Matched Filter

Matched Filter
Lecture No. 11
Dr. Aoife Moloney
School of Electronics and Communications
Dublin Institute of Technology
Lecture No. 11: Matched Filter
Overview
This lecture will look at the following:
Probability of error
Error function
Matched lter
February 2005 Slide:
1
Probability of Error
There are 2 ways an error can occur:
H
1
is chosen while s
2
was sent
H
2
is chosen while s
1
was sent
The total probability of error can be written:
P
B
= p(H
2
|s
1
)p(s
1
) + p(H
1
|s
2
)p(s
2
)
As s
1
and s
2
are equiprobable (p(s
1
)=p(s
2
)=1/2), and
because of the symmetry of the pdfs (p(H
2
|s
1
)=p(H
1
|s
2
)),
2
we have:
P
B
= p(H
1
|s
2
) = p(H
2
|s
1
)
= p
_
z(T) >
(a
1
+ a
2
)
2
|s
2
_
= p
_
z(T) <
(a
1
+ a
2
)
2
|s
1
_
We already know that the conditional probability den-
sity function of z, given that s
2
was transmitted, is given
by p(z|s
2
). So we can nd the probability that z >
(a
1
+ a
2
)/2 (eventhough an s
2
(not an s
1
) was sent) by
integrating the pdf p(z|s
2
) from (a
1
+ a
2
)/2 to innity.
3
So the total probability is given by:
P
B
=
1
_
a
1
+a
2
2
exp
_
1
2
_
z a
2
_
2
_
dz
We now let u = (z a
2
)/ and get:
P
B
=
1
_
a
1
a
2
2
exp
_
u
2
2
_
dz
There is no closed form solution to this integral. The
integral is evaluated using maths tables, numerical inte-
4
gration techniques or closedform approximations. Here
we use an error function and tables to evaluate the inte-
gral.
The probability of error P
B
is therefore given by:
P
B
= Q
_
a
1
a
2
2
_
Q is called the complementary or error function and
must be tabulated.
5
The Error Function Q
The Qfunction is an integral of a Gaussian pdf dened
as
Q(a) =
1
_
a
exp
_
x
2
2
_
dx
The Qfunction has the following properties:
Q(0) = 1/2
Q() = 0
Q(+) = 1
6
Q(a) = 1 Q(a)
There are a number of similar integral functions used for
error calculations. Common ones are:
erf(a) =
2
a
_
0
exp
_
x
2
_
dx, a 0
erfc(a) =
2
_
a
exp
_
x
2
_
dx = 1 erf
_
a
2
_
, a 0
7
The Q function is related to these functions by:
Q(a) =
1
2
_
1 erf
_
a
__
=
1
2
erfc
_
a
2
_
, a 0
Q(a) or Q(x) cannot be calculated directly and is avail-
able in tabular form
It is possible to dene a simple approximation which al-
lows a quick evaluation, when a > 3:
Q(a)
1
a
2
exp
_
a
2
2
_
, a > 3
An example of some Q function tables are given below:
8

9
Q(a)
1e-01
1e-02
1e-03
1e-04
1e-05
1e-06
1e-07
1e-08
1e-09
1e-10
a
1.2815
2.3263
3.0902
3.7190
4.2649
4.7535
5.1993
5.6120
5.9978
6.3613
a
3.5
3.6
3.7
3.8
3.9
4.0
4.5
5.0
Q(a)
2.327e-04
1.591e-04
1.078e-04
7.237e-05
4.812e-05
3.17e-05
3.40e-06
2.87e-07
a
5.5
6.0
6.5
7.0
7.5
8.0
8.5
9.0
Q(a)
1.90e-8
9.87e-10
4.02e-11
1.28e-12
3.19e-14
6.22e-16
9.48e-18
1.13e-19
10
Matched Filter
The matched lter is the linear lter providing the maxi-
mum signaltonoise power ratio at its output for a given
received symbol waveform
Let us again consider the description used for the output
of our linear lter. The sampled output of the linear lter
can be written:
z(T) = a
i
(T) + n(T)
where a
i
(T) is the signal component, and n(T), the noise
11
component.
The ratio of the instantaneous signal power to the average
noise power, at time T at the output of the receiver is:
(S/N)
T
=
a
2
i
n
2
The signal at the output of the lter can be described
using the inverse Fourier transform of its sepctrum:
a(t) =
H(f)S(f)e
j2ft
df
12
where S(f) is the Fourier transform of the signal at the
input of the lter, and H(f) is the lter transfer function
The noise power at the output of the lter can be written:
n
2
=
N
0
2
|H(f)|
2
df
Remember: The power spectral density of white noise
is N
0
/2
We can rewrite the signaltonoise ratio at the output of
13
the receiver at time T as:
_
S
N
_
T
=
H(f)S(f)e
j2fT
df
2
N
0
2
|H(f)|
2
df
Note: We can use the Schwarzs inequality:
f(x)g(x)dx
|f(x)|
2
dx
|g(x)|
2
dx
which holds for f(x) = kg
(x), with k an arbitrary con-

14
stant and

indicating complex conjugate
Using Schwarzs inequality gives:
_
S
N
_
T
=
H(f)S(f)e
j2fT
df
2
N
0
2
|H(f)|
2
df
|H(f)|
2
df
|S(f)|
2
df
N
0
2
|H(f)|
2
df
15
This simplies to:
_
S
N
_
T
=
2
N
0
|S(f)|
2
df
We can identify the energy of the input signal as:
E =
|S(f)|
2
df
Remember: The equation for energy spectral density
given in Lecture 5
16
So we can write the previous inequality as:
max
_
S
N
_
T
=
2E
N
0
The maximum signal to noise ratio depends on the en-
ergy of the input signal, and the power spectral density
of the noise
According to Schwarzs theorem this equality holds for
H(f) = kS
(f)e
j2fT
Taking the inverse Fourier transform, we get the time
17
domain response:
h(t) =
_
_
_
ks(T t), for 0 < t < T;
0, elsewhere.
Matched Filter Summary:
A lter that is matched to a signal s(t) of duration T, has
an impulse response that is a timereversed and delayed
version of the input s(t)
The impulse response of the linear lter matching the
18
signal s(t) of duration T can be written:
h(t) =
_
_
_
s(T t), for 0 < t < T;
0, elsewhere.
The peak pulse signaltonoise ratio depends only on the
ratio of the signal energy to the power spectral density
of the white noise at the lter input. The peak pulse
signaltonoise ratio is (S/N)
max
= 2E/N
0
, where N
0
/2
is the power spectral density of the input noise
19
Conclusion
This lecture has looked at the following:
Probability of error
Error function
Matched lter
20

Matched Filter

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Matched Filter

Hochgeladen von

Copyright:

Verfügbare Formate

Matched Filter

(x), with k an arbitrary con-

Das könnte Ihnen auch gefallen