Digital Communications

MATCHED FILTER Dr/ Amr Wageeh

Matched Filter

Threshold

ro(t) 𝑚
ෝ

ro(T)

• Filter that maximize instantaneous SNR at the output.

𝑆1𝑜 𝑇 −𝑆2𝑜 (𝑇) 2

• SNR = ζ2 =
𝜎𝑛2

• Filter is matched to input signal

• Design on difference signal g(t) = S1(t)- S2(t)

SNR = ζ2 is max

ro(t)
Threshold
hMF(t) ro(T)

HMF(f)
• Design on difference signal g(t) = S1(t)- S2(t)

SNR = ζ2 is max

g(t)+ N(t) go(t)+ No(t)

Threshold
hMF(t)

HMF(f) go(T)+ No(T)

Notes
• Dealing with filters: for deterministic signal

x(t) y(t)
h(t)

H(f)

y(t) = x(t)*h(t)
Y(f) = X(f).H(f)
Notes
• Dealing with filters: for random signal

n(t) no(t)
h(t)
Sn(f) So(f)

H(f)

∞
Total power at output σ no= ‫׬‬−∞ So(f)
2 𝑑𝑓
Cauchy Shuartz inequlaity
∞ ∞ ∞
‫׬‬−∞ 𝑓1 (𝑥)𝑓2 (𝑥) 𝑑𝑥 2 ≤ ‫׬‬−∞ 𝑓1 (𝑥) 2 𝑑𝑥 . ‫׬‬−∞ 𝑓2 (𝑥) 2 𝑑𝑥

if f1(x) = const f2*(x)

For Deterministic signal

• go(t) = F-1 {Go(f)}

∞
= ‫׬‬−∞ 𝐺𝑜 𝑓 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓
∞
= ‫׬‬−∞ 𝐺 𝑓 𝐻(𝑓) 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓

∞
go(T) = ‫׬‬−∞ 𝐺 𝑓 𝐻(𝑓) 𝑒 𝑗2𝜋𝑓𝑇 𝑑𝑓
For Random Signal
∞
• σ2n= ‫׬‬−∞ Sn(f) |𝐻(𝑓)|2 𝑑𝑓

2
𝑔𝑜 (𝑇)
• SNR = ζ2 =
σ2n
∞
‫׬‬−∞ 𝐺 𝑓 𝐻(𝑓) 𝑒 𝑗2𝜋𝑓𝑇 𝑑𝑓 2

=
‫׬‬−∞ Sn(f) |𝐻(𝑓)|2 𝑑𝑓
∞

∞ 𝐺 𝑓
‫׬‬−∞ 𝑆𝑛 (𝑓) 𝐻(𝑓) 𝑒 𝑗2𝜋𝑓𝑇 𝑑𝑓 2
𝑆𝑛 (𝑓)
=
‫׬‬−∞ Sn(f) |𝐻(𝑓)|2 𝑑𝑓
∞

‫׬‬−∞ Sn(f) |𝐻(𝑓)|2 𝑑𝑓

∞ 𝐺 𝑓
𝑒 𝑗2𝜋𝑓𝑇 𝑑𝑓 2
𝑆𝑛 (𝑓) ∞ 𝐺 𝑓 ∞ 𝐺 𝑓
≤ = ‫׬‬−∞ 𝑒 𝑗2𝜋𝑓𝑇 2 df = ‫׬‬−∞ 2 df
∞
‫׬‬−∞ Sn(f) |𝐻(𝑓)|2 𝑑𝑓 𝑆𝑛 (𝑓) 𝑆𝑛 (𝑓)
Matched filter design
∗
𝐺 𝑓
• 𝑆𝑛 (𝑓) 𝐻(𝑓) = const 𝑒 −𝑗2𝜋𝑓𝑇
𝑆𝑛 (𝑓)

∗
𝐺 𝑓 −𝑗2𝜋𝑓𝑇
• 𝐻𝑀𝐹 (𝑓) = const 𝑒
𝑆𝑛(𝑓)
∗ ∗
[𝑆1 𝑓 −𝑆2 𝑓 ] −𝑗2𝜋𝑓𝑇
• 𝐻𝑀𝐹 (𝑓) = const 𝑒
𝑆𝑛(𝑓)
Summary in MF

∞ 𝐺2 𝑓
• SNR = ζ2= ‫׬‬−∞ 𝑆 (𝑓) df
𝑛

∗ ∗
[𝑆1 𝑓 −𝑆2 𝑓 ] −𝑗2𝜋𝑓𝑇
• 𝐻𝑀𝐹 (𝑓) = const 𝑒
𝑆𝑛(𝑓)

ζ
𝑃𝑒 = 𝑄( )
2
Special case: in AWGN

𝑁𝑜
• 𝑆𝑛 𝑓 =
2
In case of binary Comm and equiprobable system

• ℎ𝑀𝐹 = 𝑆1 𝑇 − 𝑡 − 𝑆2 𝑇 − 𝑡 .

𝐸1 −𝐸2
• 𝑉𝑡ℎ =
2

𝐸𝑔
• 𝑃𝑒 |𝑚𝑖𝑛 = 𝑄( )
2𝑁𝑜
Proof (cont)
𝐸1 − 𝐸2
𝑉𝑡ℎ =
2

𝑆1𝑜 𝑇 +𝑆2𝑜 (𝑇)

• 𝑉𝑡ℎ =
2
∞
• 𝑆10 𝑡 = 𝑓 −1 𝑆10 𝑓 = ‫׬‬−∞ 𝑆10 (𝑓)𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓
∞
= ‫׬‬−∞ 𝑆1 𝑓 𝐻𝑀𝐹 𝑓 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓
∞
• 𝑆10 𝑇 = ‫׬‬−∞ 𝑆1 𝑓 𝐻𝑀𝐹 𝑓 𝑒 𝑗2𝜋𝑓𝑇 𝑑𝑓
∞
= ‫׬‬−∞ 𝑆1 𝑓 [ 𝑆1∗ 𝑓 − 𝑆2∗ 𝑓 ] 𝑒 −𝑗2𝜋𝑓𝑇 𝑒 𝑗2𝜋𝑓𝑇 𝑑𝑓
∞ ∞
= ‫׬‬−∞ 𝑆1 (𝑓) 2 df - ‫׬‬−∞ 𝑆1 𝑓 𝑆2∗ 𝑓 df
= E1 – E12
By the same steps for S20 (T)
• 𝑆10 𝑇 = E12– E2
𝐸1 −𝐸2
• 𝑉𝑡ℎ =
2
Parallel realization of MF
• ℎ𝑀𝐹 = 𝑆1 𝑇 − 𝑡 − 𝑆2 𝑇 − 𝑡

𝑆1 𝑇 − 𝑡
Threshold
r(t)
𝐸1 − 𝐸2
=
2
-
𝑆2 𝑇 − 𝑡
Notes 𝑃𝑒 |= 𝑄(
𝐸𝑔
)
2𝑁𝑜

• 1. Prob of error depend on energy not waveform

𝑆1 𝑡 𝑆1 𝑡

1
T t t
T

𝑆2 𝑡 𝑆2 𝑡

t t
T/2 T/4 3T/4
Notes
• Why MF is better ?

• MF benefits from entire Eg.

• Magnitude response of MF is equal to spectrum of signal.

• Choice of T

• 0 ≤ T ≤ Tb

• Causal: T must have MF causal

Optimizing Pe
• How to design S1(t) and S2(t).

• Cross correlation factor:

1 ∞ 1
• 𝜌12 =
𝐸1 𝐸2
‫׬‬−∞ 𝑆1 (𝑡)𝑆2 (𝑡)𝑑𝑡 = 𝐸1 𝐸2
E12 -1 ≤ 𝜌12 ≤ 1

• Represents the relation between signals S1(t) and S2(t).

𝐸1 +𝐸2 −2 𝐸1 𝐸2 𝜌12
• 𝑃𝑒 = Q( )
2 𝑁𝑜
 𝐸𝑔
𝑃𝑒 |= 𝑄( )
2𝑁𝑜

• Eg is the energy of g(t).

∞
• 𝐸𝑔 = ‫׬‬−∞ 𝑔(𝑡) 2 dt
∞
= ‫׬‬−∞ S1(t)−S2(t) 2 dt
∞ ∞ ∞
= ‫׬‬−∞ S1(t) 2 dt + ‫׬‬−∞ S2(t) 2 dt – 2 ‫׬‬−∞ S1(t)S2(t) dt

= E1 + E2 – 2 E12
Special cases for 𝜌12

• 𝜌12 = 1 fully correlated

• 𝜌12 = -1 antipodal signal
• 𝜌12 = 0
1. 𝜌12 = 1 fully correlated
• S1(t) = const S2(t)= +αS2(t). Two signals have the same shape, polarity , with different amplitude

S1(t)

1
t
T

S2(t)
𝐸1 +𝐸2 −2 𝐸1 𝐸2 𝐸1 − 𝐸2
1/ α 𝑃𝑒 = Q( )= Q ( )
2 𝑁𝑜 2 𝑁𝑜
t
T
2. 𝜌12 = -1 antipodal signal
• S1(t) = const S2(t)= -αS2(t). Two signals have the same shape, different polarity

S1(t)

1
t
T

S2(t)
𝐸1 +𝐸2 +2 𝐸1 𝐸2 𝐸1 + 𝐸2
𝑃𝑒 = Q( )= Q ( )
2 𝑁𝑜 2 𝑁𝑜
T t

-α
3. 𝜌12 = 0
∞
‫׬‬−∞ 𝑆1 (𝑡)𝑆2 (𝑡)𝑑𝑡= 0

𝐸1 +𝐸2
𝑃𝑒 = Q( )
2 𝑁𝑜
Correlator implementation
• Output of MF ro(T) can be implemented using correlator (fixed components).
𝑇𝑏
• 𝑟𝑜 (𝑇) = ‫𝑡𝑑 𝑡 𝑔 𝑡 𝑟 𝑜׬‬

X
r(t) 𝑇𝑏 𝑟𝑜 (𝑇) r(t) 𝑟𝑜 (𝑡) 𝑟𝑜 (𝑇)
න Equivalent to g(T-t)
𝑜

g(t)
NWGN case (whitening filter)
n(t) 𝑛(𝑡)
෤
HW(f)
NWGN WGN
PSD = Sn(f) PSD = No/2

• 𝑆𝑛𝑜 𝑓 = 𝑆𝑛 𝑓 . 𝐻𝑊 (𝑓) 2

• No/2 = 𝑆𝑛 𝑓 . 𝐻𝑊 (𝑓) 2
𝑆ሚ1(t)=S1(t)*hw(t)
𝑁𝑜ൗ
• 𝐻𝑊 (𝑓) = 2
𝑆ሚ1 𝑇 − 𝑡
𝑆𝑛 𝑓

r(t)
HW(f)
S(t)+n(t) 𝑆ሚ (𝑡) + 𝑛(𝑡)
෤ 𝑆ሚ2(t)=S2(t)*hw(t)
-
𝑆ሚ2 𝑇 − 𝑡
Proofs
Proof
• Note:
• X(-t) X*(f)
• X(t-T) X(f) e-j2πfT
∞ ∞ ∗
• E = ‫׬‬−∞ 𝑆1 (𝑡). 𝑆2 (𝑡) dt = ‫׬‬−∞ 𝑆1 (𝑓). 𝑆2 (𝑓) df cross energy (correlation between signals).

∗ ∗
[𝑆1 𝑓 −𝑆2 𝑓 ] −𝑗2𝜋𝑓𝑇
• 𝐻𝑀𝐹 (𝑓) = const 𝑒
𝑆𝑛(𝑓)
𝑁𝑜
• at AWGN 𝑆𝑛 𝑓 =
2
∗ ∗
[𝑆1 𝑓 −𝑆2 𝑓 ] −𝑗2𝜋𝑓𝑇
• 𝐻𝑀𝐹 (𝑓) = const 𝑁𝑜ൗ 𝑒
2
𝑐𝑜𝑛𝑠𝑡 ∗ ∗
• = 𝑁𝑜ൗ [𝑆1 𝑓 − 𝑆2 𝑓 ] 𝑒 −𝑗2𝜋𝑓𝑇
2

• = S1(-(t-T) - S2(-(t-T)
Proof (Cont)
ζ
• 𝑃𝑒 = 𝑄( )
2

∞ 𝐺2 𝑓 2 ∞ 2
• SNRo = ζ2= ‫׬‬−∞ 𝑆 (𝑓) df = ‫׬‬ 𝐺2(𝑓) df = E
𝑛
𝑁𝑜 −∞ 𝑁𝑜 g

1 2𝐸𝑔 𝐸𝑔
• Pe = Q( ) = Q( )
2 𝑁𝑜 2𝑁𝑜
Correlator implementation
• 𝑟𝑜 (𝑡) =r(t)*hMF(t)

= r(t)*g(T-t)
∞
= ‫׬‬−∞ 𝑟 𝜏 . 𝑔 𝑇 − 𝑡 − 𝜏 𝑑𝜏
∞
• 𝑟𝑜 (𝑡) = ‫׬‬−∞ 𝑟 𝜏 . 𝑔 𝑇 − 𝑇 − 𝜏 𝑑𝜏
∞ ∞
= ‫׬‬−∞ 𝑟 𝜏 . 𝑔 𝜏 𝑑𝜏 = ‫׬‬−∞ 𝑟 𝑡 . 𝑔 𝑡 𝑑𝑡