tsks01&digital&communication recall:&pulsemamplitude ......
TRANSCRIPT
TSKS01&Digital&CommunicationLecture&6
Dispersive&channels,&ML&sequence&detection
Emil&Björnson
Department&of&Electrical&Engineering&(ISY)Division&of&Communication&Systems
Recall:&Nyquist&Criterion
Recall:&PulseMAmplitude&Modulation&(PAM):
! " =$% & '(" − &*)�
-
Receiver&filtering&. " , Γ(1):
. ∗ ! " =$% & (. ∗ ')(" − &*)�
-
Nyquist&criterion:
$ Γ 1 −3*
4 1 −3*
5
6785
= Constant
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 2
Called:&'(") and&.(") are&Nyquist(together
Bandwidth more than 1/(2*):Many Nyquist&pulses
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 3
' " BandB. " BareBFGHIJKLBLMNOLPOQBif $ Γ 1 −3* 4 1 −
3*
5
6785
isBconstant
$ Γ 1 −3*
4 1 −3*
5
6785
Γ 1 4 1
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 4
OneMway&Digital&Communication&System
Analog(channel
Source
Destination
Channel(encoder Modulator
Channel(decoder
De8modulator
Channel&coding
Error&control
Error&correction
Digital to&analog
Analog&to&digital
Medium
Digital&modulation
Channel(propertiesImpulse response
ℎ(")Additive&noise&V(")
Channel&Filtering
! Transmitted&PAM&signal
W " =$% & '(" − &*)�
-
! Filtered&by&channel,&impulse&response&ℎ("):
! Noise&is&added
X " = W ∗ ℎ " +V(")
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 5
ℎ(")W(") W ∗ ℎ (")
How will this affect the&Nyquist&criterion?
Example
Nyquist&Criterion&with&Channel&Filtering&(1/2)
Signal&after&Receiver&filtering&. " , Γ(1):
. ∗ ! " =$% & (. ∗ ' ∗ ℎ)(" − &*)�
-
+ . ∗ V (")
Effective&pulse:. ∗ ' ∗ ℎ (")
Nyquist&criterion:
$ Γ 1 −3*
Z 1 −3*
4 1 −3*
5
6785
= Constant
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 6
Can(this be(achieved in(practice?
Nyquist&Criterion&with&Channel&Filtering&(2/2)
! If&Z 1 ≠ 0 everywhere&with&P 1 ≠ 0 and&P 1 is&Nyquist! Pick&Γ 1 = 1/Z(1) Impractical&to&depend&on&Z(1)!
! If&Z 1 ≈ constant where&P 1 ≠ 0
! Pick&Γ 1 and&4(1) as&Nyquist&together
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 7
FGHIJKLB^QJLOQJM_ $ Γ 1 −3*
Z 1 −3*
4 1 −3*
5
6785
BisBconstant
Example:Small(bandwidth:
Approximately constantOne reason to&use OFDM
|Z 1 |
What&if&Nyquist&Criterion&is¬&satisfied?
! Sampled&output
a b =$% & (. ∗ ' ∗ ℎ)(b* − &*)�
-
+ . ∗ V (b*)
! Define&c d = (. ∗ ' ∗ ℎ)(d*)
! InterMsymbol&interference&if&c d ≠ 0 for&some&d ≠ 0
! Causal&and&timeMlimited&pulses&and&impulse&response! c d ≠ 0 for&d = 0, … , f − 1
a b = $% b − d c[d]i8j
k7l
+Vm[b]
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 8
Vm[b]
fMtap&Channel&with&Gaussian&Noise
! Sampled&output
a b = $% b − d c[d]i8j
k7l
+Vm[b]
where&Vm[b] is! Gaussian&with&zero&mean&and&variance&nl/2
! Independent&for&different&values&of&b
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 9
This is&called a&dispersive channel
How to(perform detection? This is&the&topic of this and&next lecture
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 10
Detecting&Sequence&of&o symbols
ChannelSender ReceiverSource Destination
Noise
Message&set: pq q7lrs8j Size:&tu
Mapping: pq → %q 0 , %q 1 , … %q[o − 1]
for&w ∈ 0,1, … ,tu − 1
A&message consists of oBsymbols&from&an&tMsize constellation (n = 1)
% 0 , % 1 , … %[o − 1]
a 0 , a 1 , …a[o − 1]
Assume: % −f + 1 ,… , %[−1] are known
ML&Sequence&Detection&(1/3)
! Recall:
! In&this&case:
! Note
a b = $% b − d c[d]i8j
k7l
+Vm[b]
is&Gaussian&with&variance&nl/2 and mean&∑ % b − d c[d]i8jk7l
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 11
ML(decision(rule: Set&pz = pq if 1{̅|}(~̅|p-) is&maximized for&& = w.
ML&Sequence&Detection&(2/3)
! Define:
� %- b , … , %-[b − f + 1] = $%- b − d c[d]i8j
k7l! We&obtain:
! Maximized&by&minimizing
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 12
ML&Sequence&Detection&(3/3)
! Define
! We&have
! Euclidean&distance&between&~̅ and&received&signal&without&noise
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 13
ML(decision(rule:&Set&pz = pq&if&Ä ~̅, �Å- &is&minimized&for&& = w.
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 14
1 3
3
1
–1
Geometric&Illustration
?
Received&vector: z
z
Goal:&Minimize&the&error&probability
Pr É ≠ ÉÑ|a̅ = ~̅
Estimated&message
Sent&message
Received&vector
Observed&realization
Same(setup(as(beforeAre&we&done&now?
�Åj
�ÅÖ
�Ål
Complexity&Considerations
! ML&sequence&detection! Compute&Ä ~̅, �Å- for&& = 0,… ,tu − 1
! Select&the&smallest&distance
! Ways&to&reduce&complexity1. Send&one&symbol&and&be&silent&for&f − 1 symbols2. Design&some&approximate&detection&algorithm3. Find&a&way&to&implement&ML&detection&more&efficiently
2016M10M07 TSKS01&Digital&Communication&M Lecture&6 15
Number(of(messages(grows(very(fast!
This(is(what(we(will(do!www.liu.se