Вы находитесь на странице: 1из 43

HANKEL TRANSFORMS

M. RAHMAN
The object of this article is to introduce integral transform
of a particular type, called the Hankel transform, and to
illustrate the use of this method by means of examples.
The treatment is that of a review article and as such is not
meant to be exhaustive; its aim is to give a concatenated
account of known results rather than present new ones.
The emphasis throughout is on those results that are of
frequent occurrence in boundary value problems of math-
ematical physics, but some indication is also given for
possible theoretical investigations.
Proofs are either omitted entirely or only the key steps
are outlined. Readers interested in rigorous proofs of some
of the statements in this article are referred to the books
by Sneddon [1,2], Davies [3], Andrews and Shivamoggi [4],
and Zayed [5].
The organization of the article is as follows. In the rst
section, we illustrate the motivation behind introducing
the Hankel transform and then give a precise denition of
the Hankel transform and its inversion. The next two
sections are devoted to the derivation of some basic
properties of Hankel transforms. In the following section,
we explore the connection between Fourier and Hankel
transforms. Parsevals relation for Hankel transforms is
then deduced. We next introduce the modied operator of
Hankel transforms. An overview of ErdelyiKober opera-
tors and their generalization by Sneddon and Cooke is
given. We then derive Beltrami-type relations and give a
brief account of their generalization by Sneddon. An
extensive account is given of the applications of Erdelyi
Kober and Cooke operators to dual, triple, and quadruple
integral equations involving Hankel transforms. A num-
ber of issues that arise in connection with applications of
Hankel transforms to many physical problems is then
addressed. For the convenience of the readers, a compen-
dium is given in the last section of the basic theorems and
formulas of Hankel transforms that are of frequent occur-
rence in applications.
1. THE HANKEL TRANSFORM
The Hankel transform arises naturally as a result of using
the method of separation of variables to boundary value
problems of mathematical physics in cylindrical coordi-
nates, for example, boundary value problems for the La-
place and Helmholtz equations involving half-spaces and
regions bounded by parallel planes. In general, applica-
tion of this technique is relevant to problems leading to
the integration of equations of the type
@
2
f
@r
2

1
r
@f
@r

v
2
r
2
fLf=f (r; . . .)
where L is a linear operator that does not contain r, and
f (r, y) is a prescribed function.
To illustrate this, let us consider the axisymmetric
solution j(r, z) of Laplaces equation
@
2
f
@r
2

1
r
@f
@r

@
2
f
@z
2
=0 (1)
in the half-space r40, z40, which satises the boundary
condition
f(r; 0) =f (r) (2)
where f(r) is a prescribed function of r.
In addition, the solution of the problem must satisfy the
regularity conditions so that the eld decays as R-N,
where R=

r
2
z
2
_
.
Assuming that the solution can be represented in the
separated-variable form
f(r; z) =f
1
(r)f
2
(z)
we nd that Eq. (1) reduces to
1
f
1
d
2
f
1
dr
2

1
f
1
r
df
1
dr
=
1
f
2
d
2
f
2
dz
2
(3)
Since the left-hand side of Eq. (3) depends only on r while
the right-hand side only on z, we conclude that they must be
equal to a constant, say, l = s
2
, where s is a real quantity.
Thus, we obtain two ordinary differential equations
d
2
f
1
dr
2

1
r
df
1
dr
s
2
f
1
=0
df
2
dz
2
s
2
f
2
=0
(4)
The rst of these equations is that of Bessel [6], whose
solution bounded at the origin is
f
1
(r) =A
1
(s)J
0
(sr)
where A
1
(s) is an arbitrary function of s and J
0
(sr) is the
zeroth-order Bessel function of the rst kind.
On the other hand, the solution of the second relation of
Eq. (4) ensuring a decaying eld is given by
f
2
(z) =A
2
(s) e
sz
Therefore, the solution of Eq. (1) is
f(r; z) =A(s)J
0
(sr) e
sz
(5)
where A(s) is an arbitrary function of s. Readers can easily
verify that the other cases, namely, l =0 and l =s
2
(s is a
real quantity), must be ignored, since they do not ensure a
decaying eld as R o.
H
1893
The solution of Eq. (5) has the property that, if s40,
f(r; z) 0 as R o. By simple superposition, we can
therefore construct the solution of the form
f(r; z) =
_
o
0
sA(s) J
0
(sr) e
sz
ds (6)
The condition of Eq. (2) will be satised if
f (r) =
_
o
0
sA(s) J
0
(sr) ds (7)
yielding an equation for determining the unknown function
A(s). It will be shown later that A(s) is given by the formula
A(s) =
_
o
0
rf (r) J
0
(sr) dr (8)
which on substitution into Eq. (6) then formally gives the
solution of our problem.
The formulas in Eqs. (7) and (8) dene a transformation
pair called the Hankel transform of order zero. We now
give a formal denition of the Hankel transform of an
arbitrary order of a function.
Given a real function f(r) dened in the interval (0, N),
suppose that
1. f(r) is piecewise continuous and of bounded variation
in every nite subinterval [a, b], where 0oaoboN
2. The integral
_
o
0

r
_
[f (r)[droo
Then, the Hankel transformof the vth order of the function
f(r) satisfying the preceding conditions is dened as
~
ff
v
(s) =
_
o
0
rf (r) J
v
(sr) dr (9)
which we shall write as
~
ff
v
(s) =
v
[f (r); r s] (10)
Sometimes, for the sake of brevity, we shall write this
notation as
v
[f (r); s];
v
[f (r)], or simply
v
f (r).
Readers should note that since the kernel of the Hankel
transform is the Bessel function, the theory of Hankel
transforms relies heavily on the theory of the Bessel
functions. Perhaps, for this reason, in some literature,
this transform is called Bessel transformation or Fourier
Bessel transformation.
The Hankel inversion theorem states that if the func-
tion f(r) satises the preceding conditions, then
_
o
0
s
~
ff
v
(s) J
v
(sr) ds =f (r) (11)
If the function has a jump discontinuity at a point, then the
right-hand side of Eq. (11) should be replaced by the sum
1
2
[ f (r 0) f (r 0)]
We shall not give a proof of the Hankel inversion theorem
here. Interested readers are referred to the book by
Sneddon [2].
It follows from Eqs. (10) and (11) that
f (r) =
_
o
0
sJ
v
(sr) ds
_
o
0
r
0
f (r
0
) J
v
(sr
0
) dr
0
0oroo; v >
1
2
(12)
Equation (12) is called Hankels integral theorem.
Evidently, Eq. (11) can be written as
f (r) =
1
v
[
~
ff
v
(s); s r]
which, in the notation of Eq. (10), is equivalent to
f (r) =
v
[
~
ff
v
(s); s r]
hence establishing the rule
v
=
1
v
. Thus, we see that
if v >
1
2
, there is a symmetric relationship between a
function and its Hankel transform of order v, in the sense
that if
~
ff v(s) is the Hankel transform of order v of a function
f(r), then f(r) is the Hankel transform of order v of
~
ff v(s).
Extensive tables have been constructed of the Hankel
direct and inverse transforms of functions usually encoun-
tered in applications [e.g., 7].
As in the case of other types of integral transforms,
the use of Hankel transform has many advantages,
for example, it is applicable to both homogeneous and
inhomogeneous problems, it simplies calculations and
singles out the purely computational part of the solution,
and it allows us to construct an operational calculus for
a given kernel by using tables of direct and inverse
transforms.
An extensive account of applications of the Hankel
transform as well as other integral transforms to problems
in mathematical physics was given by Sneddon [1,2,8] and
Lebedev et al. [9]. Perhaps it is Sneddon who may quite
justiably be regarded as the most ardent proponent of
using the method of integral transformsin particular,
Hankel transformto various boundary value problems of
mathematical physics.
2. SOME ELEMENTARY PROPERTIES OF HANKEL
TRANSFORMS
Property 1.

m
[f (r); r s]
=(1)
m

m
[f (r); r s]
(m= 1; 2; . . . ; n; . . .)
1894 HANKEL TRANSFORMS
Proof of this property follows from the fact that [6]
J
m
(sr) =(1)
m
J
m
(sr)
Property 2.

v
[f (ar); r s] =a
2

v
f (r); r
s
a
_ _
Proof: By denition, we have

v
[ f (ar); r s] =
_
o
0
rf (ar) J
v
(sr) dr (13)
By making a change of variable ar =r, we reduce the
integral in Eq. (13) to the form

v
[ f (ar); r s] =a
2
_
o
0
rf (r) J
v
(sa
1
r) dr
=a
2

v
f (r); r
s
a
_ _
Property 3.

v
[r
1
f (r); r s]
=
s
2v
[
~
ff
v1
(s)
~
ff
v 1
(s)] (vO0)
Proof: From the recurrence relation for the Bessel func-
tions [6]
j
v1
(x)
2v
x
J
v
(x) J
v 1
(x) =0
we deduce

v
[r
1
f (r); r s] =
_
o
0
f (r) J
v
(sr) dr
=
s
2v
_
o
0
rf (r) J
v1
(sr) dr
_

_
o
0
rf (r) J
v 1
(sr) dr
_
=
s
2v
[
~
ff
v1
(s)
~
ff
v 1
(s)]
Property 4. The shift formula for the Hankel transforms is

n
[ f (r a)H(r a); r s] =

o
m=o
a
m
~
ff
m
(s)
where
a
m
=J
nm
(sa)
1
2
as (m1)
1
J
nm1
(sa)
_
(m1)
1
J
nm1
(sa)

Proof of this property is given in the book of Sneddon [2].


It should be mentioned here that it is not possible to
obtain a simple shift formula for the Hankel transforms.
This is primarily because the addition formula for the
Bessel functions, that is, the NeumannLommel addition
formula [6]
J
n
(x y) =

o
m=o
J
m
(x)J
nm
(y)
is much more complicated than the addition formula for
the exponential functions e
x
and e
ix
for the Laplace and
Fourier transforms.
3. THE HANKEL TRANSFORMS OF DERIVATIVES OF A
FUNCTION
In applications of Hankel transforms to physical problems,
it is necessary to have expressions for the Hankel trans-
forms of the derivatives of a function or a combination of
them, through the Hankel transforms of the function
itself. Using the denition of Hankels transform and the
formula for integrating by parts, we obtain

v
df
dr
; s
_ _
=
_
o
0
r
df
dr
J
v
(sr) dr
=[rf (r)J
v
(sr)]
o
0

_
o
0
@
@r
[rJ
v
(sr)]f (r) dr
(14)
The rst term on the right vanishes provided that the
function f(r) is such that
lim
r0
r
v 1
f (r) =0; lim
ro

r
_
f (r) =0
It follows from the arguments leading to the proof of the
Hankel inversion theorem [2] that the second of these
conditions holds for any f(r) whose Hankel transform
exists. Therefore, the rst term on the right in Eq. (14)
vanishes if
f (r) =o(r
v1
); r 0
where o is the Landaus symbol of order.
From the theory of Bessel functions [6,10], we have
@
@r
rJ
v
(sr) [ ] =J
v
(sr) rJ
/
v
(sr)
J
/
v
(sr) =srJ
v1
(sr) vJ
v
(sr)
HANKEL TRANSFORMS 1895
so that Eq. (14) now takes the following form:

v
df
dr
; r
_ _
=(v 1)
_
o
0
f (r) J
v
(sr) dr
s
_
o
0
rf (r)J
v1
(sr) dr
(15)
However, the integral on the right is the (v 1)th-order
Hankel transform of f(r):
_
o
0
rf (r) J
v1
(sr) dr =
v1
[f (r); r s]
Thus, Eq. (15) takes the form

v
df
dr
; r s
_ _
=(v 1)
_
o
0
f (r) J
v
(sr) dr
s
v1
[f (r); r s]
(16)
The rst term on the right is obviously the vth-order
Hankel transform of the function r
1
f(r). However, our
objective is to express everything in terms of the Hankel
transform of the function f(r). This can be achieved by
utilizing the following relation [10]:
J
v
(sr) =
1
2v
J
v1
(sr) J
v 1
(sr) [ ] (17)
Inserting Eq. (17) into Eq. (16), after some arrangements,
we nally obtain the following important relationship:

v
df
dr
; r s
_ _
= s
v 1
2v

v1
[f (r); r s]
s
v 1
2v

v 1
[f (r); r s]
(18)
Expressions for Hankel transforms of the higher deriva-
tives of the function f(r) may be deduced by repeated
application of the formula in Eq. (18). For instance, we
obtain

v
d
2
f
dr
2
; r s
_ _
=
s
2
(v 1)
4(v 1)

v2
[f (r)]

s
2
(v
2
3)
2(v
2
1)

v
[f (r)]

s
2
(v 1)
4(v 1)

v 2
[f (r)]
(19)
In applications of Hankel transforms to many physical
problems, it becomes necessary to have available the
formula for Hankel transform of the differential operator:
M
v
=
d
2
dr
2

1
r
d
dr

v
2
r
2
Integrating by parts and assuming that df =dr =o(r
1
), we
nd
_
o
0
r
d
2
f
dr
2
J
v
(sr) dr =
_
o
0
df
dr
d
dr
rJ
v
(sr) [ ] dr
so that
_
o
0
r
d
2
f
dr
2

1
r
df
dr
_ _
J
v
(sr) dr = s
_
o
0
df
dr
rJ
/
v
(sr) dr
=s
_
o
0
f (r)
d
dr
rJ
/
v
(sr)
_
dr
(20)
Equation (20) was derived on the assumption that the
function rf(r)-0 as r-0 or N.
We know from the theory of Bessel functions [6,10] that
the function J
v
(sr) satises the differential equation
d
dr
rJ
/
v
(sr)
_
= s
2

v
2
r
2
_ _
rJ
v
(sr) (21)
On substitution of Eq. (21) into Eq. (20), we obtain the
following formula:
_
o
0
r
d
2
f
dr
2

1
r
df
dr

v
2
r
2
f
_ _
J
v
(sr) dr = s
2
_
o
0
rf (r) J
v
(sr) dr = s
2

v
[ f (r); r s]
(22)
An immediate consequence of Eq. (22) is the formula
_
o
0
r
d
2
f
dr
2

1
r
df
dr
_ _
J
0
(sr) dr
= s
2

0
[f (r); r s]
(23)
To illustrate the use of the properties of Hankel trans-
forms, let us consider the classic problem of determining
the potential at any point in the eld induced by an
electried disk of radius a, whose potential is raised to
f
0
(f
0
is a constant). The problem is known as Webers
problem. A discussion of this problem can be found in the
books by Jeans [11] and Smythe [12]. The problem reduces
to that of solving Laplaces equation in Eq. (1) with the
boundary conditions
f(r; 0) =f
0;
0 _ roa
@f
@z

z =0
=0; r > a
(24)
1896 HANKEL TRANSFORMS
The second boundary condition in Eq. (24) expresses the
symmetry of the eld with respect to the plane of the disk,
that is, the plane z =0.
To solve the problem, we use the zeroth-order Hankel
transform of the function f(r, z):
f(r; z) =
0
[
~
ff(s; z); s r] (25)
Applying the transformation in Eq. (25) to Eq. (1) and
making use of the relation of Eq. (23), we obtain the
following ordinary differential equation
d
2 ~
ff
dz
2
s
2
~
ff=0
whose solution is
~
ff(s; z) =A(s) e
sz
B(s) e
sz
(26)
where A(s) and B(s) are some unknown functions of s.
Because of symmetry, it is sufcient to consider the
half-space zZ0 only. Then, since the eld must vanish at
innity (regularity conditions), we must set B=0, so that
Eq. (26) reduces to
~
ff(s; z) =A(s) e
sz
Therefore, our formal solution of the problem takes the
form
f(r; z) =
0
[A(s) e
sz
; s r] (27)
Utilizing the boundary conditions in Eq. (24), we get the
following equations to determine the unknown function
A(s):

0
[A(s); s r] =f
0
; 0 _ roa

0
[sA(s); s r] =0; r > a
or writing in integral form
_
o
0
sA(s) J
0
(sr) ds =f
0
; 0 _ roa
_
o
0
s
2
A(s) J
0
(sr) ds =0; r > a
(28)
Equations of the type in Eq. (28) are called dual-integral
equations. A systematic treatment of this kind of equa-
tions will be discussed later. Here, we give a rather
heuristic solution. Gradshteyn and Ryzhik [13] provide
the following integrals:
_
o
0
sin s
s
J
0
(sr) ds =
p
2
; 0 _ roa
_
o
0
(sin s) J
0
(sr) ds =0; r > a
(29)
A comparison of Eqs. (28) and with Eqs. (29) shows that
the solution for A(s) is
A(s) =
2f
0
p
sin s
s
(30)
Putting Eq. (30) into Eq. (27), we obtain the solution of our
problem as
f(r; z) =
2f
0
p
_
o
0
sin s
s
J
0
(sr) e
sz
ds (31)
The uniqueness of Eq. (31) follows from the physical
contents of the problem.
4. RELATION BETWEEN FOURIER AND HANKEL
TRANSFORMS
In this section, the relationship between Hankel and
Fourier transforms of a function of two variables is
explored. Specically, we shall see that there exists a close
relationship between the double Fourier transform of a
function of two variables of a particular type and its
Hankel transform.
Consider a function f(x
1
,x
2
) that is a function of
r =x
2
1
x
2
2
only. The double Fourier transform F(a
1
, a
2
) is
F(a
1
; a
2
) =
1
2p
_
o
o
_
o
o
f (

x
2
1
x
2
2
_
)
e
i(a
1
x
1
a
2
x
2
)
dx
1
dx
2
(32)
If we make the substitutions into Eq. (32)
x
1
=r cos y; x
2
=r sin y;
a
1
=s cos j; a
2
=s sin j
then, since
dx
1
dx
2
=rdrdy; a
1
x
1
a
2
x
2
=rs cos(y j)
the double integral in Eq. (32) reduces to
F(a
1
; a
2
) =
1
2p
_
o
0
rf (r) dr
_
2p
0
e
irs cos(yj)
dy (33)
Since the inner integral on the right is 2p-periodic, it does
not depend on j, that is
_
2p
0
e
irs cos(yj)
dy =
_
2p
0
e
irs cos y
dy
which is equal to 2pJ
0
(rs) [6,10], where s =

a
2
1
a
2
2
_
. We
therefore see that the function F(a
1
, a
2
) is a function of
s only and may be written as
F(s) =
_
o
0
rf (r) J
0
(sr)dr =
0
[f (r); r s] (34)
HANKEL TRANSFORMS 1897
which, of course, is the zeroth-order Hankel transform of
f(r). On the other hand, by the Fourier inversion theorem,
we have
f (x
1
; x
2
) =
1
2p
_
o
o
_
o
o
F(a
1
; a
2
)
e
i(a
1
x
1
a
2
x
2
)
da
1
da
2
Using the same substitution as before, the preceding
expression can be reduced to the following formula:
f (r) =
_
o
0
sF(s) J
0
(sr) ds =
0
[F(s); s r] (35)
Formulas in Eqs. (34) and (35) obviously express the
Hankel inversion theorem in the special case where v =0.
The preceding results can be easily generalized in case
of n-dimensional Fourier transforms. If the functions f(x
1
,
x
2
, y, x
n
) is function only of r =

x
2
1
x
2
2
x
2
n
_
, then
its Fourier transform F(a
1
, a
2
, y, a
n
) is a function of s only
where s =

a
2
1
a
2
2
a
2
n
_
. More specically, the fol-
lowing relationship holds:
s
(n1)=2
F(s) =
_
o
0
r[r
(n1)=2
f (r)] J
(n1)=2
(sr) dr (36)
For proof, the readers are referred to the book by Sneddon
[1].
It therefore follows from Eq. (36) that s
(n1)=2
F(s) is the
Hankel transform of order (n 1)/2 of the function
r
(n1)=2
f (r).
Similarly, by n-dimensional Fourier inversion theorem,
it can be shown that
r
(n1)=2
f (r) =
_
o
0
s[s
(n1)=2
F(s)] J
(n1)=2
(sr) ds (37)
If we write
f(r) =r
(n1)=2
f (r);
~
ff
v
(s) =s
(n1)=2
F(s);
v =
n 1
2
then Eqs. (36) and (37) take the following form:
~
ff
v
(s) =
_
o
0
rf(r) J
v
(sr) dr =
v
[f(r); r s]
f(r) =
_
o
0
s
~
ff
v
(s) J
v
(sr) ds =
v
[
~
ff
v
(s); s r]
These formulas obviously dene the vth-order Hankel
transformation pair for the function f(r).
5. PARSEVALS RELATION FOR HANKEL TRANSFORMS
Suppose that
~
ff
v
(s) =
v
[f (r); r s]; ~ gg
v
(s) =
v
[g(r); r s]
Then, putting formally, we obtain the equation
_
o
0
s
~
ff
v
(s) ~ gg
v
(s) ds =
_
o
0
s
~
ff
v
(s) ds
_
o
0
xg(x) J
v
(sx) dx
=
_
o
0
xg(x) dx
_
o
0
s
~
ff
v
(s) J
v
(sx) ds
(38)
in which the inner integral, by Hankels inversion theo-
rem, is obviously equal to f(r).
Equation (38) then yields the following formula:
_
o
0
s
~
ff
v
(s) ~ gg
v
(s) ds =
_
o
0
xf (x)g(x) dx (39)
The expression Eq. (39) is evidently the Parseval relation
for the Hankel transform. As in the case of other integral
transforms, such a Fourier, Laplace, Mellin, and Kantorovich-
Lebedev transforms, Parsevals relation is a very useful
too in many theoretical can practical investigations.
It should be noted here that a general Parseval relation
involving Hankel transforms of two functions of different
orders does not exist. This is primarily because the
NeumannRahman formula [6,14] for the product of two
rst-kind Bessel functions of different orders
J
mn
(sr)J
n
(sr
0
)
=
1
p
_
p
0
cos(nj)T
m
r r
0
cos j
R
_ _ _

r
0
sin nj sin j
R
U
m1
r r
0
cos j
R
_ __
J
m
(R) dj; U
1
( ) =0 (40)
where R=

r
2
r
2
0
2rr
0
cos j
_
; T
m
( ) and U
m1
( )
are the Chebyshev polynomials of the rst and second
kinds, respectively, is much more complicated than the
simplest rule for the product of two exponential functions
(kernels of Laplace and Fourier transforms) of different
powers.
As an example of application of Parsevals relation in
Eq. (39), let us evaluate the integral

v
[x
2
J
v
(ax); x s]; v >
1
2
Taking f (x) =x
v
H(a x)(a > 0) and g(x) =x
v
H(b x)
(b > 0), where (. . .) is the step function, we have
~
ff
v
(s) =
_
a
0
x
v 1
J
v
(sx) dx;
~ gg
v
(s) =
_
b
0
x
v 1
J
v
(sx) dx
1898 HANKEL TRANSFORMS
These integrals are easily evaluated [13] as
~
ff
v
(s) =
a
v 1
s
J
v 1
(sa);
~ gg
v
(s) =
b
v 1
s
J
v 1
(sb)
Now, using Parsevals relation in Eq. (39), we obtain
(ab)
v 1
_
o
0
s
1
J
v 1
(sa) J
v 1
(sb) ds
=
_
min(a;b)
0
x
2v 1
dx
Assuming that 0oaob, we nd that (13)
_
o
0
s
1
J
v 1
(sa) J
v 1
(sb) ds
=
1
2(v 1)
a
b
_ _
v 1
0oaob; v >
1
2
It therefore follows from the preceding equation that

v
x
2
J
v 1
(ax); x s
_
=
1
2v
s
a
_ _
v
; 0osoa
1
2v
a
s
_ _
v
; s > a
_

_
where, v >
1
2
.
6. THE HANKEL OPERATOR
In many theoretical investigations, it is more convenient
to use a modied operator of Hankel transform S
Z,a
instead of the operator
v
. This modied Hankel operator
is dened by the formula
S
Z;a
f (t); x [ ] =2
a
x
a

2Z a
t
a
f (t); t x [ ] (41)
so that
S
Z;a
f (t); x [ ] =2
a
x
a
_
o
0
t
1a
f (t) J
2Z a
(xt) dt (42)
If we write
~
ff
Z;a
(x) =S
Z;a
[f (t); x] (43)
then from Eq. (41), we obtain

2Z a
[t
a
f (t); x] =2
a
x
a
~
ff
Z;a
(x) (44)
Applying Hankels inversion, we deduce from Eq. (43) that
f (t) =2
a
t
a

2Z a
[x
a
~
ff
Z;a
(x); t]
or writing out the above expression in full, we obtain
f (t) =S
Z a;a
[
~
ff
Z;a
(x); t]
thus establishing the rule
S
1
Z;a
=S
Z a;a
(45)
In applications, the following relationship is useful
S
Z;a
f (x) =2
l
x
l
S
Zl=2;a l
[x
l
f (x)]
the validity of which can be easily proved by writing
out both sides of the equation using the denition in
Eq. (42).
7. THE ERDELYIKOBER OPERATORS OF FRACTIONAL
INTEGRATION
In this section, we present a brief exposition of the so-
called ErdelyiKober operators of fractional integrations
[1517] and their generalization due to Sneddon and
Erdelyi [8,18] and Cooke [19,20]. We next illustrate appli-
cations of these operators to the solution of dual-, triple
and quadruple-integral equations involving Hankel trans-
forms, which arise in many boundary value problems of
mathematical physics, especially electrostatics and elec-
tromagnetic scattering. The description here closely fol-
lows Sneddon [21].
In a series of papers [1517], Erdelyi and Kober in-
vestigated properties of the fractional integral
x
Za 1
G(a)
_
x
0
(x t)
a1
t
Z1
f (t) dt (a > 0; Z > 0)
which is a generalization of Riemanns integral
1
G(a)
_
x
0
(x t)
a1
f (t) dt
and Weyls integral
x
n
G(a)
_
o
x
(t x)
a1
t
aZ
f (t) dt (a > 0; Z > 0)
7.1. Denitions and Basic Results
If a > 0; Z >
1
2
, we dene the operator I
Z,a
by the equation
I
Z;a
f (x)
=
2x
2a2Z
G(a)
_
x
0
(x
2
u
2
)
a1
u
2Z 1
f (u) du
HANKEL TRANSFORMS 1899
I
Z
,
0
is the identity operator, and if ao0, we dene I
Z,a
by
the relation
I
Z;a
f (x) =x
2Z2a1
D
n
x
x
2Z 2a 2n1
I
Z;a n
f (x)
where n is a positive integer such that 0oano1 and D
x
is the differential operator
D
x
=
d
dx
x
1
Similarly, a > 0; Z >
1
2
, we dene the operator K
Z,a
by the
equation
K
Z;a
f (x)
=
2x
2Z
G(a)
_
o
x
(u
2
x
2
)
a1
u
2a2Z 1
f (u) du
K
Z,0
is the identity operator, and if ao0, we dene K
Z,a
by
the equation
K
Z;a
f (x) =(1)
n
x
2Z1
D
n
x
x
2n2 1
K
Zn;a n
f (x)
Operators I
Z,a
and K
Z,a
are called ErdelyiKober operators.
We next establish some properties of these operators. If
we assume that a40, b40, we have
I
Z;a
I
Z a;b
f (x)
=
2x
2Z2a
G(a)
_
x
0
(x
2
u
2
)
a1
u
2Z 1
du
2u
2Z2a2b
G(b)

_
u
0
(u
2
t
2
)
B1
t
2Z 2a 1
f (t) dt
Interchanging the order of integration and using the
result (13)
2
_
x
t
x
2
u
2
_ _
a1
(u
2
t
2
)
b1
u
2a2b 1
du
=
G(a)G(b)
G a b ( )
t
2a
x
2b
(x
2
t
2
)
a b1
we obtain
I
Z;a
I
Z a;b
f (x)
=
2x
2Z2ab
G a b ( )
_
x
0
t
2Z 1
(x
2
t
2
)
a b1
f (t) dt
The expression on the right is equal to I
Z,a b
, which
follows from its denition, thus establishing the rule
I
Z;a
I
Z a;b
=I
Z;a b
(46)
Similarly, it can be shown that
K
Z;a
K
Z a;b
=K
Z;a b
(47)
The preceding relations are valid for a40, b40, but it is a
simple exercise to show that they are also valid for
negative values of a and b. Also, it can be shown from
the theory of integral equations of Abel type [8] that the
inverse of the ErdelyiKober operators are given by the
formulas
I
1
Z;a
=I
Z a;a
; K
1
Z;a
=K
Z a;a
(48)
The following formulas hold, whose validity can be proved
very easily:
I
Z;a
{x
2b
f (x)] =x
2b
I
Z b;a
f (x)
K
Z;a
{x
2b
f (x)] =x
2b
K
Z b;a
f (x)
The following relationships hold between the Erdelyi
Kober and Hankel operators:
I
Z a;b
S
Z;a
=S
Z;a b
; K
Z;a
S
Z a;b
=S
Z;a b
S
Z a;b
S
Z;a
=I
Z;a b
S
Z;a
S
Z a;b
=K
Z;a b
S
Z a;b
I
Z;a
=S
Z;a b
; S
Z;a
K
Z a;b
=S
Z;a b
(49)
The proofs of these identities are based on the properx-
ties of Bessel functions and are given in the book by
Davies [3].
7.2. The Cooke Operators
Cooke [19,20] has dened the operators
b
a
_ _
I
Z;a
and
d
c
_ _
K
Z;a
by the formulas
b
a
_
_
_
_
I
Z;a
f (x)
=
2x
2a2Z
G(a)
_
b
a
(x
2
u
2
)
a1
u
2Z 1
f (u) du; a > 0
f (x); a =0
x
2a2Z1
G(1a)
d
dx
_
b
a
(x
2
u
2
)
a
u
2Z 1
f (u) du; 1oao0
_

_
(50)
1900 HANKEL TRANSFORMS
for 0oaoboNand
d
c
_
_
_
_
K
Z;a
f (x)
=
2x
2Z
G(a)
_
d
c
(u
2
x
2
)
a1
u
2a2Z 1
f (u) du; a > 0
f (x); a =0
x
2Z1
G(1 a)
d
dx
_
d
c
(u
2
x
2
)
a
u
2a2Z 1
f (u) du; 1oao0
_

_
(51)
for 0oxocod. It will be observed that these operators are
related to the ErdelyiKober operators by the relations
x
0
_ _
I
Z;a
=I
Z;a
;
o
x
_ _
K
Z;a
=I
Z;a
Cook [19,20] also dened the operators L and M by the
equations
x; b
c; a
_
_
_
_
L
Z;a
f (x) =
x
c
_
_
_
_
I
1
Z;a
b
a
_
_
_
_
I
Z;a
f (x)
d; b
x; a
_
_
_
_
M
Z;a
f (x) =
d
x
_
_
_
_
K
1
Z;a
b
a
_
_
_
_
K
Z;a
f (x)
(52)
and showed that if aoboco x, then
x; b
c; a
_
_
_
_
L
Z;a
f (x) =
2 sin(pa)
p
x
2Z
(x
2
c
2
)
a

_
b
a
(c
2
t
2
)
a
t
2 1
x
2
t
2
f (t) dt
(53)
and that if xodoaob, then
d; b
x; a
_
_
_
_
M
Z;a
f (x) =
2 sin(pa)
p
x
2Z 2a
(d
2
x
2
)
a

_
b
a
(t
2
d
2
)
a
t
2a2Z 1
t
2
x
2
f (t) dt
(54)
8. BELTRAMI-TYPE RELATIONS
A classic problem of electrostatics concerns that of deter-
mining the potential of the electrostatic eld due to a
circular disk whose potential is prescribed. One way to
solve this problem is to determine the charge density q on
the disk and then to calculate the potential at any eld
point r by evaluating the integral
_
S
q(R
/
)
RR
/
[ [
dS
/
over the surface of the disk. In the case of axisymmetry,
that is, when the prescribed potential f(r) is a function of r
only, Beltrami [22] showed that the density of the surface
charge is given by the formula
q(r) =
1
pr
d
dr
_
o
r
xdx

x
2
r
2
_
d
dx

_
x
0
yf(y) dy

x
2
y
2
_ ; 0 _ r _ a
(55)
where a is the radius of the disk.
Sneddon [23] showed that Beltramis relation in Eq.
(55) is a special case of a general relation between Hankel
transforms. In particular, he showed that the expression

m
[s
d

v
f (s); r]
can be expressed as a double integral involving f(r), which
is a generalization of the integral occurring on the right-
hand side of Beltramis relation in Eq. (55). By assigning
particular values of the parameters m, d, and v, we can
deduce relations that are of interest in the investigations
into axisymmetric boundary value problems of potential
theory.
If we apply the operator K
Z g,g
to both sides of the rst
equation of Eqs. (49) and make use of the second relation
of Eqs. (49), we obtain
K
Zg;g
I
Z a;b
S
Z;a
=S
Zg;a bg
(56)
Equation (56) can be written in terms of Hankel trans-
forms as follows:

2Z a bg
[t
abg
f (t); r]
=
r
2
_ _
a bg
K
Zg;g
I
Z a;b
2
a
x
a

2Z a
[t
a
f (t); x]
(57)
For a =0 b =(mn d)/2,
Z
=v/2, Eq. (57) simplies signi-
cantly

m
[s
d
~
ff (s); r]
=2
d
r
d
K
(md)=2;(vmd)=2
I
v=2;(mvd)=2
f (r)
(58)
Some special cases of formulas in Eq. (58) are of particular
interest. If we set m =v, we obtain

m
[s
d
~
ff (s); r] =2
d
r
d
K
(md)=2;d=2
I
v=2;d=2
f (r) (59)
HANKEL TRANSFORMS 1901
Special cases of particular interest are given by assigning
d =71 to Eq. (59); we then obtain

v
s
~
ff
v
(s); r
_ _
=
2
p
r
v1
d
dr
_
o
r
x
12v

x
2
r
2
_
d
dx

_
x
0
y
v 1
f (y) dy

x
2
y
2
_ (v _ 0)

v
s
1 ~
ff (s); r
_ _
=
2
p
r
v
_
o
r
x
2v

x
2
r
2
_

_
x
0
y
v 1
f (y) dy

x
2
y
2
_ (v _ 0)
(60)
On the other hand, if we put m =v 1 in Eq. (59), we
obtain the relation

v 1
[s
d
~
ff
v
(s); r]
=2
d
r
d
K
(v d 1)=2;(1d)=2
I
v=2;(1d)=2
f (r)
The special case d =1 corresponds to the well-known
formula

v 1
[s
~
ff
v
(s); r] = r
v
d
dr
[r
v
f (r)] (61)
Expressions corresponding to the particular values 0 and
1 of d are, respectively,

v 1
[
~
ff
v
(s); r] =
2
p
r
v
d
dr
_
o
r
x
2v
dx

x
2
r
2
_
_
x
0
y
v 1
f (y) dy

x
2
y
2
_ (v _ 0)

v 1
[s
1 ~
ff
v
(s); r] =r
v1
_
r
0
u
v 1
f (u) du (v _ 0)
(62)
Finally, if we set m=v 1 in Eq. (59), we obtain the
relation

v1
[s
d
~
ff
v
(s); r] =2
d
r
d
K
(v1 d)=2;(v1d)=2
f (r) (63)
The most frequently occurring special cases of the formula
in Eq. (63) are

v1
[s
~
ff
v
(s); r] =r
v
d
dr
[r
v
f (r)] (v _ 1)

v1
[
~
ff
v
(s); r] =
2
p
r
v1
_
o
r
x
12v
dx

x
2
r
2
_
d
dx
_
x
0
y
v 1
f (y) dy

x
2
y
2
_
(v _ 1)

v1
[s
1 ~
ff
v
(s); r] =r
v1
_
o
r
x
1v
f (x) dx (v _ 1) (64)
8.1. Beltramis Relation for an Electried Disk
As an application of Beltrami-type relations just derived,
let us consider the problem of an electried disk of radius
a lying in the plane z =0 with its center at the origin of the
coordinate system. Let the surface charge density be q(r).
Then in the half-space zZ0 the potential of the electro-
static eld will be f

(r, z) and in the half-space zr0, it


will be f

(r, z), where


f

(r; z) =
0
[
~
ff
0
(s) e
sz
; r]
where
~
ff
0
(s) =
0
[f(r; 0); s]
The charge density on the plane z =0 is given by the
equation
q(r) =
1
4p
@f

@z

@f
@z
_ _
z =0
and it immediately follows from this equation that
q(r) =
1
2p

0
[s
~
ff
0
(s); r] (65)
From the rst equation of Eqs. (60) then we deduce
Beltramis relation in Eq. (55). On the other hand, we
could write Eq. (65) in the form
f(r; 0) =2p
0
[s
1
~ qq
0
(s); r]
and then using the second relation of Eq. (60) deduce the
equation
f(r; 0) =4
_
o
r
dx

x
2
r
2
_
_
min(a; x)
0
yq(y) dy

x
2
y
2
_
Interchanging the order of integration, the last equation
can be written as
f(r; 0) =
_
a
0
s(y)K(r; y) dy
where
K(r; y) =4y
_
o
min(r;y)
du

(u
2
r
2
)(u
2
y
2
)
_
9. DUAL-INTEGRAL EQUATIONS INVOLVING HANKEL
TRANSFORMS
In the applications of the theory of Hankel transforms to
the solution of boundary value problems of mathematical
physics, it often happens that the problem may be reduced
to the solution of a pair of simultaneous equations of the
form
f (x) =S
m=2a;2a
[1 k(x)]c(x);
g(x) =S
v=2b;2b
c(x)
(66)
1902 HANKEL TRANSFORMS
in which
f (x) =
f
1
(x); x c I
1
={x : 0oxo1]
f
2
(x); x c I
2
={x : 1oxoo]
_
_
_
g(x) =
g
1
(x); x c I
1
={x : 0oxo1]
g
2
(x); x c I
2
={x : 1oxoo]
_
_
_
The problem is as follows. Knowing the functions
k(x)[k(x) 0; x o]; f
1
, and g
2
, is it possible to nd the
functions c, f
2
, and g
1
? In the following, we consider the
special case where k(x) =0, but it is straightforward to
generalize the results for k(x)a0.
To solve the problem, Sneddon proposed the following
trial solution:
c(x) =S
v=2 b;m=2v=2ab
h(x) (67)
Putting Eq. (67) into Eqs. (65), we obtain
S
m=2a;2a
S
v=2b;m=2v=2ab
h=f
S
v=2b;2b
S
v=2b;m=2v=2ab
h=g
which can be rewritten, using the third and fourth rela-
tions of Eq. (49), as
I
v=2 b;v=2v=2ab
h=f
K
v=2b;m=2v=2a b
h=g
when
h=I
1
v=2 b;m=2v=2 ab
f
h=K
1
v=2b;m=2v=2a b
g
(68)
Writing Eqs. (68) on the intervals I
1
and I
2
, we have
h
1
(x) =
x
0
_
_
_
_
I
1
v=2 b;m=2v=2 ab
f
1
h
2
(x) =
1
0
_
_
_
_
I
1
v=2 b;m=2v=2 ab
f
1

x
1
_
_
_
_
I
1
v=2 b;m=2v=2ab
f
2
h
2
(x) =
o
x
_
_
_
_
K
1
v=2b;m=2v=2a b
g
2
h
1
(x) =
o
1
_
_
_
_
K
1
v=2b;m=2v=2a b
g
2

1
x
_
_
_
_
K
1
v=2b;m=2v=2a b
g
1 (69)
Putting the rst and third equations of Eqs. (69) into Eq.
(67), we obtain the solution for c (x). On the other hand,
from the second and third equations of Eqs. (69), we
deduce that
x
1
_
_
_
_
I
1
v=2 b;m=2v=2 ab
f
2
=
o
x
_
_
_
_
K
1
v=2b;m=2v=2a b
g
2

1
0
_
_
_
_
I
1
v=2b;m=2v=2ab
f
1
whence it follows by use of the L operator dened by Eq.
(53) that
f
2
=
x
1
_
_
_
_
I
v=2 b;m=2v=2 ab

o
x
_
_
_
_
K
1
v=2b;m=2v=2a b
g
2

x; 1
1 0
_
_
_
_
L
v=2a;v=2m=2 ba
f
1
(70)
Thus, f
2
is determined. Similarly, using the rst and
fourth equations of Eqs. (68), we obtain for g
1
the formula
g
1
=
1
x
_
_
_
_
K
v=2b;m=2v=2a b
I
v=2 b;m=2v=2 ab
f
1

1; o
x; 1
_
_
_
_
M
v=2b;m=2v=2a b
g
2
(71)
Thus, the rst two equations in Eqs. (69) and Eqs. (70) and
(71) give the complete solution to our problem. The same
procedure, applied to the case where k(x)a0, yields
h
1
E(x) =
x
0
_
_
_
_
I
1
f
1
(x c I
1
)
h
2
E(x) =
1
0
_
_
_
_
I
1
f
1

x
1
_
_
_
_
I
1
f
2
(x c I
2
)
HANKEL TRANSFORMS 1903
h
2
=
o
x
_
_
_
_
K
1
g
2
(x c I
2
)
h
1
=
o
1
_
_
_
_
K
1
g
2

1
x
_
_
_
_
K
1
g
1
(x c I
1
)
(72)
where
E(x) =S
m=2a;v=2b am=2
kS
v=2 b;m=2v=2ab
h(x)
(73)
The subscripts with the I and K in Eqs. (72) are the same
as those in Eqs. (69). Further details are carried out for
the special case where v =m, b=0, g
2
=0, which is the
most frequently occurring case in applications. In this
case, we nd from Eqs. (72) that h
2
(x) =0 and h
1
(x) solves
the integral equation
h
1
(x) E(x) =
x
0
_ _
I
1
v=2;a
f
1
(x c I
1
) (74)
where
E(x) S
v=2a;a
kS
v=2;a
h(x)
=2
a
x
a
_
o
0
t
1a
k(t)2
a
t
a
J
va
(xt) dt
_
1
0
u
1 a
h
1
(u)J
va
(tu) du
and inverting the order of intergration, we have
E(x) =x
a
_
1
0
u
1a
K(x; u)h
1
(u) du (75)
where
K(x; u) =
_
o
0
tk(t) J
va
(xt) J
va
(ut) dt (76)
Since the functions h
1
and h
2
have been determined, it is
possible to nd the functions c, f
2
, and g
1
following the
procedure for the case k(x) =0. These details can be found
in the papers by Sneddon [8] and Cooke [19,20].
9.1. An Example: Two Coaxial Electried Circular Disks
The problem of two solid disks, each charged to a uniform
potential f
0
, was the subject of numerous research
starting with Loves paper [24] (for references, see Cooke
[19]). If the disks have different potentials the problem
may be reduced to two separate problems, in one of
which the potentials are equal and in the other they are
equal and opposite. Assume that the disks have the
same radii, equal to unity, and are situated in the planes
z =0 and z =h, where r, y, and z are cylindrical coordi-
nates. Then, the problem reduces to that of solving
Laplaces equation in Eq. (1) subject to the following
boundary conditions:
f(r; 0) =f
0
; 0oro1
f(r; 0

) =f(r; 0

); 0 _ roo
@f
@z

z =0
=
@f
@z

z =0
; r > 1
f(r; h) = f
0
; 0oro1
f(r; h

) =f(r; h

); 0 _ roo
@f
@z

z =h
=
@f
@z

z =h
; r > 1
(77)
The sign in fourth of the preceding conditions is positive or
negative according to whether the disks are of like or
unlike potentials f
0
.
The solution of the problem must satisfy the regularity
conditions at innity. Besides, in order to guarantee
uniqueness of solution, it must satisfy the edge condition
[25,26] so that the electric energy stored in any neigh-
borhood of the sharp edge r =1 be nite, which imposes
the restriction that the surface charge density not grow
more rapidly than r
1t
with t40 as r 0, where
r =1 r.
It can be shown by using zeroth-order Hankel trans-
form to Laplaces equation in Eq. (1) that the electrostatic
eld can be represented by the potential function
f(r; z) =H
0
[f
0
s
1
(e
z [ [s
e
zh [ [s
)A(s); s r] (78)
which satises the second and fth continuity conditions
in Eqs. (77), the sign in Eq. (78) being positive or negative
depending on whether the disks are of like or unlike
potentials j
0
.
We nd that the third and sixth conditions in Eqs.
(77) will be satised if the function A(s) satises the
equation

0
[A(s); s r] =0; r > 1 (79)
Using the second and fourth boundary conditions in Eqs.
(77) and Eqs. (79), we obtain the following dual integral
equations:

0
[s
1
(1 e
hs
)A(s); s r] =1; 0 _ ro1

0
[A(s); s r] =0; r > 1
(80)
Using the modied operator of Hankel transform, we
rewrite Eqs. (80) in the form
S
1=2;1
1 k(r) [ ]A(r) =1; 0 _ ro1
S
0;0
A(r) =0; r > 1
(81)
1904 HANKEL TRANSFORMS
where k(s) =e
hs
. Thus, for our problem
a =
1
2
; m=0; b =0; v =0
f
1
(r) =
r
2f
0
; g
2
(r) =0
Therefore, following the procedure outlined in the pre-
vious section, we nd that h
2
(r) =0 and h
1
(r) solves the
following integral equation
h
1
(r) r
1=2
_
1
0
u
3=2
K(r; u)h
1
(u) du
=
r
0
_
_
_
_
I
1=2;1=2
f
1
(r) 0 _ ro1
(82)
where
K(r; u) =
_
o
0
tk(t) J
1=2
(rt) J
1=2
(ut) dt
Writing rh
1
(r) =H(r), we reduce Eq. (82) to the following
Fredholm integral equation of the second kind
H(r)
_
1
0
H(u)N(r; u) du
=
r
0
_
_
_
_
I
1=2;1=2
f
1
(r)
(83)
where
N(r; u) =

ru
_
_
o
0
tk(t) J
1=2
(rt) J
1=2
(ut) dt (84)
The kernel N(r, u) in Eq. (84) can be evaluated in closed
form:
N(r; u) =
1
(r u)
2
h
2

1
(r u)
2
h
2
_ _
(85)
The integral equation dened by Eqs. (83) and (85) can be
solved numerically.
The surface density at any point of a disk in the plane z
=0 is equal to
1
4p
@f
@z
_ _
z =0
When both sides of the disk are taken into account, this
gives for the total charge Q
Q=
f
0
2p
_
1
0
2prdr
_
o
0
A(s)J
0
(sr) ds
=f
0
_
1
0
rg
1
(r) dr
=f
0
_
1
0
r
1
r
_
_
_
_
K
0;1=2
h
1
(r) dr
=
f
0

p
_
_
1
0
dr
d
dr
_
1
x
u
2
h
1
(u) du

u
2
1
_
=
2f
0
p
_
1
0
uH(u) du

u
2
1
_
(86)
Once the integral equation in Eq. (83) is solved for H(r),
the total charge can be found by evaluating the integral in
Eq. (86) numerically and hence the capacity C=Q/f
0
can
be found.
10. TRIPLE-INTEGRAL EQUATIONS INVOLVING HANKEL
TRANSFORMS
As an example of the use of Cooke operators, we consider
the solution of certain triple-integral equations involving
Hankel transforms. The problem consists in nding a
function F (x) satisfying
_
o
0
F(x)J
v
(xx) dx =G
1
(x); x c I
1
_
o
0
x
2a
1k(x) [ ]F(x)J
v
(xx) dx =F
2
(x); x c I
2
_
o
0
F(x)J
v
(xx) dx =G
3
(x); x c I
3
(87)
where I
j
( j =1,2,3) denote, respectively, the intervals (0,a),
(a, b) and (b,N) with 0oaob. The functions G
1
, F
2
, and
G
3
are assumed to be prescribed. Assuming that
F(x) =xc(x); f (x) =
2
x
_ _
2a
F(x);
g(x) =G(x)
and using the modied operator of the Hankel transform,
we rewrite Eqs. (87) in the form
S
v=2a;2a
[{1 k(x)]c(x); x] =f (x)
S
v=2;0
c(x) =g(x)
(88)
We rst consider the case where k=0, g
1
=g
3
=0, |a|o1.
There are two different ways of solving Eqs. (88), one
proposed by Sneddon and the other by Borodachev.
10.1. The Sneddon Trial Solution
Sneddon proposed the following solution for the equations
in Eq. (88):
c=S
v=2;a
h (89)
HANKEL TRANSFORMS 1905
Then, putting Eq. (89) into Eq. (88), we nd that
S
v=2;0
c=K
v=2;a
h=g(x)
S
v=2a;2a
c=I
v=2;a
h=f (x)
and solving for h, we obtain
h=I
1
v=2;a
f (x)
h=K
1
v=2;a
g(x)
(90)
Now, suppose that f (x) =f
1
(x); x c I
1
; f (x) =f
3
(x);
x c I
3
, and g(x) =g
2
(x); x c I
2
. We also write h(x) =h
j
(x);
x c I
j
.
If we evaluate Eqs. (90) on I
3
and use g
3
=0, We deduce
that h
3
=0. Similarly, if we evaluate Eq. (90) on I
1
, we
have
f
1
(x) =
x
0
_ _
I
v=2;a
h
1
(x) (91)
and if we evaluate Eq. (90) on I
2
we have
h
2
=
a
0
_ _
I
1
v=2;a
f
1

x
a
_ _
I
1
v=2;a
f
2
; x c I
2
(92)
Putting Eq. (92) into Eq. (91) and using the L operator
dened by Eq. (53), we obtain
h
2
=
x; a
a; 0
_ _
L
v=2;a
h
1

x
a
_ _
I
1
v=2;a
f
2
; x c I
2
(93)
Now, evaluating Eq. (90) on I
2
and I
1
, respectively, we
obtain the equations
g
2
=
b
x
_ _
K
v=2;a
h
2
; h
1
=
b
a
_ _
K
1
v=2;a
g
2
(94)
Putting rst of the relation in Eq. (94) into the second and
using M operator dened by Eq. (54), we obtain
h
1
=
a; b
x; a
_ _
M
v=2;a
h
2
(x c I
1
) (95)
Equations (93) and (95) form a pair of simultaneous
equations for the unknown functions h
1
and h
2
, but, by
eliminating h
1
between them, we can derive a single
Fredholm integral equation of the second kind for h
2
.
Solving it, we can determine h
1
using Eq. (95).
The same procedure applied formally to the case
in which k(x)O0; g
1
=g
3
=0; a [ [o1 leads to the set of
simultaneous equations
h
1
E=k
1
(x c I
1
)
h
2
E=
x; a
a; 0
_
_
_
_
L
v=2;a
k
1

x
a
_
_
_
_
I
1
v=2;a
f
2
(x c I
2
)
h
1
=
a; b
x; a
_
_
_
_
M
v=2;a
h
2
(x c I
1
)
(96)
where
E(x) S
v=2a;a
kS
v=2a
h
=2
a
x
a
_
o
0
t
1a
k(t)2
a
t
a
J
va
(xt) dt

_
a
0
u
1a
h
1
(u)J
va
(tu) du
_

_
b
a
u
1a
h
2
(u)J
va
(tu) du

_
o
b
u
1 a
h
3
(u)J
va
(tu) du
_
and inverting the order of integration in each of the three
repeated integrals, we have
E(x) =x
a
_
o
0
u
1 a
K(x; u)h(u) du (97)
where
K(x; u) =
_
o
0
tk(t)J
va
(xt)J
va
(ut) dt (98)
Thus, we have three equations with three unknown h
1
, h
2
,
k
1
. As before h
3
=0. Solving for them, the unknown
functions f
1
and f
3
can be found by the formulas
f
1
=
x
0
_
_
_
_
I
v;a
k
1
f =I
v=2;a
hS
v=2a;2a
kS
v=2;a
h
(99)
Equations (93) and (95) to (98) allow us to obtain the
complete solution of the problem. (For further details,
readers are referred to the papers by Cooke [19,20,27,28]).
10.2. The Borodachev Trial Solution
Borodachev [29] developed a different trial solution to
solve the triple integral equations (87). He argued as
follows. Assume that the solution of the equations has
the form
c(x) =S
b;g
h (100)
1906 HANKEL TRANSFORMS
Eqs. (88), for the case k(x) =0, may be reduced to the
following form
I
m
1
;l
1
h=f ; K
m
2
;l
2
h=g
which occur when
S
v=2a
S
b;g
=I
m
1
;l
1
; S
v=2;0
S
b;g
=K
m
2
;l
2
(101)
Using the third and fourth relations of Eqs. (49), we infer
that
b g =
v
2
a; m
1
=b; l
1
=2a g
b =
v
2
; m
2
=
v
2
; l
2
=g
which yield
b =
v
2
; g = a; m
1
=
v
2
; m
2
=
v
2
;
l
1
=a; l
2
= a
Thus, in this case Eq. (100) takes the form c=S
v=2;a
h,
that is, we have Sneddons trial solution.
On the other hand, readers might note that Eqs. (88)
can be reduced to the form
K
m
3
;l
3
H=f ; I
m
4
;l
4
H=g (102)
Carrying out calculations similar to the ones done, we
have
b =
v
2
a; g = a; m
3
=
v
2
a; m
4
=
v
2
a
l
3
=a; l
4
= a
(103)
Accordingly, in this case
c=S
v=2a;a
H (104)
Equation (104) is called Borodachevs trial solution.
We will now use Borodachevs trial solution to reduce
the triple integral equations in Eq. (88) to a Fredholm
integral equation of the second kind. Substituting Boro-
dachevs trial solution in Eq. (104) into Eqs. (88), we obtain
[see Eqs. (102) and (103)]
K
v=2a;a
H=f ; I
v=2 a;a
H=g;
whence
H=K
1
v=2a;a
f ; H=I
1
v=2 a;a
g (105)
As before, for the sake of simplicity, we consider the
case where g
1
=g
3
=0. Then writing Eq. (105) for each
interval, we obtain
H
1
=
x
0
_
_
_
_
I
1
v=2 a;a
g
1
=0
H
2
=
a
0
_
_
_
_
I
1
v=2a;a
g
1

x
a
_
_
_
_
I
1
v=2a;a
g
2
=
x
a
_
_
_
_
I
1
v=2a;a
g
2
H
3
=
a
0
_
_
_
_
I
1
v=2a;a
g
1

b
a
_
_
_
_
I
1
v=2a;a
g
2

x
b
_
_
_
_
I
1
v=2a;a
g
3
=
b
a
_
_
_
_
I
1
v=2a;a
g
2
H
3
=
o
x
_
_
_
_
K
1
v=2a;a
f
3
H
2
=
o
b
_
_
_
_
K
1
v=2a;a
f
3

b
x
_
_
_
_
K
1
v=2a;a
f
2
(106)
From the second and fourth formulas in Eqs. (106), we
deduce that
g
2
=
x
a
_
_
_
_
I
v=2 a;a
H
2
;
f
3
=
o
x
_
_
_
_
K
v=2a;a
H
3
(107)
Substituting Eq. (107) into the third and fth equations in
Eqs. (106) and making use of the operators L and M, we
obtain the following system of equations:
H
2
=
b
x
_
_
_
_
K
1
v=2a;a
f
2

b; o
x; b
_
_
_
_
M
v=2a;a
H
3
(aoxob)
H
3
=
x; b
b; a
_
_
_
_
L
v=2a;a
H
2
(boxoo)
(108)
HANKEL TRANSFORMS 1907
Using the denitions of L and Moperators, we see that the
formulas in Eq. (108) constitute a pair of coupled integral
equations, upon solving for which we can nd the func-
tions and H
2
and H
3
, while H
1
=0.
Putting the second formula of Eq. (108) into the rst
equation, we obtain a single-integral equation of the
second kind involving only H
2
:
H
2
(x) =j(x)
2
p
_ _
2
_
b
a
K(x; y)H
2
(y) dy (109)
where
j(x) =
b
x
_
_
_
_
K
v=2;a
f
2
K(x; y) = sin
2
(ap)
x
v
y
1 2a v
(b
2
x
2
)
a
(b
2
y
2
)
a

_
o
0
t
12v2a
(t
2
b
2
)
2a
(t
2
x
2
)(t
2
y
2
)
dt
(
1
2
oao1)
(110)
10.3. An Example: An Electried Annular Disk
To illustrate the application of Cookes and Borodachevs
solutions to the set of triple integral equations in Eq. (98),
we consider the electrostatic eld induced by an annular
disk with internal and external radii a and b, respectively,
the disk being charged to a potential equal to f
0
. The disk
is assumed to lie in the plane z =0.
The solution of the problem must satisfy Laplaces
equation in Eq. (1) and the following boundary conditions:
f(r; 0) =f
0
; aorob
@f
@z

z =0
=0; 0 _ roa; boroo
(111)
Furthermore, the solution must satisfy the regularity
condition and the edge conditions at the edges r =a and
r =b. As before, applying zeroth-order Hankel transform
to the Eq. (1), it can be shown that the electrostatic
potential is given by the equation
f(r; z) =f
0

0
[s
1
A(s) e
sz
; s r] (112)
Where A(s) is an unknown function of s to be determined.
Equation (112) automatically satises the radiation con-
ditions.
Making use of the boundary conditions in Eq. (111), we
obtain the following triple-integral equations:
S
1=2;1
A(r) =f (r); S
0;0
A(r) =g(r)
where f
2
(r) =2r=f
0
; g
1
(r) =0; g
3
(r) =0. Following Sneddons
trial solution in Eq. (89), we obtain the following Fredholm
integral equation of the second kind:
x
2
e
2
x
2
F(x) =1
2
p
_ _
2
_
1
e
K(x; y)F(y) dy (113)
where
x =
r
a
; e =
a
b
; F(x) =h
+
2
(xb)
K(x; y) =
1
2(x
2
y
2
)

x
2
e
2
x
log
x e
x e

y
2
e
2
y
log
y e
y e
_ _
On the other hand, making use of Borodachevs trial
solution in Eq. (104), we obtain the following Fredholm
integral equation
1 x
2
x
2
G(x) =1
2
p
_ _
2
_
1
e
M(x; y)G(y)dy (114)
where
G(x) =h
+
2
(bx); h
+
2
(r) =

p
_
r
2
2f
0

b
2
r
2
_ h
2
(r)
M(x; y) =
1
2(x
2
y
2
)

1 y
2
y
log
1 y
1 y

1 x
2
x
log
1 x
1 x
_ _
(115)
The surface charge density at any point of the disk is
q =
1
4p
@f
@z
_ _
z =0
=
f
0
4p
g
2
(r)
=
1
2p
2
r
d
dr
_
r
a

b
2
u
2
r
2
u
2
_
h
+
2
(u) du; aorob
bf
0
2p
2
r
d
dr
_
r=b
e

1 y
2
r
2
=b
2
y
2

G(y)dy; aorob:
_

_
(116)
Thus, the charge density at any point of the disk can be
calculated once the integral equation in Eq. (114) is solved.
Considering both sides of the disk, the total charge is
Q=4p
_
b
a
rq(r; 0) dr =
2f
0
b
pg
;
g
1
=
_
1
e
G(y) dy
Whence
f
0
=
pQg
2b
1908 HANKEL TRANSFORMS
so that formula in Eq. (116) takes the form
q(r; 0) =
gQ
2pr
d
dr
_
r=b
e

1 y
2
r
2
=b
2
y
2

G(y) d(y); aorob


(117)
Of great interest is to nd the asymptotic representation
of the charge density q(r, 0) as r a0 in the sense of
Erdelyi, that is, the rst term in the asymptotic expansion
of q(r, 0) as r a0. By letting r a0 in Eq. (117), we
obtain
q(r; 0) -
Qo
a
(e)
2

2
_
pb
2
r
b
e
_ _
1=2
; r a0 (118)
where
o
a
(e) =
g
e

1 e
2
e
_
G(e) (119)
Performing similar analyses on the Sneddons trial solu-
tion, it can be shown that the surface charge density
exhibits the following behavior as the outer contour of
the disk is approached
q(r; 0) =
Qo
b
(e)

2
_
pb
2
1
r
b
_ _
1=2
; r b 0 (120)
where
o
b
(e) =g

1 e
2
_
F(1) (121)
Equations (118)(121) show that the surface charge den-
sity exhibits a square-root singularity as the inner and
outer edges of the disk are approached. Thus, edge condi-
tions (Meixners conditions) are satised.
Integral equations in Eqs. (113) and (114) admit closed-
form solutions only in the special case where e =0, that is,
for the case of a circular disk:
F(x) =1;
G(x) =
x
p

1 x
2
_ log
1 x
1 x
(122)
In the context of mathematically similar elastic contact
problems, Borodachev [29] showed that the values of G(x)
do not differ practically in the range 0rer0.5. Therefore
for this range, approximate values of the surface charge
density can be calculated by using formulas in Eq. (122)
while the intergral equation in Eq. (113) can be solved to
nd the surface charge density for the range 0:5oeo1:0.
Many other application of the triple integral equations
considered here to problems of electrostatics are given in
Sneddons book [8]. It should be noted that using the same
approach, it is possible to solve a wide variety of problems
concerning diffraction of a plane electromagnetic wave by
an annular disk and by a system of coaxial annular disks.
Many examples of electromagnetic scattering by objects of
different shapes are analyzed in the books by Bowman
et al. [30] and by Uslenghi [31].
11. QUADRUPLE-INTEGRAL EQUATIONS INVOLVING
HANKEL TRANSFORMS
We now use Cooke operators to reduce certain quadruple
integral equations involving Hankel transforms to a Fred-
holm integral equation of the second kind or a system of
those. The problem is to nd a function c (x) satisfying the
equations
S
v=2a;2a
c(x) =f
1
(x); x c I
1
={x : 0oxoa]
S
v=2b;2b
c(x) =0; x c I
2
={x : aoxob]
S
v=2a;2a
c(x) =f
3
(x); x c I
3
={x : boxoc]
S
v=2b;2b
c(x) =0; x c I
4
={x : coxoo]
(123)
Taking a trial solution in the form
c(x) =S
v=2b;ab
h(x)
and then using the third and fourth relations from Eqs.
(49), we obtain
f (x) S
v=2a;2a
c(x) =I
v=2 b;ab
h(x)
g(x) S
v=2b;2b
c(x) =K
v=2b;ba
h(x)
(124)
whence
h(x) =I
1
v=2 b;ab
f (x)
h(x) =K
v=2b;ab
g(x)
(125)
Writing out Eqs. (125) on I
j
(j =1, y4), we have
h
1
(x) =
x
0
_
_
_
_
I
1
v=2 b;ab
f
1
(x c I
1
)
h
2
(x) =
a
0
_
_
_
_
I
v=2 b;ab
f
1

x
a
_
_
_
_
I
1
v=2b;ab
f
2
(x c I
2
)
h
3
(x) =
a
0
_
_
_
_
I
1
v=2 b;ab
f
1

b
a
_
_
_
_
I
1
v=2 b;ab
f
2

x
b
_
_
_
_
I
1
v=2 b;ab
f
3
(x c I
3
)
HANKEL TRANSFORMS 1909
h
4
(x) =
a
0
_
_
_
_
I
1
v=2b;ab
f
1

b
a
_
_
_
_
I
1
v=2 b;ab
f
2

c
b
_
_
_
_
I
1
v=2 b;ab
f
3

o
c
_
_
_
_
I
1
v=2 b;ab
f
4
(x c I
4
)
h
4
(x) =
o
x
_
_
_
_
K
1
v=2b;ab
g
4
=0 (x c I
4
)
h
3
(x) =
c
x
_
_
_
_
K
1
v=2b;ab
g
3
(x c I
3
)
h
2
(x) =
c
b
_
_
_
_
K
1
v=2b;ab
g
3
(x c I
2
)
h
1
(x) =
c
b
_
_
_
_
K
1
v=2b;ab
g
3

a
x
_
_
_
_
K
1
v=2b;ab
g
1
(x c I
1
)
(126)
From sixth equation of Eqs. (126), we have
g
3
=
c
b
_ _
K
v=2b;ba
h
3
which upon substitution into the seventh equation of
Eq. (126) yields
h
2
(x) =
b; c
x; b
_ _
M
v=2b;ba
h
3
(x) (127)
Writing Eq. (124) on I
3
, we obtain
f
3
(x) =
a
0
_
_
_
_
I
v=2b;ab
h
1

b
a
_
_
_
_
I
v=2 b;ba
h
2

x
b
_
_
_
_
I
v=2b;ba
h
3
(128)
Applying the operator
x
b
_ _
I
1
v=2b;ba
to both sides of Eq. (128), we obtain
h
3
(x) =L(x)
x; b
b; a
_ _
L
v=2b;ab
h
2
(x) (129)
where L(x) is the known function given by
L(x) =
x
b
_
_
_
_
I
1
v=2 b;ab
f
3
(x)

x; a
b; a
_
_
_
_
L
v=2b;ab
h
2
(x)
(130)
Equations (127) and (129) constitute a pair of coupled
integral equations for the determination of the unknown
functions h
2
and h
3
, but eliminating h
2
, we obtain a single
Fredholm equation of the second kind, namely
h
3
(x) m
_
c
b
K(x; x
0
)h
3
(x
0
) dx
0
=L(x) (boxoc)
(131)
where m =(4/p
2
) sin
2
[p(a b)] and the kernel is given by
the equation
K(x; x
0
)
=x
v2b
(x
2
b
2
)
ba
(x
2
0
b
2
)
ba
x
2av 1
0

_
b
a
(b
2
y
2
)
2a2b
y
2v2a 2b 1
(x
2
0
y
2
)(x
2
y
2
)
dy (box; x
0
oc)
(132)
Further details can be found in the article by Sneddon [21]
and the references cited therein.
Quadruple integral equations of the type in Eq. (123)
arise in many boundary value problems of mathe-
matical physics. For instance, the electrostatic problem
of three coplanar circular disks charged to a uniform
potential can be reduced to this kind of quadruple integral
equations.
Finally, it should be noted that a new set of particular
solutions can be derived for the quadruple integral equa-
tions of the type in Eq. (123) analogous to Borodachevs
trial solution for the triple integral equations in Eq. (88),
by assuming
c(x) =S
v=2 a;ab
H(x) (133)
On substituting Eq. (133) into Eqs. (123), we obtain
f (x) =K
v=2a;ab
H(x)
g(x) =I
v=2 a;a b
H(x)
1910 HANKEL TRANSFORMS
12. MISCELLANEOUS
1. There is a generalization of the Hankel integral theo-
rem in Eq. (12), known as Webers integral [32]
f (r) =
_
o
0
j
s
(r)s ds
J
2
v
(sa) Y
2
v
(sa)

_
o
0
r
0
f (r
0
)j
s
(r
0
) dr
0
aoroo
(134)
involving the linear combination
j
s
(r) =J
v
(sa)Y
v
(sr) Y
v
(sa)J
v
(sr) (135)
of Bessel functions of the rst and second kinds (v >
1
2
). A
sufcient condition for the validity of the Eq. (133) is that
f(r) be piecewise continuous and of bounded variation in
every nite subinterval [a,b], where aoaoboN, and the
integral
_
o
a

r
_
f (r)

droo
It should be noted that Webers integral reduces to Han-
kels integral in the limit as a 0. Derivation of equations
in Eqs. (134) and (135) is given in the famous book by
Titchmarsh [32]. Properties of Webers transformation are
also similar to those derived for Hankel transforms.
Webers transform is suited for solving equations of the
form in Eq. (1) for domains with an excluded circular
region. We illustrate the use of Webers integral by one
example.
Given: A cylindrical hole of radius a is drilled in an
innite body, and the walls of the hole are maintained at a
temperature T
0
starting from the time t =0. It is required
to determine the temperature distribution in the body
assuming that the initial temperature is zero. The two-
dimensional temperature distribution in the body is gov-
erned by the heat conduction equation
1
r
@
@r
r
@T
@r
_ _
=
@T
@t
; aoroo (136)
satisfying the initial condition T|
t =0
=0 and the bound-
ary and radiation conditions
T[
r =a
=T
0
; T[
ro
0
Multiplying both sides of Eq. (136) by r j
s
(r) and integrat-
ing the resulting expression from a to N, we obtain
d
~
TT
dt
s
2
~
TT =
2T
0
p
(137)
where
~
TT(s; t) =
_
o
a
r
~
TT(r; t)j
s
(r) dr
may be called the Weber transform of zeroth order of the
function T(r, t). In deriving Eq. (137), use has been made of
the relations
j
s
(a) =0; j
s
/
(a) =
2
pa
The solution of Eq. (137) satisfying the boundary condition
~
TT

t =0
is
~
TT(s; t) =
2T
0
ps
2
(1 e
s
2
t
)
Now, using Webers inversion, we nally obtain the follow-
ing formula for the temperature evolution in the body:
T(r; t) =
2T
0
p
_
o
0
(1 e
s
2
t
)j
s
(r) ds
s[J
2
0
(sa) Y
2
0
(sa)]
(138)
Many other practical applications of Webers integral are
given in the book by Lebedev et al. [9].
2. In applications of Hankel transforms to many phy-
sical problems, integrals of the following form are encoun-
tered:
_
o
0
s
~
ff
n
(s)
~
FF
m
(s)J
mn
(sr) ds (139)
The need to evaluate such integrals arises in connection
with the desire of transforming the solution for the
physical quantities given in the space of Hankel Trans-
form domain into the physical space. Using Parsevals
relation in Eq. (39), we reduce the integral in Eq. (139) to
the form
_
o
0
s
~
ff
n
(s)
~
FF
m
(s)J
mn
(sr) ds
=
_
o
0
r
0
f (r
0
)F(r
0
)dr
0
(140)
where
F(r
0
) =H
n
[
~
FF
m
(s)J
mn
(sr); s r
0
]
=
_
o
0
s
~
FF
m
(s) J
mn
(sr) J
n
(sr
0
) dr
0
(141)
For the product of Bessel functions in Eq. (141), we use
Neumanns formula generalized by Rahman [14], and
then interchanging the order of integration, we get
F(r
0
) =
1
p
_
p
0
cos (nf)T
m
r r
0
cos f
R
_ _ _

r
0
sinnfsinf
R
U
m1
r r
0
cos f
R
_ __
F(R) df
(142)
In specic physical problems, however, the cases where
m=0 and m=1, n=0 are the most frequently encountered
HANKEL TRANSFORMS 1911
ones. In these cases, formula in Eq. (142) simplies
signicantly. For instance, for m=0, we have
F(r
0
) =
1
p
_
p
0
cos(nf)F(R) df
while for m=1, n=0, we have
F(r
0
) =
1
p
_
p
0
r r
0
cos f
R
F(R) df
3. An efcient method of solving the integral equation
(74) is based on representing the unknown function h
1
(x)
in the form [33]
h
1
(x) =x
v2a

o
n=0
a
n
P
va;0
n
(1 2x
2
) (143)
where P
va;0
n
(1 2x
2
) is the Jacobi polynomial and a
n
are
the unknown expansion coefcients to be determined.
Putting the expansion in Eq. (143) into Eq. (73) and
considering the orthogonality relationship for the Jacobi
polynomials
_
1
0
P
a;b
n
(1 2x
2
)P
a;b
m
(1 2x
2
) dx
2
2ab
x
12a
(1 x
2
)
b
=
2
a b1
G(a n1)G(b n1)
n!G(na b 1)(a b 2n1)
d
mn
(d
mn
Kr oneckers delta)
we obtain the following innite system of linear algebraic
equations:
a
m
2(1 v a 2m)

o
n=0
a
n
K
mn
=r
m
(m=0; 1; 2; . . . ; o)
(144)
where
K
mn
=
_
o
0
t
1
k(t)J
1 va 2m
(t)J
1va 2n
(t) dt
r
m
=
_
1
0
x
1a
r(x)P
va;0
m
(1 2x
2
) dx
(145)
A key result that was used to obtain Eqs. (144) and (145) is
the following integral [34]:
s
a;a
[(1 x
2
)
b
P
a;b
n
(1 2x
2
)]
=
G(1 bn)
2
ab
n!x
1ab
J
1 ab2n
(x)
The innite system in Eq. (144) can be solved by trunca-
tion for the unknown expansion coefcients a
n
. In bound-
ary value problems, often it is often the case that the
quantity g
1
(x) is of prime importance. For instance, in the
charged disk problems, the function g
1
(x) is directly propor-
tional to the surface charge density q(r, 0) (0rroa),
which, in turn, is essential for nding the capacitance.
Rahman [33] showed that with the representation in Eq.
(143), the function g
1
is given by
g
1
(x) =x
v

o
n=0
a
n
n!
G(1 a n)
(1 x
2
)
b
P
v;a
n
(1 2x
2
)
The method of solution is certainly preferable to that
based on using the numerical quadrature, because it
bypasses the arduous job of evaluating Abel integrals
numerically. Furthermore, it was shown [34] that the
following relation holds:
1
x
_
_
_
_
K
v=2;a
x
v2a
(1 x
2
)
b
P
va;b
n
(1 2x
2
)
_ _
=
G(1 bn)
G(1 b a n)
x
v
(1 x
2
)
ba
P
v;ba
n
(1 2x
2
)
(146)
Formula in Eq. (146) gives a class of spectral relationship
for the operator K
v/2,a
. It can be seen by writing out
Eq. (146) in full that it gives a closed-form expression for a
class of Abel integrals involving Jacobi polynomials. It can
be used to a polynomial solution to Abel integral equa-
tions, which a number of boundary value problems of
electrostatics can be reduced to.
4. The methods described in this article for solving dual
integral equations are also applicable to a system of those
of the form
S
m
i
=2a;2a

n
j =1
c
ij
c
j
(x) =f
i
(x); x c I
1
S
v
i
=2b;2b
c
i
(x) =g
i
(x); x c I
2
By a systematic use of the properties of ErdelyiKober
operators Lowndes was able to show that the problem of
solving a system of simultaneous equations of this type
can be reduced to that of solving a system of simultaneous
integral equations. Details of these results can be found in
Sneddons book [8]. To the best of the writers knowledge,
generalization of these results has not yet been attempted
for the case of simultaneous triple- and quadruple-integral
equations.
5. The theory of Hankel transforms can also be ex-
tended to generalized functions or distributions via em-
bedding theory or adjoint method. Interested readers are
referred to consult the books by Zayed [5], Zemanian
[34,35], and Brychkov and Prudnikov [36].
1912 HANKEL TRANSFORMS
13. COMPENDIUM OF BASIC FORMULAS
For the sake of convenience of the readers, below we give a
compendium of the basic formulas that are of frequent use
in applications.
13.1. Denition of Hankel Transforms
~
ff
v
(s) =
v
[f (r); r s] =
_
o
0
rf (r) J
v
(sr) dr
f (r) =
v
[
~
ff
v
(s); s r] =
_
o
0
s
~
ff
v
(s) J
v
(sr) dr
13.2. Some Properties of Hankel Transforms

m
[f (r); r s] =(1)
m

m
[f (r); r s]
(m= 1; 2; . . . ; n; . . .)

v
[f (ar); r s] =a
2

v
f (r); r
s
a
_ _

v
[r
1
f (r); r s] =
s
2v
[
~
ff
v1
(s)
~
ff
v 1
(s)] (vO0)

n
[f (r a)H(r a); r s] =

o
m=o
a
m
~
ff
m
(s)
a
m
=J
nm
(sa)
1
2
as[(m1)
1
J
nm1
(sa)
(m1)
1
J
nm1
(sa)]

v
[M
v
f (r); r s] = s
2

v
[f (r); r s]
M
v
=
d
2
dr
2

1
r
d
dr

v
2
r
2
(v =0; 1; . . .)

v
r
v1
d
dr
[r
1v
f (r); r s
_ _
= s
v1
[ f (r)]; r s]
13.3. Parsevals Relation
_
o
0
s
~
ff
v
(s) ~ gg
v
(s) ds =
_
o
0
r
0
f (r
0
)g(r
0
) dr
0
13.4. Modied Operator of Hankel Transform and Erdelyi
Kober Operators
I
Z;0
=K
Z;0
=I
I
Z;a
x
2b
f (x) =x
2b
I
Z b;a
f (x)
I
Z;a
I
Z a;b
=I
Z;a b
K
Z;a
x
2b
f (x) =x
2b
K
Zb;a
f (x)
K
Z;a
K
Z a;b
=K
Z;a b
I
Z;n
f (x) =x
2n2Z1
D
n
x
x
2Z 1
f (x)
I
Z;a
f (x) =x
2Z2n1
D
n
x
x
2n2Z 2a 1
I
Z;a n
f (x)
K
Z;n
f (x) =(1)
n
x
2Z1
D
n
x
x
2n2Z 1
f (x)
K
Z;a
f (x) =(1)
n
x
2Z1
D
n
x
x
2n2Z 1
f (x)K
Zn;a n
f (x)
I
1
Z;a
=I
Z a;a
K
1
Z;a
=K
Z a;a
S
Z;a
f (x) =2
a
x
a
H
2Z a
[t
a
f (t); t x]
S
1
Z;a
=S
Z a;a
S
Z;a
f (x) =2
l
x
l
S
Zl=2;a l
[x
l
f (x)]
I
Z a;b
S
Z;a
=S
Z;a b
K
Z;a
S
Z a;b
=S
Z;a b
S
Z a ;b
S
Z;a
=I
Z;a b
S
Z;a
S
Z a;b
=K
Z;a b
S
Z a;b
I
Z;a
=S
Z;a b
S
Z;a
K
Z a;b
=S
Z;a b
13.5. Some Beltrami-Type Relations of Common Occurrence

0
[s
1
~
ff
0
(s); s r]
=
2
p
_
o
r
dt

t
2
r
2
_
_
t
0
xf (x) dx

t
2
x
2
_

0
[s
~
ff
0
(s); s r]
=
2
pr
d
dr
_
o
r
tdt

t
2
r
2
_
d
dt
_
t
0
xf (x) dx

t
2
x
2
_
13.6. Some Useful Relations
_
o
0
s
~
ff
n
(s)
~
FF
m
(s)J
mn
(sr) ds
=
_
o
0
r
0
f (r
0
)F(r
0
)dr
0
F(r
0
)
=
1
p
_
p
0
cos(nf)T
m
r r
0
cos f
R
_ _ _

r
0
sinnfsinnf
R
U
m1
r r
0
cos f
R
_ __
F(R) df
HANKEL TRANSFORMS 1913
J
mn
(sr)J
n
(sr
0
)
=
1
p
_
p
0
cos(nf)T
m
r r
0
cos f
R
_ _ _

r
0
sinnfsinnf
R
U
m1
r r
0
cos f
R
_ __
J
m
(sR) df
S
Z;Z
[(1 x
2
)
a
P
Z;a
n
(1 2x
2
)]
=
G(1 a n)
2
Z a
n!x
1Za
J
1Za 2n
(x)
1
x
_
_
_
_
K
v=2;a
x
v2a
(1 x
2
)
b
P
va;b
n
(1 2x
2
)
_ _
=
G(1 b n)
G(1 b a n)
x
v
(1 x
2
)
ab
P
v;ab
n
(1 2x
2
)
14. SUGGESTED LITERATURE
Readers interested in rigorous proofs of various aspects of
the theory of Hankel transforms are referred to the books
by Sneddon [1,2] Davies [3], Andrews and Shivamoggi [4],
and Zayed [5] and the papers by Erdelyi (15) and Erdelyi
and Kober (16). Many applications of the theory of Hankel
transforms to physical problems are given in the books by
Sneddon [8] and Lebedev et al. [9]. Fractional integrals
and derivatives and their applications to dual-, triple-, and
quadruple-integral equations involving Hankel trans-
forms are discussed at greater length in Sneddon [8,21],
Cooke [19,20,27,28], Borodachev [29], and Samko et al.
[37]. Extension of the theory of Hankel transforms to
generalized functions or distributions is presented in the
books by Zayed [5], Zemanian [34,35], and Brychkov and
Prudnikov [36].
Acknowledgments
The writer wishes to express his deepest gratitude to
Professor Raj Mittra (Department of Electrical Engineer-
ing, Pennsylvania State University) for his kind response
to queries with respect to some of the materials presented
herein. He also gratefully acknowledges encouragement
from his friends S. Rajeswaran and P. K. Jindal during the
period of writing of the article.
BIBLIOGRAPHY
1. I. N. Sneddon, Fourier Transforms, McGraw-Hill, New York,
1951.
2. I. N. Sneddon, The Use of Integral Transforms, McGraw-Hill,
New York, 1972.
3. B. Davies, Integral Transforms and Their Applications,
Springer-Verlag, Berlin, 1978.
4. L. Andrews and B. Shivamoggi, Integral Transform for
Engineers and Mathematicians, Macmillan, New York, 1988.
5. A. I. Zayed, Handbook of Function and Generalized Function
Transformations, CRC Press, Boca Raton, FL, 1996.
6. G. N. Watson, A Treatise on the Theory of Bessel Functions,
Cambridge Univ. Press, London, 1944.
7. A. Erdelyi et al., Tables of Integral Transforms, McGraw-Hill,
New York, 1954, Vol. II.
8. I. N. Sneddon, Mixed Boundary Value Problems in Potential
Theory, North Holland, Amsterdam, 1965.
9. N. N. Lebedev, I. P. Skalskaya, and Ya. S. Uiand, Problems of
Mathematical Physics, translated from Russian, Prentice-
Hall, Englewood Cliffs, NJ, 1965.
10. A. Erdelyi et al., Higher Transcendental Functions, McGraw-
Hill, New York, 1953, Vol. II.
11. J. Jeans, The Mathematical Theory of Electricity and Magnet-
ism, Cambridge Univ. Press, Cambridge, UK, 1908.
12. W. R. Smythe, Static and Dynamic Electricity, McGraw-Hill,
New York, 1968.
13. I. S. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series
and Products, Academic Press, New York, 1980.
14. M. Rahman, On a generalization of Neumanns formula for
the product of two rst kind Bessel functions of integral
orders, J. Appl. Math. Mech. ZAMM 77(2):156157 (1997).
15. A. Erdelyi, On fractional integration and its application to the
theory of Hankel transforms, Quart. J. Math. Oxford 11:293
303 (1940).
16. A. Erdelyi and H. Kober, Some remarks on Hankel trans-
forms, Quart. J. Math. Oxford 11:212221 (1940).
17. H. Kober, On fractional integrals and derivatives, Quart. J.
Math. Oxford 11:193211 (1940).
18. A. Erdelyi and I. N. Sneddon, Fractional integration and dual
integral equations, Can. J. Math. 14:685693 (1962).
19. J. C. Cooke, The solution of triple integral equations in
operational form, Quart. J. Mech. Appl. Math. 18(Part
3):5772 (1965).
20. J. C. Cooke, The solution of triple and quadruple integral
equations and Fourier-Bessel series, Quart. J. Mech. Appl.
Math. 25:247263 (1972).
21. I. N. Sneddon, The use in mathematical physics of Erdelyi-
Kober operators and some of their generalizations, in B. Ross,
ed., Fractional Calculus and Its Applications, Lecture Notes
in Mathematics 457, Springer-Verlag, Berlin, 1975.
22. E. Beltrami, Sulla theoria delle funzione potenziali simme-
triche, Mem. Accad. Sci. Bologna 22(IV):462 (1881).
23. I. N. Sneddon, A relation involving Hankel transforms with
applications to boundary value problems in potential theory,
J. Appl. Math. Mech. 14(1):3340 (1965).
24. E. R. Love, The electrostatic eld of two equal circular coaxial
conducting disks, Quart. J. Math. Mech. 11(2):428451 (1949).
25. J. Meixner, The Behavior of Electromagnetic Fields at
Edges, Inst. Math. Sci. Research Report EM-72, New York
Univ., 1954.
26. R. Mittra and S. W. Lee, Analytical Techniques in the Theory
of Guided Waves, Macmillan, New York, 1971.
27. J. C. Cooke, Triple integral equations, Quart. J. Mech. Appl.
Math. 16:193203 (1963).
28. J. C. Cooke, Some further triple integral equations, Proc.
Edinburgh Math. Soc. 303316 (1963).
29. N. M. Borodachev, On a particular class of solutions of
triple integral equations, J. Appl. Math. Mech. PMM 40(4):
605611 (1976).
1914 HANKEL TRANSFORMS
30. J. J. Bowman, T. B. A. Senior, and P. L. E. Uslenghi, Electro-
magnetic and Acoustic Scattering by Simple Shapes, Wiley,
New York, 1970.
31. P. L. E. Uslenghi, Electromagnetic Scattering, Academic
Press, New York, 1978.
32. E. C. Titchmarsh, Eigenfunction Expansions Associated with
Second-Order Differential Equations, Oxford Univ. Press,
London, 1946, Vol. I.
33. M. Rahman, A note on the polynomial solution of a class of
dual integral equations arising in mixed boundary value
problems of elasticity, J. Appl. Math. Phys. ZAMP
46(1):107121 (1995).
34. A. H. Zemanian, Distribution Theory and Transform Analy-
sis, McGraw-Hill, New York, 1965.
35. A. H. Zemanian, Generalized Integral Transforms, Dover,
New York, 1987.
36. Yu. A. Brychkov and A. P. Prudnikov, Integral Transforms of
Generalized Functions (transl. from Russian), Gordon &
Breach, New York, 1989.
37. S. G. Samko, A. A. Kilbas, and O. I. Marichev, Functional
Integrals and Derivatives: Theory and Applications (transl.
from Russian), Gordon & Breach, Philadelphia, 1993.
HARMONIC OSCILLATORS, CIRCUITS
BERNABE

LINARES-BARRANCO
A

NGEL RODRI

GUEZ-VA

ZQUEZ
National Microelectronics
Center (CNM)
In electronics a harmonic oscillator is an electronic
circuit that generates a sinusoidal signal. This signal
can either be a voltage, a current, or both. Harmonic os-
cillators are not restricted to electronics. They can be
found in many other disciplines. However, they always
can be described by similar mathematical equations. A
very familiar harmonic oscillator is the harmonic pendu-
lum, which is found in many high-school physics text-
books. It is a mechanical system consisting of a mass
suspended by a xed-length thread. Figure 1 illustrates
this. When mass m is slightly separated from its equilib-
rium point (so that angel y in Fig. 1 is sufciently small)
and set free, the earths gravitational force will make it
move toward its resting point. When the mass reaches
the resting point it has gained some speed that will
make it keep running toward the other side of the equi-
librium point, until it stops and comes back. And so it will
oscillate from one side of the equilibrium point to the oth-
er. What happens is that by initially departing the mass
from its equilibrium point, an external agent is increasing
its potential energy. When it is set free the action of
the earths gravitational force, together with the con-
straint imposed by the xed length thread, will gradual-
ly change this initial increase of potential energy into
kinetic energy. At the equilibrium point all potential
energy supplied initially by the external agent is in form
of kinetic energy and speed is maximum. At the points of
maximum elongation the kinetic energy (and speed) is
zero and the original potential energy is recovered. The
pendulum oscillates at constant frequency and, if there
is no friction, it keeps on oscillating indefinitely with con-
stant maximum elongation or amplitude. However, in
practice friction cannot be completely suppressed. Con-
sequently, in order to have a pendulum oscillating perma-
nently there must be a way of supplying the energy lost
by friction.
In the electronic oscillator there is also a mechanism
by which energy of one type is changed into another
type (energy can also be of the same type but interchanged
between different devices). Figure 2 shows a capacitor
connected in parallel with an inductor. At equilibrium
there is no voltage across the capacitor and no current
through the inductor. However, if by some means, an
initial voltage (or equivalently, charge) is supplied to the
capacitor, its stored energy increases. The inductor pro-
vides a path to discharge the capacitor so that a current
builds up through the inductor. However, by the time
the capacitor has zero charge the current owing through
the inductor is maximum and the inductor stores all the
original capacitor energy in the form of magnetic ux
energy. The consequence is that the current keeps owing
through the inductor, charging now the capacitor oppo-
sitely, until the current is zero. If there are no
resistive losses this process will continue indefinitely; ca-
pacitor and inductor continue to interchange their stored
energies. The voltage across the capacitor will be sinusoi-
dal in time, and so will be the current through the induc-
tor. The amplitude (or maximum elongation) of the voltage
oscillations is equal to the initial voltage supplied to the
capacitor. In practice both capacitor and inductor have
resistive losses, so that in order to keep the system oscil-
lating indefinitely there must be a way of supplying the
energy being lost.
m

Figure 1. The mechanical pendulum behaves as a harmonic os-


cillator in the limit of very small maximum angle deviations.
HARMONIC OSCILLATORS, CIRCUITS 1915
1. IDEAL RESONATOR MATHEMATICAL MODEL
In Fig. 2 the capacitor voltage v
C
and its current i
C
are
related mathematically by the expression
i
C
=C
dv
C
dt
(1)
where C is the capacitors capacitance. For the inductor,
its voltage v
L
and current i
L
are related by
v
L
=L
di
L
dt
(2)
where L is the inductors inductance. Besides this, the
circuit of Fig. 2 imposes the following topological con-
straints
v
C
=v
L
i
C
= i
L
(3)
Solving Eqs. (1)(3) yields
d
2
v
C
dt
2

1
LC
v
C
=0 (4)
The solution to this second-order time-domain differential
equation is
v
C
(t) =v
C
(0) cos (ot) i
L
(0)

L=C
_
sin(ot) (5)
where v
C
(0) is the capacitor voltage at time zero, i
L
(0) is
the inductor current at time zero, and o is the angular
frequency of the resulting oscillation whose value is
o=
1

LC
_ (6)
Using Eqs. (3), (5), and (6) in Eq. (1) results in
i
L
(t) =i
L
(0) cos (ot) v
C
(0)

C=L
_
sin(ot) (7)
By means of basic trigonometric manipulation, Eqs. (5)
and (7) can be rewritten as
v
C
(t) =V
max
cos (ot j)
i
L
(t) =V
max

C=L
_
sin(ot j)
(8)
where
V
max
=

v
2
C
(0) i
2
L
(0)L=C
_
j= arctan
i
L
(0)
v
C
(0)

L=C
_
_ _ (9)
Equation (8) reveals that v
C
(t) and i
L
(t) have a phase shift
of p/2 radians. This is usually referred to as v
C
(t) and i
L
(t)
being in quadrature, and the resonator in Fig. 2 as being a
quadrature resonator or oscillator. Note that the maxi-
mum oscillation amplitudes (V
max
or V
max

C=L
_
, respec-
tively) depend on the initial conditions v
C
(0) and
i
L
(0) = C
.
v
C
(0).
Usually, differential equations like Eq. (4) are not
solved directly in the time domain but in the frequency
domain. For this, let us take the Laplace transform of
Eq. (4)
s
2
V
C
(s) sv
C
(0)
.
v
C
(0)
V
C
(s)
LC
=0 (10)
where V
C
(s) is the Laplace transform of v
C
(t). Since
i
L
(0) = C
.
v
C
(0), Eq. (10) can be rewritten as
V
C
(s) =
s
s
2
1=LC
v
C
(0)
1
s
2
1=LC
i
L
(0)
C
(11)
Taking the inverse Laplace transform of Eq. (11) results in
Eq. (5). Usually in circuits, the initial conditions involved
in the Laplace transform are ignored and Eq. (10) is sim-
plied to
s
2

1
LC
=0 (12)
which has the following solutions
s
1
=jo; s
2
= jo
o=
1

LC
_
(13)
Solutions s
1
and s
2
are called the poles of the system, and
in this case the two poles are complex conjugate and are
purely imaginary (their real part is zero). In circuits, peo-
ple dont take the inverse Laplace transform to know the
solution. They know that if a system has a pair of purely
v
C
i
L
+

Figure 2. An ideal capacitor connected in parallel with an ideal


inductor form a harmonic oscillator.
1916 HARMONIC OSCILLATORS, CIRCUITS
imaginary poles the signals have a sinusoidal steady state
whose amplitude depends on the initial conditions.
2. REAL RESONATOR MATHEMATICAL MODEL
As mentioned earlier, the circuit of Fig. 2 is ideal. In prac-
tice there will always be resistive losses in the capacitor, in
the inductor, or in both. Either introducing a small resis-
tance R
L
in series with the inductor, a large resistance R
C
in parallel with the capacitor, or both can model this.
Solving such a circuit yields the following time-domain
differential equation
d
2
v
C
(t)
dt
2
b
dv
C
(t)
dt
o
2
v
C
(t) =0 (14)
with
b =
1
R
C
C

R
L
L
o
2
=
1 R
L
=R
C
LC
(15)
The solution to Eq. (14) is
v
C
(t) =V
max
e
bt=2
cos (o
o
t r) (16)
where o
0
2
=o
2
(b/2)
2
. Parameters V
max
and r can be
found from the initial conditions v
C
(0) and
.
v
C
(0),
V
max
=
v
C
(0)
cos r
r = arctan
.
v
C
(0)
b
2
v
C
(0)
o
o
v
C
(0)
_
_
_
_
_
_
(17)
However, circuit people prefer to solve Eq. (14) in the fre-
quency domain by taking its Laplace transform:
s
2
bs o
2
=0 (18)
The solution to this equation provides the following poles
s
1
=
b
2
jo

1
b
2o
_ _
2

=
b
2
jo
o
s
2
=
b
2
jo

1
b
2o
_ _
2

=
b
2
jo
o
(19)
which are two complex conjugate poles with a negative
real part. Circuit people know that when a system has a
pair of complex conjugate poles with a negative real part,
the system oscillates in a sinusoidal fashion with an am-
plitude that vanishes after some time. This is what
Eq. (16) shows. The amplitude of the oscillations A(t) =
V
max
e
bt/2
decreases exponentially with time. After a few
time constants 2/b the amplitude is negligible and one can
consider that the system has stopped oscillating. Conse-
quently, in practice, the circuit of Fig. 2 is not useful for
building an oscillator.
Imagine that somehow we could make R
L
or R
C
(or
both) negative, so that bo0. A negative resistance
behaves as an energy source that replaces the energy dis-
sipated by positive resistances. In this case the poles
would have a positive real part and the amplitude of the
oscillations
A(t) =V
max
e
bt=2
(20)
would increase exponentially in time, assuming V
max
a0
(due to noise V
max
cannot be exactly zero all the time).
This would yield an oscillator with initial startup but that
would be unstable, because its amplitude would be out of
control. What circuit designers do to build oscillators with
stable amplitude is to make the term b in Eq. (18) depend
on the instantaneous oscillation amplitude A(t)
b(t) =b(A(t)) (21)
and in such a way that b increases with A, b is negative for
A=0 (to ensure initial startup), and b becomes positive
above a certain amplitude. This is called amplitude con-
trol. For instance, assume that by adding some special
circuitry to Fig. 2 we are able to make
b(A) = b
0
b
1
A (22)
where b
0
and b
1
are positive constants (note that A is
always positive). Initially, if A=0, b = b
0
is negative and
the real part of the poles is positive: amplitude A(t) in-
creases exponentially with time. As A(t) increases b will
eventually become positive (poles with negative real part)
and this will decrease the amplitude A(t). The conse-
quence of these two tendencies is that a steady state will
be reached for which b =0 and the amplitude is constant.
Solving Eq. (22) for b =0 yields the value of the steady
state oscillation amplitude A
o
A
o
=
b
0
b
1
(23)
Note that, as opposed to the ideal resonator, the steady-
state amplitude is independent of any initial conditions.
In general, a harmonic oscillator does not have to be a
second order system like the case of Eq. (14) or Eq. (18). It
can have any order. What is important is that it has a pair
of complex conjugate poles whose real part can be con-
trolled by the oscillation amplitude (so that the real part
becomes zero in the steady state), and that the rest of the
poles (either complex conjugate or not) have negative real
parts. The way b depends on A does not have to be as in
Eq. (22). Strictly speaking, the conditions are
b(A=0) = b
0
o0 for initial startup
b > 0 for some A
db(A)
dA
> 0 for stable amplitude control
(24)
HARMONIC OSCILLATORS, CIRCUITS 1917
This will ensure stable oscillator operation. In what fol-
lows we will concentrate on second order systems and
will provide two different ways of performing amplitude
control.
3. AMPLITUDE CONTROL BY LIMITATION
Avery widely used method for oscillator amplitude control
is by limitation. This method usually is simple to imple-
ment, so simple that many times it is implicit in the com-
ponents used to build the resonator with initial startup.
This makes practical circuits easy to build, although many
times people dont understand the underlying amplitude
control mechanism.
Let us consider the ideal resonator of Fig. 2 with an
additional resistor R
p
in parallel to make it real. In order
to assure initial startup, let us put a negative resistor in
parallel also. Figure 3a shows a very simple way to im-
plement one using a real (positive) resistor and a voltage
amplier of gain larger than one (for example, two). Cur-
rent I
in
will be
I
in
=
V
in
R
n
(25)
and the structure resistoramplier behaves as a ground-
ed negative resistor of value R
n
. Figure 3b shows how to
build the amplier using an operational amplier and re-
sistors. Connecting this negative resistor in parallel with a
positive one of value R
p
=R
n
R
e
(with R
e
5R
n
), the equiv-
alent parallel resistance would be
R
eq
=
R
2
n
R
e
(26)
which is a very high but negative resistance. Connecting
this equivalent resistor in parallel with the ideal resonator
of Fig. 2 provides an oscillator with initial startup. This is
shown in Fig. 3c.
Because the operational ampliers output voltage can-
not go above its positive power supply V
DD
or below its
negative one V
SS
, the negative resistance emulator circuit
just described works as long as the amplier output is be-
low V
DD
and above V
SS
, or equivalently voltage V
in
is be-
tween V
DD
/2 and V
SS
/2. It is easy to compute the current
through R
eq
as a function of V
in
taking into account this
saturation effect. Figure 3d shows the resulting curve. If
V
SS
/2rV
in
rV
DD
/2 resistor R
eq
behaves as a negative re-
sistance of high value, but if V
in
is outside this range the
slope of I
in
versus V
in
is that of a positive resistance with
much smaller value. In order to analyze what happens to
the circuit of Fig. 3c when the oscillating amplitude in-
creases beyond V
DD
/2 or V
SS
/2 (whichever is smaller) the
concept of describing function can be used.
3.1. Describing Function
Figure 4 shows a sinusoidal signal x(t) applied to a non-
linear element f(x) that outputs a distorted signal y(t).
Signal y(t) is no longer sinusoidal, but it is periodic.
R
n
I
in
V
in
R
R
R
R
L C
V
in
R
n
I
in
V
DD
/2
V
SS
/2
2
V
in
R
n
R

V
o
= 2 V
in
x 2
(a)
(b)
(c)
(d)
+

2
R
n
2
R
n
R
n
Figure 3. A real harmonic oscillator can be made by adding a
negative resistor to a real resonator: (a) a negative resistor can be
emulated using a resistor and a voltage amplier of gain greater
than unity; (b) implementation of negative resistor using an op-
erational amplier and resistors; (c) oscillator composed of capac-
itor inductor and negative resistor; (d) transfer characteristics of
the negative resistor implementation of (b).
1918 HARMONIC OSCILLATORS, CIRCUITS
Consequently, a Fourier series can describe it. The rst (or
fundamental) harmonic has the same frequency as the in-
put sinusoid, while the others have frequencies which are
integer multiples of the rst one. If the block of Fig. 4 is
used in a system such that the end signals will be approx-
imately sinusoidal (as in a harmonic oscillator) then one
can neglect all higher harmonics of the Fourier expansion
of y(t) and approximate it using the rst or fundamental
harmonic only
y(t) -N(A)x(t)
x(t) =A sin(ot)
N(A) =
o
pA
_
2p=o
0
f (x(t)) sin(ot)dt
(27)
Note that this approximation makes y(t) to be linear with
x(t) so that the nonlinear block in Fig. 4 can be modeled by
a linear amplier of gain N(A). Function N(A) is the de-
scribing function of the nonlinear element f (
.
).
This approach is valid for any nonlinear function f (
.
),
but let us consider only piecewise linear functions, as in
Figs. 3 and 4, with three pieces: a central linear piece of
slope m
c
and two external linear pieces of slope m
e
. When
amplitude A is small enough so that x(t) is always within
the central piece then N(A) =m
c
. When A increases be-
yond the central piece N(A) will change gradually towards
value m
e
. In the limit of A=Nthe describing function will
be N(A) =m
e
. Computing the rst Fourier term provides
the exact expression (let us assume V
SS
= V
DD
for sim-
plicity). If AZV
DD
/2
N(A) =m
e
2
m
e
m
c
p
sin
1
V
DD
2A
_ _

V
DD
2A

1
V
DD
2A
_ _
2
_
_
_
_
(28)
and if ArV
DD
/2
N(A) =m
c
(29)
Applying the describing function method to the nonlin-
earity of Fig. 3d results in
I
in
=N(A)V
in
(30)
where N(A) = R
e
/R
n
2
for ArV
DD
/2 and N(A) tends to-
ward 2/R
n
as it increases beyond V
DD
/2. Since N(A) is con-
tinuous and monotonic, there will be a value of A (and only
one) for which N(A) =0. Let us call this value A
o
. Note that
A
o
depends only on the shape of the nonlinear function
f (
.
) of Fig. 3d. Equation (29) is the equation of a resistor of
value R
eq
=1/N(A). For small values of A, R
eq
= R
n
2
/R
e
(high resistance but negative) and the oscillator possesses
exponentially increasing amplitude. When A increases
beyond V
DD
/2, R
eq
will become more and more negative
until N(A) =0. At this point A=A
o
, R
eq
=Nand we have
the ideal resonator. If A increases further, R
eq
becomes
positive and the oscillator presents exponentially
decreasing amplitude. This represents a stable amplitude
control mechanism such that in the steady state A=A
o
and R
eq
=N.
3.2. General Formulation
In general, the block diagram of Fig. 5 describes a har-
monic oscillator with amplitude control by limitation,
where H(s) is a linear block (or lter) and f (x) is the non-
linear element responsible for the amplitude control.
Applying the describing function method to the nonlinear
block results in
y(t) =N(A) x(t) (31)
for a time domain description. For a frequency-domain
description it would be
Y(s) =N(A) X(s) (32)
x(t )
x(t )
y(t )
t
y(t )
x
y
m
e
m
e
m
c
t
Figure 4. A sinusoidal signal applied to a nonlinear element re-
sults, in general, in a distorted output signal.
HARMONIC OSCILLATORS, CIRCUITS 1919
On the other hand, input and output of the linear block or
lter are related in the frequency domain by
X(s) =H(s)Y(s) (33)
Equation (32) and (33) result in
H(s)N(A) =1 (34)
If H(s) is a second-order block it can be described by
H(s) =
a
1
s
2
a
2
s a
3
s
2
a
4
s a
5
(35)
which together with Eq. (34) yields an equation of the form
s
2
sbo
2
=0
b =
a
1
a
2
N(A)
1 a
1
N(A)
o
2
=
a
5
a
3
N(A)
1 a
1
N(A)
(36)
For small amplitudes N(A) is equal to some constant
(e.g., n
0
), and Eq. (36) is called the characteristics equa-
tion. It must be assured that b(A=0)o0. This is usually
referred to as the oscillation condition. For stable ampli-
tude control it should be
db(A)
dA
> 0 (37)
and o
2
must be kept always positive for all possible values
of A. In practice, it is desirable to make in Eq. (35) a
1
=a
3
=0, which will make o
2
and b not be coupled through a
common parameter. This way the oscillation amplitude
and frequency can be controlled independently.
3.3. A Practical Example: The Wien Bridge Oscillator
In a practical circuit it is not convenient to rely on induc-
tors because of their limited range of inductance values,
high price, and, in VLSI (very large scale integration) de-
sign, they are not available unless one operates in the GHz
frequency range. But it is possible to implement the lter
function of Fig. 5 without inductors. The Wien bridge os-
cillator of Fig. 6 is such an example. Figure 6a shows its
components: two resistors, two capacitors, and a voltage
amplier of gain k
0
. Figure 6b illustrates an implementa-
tion using an op amp and resistors for the voltage ampli-
er, and Fig. 6c shows its piecewise linear transfer
characteristics. Using the describing function, the effec-
tive gain of the amplier k(A) can be expressed as a func-
tion of the sinusoidal amplitude A at node v
1
k(A) =k
0
N(A) (38)
Y(s)
H(s)
f (x)
y(t ) x(t ) X(s)
Figure 5. A general block diagram of an oscillator with ampli-
tude control by limitation consists of a linear lter and a nonlin-
ear amplitude controlling element connected in a loop.
C
1
v
1
v
2
k
0
R
1
R
2
(a)
(b)
(c)
+

C
2
C
1
C
2
R
2
v
2
v
1
k
0
V
DD
V
DD
R
1
R
m
(k
0
1)R
m
v
1
v
2
v
+
v

Figure 6. The Wien bridge oscillator is an example of an oscil-


lator that does not require an inductor: (a) it consists of two re-
sistors, two capacitors, and a voltage amplier circuit; (b) the
voltage amplier can be assembled using an opamp and two re-
sistors; (c) the resulting voltage amplier has nonlinear transfer
characteristics.
1920 HARMONIC OSCILLATORS, CIRCUITS
where k
0
N(A) is the describing function for the function in
Fig. 6c and is given by Eq. (28) with m
c
=k
0
, m
e
=0, and
the breakpoint changes from V
DD
/2 to V
DD
/k
0
. Conse-
quently, the frequency-domain description of the circuit
in Fig. 6a is
s
2
sb o
2
=0
b=
1
R
2
C
2

1
R
1
C
1

k
0
N(A) 1
R
1
C
2
o
2
=
1
R
1
C
1
1
R
2
C
2
(39)
For initial startup it must be bo0 for A=0. The nal am-
plitude A
o
is obtained by solving b(A
o
) =0, and the fre-
quency of the oscillation is o (in radians per second) or
f =o/2p (in hertz). Optionally, the diodes in Fig. 6b, con-
nected to voltage sources v

and v

, can be added to
control the oscillation amplitude. These diodes changes
the saturation voltage V
DD
of Fig. 6c, and hence will mod-
ify the describing function N(A).
In general, when using amplitude control by limitation,
a practical advice is to make b
0
as close as possible to
zero but without endangering its sign. This way the non-
linear element will distort very little the nal sinusoid,
because it needs to use only a small portion of its nonlin-
ear nature to make b(A) become zero. If b
0
is too large
the resulting waveform will probably look more like a
triangular signal than a sinuoidal one.
4. AMPLITUDE CONTROL BY AUTOMATIC
GAIN CONTROL
Let us illustrate the amplitude control by AGC (automatic
gain control) using an OTA-C oscillator. An OTA (opera-
tional transconductance amplier) is a device that deliv-
ers an output current I
0
proportional to its differential
input voltage V
in
. Figure 7a shows its symbol and Fig. 7b
its transfer characteristics. The gain (slope g
m
in Fig. 7b)
is called the transconductance. This gain is electronically
tunable through voltage V
bias
(depending on the technol-
ogy and the design, the tuning signal can also be a cur-
rent). Using these devices, the self-starting oscillator of
Fig. 7c can be assembled. Note that g
m1
, g
m2
, and C
1
em-
ulate an inductance of value L=C
1
/(g
m1
g
m2
), g
m3
emu-
lates a negative resistance of value R
3
= 1/g
m3
, and g
m4
emulates a positive one of value R
4
=1/g
m4
. The charac-
teristics equation of the OTA-C oscillator is
s
2
bs o
2
=0
b =
g
m4
g
m3
C
2
o
2
=
g
m1
C
1
g
m2
C
2
(40)
To assure initial startup V
b3
and V
b4
must be such that
g
m3
4g
m4
. By making g
m3
(or g
m4
) depend on the oscilla-
tion amplitude A, an AGC for amplitude control can be
realized. This is illustrated in Fig. 7d, where the box la-
beled Oscillator is the circuit in Fig. 7c, the box labeled
PD is a peak detector, the large triangle represents a
differential input integrator of time constant t
AGC
, the
small triangle is an amplier of gain m necessary for sta-
bility of the AGC loop, and the circle is a summing circuit.
The output of the peak detector A
pd
(t) follows (with a little
delay) A(t), the amplitude of the sinusoid at V
0
. The error
signal resulting from subtracting A
pd
and V
ref
is integrated
and used to control g
m4
. If A
pd
4V
ref
gain g
m4
will
V
bf
V
bf
V
b4
V
ref
A
pd
PD
m
Oscillator
+

C
1
C
2
V
in
+
g
m
V
bias
I
0
I
SS
I
SS

+

+
V
in
g
m
g
m1
g
m2
+

g
m4
+

g
m3
V
b3
V
b4
V
0
V
0
(a)
(b)
(c)
I
0

Figure 7. An oscillator with amplitude control by AGC can be


made easily with OTAs and capacitors (OTA-C): (a) an OTA de-
livers an output current proportional to its differential input volt-
age; (b) it has nonlinear transfer characteristics; (c) a self-starting
OTA-C oscillator can be made with four OTAs and two capacitors;
(d) the amplitude control by AGC requires an additional peak de-
tector and integrator.
HARMONIC OSCILLATORS, CIRCUITS 1921
increase (making b positive, thus decreasing A), and if
A
pd
oV
ref
gain g
m4
will decrease (making b negative, thus
increasing A). In the steady state A=A
pd
=V
ref
and g
m4
will automatically be adjusted to make b =0. Note that
V
ref
must be such that the node voltages are kept within
the linear range of all OTAs, otherwise amplitude control
by limitation may be taking place.
OTA-C oscillators are convenient for AGC because their
gain can be adjusted electronically. In order to do this for
the Wien bridge oscillator of Fig. 6, either a capacitor or a
resistor must be made electronically tunable (using a vari-
cap or a JFET). Also, OTA-C oscillators are interesting
because they do not need resistors, and this is very at-
tractive for VLSI in CMOS technology where resistors
have very poor electrical characteristics and a limited
range of values.
4.1. Stability of Automatic Gain Control Loop
An AGC loop for amplitude control, like the one in Fig. 7d,
presents a certain dynamic behavior that can be analyzed
in order to (1) make sure it is a stable control loop and (2)
optimize its time response.
In Fig. 7d the peak detector output A
pd
(s) can be mod-
eled as a delayed version of A(s)
A
pd
(s) =A(s)(1 st
pd
) (41)
where A
pd
(s) and A(s) are the Laplace transforms of the
small-signal components of A
pd
(t) and A(t), respectively.
Signal V
b4
(s) [the Laplace transform of the small-signal
component of V
b4
(t)], according to Fig. 7d satises
V
b4
(s) =
1
st
AGC
[(1 smt
AGC
)A
pd
(s) V
ref
(s)] (42)
and controls parameter b in Eq. (40). Let us assume that
b(t) follows instantaneously V
b4
(t) so that
b(s) =aV
b4
(s) (43)
Now what is left in order to close the control loop is to
know how the amplitude A(t) (or A(s) in the frequency do-
main) at node V
0
depends on b.
This dependence can easily be obtained from the time-
domain differential equation [like Eq. (14)] in the following
way. Assume that b(t) is a time dependent signal that has
small changes around b =0 and keeps A(t) approximately
constant around A
o
. Then the solution to V
0
(t) [or v
C
(t) in
Eq. (14)] can be written as
V
0
(t) =A(t) cos (o
0
t j) (44)
where A(t) =A
o
a(t) and [a(t)[5A
o
. Substituting Eq. (44)
into v
C
(t) of Eq. (14) yields the following coefcients for the
cos(
.
) and sin(
.
) terms, respectively, which must be iden-
tically zero [if Eq. (44) is indeed a solution for Eq. (14)]:
d
2
A(t)
dt
2
b(t)
dA(t)
dt
b
2
(t)
A(t)
4
=0
2
dA(t)
dt
A(t)b(t) =0
(45)
The rst equation is not of much use, but from the second
it follows that
A(t) =A(t
0
)e
(1=2)
_
t
t
0
b(t)dt
(46)
When the AGC loop is in its steady state A(t) =A
o
a(t)
and the integral is a function that moves above and below
zero but is always close to zero. Consequently, the expo-
nential can be approximated by its rst-order Taylor
expansion resulting in
A(t) -A(t
0
) 1
1
2
_
t
t
0
b(t) dt
_ _
=a(t) -
A(t
0
)
2
_
t
t
0
b(t) dt -
A
o
2
_
t
t
0
b(t) dt
(47)
In the frequency domain this is
A(s) -
A
o
2s
b(s) (48)
From Eqs. (41)(43) and (48) a loop equation for the AGC
control can be written
A(s) =
V
ref
(s)
s
2
k
1
sk
2
1
k
1
=
2t
AGC
aA
o
mt
AGC
t
pd
k
2
=mt
AGC
t
pd
(49)
This equation represents a stable control system if the
poles have negative real part. This is achieved if k
1
Z0 and
k
2
40. Parameters k
1
and k
2
can also be optimized for op-
timum amplitude transient response [e.g., after a step re-
sponse in V
ref
(t)].
5. VOLTAGE-CONTROLLED HARMONIC OSCILLATORS
An oscillator whose frequency can be electronically con-
trolled is required in many applications. Such an oscillator
is called a voltage-controlled oscillator (VCO), although
sometimes the control parameter can also be current.
In the case of the Wien bridge oscillator of Fig. 6 the
frequency of oscillation o is controlled by R
1
, R
2
, C
1
, and
C
2
. Changing one or more of these parameters would en-
able external control of the frequency. In order to have an
electronic control there are two options: (1) continuous or
analog control and (2) discrete or digital control.
1922 HARMONIC OSCILLATORS, CIRCUITS
For analog control of the Wien bridge oscillator of Fig. 6
either a voltage-controlled resistor (JFET) or a voltage-
controlled capacitor (varicap) is needed. Digital control
can easily be implemented by using a binary-weighted ar-
ray of resistors or capacitors that are switched in and out
of the circuit by means of a digital bus. This is exemplied
in Fig. 8 for resistor R
2
. Signals s
i
are either 0 when the
switch is open or 1 when it is closed. This yields
1
R
2
=
1
r
1
2
n

n
i =1
s
i
2
i
_ _
=
1
r
d
n
(50)
where d
n
is a number that ranges from 1/2
n
to 1 in steps of
1/2
n
. Number d
n
is represented in binary format by the
bits {S
n
S
n1
yS
2
S
1
}.
The OTA-C oscillator of Fig. 7 is much better suited for
analog or continuous control of frequency. If g
m1
=g
m2
and
C
1
=C
2
, the frequency is equal to o=2pf =g
m1
/C
1
. Since
voltage V
bf
in Fig. 7d controls simultaneously g
m1
and g
m2
(making them equal), this voltage can be used directly to
control the frequency of the VCO.
Whether a VCO is made with OTAs and capacitors, or
with resistors, capacitors, and op amps, or uses some other
technique, in general it turns out that the frequency does
not have a linear dependence on the control voltage. In
practical circuits it also happens that if the control voltage
is maintained constant, the frequency may change over
long periods of time due to temperature changes which
cause device and circuit parameters (such as transcon-
ductance and resistance) to drift. Both problems can be
overcome by introducing a frequency control loop.
6. FREQUENCY-CONTROLLED LOOP
Figure 9a shows the basic concept of frequency-controlled
loop for VCOs. It consists of a VCO (e.g., the one in
Fig. 7d), a differential input voltage integrator, and a fre-
quency to voltage converter (FVC) circuit. Voltage V
CO
is
now the external control of the VCO frequency. The FVC
circuit delivers an output voltage V
FVC
that depends lin-
early on the frequency f of its input signal V
OSC
:
V
FVC
=rf V
F0
(51)
Parameters r and V
F0
must be constants and should not
depend on temperature or technological parameters that
change from one prototype to another. If such an FVC is
available, the circuit in Fig. 9a would stabilize at V
FVC
=
V
CO
. According to Eq. (51), this means that the resulting
oscillation frequency f
0
depends on V
CO
as
f
0
=
V
CO
V
F0
r
(52)
which is linear and temperature-independent.
A possible implementation with OTAs of the FVC is
shown in Fig. 9b. It uses two OTAs of transconductance g
0
,
a capacitor C, a peak detector, a switch, a temperature-
independent voltage reference V
ref
, and a monostable trig-
gered by the oscillating signal V
OSC
. During each period T
=1/f of signal V
OSC
the monostable delivers a pulse of
constant width t
0
, which must be temperature-indepen-
dent and well calibrated. Many times it is convenient to
add a sine-to-square-wave converter (and even a frequen-
cy divider) between V
OSC
and the monostable. The circuit
of Fig. 9b uses three components that are not tempera-
ture-independent and may vary over time and from one
prototype to another: the two OTAs and the capacitor.
However, provided both OTAs have the same transcon-
ductance (which is a reasonable assumption for VLSI im-
plementations), the resulting parameters r and V
F0
do not
depend on C nor g
0
. An example of the time waveforms of
u(t) and v(t) of Fig. 9b is shown in Fig 9c. During each
Digital
bus
R
2
s
1
s
2
s
n
2r 2
2
r 2
n
r
2
n
r
Figure 8. The frequency of the Wien bridge oscillator can be
digitally controlled by replacing one of the resistors by a binarily
weighted resistor array controlled through a digital databus.
Monostable
V
OSC
V
ref
VCO
+

+
+
V
bf
V
OSC
V
FVC
V
CO
t
0
t
0
u(t )
u(t )
T
t
v(t )
v(t )
g
o
g
0
C
PD
(a)
(b)
(c)
FVC

Figure 9. A frequency-controlled loop can provide a linear de-


pendence between tuning voltage and VCO frequency, and can
also make this dependence temperature and prototype indepen-
dent: (a) it can be made by adding a FVC and an integrator to a
VCO; (b) the FVC can be made with a calibrated monostable,
a reference voltage, a peak detector, two OTAs, a capacitor, and a
switch; (c) after a transient the FVC output stabilizes to a steady-
state voltage that depends linearly on the input signal frequency.
HARMONIC OSCILLATORS, CIRCUITS 1923
period Tof V
OSC
the monostable is triggered once, turning
the switch ON during a time t
0
. While the switch is in ON
state capacitor C is charged by a constant current
g
0
(V
ref
v(t)), and when the switch is OFF a constant cur-
rent of value g
0
v(t) discharges it. The output of the FVC
v(t) changes from cycle to cycle but is constant during each
cycle. If v
m
is its value during one cycle and v
m1
for the
next one, it follows that
u(t T) =u(t)
g
0
(V
ref
v
m
)
C
t
o

g
0
v
m1
C
(T t
0
)
(53)
where u(t) is taken at one of peaks: u(t) =v
m
and u(t T) =
v
m1
. Consequently, we obtain
v
m1
=
v
m
(C g
0
t
0
) V
ref
g
0
t
0
Cg
0
(T t
0
)
(54)
In the steady of the FVC v
m1
=v
m
. Applying this condi-
tion to Eq. (54) and calling V
FVC
the stabilized value of v
m
yields
V
FVC
=
V
ref
t
0
T
=V
ref
t
0
f (55)
Consequently, the circuit of Fig. 9b implements a FVC
with r=V
ref
t
0
and V
F0
=0, which both are temperature-
and prototype-independent.
7. FURTHER CONSIDERATIONS
The different concepts and considerations mentioned so
far have been illustrated with practical circuits using ei-
ther resistors, capacitors, and op amps, or using OTAs and
capacitors. There are many other circuit techniques avail-
able that can be used to implement the different blocks
and equations needed for stable harmonic oscillator cir-
cuits. Some of these techniques could be continuous cur-
rent mode, switched capacitor, switched current, digital
circuit techniques, or even any combination of these
techniques.
Depending on the frequency range of the oscillator it
may be necessary to consider circuit parasitics that have
not been mentioned so far. For example, op amps and
OTAs both have nonideal input and output impedances,
leakage currents, and most importantly gains that are
frequency-dependent. All these parasitics result in modi-
ed characteristics equations. A very sensitive parameter
to parasitics is the oscillation condition b
0
required for
initial startup. Since in practice it is desirable to have b
0
very close to zero but still guarantee its negative sign, it is
apparent that parasitics can result in either very negative
(resulting in very distorted sinusoids) or positive (result-
ing in no oscillation) values. Each circuit technique has its
own parasitics, and depending on the intended frequency
range, they will have a different impact on the nal
oscillator performance. Consequently, for good oscillator
design the dominant parasitics need to be well known and
taken into account.
Another interesting and advanced issue when design-
ing oscillators in distortion. Both amplitude control mech-
anisms, limitation and AGC, are nonlinear and will
introduce some degree of distortion. Is there a way to pre-
dict how much distortion will render an oscillator?
7.1. Distortion for Amplitude Control by Automatic
Gain Control
In an oscillator with AGC for amplitude control, as in
Fig. 7d, the element that introduces most of the distortion
is the peak detector. Figure 10 shows examples of peak
detectors based on one-phase or half-wave (Fig. 10a), two-
phase or full-wave (Fig. 10b), and four-phase (Fig. 10c)
rectifying principles. For the four-phase case, either the
oscillator should provide two phases with p/2 shift, or an
additional integrator is needed. Peak detectors with more
phases can be implemented by linearly combining previ-
ous phases. Increasing the number of phases in the peak
detector results in faster response [delay t
pd
in Eq. (41) is
smaller for more phases] and less distortion. However, all
phases have to present the same amplitude; otherwise
distortion will increase. In practice, as the number of
phases increases, it becomes more difcult (due to offsets
V
0
C
PD
V
0
(a)
(b) (c)
A
pd
A
pd
A
pd
A
pd
A
pd
t t
C
PD
C
PD
I
discharge
I
discharge
V
0
1
V
0
1
V
0
2
V
0
V
0
V
0
V
0
V
0
2
A
pd
I
discharge
+
+

+
+

t
Figure 10. Possible implementations for a peak detector: (a) one
phase-based (or half-wave-rectier based) (b) two phase-based (or
full-wave-rectier-based), and (c) four-phase-based peak detector.
1924 HARMONIC OSCILLATORS, CIRCUITS
and component mismatch) to keep the amplitude of the
phases sufciently equal.
In the peak detectors of Fig. 10, whenever one of the
phases becomes larger than A
pd
it slightly turns ON its
corresponding P transistor injecting a current into C
PD
until A
pd
increases sufciently to turn OFF the P transistor.
The discharge current ensures that A
pd
will follow the
amplitude of the oscillations if it decreases. Increasing
I
discharge
results in faster response but higher distortion.
Whatever peak detector is used, waveform A
pd
(t) is not
constant nor sinusoidal. It has a shape similar to those
shown in Fig. 10. Since A
pd
(t) is periodic its Fourier series
expansion can be computed:
A
pd
(t) =A
o

o
n=1
a
n
cos (no
0
t j
n
) (56)
The high-order harmonic components {a
2
, a
3
, y} are those
which contribute to distortion at node V
0
(in Fig. 7d). Ap-
plying the ltering functions that go from node A
pd
,
through V
b4
, to V
0
[Eqs. (41)(43) and (48)] to these high-
er-order harmonics provides their amplitudes at node V
0
:
[A
on
[ =
aA
o
2t
AGC
n
2
o
2
0
[1 jno
0
mt
AGC
[[1 jno
0
t
pd
[[a
n
[ (57)
The total harmonic distortion at the output of the oscilla-
tor is then dened as
THD(V
0
) =

o
n=2
A
on
A
o
_ _
2

_
(58)
7.2. Distortion for Amplitude Control by Limitation
This problem is computationally complicated but can be
solved by harmonic balance. Consider the general block
diagram of Fig. 5. In the steady state, periodic waveforms
x(t) and y(t) can be expressed by their respective Fourier
series expansions
x(t) =x
1
cos (o
0
t)

o
n=2
x
n
cos (no
0
t j
n
)
y(t) =

o
n=1
y
n
cos (no
0
t f
n
)
(59)
where
x
n
[ [5x
1
[ [
y
n

5 y
1

(60)
In general, this problem is solved numerically for a nite
set of harmonics. Let N be the highest harmonic to be
computed. Since f(x(t)) is also periodic, computing its Fou-
rier series expansion yields that of y(t). Therefore, given a
set of parameters for x(t) {x
1
, y x
N
, j
2
, y j
N
} the set of
parameters {y
1
, y y
N
, f
1
, y f
N
} for y(t) can be obtained
this way. By applying each component of y(t) (character-
ized by y
n
and f
n
) to lter H(s) yields the corresponding
component for x(t):
x
n
=y
n
H(jno
0
)

j
n
=f
n
phase(H(jno
0
))
(61)
By iterating this procedure until all values x
n
, y
n
, j
n
, and
f
n
converge, the distortion of x(t)
THD(x) =

N
n=2
x
n
x
1
_ _
2

_
(62)
or y(t)
THD(y) =

N
n=2
y
n
y
1
_ _
2

_
(63)
can be predicted.
FURTHER READING
K. K. Clarke and D. T. Hess, Communication Circuits: Analysis
and Design, Addison-Wesley, Reading, MA; 1978.
A. Gelb and W. Vander Velde, Multiple Input Describing Func-
tions and Nonlinear System Design, McGraw-Hill, New York,
1968.
E. J. Hahn, Extended harmonic balanced method, IEE Proc. Part
G, Circ. Devices Syst. 141:275284 (1994).
B. Linares-Barranco et al., Generation, design and tuning of OTA-
C high-frequency sinusoidal oscillators, IEE Proc. Part G, Circ.
Devices Syst. 139:557568 (1992).
E. Vannerson and K. C. Smith, Fast amplitude stabilization of an
RC oscillator, IEEE J. Solid-State Circ. SC-9:176179 (1974).
HELICAL ANTENNAS
ARLON T. ADAMS
JAY K. LEE
Syracuse University
Syracuse, New York
1. INTRODUCTION
The helix antenna has a long and fascinating history. It
was discovered in 1946 by John Kraus. Since then, new
variations have continued to arise even up to the present
day. The discovery itself is a very interesting story, which
is told in John Kraus book Antennas [1]. It all started with
an afternoon lecture at Ohio State University. John Kraus
listened as the speaker described the wave-guiding helix
used in traveling-wave tubes; he wondered if this inter-
esting helix could be used as an antenna and, after the
HELICAL ANTENNAS 1925
talk, he asked the well-known lecturer. The speaker re-
plied emphatically, No, Ive tried it, and it doesnt work.
That very evening, John Kraus went down to his base-
ment, wound a seven-turn helical coil one wavelength in
circumference and fed it by means of a coaxial line and
ground plane (Fig. 1a). He found that it produced a sharp
beam of circularly polarized radiation off the open end of
the helix. So the helix antenna was born, despite the ad-
vice of experts.
1.1. Helical Curves
The helix was well-known in ancient Greece. Geminus de-
scribed it in the rst century B.C. and there are references
to earlier work on the helix. The cylindrical helix may be
dened by considering a right circular cylinder of radius a,
whose axis is the z axis. Using a right-handed cylindrical
coordinate system (r,f,z) the equations of the helix are
x =a cos f; y =a sin f; z =aftan a (1)
where a is the pitch angle of the helix and 2a is the diam-
eter. The lines of the cylinder parallel to the z axis, that is,
the lines (r =a, f=f
o
) are considered to be the generators
of the cylinder. The cylinder is generated by rotating any
generator about the z axis. The helix cuts the generators
at a constant angle p/2 a. It also projects as a sine curve
on any plane parallel to the axis, for example, any plane
that includes the axis.
Figure 1a shows the wire helix antenna. Parameters
are dened as follows:
D is the diameter of helix, which is equal to 2a.
C is the circumference.
S is the spacing between turns.
a is the pitch angle.
L is the length of one turn.
N is the number of turns.
C
l
, S
l
, and L
l
represent the respective distances in wave-
lengths such that C
l
=C/l, and so on. When one turn of
the helix is unrolled on a at plane, the relationships be-
tween the spacing S, circumference C, turn length L, and
pitch angle a can be obtained from the triangle shown in
Fig. 1b as follows:
S=C tan a; a = tan
1
S
C
; L=

C
2
S
2
_
(2)
Thus we need only three independent parameters, C, a, N,
to describe a helix. Note that when a =01, S=0, and the
helix reduces to a planar loop. When a =901, C=0, and the
helix becomes a straight line.
The round-wire helix such as Kraus rst built consists
of a round wire whose centerline is the helix curve of Eq.
(1). The tape helix is a conducting tape of width w, which is
wound around a cylinder or a thin cylindrical tube. Its
centerline is the helix of Eq. (1). A single-wire helix is
called the monolar helix. The double, or bilar, helix
is constructed by adding an additional member, which is
formed by replacing f with fp everywhere in Eq. (1).
The quadrilar helix antenna is formed by adding three
members; f in Eq. (1) is replaced with f(p/2), p, (3p/2). A
left-handed helix can be formed by using a left-handed
coordinate system, or by replacing f in Eq. (1) with 2p f.
Figure 1a shows a right-handed helix. There are other
helical curves. The conical helix lies on the surface of a
cone and cuts the radial lines of the cone, the generators,
at a constant angle. The spherical helix lies on the surface
of the sphere and cuts the generators, for example, the
longitude lines, at a constant angle.
There are many striking examples of the helix in na-
ture, from some of the smallest to some of the largest ob-
jects. Most important of all is the DNA molecule, which is
a double helix. The marks on a snail of the family Helic-
idae resemble a spherical helix. The human ear has a
prominent helical ridge. The Heliconia is a family of herbs
with a helical shape. The helictite is similar to a stalagtite.
Some trees have helical bark. Finally, we have the largest
of the nebulae, the Helix Nebula. In addition, we see many
manmade forms of the helix around us, including auto-
mobile springs, spiral staircases, inductance coils, trans-
former coils, parking lot ramps, automobile antenna coils,
and nally, the lowly screw, which is a combination of
conical and cylindrical helices.
2. MONOFILAR HELICAL ANTENNA
We rst consider the characteristics of a monolar, or un-
ilar, helical antenna. In this section we simply call it a
helical antenna, implying that it is a monolar helix. The
helix commonly operates in two different modes, the nor-
mal mode and the axial mode, depending on the electrical
size of the helix. When the dimensions of the helix are
small compared with a wavelength (D5l, NS5l), the
maximum radiation is normal (or perpendicular) to the
s
L
C = D

(b)
s
z
D
Ground
plane
Coax
(a)
Figure 1. The helical antenna: (a) a helix fed by a coaxial line
and a ground plane; (b) one turn of the helix unrolled on a at
plane.
1926 HELICAL ANTENNAS
helix axis. This condition is called the normal mode. When
the helix circumference is on the order of one wavelength,
the maximum radiation is along the helix axis. Thus, this
type of operation is called the axial mode. The axial mode
helix is a broadband antenna. The radiation from this
axial-mode helix is close to circular polarization along the
axis. There is also a backre mode, which is discussed in a
later section.
2.1. Normal-Mode Helix
Lets consider a helix with its axis along the z axis,
centered at the origin (Fig. 2a). The geometry of the
helix reduces to a loop when the pitch angle a approaches
zero and to a straight wire when it approaches 901. Since
the limiting geometries of the helix are a loop and a dipole,
the far eld radiated by a small helix can be
described by the radiation elds of a small loop and a
short dipole when dimensions are small compared to a
wavelength. The analysis of a small short helix is facili-
tated by assuming that the helix consists of a number of
small loops and short dipoles connected in series as in
Fig. 2b. The diameter of the loops is the same as the
helix diameter (D) and the length of the dipoles is approx-
imately the same as the spacing (S) between turns of the
helix. Because the helix is small and short, the current
distribution is assumed to be uniform in magnitude and
phase over the entire length of the helix. For the same
reason, the far-eld (Fraunhofer) pattern will be indepen-
dent of the number of turns and thus can be obtained by
considering the pattern of a single-turn helix that consists
of a single small loop of diameter D and one short dipole
of length S.
Assume that the complex amplitude of current is I and
the angular frequency is o. The radiation electric eld of
the small loop of diameter D has only an E
f
component,
given by
E
f
=Zk
2
IA
e
jkr
4pr
sin y (3)
where A=pD
2
/4 is the area of the loop, k =o

me
_
is the
propagation constant, and Z =

m=e
_
is the intrinsic im-
pedance. The far eld of the short dipole, or the Hertzian
dipole of length S, has only an E
y
component, given by
E
y
=jomIS
e
jkr
4pr
sin y (4)
The total radiation eld for one turn is then given by
E= ^ aa
y
E
y
^ aa
y
E
f
={ ^ aa
y
jomS ^ aa
f
Zk
2
A]I
e
jkr
4pr
sin y (5)
The normalized radiation eld pattern f(y) of the normal-
mode helix is
f (y) = siny (6)
which is the same as that of the Hertzian dipole and the
small loop and is shown in Fig. 2c. The eld is zero along
the axis (in the endre direction) and is maximum in the
xy plane (y =901), which is normal to the helix axis.
Because E
y
and E
f
are 901 out of phase as shown in
Eqs. (3) and (4), the radiated wave is elliptically polarized.
The axial ratio (AR) of the polarization ellipse of the far
eld is obtained by dividing the magnitude of Eq. (4) by
that of Eq. (3)
AR=
[E
y
[
[E
f
[
=
omS
Zk
2p
l
A
=
Sl
2pA
=
2Sl
(pD)
2
(7)
where we have used k=2p/l, nk =om. Because E
y
and E
f
are 901 out of phase, the polarization ellipse becomes a
circle when |E
y
|=|E
f
|, indicating circular polarization.
Setting AR=1 yields
C=pD=

2Sl
_
or C
l
=

2S
_
l
(8)
Under this condition the radiation eld is circularly po-
larized in all directions except of course along the axis
where the radiation is zero. The polarization ellipse of the
radiation from a helix of constant turn length (L) changes
progressively as the pitch angle a is varied. When a =
tan
1
(S/C) =0 (the helix reduces to a loop), AR=0, E
y
=0,
E= ^ aa
f
E
f
; thus the wave is linearly polarized with hori-
zontal (or perpendicular) polarization. As a increases, the
polarization becomes elliptical, with the major axis of the
D
S
(b)

z
(c)
(a)

z
y
x
r

Figure 2. The normal-mode helix: (a) coordinate system;


(b) loopdipole model; (c) beam pattern.
HELICAL ANTENNAS 1927
ellipse horizontal. When a reaches a value such that con-
dition (8) is satised, AR=1, and the polarization is cir-
cular. With the help of Eq. (2), the condition (8) leads to the
following value of a:
a
CP
= sin
1
1

1L
2
l
_
L
l
_
_
_
_
(9)
As a increases further, the polarization again becomes el-
liptical and the major axis is vertical. Finally, when a =901
(the helix reduces to a dipole), AR=N, E
f
=0, E= ^ aa
y
E
y
;
thus the polarization is linear with vertical (or parallel)
polarization. For small pitch angles (a51), Eq. (9) is sim-
plied to
a
CP
=
C
l
2
(10)
where a
CP
is expressed in radians. For small pitch angles,
circular polarization can occur at frequencies such that
the circumference is very small compared to a wavelength
(C
l
51).
From Eqs. (3) and (4), we note that the loop eld E
f
and
the dipole eld E
y
, respectively, are proportional to the
second and rst powers of frequency. Correspondingly, ra-
diation resistance of loop and dipole are proportional to
the fourth and second powers, respectively. Thus, as fre-
quency decreases, the dipole radiation predominates and
the beam pattern is linearly polarized. In this linearly po-
larized frequency range, the normal-mode helix has some
interesting properties. Its beam pattern is essentially that
associated with the dipoles, that is, a monopole of length
NS above a ground plane. Its impedance, however, is sig-
nificantly affected by the loops.
The normal-mode helix is limited by its size. It has the
same restrictions and limitations that apply to any elec-
trically small antenna. But within those restrictions, it
has certain advantages over a dipole antenna of the same
height. These include a lower frequency for resonance and
a larger radiation resistance, because of both the longer
path of the helical structure. While the dipole may require
additional impedance-matching circuits to achieve reso-
nance, the helix is resonant without supplementary
matching elements. Another advantage over the dipole is
that the helix is exible and more resilient. The higher
radiation resistance, resonant characteristic, and exibil-
ity make the normal-mode helix suitable for small anten-
nas used in mobile communications.
C

=0.66
275 MHz
C

=0.73
300 MHz
C

=0.85
350 MHz
C

=0.97
400 MHz
C

=1.09
450 MHz
C

=1.22
500 MHz
C

=1.35
560 MHz
= 14
n = 6
Figure 3. Measured beam patterns
of the monolar axial mode helix.
(From Kraus [1], r 1988 by Mc-
Graw-Hill, Inc. Reprinted with permis-
sion of the McGraw-Hill Companies.)
Circumference, C

0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
180
150
100
50
0
E

X
R
E

0.5 0.75 1.0 1.5


C

A
x
i
a
l

r
a
t
i
o
V
S
W
R
H
a
l
f
-
p
o
w
e
r

b
e
a
m
-
w
i
d
t
h
1
2
3
4
5
6
1
2
3
4
5
6
7
(a)
(b)
(c)
Figure 4. Measured performance of the monolar axial mode
helix: (a) beamwidth; (b) axial ratio; (c) VSWR. (From Kraus [1],
r 1988 by McGraw-Hill, Inc. Reprinted with permission of the
McGraw-Hill Companies.)
1928 HELICAL ANTENNAS
2.2. Axial-Mode Helix
A very useful mode of operation for the helical antenna is
the axial or endre mode. In this mode the radiation pat-
tern has a single mainbeam along the axis of the helix ( z
direction); that is, it is an endre antenna. Experiments
have shown that the axial mode occurs when the circum-
ference of the helix is approximately one wavelength and
when the helix has several turns. A primary component of
current on the helix is a wave traveling outward from the
feed along the wire at approximately the speed of light and
the radiation is a beam off the end of the helix. Because the
electric eld vector rotates around in a circular fashion as
does the current on the helix, we expect that the radiation
eld is circularly polarized along the helix axis. One very
important feature of the axial-mode helical antenna is its
broadband character. As a rule of thumb, the approximate
bandwidth for the axial mode is given as follows [2]:
3
4
oC
l
o
4
3
(11)
The bandwidth ratio, the ratio of the upper and lower fre-
quencies, is
4
3

3
4
or 1.78, which is close to a 21 bandwidth.
The helix is usually fed axially or peripherally with the
inner conductor of the coaxial line connected to the helix
and the outer conductor attached to the ground plane. The
ground plane can be made from either solid metal or wire
mesh.
2.2.1. Analysis of Radiation Pattern. The axial-mode he-
lix has a circumference of approximately one wavelength,
so the current distribution would not be uniform, and we
assume that there is an outgoing current wave, traveling
along the helical conductor at phase velocity v =pc (where
p is the phase velocity relative to the speed of light c in free
space). Then
I(l) =I
0
e
jbl
(12)
where l is the distance measured along the helix from the
beginning of the turn closest to the ground plane, I
0
the
input current, b =k/p the phase constant of the current
wave. When the total length of one turn is approximately a
wavelength, the current distribution in Eq. (12) has oppo-
site phase (1801 out of phase) on opposite sides of a turn
because they are separated by about a half-wavelength.
Also the helical coil physically reverses current direction
for opposite points. Thus the currents at opposite points of
a turn are essentially in phase, giving rise to reinforce-
ment in the far eld along the helix axis. We can nd the
radiation pattern by using the principle of pattern multi-
plication because a helix with uniform cross section can be
considered as an array of N identical elements (or turns).
We have a uniformly excited, equally spaced array with
spacing S, so the total pattern is the product of the pattern
for one turn (the element pattern) and the pattern for an
array of N isotropic point sources (an array factor). When
the helix is long (say, NS4l), the array factor is much
sharper than the element pattern and hence determines
the shape of the total far-eld pattern.
2.2.1.1. Array Factor. The array factor (AF) of a uni-
formly excited, equally spaced, linear array of N elements
is given by
AF=
sin
N
2
c
N sin
c
2
(13)
c=kS cos y d (14)
where y is an angle measured from the array axis (z axis)
and d is the phase shift between adjacent elements. Ex-
plaining the phase relationships of the axial-mode helix is
difcult. Begin by nding the phase shift d required for
endre operation, because we know that the radiation is
endre. For ordinary endre, we nd the conditions such
that c is zero at y =01.
d = kS 2mp (mis aninteger) (15)
where the term (2mp) reects the basic ambiguity of
phase. Next we nd conditions for increased directivity
HansenWoodyard (HW) endre [3], since this is an opti-
mum form of endre:
d = kS 2mp
p
N
= 2p kS
p
N
_ _
(m=1)
(16)
To be more accurate, the term p/N should be replaced by
2.94/N [3]. However, the choice of p/N hardly changes the
radiation eld [4] and is convenient for expressing other
quantities. We have tried the arbitrary choice (m=1) be-
cause this term2p corresponds roughly to one turn of the
circumference at midband (C
l
=1). Later it will be clear
that other choices are not possible solutions. Experiments
show [1,5] that the phase shift d obtained is close to that of
Eq. (16) at midband (C
l
=1). How does this happen? We
have already pointed out that the term (2p) corresponds
roughly to one turn of the circumference. In addition, the
length around one turn of the helix is greater than a
wavelength (L
l
41) at midband. More importantly, the ve-
locity of travel is less than that of light (pE0.9) at mid-
band [5]. These two additional contributions account for
the minor term (kSp/N). Thus the phase at midband is
explained. However, the experimental data for d track Eq.
(16) fairly well over the entire bandwidth of the axial
mode. How is this possible? We cannot alter our choice of
m; one choice must work for the entire frequency range;
and if p remained constant, Eq. (16) could not be satised
over the entire band. Fortunately, p does vary quite a bit
(from 0.73 to 0.97) over the axial-mode frequency range;
the result is that HW endre described by Eq. (16) is
tracked quite well over most of the band, falling off a little
toward the high end. All in all, this is quite a remarkable
story. The phase, so to speak, locks in to HW endre over
the bandwidth of almost 21. When rst reported by
Kraus, this was called an anomalous phase progression.
It still continues to mystify succeeding generations.
HELICAL ANTENNAS 1929
To summarize, the phase progression along the helix
wire is relatively simple; it corresponds roughly to that of
speed of light along the wire. The phase progression in z,
which determines the phase difference d between turns,
follows the phase progression of the wire. Taking into ac-
count the phase ambiguity (2 mp), we see that HWendre
is obtained at midband. The relative phase velocity p then
changes with frequency just enough to maintain the HW
endre. Another point worth noting is that we have not
discussed backre (y =1801) radiation. It does, in fact, oc-
cur along with the axial mode but is usually suppressed by
the ground plane. It will return, to our advantage, with
the multilar helix:
b =

d
L
=
1
L
kS2p 1
1
2N
_ _ _ _
=
2p
L
S
l

2N1
2N
_ _
(17)
p=
k
b
=
L
l
S
l
(2N1)=2N
(18)
Using p as obtained from Eq. (18) to calculate the array
factor yields patterns in good agreement with measured
patterns. The p value calculated from Eq. (18) is in close
agreement with measured values of p [1]. Therefore, it
appears that the HansenWoodyard increased directivity
condition is a good approximation for helices radiating in
the axial mode. For a typical case where C=l, a =141, N=
10, we nd from Eqs. (2) and (18) that S=C tan a =0.249l,
L=1.031l, and p=0.79. Thus the traveling current wave
has a phase velocity less than that of free space. Finally,
substituting Eq. (16) into Eq. (14) yields
c=kS(cos y 1) 2p
p
N
_ _
(19)
Equations (13) and (19) provide the complete norma-
lized array pattern of the axial-mode helical antenna.
For the element pattern we will need an analysis of
the radiation from a circular loop, which is covered in
the next section.
2.2.2. Circular Loop Radiation. In this section we con-
sider the radiation from a circular loop carrying a current
I(f). The result is useful in understanding the operation of
the helix in both the mono- and multilar forms. In addi-
tion, it will yield an approximate element factor for a sin-
gle turn of the helix. The loop of radius a is centered at the
origin and lies in the xy plane. The current distribution
I(f) may be represented in terms of a complex Fourier se-
ries representation:
I(f) =

o
n=o
I
n
e
jnf
(20)
Consider the typical term I
n
e
jnf
of the current distri-
bution I(f). First we evaluate components A
fn
, A
yn
of
the far-eld magnetic vector potential as follows. Direc-
tions f,y are associated with the eld point rather than
the source point:
A
fn
=
e
jkr
4pr
_
2p
0
I
n
e
jnf
/
cos(f f
/
)e
jka sin y cos(ff
/
)
adf
/
(21)
A
yn
=
e
jkr
4pr
_
2p
0
I
n
e
jnf
/
[sin(f
/
f)]
cos y e
jka sin y cos(ff
/
)
adf
/
(22)
To evaluate A
f
, we introduce the change of variables c=
f
/
f and change limits to obtain
A
fn
=
e
jnf
e
jkr
I
n
a
4pr
_
2p
0
e
jnc
e
jc
e
jc
2
_ _
e
j(ka sin y) cos c
dc
Next, we use the following integral expression for the
Bessel function of the rst kind J
m
(x):
_
2p
0
e
jx cos y
e
jmy
dy =2pj
m
J
m
(x)
A
f
is then evaluated directly to obtain
A
fn
=
e
jnf
(I
n
a)e
jkr
j
n1
4r
[J
n1
(ka sin y) J
n1
(ka sin y)]
(23)
E
fn
= jomA
fn
(24)
Using similar methods, we obtain the following evaluation
of A
y
:
A
yn
=
e
jnf
(I
n
a)e
jkr
j
n1
(j cos y)
4r
[J
n1
(ka sin y)
J
n1
(ka sin y)]
(25)
E
yn
= jomA
yn
(26)
The total elds may, of course, be obtained by adding con-
tributions of all Fourier modes.
Now lets evaluate the far elds along the z axis (y =01,
1801) for each of the separate Fourier modes. We note that
J
n
(0) =0 (na0) and J
0
(0) =1. Evaluating the cases n=
71, we nd that, along the z axis
E
f
E
y
= j (27)
In other words, the modes n=71 representing traveling
waves yield circular polarization along the z axis. Note
that all other traveling-wave modes yield a null on axis. Of
all the Fourier modes, only n=71 radiate in the forward
endre or backre direction. For the helix, we dene for-
ward or backward radiation as radiation away from or to-
ward the feedpoint, respectively.
This result can also be seen by considering currents
around the loop for various modes. For n=71, each cur-
rent element is matched by its opposite across the loop,
1930 HELICAL ANTENNAS
which is in the same direction such as to add along the z
axis and to rotate polarization as time progresses. All the
other modes cancel along the axis. For even modes, each
element is canceled by its opposite across the loop. For
odd modes, a group of elements will cancel. For the general
odd case n, any group of n elements, each separated
by 1801/n, yields zero contribution. For n=5, for example,
any group of ve elements, each separated by 361, yields
zero contribution.
Any currents on the cylindrical surface may be resolved
into f- and z-directed currents. The z-directed currents do
not radiate along the axis. Thus, for currents of any di-
rection, only the n=71 Fourier modes can contribute to
endre or backre. These results will be useful when con-
sidering multilar helices.
This discussion makes it easier to understand the op-
eration of the helical antenna. At low frequencies the
zeroth mode (n=0) is strongly excited because there is
little variation of phase around the cylinder on one turn.
In addition, the impedance of the higher modes is highly
reactive. As frequency increases and C
l
approaches unity,
we have one complete cycle around the cylinder on one
turn, and we expect the e
jf
mode to be excited for a right-
hand helix. The phase velocity of the helix is lower than
that associated with the speed of light, and the impedance
of the mode n= 1 is reasonable and so the axial mode
begins at about C
l
=0.75. Similarly, as frequency increas-
es, we expect the mode n= 2 to appear; this mode would
produce beam pattern deterioration. The axial mode con-
tinues until about C
l
=1.33.
2.2.2.1. Element Pattern of the Axial-Mode Helix. For the
element pattern of one turn of the helix, the current dis-
tribution is assumed to be
I(f
/
) =I
0
e
jbl
=I
0
e
jbaf
/
(28)
where b =k/p, a=D/2, and f
/
is the angle measured from
the x axis. For accurate analysis of the element pattern,
Eq. (28) should be used to calculate the radiation integral.
However, when the helix with several turns operates
in the axial mode (C
l
E1), the array factor dominates
the endre beam pattern and the element pattern
provides minor corrections. Thus it sufces to consider
the radiation eld of a planar loop with C
l
=1, instead
of a three-dimensional one-turn helix. If we also assume
that pE1, then
I(f
/
) - I
0
e
jkaf
/
=I
0
e
j(2p=l)af
/
- I
0
e
jf
/
(29)
Using the simple form of the current distribution in Eq.
(29), we can easily calculate the radiation elds for the
element pattern from Eqs. (23)(26) for C
l
=1 (n= 1):
E
f
(y; f) =C(r) J
0
(sin y) J
2
(sin y) [ ]e
jf
(30)
E
y
(y; f) =C(r) J
0
(sin y) J
2
(sin y) [ ](j cos y)e
jf
(31)
where C(r) gives the r dependence of the elds. Note that
ka=1 when C
l
=1. If we plot the radiation patterns of
|E
y
| and |E
f
| using Eqs. (30) and (31), we obtain a g-
ure-eight pattern for E
y
with a null at 901, and a nearly
omnidirectional pattern for E
f
[6]. From the plot, it is in-
teresting to note that the normalized E
y
can be approxi-
mated by cos y. We also observe that E
y
and E
f
are 901 out
of phase. In particular, when y =01, then |E
y
|=|E
f
|;
thus the radiation eld is circularly polarized in the
endre direction. As one departs from y =01, E
y
decre-
ases more rapidly than does E
f
, so the polarization
becomes elliptical. Finally, it should be noted that Kraus
[1] has analyzed the element pattern, by using a single
turn of a three-dimensional helix with uniform traveling-
wave current.
2.2.3. Beam Patterns. The complete total far-eld pat-
tern is given by the product of the array factor shown by
Eq. (13) and the element pattern in Eq. (30) or (31). How-
ever, the array pattern is much sharper than the element
patterns. Thus the total E
y
and E
f
patterns are nearly the
same, in spite of the difference in the single-turn patterns.
The mainlobes of the E
y
and E
f
patterns are very similar
to those of the array pattern. Therefore, for long helices
(NS4l), a calculation of only the array factor is sufcient
for an approximate pattern of any eld component of the
helix.
The measured patterns of a six-turn helix with a =141
as a function of frequency are presented in Fig. 3. Patterns
are shown over a range of circumferences from approxi-
mately 0.66l to 1.35l. The solid patterns are for the hor-
izontally polarized component (E
f
) and the dashed for the
vertically polarized (E
y
). Both are adjusted to the same
maximum. We observe that the endre beam patterns are
preserved over the range of 0.73oC
l
o1.22, indicating
that the axial-mode helix is a broadband antenna.
2.2.4. Important Parameters. Four important parame-
ters for practical design of an axial-mode helical antenna
are (1) beamwidth (BW), (2) gain or directivity, input im-
pedance and axial ratio (AR). They are all functions of the
number of turns, the turn spacing (or pitch angle), and
the frequency. For a given number of turns, the behavior
of the BW, gain, impedance and AR determines the useful
bandwidth. The nominal center frequency of this band-
width corresponds to a helix circumference of about 1l.
2.2.4.1. Beamwidth. On the basis of a large number of
measurements, King and Wong [7] give the following qua-
siempirical formula for the beamwidths:
HPBW(half -power beamwidth) =
K
B
C
l

NS
l
_ (degrees)
(32)
where K
B
varies from 61 to 70, for
3
4
oC
l
o
4
3
, 121oao151,
and 8.6oNo10. Note that as N increases, the beamwidth
decreases. Figure 4a shows measured HPBWof a six-turn,
141 axial-mode helix as a function of the normalized cir-
cumference (C
l
). E
y
is shown dotted as before. We observe
that HPBW changes slowly over the range of approxi-
mately 0.7oC
l
o1.25.
HELICAL ANTENNAS 1931
2.2.4.2. Gain. The gain of the axial mode helix can be
approximately obtained [8] by
G=K
G
C
2
l
NS
l
(33)
where K
G
is the gain factor, which depends on the design
parameters. King and Wong [7] report that K
G
varies from
4.2 to 7.7. Experiments show that the gain is peak when C
is slightly larger than 1l.
2.2.4.3. Axial Ratio. We have shown from the approxi-
mate analysis described in a previous section that the ra-
diation eld is circularly polarized in the mainbeam
direction (y =01), implying AR=1. With a more accurate
analysis, including the effect of relative phase velocity for
increased directivity, Kraus [1] obtains the axial ratio
along the helix axis as follows:
AR=
2N1
2N
(y =0

) (34)
If N is large, the axial ratio approaches unity and the po-
larization is nearly circular. For example, for a six-turn
helix, AR=
13
12
=1:08 according to Eq. (34). This axial ratio
is independent of frequency or circumference. In Fig. 4b,
the measured values of the axial ratio for the six-turn, 141
axial-mode helix are plotted as a function of the circum-
ference (C
l
). We observe that AR is nearly 1 over the range
of about 0.73oC
l
o1.4. The sense of circular polarization
is determined by the sense of the helix windings.
2.2.4.4. Input Impedance. The input impedance of the
axial-mode helical antenna is nearly purely resistive. The
empirical formulas for the input resistance are given [1]
by
R
in
=140C
l
(35)
within 20% for the case of axial feed, and
R
in
=
150

C
l
_ (36)
within 10% for the case of peripheral feed. Both relations
are valid when 0.8rC
l
r1.2, 121rar141 and NZ4. With
a suitable matching section, R
in
can be made any desired
value from 50 to 150O. In the inset of Fig 4c, trends of
input resistance R and reactance X are shown as a func-
tion of the relative frequency or circumference. Note that
R is relatively constant and X is very small for
0.7oC
l
o1.5. Figure 4c also shows the voltage standing-
wave ratio (VSWR) measured on a 53 O coaxial line. We
observe that the VSWR nearly remains constant (approx-
imately 1), and equivalently the input impedance of the
helix remains unchanged, over the range of about
0.7oC
l
o1.6.
2.2.5. Broadband Characteristics. Considering all the
characteristics of beam pattern, input impedance, and po-
larization as a function of circumference, we nd that the
performance of the axial-mode helix is satisfactory over
the range of about 0.75oC
l
o1.25 within the restrictions
given on a and N.
Thus the bandwidth, dened by the ratio of upper and
lower frequencies, is almost an octave. The broadband
characteristics of the helix can be explained by the natural
adjustment of the phase velocity. As the helix size C
l
, or
equivalently the frequency, varies over a rather wide
range, the phase velocity adjusts itself automatically
such that the elds from each turn add nearly in phase
in the axial direction.
2.3. Variations and Applications of the Helical Antenna
A slight taper on the end of the helix [9,10] reduces the
axial ratio at the expense of a slight reduction in gain.
Axial ratio is improved both on and off axis. A taper is also
used at the input to improve impedance characteristics. A
circular cavity backing is sometimes used to reduce the
backradiation and increase the forward gain. Dielectric-
tube support has been used with the helix antenna. This
lowers the frequency for the onset of axial-mode operation
and has an effect on the terminal impedance. A solid di-
electric core has also been used with the helix (the polyrod
helix). A helix with an inner concentric metal core has
been used as a TV transmitter [11]. The antenna utilizes
higher-order Fourier modes such as e
7j2f
, e
7j5f
, which
radiate sidere rather than endre. This is particularly
useful with towers and masts whose circumference is
much larger than a wavelength. An array of helices is
stacked along the mast to produce the required beam pat-
tern. The helical antenna has often been used as an ele-
ment in various types of arrays. Large planar arrays of
helices have been used in radio astronomy [12]. An array
of axial-mode helices has been used for Global Positioning
System (GPS) satellite transmitters [1]. Helices are also
used as feeds for parabolic dishes. Applications of the helix
are legion.
3. MULTIFILAR HELIX ANTENNAS
The helical antenna described in the preceding section
may be termed the monolar helical antenna. It is con-
structed from a single wire, or tape, and fed from a single
source. In this section we consider the multilar helix an-
tenna, which consists of a number of wires or tapes, each
of which may be fed froma separate source. The wires may
be interleaved as shown in Fig. 5a for the quadrilar helix.
The excitations in all cases discussed is of the form e
7jf
.
Other excitations are certainly possible but have not been
thoroughly studied. Many different forms of the multilar
helix have been used, including bilar, quadrilar, and
octolar helices. Two distinct classes of multilar helices
have been used: the broadband forward-re axial mode
multilar helix and the narrowband backre multilar
helix. For both, the quadrilar helix has been widely stud-
ied and used.
We have already discussed the Fourier modes due to a
loop of circumferential current. The rigorous consider-
ation of the entire geometry of the helix yields modes
with the same f dependence but yields the phase progres-
1932 HELICAL ANTENNAS
sion in z as well. Samuel Sensiper [13] carried out this
rigorous analysis to determine the real propagation con-
stants of the normal and axial modes. He also determined
some of the characteristics of the multilar helix. Later,
Paul Klock [14] found an additional mode with a complex
propagation constant. This mode starts with backre and
then splits and scans forward as a conical beam as fre-
quency increases. It operates simultaneously with the ax-
ial mode but is usually suppressed by the ground plane.
Early experimenters of the multilar helix showed some
improvements over the monolar helix but did not always
recognize that greater bandwidth could be obtained nor
how to obtain it, as explained in the following section.
3.1. The Axial-Mode Quadrilar Helix
Gerst and Worden [15] rst described the broadband ax-
ial-mode multilar helix. They pointed out that the fre-
quency range of the axial mode could be extended by (1)
adding more wires, (2) using e
jf
excitation to maintain
that mode and suppress others, and (3) increasing the
pitch angle a. The technique is readily understood by con-
sidering the bilar helix. First, the two wires are fed 1801
out of phase. Consider a cross section perpendicular to the
helix axis. We nd, in any cross section, two wires 1801
apart in space and phase. Only the odd Fourier modes
(e
jnf
, n odd) are excited. The mode e
j2f
, which may be a
culprit in the pattern breakup of the monolar helix, is
suppressed. The mode e
j3f
is not suppressed, and the
bandwidth for the axial mode approaches 31. A pitch an-
gle of about 251301 is required. Now consider the quad-
rilar helix. A cross section through the right-hand helix
of Fig. 5a displays four wires symmetrically arranged
around the periphery, with phases 01, 901, 1801,
2701. This e
jf
excitation suppresses all the even modes,
as in the bilar helix, and mode e
j3f
as well. Mode e
j5f
is
not suppressed. The bandwidth approaches 51. A pitch an-
gle of about 401 is required. In general, with the M-lar he-
lix, mode e
jf
is excited around the periphery. All other
modes up to e
jMf
are suppressed. The bandwidth lies be-
tween M and M11.
Gerst and Worden determined that the frequency
range of the multilar axial mode helix may be approxi-
mated as follows [15]:
C
l min
oC
l
oC
l max
(37a)
C
l min
=
cos a
1 sin a
(37b)
C
l max
is the lesser of C
l max 1
and C
l max 2
:
C
l max 1
=
cos a
1 sin a
(37c)
C
l max 2
=
M
2
cot a (37d)
where Mis the number of wires. Equation (37) can be used
to predict the bandwidth of the axial-mode unilar or
multilar helix. For example, consider the monolar helix
with a =141, whose beam patterns and other characteris-
tics are given in Figs. 3 and 4. The gures indicate that the
bandwidth is approximately 0.75oC
l
o1.25 as noted pre-
viously. In comparison, Eq. (37) predicts that C
l min
=0.78
and C
lmax
=1.25 for an approximate bandwidth of
0.78oC
l
o1.28. The two bandwidth ratios are very close.
Equation (37) has been applied to bilar, quadrilar, and
octolar axial-beam helices, and the results agree well
with experiment as shown in Refs. 1517.
Figure 6 shows the beam patterns of a quadrilar helix
antenna with a ground plane (see Fig. 5a). Pitch angle a is
351, diameter D is 3in., antenna length is 24 in., ground-
plane diameter D
G
is 10 in., and tape width is
1
2
in. The feed
system [16] provides four outputs phased 01, 901, 1801,
2701, with each output connected to one of the four wires of
the quadrilar helix antenna. Equations (37b) and (37c)
yield C
l min
=0.52 and C
l max
=1.92 for a is 351 and M=4.
Thus the bandwidth of the antenna is given by
0.52oC
l
o1.92 for a bandwidth ratio of 3.71. The pro-
gression of the beam patterns may be described as follows.
At a frequency somewhat below the lower limit, backre
operation begins, as evidenced by the strong backlobe at
C
l
=0.44. The backlobe decreases rapidly as we approach
the lower limit. The axial ratio also decreases rapidly and
is less than 21 at the lower limit. Other antenna charac-
teristics such as VSWR are also acceptable [17]. The axial
mode then predominates over the 3.51 bandwidth. The
beam pattern narrows steadily, and the directivity in-
creases with frequency. The upper limit occurs at C
l
=
1.92, at a frequency just above that of Fig. 6f. Above the
upper limit, beamsplitting occurs at C
l
=2.10, with a com-
plete pattern breakup at C
l
=2.70. Thus, with the multilar
Feed region
(Phases 0, 90, 180, 270)
(b)
Ends
shorted
(a)
Ground
plane
Tape 4
Tape 3
Tape 2
Tape 1
Ends
open
D
G
D
W

Figure 5. Quadrilar helical antenna: (a) endre; (b) backre.


HELICAL ANTENNAS 1933
helix, the bandwidth of the axial mode is extended to both
lower and higher frequencies as predicted in Eq. (37).
The beam pattern at C
l
=1.80 exhibits the undesirable
characteristic of high sidelobes. This is caused by the
backre operation, which changes from backre through
sidere toward endre as frequency increases. It is some-
times called a scanning mode. The quadrilar helix may
also be used in a counterwound version with both right-
and left-hand windings. The on-axis polarization is linear
rather than circular. The backre mode is much more ef-
fectively suppressed in this version, and the sidelobe lev-
els are much lower. Bandwidths are between 41 and 51
as shown in Eq. (16). Gerst and Worden [15] describe a 531
pitch angle counterwound octolar helix with 91 band-
width. The multilar helix antenna does not radiate in a
normal mode at low frequencies because of the phase ex-
citation 01, 901, 1801, 2701 of the windings. The excitation
e
jf
is a supergain excitation at low frequencies (C
l
51).
Details on the axial multilar antenna are given in Refs.
1517 and related references.
3.2. The Backre Quadrilar Helix
In the analysis of the circular loops, we noted that, with
e
7jf
, circularly polarized radiation occurs at both y =01
and y =1801. To distinguish between these two directions,
we need additional information about the helix. A rigorous
analysis of the innite monolar helix by Paul Klock [14]
shows that there are two modes operating simultaneously
in the axial-mode region. Both involve e
jf
excitation for
right-hand helices and are circularly polarized on axis.
One is the axial mode, and the other is a backre mode,
which starts at backre and scans forward as a conical
beam as frequency increases. We may, for purposes of dis-
cussion, combine the backre and forward-scanned oper-
ations into a single backre designation. In the mono- and
quadrilar helices the backre mode shows up just before
the onset of the axial mode. The backre mode exists along
with the axial mode, but is suppressed by the presence of
the ground plane. The backre mode is favored over the
forward endre as pitch angle increases. Pitch angles in
the 40501 range with no ground plane show both forward
and backward radiation. With no ground plane, the radi-
ation is primarily backre. For the right-hand helices of
Figs. 1 and 5, the endre radiation is right-handed circu-
larly polarized and the backre radiation is left-handed
circularly polarized.
The backre biliar helix was rst studied by Patton
[18], who carried out extensive theoretical and experimental
work. He showed the range of backre beam patterns ob-
tained with the monolar and bilar helices. Later, Charles
Kilgus [19] showed that beam pattern improvements could
be obtained with the quadrilar helix. He investigated in
detail the shaped conical and cardioid patterns that are ob-
tainable with the backre quadrilar helix.
Figure 5b shows the backre quadrilar helix. It con-
sists of two bilar antennas fed 901 out of phase to produce
the e
jf
excitation. No ground plane is required, and the
ends of the helices may be shorted (short-circuited) to-
gether. The backre helix is used with pitch angles as high
as 601 and 701. With these high pitch angles, the backre
mode can operate at low frequencies, yielding a small
cross section for the antenna.
A typically shaped conical beam is shown in Fig. 7. The
beam is very wide for broad sector coverage with a dip or a
local maximum (not shown) at the center (y =1801). The
dip is appropriate for satellite coverage because it can be
chosen to yield uniform signal strength at the receiver as
the satellite passes. Kilgus [19] shows numerous beam
patterns for different designs. The beamwidths vary from
1001 to 1801. Directivities of up to 7 dB are observed.
The backre quadrilar helix has characteristics that
make it especially suitable for many satellite, spacecraft,
and navigational applications. It has been used as a trans-
mitter and receiver in satellite communication systems and
as a receiver for GPS applications [20]. It has also been con-
sidered for cellular phones and new GPS applications [21].
3.3. Smart Antennas
The quadrilar helix antenna can be used as a smart or
adaptive antenna. Complex weights are added to its four
separate channels. The weights are then adjusted to im-
270
180
90
0
270
180
90
0
(a) (b)
270
180
90
0
270
180
90
0
(f)
(h) (g)
270
180
90
0
270
270
180
180
90
90
0
0
270
180
90
0
(c) (d)
(e)
Figure 6. Beam patterns of the axial mode quadilar helix an-
tenna: C
l
=0.44, 0.52, 0.72, 1.1, 1.6, 1.8, 2.1, 2.7 in (a)(h), re-
spectively. (From Adams and Lumjiak [17], r 1971 IEEE.
Reprinted with permission of IEEE.)
1934 HELICAL ANTENNAS
prove selected antenna characteristics [22,23]. The effects
of shadowing and multipath, for example, may in some
cases be mitigated.
4. HUMAN PROXIMITY EFFECTS
The presence of a human body in close proximity to an
antenna may, of course, have significant effects. The hu-
man body may absorb radiation. The antenna character-
istics may be modied. In particular, handheld antennas
with the head and hand very close to the antenna may
require a very careful analysis of the situation.
The normal-mode helix antenna has been used for cel-
lular telephone handsets. The antenna is usually electrical-
ly small and is linearly polarized or nearly so. The head can
pick up some radiation but significantly degrades the re-
ception [24]. The axial-mode helix antenna has been pro-
posed for personal satellite communication. This antenna is
circularly polarized. The head can pick up some radiation
and significantly degrades the reception [25].
5. SUMMARY
The helical antenna, rst discovered in 1946 by John
Kraus, has evolved into many different forms with many
different applications. The normal-mode helix has some
advantages for low-frequency applications. The broad-
band, circularly polarized, axial-mode helix radiates for-
ward endre along its axis. It has been the most widely
used of all forms of the helix. The quadrilar axial-mode
helix extends the bandwidth further and can be used as a
smart antenna. The backre quadrilar helix radiates a
broad sector coverage suitable for satellite applications.
The presence of a human body in close proximity to hand-
held antennas may have significant effects on reception.
BIBLIOGRAPHY
1. J. D. Kraus, Antennas, 2nd ed., McGraw-Hill, New York,
1988, pp. 265339.
2. W. L. Stutzman and G. A. Thiele, Antenna Theory and Design,
2nd ed., Wiley, New York, 1988, pp. 231239.
3. W. W. Hansen and J. R. Woodyard, A new principle in direc-
tional antenna design, IRE Proc. 26:333345 (1938).
4. C. A. Balanis, Antenna Theory: Analysis and Design, 2nd ed.,
Wiley, New York, 1997, pp. 271276.
5. H. Nakano, Helical and Spiral Antennas: A Numerical Ap-
proach, Wiley, New York, 1987, pp. 123195.
6. R. S. Elliott, Antenna Theory and Design, Prentice-Hall, New
York, 1981, pp. 7178.
7. H. E. King and J. L. Wong, Characteristics of 1 to 8 wave-
length uniform helical antennas, IEEE Trans. Anten. Propag.
AP-28:291296 (1980).
8. E. A. Wolff, Antenna Analysis, Wiley, New York, 1967, pp.
437444.
9. R. C. Johnson and H. Jasik, Antenna Engineering Handbook,
2nd ed., McGraw-Hill, New York, 1984, pp. 13-113-23.
10. J. L. Wong and H. E. King, Broadband quasi-taper helical
antennas, IEEE Trans. Anten. Propag. AP-27:7278 (1979).
11. L. O. Krause, Sidere helix UHF-TV transmitting antenna,
Electronics 24:107119 (Aug. 1951).
12. J. D. Kraus, Radio Astronomy, 2nd ed., Cygnus-Quasar, Pow-
ell, OH, 1986.
13. S. Sensiper, Electromagnetic wave propagation on helical
structures, Proc. IRE 43: 149161 (1955); also Ph.D. thesis,
MIT, 1951.
14. P. W. Klock, A Study of Wave Propagation of Helices, Ph.D.
thesis, Univ. Illinois, UrbanaChampaign, 1963.
15. C. Gerst and R. A. Worden, Helix antennas take turn for bet-
ter, Electronics 39:100110 (Aug. 22, 1966).
16. A. T. Adams, R. K. Greenough, R. F. Wallenberg, A. Mendelo-
vicz, and C. Lumjiak, The quadrilar helix antenna, IEEE
Trans. Anten. Propag. AP-22:173178 (1974).
17. A. T. Adams and C. Lumjiak, Optimization of the quadrilar he-
lix antenna, IEEE Trans. Anten. Propag. AP-19:547548 (1971).
18. W. T. Patton, The Backre Helical Antenna, Ph.D. thesis,
Univ. Illinois, UrbanaChampaign, 1963.
19. C. C. Kilgus, Shaped-conical radiation pattern performance of
the backre quadrilar helix, IEEE Trans. Anten. Propag.
AP-23:392397 (1975).
20. J. M. Tranquilla, S. R. Best, A study of the quadrilar helix
antenna for global positioning system (GPS) applications,
IEEE Trans. Anten. Propag. AP-38:15451550 (1990).
21. N. Padros, J. Ortigosa, J. Baker, M. F. Iskander, and B. Thorn-
berg, Comparative study of high-performance GPS receiving
antenna designs, IEEE Trans. Anten. Propag. AP-45:698706
(1997).
22. M. Cooper, Antennas get smart, Sci. Am. 289:4855 (2003).
23. S. M. Leach, A. A. Agius, and S. R. Saunders, Intelligent
quadrilar helix antenna, IEE Proc. Microwave Antennas
Propag. 147:219223 (2000).
24. G. Lazzi and O. P. Gandhi, On modeling and personal dos-
imetry of cellular telephone helical antennas with the FDTD
code, IEEE Trans. Anten. Propag. AP-46:525530 (1998).
25. J. S. Colburn and Y. Rahmat-Samii, Human proximity effects
on circular polarized handset antennas in personal satellite
communications, IEEE Trans. Anten. Propag. AP-46: 813
820 (1998).
= 0
90
180
Figure 7. Beam pattern of the backre quadrilar helix.
HELICAL ANTENNAS 1935
Next Page

Вам также может понравиться