Professional Documents
Culture Documents
Feed-forward
Feed-forward
filter
filter(FFF)
(FFF)
Adjustment of
filter coefficients
Output
+
Symbol
decision
Allowed transition
between states
State
Sample time
instants
Z f B f H f E f
Transmitted
Transmitted
symbol
symbol
spectrum
spectrum
Channel
Channelfrequency
frequency
response
response
(incl.
(incl.TT&&RRfilters)
filters)
Equalizer
Equalizer
frequency
frequency
response
response
B f
H f
E f
Z f
0
fs = 1/T
Zero-forcing equalizer
Transmitted
impulse
sequence
Communication
channel
Overall
channel
h k
h k n
fk
c k
Coefficients of
equivalent FIR filter
Equalizer
z k
Input to
decision
circuit
n N
r k
m M
cm hk m
m M
cm k m
( M k M )
Zero-forcing equalizer
We want overall filter response
to be non-zero at decision time
k = 0 and zero at all other
sampling times k 0 :
fk
m M
cm hk m
1, k 0
0, k 0
h0 c M h1c M 1 ... h2 M cM 0
This leads to
a set of 2M+1
equations:
(k = M)
(k = 0)
:
h2 M 1c M h2 M 2c M 1 ... h1cM 0
h2 M c M h2 M 1c M 1 ... h0 cM 0
(k = M)
J E ek
s k
Estimate
of k:th
symbol
Channel
r k
Error
ek
Equalizer
bk
z k
b k
zk
J E ek
J
c2
c1
Wiener solution
We start with the Wiener-Hopf equations in matrix form:
Rc opt p
R = correlation matrix (M x M) of received (sampled) signal
values
rk
r k rk , rk 1 ,..., rk M 1
*
p E r k bk
M samples
Algorithms
Stochastical information (R and p) is available:
1. Direct solution of the Wiener-Hopf equations:
Rc opt p
c opt R 1p
Inverting a large
matrix is difficult!
rk M
c M
c1 M
Widrow
cM 1
rk M
ek
cM
zk
bk
zk
Equalizer
Equalizerfilter
filter
e
Coefficient
Coefficient
updating
updating
bk
Phase
Phase
synchronization
synchronization
ek
Godard
Minimize:
J E ek
cm rk m exp j bk
m M
ek zk bk
Re cn i 1 Re cn i
ek ek ek
2 2 M 1 1
equations
Phase:
i 1 i
Iteration index
Re cn
Im cn i 1 Im cn i
ek
ek
ek
Im cn
Re cn i 1 Re cn i 2 Re e
Im cn i 1 Im cn i 2 Im e
r e j
c
r
b
m k m k k n
m M
r e j
c
r
m k m
k k n
m M
j M
i 1 i 2 Im bk e cm rk m
m M
ek
smaller
larger
Slow
Slow acquisition
acquisition
Poor
Poor stability
stability
Poor
Poor tracking
tracking
performance
performance
Large
Large variation
variation
around
around optimum
optimum
value
value
q1
qQ 1
c M
c1 M
bk
FBF
rk M
bk Q
?
zk
rk M
cM 1
FFF
cM
ek
LMS
algorithm
for tap
coefficient
adjustment
ek zk bk
m M
ek
cm rk m qn bk n bk
n 1
qn
m M
cm hn m
n 1, 2, K , Q
Proakis, Ed.3, Section 10-3-1
Estimated
symbols
bk
LMS
algorithm
c0
c1 cM 1
cM
rk
rk
+
bk M
hm cm
b k
Training
symbols
(no
errors)
r k
Channel
Channel
estimation
estimation
circuit
circuit
h m
Equalizer
Equalizer
&& decision
decision
circuit
circuit
b k
bk P K bk 1
bk 1 K bk Q
Precursor
Precursorcancellation
cancellation
of
offuture
futuresymbols
symbols
Postcursor
Postcursorcancellation
cancellation
of
ofprevious
previoussymbols
symbols
rk P
Filter
Filtermatched
matchedto
to
sampled
sampledchannel
channel
impulse
impulseresponse
response
bk
+
Matched
Matched
filter
filter
NW
NW
filter
filter
y k
MLSE
MLSE
(VA)
(VA)
b k
f k
f k
Channel
Channel
estimation
estimationcircuit
circuit
p y b, f p y k b , f
exp 2
N 2
N
k 1
2
2
Objective:
find symbol
sequence that
maximizes this
probability
Length of f (k)
y f b
k 1
K 1
n 0
k n
Metric to be
minimized
(select best b
..
using VA)
f k
0
1 0 0 1 0
At time k-2
1 0 0 1 0 0
At time k-1
1 0 0 1 0 0 1
At time k
1 0 0 1 0 0 1 1
causing ISI
:
16 states
m K 1
Alphabet
size
k-3
k-2
k-1
k+1