Poularikas A. D.

The Hilbert Transform

The Handbook of Formulas and Tables for Signal Processing.
Ed. Alexander D. Poularikas
Boca Raton: CRC Press LLC,1999
 15
The Hilbert Transform

15.1 The Hilbert Transform

15.2 Spectra of Hilbert Transformation
15.3 Hilbert Transform and Delta Function
15.4 Hilbert Transform of Periodic Signals
15.5 Hilbert Transform Properties and Pairs
15.6 Differentiation of Hilbert Pairs
15.7 Hilbert Transform of Hermite Polynomials
15.8 Hilbert Transform of Product of Analytic Signals
15.9 Hilbert Transform of Bessel Functions
15.10 Instantaneous Amplitude, Phase, and Frequency
15.11 Hilbert Transform and Modulation
15.12 Hilbert Transform and Transfer Functions
of Linear Systems
15.13 The Discrete Hilbert Filter
15.14 Properties of Discrete Hilbert Transform
15.15 Hilbert Transformers (continuous)
15.16 Digital Hilbert Transformers
15.17 IIR Hilbert Transformers
References

15.1 The Hilbert Transform

15.1.1 Definition of Hilbert Transform

1 x( ) x( )

1
(t ) = H{x(t )} = P d = P d
t t

( ) 1 ( )

1
x(t ) = H1 {(t )} = P d = P d
t
t

where P stands for the Cauchy principal value of the integral.

Convolution form representation

1
v( t ) = x ( t )
t
1
x(t ) = (t )
t

1999 by CRC Press LLC

Fourier transform of (t) and x(t) and 1/t (see Table 3.1.3)

V ( ) = X ( )[ j sgn( )]

F 1{ j sgn( ) X ( )} = (t )

F = j sgn( )
1
t

Example
If x(t ) = cos t, then

H{cos t} = (t )

1 cos
=

P
t
d

1 cos[( y + t )]
=

P
y
dy

1
cos y
sin y
=

cos t P y
dy sin t P
y
dy

= sin t.

sin y
The result is due to the fact that cos y / y is an odd function and P
y
dy = .
Example
If x(t ) = pa (t ) then

t
1 d d

a
1
v (t ) = H{pa (t )} = P P
a t t t + t

1 t
1 a
1 t+a
= lim ln( t ) ln( t ) = (t ) = ln
0
a t + ta

Example
If x(t ) = a then

1 t+a
aH{1} = a lim ln =0.
a ta

Hence, if xo = constant is the mean value of a function, then x(t ) = x o + x1 (t ). Therefore H{x o + x1 (t )}
= H{x1 (t )}. This implies that the Hilbert transform cancels the mean value or the DC term in electrical
engineering terminology.

15.1.2 Analytic Signal

A complex signal whose imaginary part is the Hilbert transform of its real part is called the analytic signal.

( z ) = (t, ) = x(t, ) + j(t, ), x and are real functions

1999 by CRC Press LLC

 z = t + j

(t, ) = H{x(t, )}

The function ( z ) = x(t, ) + j(t, ) is analytic if the Cauchy-Riemann conditions

x x
= and =
t t

are satisfied.
Example
The real and imaginary parts of the analytic function

+ t
( z ) = 1 /( jz ) = +j
( + ) 2 + t 2 ( + ) 2 + t 2

satisfy Cauchy-Riemann conditions and, hence, they are Hilbert transform pairs.

(t ) + (t ) (t ) (t )
x (t ) = (t ) = ( = 0)
2 2j

15.2 Spectra of Hilbert Transformation

15.2.1 One-Sided Spectrum of the Analytic Signal

x (t ) + x ( t ) x (t ) x ( t )
x (t ) = x e (t ) + x o (t ) = +
2 2


X ( ) = Xr ( ) + jXi ( ) =

x e (t )cos t dt + j

x o (t )sin t dt

V ( ) = Vr ( ) + jVi ( ) = Spectrum of the Hilbert transform

Vr ( ) = j sgn( )[ jXi ( )] = sgn( ) Xi ( ) (see also 15.1.1)

Vi ( ) = sgn( ) Xr ( )

Example
H{cos t} = sin t, H{sin t} = cos t and, therefore,

j ( t 2 )
H{e jt } = sin t j cos t = j sgn( )e jt = sgn( ) e

Note: The operator j sgn( ) provides a /2 phase lag for all positive frequencies and /2 lead for all
negative frequencies.

15.2.2 Fourier Spectrum of the Analytic Signal

H{x(t )} = (t ); F{x(t )} = X ( ); F{(t )} = j sgn( ) X ( )

1999 by CRC Press LLC

 F{ (t )} = x(t ) + j(t ) = ( ) = X ( ) + jV ( ) = [1 + sgn( )] X ( )

2 >0

1+ sgn( ) = 1 =0
0 <0

Note: The spectrum of the analytic signal is twice that of its Fourier transform at the positive frequency
range 0 < < .
Example
1 t
If (t ) = +j then F{ (t )} = [1 + sgn( )] e where
1 + t2 1 + t2

H{1 /(1 + t 2 )} = t /(1 + t 2 ) and F{1 /(1 + t 2 )} = e .

15.3 Hilbert Transform and Delta Function

15.3.1 Complex Delta Function

If we define 21( f ) = 1( f ) + sgn( f ), then the function (see Fourier transform properties [symmetry] and
function, Chapter 3).

(t ) =


21( f )e jt df =


1( f ) e jt df +


sgn( f ) e jt df

1
= (t ) + j
t

15.3.2 Hilbert Transform of the Delta Function

From (15.3.1) implies

1
H{(t )} =
t

15.4 Hilbert Transform of Periodic Signals

15.4.1 Hilbert Transform of Period Functions

A periodic function can be written in trigonometric form

x p (t ) = Co + C cos(n t + ),
n =1
n o n o = 2 / T , T = period

Therefore we obtain

v p (t ) = H{x p (t )} = C sin(n t + )
n =1
n o n

because the Hilbert transform of a constant is zero (see 15.1.1).

1999 by CRC Press LLC

 A periodic function can also be written in complex form

x p (t ) =
n =
n e jn ot

Therefore,

v p (t ) = H{x p (t )} = H{e
n =
n
jn ot
}= j sgn(n) e
n =
jn ot

15.5 Hilbert Transform Properties and Pairs

15.5.1 Hilbert Transform Properties

TABLE 15.1 Properties of the Hilbert transformation

Original or Inverse
No. Name Hilbert Transform Hilbert Transform

1 Notations x (t ) or H 1 [ ] (t ) or x (t ) or H[ ]

1 ( ) 1 x ( )
2 Time domain definitions
x (t ) =

t d
or
(t ) =
t d
1 1
x (t ) = (t ) (t ) = x (t )
t t
3 Change of symmetry x (t ) = x1e (t ) + x 2 o (t ) ; (t ) = 1o (t ) + 2 e (t )
4 Fourier spectra
F F
x (t ) = X ( ) = X e ( ) + jX o ( ); (t ) = V ( ) = Ve ( ) + jVo ( )

X ( ) = j sgn( ) V ( ); V ( ) = j sgn( ) X ( )

For even functions the Hilbert transform is odd:

X e ( ) = 2
x
o
1e (t ) cos(t ) dt o (t ) = 2
o
X e ( )sin(t ) df

For odd functions the Hilbert transform is even:

X o ( ) = 2
x o
2o (t )sin(t ) dt e (t ) = 2
o
X o ( ) cos(t ) df

5 Linearity ax1 (t ) + bx2 (t ) a1 (t ) + b 2 (t )

6 Scaling and x ( at ); a > 0 ( at )
time reversal x ( at ) ( at )
7 Time shift x (t a ) (t a)
8 Scaling and x (bt a) (bt a)
time shift
Fourier image
9 Iteration H[ x (t )] = (t ) j sgn( ) X ( )
H[ H[ x ]] = x (t ) [ j sgn( )]2 X ( )
H[ H[ H[ x ]]] = (t ) [ j sgn( )]3 X ( )
H[ H[ H[ H[ x ]]]] = x (t ) [ j sgn( )]4 X ( )
e = even; o = odd

1999 by CRC Press LLC

 TABLE 15.1 Properties of the Hilbert transformation (continued)
Original or Inverse
No. Name Hilbert Transform Hilbert Transform

First option
1 1
10 Time derivatives x (t ) = (t ) (t ) = x (t )
t t
Second option
d 1 d 1
x (t ) = (t ) (t ) = x (t )
dt t dt t
x (t ) x 2 (t ) = x1 ( t ) 2 ( t ) =
1
11 Convolution
1 (t ) 2 (t ) 1 (t ) x 2 (t )

12 Autoconvolution
x() x(t ) d = () (t ) d
equality for = 0 energy equality

13 Multiplication by t t x (t ) t (t )

x ( ) d

14 Multiplication of x1 (t ) (low-pass signal) x 2 (t ) (high-pass signal)

signals with non-
overlapping spectra x1 ( t ) x 2 ( t ) x1 ( t ) 2 ( t )
15 Analytic signal (t ) = x (t ) + j H[ x (t )] H[ (t )] = j (t )
16 Product of (t ) = 1 (t ) 2 (t ) H[ (t )] = 1 (t ) H[ 2 (t )]
analytic signals = H[ 1 (t )] 2 (t )
17 Nonlinear x( x ) ( x )
transformations

c c 1

c x (t )
17a y= x1 ( t ) = x 1 (t ) = P dt
bt + a bt + a bt + a t

b b b b
17b y= a+ x1 ( t ) = x a + 1 (t ) = a + ( a)
t t a t

Notice that the nonlinear transformation may change the signal x(t) of finite energy to a signal x1(t)
of infinite energy. P is the Cauchy Principal Value.
18 Asymptotic value as t for even functions of finite support:

x (t) dt
1
x e (t ) = x e ( t ) lim o (t ) = a
t t s
e

a S is support of xe(t)

15.5.2 Iteration
Iteration of the HT two times yields the original signal with reverse sign.
Iteration of the HT four times restores the original signal
In Fourier frequency domain, n-time iteration translates the n-time multiplication by jsgn()

15.5.3 Parsevals Theorem

v(t ) = H{x(t )}

1999 by CRC Press LLC

 F{(t )} = V ( ) = j sgn( ) X ( )

V ( ) = j sgn( ) X ( ) = X ( )
2 2 2

since

Ex =
x 2 (t ) dt =
X ( ) df = energy of x(t )
2

E =
V ( ) df =
X ( ) df = E x
2 2

15.5.4 Orthogonality

(t ) x(t ) dt = 0

15.5.5 Fourier Transform of the Autoconvolution of the Hilbert Pairs

F{x(t ) x(t )} = X 2 ( )

F{(t ) (t )} = [ j sgn( ) X ( )]2 = X 2 ( )

x (t ) x (t ) =

x ( ) x ( t ) d =

( ) (t ) d = (t ) (t )

x1 (t ) x 2 (t ) = 1 (t ) 2 (t )

15.5.6 Hilbert Transform Pairs

TABLE 15.2 Selected Useful Hilbert Pairs

No. Name Function Hilbert Transform

1 sine sin(t ) cos(t )

2 cosine cos(t ) sin(t )
jt
3 Exponential e j sgn( ) e jt
1 t+a
4 Square pulse 2a
(t ) ln
ta

5 Bipolar pulse 2a
(t )sgn(t )
1

ln 1 ( a / t ) 2

1
6 Double triangle t (t )sgn(t ) ln 1 ( a / t ) 2
2a

1 t a t t2
7 Triangle, tri(t ) 1 t / a, t a ln + ln 2
t + a a t a 2
0, t > a

1 t
8 One-sided triangle (1 t / a) ln + 1
ta

1999 by CRC Press LLC

 TABLE 15.2 Selected Useful Hilbert Pairs (continued)
No. Name Function Hilbert Transform

1 b ( a + t )(b t ) t a2 t 2 ( a t )
9 Trapezoid ln + ln 2 2 + ln
b a ( a t )(b + t ) b a b t ( a + t )

a t
10 Cauchy pulse
a2 + t 2 a2 + t 2

e t e f 2
sin(t ) df ; = 2 f
2
11 Gaussian pulse 2
0

1 t a 2t
12 Parabolic pulse 1 (t / a ) 2 , t a

[
1 (t / a )
2
] ln
t+a

a

2a
13 Symmetric e a t 2 sin(t ) df
exponential 0 a2 2

2a
14 Antisymmetric sgn(t ) e a t 2 cos(t ) df
exponential 0 a2 2

a sin(t ) cos(t )
15 One-sided
exponential
1(t ) e a t 2

0 a2 2
df

sin(at ) sin 2 ( at / 2) 1 cos(at )

16 Sinc pulse =
at ( at / 2) at
cos 2 ( t / 2 a); t a

2a 2 sin[ /(2 a)]
17 Video test pulse
0, t > a
2

0 4a 2 2
sin(t ) df

Spectra of a(t ) and cos( 0 t) sin( 0 t ) cos( 0 t )

18 a(t ) cos( 0 t ) a( t ) sin( 0 t ) + a(t ) cos( 0 t )
overlapping t t
19 Bedrosians theorem a(t ) cos( 0 t ) a(t )sin( 0 t )
20 A constant a zero

Hyperbolic Functions: Approximation by Summation of Cauchy Functions (see Hilbert Pairs No. 10 and 45)

No. Name Function Hilbert Transform

( + 0.5)

t
21 Tangent hyp. tanh(t ) = 2 2
= 0
( + 0.5) 2 2 + t 2 = 0
( + 0.5) 2 2 + t 2

( + 0.5)
22 Part of finite
energy of tanh
sgn(t ) tanh(t ) (t ) + 2 = 0
( + 0.5) 2 2 + t 2

1 t
23 Cotangent hyp. coth(t ) = +2 (t ) + 2
t =1
( ) 2 + t 2 =1
( ) 2 + t 2
24 Secans hyp.

( + 0.5)
(1) (1)
( 1) ( 1) t
sech(t ) = 2 2
= 0
( + 0.5) 2 2 + t 2 = 0
( + 0.5) 2 2 + t 2
25 Cosecans hyp.


(1)
1 t
cosech(t ) = 2 ( 1) ( 1) (t ) + 2 ( 1)
t =1
( ) 2 + t 2 =1
( ) 2 + t 2

1999 by CRC Press LLC

 TABLE 15.2 Selected Useful Hilbert Pairs (continued)
No. Name Function Hilbert Transform

Hyperbolic Functions by Inverse Fourier Transformation; = 2 f

2 2
26 sgn(t ) tanh(at / 2) 2

0
a sinh( / a) cos(t ) df

Re a > 0

2 2
27 coth(t ) sgn(t ) 2

0 a
coth( / a) cos(t ) df


2
28 sec h( at / 2) 2

0 a cosh( /(2 a)
sin(t ) df

2
29 cos ech( at / 2) 2
0 a
tanh( /(2 a)) cos(t ) df

2
30 sec h 2 ( at / 2) 2
a sinh( /(2 a))
sin(t ) df

Delta Distribution, 1/(t) Distribution and its Derivatives: Derivation Using Successive Iteration and
Differentiation Iteration

If x (t ) v(t ) then x (t ) v(t ) H[v(t )] = HH[u(t )] = x (t )
No. Operation x (t ) v( t )

31 (t ) 1/( t )
32 Iteration 1/( t ) (t )
33 Differentiation (t ) 1 /( t 2 )

34 Iteration 1 /( t 2 ) (t )
35 Differentiation
(t ) 2 /( t 3 )

36 Iteration 1 /( t 3 ) 0.5
(t )
37 Differentiation
(t ) 6 /( t 4 )

38 Iteration 1 /( t 4 ) (1 / 6)
(t )
39 x (t ) (t ) x (0) /( t )
The procedure could be continued.

Equality of Convolution

1 1
40 (t ) (t ) (t ) = (t )
t t
1 1
41 (t ) (t ) = (t ) = (t )
t 2 t
1 1
42 (t ) (t ) =
(t ) =
(t )
t 2 t 2
6 1 2 1
43
(t ) (t ) =
(t ) =
(t ) (t ) = (t ) = 3 2
t 4 t t t

Approximating Functions of Distributions (see No. 31 to 37 of this table)

x (t ) v( t )
ln(a 2 + t 2 )
(a, t) dt =
1
44 ( a, t ) dt = tan 1 (t / a)
2

1999 by CRC Press LLC

 TABLE 15.2 Selected Useful Hilbert Pairs (continued)
No. Name Function Hilbert Transform

1 a 1 t
45 ( a, t ) = ( a, t ) =
a2 + t 2 a2 + t 2
1 2 at 1 a2 t 2
46 ( a, t ) = ( a, t ) =
(a 2 + t 2 ) 2 (a 2 + t 2 ) 2

1 6at 2 2 a 2 1 2t 2 6at 2
47 ( a, t ) = ( a, t ) =
(a 2 + t 2 ) 3 (a 2 + t 2 ) 3

1 24 a 3 t 24 at 2 1 6t 2 + 36a 2 t 2 6a 4
48 ( a, t ) = ( a, t ) =
(a + t )
2 2 4
(a 2 + t 2 ) 4
Derivation Using Successive Iteration and Differentiation (see the information above No. 31)

Trigonometric Expressions
Operation x (t ) v( t )
sin(at ) 1 cos(at ) 2 sin 2 ( at / 2)
49 =
t t t
cos(at ) sin(at )
50 Iteration (t ) +
t t
sin(at ) 1 cos(at )
51 Differentiation a(t ) +
t2 t2
cos(at ) a sin(at )
52 Iteration (t ) +
t2 t t2
sin(at ) a 2 1 cos(at )
53 Differentiation a (t ) +
t3 2t t3
cos(at ) a 2 a sin(at )
54 Iteration (t ) + (t ) 2 +
t3 2 2 t t3

Selected Useful Hilbert Pairs of Periodic Signals

Name x p (t ) p (t )

(t nT ) cos[( / T )(t nT )]
1
55 Sampling sequence
n =
T n =

56 Even square wave sgn[cos(t )], = 2 / T (2 / ) ln tan(t / 2 + / 4)

57 Odd square wave sgn[sin(t )], = 2 / T (2 / ) ln tan(t / 2)
58 Squared cosine cos (t )
2
0.5 sin(2t )
59 Squared sine sin (t )
2
0.5 sin(2t )
60 Cube cosine cos (t )
3 3
4 sin(t ) + 14 sin(3t )

61 Cube sine sin 3 (t ) 3

4 cos(t ) + 14 cos(3t )
62 cos 4 (t ) 1
2 sin(2t ) + 18 sin(4t )

63 sin 4 (t ) 12 sin(2t ) + 18 sin(4t )

64 cos 5 (t ) 5
8 sin(2t ) + 165 sin(3t ) + 161 sin(5t )
65 cos 6 (t ) 15
32 sin(2t ) + 6
32 sin(4t ) + 1
32 sin(6t )
66 cos(at + ) cos(bt + ) cos(at + )sin(bt + )
0<a<b
, = constants

1999 by CRC Press LLC

 TABLE 15.2 Selected Useful Hilbert Pairs (continued)
No. Name Function Hilbert Transform

67 Fourier Series Xo + X
n =1
n cos(nt + n ) X sin(nt + )
n =1
n n

cot[( / T )(t kT )]
xT = generating function 1
68 Any periodic function x T (t )
T k =

x T (t ) (t kT )
k =

15.6 Differentiation of Hilbert Pairs

15.6.1 Differentiation Pairs

H{x (t )} = (t )

d n x(t ) d n (t )
H n
=
dt dt n

Example

; H{ (t )} = = 2
1 d 1 1
H{(t )} =
t
dt t t

15.6.2 Derivative of Convolution

1
H{x(t )} = H (t ) (t ) =
1
x (t )
t t

d 1
H{x (t )} = H (t ) (t ) = x(t )
d 1
(see 15.6.1 and 15.5.5)
dt t dt t

1
= 2 x(t ) = 2 (t )
1
t t

H{x (t )} = H (t ) (t ) =
1 1
x (t )
t t

15.6.3 Fourier Transform of Hilbert Transform

1
(t ) = x(t ), F{(t )} = j sgn( ) X ( )
t

F{ (t )} = j[ j sgn( ) X ( )] = sgn( ) X ( )

1999 by CRC Press LLC

15.7 Hilbert Transform of Hermite Polynomials

15.7.1 Hermite Polynomials and their Hilbert Transform

2 d n t2
Hn (t ) = ( 1) n e t e n = 0,1, 2,L, < t <
dt n

Hn (t ) = 2t Hn1 (t ) 2(n 1) Hn2 (t ) n = 1, 2,L

F{e t } = e = e
2 2 2 2
f /4

v(t ) = H{x(t )} = H{e t } = F 1{V ( )} =
j sgn( ) e e jt df
2 2 2
f

e
2 2
=2 f
sin t df

H{2 te t } = 2
e
2 2 2
f
cos t df

15.7.2 Table of Hilbert Transform of Hermite Polynomials

TABLE 15.3 Hilbert Transform of Weighted Hermite Polynomials [Notation: x = exp(t 2 )]

Hermite Polynomial Hilbert Transform Energy

n Hn x H( H n x ) E

0 (1) x 2
0
exp( 2 f 2 )sin(t ) df /2

1 (2t ) x 2
exp(
0
2
f 2 ) cos(t ) df /2

2 ( 4t 2 2 ) x 2

0
2
exp( 2 f 2 )sin(t ) df 3 /2

3 (8t 3 12t ) x 2

0
3
exp( 2 f 2 ) cos(t ) df 15 / 2

4 (16t 4 48t 2 + 12) x 2

0
4
exp( 2 f 2 )sin(t ) df 105 / 2

5 (32t 5 160t 3 + 120t ) x 2

0
5
exp( 2 f 2 ) cos(t ) df 945 / 2

n
n H n x = ( 1) n [2tH n 1 (t ) ( 1) n 2

0
n
exp( 2 f 2 )sin t +
2
df

2(n 1) H n 2 (t )]

Energy =
x 2 H n2 dt =

[H ( xH n )]2 dt = 1 3 5 L (2n 1) / 2 , n 1

1999 by CRC Press LLC

15.7.3 Hilbert Transform of Orthonormal Hermite Functions (see Chapter 22)

hn (t ) = (2 n n!) 1 / 2 1 / 2 e t Hn (t )
2
n = 0,1, 2,L

H{hn (t )} = n (t )

2(n 1)! (n 2)!
1/ 2 1/ 2
= hn1 ( ) d (n 1)

1
n! t n1 (t ) n!
n 2 (t )

15.7.4 Hilbert Transform of Orthonormal Hermite Functions

TABLE 15.4 Hilbert Transforms of Orthonormal Hermite Functions (Energy = 1).

Notations: ho (t ), h1 (t ),L ho , h1 ,L; o (t ),L o , 1 ,L

e 2 a = 0.25 e t
2 2 2
g(t ) = f
sin(2 ft ) df ; /2
; b = 0.25
0

Hermite Functions Hilbert Transforms

hn (t ) n (t )

Recurrent Notation

h0 = a 0 = 2 2 bg(t )

2 b
h1 = 2 th0 1 = 2 t 0

h2 = th1 1 / 2 h0 2 = t 1 1 / 2 0

b
h3 = 2 / 3 [th2 h1 ] 3 = 2 / 3 t 2 1

h4 = 1 / 2 th3 3 / 4 h2 4 = 1 / 2 t 3 3 / 4 2

3 b
h5 = 2 / 5 th4 4 / 5 h3 5 = 2 / 5 t 4 4 / 5 3
2
............................................................................................................................................................................
2(n 1)! 2(n 1)!
hn = thn 1 + n = [t n 1
n! n!
(n 2)! (n 2)!
h
1
(n 1) hn 2 n 1 ( ) d] (n 1) n2
n! n!

Nonrecurrent Notation

h0 = a1 2 2 bg(t )

h1 = 2 at 2b[2tg(t ) 1 ]
a
h2 = ( 4t 2 2 ) 2b[(2t 2 1) g(t ) t 1 ]
8

a t2 1
h3 = (8t 3 12t ) 8 / 3b (2 t 3 3t ) g(t ) +
48 2

a t 3 2t
h4 = (16t 4 48t 2 + 12) 4 / 3b (2 t 4 6t 2 + 1.5) g(t ) +
384 2

1999 by CRC Press LLC

 TABLE 15.4 Hilbert Transforms of Orthonormal Hermite Functions (Energy = 1). (continued)
Notations: ho (t ), h1 (t ),L ho , h1 ,L; o (t ),L o , 1 ,L

e 2 a = 0.25 e t
2 2 2
g(t ) = f
sin(2 ft ) df ; /2
; b = 0.25
0

Hermite Functions Hilbert Transforms

hn (t ) n (t )

a (t 4 4t 2 ) + 1.75
h5 = (32t 5 160t 3 + 120t ) 8 / 15 b (2 t 5 10t 3 + 7.5) g(t )
3840
.................................................................................................................................................................
a
hn (t ) = H n (t ), H n (t ) = 2t H n 1 (t ) 2(n 1) H n 2 (t )
2 n n!

n 0 1 2 3 4 5

hn ( ) d 2b 0 b 0 3/ 4 b 0

15.8 Hilbert Transform of Product of Analytic Signals

15.8.1 Hilbert Transform of Product of Analytic Signals:

From

H{ (t )} = H{x(t ) + j(t )} = H{x (t ) + j H{x (t )}} = H{x (t )} jx(t )

= (t ) jx(t ) = j ( x(t ) + j(t )) = j (t )

we obtain H{ 1 (t ) 2 (t )} = j 1 (t ) 2 (t ) = 1 (t ) H{ 2 (t )} = 2 (t ) H{ 1 (t )}
since the product can be considered as an analytic function (t).

15.8.2 The nth Product of an Analytic Signal

H{ 2 (t )} = (t ) H{ (t )} = j 2 (t )

H{ n (t )} = n1 (t ) H{ (t )} = j n (t )

Example
Because H{(1 jt ) 1} = j (1 jt ) 2 , we obtain

H{(1 jt ) 2} = (1 jt ) 1 ( j (1 jt ) 1 ) = j (1 jt ) 2

15.9 Hilbert Transform of Bessel Functions

15.9.1 Hilbert Transform of Bessel Function:

Jn( n ) (t = 0) n

1
H{J n (t )} = Jn (t ) = sin(t sin n) d = t
0
n=0
n!

1999 by CRC Press LLC


1
1 2
J0 (t ) = sin t d
0 (1 2 )1 / 2

0 (t ) = J 0 (t ) + j J0 (t )

2 d 2 d

1 1
1 1 2
J0 (0) = sin(0) = 0, J0(1) (t ) = cos(0) =
0 (1 2 )1 / 2 0 (1 )
2 1/ 2

The parenthesis in the exponent indicates number of differentiations with respect to time.

15.9.2 Hilbert Transform Pairs of Bessel Functions:

TABLE 15.5 Hilbert Transform of Bessel Functions of the First Kind

Bessel Function Fourier Transform Hilbert Transform

J n (t ) Cn ( f ) J n (t ) = H[ J n (t )]

2 1
1
J 0 (t ) C0 = ; <1 C0 ( f )sin(t ) d
(1 2 ) 0.5 0

= 0; > 0

1
1
J1 ( t ) C1 = jC0 C1 ( f ) cos(t ) d
0

1
1
J 2 (t ) C2 = (2 2 1) C0 C2 ( f ) sin(t ) d
0

1
1
J 3 (t ) C3 = j ( 4 3 3 ) C0 C3 ( f ) cos(t ) d
0

1
1
J 4 (t ) C4 = (8 4 8 2 + 1) C0 C4 ( f ) sin(t ) d
0

1
1
J 5 (t ) C5 = j (16 5 20 3 + 5 ) C0 C5 ( f ) cos(t ) d
0

1 1
J 6 (t ) C6 = (32 6 48 4 + 18 2 1) C0 C ( f ) sin(t ) d
0 6
..........................................................................................................................................................................
( 1) n / 2 1
J n (t ) Cn = ( j ) n 2 n 1 Tn ( ) C0
0
Cn ( f ) sin(t ) d

for n = 0,2,4,
( 1) ( n +1) / 2

1
Cn ( f ) cos(t ) d
0

for n = 1,3,5,
Tn ( ) = cos[n cos 1 ( )] is the Chebyshev polynomial

15.10 Instantaneous Amplitude, Phase, and Frequency

15.10.1 Instantaneous Angular Frequency

(t ) = x(t ) + j(t ) = A(t ) e j ( t ) = A(t )cos (t ) + A(t )sin (t )

1999 by CRC Press LLC

 (t )
A(t ) = x 2 (t ) + 2 (t ) , (t ) = tan 1
x (t )

(t ) = (t ) = 2 F(t ) instantaneous angular frequency

(t ) (t )
F(t ) = instantaneous frequency = =
2 2
d (t ) x(t ) (t ) (t ) x (t )
(t ) = tan 1 =
dt x (t ) x 2 (t ) + 2 (t )

15.11 Hilbert Transform and Modulation

15.11.1 Modulated Signal (see 15.10.1)

(t ) = Ao (t ) e j 0 e j 0t

x (t ) = x(t ) + jx (t )

x (t ) + x (t )
x (t ) =
2

15.11.2 Instantaneous Amplitude and Angular Frequency (see 15.10.1)

m m
A(t ) = (t ) = [ x 2 (t ) + x 2 (t )]1 / 2
2 x 2

d x (t )
x (t ) = tan 1
dt x (t )

15.11.3 High-Frequency Analytic Signals (0 = 0)

upper (t ) = upper sideband = x (t ) e j0t

lower (t ) = lower sideband = x (t ) e j0t

x SSB (t ) = x(t )cos 0 t m x (t )sin 0 t

where x(t )cos( 0 t ) and x (t )sin 0 t represent double sideband (DSB) compressed carrier AM signals.

15.12 Hilbert Transform and Transfer Functions of Linear Systems

15.12.1 Causal Systems

H (s) = A(, ) + jB(, ), = + j

B( )

1
A( ) = P d

1999 by CRC Press LLC


A( )

1
B( ) = P d

15.12.2 Minimum Phase Transfer Function

H ( j ) = H ( j ) H ap ( j )

H ( j ) = minimum phase transfer function

H ap ( j ) = all-pass transfer function

H ( j ) = H ( j ) e j ( ) = A ( ) + jB ( )

H ( j ) has all the zeros lying in the left half-plane of the s-plane. The minimum phase transfer function
is analytic and its real and imaginary parts for a Hilbert pair

H{A( )} = B ( )

15.13 The Discrete Hilbert Filter

15.13.1 Discrete Hilbert Filter

j k = 1, 2,L, N2 1

H (k ) = 0 k = 0 and k = N2 ( N = even)
j k = N2 + 1, N2 + 2,L, N 1

H (k ) = j sgn k sgn(k ),
N
k = 0,1,L, N 1 ( N = even)
2

15.13.2 Impulse Response of the Hilbert Filter

N 1 N 1

j sgn 2 k sgn(k) e
1 1 N
h(i) = H (k ) e jw = jw
N k =0
N k =0

N 1
i i 2 ki
sin(w) = N sin 2 cot N ,
1 2
= 2
i = 0,1,L, N 1, w = ( N even)
N k =0
N

15.13.3 DHT of a Sequence x(i) in the Form of Convolution

i i
sin 2 cot ,
2
(i) = x(i) h(i) = x (i) i = 0,1,L, N 1
N 2 N
= circular convolution
N 1

(i) = h(i r) x(r),

r =0
i = 0,1,L, N 1 ( N even)

1999 by CRC Press LLC

15.13.4 DHT of a Sequence x(i) via DFT

FD {x(i)} = X (k )

V (k ) = j sgn k sgn(k ) X (k )
N
2

(i) = FD1{V (k )}, i, k = 0,1, 2,L, N 1 ( N even)

FD discrete Fourier transform, FD1 inverse discrete Fourier transform

15.13.5 Discrete Hilbert Filter when N is odd

j k = 1, 2,L, N21

H (k ) = 0 k=0
j k = N2 + 1, N2 + 2,L, N 1

( N 1) / 2

sin(2ki / N ) ,
2
h(i) = i = 0,1,L, N 1
N k =1

Also

h(i) =
1
1
cos(i)
N cos(i / N )
cot i
N ( )
15.14 Properties of Discrete Hilbert Transform

15.14.1 Parsevals Theorem

N 1 N 1

X (k )
1
E{x(i)} = x(i) =
2 2

i=0
N k =0

E{x(i)} E{(i)}

The reason is that the DC term (average value of x(i)) is eliminated in the DHT.

N 1

x(i) = X(0)
1
x DC =
N i=0

15.14.2 Discrete Hilbert Transform

H D {x AC (i)} = (i)

x AC (i) = x(i) x DC

where xAC(i) is the alternating part of x(i).

1999 by CRC Press LLC

15.14.3 Energies (powers) of xAC and (i)

N 1 N 1 2

2 X ( N / 2)
x AC (i) = (i) +
2
( N even)
i=0 i =1
N

where the special term X is zero, the two energies are equal.
N
2

Example
If x(i) = (i) and N = 8 we obtain (see 15.13.3)

(i) = (i) 14 sin 2 ( i / 2)cot(i / N )

Figure 15.1 shows the desired components and transforms. The xDC = 1/8 = 0.125 and the energies are:
12 12 12 2
E {x(i)} = 1, E{x AC (i)} = 1 = 0.875 , and E{(i)} = 1 = 1 = 0.75.
N N N 8

FIGURE 15.1 (a) The sequence x(i) consisting of a single sample (i), (b) its spectrum X(k) given by the DFT, (c)
the samples of the discrete Hilbert transform, (d) the corresponding spectrum V(k) , (e) the samples of the AC
component of x(i), and (f) the corresponding spectrum XAC(k).

15.14.4 Shifting Property:

FD {x(i m)} = e j 2mk / N X (k )

See 15.13.4

(i) = FD1{ j sgn k sgn(k ) e j 2 mk / N X (k )}

N
2

1999 by CRC Press LLC

15.14.5 Linearity:

H D {ax1 (i) + bx 2 (i)} = a1 (i) + b 2 (i)

15.14.6 Complex Analytic Discrete Sequence:

(i) = x(i) + j(i), (i) = H D {x (i)}

H D { (i)} = X (k ) + j[ j sgn k sgn(k )] X (k ),

N
k = 0,1,L, N 1 ( N even)
2

15.15 Hilbert Transformers (continuous)

15.15.1 Hilbert Transformer (quadratic filter)

1
H ( jf ) = F = H ( f ) e j ( f ) = j sgn f
t

j f >0

H ( jf ) = 0 f =0
j f <0


( f ) = arg H ( jf ) = sgn f
2

15.15.2 Phase-Splitter Hilbert Transformers

Analog Hilbert transformers are mostly implemented in the form of a phase splitter consisting of two
parallel all-pass filters with a common input pot and separated output ports, each having the following
transfer function respectively.

Y1 ( jf ) = e j1 ( f ) , Y2 ( jf ) = e j2 ( f )

with

( f ) = 1 ( f ) 2 ( f ) = / 2 for all f > 0

15.15.3 All-Pass Filters

R jX ( )
H ( j ) = = 2 f
R + jX ( )

2 RX ( )
( ) = arg{( R jX ( )) 2} = tan 1 2
R X ( )
2

See Figure 15.2a.

1999 by CRC Press LLC

FIGURE 15.2 An all-pass consisting of (a) a low-pass and a complementary high-pass, (b) a first-order RC low-
pass and complementary CR high-pass, and (c) a second-order RLC low-pass and complementary RLC high-pass.

1
If X ( ) = , then (see Figure 15.2b)
C

2 y
( y) = tan 1 2
, y = RC =
1 y

If X ( ) = L 1 / C, then (see Figure 15.2c)

2(1 y 2 ) qy
( y) = tan 1 2 2
, y = / r , r = 1 / LC
(1 y ) q y
2 2

q = r RC = R C / L

1999 by CRC Press LLC

15.15.4 Design Hilbert Phase Splitters
Example
Filter with two first-order all-pass filters in each branch. The phase function for the first branch is (see
Figure 15.3)

2 y 1 2 ay
1 ( f ) = tan 1 2 + tan a 2 y 2 1 , y = 2 fRC
y 1

FIGURE 15.3 Phase Hilbert splitter with two all-pass filters.

Find a to get the best linearity of 1 ( f ) in the logarithmic scale. Small changes of a introduce a
trade-off between the RMS phase error and the pass-band of the Hilbert transformer. Find shift parameter
b to yield the minimum RMS phase error

2by 1 2 aby
2 ( f ) = tan 1 2 2 + tan a 2 b 2 y 2 1
b y 1

Figure 15.4 shows an example with a = 0.08 and b = 0.24 giving the normalized edge frequencies
y1 = 1.6 and y2 = 30 ( f2 / f1 = 18.75, or more than 4 octaves) with RMS = 0.016.

15.16 Digital Hilbert Transformers

15.16.1 Digital Hilbert Transformers

Ideal discrete-time Hilbert transformer is defined as an all-pass with a pure imaginary transfer function.

H (e j ) = Hr ( ) + j Hi ( )

Hr ( ) = 0 for all f

j 0 < < 2
j
H (e ) = j Hi ( ) = 0 = 0, =
j < < 0

Equivalent Notation

H (e j ) = j sgn(sin ) = sgn(sin ) e j / 2 = H ( ) e j arg H ( )

1 0< <
H( ) = sgn(sin ) =
0 = 0, =

1999 by CRC Press LLC

FIGURE 15.4 The phase functions and the phase error of the Hilbert transformer of Figure 15.3.

arg[ H ( )] = sgn(sin )
2

= 2 fn , fn = f / fs , fs = sampling frequency

Noncausal impulse response of the ideal Hilbert transformer is

i
sin 2
2
h(i) = i = 0, 1, 2,L
i 2

15.16.2 Ideal Hilbert Transformer With Linear Phase Term

je j 0<<

H (e j ) = 0 = 0, =
je j ( 2 ) < < 2


sin 2 (i )
h(i) =
2 2 i = 0, 1, 2,L
i
h(i) = h( i) i = 0,1, 2,L

1999 by CRC Press LLC

15.16.3 FIR Hilbert Transformers:
Figure 15.5 shows a noncausal impulse response Hilbert transformer and its truncated and shifted version
so that a causal one is generated.

FIGURE 15.5 Impulse responses of (a) the ideal discrete time Hilbert transformer (see 15.16.1) and (b) a FIR
Hilbert transformer given by the truncation and shifting of the impulse response shown in (a).

Causal Filter Impulse Response

N 1

H (i1 ) = h (i ) z
i1 = 0
1 1
i1

N 1
h1 i + = h(i) i1 = i + N 1
, i = N21 ,L0,L, N21
2
2

Transfer function =

N 1
N 1
2 f
h(i) e j2h(i)sin(i),
2
j j N21 ji j N21
H (e ) = e =e =
i = N21 i =1
fs

Amplitude of Hilbert Transformer (see Figure 15.6)

N 1

2h(i)sin(i)
2
j
G( e ) =
i =1

1999 by CRC Press LLC

FIGURE 15.6 The G(ej) function of an FIR Hilbert transformer (amplitude).

Normalized Dimensionless Pass-band Hilbert Transformer

W = 2 1 = 2 , 1 , 2 = edge frequencies

2
Wf [ H z ] = fs
2

15.17 IIR Hilbert Transformers

15.17.1 IIR Ideal Hilbert Transformer (see Figure 15.7)

H HB ( z ) = 1 + z 1 G( z 2 ) ideal half band filter (see Figure 15.7a)

G( z 2 ) = all pass filter with unit magnitude

H H ( z ) = z 1 G( z 2 ) ideal IIR Hilbert transformer

F( z ) = z 1 G( z 2 ), z = e j (see Figure 15.7b)

F(e j ) = e j e jG ( ) = e j( )

( ) = 0.5 [sgn(sin(2 )) sgn ]

G ( ) = ( ) + (see Figure 15.7e)

H H (e j ) = e j e jG ( 0.5 + ) , z 2 = e j 2 , z 2 = e j 2( 0.5 + )

arg{z 1G( z 2 )} = + G (0.5 + ) (see Figure 15.7g)

IIR Hilbert transformer has an equi-ripple phase function and exact amplitude. A noncausal transfer
function may have the form

1 ai z 2
N

H(z) = z 1

i =1
z 2 ai

1999 by CRC Press LLC

FIGURE 15.7 Step-by-step derivation of the IIR transfer function of a Hilbert transformer Z1G(z2), starting from
the transfer function of the ideal half-band filter given by 1 + Z1G(z2)

Example
Let 1 = 0.02 low-frequency edge, 2 = 0.98 = high-frequency edge ( = 0.02 ), phase equi-
ripple amplitude 0.01 . Because = sin(0.5 ), = 0.0157. Using the procedure from Ansari
(1985), we find a(1) = 5.36078, a(2) = 1.2655, a(3) = 0.94167, and a( 4) = 0.53239. Inserting ais, in
H(z) above, we find the phase function.

1999 by CRC Press LLC

References
Ansari, R., IIR discrete-time Hilbert transformers, IEEE Trans., ASSP-33, 1146-1150, 1985.
Erdelyi, A., Tables of Integral Transform, McGraw-Hill Book Co. Inc., New York, NY, 1954.
Hahn, Stefan L., Hilbert Transforms, in Transforms and Applications Handbook, Ed. Alexander D.
Poularikas, CRC Press Inc., Boca Raton, FL, 1996.

1999 by CRC Press LLC