linear Receiver
Problem with zero-forcing equalizer a solution to have zero ISI causes noise enhancement !
Minimum-mean square error (MMSE) equalizer a more balanced solution to reduce the effects of both
noise and ISI in a combined manner !
1 é ù21 | ( ) | (MSE criterion)2
where ( ) ( ) ( ) for ( ) ( ) ( )
b nJ E y nT a
y t x t c t x t a q t kT w t
é ù= -ë û
= Ä = +å design the receive filter to minimize the MSE J.
where ( ) ( ) ( ) for ( ) ( ) ( )k bky t x t c t x t a q t kT w t= Ä = - +å
g
cont …
Optimum linear receiver consisting of the cascade connection of matched filter and transversal equalizer.
*( )( )( ) / 2
Q fC fS f N
=+( ) / 2q oS f N+
cont …
Block diagram of adaptive equalizer.- training mode- decision-directed mode