You are on page 1of 4

# 1

Matched Filter
Instructor: M.A. Ingram
ECE4823
Maximum SNR Property
Assume AWGN
We want the filter that yields the
highest signal-to-noise ratio (SNR) at its
output
Next, show that the optimum filter is
the matched filter
Set Up
Suppose , where n(t) is
WGN with spectral height N
o
/2, and s(t)
is a signal with a finite duration T
Let r(t) be the input to a filter with
impulse response h(t)
Let y(t) be the output
) ( ) ( ) ( t n t s t r + =
) (t h ) (t r ) (t y
Signal and Noise Parts
The signal part of the output is
The noise part of the output is

=
t
s
du u t h u s t y
0
) ( ) ( ) (

=
t
n
du u t h u n t y
0
) ( ) ( ) (
Signal-to-noise Ratio
| |
2
0
2
0
2
2
) ( ) (
) ( ) (
) (
) (
SNR

=
=

t
t
n
s
du u t h u n E
du u t h u s
t y E
t y
Denominator
| |
{ }

=

=
t t
t t
n
dudv v t h u t h v n u n E
dv v t h v n du u t h u n E
t y E
0 0
0 0
2
) ( ) ( ) ( ) (
) ( ) ( ) ( ) (
) (
2
Invoke White Noise Model
| |

=
=
t
t t
n
du u t h
N
dudv v t h u t h v u
N
t y E
0
2 0
0 0
0
2
) (
2
) ( ) ( ) (
2
) (

SNR So Far
To optimize the SNR, choose h(u) to
maximize the numerator

=
t
t
du u t h
N
du u t h u s
0
2
0
2
0
) (
2
) ( ) (
SNR
Cauchy-Schwarz Inequality
Say S and Q are two points in a Hilbert
space, then
with equality when Q=cS
2 2 2
, Q S Q S
Cauchy-Schwarz for Signals
Let S and Q be points in the Hilbert
space of square-integrable functions
Then,
Equality is reached when

t t t
du u q du u s du u q u s
0
2
0
2
2
0
) ( ) ( ) ( ) (
) ( ) ( u q u cs =
Apply Cauchy-Schwartz
Recall numerator of SNR
Pick h(t-u) to be equal to cs(u)
2
0
) ( ) (

t
du u t h u s
Simplify Optimal SNR
Substitute h(t-u)=cs(u)
2
) (
) (
2
) (
) ( SNR
0
0
2
0
2
2
0
2
0
2
opt
N
du u s
du u s
c N
du u s c
t
t
t
t

=
3
Optimize t
If s(t) has finite duration T , then SNR
is maximized by setting t=T
0
0
0
2
opt
2
2
) (
SNR
N
N
du u s
s
T

= =

Max SNR Filter = Matched
Filter
Matched filter
impulse response is a
flipped in place
version of signal
) ( ) ( u cs u T h =
0 T
s(u)
0 T
h(T-u)
u
u
0 T
h(u)
u
) ( ) ( u T cs u h =
or
Inner Product Interpretation
of Matched Filter
) (
) (
t T cs
t h

=
) (t r ) (t y
r s, ) ( ) (
) ( ) (
) ( ) ( ) (
0
0
0
c du u r u s c
du u T r u T cs
du u T r u h T y
T
T
T
= =
=
=

## Matched Filter Frequency

Response
Take Fourier Transform of the Matched
Filter impulse response
du e u T s c
du e u h f H
fu j
T
fu j
T

2
0
2
0
) (
) ( ) (

=
=
Let r=T-u
( )
( )
| |
* 2 2
0
2
2
0
2
0
) ( ) (
) (
) ( ) (
f S ce dr e r s ce
dr e r s c
du e u T s c f H
fT j fr j
T
fT j
r T f j
T
fu j
T

= =
=
=

## Matched in Frequency Domain

The magnitude of the matched filter
response is just a scaled version of the
signals F.T.
) ( ) ( f S c f H =
4
Relation to Optimum
Detection
Recall optimum detector is the
minimum distance detector
Signal Space
The black dots are
the noiseless points
choose s
2
choose s
1
Alternative Form of Minimum
Distance Detector
Expand the signal space distance
between the received vector r and the
noiseless signal point s
m
2 2 2
, 2
m m m
s s r r s r + =
| |
m m
+ s r, 2 arg =
opt
m min
m
If Signals are Equal Energy
The minimum distance receiver can be
implemented as a bank of matched
filters
| |
m m
+ s r, 2 arg =
opt
m min
m
m
s r, max
m
arg =

T
m
dt t s t r
0
) ( ) ( max
m
arg =
Summary
When the input is signal plus WGN, then the
filter that maximizes the SNR is the matched
filter
The proof is an application of the Cauchy-
Schwarz Inequality
The filter matches (has the same shape as)
the signal in magnitude in the frequency
domain
The minimum-distance receiver can be
implemented as a bank of matched filters