Part 3 Hilbert Transform and

Canonical Representation of Signals

Hilbert Transform
o How to obtain g+(t)?
o Answer: Hilbert Transformer.
Proof: Observe that
ì 1, f >0
ï
2u( f ) = 1 + sgn( f ), where sgn( f ) = í 0, f =0
ï- 1, f <0
î
Then by the next slide, we learn that

© Po-Ning [email protected] 3-2

By extended Fourier transform,

0
a|f |+j2 f t a|f |+j2 f t a|f |+j2 f t
sgn(f )e df = e df e df
0
0
(a j2 t)f
= e df e(a+j2 t)f
df
0
1 1
=
a j2 t a + j2 t
j4 t
=
a 2 + 4 2 t2
⎧ j
InverseFourier
4π t ⎪ , t≠0
sgn( f ) → lim j 2 2 2
= ⎨ πt
a↓0 a + 4π t ⎪
⎩ 0, t = 0
InverseFourier
j
2u( f ) = 1 + sgn( f ) ® d (t ) + × 1{t ¹ 0}
pt
© Po-Ning [email protected] 3-3
3
 Hilbert Transform
g + (t ) = Fourier -1{2u( f )G ( f )}
= Fourier -1{2u( f )} * Fourier -1{G ( f )}
æ 1 ö
= ç d (t ) + j × 1{t ¹ 0}÷ * g (t )
è pt ø
1
= g (t ) + j × 1{t ¹ 0} * g (t )
pt
= g (t ) + jgˆ (t ),
¥ g (t )
where gˆ (t ) = ò dt is named the Hilbert Transform of g (t ).
- ¥ p (t - t )

© Po-Ning [email protected] 3-4

Hilbert Transform

g( t ) 1 ˆ t)
g(
h(t ) =
pt

ì1, f >0
1 Fourier
ï
h(t ) = ® H ( f ) = - j sgn( f ), where sgn( f ) = í0, f =0
pt ï- 1,
î f <0

ì| G ( f ) | exp{ j[ÐG ( f ) - p / 2]}, f >0

ï
Þ Gˆ ( f ) = - j sgn( f ) × G ( f ) = í 0, f =0
ï| G ( f ) | exp{ j[ÐG ( f ) + p / 2]}, f <0
î

© Po-Ning [email protected] 3-5

Hilbert Transform
n Hence, Hilbert
Transform is
basically a 90
degree phase
shifter.

© Po-Ning [email protected] 3-6

Hilbert Transform
ì 1 ¥ g (t )
ïï gˆ (t ) = p ò-¥ t - t dt
Hilbert Transform Pair í
ï g (t ) = - 1 ¥ gˆ (t )
ïî ò
p -¥ t - t
dt

g( t ) 1 ˆ t)
g(
h(t ) =
pt

ˆ t)
g( 1 g( t )
h(t ) = Ä
pt
-1
© Po-Ning [email protected] 3-7
Hilbert Transform
o An important property of Hilbert Transform is that:

g (t ) and gˆ (t ) are orthogonal in the sense of Integration.

¥
In other words, ò
-¥
g (t ) gˆ (t )dt = 0.
(See the proof in the next slide.)

The real and imaginary parts of g+ (t) = g(t) + jĝ(t)

are orthogonal to each other.
(Examples of Hilbert Transform Pairs can be found in Table A6.4.)

© Po-Ning [email protected] 3-8

 ¥ ¥
æ ¥
ö÷dt
ò-¥ ò-¥ çè ò-¥ ˆ j 2pft
g ( t ) ˆ
g ( t ) dt = g ( t ) G ( f ) e df
ø
= ò G ( f )ç ò g (t )e j 2pft dt ö÷df
æ
¥ ¥
ˆ
-¥ è -¥ ø
¥
= ò Gˆ ( f )G ( - f )df
-¥
¥
= - j ò sgn( f )G ( f )G ( - f )df
-¥

= - j ç ò G ( f )G ( - f )df - ò G ( f )G ( - f )df ö÷
æ ¥ 0

è 0 -¥ ø
= - j ç ò G ( f )G ( - f )df - ò G ( - f )G ( f )df ö÷
æ ¥ ¥

è 0 0 ø
¥
= 0, if ò G( f )G( - f )df < ¥.
0

Note that G(f )G( f ) = G(f )G⇤ (f ) = |G(f )|2 .

© Po-Ning [email protected] 3-9
Complex Representation
of Signals and Systems
o g+(t) is named the pre-envelope, or analytical
signal, of g(t).
o We can similarly define

g - (t ) = g (t ) - jgˆ (t )

2(1 - u( f ))G ( f )
G( f )

© Po-Ning [email protected] 3-10

Canonical Representation of
Bandpass Signals

o Now let G(f) be a narrow-

band signal for which 2W
<< fc.
n Then we can obtain its pre-
envelope G+(f).
n Afterwards, we can shift the
pre-envelope to its low-pass
~
signal G ( f ) = G+ ( f + f c )

© Po-Ning [email protected] 3-11

Canonical Representation
of Bandpass Signal
o These steps give the relation between the
complex lowpass signal (baseband signal) and
the real bandpass signal (passband signal).

g (t ) = Re{g + (t )} = Re{g~(t ) exp( j 2pf c t )}

n Quite often, the real and imaginary parts of

complex lowpass signal are respectively denoted by
gI(t) and gQ(t).

© Po-Ning [email protected] 3-12

Canonical Representation
of Bandpass Signal
o In terminology,
ì g + (t ) pre - envelope
ï g~(t ) complex envelope
ï
í
ï g I (t ) in - phase component of the band - pass signal g (t )
ïî gQ (t ) quadrature component of the band - pass signal g (t )

This leads to a canonical, or standard, expression for g(t).

g(t) = Re {(gI (t) + jgQ (t))exp( j2π fc t)}
= gI (t)cos(2π fc t) − gQ (t)sin(2π fc t)
© Po-Ning [email protected] 3-13
 ( g I (t ) + jgQ (t )) exp( j 2pf ct )

Complex domain

© Po-Ning [email protected] 3-14

Canonical Representation
of Bandpass Signal
o Canonical transmitter
g (t ) = g I (t ) cos(2pf c t ) - gQ (t ) sin(2pf c t )

OSC

-90 phase
shifter

© Po-Ning [email protected] 3-15

 Canonical Representation of Bandpass
Signal 2g(t) cos(2⇡fc t)
= 2gI (t) cos2 (2⇡fc t) 2gQ (t) sin(2⇡fc t) cos(2⇡fc t)
o Canonical = gI (t) + gI (t) cos(4⇡fc t) gQ (t) sin(4⇡fc t)
receiver Low-pass
filter

OSC

2g(t) sin(2⇡fc t)
= 2gI (t) sin(2⇡fc t) cos(2⇡fc t) 90 phase
shifter
+2gQ (t) sin2 (2⇡fc t)
= gQ (t) gQ (t) cos(4⇡fc t) Low-pass
gI (t) sin(4⇡fc t) filter

© Po-Ning [email protected] 3-16

More Terminology
ì g + (t ) pre - envelope
ï g~(t ) complex envelope
ï
í
ï g I (t ) in - phase component of the band - pass signal g (t )
ïî gQ (t ) quadrature component of the band - pass signal g (t )
ì
ïa (t ) =| g (t ) |=| g~(t ) |= g 2 (t ) + g 2 (t )
ï + I Q
ï
í natural envelope or envelope of g (t )
ï
ïf (t ) = tan -1 æç gQ (t ) ö÷ phase of g (t )
ïî ç g (t ) ÷
è I ø
© Po-Ning [email protected] 3-17
Bandpass System
o Consider the case of passing a bandpass signal
x(t) through a real LTI filter h(t) to yield an
output y(t).
¥
x (t ) y (t ) = ò h(t ) x (t - t )dt
h(t ) -¥

o Can we have a lowpass equivalent system for

the bandpass system?
© Po-Ning [email protected] 3-18
o Similar to the previous analysis, we have:
Assumption : The spectrum of x (t ) is limited to within ± W Hz of the
carrier frequency f c , and W < f c .
ìï x (t ) = Re{ ~ x (t )e j 2pf t }= x I (t ) cos(2pf c t ) - xQ (t ) sin(2pf c t )
c

í~
ïî x (t ) = x I (t ) + jxQ (t )

Assumption : The spectrum of h(t ) is limited to within ± B Hz of the

carrier frequency f c , and B < f c .
{ ~
}
ìïh(t ) = Re h (t )e j 2pf t = hI (t ) cos(2pf ct ) - hQ (t ) sin(2pf ct )
c

í~
ïîh (t ) = hI (t ) + jhQ (t ) complex impulse response

© Po-Ning [email protected] 3-19

o Now, is the filter output y(t) also a bandpass
signal?
Y (f ) = X(f )H(f )
The spectrum of y(t) is limited to within ± min{W, B} Hz of the
carrier frequency fc , provided that max{W, B} < fc

y(t) = Re ỹ(t)ej2 fc t = yI (t) cos(2 fc t) yQ (t) sin(2 fc t)

ỹ(t) = yI (t) + jyQ (t)

X( f )

H( f )

Y( f )
© Po-Ning [email protected] 3-20
Bandpass System
o Question: Is the following system valid?
¥ ~
~
x (t ) ~
y (t ) = ò h (t ) ~
x (t - t )dt
~ -¥
h (t )

n The advantage of the above equivalent system is that there is

no need to deal with the carrier frequency in the system
analysis.
o The answer to the question is YES (with some
modification)! It will be substantiated in the sequel.

© Po-Ning [email protected] 3-21

 ~ ~ ~
It suffices to show that Y ( f ) = X ( f ) H ( f ).
~
ìY+ ( f ) = 2u( f )Y ( f ) ì Y ( f ) = Y+ ( f + f c )
ï ïï ~
Observe that í X + ( f ) = 2u( f ) X ( f ) and í X ( f ) = X + ( f + f c ).
ï H ( f ) = 2u ( f ) H ( f ) ï~
î + ïî H ( f ) = H + ( f + f c )
Consequently,
~ ~
X ( f )H ( f ) = X + ( f + fc )H + ( f + fc )
= (2u( f + f c ) X ( f + f c ) )(2u( f + f c ) H ( f + f c ) )
= 4u ( f + f c ) X ( f + f c ) H ( f + f c )
= 4u( f + f c )Y ( f + f c )
= 2Y+ ( f + f c )
~ There is an additional multiplicative
= 2Y ( f )
constant 2 at the output!
© Po-Ning [email protected] 3-22
Bandpass System
o Conclusion:
~ ~ 1 ¥~ ~
x (t ) y (t ) = ò h (t ) x (t - t )dt
~ 2 -¥
h (t )

© Po-Ning [email protected] 3-23

Bandpass System
o Final note on bandpass system
n Some books define H+(f) = u(f)H(f) for a filter,
instead of H+(f) = 2u(f)H(f) (as for a signal).
n As a result of this definition (i.e., H+(f) = u(f)H(f)),
~
{
h(t ) = 2 Re h (t )e j 2pf t
c
} ~ ¥ ~
and y (t ) = ò h (t ) ~
-¥
x (t - t )dt

n It is justifiable to remove 2 in H+(f) = u(f)H(f),

because a filter is used to filter out the signal; hence,
it is not necessary to make the total area constant.

© Po-Ning [email protected] 3-24

Representation of Narrowband Noise using In-
Phase and Quadrature Components

o The bandpass system representation discussed

previously is based on deterministic signals.
o How about a random process? Can we have a
low-pass isomorphism system to a bandpass
random process.
n Take the noise process N(t) as an example.
¥
Y (t ) = ò h(t ) N (t - t )dt
N (t ) h(t ) -¥

© Po-Ning [email protected] 3-25

Representation of Narrowband Noise using In-
Phase and Quadrature Components
o A WSS real-valued zero-mean noise process N(t)
is a bandpass process if its PSD SN(f) ¹ 0 only for
|f - fc| < B and |f + fc| < B, and also B < fc.
n Similar to the analysis for deterministic signals, let

ìï N (t ) = N I (t ) cos(2pf c t ) - N Q (t ) sin(2pf c t )
í~
ïî N (t ) = N I (t ) + jN Q (t )
for some joint zero-mean WSS of NI(t) and NQ(t).
n Notably, the joint WSS of NI(t) and NQ(t)
immediately imply WSS of N~ (t ).

© Po-Ning [email protected] 3-26

PSD of NI(t ) and NQ(t)
o First, we note that joint WSS of NI(t ) and NQ(t)
and the WSS of N(t) imply:
ìï RN (t ) = RN (t )
I Q

í (See the proof in the sequel.)

ïî RN , N (t ) = - RN , N (t )
I Q Q I

© Po-Ning [email protected] 3-27

RN ( ) = E[N (t + )N (t)]
= E[(NI (t + ) cos(2 fc (t + )) NQ (t + ) sin(2 fc (t + )))
(NI (t) cos(2 fc t) NQ (t) sin(2 fc t))]
= RNI ( ) cos(2 fc (t + )) cos(2 fc t)
+RNQ ( ) sin(2 fc (t + )) sin(2 fc t)
RNI ,NQ ( ) cos(2 fc (t + )) sin(2 fc t)
RNQ ,NI ( ) sin(2 fc (t + )) cos(2 fc t)
cos(2 fc ) + cos(2 fc (2t + ))
= RN I ( )
2
cos(2 fc ) cos(2 fc (2t + ))
+RNQ ( )
2
sin(2 fc (2t + )) sin(2 fc )
RNI ,NQ ( )
2
sin(2 fc (2t + )) + sin(2 fc )
RNQ ,NI ( )
2

© Po-Ning [email protected] 3-28

 (Continue from the previous slide.) These two terms
must equal zero,
1 since RN(t) is not
RN (t ) = [ RN (t ) + RN (t )] cos(2pf ct )
2 I Q
a function of t.
1
+ [ RN , N (t ) - RN , N (t )] sin(2pf ct )
2 I Q Q I

1
+ [ RN (t ) - RN (t )] cos(2pf c ( 2t + t ))
2 I Q

1
- [ RN , N (t ) + RN , N (t )] sin(2pf c ( 2t + t ))
2 I Q Q I

ìï RN (t ) = RN (t )
Þí (Property 1)
I Q

ïî RN , N (t ) = - RN , N (t )
I Q Q I

Þ RN (t ) = RN (t ) cos(2pf ct ) - RN
I Q ,N I (t ) sin(2pf ct ) (Property 2)
© Po-Ning [email protected] 3-29
PSD of NI(t ) and NQ(t)

n Some other properties

(Property 3)

© Po-Ning [email protected] 3-30

PSD of NI(t ) and NQ(t)
Properties 2 and 3 jointly imply

(Property 4)

o As the inconsistency encountered in Slide 3-24,

n some books define

n As a result of the two “inconsistent” definitions, a “simpler”

relation is obtained:
n For consistence, we let

© Po-Ning [email protected] 3-31

Summary of Spectrum Properties

By RN~ (t ) = RN*~ ( -t ).
[From 1.]

[From 4.
See the next slide.]
© Po-Ning [email protected] 3-32
© Po-Ning [email protected] 3-33
8. 𝑁! (𝑡) and 𝑁" (𝑡) are orthogonal.

ìï RN , N (t ) = - RN , N (t )
I Q Q I

í
ïî RN , N ( -t ) = E[ N I (t ) N Q (t + t )] = RN
I Q Q ,N I (t )
Þ RN , N (t ) = - RN , N ( -t )
I Q I Q

Þ RN I ,NQ (0) = - RN I ,NQ ( 0)

Þ RN I ,NQ
(0) = E[ N I (t ) N Q (t )] = 0.

© Po-Ning [email protected] 3-34

-34
 RV (t , u ) = E [VI (t )VI (u )]
I

= 4 E [ N (t ) cos(2pf c t ) N (u ) cos(2pf c u )]
= 4 RN (t , u ) cos(2pf c t ) cos(2pf c u )

Notably, VI (t ) is not WSS.

© Po-Ning [email protected] 3-35

SV ( f ) = S N ( f - f c ) + S N ( f + f c )
I

ì1, | f |£ B
S N ( f ) =| H ( f ) | SV ( f ), where | H ( f ) | = í
2 2

î0, f > B
I I

© Po-Ning [email protected] 3-36

Þ S N ( f ) = S N ( f ) = [ S N ( f - f c ) + S N ( f + f c )] | H ( f ) |2
I I

ì S N ( f - f c ) + S N ( f + f c ), | f |< B
=í
î 0, otherwise
Similarly,
ì S N ( f - f c ) + S N ( f + f c ), | f |< B
SN ( f ) = í
Q
î 0, otherwise

This result (namely, SNI (f ) = SNQ (f )) coincides with Property 1,

for which RNI (⌧ ) = RNQ (⌧ ).

© Po-Ning [email protected] 3-37

 RV ,V (t , u ) = E [VI (t )VQ (u )]
I Q

= -4 E [N (t ) cos(2pf c t ) N (u ) sin(2pf c u )]
= -4 RN (t , u ) cos(2pf c t ) sin(2pf c u )
RV ,V (t + t , t ) = -4 RN (t ) cos(2pf c (t + t )) sin(2pf c t )
I Q

© Po-Ning [email protected] 3-38

SV ,V ( f ) = j[ S N ( f + f c ) - S N ( f - f c )]
I Q

RN ,NQ (t , u ) = E [N I (t ) N Q (u )]

[ ]
I

¥ ¥
= E ò-¥ hI (t 1 )VI (t - t 1 )dt 1 ×ò-¥ hQ (t 2 )VQ (u - t 2 )dt 2
¥ ¥
= ò-¥ ò-¥ hI (t 1 )hQ (t 2 ) E[VI (t - t 1 )VQ (u - t 2 )]dt 1dt 2
¥ ¥
= ò-¥ ò-¥ hI (t 1 )hQ (t 2 ) RV ,V (t - t 1 , u - t 2 )dt 1dt 2
I Q

© Po-Ning [email protected] 3-39

R̄NI ,NQ ( )
T
1
= lim RNI ,NQ (t + , t)dt
T 2T T
T
1
= lim hI ( 1 )hQ ( 2 )RVI ,VQ (t + 1, t 2 )d 1 d 2 dt
T 2T T
T
1
= hI ( 1 )hQ ( 2 ) lim RVI ,VQ (t + 1, t 2 )dtd 1 d 2
T 2T T

= hI ( 1 )hQ ( 2 )R̄VI ,VQ ( 1 + 2 )d 1 d 2

© Po-Ning [email protected] 3-40

 ¥ ¥ ¥
SN I ,NQ ( f ) = ò-¥ ò-¥ ò-¥ hI (t 1 )hQ (t 2 ) RV ,V (t 2 + t - t 1 )e - j 2pft dtdt 1dt 2
I Q

¥ ¥ ¥
= ò-¥ ò-¥ ò-¥ hI (t 1 )hQ (t 2 ) RV ,V (u )e - j 2pf ( u -t
I Q
2 +t 1 )
dudt 1dt 2 ,
(Let u = t 2 + t - t 1.)
= H I ( f ) H Q ( - f ) SV ,V ( f )
I Q
⇤
Here, HQ ( f ) = HQ (f ).

Take HI (f ) = HQ (f ) = H(f ).
Then HI (f )HQ ( f ) = H(f )H( f ) = H(f )H (f ) = |H(f )|2 .

SN I ,NQ ( f ) = SN I ,NQ (f)

= j[ S N ( f + f c ) - S N ( f - f c )] | H ( f ) |2
ì j[ S N ( f + f c ) - S N ( f - f c )], | f |< B
=í
î 0, otherwise

© Po-Ning [email protected] 3-41

Ideal Bandpass Filtered White Noise
Revisited

© Po-Ning [email protected] 3-42

Representation of Narrowband Noise using
Envelope and Phase Components
o Now we turn to envelope R(t) and phase Y(t)
components of a random process of the form
N (t ) = N I (t ) cos(2pf c t ) - N Q (t ) sin(2pf c t )
= R (t ) cos[2pf c t + Y (t )]

where R (t ) = N I2 (t ) + N Q2 (t ) and Y (t ) = tan -1[ N Q (t ) / N I (t )].

© Po-Ning [email protected] 3-43

Pdf of R(t) and Y(t)
o Assume that N(t) is a white Gaussian process
with two-sided PSD s2 = N0/2.
o For convenience, let NI and NQ be snapshot
samples of NI(t) and NQ(t).
o Then
1 æ n 2
+ n Q ö
2

fN ( n I , nQ ) = expçç - I
÷
÷
2ps s
,NQ 2 2
I
è 2 ø

© Po-Ning [email protected] 3-44

 Pdf of R(t) and Y(t)
o By nI = r cos(y) and nQ = r sin(y),
dn I dnQ
1 æ n I2 + nQ2 ö 1 æ r 2 ö dr dr drdy
ò
A ( n I ,nQ )
2ps 2
expçç -
è 2s ø2
÷dn I dnQ = ò
÷
A ( r ,y )
2ps 2
expçç - ÷
2 ÷ dn
è 2s ø I
dnQ
dy dy
1 æ r2 ö
= ò expçç - ÷ rdrdy
2 ÷
A ( r ,y )
2ps 2
è 2s ø

æ r2 ö 1
r r æ r2 ö
So f R ,Y ( r,y ) = expçç - ÷=
2 ÷
´ 2 expçç - ÷.
2 ÷
2ps 2
è 2s ø 2p s è 2s ø
© Po-Ning [email protected] 3-45
Pdf of R(t) and Y(t)
o R and Y are therefore independent.
ì r æ r2 ö
ï f R ( r ) = 2 expçç - ÷ for r ³ 0. Rayleigh distribution.
2 ÷
ï s è 2s ø
í
ï f (y ) = 1 for 0 £ y < 2p .
ïî Y 2p
Normalized Rayleigh
distribution with s2 = 1.

© Po-Ning [email protected] 3-46

Sine Wave + Gaussian Noise
o Now suppose the Gaussian white noise is added
to a sinusoid of amplitude A.
o Then

n Uncorrelation for Gaussian nI(t) and nQ(t) implies

their independence.

© Po-Ning [email protected] 3-47

Sine Wave + Gaussian Noise
o This gives the pdf of xI(t) and xQ(t) as:
1 é ( x I - A) 2 + xQ2 ù
fX ( x I , xQ ) = exp ê - ú
2ps s
,XQ 2 2
I
ë 2 û
o By xI = r cos(y) and xQ = r sin(y), dxI dxQ
1 ⎡ (r cos(ψ ) − A)2 + r 2 sin 2 (ψ ) ⎤ dr dr
fR,Ψ (r, ψ ) = 2
exp ⎢− 2 ⎥
2πσ ⎣ 2σ ⎦ dxI dxQ
dψ dψ

r ⎛ r 2 + A 2 − 2Ar cos(ψ ) ⎞
= 2
exp ⎜ − 2 ⎟
2πσ ⎝ 2σ ⎠
© Po-Ning [email protected] 3-48
Sine Wave + Gaussian Noise
o Notably, in this case, R and Y are no longer
independent.
o We are more interested in the marginal
distribution of R.
2p
f R ( r ) = ò f R ,Y ( r,y )dy
0

2p r æ r 2 + A2 - 2 Ar cos(y ) ö
=ò expçç - ÷÷dy
0 2ps 2
è 2s 2
ø

© Po-Ning [email protected] 3-49

Sine Wave + Gaussian Noise
r æ r 2 + A2 ö 2p æ 2 Ar cos(y ) ö
f R (r) = expçç - ÷÷ ò expç ÷dy
2ps 2
è 2s ø
2 0
è 2s 2
ø
r æ r 2 + A2 ö æ Ar ö
= 2 expçç - ÷÷ I 0 ç 2 ÷,
s è 2s ø è s ø
2

1 2p
where I 0 ( x ) =
2p ò0
exp( x cos(y ))dy is the modified Bessel function

of the first kind of zero order.

This distribution is named the Rician distribution.

© Po-Ning [email protected] 3-50

Normalized Rician distribution
r A
Let v = and a = .
æ v2 + a2 ö
fV ( v ) = v × expçç - ÷÷ I 0 (av ),
è 2 ø

© Po-Ning [email protected] 3-51

 Bessel Functions
o Bessel’s equation of order n
2
d y dy
x 2
2
+ x + ( x 2
- n 2
)y = 0
dx dx

o Its solution Jn(x) is the Bessel function of the first kind of

order n.
1 π
J n (x) =
π
∫ 0
cos(x sin(θ ) − nθ )dθ
m n+2m
1 π
∞
exp ( jx sin θ − jnθ ) dθ = ∑
(−1) ( x / 2)
=
2π
∫ −π
m!(n + m)!
m=0

© Po-Ning [email protected] 3-52

Properties of the Bessel Function
1. J n ( x ) = ( -1) n J -n ( x ) = ( -1) n J n ( - x )
2n xn
2. J n -1 ( x ) + J n +1 ( x ) = J n ( x ) 3. When x small, J n ( x ) » n .
x 2 n!
2 æ p np ö
4. When x large, J n ( x ) » cosç x - - ÷
px è 4 2 ø
¥
5. lim J n ( x ) = 0.
n ®¥
6. åJ
n = -¥
n ( x ) exp( jnf ) = exp( jx sin(f ))
¥
7. å n ( x ) = 1.
J 2

n = -¥

© Po-Ning [email protected] 3-53

Modified Bessel Function
o Modified Bessel’s equation of order n
d2y dy
x 2
2
+ x + ( j 2 2
x - n 2
)y = 0
dx dx
o Its solution In(x) is the Modified Bessel function of the first
kind of order n.
−n 1 π
I n (x) = j J n ( jx) =
2π
∫ −π
exp(x cos(θ ))cos(nθ )dθ
n+2m

=∑
∞
( x / 2)
m=0 m!(n + m)!

n The modified Bessel function is monotonically increasing

in x for all n.
© Po-Ning [email protected] 3-54
 Properties of Modified Bessel Function

I2(x)
I1(x)
I0(x)
ì1, n = 0
3'. lim I n ( x ) = í .
x ¯0
î0, n ³ 1
exp( x )
4'. When x large, I n ( x ) » .
2px
¥
6'. å I ( x ) exp( jnf ) = exp( x cos(f ))
n = -¥
n

© Po-Ning [email protected] 3-55

Flat-Fading Channel
o Model of a multi-path channel
N

A cos(2pf c t ) å A cos(2pf t + Q )
k =1
k c k

Assume N paths.
Transmitter Receiver

© Po-Ning [email protected] 3-56

Flat-Fading Channel

N N
where YI = å Ak cos(Qk ) and YQ = å Ak sin(Qk ).
k =1 k =1

~ ~
Þ Input X (t ) = A induces output Y (t ) = YI + jYQ ,
which is independent of t.

Assume {( Ak , Qk )} i.i.d., and Ak uniform over [ -1,+1)

and Qk uniform over [0,2p ).

© Po-Ning [email protected] 3-57

Flat-Fading Channel
o By Central Limit Theorem,
YI A1 cos(Q1 ) + ! + AN cos(Q N )
= ® Normal (0,1)
N /6 N /6

So YI is approximately Gaussian distributed with

mean 0 and variance ( N / 6).

For any Gaussian random variable G , we have

© Po-Ning [email protected] 3-58

Flat-Fading Channel

© Po-Ning [email protected] 3-59

Flat-Fading Channel
n Normality Test: We can therefore use b1 and b2 to examine the
degree of resemblance to Gaussian for a random variable.
S N = (U1 + ! + U N ) / a N for zero - mean i.i.d. {U k }
~ E[ S N4 ] E[(U1 + ! + U N ) 4 ] NE[U 4 ] + 3N ( N - 1) E 2 [U 2 ] E[U 4 ] / E 2 [U 2 ] - 3
Þ b2 = 2 2 = 2 = = 3+
E [ S N ] E [(U1 + ! + U N ) 2 ] N 2 E 2 [U 2 ] N
ì E[ A cos(Q)] = E[ A sin(Q)] = 0 ⎧ N = 10 100
ï 1000 10000
Þ í E[ A2 cos2 ( Q)] = E[ A2 sin 2 ( Q)] = 1 / 6 ⎨
ï E[ A4 cos4 ( Q)] = E[ A4 sin 4 ( Q)] = 3 / 40 ⎩ 2.97 2.997 2.9997 2.99997
î

© Po-Ning [email protected] 3-60

Summary
o Narrowband process
n Hilbert transform
n Bandpass system
o Gaussian, Rayleigh and Rician
n Central Limit Theorem

© Po-Ning [email protected] 3-61