Vous êtes sur la page 1sur 4

Bode Gain-Phase Relationship (Bode, 1945)

Equation (2.11) in Skogestads book is an approximation at local frequency


o by using Bodes gain-phase relationship, which is as follows and shown
in H. W. Bodes book Network Analysis and Feedback Design of Amplifiers,
Chapter 14.
Bodes gain-phase relation (for a minimum phase system),
1
G(j0 ) =



d(ln |G(j)|) + 0 d
ln
,
d(ln )
0
{z
}
|

(1)

N ()


d(ln |G(j)|)
is the local slope. If we know the local
where N (0 ) =
d(ln ) =o
slope at gain crossover frequency is constant
40dB
per decade),


Z + (e.g.


+ 0 d
2
ln
and further make use of the fact1 that
=
, then
0
2

Equation (1) can be approximated by


G(j0 )

N (0 ) [in radians].
2

Proof of Equation (1) as explained in H. W. Bodes book.


Theorem 1. Bodes gain-phase relationship (1945). Equation (1)
holds for minimum phase systems.
Proof. Starting from Chapter 8, Bode used the notation for a complex
valued function = A + iB where A is the real component of and an
1
Bode pointed this fact to Bierenes de Haan, Nouvelles Tables DInt
R egrales Definies,
Table 256, p. 377, equations 10) and 11). See also How to compute 0 ln(1 + ex ) dx?
on math.stackexchange.com.

even function; B is the imaginary part and an odd function due to the fact
that one can write () = |()| ei() . By Euler formula,
() = |()| ei()
= |()| cos (()) + i |()| sin (()) .
{z
}
|
{z
}
|
A, even function

B, odd function

Then Bode investigated the value of B at = c . In order to calculate


the value, first divide () by c . Also introduce the complementary
point c to preserve symmetry (why?). For convenience, only () Ac
is considered in the integrand, where Ac is the value assumed by A at
frequency c . So Bode constructed a contour integral as below,

Z 
() Ac () Ac

d = 0,
(2)
c
+ c

where is the usual Nyquist contour consisting of imaginary axis (with


small semicircle indents around c ) and indefinitely large semicircle
which together surround the open right half plane (ORHP). The integral of the left hand side of Equation (2) is zero because the integrand is
analytic within the region formed by contour .
At high frequencies, the integrand goes to zero faster than 1/2 so at infinity
the integral along the indefinitely large semicircle is zero (why?). The
remaining part of the integral is taken along the imaginary axis except that
at two frequencies c two semicircle indents are used. So the contour
integral in Equation (2) could be further broken up into three terms,


Z +
Z
1
1
2c
(A Ac + iB) d
iBc

d
2
2
c + c
1
c


Z
1
1
+
iBc

d = 0,
c + c
2
(3)
where the first integral is taken along the imaginary axis and the latter
two along two indentations 1 (centered at = c ) and 2 (centered
at = c ). Since B is an odd function, B term in the first integral in
Equation (3) can be ignored. Also we can restrict the integral limits to
the positive side [0, +] and double this integral to get integration over
[, +] since A is an even function.
2

In the second integral, when we calculate integral along the indentation


1
term can be ignored because it is analytic around the neigh1 ,
c
bourhood of = c . By Residue Theorem ,



Z 
1
1
d = 2i W (1 , c )Res
, c
+ c
+ c
1
1
= 2i 1 = i,
2
where W (1 , c ) is the winding number of 1 around pole = c and
Res(f (z), zk ) is the residue of f (z) at zk . Similarly we can calculate the
third integral.
So finally the value of B at frequency c is,
Z
2c + A Ac
d.
Bc =
0
2 c2

(4)

Equation (4) has an alternative form if logarithmic scale is chosen. Suppose


we define u = ln(/c ), we rewrite Bc as,
Z
2 + A Ac d
Bc =
/c c/
0
Z +
2
A Ac
=
du
eu eu
Z
1 + A Ac
du.
=
sinhu
Bode showed with heavy algebra by integrating by parts, the above equation can be also expressed as,


Z
1 + dA
|u|
Bc =
ln coth
du.
(5)
du
2




+ c
|u|
. If we compare Equation (5) with
Notice that ln coth
= ln
2
c
Equation (1), they are essentially the same except that A and B are real
and imaginary components of a complex valued function (). To see they
3

are equivalent, lets do a simple transformation on G(j). Since G(j) is


a complex number, we rewrite it according to Euler formula,
G(j) = |G(j)| ejG(j)
ln (G(j)) = ln |G(j)| +j G(j) .
| {z } | {z }
| {z }
A

()

Hence it completes the proof.

typeset with AMS-LATEX

Vous aimerez peut-être aussi