Professional Documents
Culture Documents
1.
dB
= 20 log 10 G
(1)
= 20 log 10
dB
Magnitude in dB
1/2
6dB
0 dB
6 dB
5 = 10/2
20 dB 6 dB = 14 dB
10
1000 = 10
20dB
3
3 20dB = 60 dB
Z
Rbase
(2)
The value of Rbase is arbitrary, but we need to tell others what value we have used. So if
|| Z || is 5, and we choose Rbase = 10, then we can say that || Z ||dB = 20 log10(5/10)
= 6dB with respect to 10. A common choice is Rbase = 1; decibel impedances
expressed with Rbase = 1 are said to be expressed in dB. So 5 is equivalent to 14dB.
Current switching harmonics at the input port of a converter are often expressed in dBA,
or dB using a base current of 1A: 60dBA is equivalent to 1000A, or 1mA.
The magnitude Bode plots of functions equal to powers of f are linear. For
example, suppose that the magnitude of a dimensionless quantity G(f) is
G =
f
f0
(3)
= 20 log 10
dB
f
f0
= 20n log 10
f
f0
(4)
60dB
f
f0
40dB/decade
40dB
20dB
f
f0
20dB/decade
1
at frequency f = f 0 . They are
n=
0dB
linear functions of log10(f). The
n=
1
20 dB/decade
f 1
slope is the change in || G ||d B 20dB
f0
n
=
arising from a unit change in 40dB
2
f 2
40dB/decade
log10(f); a unit increase in
f0
log10(f) corresponds to a factor 60dB
0.1f0
f0
10f0
f
of 10, or decade, increase in f.
log scale
Fig. 1.
Magnitude Bode plots of functions which vary
From Eq. (4), a decade increase
n
as f are linear, with slope n dB per decade.
in f leads to an increase in
|| G ||dB of 20n dB. Hence, the
slope is 20n dB per decade. Equivalently, we can say that the slope is 20n log10(2) 6n dB
1
v2(s)
sC
G(s) =
=
1 +R
v1(s)
sC
R
+
v1(s)
v2(s)
Fig. 2.
Simple R-C lowpass filter example.
(5)
This transfer function is a ratio of voltages, and hence is dimensionless. By multiplying the
numerator and denominator by sC, we can express the transfer function as a rational
fraction:
G(s) =
1
1 + sRC
(6)
The transfer function now coincides with the following standard normalized form for a
single pole:
G(s) =
1
1 + s
(7)
0 = 1
RC
(8)
Since R and C are real positive quantities, 0 is also real and positive. The denominator of
Eq. (7) contains a root at s = 0, and hence G(s) contains a real Im(G(j))
G
(j
)|
G(j)
||
G(j)
Re(G(j))
magnitude and phase of the resulting complex-valued function.
With s = j, Eq. (7) becomes
Fig. 3.
Magnitude and
1 j
0
1
phase of the
G( j) =
=
2
1+
1+ j
complex-valued
0
0
(9)
function G(j).
The complex-valued G(j) is illustrated in Fig. 3, for one value of . The magnitude is
G( j) =
=
(10)
G( j)
dB
= 20 log 10
1+
0
dB
(11)
The easy way to sketch the magnitude Bode plot of G is to investigate the asymptotic
behavior for large and small frequency.
For small frequency, << 0 and f << f0, it is true that
0 << 1
(12)
The (/0) term of Eq. (10) is therefore much smaller than 1, and hence Eq. (10) becomes
2
G( j) 1 = 1
1
(13)
G( j)
dB
0dB
(14)
|| G(j) ||dB
0dB
0dB
20dB
20dB/decade
f
f0
40dB
60dB
0.1f0
Fig. 4.
10f0
f0
0 >> 1
(15)
1+
0
(16)
G( j)
f
=
f0
(17)
This expression coincides with Eq. (3), with n = 1. So at high frequency, || G(j) ||dB has
slope 20dB per decade, as illustrated in Fig. 4. Thus, the asymptotes of || G(j) || are
equal to 1 at low frequency, and (f / f0)1 at high frequency. The asymptotes intersect at f 0 .
The actual magnitude tends towards these asymptotes at very low frequency and very high
frequency. In the vicinity of the corner frequency f 0 , the actual curve deviates somewhat
from the asymptotes.
The deviation of the exact curve from the asymptotes can be found by simply
evaluating Eq. (10). At the corner frequency f = f0, Eq. (10) becomes
G( j0) =
1 + 0
0
= 1
2
(18)
G( j0)
= 20 log 10
dB
1 + 0
0
3 dB
(19)
So the actual curve deviates from the asymptotes by 3dB at the corner frequency, as
illustrated in Fig. 5. Similar arguments show that the actual curve deviates from the
asymptotes by 1dB at f = f0/2 and at f = 2f0.
|| G(j) ||dB
G(j)
0 asymptote
0
-15
0dB
3dB
1dB
0.5f0
10dB
-30
1dB
f0
-45
2f0
20dB
-45
f0
-60
20dB/decade
-75
90 asymptote
30dB
-90
0.01f0
Fig. 5.
0.1f0
f0
10f0
Fig. 6.
G( j) = tan 1
Im G( j)
Re G( j)
(20)
Insertion of the real and imaginary parts of Eq. (9) into Eq. (20) leads to
G( j) = tan 1
(21)
This function is plotted in Fig. 6. It tends to 0 at low frequency, and to 90 at high
frequency. At the corner frequency f = f0, the phase is 45.
Since the high-frequency and low-frequency phase asymptotes do not intersect, we
need a third asymptote to approximate the phase in the vicinity of the corner frequency f 0 .
One way to do this is illustrated in Fig. 7, where the slope of the asymptote is chosen to be
identical to the slope of the actual curve at f = f0. It can be shown that, with this choice, the
asymptote intersection frequencies fa and fb are given by
fa = f0 e / 2 f0 / 4.81
fb = f0 e / 2 4.81 f0
(22)
fa = f0 / 10
fb = 10 f0
(23)
100f0
fa = f0 / 4.81
G(j)
-15
-30
-45
-45
f0
-60
-75
-90
0.01f0
0.1f0
f0
fb = 4.81 f0
100f0
Fig. 7.
One choice for the midfrequency phase asymptote,
which correctly predicts the actual slope at f = f0.
fa = f0 / 10
G(j)
0
-15
-30
-45
-45
f0
-60
-75
-90
0.01f0
0.1f0
fb = 10 f0
f0
100f0
Fig. 8.
A simpler choice for the midfrequency phase
asymptote, which better approximates the curve over the
entire frequency range.
0dB
|| G(j) ||dB
3dB
1dB
0.5f0
1dB
f0
2f0
20dB/decade
0
G(j)
f0 / 10
5.7
-45/decade
-45
f0
-90
5.7
10 f0
Fig. 9.
Summary of the magnitude and phase Bode
plot for the single real pole.
(s/0)-1. This leads to a magnitude of (f/f0)-1. The corner frequency is f0 = 0/2. So the
transfer function is written directly in terms of its salient features, i.e., its asymptotes and
its corner frequency.
1.2.
G(s) = 1 + s
0
(24)
G( j) =
1+
0
(25)
At low frequency, f << f0 = 0/2, the transfer function magnitude tends to 1 0dB. At
high frequency, f >> f 0 , the transfer function magnitude tends to (f/f0). As illustrated in
Fig. 10, the high frequency asymptote
+20dB/decade
has slope +20dB/decade.
The phase is given by
2f0
G( j) = tan 1
(26)
With the exception of a minus sign,
the phase is identical to Eq. (21).
Hence, suitable asymptotes are as
illustrated in Fig. 10. The phase tends
to 0 at low frequency, and to +90 at
high frequency. Over the interval
f0/10 < f < 10f0, the phase asymptote
has a slope of +45/decade.
f0
1dB
0.5f0
0dB
|| G(j) ||dB
1dB
3dB
10 f0
+90
5.7
f0
45
+45/decade
G(j)
5.7
f0 / 10
Fig. 10.
1.3.
Frequency inversion
Two other forms arise, from
inversion of the frequency axis. The
inverted pole has the transfer function
G(s) =
1 + s0
(27)
As illustrated in Fig. 11, the inverted
pole has a high-frequency gain of 1,
and a low frequency asymptote having
a +20dB/decade slope. This form is
useful for describing the gain of highpass filters, and of other transfer
functions where it is desired to
emphasize the high frequency gain,
with attenuation of low frequencies.
Equation (27) is equivalent to
s
0
G(s) =
1 + s
0
0dB
3dB
1dB
1dB
f0
2f0
0.5f0
|| G(j) ||dB
+20dB/decade
G(j)
+90
f0 / 10
5.7
-45/decade
+45
f0
0
5.7
10 f0
Fig. 11.
Inversion of the frequency axis: summary
of the magnitude and phase Bode plot for the
inverted real pole.
(28)
However, Eq. (27) more directly emphasizes that the high frequency gain is 1.
The inverted zero has the form
G(s) = 1 + s0
(29)
As illustrated in Fig. 12, the inverted
zero has a high-frequency gain
asymptote equal to 1, and a lowfrequency asymptote having a slope
equal to 20dB/decade. An example of
the use of this type of transfer function
is
the
proportional-plus-integral
controller, discussed in connection
with feedback loop design in the next
chapter. Equation (29) is equivalent to
20dB/decade
|| G(j) ||dB
0.5f0
f0
1dB
2f0
3dB
1dB
0dB
10 f0
5.7
f0
45
+45/decade
G(j)
90
5.7
f0 / 10
Fig. 12.
1 + s
0
G(s) =
s
0
(30)
However, Eq. (29) is the preferred form when it is desired to emphasize the value of the
high-frequency gain asymptote.
The use of frequency inversion is illustrated by example in the next section.
1.4.
Combinations
The Bode diagram of a transfer function containing several pole, zero, and gain
terms, can be constructed by simple addition. At any given frequency, the magnitude (in
decibels) of the composite transfer function is equal to the sum of the decibel magnitudes of
the individual terms. Likewise, at a given frequency the phase of the composite transfer
function is equal to the sum of the phases of the individual terms.
For example, suppose that we have already constructed the Bode diagrams of two
complex-valued functions of , G1() and G2(). These functions have magnitudes R1()
and R2(), and phases 1() and 2(), respectively. It is desired to construct the Bode
diagram of the product G3() = G1() G2(). Let G3() have magnitude R3(), and phase
3(). To find this magnitude and phase, we can express G1(), G2(), and G3() in
polar form:
(31)
(32)
Simplification leads to
(33)
(34)
(35)
R3()
dB
= R1()
dB
+ R2()
10
dB
(36)
So the composite phase is the sum of the individual phases, and when expressed in
decibels, the composite magnitude is the sum of the individual magnitudes. The composite
magnitude slope, in dB per decade, is therefore also the sum of the individual slopes in dB
per decade.
For example, consider construction of the Bode plot of the following transfer
function:
G(s) =
G0
1 + s
1
1 + s
2
(37)
where G0 = 40 32dB, f1 = 1/2 = 100Hz, f2 = 2/2 = 2kHz. This transfer function
contains three terms: the gain G0, and the poles at frequencies f1 and f2. The asymptotes for
each of these terms are illustrated in Fig. 13. The gain G0 is a positive real number, and
therefore contributes zero phase shift with the gain 32dB. The poles at 100Hz and 2kHz
each contribute asymptotes as in Fig. 9.
At frequencies less than 100Hz, the G0 term contributes a gain magnitude of 32dB,
while the two poles each contribute magnitude asymptotes of 0dB. So the low-frequency
composite magnitude asymptote is 32dB + 0dB + 0dB = 32dB. For frequencies between
100Hz and 2kHz, the G0 gain again contributes 32dB, and the pole at 2kHz continues to
contribute a 0dB magnitude asymptote. However, the pole at 100Hz now contributes a
magnitude asymptote that decreases with a 20dB/decade slope. The composite magnitude
asymptote therefore also decreases with a 20dB/decade slope, as illustrated in Fig. 13.
For frequencies greater than 2kHz, the poles at 100Hz and 2kHz each contribute decreasing
40dB
|| G ||
20dB
G0 = 40 32dB
|| G ||
20 dB/dec
0dB
0dB
20dB
f1
100Hz
40dB
f2
2kHz
0
f1 / 10
10Hz
90/dec
10 f1
1kHz
10Hz
100Hz
1kHz
90
10 f2
20kHz
135
45/dec
10kHz
180
100kHz
f
Fig. 13.
Construction of magnitude and phase asymptotes for the transfer
function of Eq. (37). Dashed lines: asymptotes for individual terms. Solid
lines: composite asymptotes.
11
0
45
45/dec
60dB
1Hz
40 dB/dec
f2 / 10
200Hz
A(s) = A 0
1 + s
1
1 + s
2
(38)
A0
s
1+
= A0 1 = A0
1
s
1+
2
(39)
s = j
12
For f1 < f < f 2 , the numerator term (s/1) has magnitude greater than 1, while the
denominator term (s/2) has magnitude less than 1. The asymptote is derived by neglecting
the smaller terms:
A0
1 + s
s
1+
= A0
s
1
s = j
=A f
= A0
0
f1
1
(40)
This is the expression for the midfrequency magnitude asymptote of A(s). For f > f 2 , the
(s/1) and (s/2) terms each have magnitude greater than 1. The expression for the highs = j
A0
1 + s
1 + s
= A0
s
1
s
2
s = j
s = j
f
= A 0 2 = A 0 2
f1
1
s = j
(41)
A = A0
f2
f1
(42)
A(s) = A
1 + s1
1 + s2
(43)
G(s) =
v2(s)
1
=
v1(s) 1 + s L + s 2LC
R
13
L
+
v1(s)
v2(s)
(44)
Fig. 15.
Two-pole low-pass filter
example.
This transfer function contains a second-order denominator polynomial, and is of the form
G(s) =
1
1 + a 1s + a 2s 2
(45)
G(s) =
1
1 ss
1
1 ss
2
(46)
Use of the quadratic formula leads to the following expressions for the roots:
s1 =
s2 =
a1
1
2a 2
a1
1+
2a 2
4a 2
a 21
(47)
4a 2
a 21
(48)
If 4a2
then the roots are real. Each real pole then exhibits a Bode diagram as derived
in section 1.1, and the composite Bode diagram can be constructed as described in section
1.4 (but a better approach is described in section 1.6).
If 4a2 > a12, then the roots (47) and (48) are complex. In section 1.1, the
assumption was made that 0 is real; hence, the results of that section cannot be applied to
a12,
this case. We need to do some additional work, to determine the magnitude and phase for
the case when the roots are complex.
The transfer functions of Eqs. (44) and (45) can be written in the following
standard normalized form:
G(s) =
1
s
1 + 2 + s
0
0
(49)
If the coefficients a1 and a2 are real and positive, then the parameters and 0 are also real
and positive. The parameter 0 is again the angular corner frequency, and we can define f0
= 0 / 2. The parameter is called the damping factor: controls the shape of the transfer
function in the vicinity of f = f0. An alternative standard normalized form is
G(s) =
1
s
1+
+ s
Q0
0
(50)
where
Q= 1
2
(51)
14
The parameter Q is called the quality factor of the circuit, and is a measure of the dissipation
in the system. A more general definition of Q, for sinusoidal excitation of a passive element
or network, is
(peak stored energy)
Q = 2
(energy dissipated per cycle)
(52)
For a second-order passive system, Eqs. (51) and (52) are equivalent. We will see that the
Q-factor has a very simple interpretation in the magnitude Bode diagrams of second-order
transfer functions.
Analytical expressions for the parameters Q and 0 can be found by equating like
powers of s in the original transfer function, Eq. (44), and in the normalized form, Eq.
(50). The result is
1
f0 = 0 =
2 2 LC
Q=R C
L
(53)
The roots s1 and s2 of Eqs. (47) and (48) are real when Q 0.5, and are complex when Q
> 0.5.
The magnitude of G is
G( j) =
1
0
2 2
+ 12
0
Q
(54)
Asymptotes of ||G || are illustrated in Fig. 16. At low frequencies, (/0) << 1, and hence
G 1 for << 0
(55)
At high frequencies, (/0) >> 1, the (/0) term dominates the expression inside the
4
f
f0
for >> 0
(56)
This expression coincides with Eq. (3), with n = 2. Therefore, the high-frequency
asymptote has slope 40dB/decade. || G(j) ||dB
0dB
The asymptotes intersect at f = f 0 ,
0dB
and are independent of Q.
f 2
20dB
f
0
The parameter Q affects the
40dB
deviation of the actual curve from
40dB/decade
60dB
the asymptotes, in the neighborhood
0.1f0
10f0
f0
f
of the corner frequency f0. The exact
Fig. 16. Magnitude asymptotes for the two-pole
magnitude at f = f0 is found by
transfer function.
15
|| G ||
| Q |dB
0dB
G( j0) = Q
(57)
So the exact transfer function has magnitude Q at the
corner frequency f0. In decibels, Eq. (57) is
G( j0)
= Q
f0
40dB/dec
Fig. 17.
Important features
of the magnitude Bode plot,
for the two-pole transfer
function.
(58)
So if, for example, Q = 2 6dB, then the actual curve
dB
dB
G( j) = tan
1
Q 0
1
0
fa = e
1
/ 2 2Q
fb = e / 2
1
2Q
(59)
0
increasing Q
-90
-180
0.1
10
f / f0
Fig. 18.
Phase plot, second-order poles.
Increasing Q causes a sharper phase change.
0
fa
G
-90
-90
f0
f0
f0
(60)
A better choice, which is consistent with
the approximation (23) used for the real
single pole, is
-180
-180
0.1
fb
f / f0
Fig. 19.
One choice for the midfrequency
phase asymptote of the two-pole
response, which correctly predicts the
actual slope at f = f0.
16
10
fa = 10 1 / 2Q f0
fb = 10 1 / 2Q f0
fa
(61)
G
With this choice, the midfrequency
asymptote has slope 180Q degrees per
-90
-90
f0
decade. The phase asymptotes are
summarized in Fig. 20. With Q = 0.5, the
phase changes from 0 to 180 over a
-180
-180
frequency span of approximately two
f
0.1
1
b
f
/
f
decades, centered at the corner frequency
0
Fig.
20.
A
simpler
choice
for the
f0. Increasing the Q causes this frequency
midfrequency phase asymptote, which
span to decrease rapidly.
better approximates the curve over the
entire frequency range and is consistent
Second-order response magnitude
with the asymptote used for real poles.
curves, Eq. (54), and phase curves, Eq.
(59), are plotted in Figs. 21 and 22 for several values of Q.
17
10
10dB
Q=2
Q=1
Q = 0.7
0dB
Q = 0.5
|| G ||dB
-10dB
Q = 0.2
Q = 0.1
-20dB
0.3
0.5
0.7
f / f0
Fig. 21.
-45
Q = 0.2
Q = 0.1
-90
-135
-180
0.1
10
f / f0
Fig. 22.
18
1.6.
G(s) =
1 + s
1
1 + s
2
(61)
This is a particularly desirable approach when the corner frequencies 1 and 2 are wellseparated in value.
The difficulty in this procedure lies in the complexity of the quadratic formula used
to find the corner frequencies. Expressing the corner frequencies 1 and 2 in terms of the
circuit elements R, L, C, etc., invariably leads to complicated and unilluminating
expressions, especially when the circuit contains many elements. Even in the case of the
simple circuit of Fig. 15, whose transfer function is given by Eq. (44), the conventional
quadratic formula leads to the following complicated formula for the corner frequencies:
1, 2 =
L/R
L / R 4 LC
2 LC
(62)
This equation yields essentially no insight regarding how the corner frequencies depend on
the element values. For example, it can be shown that when the corner frequencies are
well-separated in value, they can be expressed with high accuracy by the much simpler
relations
1 R ,
L
2 1
RC
(63)
In this case, 1 is essentially independent of the value of C, and 2 is essentially
independent of L, yet Eq. (62) apparently predicts that both corner frequencies are
dependent on all element values. The simple expressions of Eq. (63) are far preferable to
Eq. (62), and can be easily derived using the low-Q approximation [2].
Let us assume that the transfer function has been expressed in the standard
normalized form of Eq. (50), reproduced below:
G(s) =
1
s
+ s
1+
Q0
0
(64)
For Q 0.5, let us use the quadratic formula to write the real roots of the denominator
polynomial of Eq. (64) as
19
1
1 = 0
Q
1 4Q 2
2
0 1 +
Q
2 =
1 4Q
2
F(Q)
(65)
0.75
0.5
(66)
can be
0.25
expressed
0
2 = 0 F(Q)
Q
0.2
(67)
F(Q) = 1 1 +
2
0.1
0.3
0.4
0.5
Fig. 23.
F(Q) vs. Q, as given by Eq. (8-72). The
approximation F(Q) 1 is within 10% of the
exact value for Q < 0.3.
1 4Q 2
(68)
Note that, when Q << 0.5, then 4Q << 1, and F(Q) is approximately equal to 1. We then
obtain
2 0 for Q << 1
Q
2
(69)
The function F(Q) is plotted in Fig. 23. It can be seen that F(Q) approaches 1 very rapidly
as Q decreases below 0.5.
To derive a similar approximation for 1, we can multiply and divide Eq. (65) by
2
1 =
Q 0
F(Q)
(70)
1 Q 0
for Q << 1
2
(71)
Magnitude asymptotes for the low-Q || G ||dB
0dB
case are summarized in Fig. 24. For Q < 0.5,
the two poles at 0 split into real poles. One
real pole occurs at corner frequency 1 < 0,
while the other occurs at corner frequency 2 >
0. The corner frequencies are easily
approximated, using Eqs. (69) and (71).
For the filter circuit of Fig. 15, the
parameters Q and 0 are given by Eq. (53). For
Q f0
F(Q)
Q f0
f1 =
f0
f0F(Q)
Q
f0
f2 =
20dB/decade
40dB/decade
Fig. 24.
Magnitude asymptotes predicted
by the low-Q approximation. Real poles
occur at frequencies Qf0 and f0/Q.
the case when Q << 0.5, we can derive the following analytical expressions for the corner
frequencies, using Eqs. (70) and (71):
20
1 Q 0 = R C 1 = R
L LC L
0
1
2
= 1
= 1
Q
RC
LC R C
L
(72)
So the low-Q approximation allows us to derive simple design-oriented analytical
expressions for the corner frequencies.
1.7.
P(s) = 1 + a 1 s + a 2 s 2 +
+ an s n
(73)
P(s) = 1 + 1 s 1 + 2 s
1 + n s
(74)
In a real circuit, the coefficients a1,...,an are real, while the time constants 1 ,..., n may be
either real or complex. Very often, some or all of the time constants are well separated in
value, and depend in a very simple way on the circuit element values. In such cases, simple
approximate analytical expressions for the time constants can be derived.
The time constants 1 ,..., n can be related to the original coefficients a 1,...,a n by
multiplying out Eq. (74). The result is
a1 = 1 + 2 +
a2 = 1 2 +
+ n + 2 3 +
a 3 = 1 2 3 +
a n = 1 2 3
+ n
+ n +
+ n + 2 3 4 +
+ n +
(75)
General solution of this system of equations amounts to exact factoring of the arbitrary
degree polynomial, a hopeless task. Nonetheless, Eq. (75) does suggest a way to
approximate the roots.
Suppose that all of the time constants 1 ,..., n are real and well separated in value.
We can further assume, without loss of generality, that the time constants are arranged in
decreasing order of magnitude:
1 >> 2 >>
>> n
(76)
When the inequalities of Eq. (76) are satisfied, then the expressions for a 1,...,a n of Eq.
(75) are each dominated by their first terms:
21
a1 1
a 2 1 2
a 3 1 2 3
a n = 1 2 3
(77)
These expressions can now be solved for the time constants, with the result
1 a1
a2
a1
a
3 3
a2
2
an
an 1
(78)
Hence, if
a 1 >>
a2
a
>> 3 >>
a1
a2
an
an 1
>>
(79)
then the polynomial P(s) given by Eq. (73) has the approximate factorization
P(s) 1 + a 1 s 1 +
a2
s
a1
1+
a3
s
a2
1+
an
s
an 1
>>
ak + 1
>>
ak
(80)
Note that if the original coefficients in Eq. (73) are simple functions of the circuit elements,
then the approximate roots given by Eq. (80) are similar simple functions of the circuit
elements. So approximate analytical expressions for the roots can be obtained. Numerical
values are substituted into Eq. (79) to justify the approximation.
In the case where two of the roots are not well separated, then one of the
inequalities of Eq. (79) is violated. We can then leave the corresponding terms in quadratic
form. For example, suppose that inequality k is not satisfied:
a 1 >>
a2
>>
a1
>>
ak
ak 1
>>
an
an 1
not
satisfied
22
(81)
P(s) 1 + a 1 s 1 +
a2
s
a1
1+
ak
a
s + k + 1 s2
ak 1
ak 1
1+
an
s
an 1
(82)
a 1 >>
a2
>>
a1
>>
ak
a a
>> k 22 k + 1 >>
ak 1
ak 1
>>
an
an 1
(83)
>>
a1
a2
a
>> 3 >>
a1
a2
>>
an
an 1
not
satisfied
(84)
P(s) 1 + a 1 s + a 2 s 2
1+
a3
s
a2
1+
an
s
an 1
(85)
a 22
a
a
>> a 1 >> 3 >> 4 >>
a3
a2
a3
>>
an
an 1
(86)
If none of the above approximations is justified, then there are three or more roots that are
close in magnitude. One must then resort to cubic or higher-order forms.
As an example, consider the following transfer function:
G0
G(s) =
1+s
L1 + L2
LLC
+ s 2 L 1C + s 3 1 2
R
R
(87)
This transfer function contains a third-order denominator, with the following coefficients:
a1 =
L1 + L2
R
a 2 = L 1C
LLC
a3 = 1 2
R
(88)
It is desired to factor the denominator, to obtain analytical expressions for the poles. The
correct way to do this depends on the numerical values of R, L 1 , L 2, and C. When the
roots are real and well separated, then Eq. (80) predicts that the denominator can be
factored as follows:
23
1+s
L1 + L2
R
1 + sRC
L1
L1 + L2
1+s
L2
R
(89)
(90)
These inequalities cannot be satisfied unless L1 >> L 2. When L1 >> L 2, then Eq. (90) can
be further simplified to
L1
L
>> RC >> 2
R
R
(91)
The approximate factorization, Eq. (89), can then be further simplified to
1+s
L1
R
1 + sRC 1 + s
L2
R
(92)
Thus, in this case the transfer function contains three well separated real poles.
When the second inequality of Eq. (90) is violated,
L1
L1 + L2
>> RC
R
L1 + L2
>>
L2
R
not
satisfied
(93)
then the second and third roots should be left in quadratic form:
1+s
L1 + L2
R
1 + sRC
L1
+ s 2 L 1||L 2 C
L1 + L2
(94)
This expression follows from Eq. (82), with k = 2. Equation (83) predicts that this
approximation is justified provided that
L1
L ||L
L1 + L2
>> RC
>> 1 2 RC
R
L1 + L2
L1 + L2
(95)
In application of Eq. (83), we take a0 to be equal to 1. The inequalities of Eq. (95) can be
simplified to obtain
L1
L 1 >> L 2, and
>> RC
R
(96)
Note that it is no longer required that RC >> L2 / R. Equation (96) implies that factorization
(94) can be further simplified to
1+s
L1
R
1 + sRC + s 2L 2C
(97)
Thus, for this case, the transfer function contains a low-frequency pole that is well
separated from a high-frequency quadratic pole pair.
24
L1 + L2
R
>>
RC
L1
L
>> 2
L1 + L2
R
not
satisfied
(98)
then the first and second roots should be left in quadratic form:
1+s
L1 + L2
+ s 2 L 1C
R
1+s
L2
R
(99)
This expression follows directly from Eq. (85). Equation (86) predicts that this
approximation is justified provided that
L 1RC
L + L2
L
>> 1
>> 2
L2
R
R
(100)
i.e.,
L2
R
(101)
For this case, the transfer function contains a low-frequency quadratic pole pair that is wellseparated from a high-frequency real pole. If none of the above approximations are
justified, then all three of the roots are similar in magnitude. We must then find other means
of dealing with the original cubic polynomial.
2.
di(t)
= Dvg(t) + (1 D)v(t) + Vg V vc(t)
dt
dv(t)
v(t)
C
= (1 D) i(t)
+ Ivc(t)
R
dt
L
(102)
The system contains two independent ac inputs: the control input vc(t) and the input voltage
vg(t). The capitalized quantities are given constants. In the Laplace transform domain, the
ac output voltage v(s) can be expressed as the superposition of terms arising from the two
inputs:
25
(103)
Gvc(s) =
v(s)
vc(s)
and Gvg(s) =
v g(s) = 0
v(s)
vg(s)
v c(s) = 0
(104)
An algebraic approach to deriving these transfer functions begins by taking the Laplace
transform of Eq. (102), letting the initial conditions be zero:
v(s)
+ I vc(s)
R
(105)
To solve for the output voltage v(s), we can use the top equation to eliminate i(s):
i(s) =
(106)
sCv(s) =
(1 D)
v(s)
D vg(s) + (1 D) v(s) + Vg V vc(s)
+ I vc(s)
sL
R
(107)
Vg V s LI
D (1 D)
v
vc(s)
g(s)
(1 D) 2 + s L + s 2 LC
(1 D) 2 + s L + s 2 LC
R
R
(108)
We arent done yet the next step is to manipulate these expressions into normalized form,
such that the coefficients of s0 in the numerator and denominator polynomials are equal to
one:
v(s) =
v(s) =
D
1D 1+s
1
vg(s)
L
+ s 2 LC 2
2
(1 D)
(1 D) R
Vg V
(1 D) 2 1 + s
1s
LI
Vg V
L
+ s 2 LC 2
(1 D) 2 R
(1 D)
vc(s)
(109)
This result is similar in form to Eq. (103). The transfer function from vg(s) to v(s) is
Gvg(s) =
v(s)
vg(s)
=
v c(s) = 0
D
1D 1+s
1
L
+ s 2 LC 2
(1 D)
(1 D) 2 R
Thus, this transfer function contains a dc gain Gg0 and a quadratic pole pair:
26
(110)
1
s
1+
+ s
Q0
0
Gvg(s) = Gg0
(111)
Analytical expressions for the salient features of the transfer function from vg(s) to v(s) are
found by equating like terms in Eqs. (110) and (111). The dc gain is
Gg0 =
D
1D
(112)
By equating the coefficients of s in the denominators of Eqs. (110) and (111), one obtains
2
1 = LC
20 D' 2
(113)
0 = D'
LC
(114)
By equating coefficients of s in the denominators of Eqs. (110) and (111), one obtains
1 = L
Q0 D' 2R
(115)
Q = D'R
C
L
(116)
Equations (114), (112), and (116) are the desired results in the analysis of the voltage
transfer function from vg(s) to v(s). These expressions are useful not only in analysis
situations, where it is desired to find numerical values of the salient features G g0, 0, and
Q, but also in design situations, where it is desired to select numerical values for R, L, and
C such that given values of the salient features are obtained.
Having found analytical expressions for the salient features of the transfer function,
we can now plug in numerical values and construct the Bode plot. Suppose that we are
given the following values:
D = 0.6
R = 10
Vg = 30V
L = 160H
C = 160F
(117)
We can evaluate Eqs. (112), (114), and (116), to determine numerical values of the salient
features of the transfer functions. The results are:
27
Gg0 = 1.5
3.5dB
|| Gvg ||
0dB
|| Gvg ||
f0
20dB
60dB
40dB/dec
400Hz
40dB
Gvg
Q = 4 12dB
10 1 / 2Q 0 f0
300Hz
0
Gvg
80dB
90
180
10 1 / 2Q 0 f0
533Hz
10Hz
100Hz
1kHz
10kHz
180
270
100kHz
f
Fig. 25.
D = 1.5 3.5dB
1D
f0 = 0 = D' = 400Hz
2 2 LC
Q = D'R C = 4 12dB
L
Gg0 =
(118)
The Bode plot of the magnitude and phase of the transfer function Gvg is
constructed in Fig. 25. This transfer function contains a dc gain of 3.5dB and resonant
poles at 400Hz having a Q of 4 12dB. The resonant poles contribute 180 to the high
frequency phase asymptote.
3.
28
3.1.
Z(s) = R + 1
(119)
sC
Let us first sketch the magnitudes of the individual impedances.
The 10 resistor has an impedance magnitude of 10 20dB.
1F
Fig. 26.
Series R-C
network example.
1 = 1
C
(120)
i.e., at
1 =
1
= 10 6 rad/sec
1 C (1) (10 -6F)
(121)
f = = 10 = 159kHz
2 2
So the capacitor impedance 80dB
magnitude is a line with slope
1
60dB
C
20dB/dec, and which passes
20dB/decade
40dB
through 0dB at 159kHz, as
shown in Fig. 27. It should be 20dB R = 10 20dB
noted that, for simplicity, the 0dB
1 = 1 at 159kHz
asymptotes in Fig. 27 have been
C
20dB
100Hz
1kHz
10kHz
100kHz
labeled R and 1/C. But to draw
6
(122)
Fig. 27.
Impedance magnitudes of the individual
elements in the network of Fig. 26.
29
10k
1k
100
10
1
0.1
1MHz
Let us now construct the magnitude of Z(s), given by Eq. (119). The magnitude of
Z can be approximated as follows:
Z( j) = R + 1
jC
R for R >> 1 / C
1 for R << 1 / C
C
(123)
The asymptotes of the series combination are simply the larger of the individual resistor and
capacitor asymptotes, as illustrated by the heavy lines in Fig 28. For this example, these are
in fact the exact asymptotes of || Z ||. In the limiting case of zero frequency (dc), then the
10k
|| Z ||
capacitor tends to an open circuit. 80dB
1k
The series combination is then 60dB
dominated by the capacitor, and 40dB
100
the
exact
function
tends
R
10
20dB
f0
asymptotically to the capacitor
1 = 16kHz
1
1
2RC
C
impedance magnitude. In the 0dB
0.1
limiting case of infinite frequency, 20dB
100Hz
1kHz
10kHz
100kHz
1MHz
then the capacitor tends to a short Fig. 28. Construction of the composite asymptotes of
|| Z ||. The asymptotes of the series combination can
circuit, and the total impedance
be approximated by simply selecting the larger of the
becomes simply R. So the R and
individual resistor and capacitor asymptotes.
1/C lines are the exact
asymptotes for this example.
The corner frequency f 0 , where the asymptotes intersect, can now be easily
deduced. At angular frequency 0 = 2f0, the two asymptotes are equal in value:
1 =R
0C
(124)
1
(125)
0 = 1 =
= 10 5 rad/sec
RC (10) (10 -6F)
f0 = 0 = 1 = 16kHz
2 2RC
So if we can write analytical expressions for the asymptotes, then we can equate the
expressions to find analytical expressions for the corner frequencies where the asymptotes
intersect.
The deviation of the exact curve from the asymptotes follows all of the usual rules.
The slope of the asymptotes changes by +20dB/decade at the corner frequency f0 (i.e.,
from 20dB/decade to 0dB/decade), and hence there is a zero at f = f 0 . So the exact
30
R
Z(s)
L
Z(s) = R + sL + 1
sC
(126)
The magnitudes of the individual resistor, inductor, and capacitor
Fig. 29. Series Rasymptotes are plotted in Fig. 30, for the values
L-C network
R = 1k
example.
L = 1mH
C = 0.1F
(127)
The series impedance Z(s) is dominated by the capacitor at low frequency, by the resistor at
mid frequencies, and by the inductor at high frequencies, as illustrated by the bold line in
Fig. 30. The impedance Z(s) contains a zero at angular frequency 1, where the capacitor
and resistor asymptotes intersect. By equating the expressions for the resistor and capacitor
asymptotes, we can find 1:
R=
1
1 = 1
1 C
RC
(128)
A second zero occurs at angular frequency 2, where the inductor and resistor asymptotes
intersect. Upon equating the expressions for the resistor and inductor asymptotes at 2, we
obtain the following:
R = 2 L 2 = R
L
(129)
So simple expressions for all important features of the magnitude Bode plot of Z(s) can be
obtained directly. It should be 100dB
100k
|| Z ||
noted that Eqs. (128) and (129)
10k
80dB
are approximate, rather than
1k
60dB
R
f1
f2
exact, expressions for the corner
100
frequencies 1 and 2. Equations 40dB
1
31
|| Z ||
100dB
100k
10k
80dB
1k
60dB
f0
R0
40dB
20dB
10
1
C
0dB
100Hz
100
1kHz
10kHz
100kHz
1
1MHz
Fig. 31.
Graphical construction of impedance asymptotes
for the series R-L-C network example, with R
decreased to 10.
0L = 1 = R0
0C
(130)
0 =
1
LC
(131)
100k
10k
80dB
1k
60dB
actual curve
40dB
0dB
100Hz
f0
20dB
Z( j0) = R + j0L +
|| Z ||
100dB
R0
Q = R0 / R
100
10
1
C
L
1kHz
10kHz
100kHz
1
1MHz
Fig. 32.
Actual impedance magnitude (solid line) for the
series R-L-C example. The inductor and capacitor
impedances cancel out at f = f0, and hence Z(j0) = R.
1
j0C
(132)
Z( j0) = R + jR0 +
R0
= R + jR0 jR0 = R
j
(133)
At = 0, the inductor and capacitor impedances are equal in magnitude but opposite in
phase. Hence, they exactly cancel out in the series impedance, and we are left with Z(j0) =
32
R, as illustrated in Fig. 32. The actual curve in the vicinity of the resonance at = 0 can
deviate significantly from the asymptotes, because its value is determined by R rather than
L or 1/C.
We know from section 1.5 that the deviation of the actual curve from the
asymptotes at = 0 is equal to Q. From Fig. 32, one can see that
= R0
R
Q= 0
R
dB
dB
dB
, or
(134)
Equations (130) - (134) are exact results for the series resonant circuit.
The practice of adding asymptotes by simply selecting the larger asymptote can be
applied to transfer functions as well as impedances. For example, suppose that we have
already constructed the magnitude asymptotes of two transfer functions, G1 and G 2, and
we wish to find the asymptotes of G = G1 + G 2. At each frequency, the asymptote for G
can be approximated by simply selecting the larger of the asymptotes for G1 and G2:
G 1,
G1 >> G2
G 2,
G2 >> G1
(135)
Corner frequencies can be found by equating expressions for asymptotes as illustrated in
the preceeding examples. In the next chapter, we will see that this approach yields a simple
and powerful method for determining the closed-loop transfer functions of feedback
systems. The accuracy of the result is discussed in section 3.3.
3 . 2 . Parallel impedances: inverse addition of asymptotes
A parallel combination represents inverse addition of impedances:
1
(136)
1 + 1 +
Z1 Z2
If the asymptotes of the individual impedances Z1, Z2, , are known, then the asymptotes
of the parallel combination Zpar can be found by simply selecting the smallest individual
Z par =
impedance asymptote. This is true because the smallest impedance will have the largest
inverse, and will dominate the inverse sum. As in the case of the series impedances, this
procedure will often yield the exact asymptotes of Zpar.
Let us construct the magnitude asymptotes for the Z(s)
parallel R-L-C network of Fig. 33, using the following element
values:
Fig. 33.
R = 10
Parallel R-L-C
network example.
33
at = 1, R = 1L 1 = R
L
1
at = 2, R =
2 = 1
RC
2C
(138)
These expressions could have been obtained by conventional analysis, combined with the
low-Q approximation of section 1.6.
Figure 35 illustrates what 80dB
10k
1
L
C
happens when the value of R is
1k
60dB
R
increased to 1k. The asymptotes
100
40dB
R0
for || Z || then become independent of
f0
10
R, and change directly from L to 20dB
1/C at angular frequency 0. The
|| Z ||
1
0dB
corner frequency 0 is now the
frequency where the inductor and
capacitor asymptotes have equal
value:
20dB
100Hz
1kHz
10kHz
100kHz
0.1
1MHz
Fig. 35.
Graphical construction of impedance asymptotes
for the parallel R-L-C example, with R increased to 1k.
0L = 1 = R0
0C
(139)
0 =
1
LC
(140)
34
80dB
Z( j0) = R || j0L ||
60dB
1k
Q = R / R0
R0
actual curve
40dB
100
f0
10
20dB
0dB
20dB
100Hz
1
C
|| Z ||
L
1kHz
10kHz
100kHz
0.1
1MHz
Fig. 36.
Actual impedance magnitude (solid line) for the
parallel R-L-C example. The inductor and capacitor
impedances cancel out at f = f0, and hence Z(j0) = R.
1 =
1
j0C 1 + 1 + j0C
R j0L
(141)
Z( j0) =
1
1
=
=R
j
j
j
1+ 1 +
1
+
R jR0 R0 R R0 R0
(142)
So at = 0, the impedances of the inductor and capacitor again cancel out, and we are left
with Z(j0) = R. The values of L and C determine the values of the asymptotes, but R
determines the value of the actual curve at = 0.
The actual curve is illustrated in Fig. 36. The deviation of the actual curve from the
asymptotes at = 0 is
= R
Q= R
R0
dB
dB
R0
dB
, or
(143)
Equations (139) - (143) are exact results for the parallel resonant circuit.
3.3.
Another example
Now let us consider a slightly more complicated
example, Fig. 37. This example illustrates the multiple
application of the series and parallel rules, and it also
illustrates why the algebra-on-the-graph approach in general
yields approximate, rather than exact, asymptotes.
For this circuit, the impedance Z consists of the seriesparallel combination,
Z(s) = R1 || 1 + R2 || 1
sC 1
sC 2
35
Z(s)
R1
20k
C1
33nF
R2
5k
C2
3.3nF
Fig. 37.
Series-parallel
network example.
(144)
R1 || 1
sC 1
1
and R2 || sC
2
As
before,
the
parallel
combinations are constructed by
selecting the smaller of the
individual asymptotes. The result
is illustrated in Fig. 39. The
parallel combination of R1 and C1
contains a pole at frequency f 1 ,
where the R 1 and 1/C1
asymptotes intersect. By equating
these
two
asymptotes
at
frequency f1, we obtain
R1 =
1
1C 1
(145)
120dB
1
C 1
100dB
33nF
1
C 2
1M
3.3nF
100k
R1 20k
80dB R 5k
2
10k
60dB
1k
40dB
100
20dB
10Hz
100Hz
1kHz
10kHz
10
100kHz
Fig. 38.
Impedance magnitudes of the individual elements
in the network of Fig. 37.
120dB
1
C 1
1M
1
C 2
100dB
100k
R1
f1
80dB
R2
f2
60dB
1k
R1 || 1
sC 1
R2 || 1
sC 2
40dB
20dB
10Hz
10k
100
100Hz
1kHz
10kHz
10
100kHz
Fig. 39.
Graphical construction of asymptotes for
parallel combinations (solid and shaded lines).
120dB
1
C 1
100dB
1M
1
C 2
100k
f1
R1
f2
80dB
R2
10k
f3
60dB
1k
1
= 240Hz
2R1C 1
|| Z ||
100
(146) 40dB
10
Likewise,
the
parallel 20dB
10Hz
100Hz
1kHz
10kHz
100kHz
combination of R 2 and C2
Fig. 40.
Graphical construction of composite asymptotes
for
||
Z
||.
contains a pole at frequency f 2 ,
where the R 2 and 1/C2 asymptotes intersect. By equating these two asymptotes at
frequency f2 and proceeding in a similar manner, we obtain
f1 =
f2 =
1
= 9.6kHz
2R2C 2
(147)
36
R2 =
1
3C 1
(148)
f3 =
1
= 965Hz
2R2C 1
(149)
or,
The actual curve deviates from the asymptotes, according to the usual rules. For example,
at f2, the actual curve deviates from the asymptotes by 3dB.
The asymptotes derived in this example are approximate rather than exact. For
example, the actual dc asymptote is R1 + R2 = 25k, rather than simply R 1 = 20k. This
can be seen from the original circuit, letting the two capacitors become open circuits. On a
dB scale, 25k represents 28dBk while 20k represents 26dBk, so the approximate
dc asymptote deviates by 2dB from the exact value. Likewise, the high-frequency
asymptote should be
1
C 1C 2
C 1+C 2
(150)
rather than simply 1/C2. The actual corner frequencies may in general also be slightly
displaced from the values shown. For this example, the pole frequencies f 1 and f2 given
above are the exact values, while the exact zero frequency f3 is 1096 Hz rather than 965
Hz.
When adding asymptotes having the same slopes, such as R 1 and R 2 in this
example, the actual asymptote is the sum (R 1 + R 2) rather than simply the largest.
Likewise, the parallel combination of two asymptotes having the same slope is actually
given by the inverse addition formula, rather than simply the smaller asymptote. So the
algebra on the graph method is an approximation. The worst case, having least accuracy,
occurs when the asymptotes having the same slope also have the same magnitude the
approximate asymptote will then differ by 6dB. When the two impedances have different
values, then the deviation will be less than 6dB. Also implicit in the algebra on the graph
method is the approximate factorization of the numerator and denominator polynomials of
the impedance or transfer function. The accuracy of the result is high when the corner
frequencies are well separated in magnitude.
The fact that the graphical construction method yields an approximate answer
should not be viewed as a disadvantage. Indeed, this is a great advantage of the method:
Bode diagrams of reasonable accuracy can be obtained very quickly. Much physical
understanding of the circuit can be gained, because simple analytical expressions are found
37
for how the salient features (corner frequencies and asymptote values) depend on the
component values. Suitable approximations become obvious.
In the majority of design situations, absolute accuracy is not so important. Instead,
it is more important for the design engineer to gain insight into how the circuit works, so
that he or she can figure out how to make the circuit behave as desired. Simple
approximations are the key to obtaining this insight; more exact (and more complicated)
equations may remain as enigmatic as the original circuit. In those cases where more
accuracy is needed, and where it is judged worthwhile to expend the effort, a less
approximate analysis can still be performed.
The graphical construction method for impedance magnitudes is well known, and
impedance graph paper can be purchased commercially. As illustrated in Fig. 41, the
magnitudes
of
the
80dB
10k
100
pF
impedances of various
10H
60dB
1k
1nF
inductances,
capacitances,
and 40dB 1H
100
10n
F
H
resistances are plotted
m
100
20dB
10
100
nF
on
semilogarithmic
H
10m
0dB
1
1F
axes. Asymptotes for
the impedances of R-L- 20dB 1mH
100m
10
F
C networks can be
H
100
40dB
10m
100
F
sketched directly on
100
H 10m
1F
1m
n
m
H
0
H
H
F 1nH
F 10n
F 10
0
1
1m
these
axes,
and 60dB 1
10Hz
100Hz
1kHz
10kHz
100kHz
1MHz
numerical values of
Fig. 41. Impedance graph: an aid for graphical construction of
corner frequencies can
impedances, with the magnitudes of various inductive,
capacitive, and resistive impedances pre-plotted.
then be graphically
determined.
3.4.
38
a)
b)
He(s)
sLe
ve(s)
Zin(s)
1/sC
+
v(s)
Zout(s)
Le
Zout
Z2(s)
Z1(s)
Z1
Z2
Fig. 42. Transfer function He(s) and output impedance Zout(s) of a single-section L-C filter: (a) circuit,
with He and Zout identified; (b) determination of Zout, by setting independent sources to zero.
Z2
Z
v(s)
=
= 2
ve(s) Z 1+Z 2 Z in
(151)
For the example of Fig. 42, Z1(s) = sL e, and Z2(s) is the parallel combination of R and
1/sC. Hence, we can find the transfer function asymptotes by constructing the asymptotes
of Z 2 and of the series combination represented by Z in , and then dividing. Another
approach, which is easier to apply in this example, is to multiply the numerator and
denominator of Eq. (151) by Z1:
ZZ
Z
v(s)
= 2 1 1 = out
ve(s) Z 1+Z 2 Z 1 Z 1
(152)
where Zout = Z 1 || Z 2 is the output impedance of the voltage divider. So another way to
construct the voltage divider transfer function is to first construct the asymptotes for Z 1 and
for the parallel combination represented by Z out, and then divide. This method is useful
when the parallel combination Z1 || Z2 is easier to construct than the series combination Z 1
+ Z2. It often gives a different approximate result, which may be more (or sometimes less)
accurate than the result obtained using Zin.
The output impedance Zout in Fig. 42(b) is
Z out(s) = R || 1 || sL e
(153)
sC
The impedance of the parallel R-L-C network is constructed in section 3.2, and is
illustrated in Fig. 43(a) for the high-Q case.
According to Eq. (152), the voltage divider transfer function magnitude is
|| H e || = || Zout || / || Z 1 ||. This quantity is constructed in Fig. 43(b). For < 0, the
asymptote of || Zout || coincides with || Z1 ||: both are equal to L e. Hence, the ratio is
|| Z out || / || Z 1 || = 1. For > 0, the asymptote of || Z out || is 1/C, while || Z1 || is equal
to Le. The ratio then becomes || Zout || / || Z 1 || = 1/2LeC, and hence the high-frequency
asymptote has a40dB/decade slope. At = 0, || Zout || has exact value R, while || Z 1 || has
39
a)
b)
|| Z1 || = Le
1
C
L e
=1
L e
Q = R / R0
R0
Q = R / R0
L e
He =
|| Zout ||
Fig. 43.
1 /C
f0
f0
1
2L eC
Z out
Z1
Graphical construction of He and Zout: (a) output impedance Zout; (b) transfer function He.
exact value R0. The ratio is then || He(j0) || = || Zout(j0) || / || Z1(j0) || = R / R0 = Q. So the
filter transfer function He has the same 0 and Q as the impedance Zout.
It now becomes obvious how
variations in element values affect the
salient features of the transfer function and
output impedance. For example, the effect
of increasing the inductance value L e on
1
C
increasing
Le
Le
R
Q = R / R0
R0
f0
|| Zout ||
Fig. 44.
Variation of output impedance
asymptotes, corner frequency, and Q-factor
with increasing inductance Le.
134.7
vx
known as a network
analyzer, or frequency
response analyzer. The
key inputs and outputs of
Fig.45.
Key features and functions of a network analyzer: sinusoidal
a basic network analyzer
source of controllable amplitude and frequency, two analog inputs,
are illustrated in Fig. 45.
and determination of relative magnitude and phase of the input
+
4.
The network analyzer provides a sinusoidal output voltage vz of controllable amplitude and
frequency. This signal can be injected into the system to be measured, at any desired
location. The network analyzer also has two (or more) inputs, vx and vy . The return
electrodes of vz , vx and vy are internally connected to earth ground. The network analyzer
performs the function of a narrowband tracking voltmeter: it measures the components of
vx and vy at the injection frequency, and displays the magnitude and phase of the quantity
vy / vx . The narrowband tracking voltmeter feature is essential for measurements over a
wide range of magnitudes; otherwise, noise and interference from neighboring circuits
corrupt the desired sinusoidal signals and make accurate measurements impossible [3].
Modern network analyzers can automatically sweep the frequency of the injection source
vz to generate magnitude and phase Bode plots of the transfer function vy / vx .
vz
magnitude
Measured inputs
Data
vz
frequency
vz
output
vy
vx
vx
input
vy
input
vy
vx
4.7 dB
Data bus
to computer
162.8
DC
blocking
capacitor
VCC
input
DC
bias
adjust
G(s)
output
Network Analyzer
Injection source
Device
under test
Fig. 46.
and output terminals of the amplifier. Hence, the measured transfer function is
vy(s)
= G(s)
vx(s)
(154)
Note that the blocking capacitance, bias potentiometer, and vz amplitude have no effect on
the measured transfer function.
41
An impedance
Z(s) =
v(s)
i(s)
(155)
can be measured by treating the impedance as a transfer function from current to voltage.
For example, measurement of the output impedance of an amplifier is illustrated in Fig. 47.
The quiescent operating condition is again established by a potentiometer which biases the
amplifier input. The injection source vz is coupled to the amplifier output through a dc
blocking capacitor. The injection source voltage vz excites a current i out in impedance Z s.
This current flows into the output of the amplifier, and excites a voltage across the amplifier
output impedance:
Z out(s) =
vy(s)
i out(s)
amplifier
=0
ac input
(156)
A current probe is used to measure i out . The current probe produces a voltage proportional
to i out ; this voltage is connected to the network analyzer input vx . A voltage probe is used
i out
input
DC
bias
adjust
Zs
Device
under test
G(s)
output
VCC
Zout
DC blocking
capacitor R
source
current
probe
voltage
probe
+ +
not
affect
the
vy
vx
measurement of Zout.
Fig. 47. Measurement of the output impedance of a system.
It is sometimes necessary to measure impedances that are very small in magnitude.
Grounding problems cause the test setup of Fig. 47 to fail in such cases. The reason is
illustrated in Fig. 48(a). Since the return connections of the injection source vz and the
analyzer input vy are both connected to earth ground, the injected current i out can return to
the source through the return connections of either the injection source or the voltage probe.
In practice, i out divides between the two paths according to their relative impedances.
Hence, a significant current (1 k) out flows through the return connection of the voltage
probe. If the voltage probe return connection has some total contact and wiring impedance
Zprobe, then the current induces a voltage drop (1 k) out Zprobe in the voltage probe wiring,
as illustrated in Fig. 48(a). Hence, the network analyzer does not correctly measure the
42
vz
a)
injection
source
return
connection
Impedance
under test
i out
Z(s)
Injection source
Rsource
+
Zrz
k i out
i out
Network Analyzer
(1 k) i out
vz
Measured
inputs
voltage
probe
voltage
probe
return
connection
Zprobe
vx
+
vy
(1 k) i out Z probe
b)
injection
source
return
connection
Impedance
under test
i out
Network Analyzer
Injection source
1:n
Z(s)
Zrz
i out
Rsource
+
vz
0
Measured
inputs
voltage
probe
voltage
probe
return
connection
Zprobe
{
+
0V
vx
vy
Fig. 48.
Measurement of a small impedance Z(s): (a) current flowing in the return connection of the
voltage probe induces a voltage drop which corrupts the measurement; (b) an improved experiment,
incorporating isolation of the injection source.
voltage drop across the impedance Z. If the internal ground connections of the network
analyzer have negligible impedance, then the network analyzer will display the following
impedance:
Z + (1 k) Z probe = Z + Z probe || Z rz
43
(157)
Here, Zrz is the impedance of the injection source return connection. So to obtain an
accurate measurement, the following condition must be satisfied:
Z >>
Z probe || Z rz
(158)
44
8.
REFERENCES
[1] R.D. Middlebrook, Low Entropy Expressions: The Key to Design-Oriented Analysis, IEEE Frontiers
in Education Conference, 1991 Proceedings, pp. 399-403, Sept. 1991.
[2] R. D. Middlebrook, Methods of Design-Oriented Analysis: The Quadratic Equation Revisited, IEEE
Frontiers in Education Conference, 1992 Proceedings, pp. 95-102, Nov. 1991.
[3] F. Barzegar, S. Cuk, and R. D. Middlebrook, Using Small Computers to Model and Measure
Magnitude and Phase of Regulator Transfer Functions and Loop Gain, Proceedings of Powercon
8, April 1981. Also in Advances in Switched-Mode Power Conversion, Irvine: Teslaco, vol. 1,
pp. 251-278, 1981.
P ROBLEMS
1.
Express the gains represented by the asymptotes of Figs. 49(a)-(c) in factored pole-zero form. You
may assume that all poles and zeroes have negative real parts.
2.
Derive analytical expressions for the low-frequency asymptotes of the magnitude Bode plots shown
in Fig. 50(a)-(c).
a)
a)
Gm
f0
f1
f0
20dB/decade
+20dB/decade
+20dB/decade
b)
b)
Q1
f1
f1
20 dB/decade
Gm
40dB/dec
f2
f2
f3
Q2
20 dB/decade
c)
f1
c)
20dB/dec
Q
f1
f2
20dB/dec
+20dB/dec
40dB/dec
f2
f3
Q
Fig. 49.
Fig. 50.
45
3.
Express the gains represented by the asymptotes of Figs. 50(a)-(c) in factored pole-zero form. You
may assume that all poles and zeroes have negative real parts.
4.
Derive analytical expressions for the three magnitude asymptotes of Fig. 14.
5.
40dB
30dB
|| A ||
20dB
90
10dB
45
A
0
0dB
-45
-90
-135
10Hz
100Hz
1kHz
10kHz
100kHz
-180
1MHz
Fig. 51.
turn in: (1) your analytical expression for A(s), with numerical values given, and (2) a copy of
Fig. 51, with your magnitude and phase asymptotes superimposed and with all break frequencies
and slopes clearly labeled.
6.
46
v1
30dB
20dB
|| Z ||
10dB
0dB
90
-10dB
45
Z
-45
10Hz
100Hz
1kHz
-90
10kHz
Fig. 52.
L1
10mH
+
C2
220F
C1
47nF
R2
100
R3
1k
Zout
v2
Fig. 53.
8.
Operational amplifier filter circuit. The op amp circuit shown in Fig. 54 is a practical realization
of what is known as a PID controller, and is sometimes used to modify the loop gain of feedback
circuits to improve their performance. Using semilog graph paper, sketch the Bode diagram of the
magnitude of the transfer function v2(s) / v1(s) of the circuit shown. Label all corner frequencies,
flat asymptote gains, and asymptote slopes, as appropriate, giving both analytical expressions and
R4
numerical values. You may assume that
C2
2k
8F
the op amp is ideal.
9.
R3 20k
R2
1k
C3 800pF
v1
R1
100
C1
24nF
+
v2
Fig. 54.
47
10.
v1
R2
C1
L2
R1
Zout
C2
Fig. 55.
11.
12.
L1 = 100H
L2 = 16mH
C1 = 1000F
C 2 = 10F
R 1 = 5
R 2 = 50
L1
R1
vg
L3
L2
R2
C1
L4
C2
Fig. 56.
L1 = 32mH
C 1 = 32F
L2 = 400H
C 2 = 6.8F
L3 = 800H
R 1 = 10
L4 = 1H
R 2 = 1
(a)
Construct || Z s || using the algebra on the graph method. Give simple approximate
analytical expressions for all asymptotes and corner frequencies.
(b)
1
L1 + L2
L LC
1+s
+ s 2 L 1 C 1 + C 2 + L 2C 2 + s 3 1 2 1 + s 4 L 1L 2C 1C 2
R
R
C2
L2
48
4.7F
50H
Zs
13.
14.
(a)
Factor G(s) into approximate real and quadratic poles, as appropriate. Give analytical
expressions for the salient features. Justify your approximation using the numerical
element values.
(b)
(c)
R1
L1
R2
L3
v1
L2
L4
C1
C2
Zout
Fig. 57.
R1
R2
10
C1
10F
C2
1F
L1
32H
L2
8H
L3
3.2mH
L4
320H
sC 1R
1 + sR (C 1 + C 2) + s 2LC 1 + s 3LC 1C 2R
(a)
When C1 is sufficiently large, this transfer function can be expressed as an inverted pole
and a quadratic pole pair. Derive analytical expressions for the corner frequencies and Qfactor in this case, and sketch typical magnitude asymptotes. Determine analytical
conditions for validity of your approximation.
(b)
When C 2 is sufficiently large, the transfer function can be also expressed as an inverted
pole and a quadratic pole pair. Derive analytical expressions for the corner frequencies and
Q-factor in this case, and sketch typical magnitude asymptotes. Determine analytical
conditions for validity of your approximation in this case.
(c)
When C 1 = C 2 and when the quadratic poles have sufficiently high Q, then the transfer
function can again be expressed as an inverted pole and a quadratic pole pair. Derive
analytical expressions for the corner frequencies and Q-factor in this case, and sketch
typical magnitude asymptotes. Determine analytical conditions for validity of your
approximation in this case.
49