This preview shows page 1. Sign up to view the full content.
Unformatted text preview: ms Matched Filters
In a digital communication system it is usual for the received ﬁlter to be
matched to the transmitted signal. In this case, if s(t ) is the transmitted
signal and is of duration T beginning at 0, we sample the ﬁlter output
at time T and h(t ) = s(T − t ). This is called the matched ﬁlter. The
ﬁlter output is
y (t ) = ∞
−∞ t h(t − α)s(α)d α = t −T
t =
t −T
t =
t −T h(t − α)s(α)d α
s(T − (t − α))s(α)d α
s(α − (t − T ))s(α)d α (This is the autocorrelation of the signal s(t )).
EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 61 / 174 Lecture Notes 2 Linear Systems Matched Filter The desired signal is the output sampled at time T .
T T y (T ) =
0 h(T − α)s(α)d α = s(α)s(α)d α
0
T = s2 (α)d α 0 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 62 / 174 Lecture Notes 2 Linear Systems Multiuser System
In a spreadspectrum system the signal has the form
s(t ) = N −1
l =0 al ψ (t − lTc ) so the impulse response of the matched ﬁlter has the following form
h(t ) = s(T − t ) = N −1
l =0 al ψ (T − t − lTc ) where N is the number of “chips” per bit, 1/Tc is the chip rate, and
NTc = T is the data bit duration or the inverse data rate. In this case
the implementation of the matched ﬁlter can be simpliﬁed as follows. EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 63 / 174 Lecture Notes 2 Linear Systems Multiuser System Let s(t ) be the ﬁlter input then
y (t ) = ∞
−∞ = h(t − α)s(α)d α ∞ N −1 −∞ l =1 = N −1
l =0 EECS 455 (Univ. of Michigan) al al ψ (T − t + α − lTc )s(α)d α
∞ −∞ ψ (T − t + α − lTc )s(α)d α. Fall 2012 September 7, 2012 64 / 174 Lecture Notes 2 Linear Systems Matched Filter Let x (t ) be the output of a ﬁlter with impulse response ψ (Tc − t ) then
x (t ) = ∞
−∞ ψ (Tc − t + α)s(α)d α. Now it is clear that
y (t ) = N −1
l =0 EECS 455 (Univ. of Michigan) al x (t − (N − 1 − l )Tc ). Fall 2012 September 7, 2012 65 / 174 Lecture Notes 2 Linear Systems Matched Filter Thus the matched ﬁlter can be implemented as a ﬁlter matched to the chip
waveform followed by a weighted sum. Since we are interested in the sample
only at time t = mT we only need the samples of the chip matched ﬁlter at
multiples of Tc . For example at time t = T the output is
N −1 al x ((l + 1)Tc ) y (T ) =
l =0 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 66 / 174 Lecture Notes 2 Linear Systems Matched Filter
At time t = mT the output is
y (mT ) = N −1
l =0 al x ((m − 1)T + (l + 1)Tc )
mN y (mT ) =
j =(m−1)N +1 aj −1−(m−1)N x (jTc ) If the spreading sequence is periodic with period N so that aj +N = aj
then
mN y (mT ) =
j =(m−1)N +1 EECS 455 (Univ. of Michigan) Fall 2012 aj −1 x (jTc ). September 7, 2012 67 / 174 Lecture Notes 2 Linear Systems Matched Filter t = jTc
ψ (t ) × mN
j =(m−1)N +1 y (MT ) x (jTc )
a j −1 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 68 / 174 Lecture Notes 2 Linear Systems Matched Filtering
s(t ) +1 t
1 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 69 / 174 Lecture Notes 2 Linear Systems Matched Filtering
h(t ) +1 t
1 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 70 / 174 Lecture Notes 2 Linear Systems Matched Filtering
7
6
5
4
3
2
1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
1
2
3 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 71 / 174 Lecture Notes 2 Linear Systems Multiple Bits
s(t ) +1 t
1 EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 72 / 174 Lecture Notes 2 Linear Systems Multiple Bits
y (t ) t EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 73 / 174 Lecture Notes 2 Linear Systems Multiple Bits
y (t ) t EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 74 / 174 Lecture Notes 2 Random Variables, Random Processes, Noise Noise
Communications systems have some amount of noise.
Electrical systems have random thermal noise due to motion of
electrons because the system is not at absolute zero temperature.
This noise is usually modeled as having power at all frequencies
but in actuality at very high frequencies the power decreases (in
the optical range of frequencies).
The model widely used for thermal noise is that of zero mean
white Gaussian noise.
Since only the frequency band of the transmitted signal is of
interest the noise outside this band is not important.
For all systems considered here we model the noise as having
equal power at all frequencies (white noise).
In addition the noise will have zero mean or average.
EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 75 / 174 Lecture Notes 2 Random Variables, Random Processes, Noise Probability Because we can not know the value of the noise ahead of time we
can only characterize it by its statistical properties such as
average values.
Using probability we can determine likelihoods of events. EECS 455 (Univ. of Michigan) Fall 2012 September 7, 2012 76 / 174 Lecture Notes 2 Random Variables...
View
Full
Document
This note was uploaded on 02/12/2014 for the course EECS 455 taught by Professor Stark during the Fall '08 term at University of Michigan.
 Fall '08
 Stark
 Frequency

Click to edit the document details