Documente Academic
Documente Profesional
Documente Cultură
MICROWAVE FILTERS
9.1
9.2
9.3
Filter Transformations
9.3.1 Frequency Transformations
9.3.2 Load Transformation
9.4.1
9-1
Chapter 9
MICROWAVE FILTERS
Microwave filter is one of the very useful components for microwave systems. It can be
used for excluding undesired frequencies as in UWB systems, for combining or sorting
power at different frequencies as in multiplexers, and for impedance matching over
broad-bandwidths. Filters can be classified into four categories on the basis of their
insertion loss as a function of frequency; low-pass, band-pass, band-stop and high-pass.
The design of microwave filters starts with a prototype filter whose low-pass lumpedelement ladder network is normalized to 1-ohm terminations and a cut-off frequency of 1
rad/sec. Equations and/or tables are used for determining the prototype element values.
Frequency and load transformations are then used to derive high-pass, band-pass, bandstop filters and low-pass filters with higher cut-off frequency. The filter with new element
values may be implemented in lumped element form. If the lumped element realization is
not possible, relationship between the transmission line section and the equivalent
lumped elements are used to realize the filter in the distributed element form. Suitable
techniques may be employed for changing the microwave filter configuration thus arrived
at into a form suitable for practical implementation. Schematic of the above procedure is
shown in Fig. 9.1.
There are two low-frequency synthesis techniques in common use. These are: image
parameter method (and the variations thereof, such as the constant-k and m-derived
filters) and the insertion loss method. We shall use the insertion loss method because it is
based on complete specification of a physically realizable frequency characteristic. These
characteristics are: (a) width of pass-band frequencies, (b) maximum insertion loss or
VSWR in the pass-band, and (c) minimum insertion loss or isolation in the stop band at
one or more than one frequency. We shall use the frequency variable
for the prototype
filter and is reserved for later use with microwave filters.
9.1
INSERTION-LOSS METHOD OF FILTER DESIGN
This method of filter design has been very well described in the text book by Collin*. A
filter is specified by the insertion-loss as a function of frequency. The insertion loss can
, defined as the ratio of the power
also be expressed in terms of power loss ratio
delivered by the generator to the load when connected directly, to the power delivered
when the filter is inserted in between. If is the input reflection coefficient of the filter
terminated in a matched load, then
(9.1)
|
|
9-2
(9.2)
and
are real, positive and even polynomials. Although an unlimited
where
number of power loss ratios can be specified satisfying (9.2), it has been found in practice
Specs
Low-pass lumped
element prototype filter
for desired response
Design eqns
and/or Tables
Distributed
network conversion
Lumped element
realization
Practical realization
e.g. stripline
that for microwave filter design only three types are sufficient. These are: maximally flat
or Butterworth response in the passband, an equi-ripple or Chebyshev response in
passband, and an elliptic function filter. Design of first two types of filter is described
next. The design of elliptic function filter is beyond the scope of the text.
9.1.1 Butterworth Filter
The power loss ratio for a maximally flat or Butterworth filter is described by the
following equation
1
/
(9.3)
1
/ )
where
. is the Chebyshev polynomial of degree N, and described as
(9.4)
for
for
(9.5)
The behavior of
. is shown in Fig. 9.3.
. oscillates between 1 for
and
. Therefore, power loss ratio will oscillate between 1
increases monotonically for
in the passband (
), equals 1
at
, and will increase
and 1
monotonically for
. The number of ripples in the pass-band equals N/2 (see Fig.
9.2).
When compared with maximally flat filter, (9.3), this power loss ratio is larger by a factor
2
. It means that the Chebyshev filter has a sharper selectivity or a much smaller
frequency range separating the passband and the stopband, and is a desirable feature. For
1.7 for the maximally flat
example for N=15, 70 dB attenuation is reached at
filter case and at
1.18 for the Chebyshev filter case.
9-4
Fig. 9.2: Low-pass filter response for maximally flat and Cheyshev filters; N =2,3.
Passband tolerance k 2 and VSWR
Setting the power loss ratio (9.1) to 1
at
, we obtain the following
relationship between passband tolerance k and the input reflection coefficient
(9.7)
(9.8a)
For k = 0.509 the insertion loss equals 1 dB and the corresponding VSWR is 2.66; and
VSWR = 5.83 for an insertion loss of 3dB. The passband tolerance
may also be
expressed in terms of passband ripple
in dB using the following equation
10
10
(dB)
dB /10
(9.8b)
The insertion loss value in the stopband is called isolation or stopband attenuation, and is
given by the following expression
for the Butterworth filter case
Isolation (dB) = 10
9-5
9-6
9-7
=1+
(9.10)
2 2
4 2 2 2
(9.11)
Comparison at
0 gives R = 1; comparison of coefficient of
imply L = C and
yield
comparison of coefficient of
4
or
L = C = 2
(9.12)
9.2.2 Chebyshev Filter Design
is described by (9.4). Therefore,
For this type of frequency characteristics
2
2
Since
2 2
1, (9.13) becomes
2
1 21
2 2
0 gives
Comparison at
4 2 2 2
4 2 2 2
(9.13)
(9.14)
or
2 1
R = 1+2
(9.15a)
,
,
2
4
(9.15b)
(9.15c)
9-8
The above equations can be solved to determine L and C. It may be noted that R 1
unlike the case of Butterworth filter. The reason is that the value of Chebyshev function
( ) is not zero at
0, in general, and depends on the order N according to
0 for N odd
(0) =
(9.16)
1 for N even
Therefore, Chebyshev filters require
1
for N odd
(9.17)
!
2
2
1 2
2 1
for N even
If the Chebyshev filter is to be designed for load R = 1, the mismatch between R 1 and
R=1 for N even can be bridged by employing an impedance transformer in between. An
alternative is to increase the order of filter to make N odd.
The analysis procedure becomes cumbersome for large values of N. Instead, one can use
the Tables available or the general expressions given next.
(a) Maximally flat lowpass filter element values [Collins, Matthaei]
1
(9.18a)
!
and
2
1,2,3,
(9.18b)
where is the value either of inductance in henry or capacitance in farad. Table 9.1
gives the element values for = 1,
1 and k = 1 for N = 1 to 10.
(b) Equi-ripple filter element values [Collins, Matthaei]
The general expressions are given below for a low-pass filter with
1
!
or
1:
for N odd
for N even
(9.19a)
(9.19b)
and
where
2,3,
(9.19c)
1,2,3,
(9.19d)
(9.19e)
1,2,3,
with
(9.19f)
1, and
0.5
and 3dB for N =
Table 9.2 gives the element values for = 1,
1 to 10. The element values are numbered from
as the generator resistance or
conductance to
as
the
load
resistance
or
conductance
for a filter with N reactive
!
elements. The elements are defined as:
9-9
(9.20a)
(9.20b)
1,2,3,
Table 9.1: Element values for maximally flat low-pass filter prototype with k = 1,
1
N
1
2
3
4
2.000
1.414
1.000
0.7654
1.0
1.414 1.0
2.000 1.000 1.0
1.848 1.848 0.765
1.0
5
6
7
8
9
10
0.6180
0.5176
0.4450
0.3902
0.3473
0.3129
1.618
1.414
1.247
1.111
1.000
0.908
0.6180
1.414
1.802
1.962
2.000
1.975
2.000
1.932
1.802
1.663
1.532
1.414
1.618
1.932
2.000
1.962
1.879
1.782
1.0
0.5176
1.247
1.663
1.879
1.975
1.0
0.445
1.111
1.532
1.782
1.0
0.3902 1.0
1.000 0.3473 1.0
1.414 0.908 0.3129 1.0
Table 9.2: Element values for equi-ripple low-pass filter prototype with
0.5 , k = 0.3493
N
1
2
3
4
5
6
7
8
9
10
0.6986
1.4029
1.5963
1.6703
1.7058
1.7254
1.7372
1.7451
1.7504
1.7543
1.0
0.7071
1.0967
1.1926
1.2296
1.2479
1.2583
1.2647
1.2690
1.2721
1.9841
1.5963
2.3661
2.5408
2.6064
2.6381
2.6564
2.6678
2.6754
1.0
0.8419
1.2296
1.3137
1.3444
1.3590
1.3673
1.3725
1.9841
1.7058
2.4758
2.6381
2.6964
2.7239
2.7392
1.0
0.8696
1.2583
1.3389
1.3673
1.3806
= 1,
1.9841
1.7372
2.5093
2.6678
2.7231
= 1,
1.0
0.8796 1.9841
1.2690 1.7504 1.0
1.3485 2.5239 0.8842 1.9841
9-10
3
N
1
2
3
4
5
6
7
8
9
10
,k=1
0.6986
1.4029
1.5963
1.6703
1.7058
1.7254
1.7372
1.7451
1.7504
1.7543
1.0
0.7071
1.0967
1.1926
1.2296
1.2479
1.2583
1.2647
1.2690
1.2721
1.9841
1.5963
2.3661
2.5408
2.6064
2.6381
2.6564
2.6678
2.6754
1.0
0.8419
1.2296
1.3137
1.3444
1.3590
1.3673
1.3725
1.9841
1.7058
2.4758
2.6381
2.6964
2.7239
2.7392
1.0
0.8696
1.2583
1.3389
1.3673
1.3806
1.9841
1.7372
2.5093
2.6678
2.7231
1.0
0.8796 1.9841
1.2690 1.7504 1.0
1.3485 2.5239 0.8842 1.9841
for k = 1,
(9.21)
(ii)
(iii)
for
(9.22)
For the value of N determined above, read the element values of the prototype
filter from the appropriate Table 9.1 or 9.2. Otherwise, (9.18)-(9.19) can be
programmed for the purpose.
Scale the element values according to the frequency transformation and load
1 and = 1 for the prototype filter.
transformation discussed next, since
9.3
FILTER TRANSFORMATIONS
The design of low-pass prototype filter was discussed in the last section. The element
1 for Butterworth and Chebyshev characteristics. We
values correspond to = 1,
need to transform this filter into another filter with either higher cut-off low-pass filter, or
band-pass, band-stop or high-pass filter. The element values therefore need to be properly
scaled. We employ frequency and impedance transformations for the purpose.
9.3.1 Frequency Transformations
(a) Low-pass Cut-off Translation: If the cut-off frequency of the new low-pass filter is
to be raised from
1 to , we apply the frequency translation
/
(9.23)
The frequency translation of (9.23) maps the points 1 to the points
. This
is shown in Fig. 9.7. The series reactances and shunt susceptances for the new filter are
9-11
obtained by replacing by /
value
should be
Writing
so that this impedance value does not disturb the impedance match
with source and load gives the new inductance
(9.25a)
Similarly, the new capacitance value
is related to the prototype element value
as
(9.25b)
and
for
1. The power loss ratio for the new
It may be noted that
filter is shown in Fig. 9.7(b).
(b) Low-pass to High-pass Transformation: A high-pass filter comprises of series
capacitors and shunt inductors because of their favourable impedance behaviour at high
frequencies. This is proved next by using the following frequency transformation of the
prototype low-pass filter
/
(9.26)
This transformation maps the points 0 to the points
, the points 1 to
(9.27)
9-12
such
(9.28)
The high-pass transformation of the prototype filter for N = 3 is shown in Fig. 9.9.
(c) Low-pass to Band-pass Transformation: A band-pass filter allows a band of
frequencies to pass through while rejecting the frequencies on either side. This response
is similar to that of a resonator. Therefore, it is natural that a band-pass filter is realized
using resonant elements. The appropriate frequency transformation of the low-pass filter
is
(9.29)
9-13
where
and
are the lower and upper frequencies of the desired passband, and
the geometric mean of
and . Equation (9.29) can be solved for to give
is
(9.30)
, we obtain
(9.31)
(9.33a)
(9.33b)
9-14
If we choose
(9.34)
The required series reactance of (9.33a) can now be seen to be realized if we replace the
reactance (9.32a) of the prototype filter by a series circuit such that
or
for the series element
(9.35a)
(9.35b)
Similarly,
(9.36a)
and
and
(9.36b)
Thus, in a band-pass filter the series inductor are replaced by series tuned L-C circuits
and the shunt capacitors are replaced by parallel tuned L-C circuits as shown in Fig. 9.12.
The relationship between the low-pass prototype element values and the band-pass prototype filter element values is given by (9.36).
9.3.2 LOAD TRANSFORMATION
The design of low-pass, high-pass and band-pass prototype filters worked out so far
assumes load and source impedances of 1 ohm, except for Chebyshev filters with N even
for which load resistance is not unity. For realistic filters with a source or load impedance
of , the new values of inductances and capacitances should be such that impedances of
series and shunt elements are multiplied by , i.e.
(9.37a)
/
(9.37b)
Fig. 9.11: Series and parallel tuned circuits for the realization of band-pass filters.
9-15
- low
9-16
L/2, (Z11-Z12)
L/2
Z0 , l
C, (Z12)
Shunt element:
For
/2 and 1
and
/
, the above expressions may be approximated as
and
Correspondence
/2
gives
(9.38)
(9.39)
The equivalent circuit of Fig. 9.14 may be approximated as lumped inductor or capacitor
alone if the effect of the other element is made negligible. For this, if we choose high
for the inductor, according to (9.39) we need small for a given value of L. Also, high
and small reduce the associated C. Therefore, we choose highest value of
consistent with fabrication limitations to realize series inductor. Similarly, lowest value
of
is employed to realize shunt capacitor. The approximate expressions for the line
lengths are (
/4 :
/
for the inductor
(9.41)
Example 9.1: Let us design an N = 3 maximally flat low-pass filter with cut-off at 2
GHz in stripline configuration. The maximum realizable
is 118 ohm and the minimum
is 30.5 ohm.
realizable
Solution: Let us choose -configuration for the prototype filter with 1 ohm source and
load impedances as shown in Fig. 9.16. From the Table 9.1 we obtain
1
and
= 2H
Frequency translation to
2
gives the following filter element values
0.159
and
79.5
9-17
w1
w2
Z 01
jX
Z 02
2.7, b
Fig. 9.18: Transfer characteristics of the low-pass filter shown in Fig. 9.12
2GHz
2.7, b 6.35mm,
9-19
The insertion loss versus frequency behaviour of the realized stripline filter is plotted in
Fig. 9.18. The rate of increase in attenuation with frequency is slow in the stop-band. The
stop-band attenuation is about 10dB at 4 GHz (2 whereas the expected value for
lumped element design is 18dB. The distributed element design gives poor stop-band
attenuation because the equivalent lumped values depart significantly from the design
outside the pass-band (Pozar, p.473).
where
sin
m ,
and
1 for
0
0
2 for
Here, b, l,
and
are the effective dimensions of the filter geometry and are greater
than the corresponding physical dimensions as they include the energy stored in the
fringing fields, and can be determined using the planar waveguide model of
microstrip/stripline.
Since the dimensions of the filter element to be considered are very small compared to
the minimum wavelength of operation we can reduce the number of modes contributing
at a cost of small error. By retaining only the first two terms, corresponding to the
to
and
modes, the impedance matrix of the filter reduces to
/
(9.43)
(9.44)
gives
and
2
1
/
is the resonance frequency for the
i = 1,2
mode and is given by
9-20
Fig. 9.19: Geometry of a planar elliptic function low-pass filter; b and l are
effective dimensions.
1
1
2
where
(9.45)
It is seen from (9.45) that the area lb of the strip affects the values of capacitances. The
inductance depends on the aspect ratio l/b, and is intuitively expected that narrower
the strip the larger its inductance value. The values of filter elements , , and
can be controlled independently by varying the dimensions of the strip and the location of
the ports and . The physical realizability of the lumped circuit of Fig. 9.20 imposes
----------------------------------------------* Dinzeo, et. al., Novel microwave integrated low-pass filters, Electron. Lett., vol. 15,
1979, pp. 258-260
9-21
Fig. 9.20: A simple lumped equivalent -network of the filter geometry of Fig. 9.19.
some constraints on the parameters of the strip and ports. For the case ,
2 is nearly unity, the following conditions should be satisfied
,
Also, the capacitors
with respect to l/2.
and
, so that
(9.46)
The
mode resonance frequency
also corresponds to the open circuit resonance
of the equivalent circuit of Fig. 9.20, and is given by
(9.47)
parameter of the filter and is
The parallel resonance produces a zero in the
responsible for the sharp increase in insertion loss about
.
Example 9.2: A low-pass Cauer-Chebyshev 03-02-41 type of filter using the twodimensional circuit approach is described*. Here, the first two digits define the order of
the filter (n = 03), the next two digits determine the reflection coefficient (in this case
20%), and the last two digits specify the modular angle , that describes the sharpness of
the filter response. In this case
1.5243
and
2.5 and
9-22
= 17.82mm, = 2.557mm,
12.03
4.51
for a cutoff frequency of 5 GHz. The computed insertion loss and return loss of the designed filter
is plotted in Fig. 9.21. The stopband attenuation is about 15 dB at 7.5GHz (1.5 ..
9.4.2 Bandpass Filters
A bandpass filter consists of series resonators alternating with parallel resonators as
shown in Fig. 9.12 for the prototype filter. This arrangement is, however, difficult to
realize using distributed parameters at microwave frequencies. It is much more practical
to realize a circuit that employs only one type of resonant circuit either series or parallel.
This limitation arises from the transmission line technology. The new configuration can
be obtained from the configuration of Fig. 9.12 by using impedance or admittance
inverters to convert one type of resonant circuit to another type.
= 17.82mm,
(a)
= 2.557mm
.
.
. , h = 1.6mm
9-23
= 17.82mm,
. ,
= 2.557mm
.
.
.
(b)
Fig. 9.21: Frequency response of designed low-pass elliptic function filter of Fig. 9.19.
(a) Insertion loss, (b) return loss
9-24
.
9.4.4 Impedance/admittance inverters: The word immittance is a common word used
for both impedance and admittance. An immittance inverter in its simplest form is a
quarter-wave transformer. It can be used to transform a series element into a shunt
element with same frequency dependence, e.g. a shunt admittance such as jC can be
transformed to a series element like jL by an impedance inverter K. Further, the
impedance value jL can be controlled by the value of K and is given by K =
/ .
Similarly, a series resonant circuit may be transformed into a parallel resonant circuit and
vice versa by an immittance inverter. For example, a shunt admittance Y preceded and
followed by admittance inverters as in Fig. 9.22 will perform like a series impedance, i.e.,
the two networks will have identical transmission coefficient provided
(9.48)
where J is the characteristic admittance of the admittance inverters. For the dual circuit
shown in Fig. 9.22 we have
(9.49)
where K is the characteristic impedance of the impedance inverters. The admittance
(impedance) level of the transformed network can be controlled through the value of J
(K).
The quarter-wave section based inverters will have narrow bandwidth (1-2%) because the
variation of electrical length with frequency. An alternative is to employ lumped element
based immittance inverters. Impedance (admittance) inverter realized using a shunt
inductor (series capacitor) and its equivalence is given in Fig. 9.23.
Fig. 9.22: Use of immittance inverters to convert one type of resonant circuit to
another type of resonant circuit.
9-25
/4
/2
/2
jX
| /2|
/4
/2
/2
jB
| /2|
Fig. 9.23: Impedance (admittance) inverters and their realization using shunt inductor
(series capacitor) surrounded by line length /2.
The relationships between the equivalent parameters X(B), ( ,), and K (J) can be
determined by comparing the ABCD-parameters of the equivalent networks at a quarterwave frequency. Some more inverters based on lumped elements are given in the book,
Foundations for Microwave Engineering, Sec. 8.18.
9-26