Transcript HVT

Communication Theory
I. Frigyes
2009-10/II.
http://docs.mht.bme.hu/~frigyes/hirkelm
hirkelm01bEnglish
Frigyes: Hírkelm
2
Topics
• (0. Math. Introduction: Stochastic processes, Complex
envelope)
• 1. Basics of decision and estimation theory
• 2. Transmission of digital signels over analog channels:
noise effects
• 3. Transmission of digital signels over analog channels:
dispersion effects
• 4. Analóg jelek átvitele – analóg modulációs eljárások (?)
• 5. Channel characterization: wireless channels, optical
fibers
•
•
•
•
•
6. A digitális jelfeldolgozás alapjai: mintavételezés, kvantálás, jelábrázolás
7. Elvi határok az információközlésben.
8. A kódelmélet alapjai
9. Az átvitel hibáinak korrigálása: hibajavító kódolás; adaptív kiegyenlítés
10. Spektrális hatékonyság – hatékony digitális átviteli eljárások
Frigyes: Hírkelm
3
(0. Stochastic processes, the
complex envelope)
Stochastic processes
• Also called random waveforms.
• 3 different meanings:
• As a function of ξ number of realizations:
a series of infinite number of random
variables ordered in time
• As a function of time t: a member of a
time-function family of irregular variation
• As a function of ξ and t: one member of a
family of time functions drawn at random
Frigyes: Hírkelm
5
Stochastic processes
• example:
f(t,ξ3)
2
f(t,ξ2)
ξ
f(t,ξ1)
f(t1,ξ)
t
Frigyes: Hírkelm
f(t2,ξ)
1
f(t 3,ξ)
3
6
Stochastic processes: how to
characerize them?
• According to the third definition
• And with some probability distribution.
• As the number of random variables is
infinite: with their joint distribution (or
density)
• (not only infinite but continuum cardinality)
• Taking these into account:
Frigyes: Hírkelm
7
Stochastic processes: how to
characerize them?
• (Say: density)
• First prob. density of x(t) p xt t
x
• second:
joint t1,t2
p x1, x 2 xt1 , xt 2 t1 , t 2
• nth:n-fold joint
p x1 , x2 ,... xn xt1 , xt 2 ,.... xt n t1 , t 2 ,...t n
• The stochastic process is completly
characterized, if there is a rule to compose
density of any order (even for n→).
• (We’ll see processes depending on 2
parameters)
Frigyes: Hírkelm
8
Stochastic processes: how to
characerize them?
• Comment: although precisely the process
(function of t and ξ) and one sample
function (function of t belonging to say ξ16)
are distinguished we’ll not always make
this distinction.
Frigyes: Hírkelm
9
Stochastic processes: how to
characerize them?
•
•
•
•
•
Example: semi-random binary signal:
Values : ±1 (P0=P1= 0,5)
Change: only at t=k×T
1
1
p x   x  1   x  1
2
2
First density_:
1
1

Second::






 x1  1,  x2  1,

x

1
,

x

1

1
2

2
2

if t1 , t 2 in the same time slot ;

p x1, x 2 xt1 , xt 2    1
1
 4   x1  1,   x2  1  4   x1  1,   x2  1 

 1   x  1,   x  1  1   x  1,   x  1, otherwise
1
2
1
2
 4
4
Frigyes: Hírkelm
10
Continuing the example:
In the same time-slot
45o
Frigyes: Hírkelm
In two distinct time-slots
45o
11
Stochastic processes: the
Gaussian process
• A stoch. proc. is Gaussian if its n-th density is
that of an n-dimensional vector random variable
px ti  X  
1
det K 2 
n
e

1
 X m T K 1  X m 
2
• m is the expected value vector, K the covariance
matrix.
• nth density can be produced if are given
mx t   Ext  és K x t1 , t2   Ext1   mx t1  xt2   mx t2 
• are given
Frigyes: Hírkelm
12
Stochastic processes: the
Gaussian process
• An interesting property of Gaussian
processes (more precisely: of Gaussian
variables):
Ex  y  z  w  Ex  y Ez  w 
 Ex  z Ey  w  Ex  w  Ez  y
• These can be realizations of one process
at different times
Frigyes: Hírkelm
13
Stochastic processes: stationary
processes
• A process is stationary if it does not
change (much) as time is passing
• E.g. the semirandom binary signal is
(almost) like that
• Phone: to transmit 300-3400 Hz sufficient
(always, for everybody). (What could we
do if this didn’t hold?)
• etc.
Frigyes: Hírkelm
14
Stochastic processes: stationary
processes
• Precise definitions: what is almost unchanged:
• A process is stationary (in the strict sense) if for
the distribution function of any order and any at
any time and time difference
Fx t1 , t 2 ...t n , t n 1 ,...   Fx t1   , t 2   ...t n   , t n 1   ,... 
• Is stationary in order n if the first n distributions
are stationary
• E.g.: the seen example is first order stationary
• In general: if stationary in order n also in any
order <n
Frigyes: Hírkelm
15
Stochastic processes: stationary
processes
• Comment: to prove strict sense stationarity
is difficult
• But: if a Gaussian process is second order
stationary (i.e. in this case: if K(t1,t2) does
not change if time is shifted) it is strict
sense (i.e. any order) stationary. As: if we
know K(t1,t2) nth density can be computed
(any n)
Frigyes: Hírkelm
16
Stochastic processes:
stationarity in wide sense
• Wide sense stationary: if the correlation
function is unchanged if time is shifted (to
be defined)
• A few definitions:.
• a process is called a Hilbert-process if
Ext   
2
• (That means: instantaneous power is
finite.)
Frigyes: Hírkelm
17
Stochastic processes: wide
sense stationary processes
• (Auto)correlation function of a Hilbertprocess:
Rt1 , t 2   Ext1 .xt 2 
• The process is wide sense stationary if
• the expected value is time-invariant and
• R depends only on τ=t2-t1 for any time and
any τ.
Frigyes: Hírkelm
18
Stochastic processes: wide sense
– strict sense stationary processes
• If a process is strict-sense stationary then also
wide-sense
• If at least second order stationary: then also
wide sense.
Rt1 , t2   Ext1 , xt 2  

 X



 X

• I.e.:
Frigyes: Hírkelm
t1

t1


X t2 p x1x2 X t 1 , X t 2 dX t1 dX t2

X t2 p x1x2 X  t 1 , X   t dX t1 dX t2 Rt1   , t2   
2
Rt1 , t 2   Rt1 , t1     R 
19
Stochastic processes: wide sense
– strict sense stationary processes
• Further: if wide sense stationary, not strict
sense stationary in any sense
• Exception: Gaussian process. This: if wide
sense stationary, also in stict sense.
Frigyes: Hírkelm
20
Stochastic processes: once
again on binary transmission
• As seen: only first order stationary (Ex=0)
• Correlation:
•
if t1 and t2 in the same time-slot:
Rt1 , t 2   Ext1 .xt 2   1
•
if in different:
Rt1 , t 2   Ext1 Ext 2   0
Frigyes: Hírkelm
21
Stochastic processes: once
again on binary transmission
• The semi-random binary transmission can
be transformed in random by introducing a
dummy random variable e distributed
uniformly in (0,1)
yt   xt  e
T
• like x:
Frigyes: Hírkelm
e
Eyt   0
22
Stochastic processes: once
again on binary transmission
• Correlation:
• If |t1-t2|>T,
(as e T)
• if |t1-t2|  T
Eyt1 yt2   EEyt1 .yt2 e  0
1; e  T  t1  t2
Ry t1 , t2   
 0; otherwise
• so
Ry    1 Ee  T     1 
Frigyes: Hírkelm

T
23
Stochastic processes: once
again on binary transmission
• I.e. :
-T
Frigyes: Hírkelm
T
τ
24
Stochastic processes: other
type of stationarity
• Given two processes, x and y, these are
jointly stationary, if their joint distributions
are alle invariant on any τ time shift.
• Thus a complex process zt   xt   jy t 
is stationary in the strict sense if x and y
are jointly stationary.
• A process is periodic (or ciklostat.) if
distributions are invariant to kT time shift
Frigyes: Hírkelm
25
Stochastic processes: other
type of stationarity
• Cross-correlation:
Rx , y t1 , t 2   Ext1  y t 2 
• Two processes are jointly stationary in the
wide sense if their cross correlation is
invariant on any time shift   t 2  t1
Frigyes: Hírkelm
26
Stochastic processes: comment
on complex processes
• Appropriate definition of correlation for
these:






R t1 , t2  E x t1 .x t 2 
• A complex process is stationary in the
wide sense if both real and imaginary
parts are wide sense stationary and they
are that jointly as well
Frigyes: Hírkelm
27
Stochastic processes: continuity
• There are various definitions
• Mean square continuity


E xt     xt    , ha   
2
2
• That is valid if the correlation is continuous
Frigyes: Hírkelm
28
Stochastic processes:
stochastic integral
• x(t) be a stoch. proc. Maybe that Rieman
integral exists for all realizations:
b
s   x t dt
a
• Then s is a random variable (RV). But if
not, we can define an RV converging (e.g.
mean square) to the integral-approximate
sum:
Frigyes: Hírkelm
29
Stochastic processes:
stochastic integral
2
n

 


s   xt dt if lim E  s   xti ti    0
ti 0

 
a
 i 1

b
• For this
b
Es    Ext dt
a
b b
 s 2    Rt1 , t 2   Ext1 Ext 2 dt1dt2
a a
Frigyes: Hírkelm
30
Stochastic processes:
stochastic integral - comment
• In σs2 the integrand is the
(auto)covariancie-function:
C t1 , t 2  ˆ Ext1 xt 2   Ext1 Ext 2 
• This depends only on t1-t2=τ if x is
stationary (at least wide sense)
Frigyes: Hírkelm
31
Stochastic processes: time
average
• Integral is needed – among others –to define
time average
• Time average of a process is its DC component;
• time average of its square is the mean power
• definition:
1

n x  lim
T   2T
Frigyes: Hírkelm
T
 xt dt
T
32
Stochastic processes: time
average
• In general this is a random variable. It
would be nice if this were the statistical
average. This is really the case if

2
nx  Ext  and  n  0
• Similarly we can define

1
R   lim
T   2T
Frigyes: Hírkelm
T
 xt xt   dt
T
33
Stochastic processes: time
average
• This is in general also a RV. But equal to
the correlation if

2

Rx    Rx  , with  R  0
• If these equalities hold the process is
called ergodic
• The process is mean square ergodic if
1
lim
T  2T
Frigyes: Hírkelm
T
 R  d Ext 
2
x
T
34
Stochastic processes: spectral
density
• Spectral density of a process is, by
definition the Fourier transform of the
correlation function
S x   

 R  e
x
 j
d

Frigyes: Hírkelm
35
Stochastic processes: spectral
density
• A property:

E xt 
2

1

2

 S  d
x
as


1
Fu  d  u   0 ;   , : Fourier va riable - pairs 

2 -

But : R  0  E xt 
2

• Consequently this integral >0; (we’ll
see: S˙(ω)>0)
Frigyes: Hírkelm
36
Spectral density and linear
transformation
• As known in time functions output function
is convolution
x(t)
FILTER
y(t)
h(t)
y t   xt   ht  

 xu ht  u du

• h(t): impulse response
Frigyes: Hírkelm
37
Spectral density and linear
transformation
• Comment.: h(t<0)≡ 0; (why?);
and: h(t) = F-1[H(ω)]
• It is plausible: the same for stochastic
processes
• Based on that it can be shown :
S y    H   S x  
2
• (And also
Frigyes: Hírkelm
Eyt   Ext H 0
)
38
Spectral density and linear
transformation
• FurtherS(ω) ≥ 0 (all frequ.)
• For: if not, there is a domain where
S(ω) <0 (ω1, ω2)
x(t)
FILTER
y(t)
h(t)
H(ω)
Frigyes: Hírkelm
Sx(ω)
Sy(ω) (its integral is negative)
39
Spectral density and linear
transformation
• S(ω) is the spectral density (in rad/sec).
As:
2
H(ω)
S y    S x   H   
Power
ω
Frigyes: Hírkelm
1
2

 S  d  2S  B
x
Hz
x

40
Modulated signals – the complex
envelope
• In previous studies we’ve seen that in
radio, optical transmission
• one parameter is influenced (e.g. made
proportional)
•
of a sinusoidal carrier
•
by the modulating signal
.
• A general modulated signal:
xt   2 Ad t cosc t   t   
Frigyes: Hírkelm
41
Modulated signals – the
complex envelope
• Here d(t) and/or (t) carries the
information – e.g. are in linear relationship
with the modulating signal
• An other description method (quadrature
form):
xt   Aat cosc t     Aqt sinc t   
• d, , a and q are real time functions –
deterministic or realizations of a stoch.
proc.
Frigyes: Hírkelm
42
Modulated signals – the
complex envelope
• Their relationship:
d t  
at   qt 
2
2
2
qt 
; t   arctg
at 
at   2d t cos t ; qt   2d t sin  t 
• As known x(t) can also be written as:
xt   Reat   jqt e 
jct
Frigyes: Hírkelm
43
Modulated signals – the
complex envelope
• Here a+jq is the complex envelope.
Question: when, how to apply.
• To beguine with: Fourier transform of a
real function is conjugate symmetric:
X   ˆ Fxt ; and Fx- t   X  

• But if so: X(ω>0) describes the signal
completly: knowing that we can form the
ω<0 part and, retransform.
Frigyes: Hírkelm
44
Modulated signals – the
complex envelope
• Thus instead of X(ω) we can take that:
X   ˆ X    sign  X    X    j jsign  X  

• By the way:
2 X  ;   0
X    
 0;   0
↓
„Hilbert” filter

• The relevant time function:


xt   F  X    xt   jF1  jsign  X  



Frigyes: Hírkelm
1
45
Modulated signals – the
complex envelope

We can write: xt   xt   jx t  F1 - jsign 
•
• The shown inverse Fourier transform is
1/t.


x 
xt   xt   j 
d
• So

t 
xt   xt   jHxt   xt   jxˆ t 

• Imaginary part is the so-callerd Hilberttranszform of x(t)
Frigyes: Hírkelm
46
Modulated signals – the
complex envelope

• Now introduced function xt  is the analytic
function assigned to x(t) (as it is an
analytic function of the z=t+ju complex
variable).
• An analytic function can be assigned to
any (baseband or modulated) function;
relationship between the time function and
the analytic function is
Frigyes: Hírkelm
47
Modulated signals – the
complex envelope
• It is applicable to modulated signals: analytic signal of
cosωct is ejωct. Similarly that of sinωct is jejωct. So if
quadrature components of the modulated signal a(t), q(t)
are
•
band limited and
•
their band limiting frequency is < ωc/2π (narrow band

signal)
xt   at  cos ct  xt   at e jct
• then
or
NB. Modulation is a linear

operation in a,q: frequencyxt   qt sin  t  xt   jqt e jct
c
displacement.
Frigyes: Hírkelm
48
Modulated signals – the
complex envelope



• Thus ~
x t ̂ a t  jq t complex envelope
determines uniquely the modulated signals. In the time
domain

~
x t   at   jq t   xt   ~
x t e jct  xt  
 
 Re  xt   at cos c t  qt sin c t
 
• Comment: ~
x t  according to its name can be complex.
(X(ω) is not conjugate symmetricaround ωc.)
• Comment 2: if the bandwidt B>fc,xt  is not analytic, its
real part does not define the modulated signal.)
• Comment 3: a és q can be independent signals (QAM)
or can be related (FM or PM).
Frigyes: Hírkelm
49
Modulated signals – the
complex envelope
• In frequency domain? On analytic signal we saw.

~
x t   xt   e
 j c t

~
and so X    X   c 
X(ω)
X(ω)
X˚(ω)
X̃(ω)
Frigyes: Hírkelm
ω
50
Modulated signals – the complex envelope
• Linear transformation – bandpass
filter –acts on x̃(t) as
a lowpass filter.
• If H(ω) is asymm:
x̃(t) is complex – i.e.
is a crosstalk
between a(t) és q(t)
között
• (there was no sin
component – now
there is.)
Frigyes: Hírkelm
M(ω)
H(ω)
X(ω)=F[m(t)cosωct]
Y(ω)
Y˚(ω)
X˚(ω)
X̃ (ω)
Ỹ(ω)
51
Modulated signals – the complex envelope,
stochastic processes
• Analytic signal and complex envelope are
defined for deterministic signals
• It is possible for stochastic processes as
well
• No detailed discussion
• One point:
Frigyes: Hírkelm
52
Modulált jelek – a komplex burkoló;
sztochasztikus folyamatok
• x(t) is stationary (Rx is independent from t),
iff
Eat   Eqt   0; Ra    Rq  ; Raq     Rqa  

xt   at cos c t  qt sin c t és



Rx t ,   Ra    Rq   cos c  Rqa    Raq   sin c 




 Ra    Rq   cos c 2t     Raq    Rqa   sin c 2t   
Frigyes: Hírkelm
53
Narrow band (white) noise
• White noise is of course not narrow band
• Usually it can be made narrow band by a
(fictive) bandpass filter:
H(ω)
X(ω)
X(ω)
Sn(ω) =N0/2
Frigyes: Hírkelm
ω
54
Narrow band (white) noise
properties
jc t
jc t
~
nt   n t e
 nc t   jn s t e

Rc    Rs  ; Rc , s     Rs ,c  


Rn~    Rc    jR c , s  
1
S n    S n~   c   S n~   c 
2
Frigyes: Hírkelm
55
1. Basics of decision theory and
estimation theory
Detection-estimation problems in
communications
• 1. Digital communication: one among signals
known (by the receiver) – in presence of noise
DIGITAL
SOURCE
Transmission
channel
SINK
• E.g.: (baseband binary communication)
s1 t   U  nt ; s0 t   nt 
• Decide: which was sent?
Frigyes: Hírkelm
57
Detection-estimation problems in
communications
• 2. (Otherwise) known signal has unknown
parameter(s) (statistics are known) • Same block schematic; example: noncoherent FSK




s1 t   2 A cos c1t  1  nt ; s2 t   2 A cos c 2t  2  nt 
1
 ;     ,  
p1    p 2     2
 0; otherwise
Frigyes: Hírkelm
58
Detection-estimation problems
in communications
• Other example: non-coherent FSK, over nonselective Rayleigh fading channel




s1 t   2 A cos c1t  1  nt ; s2 t   2 A cos c 2t  2  nt 
1
 ;     ,  
p1    p 2     2
 0; otherwise
Frigyes: Hírkelm
  2
p A    2 e

2 2
59
Detection-estimation problems
in communications
• 3. Signalshape undergoes random changes
• Example: antipodal signal set in very fast fading
DIGITAL
SOURCE
Transmission
channel
T(t)
SINK
s1 t   st  T t   nt ; s2 t   st  T t   nt 
Frigyes: Hírkelm
60
Detection-estimation problems
in communications
• 4. Analog radio communications: one parameter
of the carrier is proportional to the time-contous
modulating signal.
• E.g..: analog FM; estimate: m(t)


st   2 A cos c t  2 .F.mt   nt 
• Or: digial transmission over frequency-selective
fading. For decision: h(t) must be known (i.e.
estimated
r t  

 ht   si  d  nt ; i  1,2,... M

Frigyes: Hírkelm
61
Basics of decision theory
• Simplest example: simple binary transmission;
decision is based on N independent samples
• Model:
H0
SOURCE
CHANNEL
H1
OBSEVATION
SPACE
(OS)
(Only statistics are known)
Comment: now ˆ has nothing to
do with Hilbert transform
Decision rule
H0? H1?
DECIDER
Frigyes: Hírkelm
Ĥ
62
Basics of decision theory
• Two hypothesis (H0 és H1)
• Observation: N samples→the OS is of Ndimensions
• Observation: rT=(r1,r2…,rN)
• Decision: which was sent
• Results: 4 possibilities
•
1. H0 sent & Ĥ=H0 (correct)
•
2. H0 sent & Ĥ=H1 (erroneous)
•
3. H1 sent & Ĥ=H1 (correct)
•
4. H1 sent & Ĥ=H0 (erroneous)
Frigyes: Hírkelm
63
Bayes decision
• Bayes decision :
•
a.) probabilities of sending H0 or H1 are apriori known:
Pr H 0  ˆ P0 ; Pr H1  ˆ P1
•
• b.) each decision has some cost (Cik) (we decide
in favor of i while sent was k)
• c.) it is sure: false decision is more expensive
than correct:
C10  C00 ; C01  C11
Frigyes: Hírkelm
64
Bayes decision
•
d.) decision rule: the average cost (so called
risk, K) should be minimal




K  C11 P1 Pr Hˆ  H1 | H1  C01 P1 Pr Hˆ  H 0 | H1 
 C P Pr Hˆ  H | H  C P Pr Hˆ  H | H
00 0

0
0

10 0

1
0

OS
pr|H1(R|H1)
(Z1)
FORRÁS
„H1”
pr|H0(R|H0)
„H0”
(Z0)
domain of r; two pdf-s
correspond to each point.
Frigyes: Hírkelm
65
Bayes decision
• Question: how to partition OS in order to
get minimal K?
• For that: K in detail:
K  C00 P0
C11P1
 pr|H 0 R | H 0 dR C10 P0  pr|H 0 R | H 0 dR 
Z 0 
Z1 
 pr|H 1 R | H1 dR C01P1  pr|H 1 R | H1 dR
Z1 
Z 0 
• As some decision is taken:
• And so 
 1

Z1 
Frigyes: Hírkelm

Z1  Z 0  Z 0 

 p   1
Z 0 Z1 
Z 0 
66
Bayes decision
• From that:
K  P0C10  P1C11 

 P1 C01  C11  pr|H 1 pR | H1 dR 
Z 0 
 P0 C10  C00  pr|H 0 pR | H 0 dR
Z 0 
• Term 1 & 2 are constant
(Z )
FORRÁS
„H ”
„H ”
• And: both integrands >0
(Z )
p (R|H )
• Thus: Z1, where the first integrand is larger
Z0, where the second
pr|H1(R|H1)
1
0
1
r|H0
Frigyes: Hírkelm
0
0
67
Bayes decision


Z1 : P1 C01  C11  pr | H 1 pR | H1  


 P0 C10  C00  pr | H 0 pR | H 0 
• And here we decide in favor of H1: Hˆ  H1


Z 0 : P1 C01  C11  pr | H 1 pR | H1  


 P0 C10  C00  pr | H 0 pR | H 0 
• decision: H0
Frigyes: Hírkelm
Hˆ  H 0
68
Bayes decision
• It can also be written: decide for H1 if
pr | H 1 R | H1 
P0 C10  C00 

pr | H 0 R | H 0  P1 C01  C11 
•
•
•
•
Otherwise for H0
Lefthand side: likelyhood ratio, Λ(R)
Righthand: (from certain aspect) the treshold, η
Comment: Λ depends only on the realisation of r
(on: what did we measure?)
•
η only on the a-priori probabilities and costs
Frigyes: Hírkelm
69
Example on Baysean decision
• H1: constant voltage + Gaussian noise
• H0: Gaussian noise only (designation:
φ(r;mr,σ2)
• Decision: on N independent samples of r
• At sample #i



pri | H 1   Ri ; m,  2 ; pri | H 0   Ri ;0,  2
N





p r | H 1 R | H 1     Ri ; m,  2 ;
• This resulting in
i 1
N
p r | H 0 R | H 0     Ri ;0,  2
Frigyes: Hírkelm
i 1
70
Example on Baysean decision
 Ri  m 2
 2  exp   2 2
i 1

R  
N
 Ri 2 
1
 2  exp   2 2 
i 1


N
• its logaritm
1
ln R  
• resulting in
m

2
N




Nm 2
 Ri  2 2
i 1
2
N
Nm

Nm
ˆ
ˆ
H  H1 :  Ri 
ln  
; H  H 0 :  Ri 
ln  
m
2
m
2
i 1
i 1
N
2
threshold
Frigyes: Hírkelm
71
Comments to the example
• 1. The threshold contains known quantities only,
independent of the measured values
• 2. Result depends only on the sum of ri-s – we
have to know only that; so called sufficient
statistics l(R):
N
l R    Ri
i 1
• 2.a Like in this example: OS dimension is
whatever, l(R) is always 1D
• i.e.„1 coordinate” – the others are independent
on hypothesis
Frigyes: Hírkelm
72
Thus the decision process
H0
SOURCE
CHANNEL
H1
OBSERVATION
SPACE
(OS)
(Only statistics known)
l R 
DECISION
SPACE
(DS)
DECIDER
Ĥ
Frigyes: Hírkelm
Decision rule
73
Comments to the example
• 3. Special case: C00=C11=0 és C01=C10=1
K  P0
 pr|H 0 R | H 0 dR  P1  pr|H 1 R | H1 dR
Z1 
Z 0 
• (i.e. probability of erroneous decision)
• If P0,1≡0,5, the treshold N.m/2
Frigyes: Hírkelm
74
An other example, for home
• Similar but now the signal is not constant
but Gaussian noise with variance σS2
• I.e.
H1:Π φ(Ri;0,σS2+σ2)
•
H0:Π φ(Ri;0,σ2)
• Questions: threshold; sufficient statistics
Frigyes: Hírkelm
75
Third example - discrete
• Given two Poisson sources with different
expected values; which was sent?
• Remember: Poisson-distribution:
mn e n
Pr n 
n!
• Hypotheses:
n  m1
m1 e
H1 : Pr n  
n!
Frigyes: Hírkelm
n  m0
m0 e
; H 0 : Pr n  
n!
76
Third example - discrete
n
 m1   m1  m0
 e
n   
• Likelihood-ratio:

m
 0
ln   m1  m0
if n 
: H1 ;
• Decision rule:
ln m1 m0 
(m1>m0)
ln   m1  m0
if n 
: H0
ln m1 m0 
• For sake of precision:
ln   m1  m0
if n 
: H1Q  H 0 1  Q ;
ln m1 m0 
Pr(Q  0)  Pr(Q  1)  0,5
Frigyes: Hírkelm
77
Comment
• A possible situation: a-priori probebilities
are not known
• A possible method then:
compute maximal K (as a function of Pi;
and chose the decision rule wich
minimizes that (so called minimax
decision)
• (Note: this is not optimal at any Pi )
• But we don’t deal with this in detail.
Frigyes: Hírkelm
78
Probability of erroneous decision
• For that: compute relevant integral
• In example 1 (with N=1): Gs pdf should be
integrated over the hatched domains
d0
Threshold:
Frigyes: Hírkelm
d1

m
d 0  ln  
m
2
79
Probability of erroneous decision
• Thus:
PE  P0 P1 | 0   P1P0 | 1

1
 d0 
P1 | 0    R ;0,  dR  erfc  
2

d0
2
1
 m  d0 
P0 | 1   R ; m,  dR  erfc 

2




d0
2
• If lnη=0: d0=d1=m/2 (threshold: the point of
intersection)
• Comment: N samples:

m N
d0 
ln  
2
m N
Frigyes: Hírkelm
80
Decision with more than 2
hypotheses
• M possible outcomes (e.g.:non-binary
digital communication – we’ll see: why for)
• Like before: each decision has a cost
• Their average is the risk
• With Bayes-decision: this is minimized
• Like before: Observation Space
•
decision rule: partitioning of the OS
Frigyes: Hírkelm
81
•
Decision with more than 2
hypotheses
M 1M 1
Like before, risk:
K ˆ   Cij Pj  p R| Hj R | H j dR
i 0 j 0
• From that (with M = 3)
Z i 
K  P0 C00  P1C11  P2 C22 
 P2 C02  C22  pr|H 2 R | H 2   P1 C01  C11  pr|H 1 R | H1 dR 
Z 0 
 P2 C12  C22  pr|H 2 R | H 2   P0 C10  C00  pr|H 0 R | H 0 dR 
Z1 
 P1 C21  C11  pr|H 1 R | H1   P0 C20  C00  pr|H 0 R | H 0 dR
Z 2 
Frigyes: Hírkelm
82
Döntés kettőnél több hipotézisnél
• Likelyhood ratio-series :
1 R  
pr| H 1 R | H1 
pr| H 0 R | H 0 
;  2 R  
pr| H 2 R | H 2 
pr| H 0 R | H 0 
• Decision rule(s):
Hˆ  H1 or H 2 : P1 C01  C11 1 R   P0 C10  C00   P2 C12  C02  2 R 
Hˆ  H or H if less;
0
2
Hˆ  H 2 or H1 : P2 C02  C22  2 R   P0 C20  C00   P1 C21  C01 1 R 
Hˆ  H or H if less;
0
1
Hˆ  H 2 or H1 : P2 C12  C22  2 R   P0 C20  C10   P1 C21  C11 1 R 
Hˆ  H or H if less;
Frigyes: Hírkelm
1
0
83
Döntés kettőnél több hipotézisnél
(M =3)
• This defines 3 streight lines (in the 2D decision
space)
Λ2(R)
H0
H2
H1
Λ1(R)
Frigyes: Hírkelm
84
Ecample: special case – error
probability
• The average error
probability is
minimized
Cii  0; Ci  j  1
Λ2(R)
H2
• Then we get:
Λ2 = (P1/P2)Λ1.
P0 /P2
H1
H0
Frigyes: Hírkelm
P0 /P1
Λ1(R)
85
The previous, detailed
Hˆ  H 1 vagy H 2 : P11 R   P0
Hˆ  H vagy H ha kisebb;
0
Cii  0; Ci  j  1
2
Hˆ ˆ H 1 vagy H 2 : P1 C01  C11 1 R   P0 C10  C00   P2 C12  C02  2 R 
H  H 2 vagy H 1 : P2  2 R   P0
Hˆ  H vagy H ha kisebb;
0
Hˆ  H 0 vagy H 1 ha kisebb;
2
C20  CH00   P1 C21  C01 1 R 
Hˆ  H 2 vagy H 1 : P2 C02  C 22  2 R   ΛP20(R)
2
Hˆ  H 2 vagy H 1 : P2 ˆ2 R   P11 R 
Λ2 = (P1/P2)Λ1.
H  H vagy H ha kisebb;
ˆ
0
P01 /P2
H vagy
H1 H
vagy
H
kisebb;
0 ha

Hˆ  H
:
P
C

C
 C  R 
2
1
2
12
22  2 R   P0 C 20  C10   P1 C 21
H1 11 1
H0
Hˆ  H vagy H ha kisebb;
1
Frigyes: Hírkelm
0
P0 /P1
Λ1(R)
86
Example –special case, error
probability: a-posteriori prob.
• Very important, based on the precedings: we have
H1 vagy H 2 : P1 pr| H 1 R | H 1   P0 pr| H 0 R | H 0 
H 2 vagy H1 : P2 pr| H 2 R | H 2   P0 pr| H 0 R | H 0 
H 2 vagy H 0 : P2 pr| H 2 R | H 2   P1 pr| H 1 R | H1 
• If we divide each by pr(R) and apply Bayes theorem we
get:
Pr b | a  Pr a 
Bayes theorem : Pr a | b  
Pr b 
Pr H1 | R  Pr H 0 | R; Pr H 2 | R  Pr H 0 | R; Pr H 2 | R  Pr H1 | R
Frigyes: Hírkelm
(these
are a-posteriori probabilities)
87
Example –special case, error
probability: a-posteriori prob.
• I.e. we have to decide on the max. aposteriori probabilities.
• Rather plausible: probability of correct
decision is the highest if we decide on
what is the most probable
Frigyes: Hírkelm
88
Bayes- theorem (conditional
probabilities)
Pr a, b 
Pr a, b 
Pr a | b  ˆ
; Pr b | a  ˆ
Pr b 
Pr a 
Pr b | a . Pr a 
 Pr a | b  
Pr b 
• For discrete variables:
pb | a . pa 
pa | b  
pb 
• Continuous:
• a discrete, b continuous:
pb | a . Pr a 
Pr a | b  
pb 
Frigyes: Hírkelm
89
Comments
• 1. The Observation Space is N dimensional (N
is number of the observations).
•
The Decision Space is M-1 dimensional (M is
number of the hypotheses).
• 2. Explicitly we dealt only with independent
samples; investigation is much more
complicated is these are correlated
• 3. We’ll see that in digital transmission the case
N>1 is often not very important
Frigyes: Hírkelm
90
Basics of estimation theory:
parameter-estimation
• Task is to estimate unknown parameter(s)
of an analog or digital signal
• Examples: voltage measurment in noise
x  s; r  s  n
•
digital signal; phase measurement
2


si t   ai t  cos c t 
i   ;
M


(i  0,1,... M  1)
Frigyes: Hírkelm
91
Basics of estimation theory:
parameter-estimation
• Frequency estimation
•
synchronizing an unaccurate oscillator
• Power estimation of interfering signals
•
interference cancellation (via the
antenna or multiuser detection)
• SNR estimation
• etc
Frigyes: Hírkelm
92
Basics of estimation theory:
parameter-estimation
• The parameter can be:
•
i. a random variable (pdf is assumed to be known) or
ii. means: we
•
ii. an unknown deterministic value
have no a-priori
• Model:
i. means: we have
knowledge about
some a-priori knowledge
its magnitude
Mapping to
OS
Domain of
the estimated
parameter
a
PARAMETER
SPACE
Frigyes: Hírkelm
pa(A)
Becslési
szabály
OBSERVATION
SPACE (OS)
â R 
ESTIMATION
SPACE
93
Example – details (estimation 1)
• We want to measure
voltage a
• We know that a  V
• And that Gaussian
noise is added
φ(r;0,σn2)
• Observable
parameter is r = a+n
 R  A2 
1
• Mapping of the
p r |a  R | A  
exp 

2
2  n
parameter to OS:
 2 n 
Frigyes: Hírkelm
94
Parameter estimation –parameter
is a RV
• Similar principle: estimation ha some cost; its
average is the risk; we want to minimize that
• realization of the parameter : a
• Observation vector: R
• Estimated value: â(R)
• Cost is in the general case a 2-variable function:
C(a,â)
• Error of the estimation is ε = a-â(R)
• Often the cost depnds only on that: C = C (ε)
Frigyes: Hírkelm
95
Parameter estimation –
parameter is a RV
• Examples:

0;    / 2
C     ; C     ; C    

1;    / 2
2
• Risk is
K  EC   
 
  CA  aˆ R pa,r  A, R dRdA
 
• Joint pdf can be
written:
pa ,r  A, R   pr R  pa|R  A | R 
Frigyes: Hírkelm
96
Parameter estimation –
parameter is a RV
• Applying to the square cost function
(subscript ms: mean square)


2
  pR   A  aˆ R  pa|r  A | R dAdR

 


K ms
• K=min (i.e. dK / daˆ  0 )where the inner
integral = min (as the outer is i. pozitiv and
ii. does not depend on A); this holds where
Frigyes: Hírkelm
97
Parameter estimation –
parameter is a RV
d 
2
ˆ




A

a
R
pa|r  A | R dA 

daˆ  




 2  Ap a|r  A | R dA  2aˆ R   pa|r  A | R dA  0
• The second integral =1, thus
aˆ R ms 

 Ap  A | R dA
a|r

• i.e. a-posteriori expected value of a.
• (According to the previous definition : a-posteriori knowledge: what
is gained from measurement/investigation.)
Frigyes: Hírkelm
98
Comment
• Comming back to the risk:
K ms 




2
ˆ
pr R dR  A  aR  pa|r  A | R dA

• Inner integral is now: the conditional
variance, σa2(R). Thus

2
2
•
K  p R  R dR  E 
ms

r
a
 
a

Frigyes: Hírkelm
99
Parameter estimation –
parameter is a RV
• An other cost function: 0 in a band Δ
elsewhere 1. The risk is now
Δ
 aˆ R   2

K un   pr R dR 1   pa|r  A | R dA
 aˆ R   2



• (un: uniform)
• K=min, if result of the estimation is
maximum of the conditional pdf (if Δ is
small): max. a-posteriori – MAP-estimation
Frigyes: Hírkelm
100
Parameter estimation –
parameter is a RV
• Than derivative of the log conditional
pdf=0
 ln p  A | R 
0
a|r
A
A aˆ R 
• (so called MAP-equation)
• Applying Bayes-theorem log-cond. pdf
can be written
p R | A. p  A 

 pa|r  A | R  

r|a
pr R 
a


ln pa|r  A | R   ln pr|a R | A  ln pa  A  ln pr R 
Frigyes: Hírkelm
101
Parameter estimation –
parameter is a RV
• First term is the statistical relationship between
A-R
• Secnd is the a-priori knowledge
• Last term does not depend on A.
• Thus what is maximal is
l  A ˆ ln pr|a R | A  ln pa  A
• And so the MAP equation is:
 ln pr|a R | A  ln pa  A
A  aˆ R , ahol :

0
A
A
Frigyes: Hírkelm
102
Once again
• Minimum Mean Square Error (MMSE)
estimate is the average of the a-posteriori.
pdf
• Maximum a-posteriori (MAP) estimate is
the maximum of the a-posteriori. pdf
Frigyes: Hírkelm
103
Example (estimate-2)
• Again we have Gauss
ian a+n, but N
independent samples 1
 A2 
pa ( A) 
exp 
2 
2  a
 2 a 
N
pr|a R | A  
i 1
2


Ri  A 
1
exp 

2
2 a 
2  n

• What we need to p  A | R   pr|a R | A pa  A
a |r
pr R 
(any) estimate is
Frigyes: Hírkelm
104
Example (estimate-2)
• Note that p(R) is constant from the point of
view of the conditional pdf, thus its form is
irrelevant.
• Thus

N

2
Ri  A



1
A2  
 1 i 1

pa|r  A | R   k1 R 
exp  
 2 
N
2
n
 a 
 2
2  a  2  n



i 1
Frigyes: Hírkelm



105
Example (estimate-2)
• This is a Gaussian distribution and we need its
expected value. For that the square must be
completed in the exponent, resulting in

1

pa|r  A | R   k 2 R exp 
2
2


2

with  2
Frigyes: Hírkelm

a
1

A  2
2
a n N  N

2
a n

2
2
N a   n
2
2

 Ri 
i 1

N
2
106
2





Example (estimate-2)
• In Gaussian pdf average = mode thus
 a2
1 N
aˆms R   aˆmap R   2
  Ri
2
 a   n / N N i 1
Frigyes: Hírkelm
107
Example (estimate-3)
• a is now also φ(A;0,σn2), but only
s(A), a nonlinear function of it can be
observed (e.g. phase of a carrier);
noise is added: ri  s A  ni
• Az a-posteriori sűrűségfüggvény

N

2


Ri  s A

2 

1  i 1
A 

pa|r  A | R   k R exp   
 2 
2
2
n
a 







Frigyes: Hírkelm
108
Példa (becslés-3)
• Remember: the MAP equation:
 ln pr|a R | A  ln pa  A
A  aˆ R , ahol :

0
A
A
• Aplying that to the preceding
a
aˆ MAP R   2
n
2
Frigyes: Hírkelm
s A
 Ri  s A A
i 1
N
109
Parameter estimation – the
parameter is a real constant
• In that case it is the measurement result what is
a RV. E.g. in the case of a square cost function
the risk is:

2
ˆ
K  A   aR   A pr|a R | AdR

• This is minimal if â(R)=A. But this has no sense:
that’s the value what we want to estimate.
• In that case – i.e. if we have no a-priori
knowledge – the method can be:
• we search for a function of A – an estimator –
which is „good” (has average close to A and low
Frigyes: Hírkelm
110
variance)
Közbevetve: a tételek eddig
• 1. Sztohasztikus folyamatok: főként a
fogalmak definiciója
• (sztoh. foly.; val. sűrűségek-eloszlások, erős
stacioanaritás; korrelációs fv. gyenge
stacionaritás, időátlag, ergodicitás spektrális
sűrűség, lin. transzformáció)
• 2. Modulált jelek – komplex burkoló
• (mod. jelek leírása, időfüggv., analitikus függv.
(frekv-idő) komplex burkoló, egyikből a másik,
szűrő)
Frigyes: Hírkelm
111
Közbevetve: a tételek eddig/2
• 3. A döntéselmélet alapjai
• (Milyen feladatok; költség, kockázat, a-priori val,
Bayes-f. döntés; a megfigyelési tér opt.
felosztása, küszöb, elégséges statisztika, M
hipotézis (nem kell rész-letezni, csak az
eredmény), a-post. val.)
• 4. A becsléselmélet alapjai
• (Val.-vált paraméter, költségfüggv. min. ms. max.
a-post.; determ. param, likelyhood fv. ML, MVU
becslés, torzítatlan; Cramér-Rao; hatékony)
Frigyes: Hírkelm
112
Parameter estimation – the parameter is
a real constant – criteria for the estimator
• Average of the – somehow chosen – estimator:
 A

Eâ R  ˆ  â R p r|A R | A dR   A  B

A  B(A)


• If this is = A: unbiased estimate; operation of
estimation is searching of the average
• If bias B is constant: can be subtracted from the
average (e.g. in optics: background radiation)
• if B=f(A): biased estimate
Frigyes: Hírkelm
113
Parameter estimation – the parameter is
a real constant – criteria for the estimator
• Error variance:
    aˆ R  A
2
•
•
•
•
•
•
•
•
2


2
ˆ
 E aR   A  B 2  A
Best would be:
unbiased and
low variance
more precisely: unbiased and minimális variance
for any A (MVU)
Such estimator does or does not exist
An often applied estimator: maximum likelihood
Likelihood function.: pr|a R | A as a function of A
Estimator of A then: maximum of the likelihood function
Frigyes: Hírkelm
114
A few of details
This is certainly better
than that:
variance of the
measuring result is lower
Θ
Further:
Frigyes: Hírkelm
115
Max. likelihood (ML) estimation
• Necessary condition of max of ln likelihood
A  aˆ R , ahol :
 ln pr|a R | A
A
0
• Remember: in the case of a RV: MAP estimate
 ln pr|a R | A  ln pa  A
A  aˆ R , ahol :

0
A
A
• Thus: ML is the same without a-priori knowledge
Frigyes: Hírkelm
116
Max. likelihood (ML) estimation
• For any unbiased estimate: variance cannot be lower
than Cramér-Rao lower bound, CRLB,
 2 aˆ R  A
   ln p R | A 

r|a

  E 


A
 

 
2
 


 
1
• or an other form:

   ln pr|a R | A  

 -E 

2

A


 



2
 2 aˆ R  A
Frigyes: Hírkelm
1
117
Max. likelihood (ML) estimation
• Proof of CRLB: via Schwartz- unequality
• If the estimate is equal to CRLB: efficient
estimate.
Example of existing MVU
but no (or unknown)
efficient estimate
(var θ^1>CRLB)
Frigyes: Hírkelm
118
Example 1 (estimation,
nonrandom)
• Voltage + noise, but the voltage is a real,
nonrandom
• Max likelihood estimation

ri  A  ni ; i  1,2,... N ; pn (n)   n;0,  n
 ln pr| A
N 1
 2
A
n  N
1 N
aˆ ML R    Ri
N i 1
Frigyes: Hírkelm
2


 Ri  A   0
i 1

N
119
Example 1 (estimation,
nonrandom)
• Is it biased??
1
Eaˆ ML R  
N
N
1
 ERi   N NA  A
i 1
• Expected value of â is the (a-priori not
known) true value; i.e. it is unbiased
Frigyes: Hírkelm
120
Example 2 (estimation,
nonrandom): phase of a sinusoid
• What can be measured is a function of the
quantity to be estimated
• (Independent Samples are taken at equal
times)
• A likelyhood function to be maximized:
Frigyes: Hírkelm
121
Example 2 (estimation,
nonrandom): phase of a sinusoid
• This is = max, if the function below = min
•
=0
• The righthand side tends to 0, with large N
Frigyes: Hírkelm
122
Example 2 (estimation,
nonrandom): phase of a sinusoid
• And then finally:
Frigyes: Hírkelm
123
2. Transmission of digital signals
over analog channels: effect of
noise
Introductory comments
• Theory of digital transmission is (at least
partly) application of decision theory
• Definition of digital signals/transmission:
•
Finite number of signal shapes (M)
•
Each has the same finite duration (T)
•
The receiver knows (a priori) the signal
shapes (they are stored)
• So the task of the receiver is hypothesis
testing.
Frigyes: Hírkelm
125
Introductory comments –
degrading effects
ωc
s(t)NONLINEAR
AMPLIFIER
n(t)
BANDPASS
FILTER
FADING
CHANNEL
z0(t)
ωc
z1(t)
ω1
ACI
ω2
ACI
INTERFERENCE
INTERFERENCE
z2(t)
INTERFERENCE
Frigyes: Hírkelm
+
BANDPASS
FILTER
DECISION
MAKER
CCI
126
Introductory comments
• Quality parameter: error probability
• (I.e. the costs are:
•
Cii  0; Ci  k  1; i, k  1,2,..., M )
• Erroneous decision may be caused by:
•
additíve noise
•
linear distortion
•
nonlinear distortion
•
additive interference (CCI, ACI)
•
false knowlledge of a parameter
•
e.g. synchronizing error
Frigyes: Hírkelm
127
Introductory comments
• Often it is not one signal of which the error
probability is of interest but of a group of
signals – e.g. of a frame.
• (A secondquality parameter: erroneous
recognition of T : the jitter.)
Frigyes: Hírkelm
128
Transmission of single signals in
additive Gaussian noise
• Among the many sources of error now we
regard only this one
• Model to be investigated:
TIMING (T)
n(t)
SIGNAL
GENERATOR
SOURCE
{mi}, Pi
Frigyes: Hírkelm
mi
+
si(t)
ˆm
DECISION
MAKER
SINK
r(t)= si(t)+n(t)
129
Transmission of single signals in
additive Gaussian noise
•
Specifications:
• a-priori probabilities Pi are known
• support of the real time finctions s t  is
i
•
(0,T)
•
their energy is finite (E: square integral
of the time functions)
mi  si t  relationship is mutual and
•
unique (i.e. there is no error in the
transmitter)
Frigyes: Hírkelm
130