Scribd
Upload
33 views26 pages

Matched Filter Maximizes The SNR at The Output of A FIR Filter

The document discusses the matched filter. The matched filter maximizes the SNR at the output of an FIR filter. It is defined over a finite time interval and is zero otherwise. When noise is absent, the matched filter output is equal to the signal energy. The matched filter can be used for signals with known arrival times. The shape of the signal does not affect detection performance.

Uploaded by

Erdem
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
33 views26 pages

Matched Filter Maximizes The SNR at The Output of A FIR Filter

The document discusses the matched filter. The matched filter maximizes the SNR at the output of an FIR filter. It is defined over a finite time interval and is zero otherwise. When noise is absent, the matched filter output is equal to the signal energy. The matched filter can be used for signals with known arrival times. The shape of the signal does not affect detection performance.

Uploaded by

Erdem
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 26

Matched Filter maximizes the SNR at the output of a

FIR filter

 h[n] is defined over [0,N-1] and zero otherwise

 When the noise is absent the matched filter output is just the signal energy

 For the signals with known arrival times we can use the matched flter
in its present form
The shpe of the signal does not effect the detection performance

Energies of the two signals are the same!!!


19/12/16

Deterministic Signals
Binary detection problem:
H0 x[n] = w[n]
H1 x[n] = s[n] + w[n]
Assumptions

• s[n] is deterministic and known.

• w[n] is white Gaussian noise with variance 2


.

Question: What is the correlation function of w[n]?

17

Deterministic Signals
The NP detector decides H1 if the likelihood ratio exceeds a threshold or Notice that presence of s[n]
implies change in mean of
p(x; H1 )
L(x) = >
p(x; H0 ) observe signal. Optimal detector

where x = [x[0], x[1], . . . , x[N 1]]T . Since will test whether there is a
" # change in the mean of the test
N 1
1 1 X 2
p(x; H1 ) = N exp 2
(x[n] s[n]) statistic.
(2⇡ 2) 2 2 n=0
" N 1
#
1 1 X 2
p(x; H0 ) = N
exp 2
x [n]
(2⇡ 2) 2 2 n=0

we have " !#
N
X1 N
X1
1 2 2
L(x) = exp 2
(x[n] s[n]) x [n] > .
2 n=0 n=0

18

9
Example 4.3 Uncorrelated Noise with Unequal
Variances

 If a data sample has small variance, then we weight its


contribution to the sum (decision statistics) more heavily
 In the WGN case C=   thus observations have the same weights
 The noise samples have been equalized or prewhitened thus
 The general matched filter first prewhitens the noise samples!
 In doing so it also distorts the signal , if present, to be
 As a result, after prewhitening the detector correlates against the
distorted signal
 The generalized matched filter can be expressed as

 The generalized matched filter can be thought of as a prewhitener,


which is then followed by a correlator or matched filter to the
distorted signal
Generalized Matched Filters
 The matched filter is an optimal detector for a known signal in WGN
 In many situations, however, the noise is more accurately modeled as
correlated noise
 Let C is the covariance matrix of the noise, we assume that
 If the noise is modeled as wide sense stationary (WSS), then C has the
special form of a symmetric Toeplitz matrix
 the elements along any NW to SE diagonal of C are the same


 Observation vector x[x[0] x[1]...x[N-1]] has the following PDFs under two
hypothesis

 In the WGN case C=  


 In the correlated noise case the detector is again referred as a
generalized replica-correlator or matched filter where the replica is
the modified signal     
For any C that is positive definite

  is exists and also positive definite thus it can


be factorized as   =   where D is a non-
singular NxN matrix referred as prewhitening
matrix
  
 D=diag , ,…, for the previous Exp.4.3
  

Thus the test statistics becomes


How to factorize the correlation matrix after
whitening
The prewhitening form of the generalized matched filter
Details of calculations
Why shape of the signal is important?

In the WGN case C=   and probability of


detection increases monotonically with the energy
of the signal not with its shape
In colored noise case the probability of detection
increases monotonically with thus the
signal can be designed to maximize

How to optimize the signal??


Detection OR Classification?

The problem we have been addressing up to


now is the detection of a known signal in noise

In communication / pattern recognition at the


receiver / classifier there is no question as to
whether a signal present but only which one

The problem then is a classification problem


(you may still refer to it as detection)
Example: Phase Shift Keying (PSK)

Called coherent PSK since we assume


perfect knowledge of the transmitted signal
To minimize the probability of error at the
reciver we transmit a sinusoidal with one of
two phases where    for n=0,1,...,N-1.
 the average signal enery:
 signal correlation
coefficient
The probability of error in a binary communication
system can be written as

 the average signal enery:


 signal correlation coefficient
 To minimize the error
subject to an average
energy constraint, we
choose the signals to
ensure   1

You might also like