0% found this document useful (0 votes)
65 views25 pages

Optimum Receiver

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
0% found this document useful (0 votes)
65 views25 pages

Optimum Receiver

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/ 25

Optimum Receiver for

Signals Corrupted by AWGN


©Dr. James S. Kang
Professor
ECE Department
Cal Poly Pomona

1
Received Signal
• One of M signal waveforms [sm(t), m = 1, 2, ….,. M], 0 ≤ t < T, is
transmitted.
• 𝑠𝑠𝑚𝑚 𝑡𝑡 = ∑𝑁𝑁𝑘𝑘=1 𝑠𝑠𝑚𝑚𝑘𝑘 𝑓𝑓𝑘𝑘 (𝑡𝑡), linear combination of orthonormal basis
functions. AWGN n(t) is added to signal.
• r(t) = sm(t) + n(t), 0 ≤ t < T, received signal.
𝑁𝑁𝑜𝑜
• Φ𝑛𝑛𝑛𝑛 𝑓𝑓 = 𝑊𝑊/𝐻𝐻𝐻𝐻, noise PSD.
2
sm(t) r(t) No/2

n(t) 0 f

2
Model for Optimum Receiver
• Design a receiver that is optimum in the sense that it minimizes the
probability of making an error.

r(t) Signal r Output


Demodulator Detector
𝒓𝒓 = 𝑟𝑟1 , 𝑟𝑟2 , … , 𝑟𝑟𝑁𝑁 Decision

• Signal demodulator: correlator demodulator or matched filter


demodulator. Changes waveform r(t) to a vector r.
• Detector decides which of the M signals was transmitted based on r.
• Maximum likelihood (ML) detector or maximum a posteriori (MAP)
detector.
3
Correlator Demodulator

𝑇𝑇
� ( )𝑑𝑑𝑑𝑑 r1
0 t=T
f1(t)
r(t) 𝑇𝑇
� ( )𝑑𝑑𝑑𝑑 r2
0 t=T
f2(t)

𝑇𝑇
� ( )𝑑𝑑𝑑𝑑 rN
0
fN(t) t=T

4
Correlator Demodulator
𝑇𝑇 𝑇𝑇
• 𝑟𝑟𝑘𝑘 = ∫0 𝑟𝑟𝑘𝑘 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑 = ∫0 𝑠𝑠𝑚𝑚 𝑡𝑡 + 𝑛𝑛(𝑡𝑡) 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑
𝑇𝑇 𝑇𝑇
• = ∫0 𝑠𝑠𝑚𝑚 𝑡𝑡 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛(𝑡𝑡)𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑
• 𝑟𝑟𝑘𝑘 = 𝑠𝑠𝑚𝑚𝑘𝑘 + 𝑛𝑛𝑘𝑘 , 𝑘𝑘 = 1, 2, 3, … , 𝑁𝑁
𝑇𝑇
• 𝑠𝑠𝑚𝑚𝑘𝑘 = ∫0 𝑠𝑠𝑚𝑚 𝑡𝑡 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑, 𝑘𝑘 = 1, 2, 3, … , 𝑁𝑁 signal component
𝑇𝑇
• 𝑛𝑛𝑘𝑘 = ∫0 𝑛𝑛(𝑡𝑡)𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑, 𝑘𝑘 = 1, 2, 3, … , 𝑁𝑁 noise component
• 𝒔𝒔𝑚𝑚 = [𝑠𝑠𝑚𝑚1 , 𝑠𝑠𝑚𝑚2 , … , 𝑠𝑠𝑚𝑚𝑁𝑁 ] = one of the vectors in the constellation
• 𝒓𝒓 = 𝑟𝑟1 , 𝑟𝑟2 , … , 𝑟𝑟𝑁𝑁 = received vector

5
Noise
• 𝑟𝑟 𝑡𝑡 = 𝑠𝑠𝑚𝑚 𝑡𝑡 + 𝑛𝑛 𝑡𝑡 = ∑𝑁𝑁 𝑠𝑠 𝑓𝑓
𝑘𝑘=1 𝑚𝑚𝑘𝑘 𝑘𝑘 𝑡𝑡 + ∑𝑁𝑁
𝑘𝑘=1 𝑛𝑛𝑘𝑘 𝑓𝑓𝑘𝑘 𝑡𝑡 + 𝑛𝑛𝑛(𝑡𝑡)
• 𝑟𝑟 𝑡𝑡 = ∑𝑁𝑁 (𝑠𝑠
𝑘𝑘=1 𝑚𝑚𝑘𝑘 + 𝑛𝑛 )𝑓𝑓
𝑘𝑘 𝑘𝑘 𝑡𝑡 + 𝑛𝑛𝑛(𝑡𝑡) = ∑𝑁𝑁
𝑘𝑘=1 𝑟𝑟𝑘𝑘 𝑓𝑓𝑘𝑘 𝑡𝑡 + 𝑛𝑛𝑛(𝑡𝑡)
• 𝑛𝑛 𝑡𝑡 = ∑𝑁𝑁 𝑘𝑘=1 𝑛𝑛𝑘𝑘 𝑓𝑓𝑘𝑘 𝑡𝑡 + 𝑛𝑛𝑛(𝑡𝑡)
• 𝑛𝑛′ 𝑡𝑡 = 𝑛𝑛 𝑡𝑡 − ∑𝑁𝑁 𝑘𝑘=1 𝑛𝑛𝑘𝑘 𝑓𝑓𝑘𝑘 𝑡𝑡
• 𝑛𝑛′ 𝑡𝑡 is a zero-mean Gaussian noise process that represents the
difference between the original noise process n(t) and the projection
of n(t) onto the basis functions [fk(t)]. 𝑛𝑛′ 𝑡𝑡 is irrelevant to the
decision. 𝑛𝑛′ 𝑡𝑡 is the projection of n(t) onto other dimensions, other
than f1(k), f2(k), ….., fN(t).
6
Noise
• The decision may be based entirely on
• 𝑟𝑟𝑘𝑘 = 𝑠𝑠𝑚𝑚𝑘𝑘 + 𝑛𝑛𝑘𝑘 , 𝑘𝑘 = 1, 2, 3, … , 𝑁𝑁
𝑇𝑇
• 𝑛𝑛𝑘𝑘 = ∫0 𝑛𝑛(𝑡𝑡)𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑, 𝑘𝑘 = 1, 2, 3, … , 𝑁𝑁
• [nk] are Gaussian, since nk can be viewed as the sampled output of a
linear filter excited by Gaussian noise.
• Mean value of nk:
𝑇𝑇
• 𝐸𝐸[𝑛𝑛𝑘𝑘 ] = ∫0 𝐸𝐸[𝑛𝑛 𝑡𝑡 ]𝑓𝑓𝑘𝑘 𝑡𝑡 𝑑𝑑𝑑𝑑 = 0

7
Covariances of Noise
• Covariances
𝑇𝑇 𝑇𝑇
• 𝐸𝐸 𝑛𝑛𝑘𝑘 𝑛𝑛𝑚𝑚 = ∫0 ∫0 𝐸𝐸 𝑛𝑛 𝑡𝑡 𝑛𝑛 𝜏𝜏 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑓𝑓𝑚𝑚 𝜏𝜏 𝑑𝑑𝑑𝑑𝑑𝑑𝜏𝜏
• n(t) is wide-sense stationary process.
𝑁𝑁𝑜𝑜
• 𝐸𝐸 𝑛𝑛 𝑡𝑡 𝑛𝑛 𝜏𝜏 = 𝑅𝑅𝑁𝑁 𝑡𝑡 − 𝜏𝜏 = 𝛿𝛿(𝑡𝑡 − 𝜏𝜏)
2
−1 𝑁𝑁𝑜𝑜 𝑁𝑁𝑜𝑜
• 𝑅𝑅𝑁𝑁 𝜏𝜏 = 𝐹𝐹 = 𝛿𝛿(𝜏𝜏)
2 2
𝑇𝑇 𝑇𝑇 𝑁𝑁𝑜𝑜
• 𝐸𝐸 𝑛𝑛𝑘𝑘 𝑛𝑛𝑚𝑚 = ∫0 ∫0 𝛿𝛿(𝑡𝑡 − 𝜏𝜏) 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑓𝑓𝑚𝑚 𝜏𝜏 𝑑𝑑𝑑𝑑𝑑𝑑𝜏𝜏
2

• Sifting property: ∫−∞ 𝛿𝛿 𝑡𝑡 − 𝑎𝑎 𝑓𝑓 𝑡𝑡 𝑑𝑑𝑑𝑑 = 𝑓𝑓(𝑎𝑎)
8
Covariances of Noise
𝑁𝑁𝑜𝑜 𝑇𝑇 𝑇𝑇 1, 𝑘𝑘 = 𝑚𝑚
• 𝐸𝐸 𝑛𝑛𝑘𝑘 𝑛𝑛𝑚𝑚 = ∫ 𝑓𝑓 𝑡𝑡 𝑓𝑓𝑚𝑚 𝑡𝑡 𝑑𝑑𝑑𝑑, ∫0 𝑓𝑓𝑘𝑘 𝑡𝑡 𝑓𝑓𝑚𝑚 𝑡𝑡 𝑑𝑑𝑑𝑑 = �
2 0 𝑘𝑘 0, 𝑘𝑘 ≠ 𝑚𝑚
𝑁𝑁𝑜𝑜
𝑁𝑁𝑜𝑜 , 𝑘𝑘 = 𝑚𝑚
• 𝐸𝐸 𝑛𝑛𝑘𝑘 𝑛𝑛𝑚𝑚 = 𝛿𝛿 = �2
2 𝑚𝑚𝑚𝑚
0, 𝑘𝑘 ≠ 𝑚𝑚
• The N noise components [nk] are zero-mean, uncorrelated Gaussian
random variables with a common variance
𝑁𝑁𝑜𝑜
• 𝜎𝜎𝑛𝑛2 =
2

9
Correlator Output
• The correlator outputs [rk] conditioned on the mth signal being
transmitted are Gaussian random variables with mean
• 𝐸𝐸(𝑟𝑟𝑘𝑘 ) = 𝐸𝐸(𝑠𝑠𝑚𝑚𝑘𝑘 ) + 𝐸𝐸 𝑛𝑛𝑘𝑘 = 𝑠𝑠𝑚𝑚𝑘𝑘 + 0 = 𝑠𝑠𝑚𝑚𝑘𝑘
• 𝑛𝑛𝑘𝑘 has Gaussian pdf. 𝑟𝑟𝑘𝑘 has Gaussian pdf with mean 𝑠𝑠𝑚𝑚𝑘𝑘 and variance
𝑁𝑁𝑜𝑜
• 𝜎𝜎𝑟𝑟2 = 𝜎𝜎𝑛𝑛2 =
2
• {nk} are uncorrelated Gaussian random variables. Then, {nk} are
statistically independent Gaussian random variables.
• 𝒓𝒓 = 𝑟𝑟1 , 𝑟𝑟2 , … , 𝑟𝑟𝑁𝑁 = received vector

10
Conditional pdf
• 𝑝𝑝 𝒓𝒓|𝒔𝒔𝑚𝑚 = ∏𝑁𝑁
𝑘𝑘=1 𝑝𝑝 𝑟𝑟𝑘𝑘 |𝑠𝑠𝑚𝑚𝑘𝑘 , 𝑚𝑚 = 1, 2, … , 𝑀𝑀
2
1 𝑟𝑟𝑘𝑘 −𝑠𝑠𝑚𝑚𝑘𝑘
• 𝑝𝑝 𝑟𝑟𝑘𝑘 |𝑠𝑠𝑚𝑚𝑘𝑘 = 𝑒𝑒𝑒𝑒𝑒𝑒 − 𝑁𝑁 , 𝑘𝑘 = 1, 2, … , 𝑁𝑁
𝑁𝑁
2𝜋𝜋 2𝑜𝑜 2 2𝑜𝑜
2
1 𝑟𝑟𝑘𝑘 −𝑠𝑠𝑚𝑚𝑘𝑘
• 𝑝𝑝 𝒓𝒓|𝒔𝒔𝑚𝑚 = 𝑁𝑁 𝑒𝑒𝑒𝑒𝑒𝑒 − ∑𝑁𝑁
𝑘𝑘=1 , 𝑚𝑚 = 1, 2, … , 𝑀𝑀
𝑁𝑁𝑜𝑜
𝜋𝜋𝑁𝑁𝑜𝑜 2
• [r1, r2, ….., rN] are sufficient statistics. No additional relevant
information can be extracted from the remaining noise process n′(t).
n′(t) are uncorrelated with the N correlator outputs {rk}, that is,

11
Conditional pdf
• 𝐸𝐸 𝑛𝑛𝑛(𝑡𝑡)𝑛𝑛𝑘𝑘 = 𝐸𝐸 𝑛𝑛 𝑡𝑡 − ∑𝑁𝑁
𝑗𝑗=1 𝑛𝑛𝑗𝑗 𝑓𝑓𝑗𝑗 𝑡𝑡 𝑛𝑛𝑘𝑘
𝑇𝑇
• 𝑛𝑛𝑘𝑘 = ∫0 𝑛𝑛(𝜏𝜏)𝑓𝑓𝑘𝑘 𝜏𝜏 𝑑𝑑𝜏𝜏
𝑇𝑇
• 𝐸𝐸 𝑛𝑛𝑛(𝑡𝑡)𝑛𝑛𝑘𝑘 = ∫0 𝐸𝐸[𝑛𝑛 𝑡𝑡 𝑛𝑛 𝜏𝜏 ]𝑓𝑓𝑘𝑘 𝜏𝜏 𝑑𝑑𝜏𝜏 − ∑𝑁𝑁
𝑗𝑗=1 𝐸𝐸[𝑛𝑛𝑗𝑗 𝑛𝑛𝑘𝑘 ]𝑓𝑓𝑗𝑗 𝑡𝑡
𝑁𝑁𝑜𝑜
• 𝐸𝐸 𝑛𝑛 𝑡𝑡 𝑛𝑛 𝜏𝜏 = 𝑅𝑅𝑁𝑁 𝑡𝑡 − 𝜏𝜏 = 𝛿𝛿(𝑡𝑡 − 𝜏𝜏),
2
𝑁𝑁𝑜𝑜
𝑁𝑁𝑜𝑜 , 𝑗𝑗 = 𝑘𝑘 ∞
• 𝐸𝐸 𝑛𝑛𝑗𝑗 𝑛𝑛𝑘𝑘 = 𝛿𝛿𝑗𝑗𝑘𝑘 = � 2 , ∫−∞ 𝛿𝛿 𝑡𝑡 − 𝑎𝑎 𝑓𝑓 𝑡𝑡 𝑑𝑑𝑑𝑑 = 𝑓𝑓(𝑎𝑎)
2
0, 𝑗𝑗 ≠ 𝑘𝑘
𝑇𝑇 𝑁𝑁𝑜𝑜 𝑁𝑁 𝑁𝑁𝑜𝑜
• 𝐸𝐸 𝑛𝑛𝑛(𝑡𝑡)𝑛𝑛𝑘𝑘 = ∫0 2 𝛿𝛿(𝑡𝑡 − 𝜏𝜏) 𝑓𝑓𝑘𝑘 𝜏𝜏 𝑑𝑑𝜏𝜏 − ∑𝑗𝑗=1 𝛿𝛿𝑗𝑗𝑘𝑘 𝑓𝑓𝑗𝑗 𝑡𝑡
2
12
Conditional pdf
𝑁𝑁𝑜𝑜 𝑁𝑁𝑜𝑜
• 𝐸𝐸 𝑛𝑛𝑛(𝑡𝑡)𝑛𝑛𝑘𝑘 = 𝑓𝑓 𝑡𝑡 − 𝑓𝑓 𝑡𝑡 = 0
2 𝑘𝑘 2 𝑘𝑘
• n′(t) and {rk} are Gaussian and uncorrelated. They are statistically
independent.
• n′(t) belongs to other dimensions, N+1, N+2, …...
• n′(t) does not contain any information relevant to the decision as to
which signal waveform was transmitted.
• n′(t) can be ignored.

13
EXAMPLE
• M-ary PAM: Dimension N =1 a

• g(t) = a, 0 ≤ t < T. (Square pulse)

g(t)
0

• Energy of g(t) = Eg = a2T 0


t
T

• Basis function
1/sqrt(T)
1 1
• 𝑓𝑓 𝑡𝑡 = 𝑔𝑔 𝑡𝑡 = ,0 ≤ 𝑡𝑡 < 𝑇𝑇

f(t)
𝐸𝐸𝑔𝑔 𝑇𝑇 0
0 T

• Received vector r t

𝑇𝑇 1 𝑇𝑇 1 𝑇𝑇
• 𝑟𝑟 = ∫0 𝑟𝑟 𝑡𝑡 𝑓𝑓 𝑡𝑡 𝑑𝑑𝑑𝑑 = ∫0 𝑟𝑟 𝑡𝑡 𝑑𝑑𝑑𝑑 = ∫0 𝑠𝑠𝑚𝑚 𝑡𝑡 + 𝑛𝑛(𝑡𝑡) 𝑑𝑑𝑑𝑑
𝑇𝑇 𝑇𝑇
1 𝑇𝑇 1 𝑇𝑇
• = ∫0 𝑠𝑠𝑚𝑚 𝑡𝑡 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛(𝑡𝑡)𝑑𝑑𝑑𝑑 = 𝑠𝑠𝑚𝑚 + 𝑛𝑛
𝑇𝑇 𝑇𝑇

14
EXAMPLE
• 𝐸𝐸 𝑛𝑛 = 0
• Variance of n is
1 𝑇𝑇 𝑇𝑇
• 𝑣𝑣𝑣𝑣𝑣𝑣 𝑛𝑛 = 𝜎𝜎𝑛𝑛2 = 𝐸𝐸 ∫0 ∫0
𝑛𝑛 𝑡𝑡 𝑛𝑛 𝜏𝜏 𝑑𝑑𝑑𝑑𝑑𝑑𝜏𝜏
𝑇𝑇
1 𝑇𝑇 𝑇𝑇 1 𝑇𝑇 𝑇𝑇 𝑁𝑁𝑜𝑜
•= ∫ ∫ 𝐸𝐸[𝑛𝑛 𝑡𝑡 𝑛𝑛 𝜏𝜏 ]𝑑𝑑𝑑𝑑𝑑𝑑𝜏𝜏 = 𝑇𝑇 ∫0 ∫0 2 𝛿𝛿(𝑡𝑡 − 𝜏𝜏)𝑑𝑑𝑑𝑑𝑑𝑑𝜏𝜏
𝑇𝑇 0 0
1 𝑁𝑁𝑜𝑜 𝑇𝑇 𝑁𝑁𝑜𝑜 𝑇𝑇
•= ∫0
1𝑑𝑑𝑑𝑑 = , ∫0 𝛿𝛿(𝑡𝑡 − 𝜏𝜏)𝑑𝑑𝜏𝜏 = 1
𝑇𝑇 2 2
1 𝑟𝑟−𝑠𝑠𝑚𝑚 2
• 𝑝𝑝 𝑟𝑟|𝑠𝑠𝑚𝑚 = 𝑒𝑒𝑒𝑒𝑒𝑒 − 𝑁𝑁
𝑁𝑁
2𝜋𝜋 2𝑜𝑜 2 2𝑜𝑜

15
Matched Filter Demodulator

y1(t)
f1(T-t) r1
t=T

r(t) y2(t)
f2(T-t) r2
t=T

y3(t)
f3(T-t) rN
t=T

16
Matched Filter Demodulator
• Matched filter demodulator has the same performance as the
correlation demodulator.
• The received signal r(t) is applied to N linear filters with impulse
responses f1(T-t), f2(T-t), ….., fN(T-t). Let the output of the filters be
y1(t), y2(t), ….., yN(t) respectively. The outputs are sampled at t = T to
produce r1, r2, ….., rN.
• ℎ𝑘𝑘 𝑡𝑡 = 𝑓𝑓𝑘𝑘 𝑇𝑇 − 𝑡𝑡 , 0 ≤ 𝑡𝑡 < 𝑇𝑇
• 𝑓𝑓𝑘𝑘 (𝑡𝑡) are the N basis functions
𝑇𝑇 𝑇𝑇
• 𝑦𝑦𝑘𝑘 𝑡𝑡 = ∫0 𝑟𝑟(𝜏𝜏)ℎ𝑘𝑘 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝜏𝜏 = ∫0 𝑟𝑟(𝜏𝜏)𝑓𝑓𝑘𝑘 𝑇𝑇 − 𝑡𝑡 + 𝜏𝜏 𝑑𝑑𝜏𝜏

17
Matched Filter Demodulator
• Sample at t = T.
𝑇𝑇 𝑇𝑇
• 𝑦𝑦𝑘𝑘 𝑇𝑇 = ∫0 𝑟𝑟 𝜏𝜏 𝑓𝑓𝑘𝑘 𝑇𝑇 − 𝑇𝑇 + 𝜏𝜏 𝑑𝑑𝜏𝜏 = ∫0 𝑟𝑟 𝜏𝜏 𝑓𝑓𝑘𝑘 𝜏𝜏 𝑑𝑑𝜏𝜏 , 𝑘𝑘 = 1,2, … , 𝑁𝑁
• This is the same result as that of correlator demodulator.
• A filter with impulse response h(t) = s(T-t) is called the matched filter
to the signal s(t).
• Flip s(t) to get s(-t).
• shift s(-t) to the right by T to get s(-(t-T)) = s(T-t)

18
EXAMPLE
𝐴𝐴
• 𝑠𝑠 𝑡𝑡 = 𝑡𝑡, 0 ≤ 𝑡𝑡 < 𝑇𝑇 A

𝑇𝑇

s(t)
𝐴𝐴
• 𝑠𝑠 −𝑡𝑡 =
0

− 𝑡𝑡, −𝑇𝑇 ≤ 𝑡𝑡 < 0 0 T

𝑇𝑇 t

𝐴𝐴
• ℎ 𝑡𝑡 = 𝑠𝑠 𝑇𝑇 − 𝑡𝑡 = − 𝑡𝑡 − 𝑇𝑇 A

𝑇𝑇

s(-t)
𝐴𝐴 0

•= − 𝑡𝑡 + 𝐴𝐴, 0 ≤ 𝑡𝑡 < 𝑇𝑇
-T 0
t
𝑇𝑇
A

s(T-t)
0
0 T
t

19
Matched Filter

s(t) y(t)
h(t) = s(T-t)

𝑡𝑡 𝑡𝑡
• 𝑦𝑦 𝑡𝑡 = ∫0 𝑠𝑠 𝜏𝜏 ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑 = ∫0 𝑠𝑠 𝜏𝜏 𝑠𝑠 𝑇𝑇 − 𝑡𝑡 + 𝜏𝜏 𝑑𝑑𝑑𝑑 = 𝑅𝑅𝑠𝑠 (𝑇𝑇 − 𝑡𝑡)
• 𝑦𝑦 𝑡𝑡 is the time autocorrelation function of s(t).
𝑇𝑇 2
• 𝑦𝑦 𝑇𝑇 = ∫0 𝑠𝑠 𝜏𝜏 𝑑𝑑𝜏𝜏 = 𝑅𝑅𝑠𝑠 0 = energy of s(t)
• 𝑅𝑅𝑠𝑠 0 ≥ 𝑅𝑅𝑠𝑠 𝜏𝜏
• The sample at t = T is the maximum value of autocorrelation of s(t).
20
Matched Filter

s(t) r(t) y(t) y(T)


h(t) = s(T-t)
t=T

n(t)

• Under AWGN, matched filter provides maximum output SNR.


𝑡𝑡 𝑡𝑡
• 𝑦𝑦 𝑡𝑡 = ∫0 𝑟𝑟 𝜏𝜏 ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑 = ∫0 [𝑠𝑠 𝜏𝜏 + 𝑛𝑛(𝜏𝜏)]ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑
𝑡𝑡 𝑡𝑡
•= ∫0 𝑠𝑠 𝜏𝜏 ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛(𝜏𝜏)ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑
𝑡𝑡 𝑡𝑡
•= ∫0 𝑠𝑠 𝜏𝜏 𝑠𝑠 𝑇𝑇 − 𝑡𝑡 + 𝜏𝜏 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛 𝜏𝜏 𝑠𝑠 𝑇𝑇 − 𝑡𝑡 + 𝜏𝜏 𝑑𝑑𝑑𝑑
21
Matched Filter
𝑇𝑇 𝑇𝑇
• 𝑦𝑦 𝑇𝑇 = ∫0 𝑠𝑠 𝜏𝜏 𝑠𝑠 𝑇𝑇 − 𝑇𝑇 + 𝜏𝜏 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛 𝜏𝜏 𝑠𝑠 𝑇𝑇 − 𝑇𝑇 + 𝜏𝜏 𝑑𝑑𝑑𝑑
𝑇𝑇 2 𝑇𝑇
• 𝑦𝑦 𝑇𝑇 = ∫0 𝑠𝑠 𝜏𝜏 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛 𝜏𝜏 𝑠𝑠
𝜏𝜏 𝑑𝑑𝑑𝑑 = 𝑦𝑦𝑠𝑠 𝑇𝑇 + 𝑦𝑦𝑛𝑛 (𝑇𝑇)
• Find h(t) that maximizes (SNR)o.
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑦𝑦𝑠𝑠2 (𝑇𝑇)
• (𝑆𝑆𝑆𝑆𝑆𝑆)𝑜𝑜 = =
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝐸𝐸 𝑦𝑦𝑛𝑛2 (𝑇𝑇)
• 𝐸𝐸 𝑦𝑦𝑛𝑛2 (𝑇𝑇) = variance of the noise term
𝑡𝑡 𝑡𝑡
• 𝑦𝑦 𝑡𝑡 = ∫0 𝑠𝑠 𝜏𝜏 ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛(𝜏𝜏)ℎ 𝑡𝑡 − 𝜏𝜏 𝑑𝑑𝑑𝑑
𝑇𝑇 𝑇𝑇
• 𝑦𝑦 𝑇𝑇 = ∫0 𝑠𝑠 𝜏𝜏 ℎ 𝑇𝑇 − 𝜏𝜏 𝑑𝑑𝑑𝑑 + ∫0 𝑛𝑛(𝜏𝜏)ℎ 𝑇𝑇 − 𝜏𝜏 𝑑𝑑𝑑𝑑 = 𝑦𝑦𝑠𝑠 𝑇𝑇 + 𝑦𝑦𝑛𝑛 (𝑇𝑇)
22
Matched Filter
𝑇𝑇 𝑇𝑇
• 𝐸𝐸 𝑦𝑦𝑛𝑛2 (𝑇𝑇) = 𝐸𝐸 ∫0 𝑛𝑛(𝜏𝜏)ℎ 𝑇𝑇 − 𝜏𝜏 𝑑𝑑𝑑𝑑 ∫0 𝑛𝑛(𝑡𝑡)ℎ 𝑇𝑇 − 𝑡𝑡 𝑑𝑑𝑡𝑡
𝑇𝑇 𝑇𝑇
• = ∫0 ∫0 𝐸𝐸 𝑛𝑛 𝜏𝜏 𝑛𝑛(𝑡𝑡) ℎ 𝑇𝑇 − 𝜏𝜏 ℎ 𝑇𝑇 − 𝑡𝑡 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
𝑁𝑁𝑜𝑜 𝑇𝑇 𝑇𝑇
• = ∫0 ∫0 𝛿𝛿(𝑡𝑡 − 𝜏𝜏) ℎ 𝑇𝑇 − 𝜏𝜏 ℎ 𝑇𝑇 − 𝑡𝑡 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑
2
𝑇𝑇
• ∫0 𝛿𝛿(𝑡𝑡 − 𝜏𝜏) ℎ 𝑇𝑇 − 𝜏𝜏 𝑑𝑑𝑑𝑑 = ℎ 𝑇𝑇 − 𝑡𝑡
2 𝑁𝑁𝑜𝑜 𝑇𝑇 2
• 𝐸𝐸 𝑦𝑦𝑛𝑛 (𝑇𝑇) = ∫0 ℎ 𝑇𝑇 − 𝑡𝑡 𝑑𝑑𝑑𝑑
2
𝑇𝑇 2
• ∫0 ℎ 𝑇𝑇 − 𝑡𝑡 𝑑𝑑𝑑𝑑 = energy of h(t) = energy of s(t)

23
Matched Filter
𝑇𝑇
• 𝑦𝑦𝑠𝑠 𝑇𝑇 = ∫0 𝑠𝑠 𝜏𝜏 ℎ 𝑇𝑇 − 𝜏𝜏 𝑑𝑑𝑑𝑑
𝑇𝑇 2
𝑦𝑦𝑠𝑠2 (𝑇𝑇) ∫0 𝑠𝑠 𝜏𝜏 ℎ 𝑇𝑇−𝜏𝜏 𝑑𝑑𝑑𝑑
• (𝑆𝑆𝑆𝑆𝑆𝑆)𝑜𝑜 = = 𝑁𝑁𝑜𝑜 𝑇𝑇 2
𝐸𝐸 𝑦𝑦𝑛𝑛2 (𝑇𝑇) ∫ ℎ 𝑇𝑇−𝑡𝑡 𝑑𝑑𝑑𝑑
2 0
• Maximize the numerator while the denominator is held constant. Use
Cauchy-Schwarz inequality.
∞ 2 ∞ ∞
• ∫−∞ 𝑔𝑔1 𝑡𝑡 𝑔𝑔2 𝑡𝑡 𝑑𝑑𝑡𝑡 ≤ ∫−∞ 𝑔𝑔12 𝑡𝑡 𝑑𝑑𝑑𝑑 ∫−∞ 𝑔𝑔22 𝑡𝑡 𝑑𝑑𝑑𝑑
• Equality holds when 𝑔𝑔1 𝑡𝑡 = 𝑐𝑐𝑔𝑔2 𝑡𝑡
• Let 𝑔𝑔1 𝑡𝑡 = ℎ 𝑡𝑡 , 𝑔𝑔2 𝑡𝑡 = 𝑠𝑠(𝑇𝑇 − 𝑡𝑡).
24
Matched Filter
• SNR is maximized when
• ℎ 𝑡𝑡 = 𝑐𝑐𝑐𝑐(𝑇𝑇 − 𝑡𝑡)
• ℎ 𝑇𝑇 − 𝑡𝑡 = 𝑐𝑐𝑐𝑐(𝑡𝑡)
• The scale factor c2 drops out or set c = 1.
• The maximum SNR is
𝑇𝑇 𝑇𝑇
∫0 ℎ2 𝜏𝜏 𝑑𝑑𝜏𝜏 ∫0 𝑠𝑠 2 𝑇𝑇−𝜏𝜏 𝑑𝑑𝜏𝜏 2 𝑇𝑇 2 2 𝑇𝑇 2 2𝐸𝐸
• (𝑆𝑆𝑆𝑆𝑆𝑆)𝑜𝑜,𝑚𝑚𝑚𝑚𝑚𝑚 = 𝑁𝑁𝑜𝑜 𝑇𝑇 2 = ∫0
𝑠𝑠 𝑇𝑇 − 𝜏𝜏 𝑑𝑑𝑑𝑑 = ∫0
𝑠𝑠 𝑡𝑡 𝑑𝑑𝑑𝑑 =
∫ ℎ 𝑇𝑇−𝑡𝑡 𝑑𝑑𝑑𝑑 𝑁𝑁𝑜𝑜 𝑁𝑁𝑜𝑜 𝑁𝑁𝑜𝑜
2 0

• where E is the energy of s(t). The maximum output SNR depends on


energy of s(t), not on the shape of s(t).
25

You might also like