Chap4 PAM

1
Chapter 4 Baseband Pulse

Transmission
Techniques for the transmission of (originally)
digital data over a baseband channel are the
main focus of this chapter.
Po- Ning Chen@cm.nct u Chapt er 4- 2
4.1 Introduction
Transmission of digital data (bit stream) over a noisy
baseband channel typically suffers two channel
imperfections
Intersymbol interference (ISI)
Background noise (e.g., AWGN)
These two interferences/noises often occur simultaneously.
However, for simplicity, they are often separately
considered in analysis.
2
4.1 ISI
ISI
channel
ISI
channel
ISI
channel
ISI
channel
ISI
(t)
Impulse response
ISI
channel
ISI
channel
ISI
4.2 Matched filter
Matched filter is a device for the optimal detection of a
digital pulse. It is named so because the impulse response
of the matched filter matches the pulse shape.
System model without ISI
channel
3
4.2 Design criterion
To find h(t) such that the output signal-to-noise ratio SNR
O
is maximized.
) ( ) (
) ( * ) ( ) ( * ) (
) ( * )] ( ) ( [ ) (
0 for ) ( ) ( ) (
t n t g
t h t w t h t g
t h t w t g t y
T t t w t g t x
o
+ =
+ =
+ =
< + =
)] ( [
| ) ( |
2
2
T n E
T g
SNR
o
O
=
4.2 Analysis of matched filter
2
2
) 2 exp( ) ( ) ( | ) ( |
) 2 exp( ) ( ) ( ) (

=
=
df fT j f G f H T g
df ft j f G f H t g
o
o
, 2 / PSD with white being ) ( With

0
N t w

= =
= =
df f H
N
df f S T n E
f H
N
f H f S f S
N
W N
2
0
2
2
0
2
| ) ( |
2
) ( )] ( [
| ) ( |
2
| ) ( | ) ( ) (
4

=
df f H
N
df fT j f H f G
2
0
2
| ) ( |
2
) 2 exp( ) ( ) (
. constant some for ) ( ) ( if, only and if, holds equality with
| ) ( | | ) ( | ) ( ) (
*
2 1
2
2
2
1
2
2 1
k x k x
dx x dx x dx x x

=
|
.
|
\
|
|
.
|
\
|

Cauchy-Schwarz inequality
By Cauchy-Schwarzs inequality,

df fT j f G df f H df fT j f G f H
2 2
2
| ) 2 exp( ) ( | | ) ( | ) 2 exp( ) ( ) (
df f G
N
df f H
N
df f G df f H
2
0
2
0
2 2
| ) ( |
2
| ) ( |
2
| ) ( | | ) ( |
This is a constant bound, independent of the choice of h(t).

Hence, the optimal is achieved by:
) 2 exp( ) ( ) (
*
fT j f G k f H =
5
Hence, under additive white noise, the optimal received
filter matches the input signal in the sense that it is a time-
inversed and delayed version of the complex-conjugated
input signal g(t).
). (
) ( 2 exp( ) (
) 2 exp( ) 2 exp( ) ( ) (
*
*
*
opt
t T kg
df t T f j f G k
df ft j fT j f G k t h
=
|
.
|
\
|
=
=

The maximum output signal-to-noise ratio only depends on
the energy of the input, and is nothing to do with the pulse
shape itself.
Namely, whether the pulse shape is sinusoidal,
rectangular, triangular, etc is irrelevant to the maximum
output signal-to-noise ratio, as long as these pulse shapes
have the same energy.
4.2 Properties of matched filter
. | ) ( | where ,
2
2
0
max

= = df f G E
N
E
s
s
6
Example 4.1 Matched filter for rectangular pulse
Also h(t)
Example 4.1 Matched filter for rectangular pulse
h
opt
(t) in this example can be implemented as integrate-and-
dump circuit
7
4.3 Error rate due to noise
In what follows, we analyze the error rate of polar non-
return-to-zero (NRZ) signaling in a system with optimal
matched filter receiver over AWGN channel.
}. 1 , 1 { where ), ( ) ( + = I t g I t s
. ) ( ) ( and | ) ( | where ,
) ( ) ( | ) ( |
) ( ) ( ) ( ) (
) ( ) ( ) ( ) (
) ( * ) ( ) ( * )] ( [ ) (
* 2
* 2
* *

= =
= = + =
+ =
+ =
+ =
+ =

d w g n d g E kn kE I
d w g k d g k I
d T w T kg d T g T kg I
d T w h d T g h I
t h t w t h t g I T y
g g
T t T t
. by / ) ( brief e, convenienc notational For y k T y
e.) convenienc for here range n integratio the as line real
entire the use however I text. in does as range, this outside
zero is ) ( since ) , 0 [ over taken be can n integratio (The t g T
8
0. )] ( [ ) ( ] [
*
= =

d w E g n E
g
E
N
ds s g
N
dsdt t s
N
t g s g
dsdt t w s w E t g s g n E
2
) (
2
) (
2
) ( ) (
)] ( ) ( [ ) ( ) ( ] [
0
2
0
0
2
= =
=
=
with d distribute Gussian is ) ( ) (

, assumption ) ( real and ), ( of assumption AWGN By
*

= d w g n
t g t w
= =
+ = =
+ =
+
1 if ), 2 / , ( ) (
; 1 if ), 2 / , ( ) (
0 1
0 1
I N E E Normal y
I N E E Normal y
n E I y
g g
g g
g

made. is 1 favors decision a for which set the be Let +
( )
]. 1 Pr[ where , )] ( ) ( ) 1 [(
} 1 | Pr{ } 1 | Pr{ ) 1 (
} 1 | Pr{ ) 1 ( } 1 | Pr{ 1
} 1 | Pr{ ] 1 Pr[ } 1 | Pr{ ] 1 Pr[
} 1 | ) 1 ( guess Pr{ ] 1 Pr[ } 1 | ) 1 ( guess Pr{ ] 1 Pr[
1 1
+ = = + =
+ = = + =
= + + = =
= = + + = + = =
= + = + + = + = =
+
I p dy y p y p p
I y p I y p p
I y p I y p
I y I I y I
I I I I BER

{ }. 0 ) ( ) ( ) 1 ( : set optimal the , minimize To
1 1 opt
< =
+
y p y p y BER
.
) ( ) ( ) 1 ( , 1
) ( ) ( ) 1 ( , 1
) (
: by given is maker decision optimal the Thus,
1 1
1 1

< +
=
+
+
y p y p
y p y p
y d

9
)
`
=
)
`
=
)
`
)
`
=
>
<
+
0 0
2
2
2
2
2
2
2
1
1
1
1
4
exp
2 /
2
exp
2
exp
2
) (
exp
2
1
2
) (
exp
2
1
) (
) ( ) 1 (
N
y
N E
y E
y
y
y
y
y
p
p
g
g
= =
+ = =
+
1 if ), 2 / , ( ) (
; 1 if ), 2 / , ( ) (
0 1
0 1
I N E E Normal y
I N E E Normal y
g g
g g
. 2 / and Let
0
2
N E E
g g
= =
(

<
>
+
p
p N
y
) 1 (
log
4
0
1
1
This threshold depends on N
0
; hence, the best
decision relies on the accuracy of N
0
estimate.
4.3 Error rate due to noise under uniform input
In order to free the system dependence on N
0
estimate, a
uniform I is transmitted in which case, p = .
The best decision now becomes y 0.
1 +
>
)
`

+
)
`
+
=
+ =
0
2
2
2
0
2
2
2
0
1
0
1 opt
2
) (
exp
2
1
2
1
2
) (
exp
2
1
2
1
) (
2
1
) (
2
1
dy
y
dy
y
dy y dy y BER

1
<
10
{ }
|
|
.
|
\
|
=
|
|
.
|
\
|
=
=
=
)
`
=
)
`
+
)
`
0
2
2 /
2
2
2
2
2
2
2
2
2
2
2
opt
erfc
2
1
2
erfc
2
1
exp
1
2
,
2
exp
2
1
2
exp
2
1
2
1
2
exp
2
1
2
1
2
N
E
dz z
y
z dy
y
dy
y
dy
y
BER
g
function. error ary complement the is ) exp(

2
) ( erfc where
2
=
u
dz z u

4.3 Error function
Error function
Complementary error function
Q-function
) exp(
2
) ( erf
0
2
=
u
dz z u
=
u
dz z u ) exp(
2
) ( erfc
2
|
|
.
|
\
|
=
u
dz
z
u Q
2
exp
2
1
) (
2
|
.
|
\
|
=
=
=
2
erfc
2
1
) (
) ( erf 1 ) ( erfc
) ( erf ) ( erf
u
u Q
u u
u u
11
4.3 Error function
Bounds for error function
|
.
|
\
|
+

+ =

L
6 3 4 2 2
2
5 3 1
2
3 1
2
1
1
1
) ( erfc
2
x x x
e
x
x
x
2 2 1
) ( erfc
2
1
1
1
, 0 For
2
x x
e
x
x
x
e
x
x

< <
|
.
|
\
|
>

(The bound is good when x is large.)
4.3 Error rate due to noise
The optimal BER formula is important in communications:
The best decision is y 0.
|
|
.
|
\
|
=
|
|
.
|
\
|
=
0 0
opt
2
erfc
2
1
N
E
Q
N
E
BER
g g
1 +
>
1
<
12
g
T T
b
E dt t g I E dt t s E E
I t g I t s
= = =
+ =

0
2 2
0
2
) ( ] [ )] ( [ case, this In
}. 1 , 1 { where ), ( ) (
4.4 Intersymbol interference
The channel is usually dispersive in nature.
In this section, we only consider discrete pulse-amplitude
modulation (PAM). Consideration of PDM and PPM will
be out of the scope of this section.
. ) ( ) ( and 1 2 }, 1 , 0 {

=
= =
k
b k k k k
kT t g a t s b a b
13
4.4 Intersymbol interference
Notably, in the previous section, we only consider one
interval of input.
This is justifiable because of no ISI.
However, in this section, we have to consider
since ISI is involved.
We also assume perfect synchronization to simplify the
analysis.
. ) ( ) (

=
=
k
b k
kT t g a t s
) ( ) ( t g I t s =
) ( )] ( * ) ( [ ) ( ) ( * ) ( ) (
) ( ) (
t w t h kT t g a t w t h t s t x
kT t g a t s
k
b k
k
b k
+ = + =
=
=
b
b
iT t
iT t
k
b k b
k
b k
t c t w t c t h kT t g a iT y
t c t w t c t h kT t g a t c t x t y
=
=
=
+ =
+ = =

) ( * ) ( )] ( * ) ( * ) ( [ ) (
) ( * ) ( )] ( * ) ( * ) ( [ ) ( * ) ( ) (
Information of a
k
is carried at [kT
b
, (k+1)T
b
).
14
) (
)} ( 2 exp{ ) ( ) ( ) (
} 2 exp{ ) ( ) ( } 2 exp{ ) ( ) ( * ) ( * ) (
b
b
b b
kT t p
dt kT t f j f C f H f G
dt ft j f C f H fkT j f G t c t h kT t g
=
=
=

. } 2 exp{ ) ( ) ( ) ( ) ( where

= dt ft j f C f H f G t p
) ( * ) ( ) ( where ), ( ) ( ) ( t c t w t n t n kT t p a t y
k
b k
= + =

=
) ( ) ( ) (
b
k
b b k b
iT n kT iT p a iT y + =

=
4.4 ISI and noise
Without ISI,
. } 2 exp{ ) ( ) ( ) ( 1 ) (

= = dt ft j f C f G t p f H
), ( ) ( or }, 2 exp{ ) ( ) ( filter matched With
* *
t T g t c fT j f G f C
b b
= =
1 if , 0
) ) 1 ( ( ) (
, ) ( ) (
) ( ) ( ) (
*
*
=
+ =
= =
=

i
ds T i s g s g
T s d iT g T g
d iT g c iT p
b
b b b
b b

15
) ) 1 (( ) ( ) ) 1 (( ) ) 1 (( ) ) 1 ((
b b i b
k
b k b
T i n T p a T i n T k i p a T i y + + = + + + = +

=
) ( ) ) 1 (( ) ) 1 ((
noise, additive and ISI without result, a As
b i
k
b k b
T p a T k i p a T i y = + = +

=
)}. ) 1 (( { by ted reconstruc completely be can } { and
b i
T i y a +
The text sets p(0) = 1 for simplicity, but is a little confusion!
The text is correct when information of a
i
is carried at [(i1)T
b
, iT
b
).
In my notations, information of a
i
is actually carried at [iT
b
, (i+1)T
b
).
So in order to recover a
i
, correlation (convolution) operation should
start at iT
b
, and end (is sampled) at (i+1)T
b
.
Hence, y((i+1)T
b
) is used to reconstruct a
i
.
4.5 Nyquists criterion for distortionless baseband
binary transmission
Is it possible to completely eliminate ISI (in principle) by
selecting a proper g(t)?
. 0 ) 0 ( and , 0 all for 0 ) ( satisfies
} 2 exp{ ) ( ) ( ) ( ) ( that such ) ( and ) ( Choose
=
=

p i iT p
dt ft j f C f H f G t p t c t g
b
(Here, I assume that information of a

i
is carried at [(i1)T
b
, iT
b
). )
16
binary transmission
Let P(f) = G(f)H(f)C(f).
Sample p(t) with sampling period T
b
to produce P
(f).
(Aliasing) From slide Chapter 3-4, we get:
Also from slide Chapter 3-4, we have:
=
|
|
.
|
\
|
=
n
b b
T
n
f P
T
f P
1
) (
( ) 1 2 exp ) ( ) ( = =

= n
b b
f nT j nT p f P

binary transmission
This concludes that the condition for zero ISI is:
This is named the Nyquist criterion.
The overall system frequency function P(f) suffers no
ISI for samples taken at interval T
b
if it satisfies the
above equation.
Notably, P(f) represents the overall accumulative effect
of transmit filter, channel response, receive filter, etc.
b
n
b
T
T
n
f P =
|
|
.
|
\
|
=
17
4.5 Ideal Nyquist channel
The simplest P(f) that satisfies Nyquist criterion is the
rectangular function:
. ) ( ) ( and
2
1
| | , 0
2
1
| | ,
) (
b
b
b
b
T W P W P
T
W f
T
W f T
f P = +
= >
= <
=
) 2 ( sinc
2
) 2 sin(
) ( Wt
Wt
Wt
t p = =

18
1
0
1
1
0
1
0
b b
iT T i ~ ) 1 (
4.5 Infeasibility of ideal Nyquist channel
Rectangular P(f) is infeasible because:
p(t) extends to negative infinity, which means that each
a
k
have already been transmitted at t = !
A system response being flat from W to W, and zero
elsewhere is physically unrealizable.
The margin of error is quite small, as a slight shift (error)
in sampling time (such as, iT
b
+) would cause very
large ISI.
Note that p(t) decays to zero at a very slow rate of
1/|t|.
19
4.5 Infeasibility of ideal Nyquist channel
Examination of timing error margin
Let t be the sampling time difference between
transmitter and receiver.
For simplicity, set i = 0.
=
+ = +
k
b k b
t T k i p a t iT y ) ) (( ) (
=

=
=
k
b
b
k
k
b k
kT t W
kT t W
a
kT t p a t y
) ( 2
)] ( 2 sin[
) ( ) (
=

+
=

=

=
0
0
2
) 1 ( ] 2 sin[
2
] 2 sin[
2
] 2 sin[ ) 1 (
2
] 2 sin[
) (
k
k
k
k
k
k
k
k
k
k t W
a t W
t W
t W
a
k t W
t W
a
k t W
k t W
a t y

. 0 small fixed any for
2
) 1 (
that such } { exists There
0
> =

=
t
k t W
a
a
k
k
k
k
k
Question: How to make p(t) decays faster?
Answer: Make P(f) smoother.
20
4.5 Raised Cosine Spectrum
exists.
) (
then , ) ( if
), ( function e nonnegativ a For
k
k
k
f
f P
dt t p t
t p
<

We extend the bandwidth of p(t) from W to 2W, and require
that
So the price to pay here is a larger bandwidth.
One of the P(f) that satisfies the above condition is the
raised cosine spectrum.
. | | for
2
1
) 2 ( ) 2 ( ) ( W f
W
W f P W f P f P < = + + +
W f
W f W
W
W f
W
W f
W
f P
) 1 ( | | , 0
) 1 ( | | ) 1 ( ,
2
) ) 1 ( | (|
cos 1
4
1
) 1 ( | | 0 ,
2
1
) (

+
+ <
)
`

+
<
=
21
The transmission bandwidth of the raised cosine spectrum
is equal to:
where is the rolloff factor, which is the excess bandwidth
over the ideal solution.
) 1 ( 2 + = W B
T
The text puts B

T
= W(1+)
may not be justifiable!
large | | as
| |
1
~
16 1
) 2 cos(
) 2 ( sinc ) (
3
2 2 2
t
t
t W
Wt
Wt t p
|
.
|
\
|
22
consists of two terms:
The first term ensures the desired zero crossing of p(t).
The second term provides the necessary tail
convergence rate of p(t).
The special case of = 1 is known as the full-cosine rolloff
characteristic.
|
.
|
\
|
=
2 2 2
16 1
) 2 cos(
) 2 ( sinc ) (
t W
Wt
Wt t p
2 2
16 1
) 4 ( sinc
) (
t W
Wt
t p
=
Useful property of full-cosine spectrum.
We have more zero-crossing at 3T
b
/2, 5T
b
/2, 7T
b
/2,
in addition to the desired T
b
, 2T
b
, 3T
b
This is useful in synchronization. (Think of when

synchronized, the quantity should be small both at 3T
b
/2,
5T
b
/2, 7T
b
/2,and at T
b
, 2T
b
, 3T
b
)
However, the price to pay for this excessive synchronization
information is to double the bandwidth.
=
=
=
|
.
|
\
|
2 , 0
1 ,
2
1
0 , 1
2
i
i
i
iT
p
b
23
Example 4.2 Bandwidth requirement of the T1
system
For T1 transmission, a frame consists of 24 PCM-encoded
voice channels and 1 framing bit.
The resultant number of bits in a frame is 24 8 + 1 =
193.
The duration of each frame is 125s.
Hence,
kHz ) 1 ( 772 ) 1 (
. kHz 772
2
1
647688 . 0
193
125
Baseband) ,(

+ = + =
= =
= =
W B
T
W
s
s
T
T
b
b
W W
4.6 Correlative-level coding
ISI, when generated in an uncontrolled manner, is an
undesirable phenomenon.
However, ISI may become a friend if it is added to the
transmitted signal in a controlled manner.
Known fact: A signal of bandwidth W can be
distortionlessly transmitted using its samples with
sampling rate 2W.
Conversely, in a channel with bandwidth WHz, the
theoretical maximum signal rate is 2W symbols per
second.
24
W W
B B
W W
W W
. second per samples 2
is rate signal maximum The
W
A channel with bandwidth W Hz
Why intentionally adding ISI? Answer: To have better
bandwidth efficiency.
Ideal Nyquist pulse shaping is efficient; it cannot be
realized.
Raised consine pulse shaping is realizable; it is
bandwidth inefficient.
By adding ISI to the transmitted symbols in a controlled
manner, we can achieve the Nyqusit rate 2Win a
channel bandwidth of W Hertz.
Correlative-level coding or Partial-response
signaling
25
4.6 One example of Correlative-level coding
Duobinary signaling (or class I partial response)
i.i.d. } { where
k
b
) ( f H
duoB
Lets ignore the effect of H
Nyquist
(f) first in the block
diagram in the previous slide. We directly obtain:
Note that c
k
has three levels (2,0,2).
The transfer function of the overall system is thus:
) 2 exp( 1 ) (
b DuoB
fT j f H + =
4.6 Duobinary signaling
26
H
Nyqusit
(f):
Only for derivation purpose (do not need it finally)
As shown in the next slide, the response H
I
(f) is
realizable.

H
I
(f)
27
h
I
(t):
) ( ) (
Nyquist Nyquist b
T t h t h + =

h
I
(t):
28
Bandwidth efficiency of duobinary signaling
Example.
) ( * ) ( ) ( signal ed transmitt The t g kT t a kT t g a
k
b k
k
b k
(
= =

=
=

k
b k
kT t a ) (
) (t g
The input to this filter is not WSS!
Then we can introduce time-average autocorrelation function.
4.6 Time-average autocorrelation function
) ( f H
) (t X ) (t Y

+ = + =
T
T
X
T
T
T T
X
dt t t R
T
dt t X t X E
T
R ) , (
2
1
lim )] ( ) ( [
2
1
lim ) (
Function ation Autocorrel Average Time

+ =
+ =
T
T
X
T
T
T
Y
T
Y
dt d d t t R h h
T
dt t t R
T
R
2 1 2 1 2 1
) , ( ) ( ) (
2
1
lim
) , (
2
1
lim ) (

(Assume that limit and integration are interchangeable.)
29
2 1 1 2 2 1
2 1 2 1 2 1
) ( ) ( ) (
) , (
2
1
lim ) ( ) ( ) (

d d R h h
d d dt t t R
T
h h R
X
T
T
X
T
Y
+ =
|
.
|
\
|
+ =

) ( | ) ( |
real is ) ( if ), ( ) ( ) (
' ) ' ( ) ( ) (
' ]) ' [ 2 exp( ) ' ( ) ( ) (
) 2 exp( ) ( ) ( ) (
) 2 exp( ) ( ) (
2
*
' 2
1
2
1 2
2
2
2 1 1 2 2 1
2 1 1 2 2 1
1 2
f S f H
h f S f H f H
d e R d e h d e h
d d d f j R h h
d d d f j R h h
d f j R f S
X
X
f j
X
f j f j
X
X
Y Y
=
=
=
+ =
+ =
=

Now back to the example.
=
=
k
b k
kT t a t X ) ( ) (
) (t g
dt kT t jT t a a E
T
dt jT t a kT t a E
T
R
T
T
k
b
j
b k j
T
T
T
j
b j
k
b k
T
X
=

+ =
(
(
|
|
.
|
\
|
+ |
.
|
\
|
=
) ( ) ( ] [
2
1
lim
) ( ) (
2
1
lim ) (

=
=
k
b k
kT t g a t Y ) ( ) (
(to channel)
30
) (
1
) (
2
1
lim ) (
) ( ) (
2
1
lim ) (

b
T
T
k
b
T
T
T
k
b b
T
X
T
dt kT t
T
dt kT t kT t
T
R
=
|
.
|
\
|
=
|
.
|
\
|
+ =
=

0 T T
them of
2
ely Approximat
b
T
T
2
| ) ( |
1
) ( f G
T
f S
b
Y
=
=
=
k
b k
kT t a t X ) ( ) (
) (t g

=
=
k
b k
kT t g a t Y ) ( ) (
) (t h
DuoB
2 2
| ) ( | | ) ( |
1
) ( f H f G
T
f S
DuoB
b
Y
=
(to channel)
31
( ) ( )
b b
b
fT T f G
T t
t g
2 2 2
sinc | | .
otherwise , 0
0 , 1
) ( Assume =
<
=
=
ISI Signal With ), 2 ( sinc
ISI Signal No ), ( sinc

ISI Signal With ), ( )sinc ( cos
ISI Signal No ), ( sinc
) 0 (
) (
2
2
2 2
2
b
b
b b
b
Y
Y
fT
fT
fT fT
fT
S
f S
) exp( ) cos( 2 ) (
b b DouB
fT j fT f H =
0
0.2
0.4
0.6
0.8
1
1.2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
ISI Signal with ), 2 ( sinc
2
b
fT
ISI Signal without ), ( sinc
2
b
fT
b
fT
32
Conclusions
By adding ISI to the transmitted signal in a controlled (and
reversible) manner, we can reduce the requirement of bandwidth of
the transmitted signal.
Hence, in the previous example, {c
k
} can be transmitted in every
T
b
/2 seconds!
Doubling the transmission capacity without introducing
additional requirement in bandwidth!
Duobinary signaling : Duo means doubling the transmission
capacity of a straight binary system.
A large SNR is required to yield the same error rate because of an
increase in the number of signal levels (from 1, +1 to 2, 0, 2).
Detailed discussion on error rate impact is omitted here!
Conclusions (cont.)
The duobinary signaling is also named class I partial
response.
Full response: The transmission wave at each time
instance is fully determined by a single information
symbol.
Partial response: The transmission wave at each time
instance is only partially determined by one
information symbol (i.e., is fully determined by two
or more information symbols).
33
4.6 Decision feedback for correlative-level coding
Recovering of {a
k
} from {c
k
}
It requires the previous decision to determine the current
symbol.
So the system should feedback the previous decision.
Error therefore may propagate!
How to avoid error propagation? Answer: Precoding.
1

=
k k k
a c a
4.6 Precoding of correlative coding
1 1
1
~
2
~ ~
} i.i.d. } 1 , 0 { {

+ = = =
k k k k k k k k k
a a c b a b b b b
1
1 2 } i.i.d. } 1 , 0 { {

+ = =
k k k k k k
a a c b a b
Without precoding
With precoding
2
~
2
~
2
) 1
~
2 ( ) 1
~
2 (
1
1
1
+ =
+ =
+ =
k k
k k
k k k
b b
b b
a a c
1
~ ~
=
k k k
b b b
k
b
~
1
~
k
b
k
c
k
b
0
0
1
1
0
1
0
1
0
1
1
0
2
0
0
2
34
Final notes
The precode must not change the duo- of the
transmission capacity of a straight binary system.
Hence, i.i.d. be must }
~
{
k
b
I.i.d. of
It suffices to show that:
}
~
{
k
b
( ) ( )
k k k k
b b b b
~
Pr ,...
~
,
~
|
~
Pr
2 1
=

( ) ( )
1 2 1 1
~
|
~
Pr ,...
~
,
~
|
~
Pr
~ ~

= =
k k k k k k k k
b b b b b b b b
= = = = =
= = = = =
= = = = =
= = = = =
2 / 1 ) 0 Pr( ) 1
~
| 1
~
Pr(
2 / 1 ) 1 Pr( ) 0
~
| 1
~
Pr(
2 / 1 ) 1 Pr( ) 1
~
| 0
~
Pr(
2 / 1 ) 0 Pr( ) 0
~
| 0
~
Pr(
1
1
1
1
k k k
k k k
k k k
k k k
b b b
b b b
b b b
b b b
35
=
= + = =
= = = +
= = = = =
=
= + = =
= = = +
= = = = =

2
1

2
1
) 1
~
Pr(
2
1
) 0
~
Pr(
) 1
~
| 1
~
Pr( ) 1
~
Pr(
) 0
~
| 1
~
Pr( ) 0
~
Pr( ) 1
~
Pr(
2
1

2
1
) 1
~
Pr(
2
1
) 0
~
Pr(
) 1
~
| 0
~
Pr( ) 1
~
Pr(
) 0
~
| 0
~
Pr( ) 0
~
Pr( ) 0
~
Pr(
1 1
1 1
1 1
1 1
1 1
1 1
k k
k k k
k k k k
k k
k k k
k k k k
b b
b b b
b b b b
b b
b b b
b b b b
Q.E.D.
Example 4.3 Duobinary coding with precoding
Table 4.1 in text
0 1 1 0 1 0 0
2 0 0 2 0 +2 +2 {c
k
}
1 1 +1 1 1 +1 +1 +1 {a
k
}
0 0 1 0 0 1 1 1
0 1 1 0 1 0 0 {b
k
}
}
~
{
k
b
}
{
k
b
36
4.6 Modified duobinary signaling
The PSD of the signal is nonzero at the origin.
The is considered to be an undesirable feature in some
applications, since many communication channels cannot
transmit a DC component.
Solution: Class IV partial response or modified duobinary
technique.
0
0.2
0.4
0.6
0.8
1
1.2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
2 2
1
~
2
~ ~
} i.i.d. } 1 , 0 { {

= = =
k k k k k k k k k
a a c b a b b b b
) ( f H
MDuoB
) 4 exp( 1 ) (
b MDuoB
fT j f H =
37
( ) ( )
b b
b
fT T f G
T t
t g
2 2 2
sinc | | .
otherwise , 0
0 , 1
) ( Assume =
<
=
=
=
Duobinary Modified ), ( )sinc (2 sin ) 4 /( ) (

Duobinary ), 2 ( sinc / ) (
2 2
2
b b b Y
b b Y
fT fT T f S
fT T f S
) 2 exp( ) 2 sin( 2
) 2 exp( )] 2 exp( ) 2 [exp(
) 4 exp( 1 ) (
b b
b b b
b MDuoB
fT j fT j
fT j fT j fT j
fT j f H

=
=
=
0
0.2
0.4
0.6
0.8
1
1.2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Duobinary ), 2 ( sinc
2
b
fT
duobinary Modified
), ( )sinc (2 sin
2 2
b b
fT fT
b
fT
38
Precoding is added to eliminate error propagation in
decision system.
2
~ ~
=
k k k
b b b
k
b
~
2
~
k
b
k
c
k
b
0
0
1
1
0
1
0
1
0
1
1
0
0
2
2
0
2
2
2
~
2
~
2
) 1
~
2 ( ) 1
~
2 (
=
=
=
k k
k k
k k k
b b
b b
a a c
4.6 Generalized form
of correlative level
coding (or partial
response signaling)
). 2 exp( where
, ) (
1
1
1
1 0
b
N
N CLC
fT j z
z w z w w f H
=
+ + + =

L
39
4.6 Generalized form of correlative-level coding
(or partial-response signaling)
1 0 2 0 1 5 V
Modified duobinary coding 1 0 1 3 IV
1 1 2 3 III
1 2 1 3 II
Duobinary coding 1 1 2 I
Comments w
4
w
3
w
2
w
1
w
0
N Type of Class
+ =
V fT
IV fT
III fT fT
II fT
I fT
T
f G
f S
b
b
b b
b
b
b
Y
) 2 ( sin 16
) 2 ( sin 4
) 2 ( sin 8 ) ( cos 4
) ( cos 16
) ( cos 4
| ) ( |
) (
4
2
2 2
4
2
2
( ) ( )
b b
b
fT T f G
T t
t g
2 2 2
sinc | | .
otherwise , 0
0 , 1
) ( Assume =
<
=
0
0.5
1
1.5
2
-1.5 -1 -0.5 0 0.5 1 1.5
III
I
II
IV
V
40
4.7 Baseband M-ary PAM transmission
Gray code
Any dibit differs from
an adjacent dibit in a
single bit position.
For M-ary PAM transmission, there are M possible symbols
with symbol duration T.
1/T is referred to as the signaling rate or symbol rate or
symbols per second or baud.
Some equivalences
Each symbol can be equivalently identified with log
2
M
bits.
So the baud rate 1/T can be equivalently transformed to
bps as:
) ( log
2
M T T
b
=
41
Some equivalences
Virtually fix the symbol error, namely, fix the level
distance (to be 2). For example, (+1, 1) for M = 2, and
(+3, +1, 1, 3) for M = 4. Then the transmitted power
per unit time for M-ary PAM transmission becomes:
( )
) ( log 3
) 1 ( 1
) ( log 3
) 1 (
) ( log
) 1 ( ) 3 ( )] 3 ( [ )] 1 ( [
1
] [
2
2
2
2
2
2 2 2 2
2
M
M
T M T
M
M T
M M M M
M
T
S E
b b
b
|
|
.
|
\
|
=
=
+ + + +
=
L
) ( log 3
) 1 ( 1 ] [
2
2 2
M
M
T T
S E
b
|
|
.
|
\
|
=
For fixed bps, the transmitted power for M-ary
transmission must be increased by a factor M
2
/log
2
M.
42
4.8 Digital subscriber lines
A DSL operates over a local loop (often less than 1.5km) that
provides a direct connection between a user terminal (e.g.,
computer) and a telephone companys central office (CO).
Since it is a direct connection, no dialup is necessary.
The information-bearing signal is kept in the digital domain
all the way from the user terminal to an Internet service
provider.
DSL is intended to provide high data-rate, full-duplex,
digital transmission capability using local cost
configuration (such as twisted pairs for ordinary telephonic
communications).
One of two possible modes can be used to achieve the full-
duplex goal.
Time compression (TC) multiplexing
Echo-cancellation mode
43
Time-compression (TC) multiplexing
A guard time is often inserted between bursts in the
two opposite directions of data.
So the required line rate is slightly greater than twice
the data rate.
Transmitter
Receiver
Transmitter
Receiver
Echo-cancellation (EC) mode
Support the simultaneous flow of data along the
common line in both directions.
In this mode, the line rate is the same as the data rate.
44
Hybrid transformer
Two-to-four-wire conversion
Comparison between TCM mode and EC mode
EC offers a much better data-transmission performance
at the expense of increase complexity.
However, with the recent advance in VLSI, complexity
is no longer a main system concern. So in North
America, the EC mode has been adopted as the basis for
designing the transceiver.
45
Other impairments to DSL
ISI and Crosstalk
The transfer function of a twisted pair line can be
approximated by
( ) f f H = exp | ) ( |
2
pair twist
pair. twisted the of
length actual and length reference ly the respective are and
and pair, twisted the of constant physical a is , where
0
0
l l
k
l
l
k =
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
-1 0 1 2 3 4 5
ISI
( )
0 pair twist
2 exp | |
2
exp ) (
f j f f H
|
.
|
\
|
=
2
0
2 2
pair twist
) ( 16
4
) (

+
= h
2 2
pair twist
) 1 . 0 ( 16 1
4
) (
+
=

h
) 1 ( ) ( + t t
46
Crosstalk
Capacitive coupling that exists between adjacent
twisted pairs in a cable
Near-end crosstalk and Far-end crosstalk
Near-end crosstalk Far-end crosstalk
Crosstalk (cont.)
FEXT suffers the same line loss as the signal,
whereas NEXT does not.
This is close to the phenomenon of near-far effect
of wireless channel.
Accordingly, NEXT will be a more serious problem
than FEXT. So we can ignore the effect of FEXT,
and add NEXT filter to the twisted pair channel
model (as shown in the figure in the next slide).
47
2 / 3
NEXT
) ( f f H
Interference (input of H
NEXT
(f)) often
assumes to have the same PSD as the
transmitted signal, but is Gaussian
distributed.
Other features of DSL channel
The PSD of the transmitted signal should be zero at zero
frequency because no DC transmission through a hybrid
transformer is possible.
The PSD of the transmitted signal should be low at high
frequencies because
transmission attenuation in a twisted pair is most
severe at high frequency;
crosstalk due to capacitive coupling between
adjacent twisted pairs increases dramatically at high
frequency (recall that the impedance of a capacitor is
inversely proportional to frequency).
48
Possible candidates for line codes that are suitable for DSL
Manchester code
Zero DC component but large spectrum at high
frequency so it is vulnerable to NEXT and ISI.
Bipolar (Alternate mark inversion or AMI) code
Successive 1s are represented alternately by positive
and negative but equal levels, and 0 is represented by
a zero level.
Zero DC component. Its NEXT and ISI performance
is slightly inferior to the modified duobinary code on
all digital subscriber loops.
Modified duobinary code
Of no DC component and moderately spectrally efficient.
However, its robustness against NEXT and ISI is about 2
to 3 dB poorer than that of (2B1Q) block codes on worst-
case subscriber lines.
2B1Q code
Two binary bits encoded into one quaternary symbol
(four-level PAM signal).
Zero DC component, and offers the best performance
among all the codes introduced. So it is adopted as the
standard as the North American standard for DSL.
49
2B1Q code (cont.)
2B1Q code (cont.)
With 2B1Q line coding, adaptive equalizer and echo
cancellation, it is possible to achieve BER = 10
7
operating full duplex at 160 kb/s.
50
4.8 Asymmetric digital subscriber lines
ADSL is targeted to simultaneously support three services
at a single twisted-wire pair
Data transmission downpstream at 9 Mbps
Data transmission upstream at 1Mpbs
Plain old telephone service (POTS)
Some notes
It is named asymmetric because the downstream bit rate
is much higher than the upstream bit rate.
The actually achievable bit rates depend on the length of
the twisted pair used to do the transmission.
Frequency-division multiplexing (FDM) technique is used
to combine analog voice and DSL data.
Upstream and downstream data transmission are placed in
different frequency band to avoid crosstalk.
splitter
51
Various applications can be applied to asymmetric
transmissions, such as video-on-demand (VoD).
For example
Downstream = 1.544 Mbps (DS1) for video data
Upstream = 160 kbps for real-time control
commands.
4.9 Optimum linear receiver
Zero-forcing equalizer
One receiver design is to use a zero-forcing equalizer
followed by a decision-making device.
The design objective of zero-forcing equalizer is to
force the ISI to zero at all sampling instances t = kT
b
for
k 0, provided that the channel noise w(t) is zero.
52
Zero-forcing equalizer (cont.)
This reduces to the Nyquist criterion.
b
n
b
T
T
n
f P =
|
|
.
|
\
|
=
where P(f) = G(f)H(f)C(f).
=
=
0 , 0
0 , 1
) (
n
n
nT p
b
or
Zero-forcing equalizer (cont.)
A serious consequence of the ignorance of w(t) in the
design of zero-forcing equalizer is the performance
degradation due to noise enhancement.
53
Example of noise enhancement.
Suppose that the receiver filter is a tapped-delay-line
equalizer, which is of the form
Assume ideally that G(f) = 1.
Hence, the Nyquist criterion becomes:
=
=
0
) ( ) (
k
b k
kT t c t c
where P(f) = H(f)C(f).
=
=
0 , 0
0 , 1
) (
n
n
nT p
b

=
=
|
.
|
\
|
=
=
0
0
0
) (
) ( ) (
) ( ) (
) ( ) ( ) (
k
b k
k
b k
k
b k
kT t h c
d kT t h c
d kT t c h
d t c h t p

=
= = = =

=

=
0 , 0
0 , 1
) ) (( ) (
0 0
n
n
h c T k n h c nT p p
k
k n k
k
b k b n
54
. 0 arbitrary for
1
0 1
0 0 1
0 0 0 1
0
0
0
1
1
1
0
2 1
1 2
1
>
(
(
(
(
(
(
(
(
(
(
(
(
=
(
(
(
(
(
(

N
c
c
c
c
h h h
h h
h
N
N
N N N
M
L
M M M M M
L
L
L
M
. 1 and , 0 for 0 that assume to reasonable is It
0
= < = h n h
n
<
=
otherwise , 0
2 0 ), 2 /( | | 1
) (
b b
T T
h

Suppose
1
b
T 2
. 1 , 0 for 0 and ,
2
1
, 1
1 0
= = = n h h h
n
otherwise. zero, and , 0 for 2 ) 1 ( =

n c
n n
n

=
0
) ( ) (
becomes noise sampled the , instance any time At
k
k n k b
b
w c d nT c w
nT

The above c(t) can successfully remove ISI, provided w(t) = 0.
Now add the additive white Gaussian noise w(t), which also
passes the filter c(t).
| |
2 2
0
2 2
0
2
0
3
4
2 Var Var
: becomes then variance noise sampled The
w w
k
k
w
k
k n k
k
k n k
w c w c > = = =
(
=

55
An easier way to interpret the noise enhancement
phenomenon.
Nyquist criterion requires that:
A sufficient condition for the Nyquist criterion is that:
When H(f) is very small (or zero) at some frequency
range, C(f) has to be very large (or infinity) at the same
frequency range in order to equalize the spectrum.
Thus, the noise spectrum S
W
(f)|C(f)|
2
after passing
through C(f) will be enhanced.
b
b
n
b
n
b
T
T
n
f C
T
n
f H
T
n
f P =
|
|
.
|
\
|
|
|
.
|
\
|
=
|
|
.
|
\
|
=
Spectrum Cosine Raised ) ( ) ( = f C f H
To alleviate noise enhancement phenomenon, it is better to
simultaneously consider the ISI and channel noise.
An approach of this kind is to use the mean-square error
criterion, and find a balanced solution to the problem of
reducing the effects of both channel noise and intersymbol
interference.
56
=
+ =
= =

) ( * ) ( ) (
) ( ) ( ) (
) ( ) ( ) ( * ) ( ) (
t h t g t q
t w kT t q a t x
d t x c t x t c t y
k
b k

i i b b b
k
k b
n d iT w c d kT iT q c a iT y + = + =

) ( ) ( ) ( ) ( ) (
. ) ( receiver, perfect For
i b
a iT y =
. ) ( error the So,
i i i i
a n e + =
: minimize to wishes then criterion error squared mean The
e. convenienc for added is
2
1
factor the where
], } ) ( [{
2
1
] [
2
1

2
i
2
i i i i
a n E e E J + = =
] [ ] [ ] [ ] [
2
1
] [
2
1
] [
2
1
2 2 2
i i i i i i i i i i
a E a n E n E a E n E E J + + + =
=
k
b b b b q
kT iT q kT iT q i R ) ( ) ( ) ; , ( where
2 1 2 1

For i.i.d. {a
k
} where a
k
= 1,

=
=
2 1 2 1 2 1
2 1 2 1 2 1
2
) ; , ( ) ( ) (
) ( ) ( ) ( ) ( ] [ ] [

d d i R c c
d d lT iT q kT iT q c c a a E E
q
k l
b b b b l k i
1st term
57
Assume white w(t) with PSD N
0
/2.

=
=
=
1 1
2 0
2 1 2 1
0
2 1
2 1 2 1 2 1
2
) (
2
) (
2
) ( ) (
)] ( ) ( [ ) ( ) ( ] [

d c
N
d d
N
c c
d d iT w iT w E c c n E
b b i
2nd term
3rd term
. 1 ] [
2
=
i
a E For i.i.d. {a
k
} where a
k
= 1,
4th and 5th term
By the independence of {a
k
} and w(t), and zero mean of n
i
,
. 0 ] [ ] [ ] [ and 0 ] [ ] [ ] [ = = = =
i i i i i i i i
a E n E a n E n E E n E
6th term

= = d q c d kT iT q c a a E a E
k
b b i k i i
) ( ) ( ) ( ) ( ] [ ] [
Substitute all six terms into J
i
.

+ +
=
1 1 1 1 1
2 0
2 1 2 1 2 1
) ( ) (
2
1
) (
4
) ; , ( ) ( ) (
2
1

d q c d c
N
d d i R c c J
q i
Taking the derivative of J
i
with respect to c(s), we obtain:
) ( ) (
2
) ; , ( ) (
) ; , ( ) (
2
1
) ; , ( ) (
2
1
) (
0
} /{
1 1 1
} /{
2 2 2
1 2
s q s c
N
i s s R s c
d i s R c d i s R c
s c
J
q
s
q
s
q
i
+ +
+ =

58
), ; , ( ) ( ) ( ) ; , ( Since
1 2 2 1 2 1
i R kT iT q kT iT q i R
q
k
b b b b q
= =
0 ) ( ) (
2
) ; , ( ) (
) ( ) (
2
) ; , ( ) (
) (
0
0
=
|
.
|
\
|
+ =
+ =

s q d s
N
i s R c
s q s c
N
d i s R c
s c
J
q
q
i

So the optimal design for equalizer (or received filter) c() is:
) ( ) (
2
) ; , ( ) (
0
s q d s
N
i s R c
q
=
|
.
|
\
|
+

). ( * ) ( ) ( and ) ( ) ( ) ; , ( where t h t g t q kT iT q s kT iT q i s R
k
b b b b q
= =

An equalizer designed based on the above equality is referred to
as the minimum-mean square error (mmse) equalizer.
: yield to equalizer mmse for criterion design the into this Take
). ( as it reexpress then can We
. respect to
with invariant is and , and between difference the on depends
only ) ( ) ( ) ; , ( that Observe
2 1
2 1
2 1 2 1

q
k
b b b b q
R
i
kT iT q kT iT q i R
) ( ) (
2
) ( ) (
0
s q d s
N
s R c
q
=
|
.
|
\
|
+

We can then take the Fourier transform of both sides, and obtain:
) (
2
) ( ) (
* 0
f Q
N
f S f C
q
=
|
.
|
\
|
+
equalizer. mmse for solution
2 / ) (
) (
) (
0
*
N f S
f Q
f C
q
+
=
59
4.9 MMSE equalizer
Summary
The MMSE equalizer can be viewed as the
concatenation of two filters:
a matched filter Q
*
(f) to Q(f) = G(f)H(f)
an equalizer whose frequency response is the inverse
of S
q
(f) + N
0
/2.
4.9 MMSE equalizer
Property of S
q
(f)
The text wrote that , which
is periodic with period 1/T
b
. This implies that R
q
()
consists of a series of pulse train with width T
b
, which is
not true.
|
|
.
|
\
|
+ =
k
b b
q
T
k
f Q
T
f S
2
1
) (
=
k
b b q
kT q kT q R ) ( ) ( ) (
2 1 2 1

60
|
|
.
|
\
|
+ =
|
.
|
\
|
=
=
=
=
=
|
.
|
\
|
= =

k
b b
k
b
k
b b
k
b b
k
b b
k
b b
k
b b q q
T
k
f Q
T
f Q
dt ft j kT t t q f Q
fkT j kT q f Q
dv fv j v q fkT j kT q
dv v kT f j v q kT q
d f j kT q kT q
d f j kT q kT q d f j R f S
1
) (
) 2 exp( ) ( ) ( ) (
) 2 exp( ) ( ) (
) 2 exp( ) ( ) 2 exp( ) (
)) ( 2 exp( ) ( ) (
) 2 exp( ) ( ) (
) 2 exp( ) ( ) ( ) 2 exp( ) ( ) (
*
*
*

4.9 Implementation of MMSE equalizer
One can approximate 1/[S
q
(f) + N
0
/2] by a periodic function
with:
Since
q
(f) = is now periodic with period
1/T
b
, we obtain by Fourier series that
) (
~ 1 1
) ( ) (
2
*
f S
T
k
f Q
T T
k
f Q
T
f Q f S
q
k
b b
k
b b
q

|
|
.
|
\
|
+
|
|
.
|
\
|
+ =

. ) 2 exp( ) ( where
) 2 exp( ) (
) 2 /( 1
) 2 /( 1
=
=
=
b
b
T
T
b q b k
k
b k q
df kfT j f T c
kfT j c f
] 2 / ) (
~
/[ 1
0
N f S
q
+
61
We can approximate Q
q
(f) by its main 2N+1 terms as:
=

N
N k
b k q
kfT j c f ) 2 exp( ) (

=
+
N
N k
b k q
kT t c ) ( ) (
One can therefore approximate 1/[S
q
(f) + N
0
/2] by a transversal tapped-delay-line equalizer
Final notes
In real-life telecommunications environment, the
channel is usually time-varying.
Therefore, an adaptive receiver that provides for the
adaptive implementation of both the matched filter and
the equalizer in a combined manner is usually necessary.
It may be desirable to have the taps of the equalizer
(named fractionally spaced equalizer or FSE) spaced by
a fraction of the symbol period T
b
, e.g., T
b
/2.
S
q
(f) is now approximated by a periodic spectrum
with period 2/T
b
. Details are omitted in this book.
62
4.10 Adaptive equalization
The equalizer is adjusted under the guidance of a training
sequence transmitted through the channel.
The training sequence commonly used in practice is the pseudonoise
(PN) sequence, which consists of a deterministic periodic sequence with
noise-like characteristics. This subject will be discussed in Chapter 7.
Least-mean-square (LMS) algorithm
Design objective
To find the filter coefficients w
0
, w
1
, , w
N
so as to
minimize index of performance J:
=
= =
N
k
k
k n x w n d n y n d n e
0
] [ ] [ ] [ ] [ ] [
] [
2
n e J =
63
To minimize J, we should update w
i
toward the bottom of
the J-bowel.
So when g
i
> 0, w
i
should be decreased.
On the contrary, w
i
should be increased if g
i
< 0.
Hence, we may define the update rule as:
where is a chosen constant step size, and is
included only for convenience of analysis.
i
i
w
J
g
i i i
g w w =
2
1

current , next ,
] [ ] [ ] [ ] [ 2 ] [
] [ ] [
0 0 0
2
2
0
j n x k n x w w k n x n d w n d
k n x w n d J
N
k
N
j
j k
N
k
k
N
k
k
+ =
|
.
|
\
|
=

= = =
=
] [ ] [ 2
] [ ] [ ] [ 2
] [ ] [ 2 ] [ ] [ 2
0
0
n e i n x
k n x w n d i n x
i n x k n x w i n x n d
w
J
g
N
k
k
N
k
k
i
i
=
|
.
|
\
|
=
+ =
=
=
64
Some notes on LMS algorithm
There is no guarantee that the algorithm converges to a
local minimum (could converge to a saddle point).
There is even no guarantee that the algorithm converges.
.
, 0 For
] [ ] [ , 0 For
] [ ] [ ] [
Repeat
current , current ,
current , next ,
0
current ,
=
+ =
=

=
i i
i i
N
k
k
w w N i
n e i n x w w N i
k n x w n d n e

Some notes on LMS algorithm (cont.)
If is too large, high excess mean-square error may
occur.
If is too small, a slow rate of convergence may arise.
65
4.10 Operation of the equalizer
Two modes of operations for adaptive equalizer
Training mode (position 1)
Decision-directed mode (position 2)
4.10 Decision-directed mode
In normal operation, the decisions made by the receiver are
correct with high probability.
Under such premise, we can use the previous decisions to
calibrate or track the tap coefficients.
In this mode,
if is too large, high excess mean-square error may
occur.
if is too small, a too-slow tracking may arise.
We can further extend the idea of decision-directed or
decision-feedback to the decision-feedback equalizer (DFE).
66
4.10 Decision-feedback equalizer
Subtract out ISI
. at time sample where ,
and
Let
) 2 (
) 1 (
nT n
n
n
n
n
n
n
=
(
=
(
=
a
x
v
w
w
c
symbol. ed transmitt th the is where , Denote n a a e
n n
T
n n n
v c =
.
n
T
n n n
a e v c =
4.10 Decision-feedback equalizer
Then DFE gives:
As anticipated, DFE suffers from error propagation due to
incorrect decisions.
However, error propagation will not persist indefinitely;
rather, it tends to occur in bursts.
E.g., if the number of taps in the feedback section is L,
then the influence of one decision error will be flushed
out after subsequent L correct decisions.
.
2
1
) 2 (
) 1 (
) 2 (
1
) 1 (
1
(
+
(
=
(
+
+
n n
n n
n
n
n
n
e
e
a
x
w
w
w
w
67
4.11 Computer experiments: Eye patterns
Eye pattern: The synchronized superposition of all possible
realizations of the signal of interest viewed within a
particular signaling interval.
-1
0
1
0 1 2 3 4 5 6 7 b
T t /
Eye pattern for pulse shaping function p(t) is half-cycle sine
wave with duration T
b
, and error-free BPSK transmission.
-1
0
1
0 0.5 1 1.5 2
b
T t /
68
-1
0
1
0 0.5 1 1.5 2
Eye pattern for pulse shaping function p(t) is half-cycle sine
wave with duration 2T
b
, and error-free BPSK transmission.
b
T t /
Interpretation of eye pattern
69
-1
0
1
0 0.5 1 1.5 2
Experiment 1: Effect of
channel noise
(Raise-cosine pulse-shaping with
roll-off factor = 0.5, W= 0.5 Hz,
M= 4)
(a) Eye diagram for noiseless
quaternary system.
(b) Eye diagram for quaternary
system with SNR = 20 dB.
(c) Eye diagram for quaternary
system with SNR = 10 dB.
70
Experiment 2: Effect of
bandwidth limitation
(Raise-cosine pulse-shaping
with roll-off factor = 0.5, W
= 0.5 Hz, M= 4)
(a) Eye diagram for noiseless
band-limited quaternary system:
cutoff frequency f
o
= 0.975 Hz
(b) Eye diagram for noiseless
band-limited quaternary system:
cutoff frequency f
o
= 0.5 Hz
(The channel is now modeled by
a low-pass Butterworth filter with
50
0
2
) / ( 1
1
| ) ( |
f f
f H
+
=

Chap4 PAM

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chap4 PAM

Uploaded by

Copyright:

Available Formats

1

Chapter 4 Baseband Pulse

, 2 / PSD with white being ) ( With

This is a constant bound, independent of the choice of h(t).

with d distribute Gussian is ) ( ) (

Po- Ning Chen@cm.nct u Chapt er 4- 16

function. error ary complement the is ) exp(

Po- Ning Chen@cm.nct u Chapt er 4- 19

(Here, I assume that information of a

Po- Ning Chen@cm.nct u Chapt er 4- 32

Po- Ning Chen@cm.nct u Chapt er 4- 34

Po- Ning Chen@cm.nct u Chapt er 4- 40

The text puts B

This is useful in synchronization. (Think of when

Po- Ning Chen@cm.nct u Chapt er 4- 52

Po- Ning Chen@cm.nct u Chapt er 4- 54

Duobinary Modified ), ( )sinc (2 sin ) 4 /( ) (

Po- Ning Chen@cm.nct u Chapt er 4- 128

You might also like