Electronic filters design tutorial

PREFACE

I designed my first low pass filter in 1975. It was I’m not saying the simulator is not useful , but it
a T filter, used to reduce the harmonics of a 5 W will be more difficult to obtain good result.
transmitter. The transmitter was already built in
mass production, but it did not pass If you have chosen a “closed box”
homologation tests. Inserting an external boxed electromagnetic simulator for the development of
filter on the antenna output, it passed all antennas, probably you do not have enough
homologation tests, and I gained a solemn specific knowledge about electromagnetic
encomium. simulators for the design of antennas, and you
hope to compensate your loss of specific
Since then, in a third of a Century, I designed knowledge by means of an EM simulator, that will
hundreds of filters, with different topologies and unfortunately be of scarce help.
use. Starting from hand calculation the first years
and going thru the various and more refined CAD Instead, if you read at least one book on EM
systems, decade to decade: before SPICE simulator you will be able to choose the best
simulators, and after EM ones. program that fit your application!

In all these years I spent a lot of time to

understand the concepts behind filters.
Mathematic approach is important, but only a This tutorial is divided in five sections:
number of filter approximations are available for
the design. If your needs do not fall into 1)Band pass passive filters
someone of these classes your design will be non
optimal. 2)Band pass filter with distributed
components
In my opinion the best simulator available is our
brain! All available simulators today are number 3)High pass, low pass and notch passive
crunching, but if you don’t have in your mind the filters
right model of the circuit, the simulated circuit
could be anything from perfect to useless. It is 4) Active filters
not the case to came up with Murphy’s law!
Simply if the result is perfect you have chosen 5) Time domain filters
the right model/circuit for your application
(maybe you are a clever engineer, or maybe you It is expected from the reader a general
are a bit lucky!), but if the circuit do not operate electronic knowledge, but the use of mathematic
properly this means only that the model is is purposely reduced to a minimum, the scope of
incorrect or incomplete in your application, or the the tutorial is to improve the practical knowledge
circuit chosen do not fit the application. of filters, mathematic is demanded to the
simulators used in many examples. In the
In every case you have to increase your bibliography, books are listed that investigate
knowledge of the circuit/system before designing also mathematical aspect of filters.
it. This seems very simple, and is the basis for
every engineering work, but unfortunately the
today culture is to consider circuit blocks as
“black box” and to avoid the study inside them,
merely connecting blocks one to the other, and
hoping Mr Murphy will be far for holidays and
anything will go well!

This is true for every circuit, system and situation

in life.
A simple example? If you buy an electromagnetic
simulator program which operate in “closed box”
environment, it will be perfect for simulating
filters and circuit matching, but it will be poor in
antenna simulation, since the closed box will
alter antenna impedance and radiation pattern.

ANGELO BONI, Redox s.r.l. Page 1 of 13 [Link] 13/02/2009

 Electronic filters design tutorial
Par 1) THE RESONAT CIRCUIT

The circuit of Fig.1 consist of a parallel resonant Note that impedance of inductor and capacitor
circuit at 100 MHz, loaded with a 50 Ω resistor, has to be changed simultaneously, otherwise a
the generator is a perfect voltage source. Since resonance shift in frequency will occur.
both the inductor and the capacitor exhibit a
reactance of 50 Ω the Q of the circuit is 1 for
definition (Q = R/XC, or = R/XL).

Fig.3
Per definition the BW of the circuit at 3dB point is
equal to: BW = 1/Q Now we have all the elements to realise a band
pass filter; we have only to add two reactive
and hence the bandwidth is equal to the central elements in the circuit and place the load at
frequency, 100 MHz. output to 50Ω, in order to maximize the power
transfer. The two reactance in series with
Running the simulation (fig 2) it can be seen that generator and load determine the Q of the filter.
Here the relationship is more complex, since
input and output coupling are now complex
number, due to the sum of 50Ω resistors with
series reactance. Doubling the reactance do not
exactly halves the bandwidth.

the BW goes from 62 MHz to 162MHz, or exactly

the 100 MHz expected.
Running a subsequent simulation we step the
resistor from 50 Ω to 250 Ω, enhancing the Q
from 1 to 5. The BW should decrease in
proportion, moving from 100 MHz to 20 MHz.
In Fig 4 appears such a filter where two different
It can be seen in Fig3 that the simulation circuits are simulated with series reactance of
displays as expected. The same result can be 50Ω and 100Ω. The response in Fig.5 makes
obtained lowering the impedance of the reactive evidence of the two bandwidth.
elements, for example increasing the capacitor
value five times and decreasing the inductor
value at 1/5 of its value will yield to an
impedance of 10 Ω and the Q will be again 5.

ANGELO BONI, Redox s.r.l. Page 2 of 13 [Link] 13/02/2009

 Electronic filters design tutorial
The response show that while the filter with
series inductor and capacitor has a symmetrical
response (blue trace), the two inductor filters has
steeper response at higher frequencies (red
trace), this network is normally referred as low
pass coupling. The two capacitor filter has
steeper response at low frequencies (green
trace), this type of coupling is referred as high
pass coupling.

Fig.5

Moving the values of C2 and L2 we are able to

set the bandwidth of the filter, note that the two
reactance must be equal in amplitude (and
opposite in sign, in this case) otherwise central
frequency will shift and insertion loss won’t be
zero (simulation is made with lossless
components)

Till now the circuit is made with two Fig. 7

complementary series reactance. Note that in the two coupling capacitor filter, the
value of resonant capacitor C5 has been
What happens if the two series reactance are decreased in order to compensate the frequency
equal? shift caused by the two coupling capacitors, in
the two inductor filter inductance L1 has been
In Fig 6 three circuits appear simulating the increased to compensate for the frequency shift
different possibilities. It is very interesting to caused by the two inductors.
check the effects of this modifications on the
filter response in Fig.7 ∗ these different frequency responses lead
to a great flexibility in applications.

For example, if you have to design a front end

filter for a receiver with an image to reject below
the operating frequency, a filter with capacitor
coupling will maximize image rejection.
If the first local oscillator is operating beyond the
operating frequency, the image frequency will
appear also over the operating frequency; in this
case the two inductor coupled filter will give the
maximum image rejection.

∗ To recapitulate:
1) With a very simple calculation we are able
to set a resonant circuit with a loaded Q of
1 to start the design of a filter
2) Placing the proper reactance in series with
the generator and the load we are able to
change the Q of the filter, and
consequently the BW
3) Changing the type of reactance in series
we are able to manipulate the response of
the filter, enhancing cutoff at higher or

ANGELO BONI, Redox s.r.l. Page 3 of 13 [Link] 13/02/2009

 Electronic filters design tutorial
lower frequency, or making the filter The two inductor are substituted by a single one,
symmetrical. L2, with twice the value. Running the simulation
Spice simulators: all Spice based simulator are we should expect a filter with a sharper response
useful for this type of analysis, even a linear and a bandwidth slightly reduced with respect to
simulator can be used for this first section with the single pole. Running the simulation (Fig.9)
passive filters.

If you do not have a Spice simulator you can

freely download one of the following:
A) Spectrum Software: Microcap-9
[Link]
free demo version downloadable with reduced
performances, but adequate for the trials with
this tutorial. For learning purpose the
evaluation version can be used indefinitely.
All the simulation presented here are made
whit the full version of Microcap-8

B) Texas Instrument: Tina-TI

[Link]
[Link] a free version of TINA developed Fig.9
for Texas Instrument. Actual version is 7.0 we can see that the filter has a ripple of 3 dB in
the passband and a slight center frequency shift
C) Simetrix: SIMetrix of approximately 7 MHz.
[Link] Both are caused by the termination between the
[Link] another free version of spice two resonators that now differ from the resistive
simulator 50Ω seen in the single stage filter.

C) Linear Technology: LT spiceIV Editing the value of L2 we can correct the ripple
[Link] in the BW. In Fig.10 the L2 value is stepped from
#Spice spice simulator developed for Linear 600 to 1000 nH in 100 nH steps.
The two tuning capacitors C1 and C6 has been
reduced to 25 pF to compensate for frequency
Multiple resonators filters drift. The vertical divisions are set to 1dB.
Filters used in real life rarely are constructed with
a single resonator, because the rejection is poor
in most applications. Moreover the shape of the
response is gaussian and again in many
application the shape of the filter is requested
more rectangular.

In Fig.8 two single resonators are connected

together:

Fig.10
Stepping the inductor L2 has three effects
simultaneously:
1) change the bandwidth of the filter
2) change the ripple in the bandwidth
3) shift slightly the center frequency
The point 3) can easily be overcame tuning
capacitors C1 and C6 to the right frequency.
About the point 1) and 2) you can select L2 for
the best ripple in band, and adjust the bandwidth
of the filter as will be explained later, or if the
ANGELO BONI, Redox s.r.l. Page 4 of 13 [Link] 13/02/2009
 Electronic filters design tutorial
ripple is acceptable for a given BW, you can fix
the filter as is.
In Fig.11 the response of the filter is show, with
L2 set at 800nH for maximum in-BW flatness.

Fig.11
Both the close-in response and the broad band
attenuation are shown. You can see the dramatic
improvement in response compared with the
single resonator filter (blue trace). The in- band
response is more flat, and the attenuated BW is
extremely sharp (red trace)

What happen now if we want to change the

bandwidth of the filter? Simply you have to edit
the parameter of the series components, C5–C7–
L2, changes must be done together, decreasing
the reactance of L2 (decreasing the inductance of
L2) and simultaneously decreasing the reactance
of C5 and C7 (increasing the value of the
capacitors). Unfortunately there isn’t a perfectly
linear relationship between the BW and the value Fig.12
of these components. In fig.12 we initially halves
the value of L2 to 400 nH and increased C5 and The blue trace is the original filter, the red one is
C7 to 20 pF. As expected the BW did not double, the OUT2 of the drawing.
and the filter exhibits a poor response. Changing
C5 and C7 to 15 pF restore the shape of the filter We can verify another parameter in order to
and the BW move from 13 MHz to 25 MHz. check the proper fitting of the filter we are
See the diagram in Fig. 12 and the relative designing. This parameter is called in many
responses. ways: return loss, VSWR, source and load match,
etc.
ANGELO BONI, Redox s.r.l. Page 5 of 13 [Link] 13/02/2009
 Electronic filters design tutorial
Return loss is an important indicator that the A common subject of misunderstanding is the
filter is matched to the generator. An alternate definition of “Q” or quality factor, that is used
indicator is the insertion loss, that for a perfectly both for the filter bandwidth and for the quality
matched lossless filter must be zero. of components. When Q is used for filters we can
But when we will use real component with its use the term “loaded Q”; for components the
own loss it will be difficult to understand if the quality factor is expressed simply as Q or
insertion loss of the filter will be generated by “unloaded Q”.
mismatch or by component loss.
Monitoring the return loss, we can distinguish When Q is used for filters with a single resonator
between the two types of loss. the relationship is Q= 1/BW, as explained before.
In Fig.13 the return loss in dB of the filter from If the filter has many resonators the relationship
Fig. 12 is simulated. became complex, because normally the various
resonator do not exhibit the same loaded Q in
the circuit. Usually in this case the loaded Q of
the filter is referred as the highest Q present in
the filter. This no longer gave information on the
BW of the filter, but is useful for evaluating the
minimum Q requested by the components used
in the filter.

And many times the misunderstanding is here!

To build a filter with a loaded Q of 10 you don’t
need inductors an capacitors with an unloaded Q
of 10, but as a rule of thumb, you need at
least a Q of 100.

Many times people misunderstand loaded and

unloaded Q and try to assembly a filter with
components totally unsuited for that application.

∗ physical components
In a third of a century I have heard every type of
story about component suited or not suited for
filter design.
One example for all: it isn’t possible to use
ceramic capacitors in audio filters, they are not
stable and they “sound” (makes acoustical noise)
when used at relatively high power.
Evidently upholders of this theory makes their
experiments and found that some capacitors are
not stable with temperature, applied voltage, and
aging. Moreover some of this capacitors exhibit a
piezoelectric effect, and they “sound” when music
is applied. Poor quality ceramics as Z5U and Y5V
exhibit astonishing tolerances, and some of them
Fig.13 has piezoelectric effect.

The return loss of the filter in the pass band is in But on the other way exist ceramic materials that
excess of 20 dB, a good result for a lossless has very tight tolerance (5% to 1%), zero
filter. temperature, voltage and aging drift. Materials as
The loss in the filter generated by a return loss of COG or NPO has these characteristics and are
20 dB is less then 0,05 dB. As we will see later perfectly suited for audio frequency filters.
the loss induced by the parasitic of components
is greater at least one order of magnitude. My position is to avoid the suggestions of people
who made a work or two in the field, but they are
∗ Q of the filter and Q of components
not expert. If you makes a design or two and find
a solution, you are probably going to consider
your circuit as the only possible one. But since

ANGELO BONI, Redox s.r.l. Page 6 of 13 [Link] 13/02/2009

 Electronic filters design tutorial
you aren’t an expert in this field, maybe a dozen You can download SW evaluating the
of circuit with best performances exist. performance of capacitors from the following
As before, my suggestion is to learn about the sites:
design you are going to made from many and
different sources, and after verify experimentally 1) Murata: Mcsil
if the knowledge is true. [Link]
ml a great SW for understanding the capacitor
A customer requested me to review a design of a properties. There you can see series resistance,
system that has production yield problems. I frequency of auto-resonance, Q of capacitors and
found that each stage was buffered by an so on. It contains also parameters for small SMD
additional one, being each stage an operational inductors
amplifier or an RF amplifier, each one was
followed by an additional buffering stage. In 70% 2) Kemet: Kemet spice
of location these buffers were not only [Link]
unnecessary but they spread the production [Link]/Weben/kemsoft#fit it contains data
parameters and gave a poor yield in production. for ceramic and tantalum capacitors.
I asked to the designer why he used that design
methodology and his answer was that his chief
engineer many years ago teach him to design in ∗ Inductors:
that manner because “ a redundancy in design is The greatest part of inductor used today in filters
always advisable, so you have room for fixing are SMD wire-wounded. Monolithic type usually
bugs and problems”. has lower Q and rarely are used in filters, where
I believe this is the perfect example to cite for to keep losses low, maximum Q is needed.
indicating a person which do not work with his Thru- hole inductors are generally used only for
head! power applications.

∗ Capacitors:
Ferrite or air core Inductors?
Air inductor are used for higher frequency, where
Radio frequency filters employed in the majority ferrite cores had unacceptable loss. On the
of applications use multilayer ceramic capacitors contrary at moderate frequency (up to 10-50
of the SMD type; the dielectric used is usually MHz) the ferrite core could increase the Q of the
COG or NPO which exhibit a zero drift over inductor.
temperature.
Only in high power applications porcelain, teflon One of the best WEB site for information on
and mica capacitors are employed. inductor is Coilcraft:
In microwaves circuits single layer types are [Link] where you
used. can find a lot of data on Q, inductance vs.
frequency and frequency of auto-resonance of
At lower frequency you can use ceramic inductors.
capacitor, polypropylene film capacitors
(especially at high voltages), mica capacitor in Web sites represented here are only suggestion
special cases (they are very expensive). The use to consult, the directory would be very long, and
of polyester film capacitors isn’t recommended, the choose to present only these is personally.
since they are not stable with temperature.

Manufacturers in their WEB sites have useful ∗ Other passive component used in filters:
tools to investigate the properties of their
capacitors. Spice models are available also for 1)Quartz crystal, SAW resonator and ceramic
many simulators brand. However in the first resonator are used in narrowband filters, the
phase I’m suggesting you to avoid this models design of these filters are usually monolithic and
and to check manually the behaviour of they are outside the purpose of this tutorial and
capacitors in the environment. Manual insertion will not be treated here. However in the
of the parasitic element in the simulator will bibliography a book on this topic is listed.
improve your knowledge of components.
Of course when you will be skilled this approach 2) strip-line, coaxial resonator and other
will be no longer necessary, but if you gain this distributed components: these components will
component knowledge, you will became able to be treated in the second part of this tutorial,
see errors or problems in the models when they since the design of filter with this class of
don’t fit your application. component is quite different.
ANGELO BONI, Redox s.r.l. Page 7 of 13 [Link] 13/02/2009
 Electronic filters design tutorial

Starting the design with real components.

Using all the tools learned in the previous

section, we can start the design of a practical
filter. The example is a realization of a bandpass
filter used to separate the FM audio from the
video signal. The FM audio channels are from 6
to 7.5 MHz, where the maximum allowable loss is
3 dB, and the bandpass ripple allowed is +/-2 dB
maximum. An attenuation of at least 50 dB is
requested from DC to 4 MHz (the BW of the video
signal), the attenuation requested in the upper
band from 10 to 100 MHz is at least 15 dB, since
strong signals are not present here, there is only Fig.15
to reject noise and some spurious frequencies.
Response appear in fig 15, it can be seen that
Specifications of design also request that the some ripple in bandwidth already exist, indicating
filter will be produced with all SMD components that the design is not perfect, but the use of
and could be assembled with no post-production standard value in components is highly regarded
tuning. in production.

Starting from the design of Fig.6 the “high pass” Now we can start to insert the value of parasitic
response (out 3) has been chosen, in order to in components, since the design is at few MHz,
increase cutoff at low frequency. we can insert the series resistance only, because
at these frequencies self inductance is negligible.
To obtain the required attenuation at least three Capacitors has been chosen as 0603 size, COG
resonator must be used. This is a rule of thumb, ceramic with 2% tolerance, and inductors are in
dictated by experience, but it is quite easy for 1008 size and 5% tolerance. In Fig. 16 the
the experimenter to start the design with a two parasitic resistances are indicated in red.
poles filter and verifying that specifications are
not met. (I believe a single pole filter, checking
Fig.6, is discarded immediately)

The filter with ideal components has been

simulated in Fig.14. Note that components with
standard value has been used, in order to
simplify the following steps.

Fig.16

Fig.14

The input and output impedances are 75Ω, the

value of components are all standardized, and
luckily only two values of capacitors are needed,
and the three inductors has the same value.
Fig.17

ANGELO BONI, Redox s.r.l. Page 8 of 13 [Link] 13/02/2009

 Electronic filters design tutorial
It can be seen that coil losses dominate all over
the other losses. The response of the filter suffer
from such a high losses, see Fig.17. It is
important to decrease the loss of inductors,
maybe using a larger case model.

Running the Monte-Carlo simulation at Fig.18

(100 iterations, worst case distribution) it can be
seen that we are outside specification for BW loss
at 6 MHz: some iterations show a 7.5 dB loss
instead of 3+2 dB maximum allowed.

Fig.20

We can see in Fig.21 that a production lot of

even 1000 pieces with Gaussian distribution meet
the specification even if the worst case
distribution is slightly outside specification
(Fig.20)

Fig.18

We made two more simulation with Monte-Carlo

analysis, the first (Fig.19) with the same inductor
but with a 2% tolerance, the second (Fig.20)
with a 1210 size inductor which exhibit a 2Ω
series resistance.

Fig.21

So the filter goes into production without

problems with the original inductors, but with a
2% tolerance.
In that case I did not win a Solemn Encomium as
in the previous case, but only a coffee from a
colleague that bet with me that it was impossible
to manufacture such a filter without tuning. I
won only a coffee, but it was an Italian coffee,
very strong and very thick!
Fig.19
It can be seen that the best solution is to employ Next step we will increase the operating
a 2% tolerance inductor, in this way we met the frequency and will start to introduce the parasitic
specification, while in Fig 20 you can see that few of the PCB and the surrounding environment.
iterations are outside the specification.
It is to be noted however that we are using in the Till now we have only inserted the parasitic
simulation the worst case distribution, using resistance of capacitors and inductors. But
gaussian distribution instead, which is the normal capacitors exhibit also parasitic inductance, which
spread in components (or at least it should be). is, for ceramic capacitors mainly function of the
size.

ANGELO BONI, Redox s.r.l. Page 9 of 13 [Link] 13/02/2009

 Electronic filters design tutorial

∗ As a rule of thumb here are the parasitic ∗ If in a certain point you have a 100 pF to GND
inductance for capacitors in the following size: you can ignore the parasitic capacitance of say
0.25 pF, but if you are operating at 1 GHz, the
size 0201 0402 0603 0805 1206 1812 impedance in that point will be 1.6 Ω and if you
Induct. .4nH .5 n .63 n .55 n .74 n .9 n have a connection to GND with a via hole its
parasitic inductance of 1 nH will exhibit a
These value can change varying the value of the reactance of 6.28 Ω. Of course your circuit will
capacitor and its voltage, however in the table not operate properly, so you will have to choose
are listed the mean values and typically the a different grounding mechanism, or change the
difference is less than a +/- 15%. If you need a working impedance of 1.6 Ω, or even better do
more precise value data are available in the Web both the remedies.
sites of various manufacturer.
Till now the filter presented here are all poles
Inductors: of course they exhibit parasitic type, they have zero response at DC or at
capacitance, but the value is not so case- infinity. A different class of filters exist that
dependent. The value of capacitance strongly places zero responses at some frequencies of
depend on manufacturing technology, number of interest, they are called elliptic filters.
turns and inductor value, core material and so
on. In the next example we will insert not only
The only way to have reliable data is to verify in parasitic but we will use an elliptic filter.
the data sheet the resonating frequency and
calculate the capacity from this data. One of our customers gave us specifications of an
Usually if the frequency of resonance is 3-5 times input filter for a receiver operating from 250 to
higher than the operating one, this capacitance 300 MHz. The loss of the filter should be less
can be neglected. than 1dB in order to keep the receiver sensitive.
Cutoff at low frequencies should be at least 25 db
Printed circuits: three main sources of parasitic at 100 MHz and 50 dB at 30 MHz.
exist on a PCB, first the capacitances of PADs The local oscillator was beating in the upper
which are functions of PCB thickness and mode and the fist IF was at 110.6 MHz. Therefore
material. Second vias to GND and vias to other the image was 250 + 2 X IF = 471,2 MHz as
layers of the PCB which behave like inductors. minimum and 300 + 2 X IF = 521,2 MHz. The
Third the tracks that connect the various required cutoff in the image band was 45 dB,
component also behave like inductors. since the tuned filter between the low noise
In the following table a list of typical values for amplifier and the mixer cut the rest. The receiver
parasitic capacitances and inductances are shown must have a maximum height of 7 mm and
hence coils must be miniaturized.
Thickness PAD Via Via Track A three resonators filter was chosen, and the
to GND Capacity ∅ .5 ∅1 .5 mm three resonant inductors were chosen of the
Layer Epoxy mm mm 1cm long Mini-spring air core inductor, from Coilcraft,
.4mm .7 pF .25nH .2nH 2.5nH which exhibit a Q of 150 at the frequency of
.8mm .3 pF .5nH .4nH 3.3nH operation. The inductors has a nominal value of
1.2mm .2 pF .8nH .6nH 3.6nH 18.5 nH, but the value inserted in the simulator
1.6mm .14pF 1.2nH .9nH 4.1nH is 20,5 nH to keep count of parasitic inductance.
All the parasitic resistances are indicated in red.
A capacitance per square millimetre is indicated The parasitic capacitances are not inserted in the
in the table, to obtain the value of capacitance simulation since all them are referred to ground
for a given PAD you have to calculate the actual and hence can be inserted in the value of the
area. Please note that the thickness do not three resonating capacitors C1,C2 and C3.
necessary correspond to the thickness of the
PCB, in the case of a multilayer PCB the
thickness correspond to the distance from the
populated layer (normally the top layer) to the
ground plane (normally the inner layer 1). The
inductance of the track is indicated for a length of
one centimetre, over a ground plane; it is
measured on a 50Ω impedance.
Fig.22

ANGELO BONI, Redox s.r.l. Page 10 of 13 [Link] 13/02/2009

 Electronic filters design tutorial
The tolerances of components are set to 2%,
except for the central resonator, where
tolerances are set to zero (the capacitor C3 will
be tuned), otherwise the spread of parameters
will be too high. In Fig 23a appear the analysis of
the filter, where you can see that the loss in the
passband meet the specification even with
Monte-Carlo analysis (500 run) Fig 23b.

Fig 24

Fig. 23a

Fig.25
It appears that the cutoff at higher frequencies
has greatly increased (green trace versus red
trace). Now the specification of 45 dB cutoff at
frequencies over 471.2 MHz is easily met. The
insertion loss instead is slightly increased, but it
Fig.23b is still inside specification.

At low frequencies the specification is also met, Many people believe my way to design filter is a
but in the image band the attenuation is 15 dB loss of time. Usually you can synthesize the filter
out of specification. directly with a program or from a data table.
The solution, instead of redesign the filter with a
higher resonator number, is to replace the two I don’t use only this method to design filters, the
coupling inductors L4 and L5 with two parallel approach is useful for teaching people how to
resonant circuits, in this case the filter became learn about filters, and is the only method useful
elliptical and two zeros are placed near the image when the design needs fall outside the class of
band, increasing the attenuation. In Fig.24 you filters available in the programs of synthesis or in
can see the modified circuit, with the two the data tables.
capacitors added. The capacitor C8 has been
decreased to retune the central resonator. In the following 2 example we will synthesize two
The two inductors L9 and L10 has been filters, one perfectly fitting real components, the
decreased in value to compensate for the other impossible to manufacture with that
fictitious increasing of the apparent inductance in design.
the passband caused by the two 2.4 pF If you will understand immediately why the
capacitors. second design couldn’t be manufactured, I hope
In Fig. 25 you can see the simulation results. this tutorial has increased your knowledge.
ANGELO BONI, Redox s.r.l. Page 11 of 13 [Link] 13/02/2009
 Electronic filters design tutorial
The first design is a filter with a bandpass of 480 Now instead we will design an 870 MHz filter with
to 880 MHz with a maximum of 3dB loss and a 100 MHz BW and a 3 dB maximum loss.
rejection of at least 20 db from 100 MHz to both Rejection of at least 25 dB is requested from 50
band edges. MHz to both edges.
The spice simulator (Micro cap) directly made the
synthesis, (Fig. 26) the ripple in BW is set to 0,5 The design is in Fig 28 and the ideal response in
dB as a design parameter. Fig.29.

Fig.26 Fig.28

The design is reasonable, parallel inductors are of

acceptable values, even if great care must be
taken in the central resonator, to keep parasitic
low, due to the 2,5 nH inductor. Series
resonators also has acceptable values and the
filter works with real value components: Mini-
spring high Q inductors and low loss ceramic
capacitors. Tolerances are 2% for inductors, 1%
for 12 and 24 pF capacitors and 5% for 2.7 pF
type.
The simulation appears in Fig.27.

Fig.29
It can be seen that the simulated values are
perfect, but the value of the components used in
the filter are hard or impossible at all to find in
the physical word. The three resonance inductors
has sub-nH values (0.494 and 0.351 nH) even if
you find them with the right value, it seems
difficult to place them in a real word PCB where
the mere distance between two components has
nearly the same inductance value. The same is
true for capacitor C3, which has a reactance of
1,7 Ω; its self inductance generate a reactance
grater than this value. Moreover the series
Fig.27 capacitors C2 and C4 has a value of 0.386 pF;
Red trace is the theoretical simulation and the the nearest commercial value available is 0.39
blue one is a Monte-Carlo analysis of 100 runs. It pF, the tightest available tolerance is +/- 0.1pF
can be seen that at the band edge at 480 MHz or a +/- 25%! Again the capacitance versus
the trace is slightly outside the specification, but ground of the two pads in the hot points between
this will be correct in the physical sample. L2-C2 and L4-C4 exhibit a parasitic capacitance
We can see that in this design the synthesis tool nearly the same value of C2 and C4. It seems
work properly and the result is directly applicable
to a prototype.
ANGELO BONI, Redox s.r.l. Page 12 of 13 [Link] 13/02/2009
 Electronic filters design tutorial
clear that in this case the circuit synthesis is Disclaimer
useless.
With this last simulation the first section of this This document is provided for your personal use only
tutorial ends. and may not be retransmitted or redistributed without
The second part will threats of filters with written permission from Redox srl. You may not
distributed elements. distribute this document or part of it without prior written
permission from Redox. You may not make copies for
Bibliography any commercial purpose. You do not obtain any
ownership right, title, or other interest by downloading,
copying, or otherwise using these materials.
1) MICROWAVE FILTERS, IMPEDANCE-MATCHING
You assume the entire risk related to your use of this
NETWORKS, AND COUPLING STRUCTURES data. Redox srl is providing this data "as is," and Redox
G. Matthaei, [Link], E.M.T. Jones srl disclaims any and all warranties, whether express or
1st edition 1964 ISBN:0-89006-0991. implied, including (without limitation) any implied
A Historical book, but still actual today warranties of merchantability or fitness for a particular
purpose. In no event will Redox srl be liable to you or to
2) HANDBOOK OF FILTER SINTHESIS any third party for any direct, indirect, incidental,
Anatol I. Zverev ISBN:0-471-74942-7 consequential, special or exemplary damages or lost
1st edition 1967 profit resulting from any use or misuse of this data.
A classical book on filter, full of tables for
filter design

3) FILTERING IN THE TIME AND FREQUENCY

DOMAIN [Link], H.J. Blinchikoff
1st edition 1976 ISBN: 1-884932-17-7
Another classical book, perfectly useful today,
you couldn’t miss it in your library

4) INTRODUCTION TO THE THEORY AND DESIGN

OF ACTIVE FILTERS
[Link], [Link] ISBN:0-07-030854-3
1st edition 1980
Another dated book, but still actual.

5) ELECTRONIC FILTER DESIGN HANDBOOK

[Link], [Link] ISBN:0-07-070434-1
1st edition 1981
A practical book, containing a lot of tables.

6) CRYSTAL FILTERS [Link]

1st edition 1987 ISBN:0-471-88478-2
A useful book for the engineers interested in
crystal filters design.

7) COMPUTER-AIDED CIRCUIT ANALYSIS USING

SPICE, [Link] ISBN:0-13-162579-9
1st edition 1989
A useful book about Spice simulation

8) MICROWAVE CIRCUIT MODELING USING

ELECTROMAGNETIC FIELD SIMULATION
[Link], [Link]
1st edition 2003 ISBN:1-58053-308-6
A great book on electromagnetic simulation.

9) STRIPLINE CIRCUIT DESIGN Harlan Hove, Jr.

1st edition 1974 ISBN:0-89006-020-7
A good book on stripline circuits and
components.

ANGELO BONI, Redox s.r.l. Page 13 of 13 [Link] 13/02/2009