Keysight Technologies

Simpliied Filter Tuning Using Time Domain

Application Note
Introduction

The increase in wireless communications services is forcing more and more channels into less fre-
quency spectrum. To avoid interference, very stringent iltering requirements are being placed on
all systems. These systems usually employ coupled resonator ilters to handle the power levels and
provide the needed isolation. The dificulty of tuning these ilters quickly and accurately often limits
manufacturers from increasing their production volumes and reducing manufacturing cost.

In a coupled-resonator cavity-tuned ilter, the center frequency of each resonator must be precisely
tuned. The couplings between resonators must also be precisely set to achieve the proper passband
response, low return loss (relection), and small passband ripple. Setting coupling coeficients and
tuning the resonators are as much art as science; often a trial-and-error adjustment process. Until
now, there has been no alternative.

This application note describes a method of tuning a ilter using the time-domain response of its
return loss, which makes ilter tuning vastly easier. It is possible to tune each resonator individually,
since time-domain measurements can distinguish the individual responses of each resonator and cou-
pling aperture. Such clear identiication of responses is extremely dificult in the frequency domain.
Coupling coeficients may be precisely set to provide a desired ilter response, and any interaction
caused by adjustment of the coupling structures and resonators can be immediately determined and
accounted for.

Perhaps the most important advantage of the time-domain tuning method is that it allows inexpe-
rienced ilter tuners to successfully tune multiple-pole ilters after only brief instruction. Such rapid
proiciency is impossible with previous tuning methods. This technique also lends itself well to the
automated production environment, which has always been a challenge.

Dificulties of ilter tuning

The interactive nature of coupled-resonator ilters makes it dificult to determine which resonator or
coupling element needs to be tuned. Although some tuning methods can achieve an approximately
correct ilter response, inal tuning often requires the seemingly random adjustment of each element
until the inal desired ilter shape is obtained. Experienced tuners can develop a feel for the proper
adjustments, but months are often required before a novice can be proicient at tuning complex ilters.
The time and associated cost of tuning, and the dificulty and cost in training new personnel can limit
a company’s growth and responsiveness to changing customer needs.

Some companies have attempted to automate the tuning process, using robotics to engage and turn
the tuning screws, and an algorithmic process to accomplish the tuning. The tuning algorithms are
a particular problem, especially when a ilter is nearly tuned, at which point the interaction between
stages can be so great that inal tuning cannot be achieved. New ilter designs may require entirely
new algorithms, making it even more dificult for test designers to keep up with changing require-
ments. Manufacturing changes that affect the ilter components, such as tool wear or changing
vendors, may also cause algorithms and processes to become less effective.

In some cases, tuned ilters go through temperature cycling or other environmental stress as part of
the manufacturing process, and their characteristics may change as a result. It can be very dificult
to identify which resonators or coupling apertures need to be retuned using conventional ilter tuning
methods.
Ideal tuning method
The solution to these dificulties would be a tuning
method that is simple, lexible, and deterministic. That
is, one in which the individual adjustment goals for each
tuning element, resonator, and coupling aperture would
not depend upon the other elements in the ilter. The
response to each tuning screw would be easily identiied,
and any interactive effect would be immediately seen and
accounted for. Ideally, each screw would only need to
be adjusted once. Finally, the tuning method would not
depend on ilter type or shape, or number of ilter poles.

This application note presents a technique that clearly

identiies the resonator or coupling aperture that needs
to be tuned, and enables the operator to see and correct
for interactions. Filters can be tuned to match any ilter
shape within their tuning ranges. Although this technique
does not meet the ideal goal of requiring only a single
adjustment of each screw, it greatly simpliies and speeds
up the ilter-tuning process.

3
Basic characteristics of bandpass ilters

First, let’s review some basic information and characteris-

tics about bandpass ilters.

Bandpass ilters are commonly designed by transforming

a low-pass ilter response to one that is centered about
some new frequency. Coupled resonators, which may be
lumped LC resonators, coaxial line resonators, cavity res-
onators, or microwave waveguide resonators, are used to
create the upward shift in frequency. The terms resonator,
cavity resonator, and cavity will be used interchangeably
in this application note. More details on bandpass ilter
design can be found in Appendix A.

The center frequency of the ilter is determined by setting

the resonators. In most designs, all resonators are set
exactly to the center frequency, with the effects of adja-
cent coupling included in the calculation of the resonant
frequency.

The ilter shape, bandwidth, ripple, and return loss are

all set by the coupling factors between the resonators.
When properly tuned, the resonators have almost no
effect on the ilter shape. The only exception is that the
input and output resonators set the nominal impedance
of the ilter. Usually an input or output transformer is used
to match to a desired impedance. Of course, when the
resonators are not properly tuned, the return loss and
insertion loss will not be at the optimal levels.

Because the resonators are coupled to each other, tuning

one resonator will have the most effect on the adjacent
resonators, but it will also have some smaller effect on
the remaining resonators. The extent of the effect de-
pends on the coupling factor.

With this information in mind, we are ready to explore the

new time-domain tuning technique.

4
Time-domain response of simulated ilters

To introduce this tuning method, we will use simulations Figure 2 shows the frequency response and time re-
to examine what happens to the time-domain response sponse of the ilter. Notice the distinctive dips in the
of a bandpass ilter when it is tuned. We will start with a time-domain S11 response of the ilter. These are char-
relatively simple ilter: a ive-pole coupled resonator ilter acteristic nulls that occur if the resonators are exactly
with four coupling structures, designed for a Chebyshev tuned. The peaks between the nulls relate to the coupling
response with 0.25 dB of passband ripple. In this exam- factors of the ilter, as we will see later. Markers 1 through
ple, a ilter response will be simulated by Keysight Tech- 5 have been placed to show the characteristic dips corre-
nologies, Inc. Advanced Design System (ADS) microwave sponding to resonators 1 through 5 in the ilter. Although
design software, so that the exact values of constituent there are some dips to the left of marker 1, those are not
components are known. The frequency sweeps will part of the ilter response. Generally the peaks corre-
be performed in the simulator, and the results will be sponding to the ilter response will be much higher in
downloaded to the vector network analyzer (VNA), where magnitude than the ones in the t<0 region, which are not
the instrument’s time-domain transform application can meaningful, and usually the dip corresponding to the irst
show the effects of ilter tuning. The schematic for the resonator will occur near t=0.
ilter is shown in Figure 1.

To set up the measurement for time-domain tuning, the

frequency sweep MUST be centered at the desired center
frequency of the bandpass ilter. This is critical, since the
tuning method will tune the ilter to exactly that center
frequency. Next, the span should be set to approximately
two to ive times the expected bandwidth.

Figure 1. Schematic for ive-pole coupled resonator bandpass ilter

S21

S11

Figure 2. The frequency and time-domain response of a bandpass ilter

5
Effect of tuning resonators The lower plots show one response with only the third
resonator mistuned 2% high and another one with only
The example ilter starts out with the ideal design values, the fourth resonator mistuned 2% low. Again, it is easy
which yields the desired response since it is properly to identify which resonator is mistuned by looking for the
“tuned” by deinition. To understand the time-domain irst dip that is no longer minimized. Additional simula-
response to tuning the resonators, we will monitor the tions have shown that the characteristic dips are mini-
time-domain response while changing (mistuning) the mized only when the corresponding resonators are set to
resonator components in the simulation. Figure 3 shows their correct values. Changing the tuning in either direc-
the time-domain traces for three conditions (with the tion causes the dips to rise from the minimum values.
ideal response in the lighter trace). The upper plots show
the ilter with the second resonator mistuned 2% low in The key to this tuning technique is to adjust the resona-
frequency. Note that the irst dip has not changed, but tors until each null is as low as possible. The adjustment
the second dip is no longer minimized, and neither are will be mostly independent, although if all the resona-
the following dips. If a resonator is substantially mis- tors are far from the inal value the irst time through,
tuned (more than 1%), it will signiicantly mask the dips of adjusting a succeeding resonator may cause the null of
following resonators. Therefore, to identify the mistuned the previous resonator to rise from its minimum. If this
resonator, look for the irst dip that is no longer at a occurs, the null for the previous resonator should be
minimum. In this case, we see that mistuning resonator 2 optimized again. Once the succeeding resonator has been
causes the second null to move away from its minimum tuned and the previous one optimized, additional smaller
value. adjustment to the second resonator will have very little
effect on the dip corresponding to the irst resonator.

Resonator 2 mistuned

Resonator 2 mistuned

Ideal (tuned) response

Ideal response

Resonator 3 mistuned
Resonator 4 mistuned

Resonator 4
mistuned
Ideal Resonator 3
mistuned
Ideal

Figure 3. The response of a bandpass ilter to tuning the resonators

6
Those who are familiar with the resolution limits of In the time-domain, there is no change in the irst peak,
time-domain measurements will know that time-domain but the second peak is smaller. While it might seem that
resolution is inversely proportional to the frequency the irst peak would be associated with the irst coupling
span being measured, and they may wonder how it is factor, remember that the irst coupling factor comes
possible to resolve individual resonators in a ilter when after the irst resonator in the ilter, and we have already
the frequency span is only two to ive times the ilter’s seen that the irst dip after the irst peak is related to
bandwidth. Appendix B explains how the time-domain the irst resonator. It turns out that the irst peak can be
transform relates to bandpass ilter measurements in associated with the input coupling, which has not been
more detail. adjusted in this ilter.

One more thing to note from Figure 3 is that the S11 The reduction in height of the second peak when cou-
frequency response when resonator 2 is mistuned looks pling is increased makes sense, because increasing
almost identical to S11 response when resonator 4 is mis- the coupling means more energy is coupled to the next
tuned. This illustrates why it can be dificult to determine resonator. Thus less energy is relected, so the peak cor-
which resonator requires tuning when viewing only the responding to relected energy from that coupling should
frequency-domain measurements. decrease. Note that the following peaks are higher than
before. More energy has been coupled through the irst
Effect of tuning coupling apertures coupling aperture, so there is more energy to relect off
the remaining coupling apertures.
Although simple ilters may only allow adjustments of the
resonators, many ilters also have adjustable couplings. It is important to recognize that changing the irst cou-
To understand the effects of adjusting the coupling , we pling factor will affect the responses of all the following
will go back to our original “tuned” simulated ilter. First, peaks. This suggests that coupling factors should be
we will examine what happens when we increase the irst tuned starting with the coupling closest to the input and
coupling factor by 10%. Figure 4 shows the S11 response moving towards those in the center of the ilter. Other-
in both frequency and time domains, both before and af- wise, improperly tuned coupling near the input can mask
ter changing the coupling factor. In the frequency domain, the real response of the inner coupling factors.
we see that the ilter bandwidth is slightly wider and the
return loss has changed. This makes intuitive sense, be-
cause increasing the coupling means more energy should
pass through the ilter, resulting in a wider bandwidth.

First coupling factor First coupling factor

increased 10% increased by 10%
Ideal

Ideal

Figure 4. Effect of increasing irst coupling factor (darker trace is after adjustment)

7
Now consider what happens if we take the original Thus, we have seen that the coupling factor can be
ilter and decrease the second coupling coeficient by related to the height of the time-domain relection trace
10%. Figure 5 shows that in the frequency domain, the between each of the resonator nulls. The exact relation-
bandwidth of the ilter has been reduced slightly and ship also depends on the ratio of the ilter bandwidth to
the return loss has changed. Again, this makes sense the frequency sweep used to compute the time-domain
because decreasing the coupling means less energy transform. The wider the frequency sweep (relative to the
will pass through the ilter, corresponding to a narrower ilter’s bandwidth), the more total energy is relected, so
bandwidth. the higher the peaks.

Examining the time-domain trace, we see no change in The magnitudes of the peaks are dificult to compute
the irst 2 peaks, but the third peak is higher, consis- because changing the coupling of one stage changes the
tent with more energy being relected as a result of the height of the succeeding peaks. A detailed explanation of
decreased coupling. Since the amount of energy coupled relationship between the time-domain response and cou-
to the following resonators and apertures is reduced, the pling coeficients is beyond the scope of this application
following peaks are all lower in value. Note how well the note. Even though it may not be easy to calculate these
time-domain response separates the effects of changing peaks simply from the coupling coeficients, once the
each coupling, allowing the couplings to be individually desired values of the peaks are determined, the apertures
adjusted. In contrast, the S11 frequency response trace in may be tuned directly in the time domain. One method
Figure 4 is very similar to the one in Figure 5, so it would for determining the desired magnitudes of the peaks is by
be very dificult to know which coupling changed from using a template as described in the next section.
looking at the frequency-domain response.

2nd coupling factor

2nd coupling factor
decreased by 10%
decreased by 10%

Ideal

Ideal

Figure 5. Effect of decreasing second coupling factor (darker trace is after adjustment)

8
Practical examples of tuning ilters

Now that we have an understanding of the relationship Setting up the network analyzer
between tuning resonators or coupling apertures and the
corresponding results in the time-domain response, we It is essential to set the center frequency of the analyzer’s
are ready to to put the theory into practice. frequency sweep to be equal to the desired center fre-
quency of the ilter, since tuning the ilter in the time do-
For multi-pole cavity ilters that have ixed apertures, it main will set the ilter’s center to this frequency. Choose
is only necessary to tune for the characteristic dips in a frequency span that is 2 to 5 times the bandwidth of the
the time domain in order to achieve optimal tuning of the ilter. A span that is too narrow will not provide suficient
ilter. To tune a ilter with variable coupling coeficients, resolution to discern the individual sections of the ilter,
it is easiest to tune the coupling to a target time-domain while too wide a span will cause too much energy to be
trace or template. This target time-domain response for relected, reducing the tuning sensitivity.
any ilter type may be determined in several ways. One
method is to use a “golden” standard ilter that has the The primary parameter to be measured is S11 (input
same structure and is properly tuned for the desired ilter match). However, for time-domain responses more than
shape. This ilter can be measured and the data placed halfway through the ilter, the responses often get more
in the analyzer’s memory. Each subsequent ilter can be dificult to distinguish. Even in low-loss ilters, there can
tuned to obtain the same response. be signiicant return loss differences between the input
and output due to loss in the ilter. In addition, there is
An alternative is to create a ilter from a simulation a masking effect that tends to make relections from
tool, such as Keysight’s Advanced Design System. The couplings and resonators farther from the input or output
simulated response can be downloaded into the network appear smaller, since some of the incident energy has
analyzer and used as a template. This is a very effective been lost due to earlier relections in the device. For
approach, as there is great lexibility in choosing ilter these reasons, the most effective way to tune is to look
types. The only caution is that each real ilter has limits at both sides of the ilter at once, so a network analyzer
on the Q of the resonators and the tuning range of the with an S-parameter test set is recommended. To aid in
coupling structures and resonators. It is important to tuning, the instrument’s dual-channel mode can be used
make the attributes of the simulation consistent with the to measure the reverse return loss (S22) on a second
limitations of the structures used in the real ilters. channel. With this setup, you will tune the irst half of the
resonators and couplings using the S11 response, and
In this section, we will begin with a discussion of how to tune the remaining ones using the S22 response. Keep in
set up the network analyzer to tune bandpass ilters in mind that you need to count resonators and coupling ap-
the time domain, and then we will show three examples ertures starting from the port where the signal is entering
to illustrate how to tune both resonators and coupling the ilter for that measurement. Thus for S11, the irst dip
apertures in real ilters. would correspond to the resonator closest to the input
port of the ilter. For S22, the irst dip would correspond
to the resonator closest to the output port of the ilter.

9
For the network analyzer time-domain setup, the band- Experience has shown that it is best to begin tuning from
pass mode must be used. The start and stop times need the input/output sides and move toward the middle.
to be set so that the individual resonators can be seen. Figure 6 shows the time-domain response after the irst
For most ilters, the start time should be set slightly and ifth resonators have been tuned to obtain the lowest
before zero time, and the stop time should be set some- dips. Note that the irst resonator closest to the input
what longer than twice the group delay of the ilter. If the corresponds to the irst dip in S11, while the ifth reso-
desired bandwidth is known, the correct settings can be nator, which is the irst one when looking in the reverse
approximated by setting the start time at t=-(2/πBW) and direction, corresponds to the irst dip in S22. These
the stop time at t=(2N+1)/(πBW), where BW is the ilter’s responses are good illustrations of masking. Even though
expected bandwidth, and N is the number of ilter sec- the ifth resonator is correctly tuned, you cannot see that
tions. This should give a little extra time-domain response from looking at the S11 response. Similarly, you cannot
before the start of the ilter and after the end of the ilter see that the irst resonator is tuned by looking only at the
time response. If you are tuning using both the S11 and S22 response.
S22 responses of the ilter, you can set the stop time to a
smaller value, since you will use the S22 response to tune
the resonators that are farther out in time (and closer to CH1 S11 LOG 8 dB/ REF 0 dB
the output port).

The format to use for viewing the time-domain response

is log magnitude (dB). It may be helpful to set the top of
PRm
the screen at 0 dB. C

Example 1: Tuning resonators only

The irst example is a simple ive-pole cavity ilter with CH2 S22 LOG 8 dB/ REF 0 dB

ixed apertures, so only the resonators can be tuned

to adjust the center frequency. This ilter has a center
frequency of 2.414 GHz and a 3 dB bandwidth of 12 MHz.
The network analyzer is set up for this same center fre- PRm
quency and a span of 50 MHz. Dual channel mode is used C

to display both S11 and S22. The time-domain response

is set up to sweep from –50 ns to 250 ns.
START -50 ns STOP 250 ns

Figure 6. Time-domain response of 5-pole ilter after tuning resonators

1 and 5

10
Next, we tune the second resonator, readjusting CH1 S11 LOG 2 dB/ REF 0 dB

the irst one as needed to keep its dip minimized. Then we

go back to the output side and tune the fourth resona-
tor, readjusting the ifth one as needed. Finally, we tune
the third resonator in the middle, readjusting the second PRm
and fourth resonators as needed. It may be necessary C

to go back and readjust each of the resonators again to

ine-tune the response. Figure 7 shows the time-domain
response after the ilter has been tuned. Figures 8 and 9 CH2 S22 LOG 2 dB/ REF 0 dB
show the frequency domain relection and transmission
responses. Note that the center frequency has been set
precisely to 2.414 GHz without looking at the frequen-
cy domain while tuning. With frequency domain tuning
PRm
methods, it is often possible to tune the ilter to have the C
correct shape while the center frequency is slightly off.
The time-domain tuning method centers the ilter very
accurately.
CENTER 2 414 . 000 000 MHz SPAN 50 . 000 000 MHz

Figure 8. Final relection frequency response

CH1 S11 LOG 8 dB/ REF 0 dB

CH1 S21 LOG 2 dB/ REF 0 dB 1: -1 . 5834 dB 2 414 . 500 000 MHz

1
CH1 Markers
2 BW: 12 . 813259 MHz
PRm 3 4
cent : 2414 . 001287 MHz
C Q: 188 . 40
PRm
1_loss : -1 . 5834 dB
C

CH2 S22 LOG 8 dB/ REF 0 dB

CH2 S12 LOG 2 dB/ REF 0 dB 1: -1 . 5871 dB 2 414 . 500 000 MHz

1
CH2 Markers
2 2: -1 . 5838 dB
3 4 2. 41400 GHz
PRm 3: -2 . 9687 dB
C 2. 40759 GHz
PRm
4: -3 . 0301 dB
C 2. 42040 GHz

START -50 ns STOP 250 ns

CENTER 2 414 . 000 000 MHz SPAN 50 . 000 000 MHz

Figure 7. Time-domain response of 5-pole ilter after tuning all
resonators Figure 9. Final transmission frequency response

11
Now, what if we want to change the center frequency Example 2: Tuning to a “golden” ilter
of the ilter, for example to 2.42 GHz? We simply need
to repeat the tuning process with the analyzer’s center The second example uses a ilter that has eight poles with
frequency set to the new frequency. Figure 10 shows the seven tunable interstage coupling structures, along with
time-domain response (in bold) that results from mea- input and output coupling. In the discussion that follows,
suring the 2.414 GHz ilter after changing the network we use a “golden” ilter that was tuned by an experienced
analyzer’s center frequency to 2.42 GHz. The original engineer to obtain the desired frequency response and
time-domain response is shown in the lighter trace. It is return loss. A second, untuned test ilter, shown in Figure
clear that the resonator dips are no longer at their min- 11, was used as a test example. Figure 12 shows the
imums, so the resonators need to be retuned. Adjusting time-domain and frequency-domain plots of both ilters.
the resonators to minimize the dips again will result in a A four-parameter display mode is used to show both the
ilter tuned to a center frequency of 2.42 GHz. S11 and S22 (input and output return loss) in both the
time and frequency domains.

CH1 S11 &M LOG 8 dB/ REF 0 dB

PRm
C

CH2 S22 &M LOG 8 dB/ REF 0 dB

PRm

START -50 ns STOP 250 ns

Figure 10. Time-domain response with center frequency changed Figure 11. Eight-pole, seven-aperture ilter used for Examples 2 and 3

12
 CH1
S11 &M
LOG 5 dB/ REF 0 dB CH2
S11 &M
LOG 5 dB/ REF 0 dB
Following example 1, each of the eight resonators are
tuned, starting with the two outside resonators and
continuing until the center resonators are tuned. Each is
tuned by minimizing the response (making the deepest
dip). Again we begin by irst tuning the two outside reso-
PRm PRm
nators (numbers one and eight), looking at both S11 and
Cor Cor S22, then retuning them after the next inside resonators
(two and seven) are tuned. After the third set of reso-
nators are tuned (three and six) the second resonators
START-20 ns STOP 80 ns START 1060 . 000 MHz STOP 1380 . 000 MHz (two and seven) are re-tuned. This continues one more
CH3
S22 &M
LOG 5 dB/ REF 0 dB CH4
S22 &M
LOG 5 dB/ REF 0 dB
time for the fourth and ifth resonators. After this initial
tuning is complete, the ilter exhibits a very nice frequen-
cy response (Figure 13), but does not match the desired
response. Now it is time to tune the coupling structures.

CH1 LOG 5 dB/ REF 0 dB CH2 LOG 5 dB/ REF 0 dB

S11 &M S11 &M

START-20 ns STOP 80 ns START 1060 . 000 MHz STOP 1380 . 000 MHz

Figure 12. The response of a “golden” ilter (lighter trace) and an untuned PRm PRm

ilter of the same type (darker trace)

Cor Cor

The test ilter was pre-tuned by arranging the coupling START-20 ns STOP 80 ns START 1060 . 000 MHz STOP 1380 . 000 MHz

screws (the long screws in the picture) to about the same CH3 LOG 5 dB/ REF 0 dB CH4 LOG 5 dB/ REF 0 dB
S22 &M S22 &M
height as the “golden” ilter. Such pre-tuning is commonly
done to get the coupling apertures closer to the correct
value before beginning to tune, but it doesn’t work for
situations where a previously tuned ilter is not available.

The irst step in tuning this ilter is to assume that the

inter-stage coupling is close to correct, and adjust the
resonators to optimally tune the ilter without adjusting
the coupling screws. The setup for this ilter is a center
START-20 ns STOP 80 ns START 1060 . 000 MHz STOP 1380 . 000 MHz
frequency of 1220 MHz and a span of about 320 MHz.
The ilter bandwidth is about 80 MHz, so the time Figure 13. The response of a “golden” ilter (lighter trace) and a ilter
domain is initially set up from about -8 ns (-2/πBW) to where only the resonators are tuned (darker trace)
about 70 ns ((2N+1)/πBW). After the irst tuning, -20 ns
and 80 ns are determined to be a good choice for time
settings.

13
To tune the coupling structures, the scale can be This ilter response is nearly identical to the template
changed so that it is easy to see the peaks of the ilter. The coupling (and hence return loss) is not symmet-
time-domain response. For this example, four-parameter rical for input and output, but it is also not symmetrical
display capability is used to show the time domain in full for the “golden” ilter used as a template. If the ilters
scale with a close-up view of the peaks. With this display had no loss, the input and output match would be the
it is easier to adjust both the peaks and the dips. To tune same. The loss in the ilter causes the input match to be
the coupling, start by tuning the coupling apertures different from the output match. It is possible to tune this
that are closest to the input and output of the ilter and ilter to have exactly the same input and output match,
work towards the center, to avoid masking effects from but with a lossy ilter, one match may be improved only at
improperly tuned outer couplings. Turn the screw in to the expense of the other.
increase the coupling (reduce the peak). After each cou-
pling screw is adjusted, readjust the resonators on each Also, note that the ilter tuned in the time domain has
side to make the dip as low as possible, starting from the better return loss than the “golden” ilter, and that from
outside and working in. Figure 14 shows the result after the time-domain trace, we can see that the irst resona-
the irst pass of adjusting the coupling structures and tor is not optimally tuned according to the time-domain
resonators from the outside in. tuning process, even though the ilter has been tuned by
an expert.

CH1 LOG 5 dB/ REF 0 dB CH2 LOG 5 dB/ REF 0 dB

S11 &M S11 &M

PRm PRm

Cor Cor

START-20 ns STOP 80 ns START 1060 . 000 MHz STOP 1380 . 000 MHz

CH3 LOG 5 dB/ REF 0 dB CH4 LOG 5 dB/ REF 0 dB

S22 &M S22 &M

START-20 ns STOP 80 ns START 1060 . 000 MHz STOP 1380 . 000 MHz

Figure 14. The response of a “golden” ilter (lighter trace) and another
ilter with both couplings and resonators tuned (darker trace)

14
Example 3: Using simulated results for a Each coupling aperture and resonator is tuned to achieve
the same time response as the simulated template,
template
following the procedure described in Example 2. The last
Using a simulated ilter response to create a template for coupling structure is not tunable, but it is close enough to
tuning the ilter is the basis for the inal example of tuning. avoid distorting the overall response.
An ideal eight-pole Chebyshev ilter is simulated, and
any value can be chosen for bandwidth or ripple. For this Figure 15 shows the result with the simulated trace, and
ilter, a wider bandwidth with larger ripple was chosen. the inal tuned ilter. The results are remarkably close,
We will attempt to tune the same ilter used in example considering that the ilter was tuned only in the time do-
2 to yield this new ilter shape. Since the example ilter main, and that the simulation used capacitively-coupled
does not have adjustable input and output coupling, there lumped elements, while the real ilter had magnetically
are limits on the ilter shape that can be achieved. In this coupled distributed elements. Using this technique, virtu-
case, the bandwidth was ixed, and a return loss value ally any ilter shape that can be simulated can be used as
that yields the same value for input coupling in the time a template for a real ilter that can be easily and deter-
domain as that of the example ilter was chosen. ministically tuned, as long as the ilter elements have the
tuning lexibility. Even inexperienced tuners can follow
The frequency response of the simulation was download- this simple tuning technique because each coupling and
ed into the network analyzer and used as a template. In resonator structure can be distinguished in the time
the simulation, loss was added to the resonator struc- domain.
tures to approximate the total loss of the real ilter. This
allows the S11 and S22 from the simulation to better
match the actual time-domain response of the ilter. The
effects of loss are discussed in more detail in the next
section.

Real S11
Real S21

Simulated
Simulated S11 Real
Simulated S21

Figure 15. An example of a simulated ilter and a real ilter tuned to match the time-domain response.

15
Effects of loss in ilters

Earlier, there was a caution about considering the effects cases, tuning from the S11 side will provide better results.
of loss when using simulation to generate the time-do- If a template for a lossless ilter must be used, you may
main trace. A lossy ilter has peaks in the time-domain need to adjust the coupling apertures so they don’t com-
trace that are lower than those of a lossless ilter, and pletely match the peaks; that is, allow them to be a little
the differences in the peak levels are greater for the lower to account for the loss in the ilter.
apertures that are farther into the ilter. Therefore, tuning
a lossy ilter to a template based on the simulation of a
lossless ilter will probably result in incorrect settings of S11 Frequency Response
the coupling factors.

Trying to set the coupling apertures in the lossy ilter to

match the template of a lossless ilter requires increas- Lossy filter
ing the peaks in the time-domain trace higher than the response
proper value, so the coupling must be reduced to get
more relection. Usually it will not be possible to match Lossless
all of the peaks, especially the ones for the apertures that template
are farther into the ilter, because as we observed earlier
in Figure 5, decreasing one coupling factor will cause the
corresponding peak to increase, but the following peaks
will all decrease.

In the frequency domain, the result is that you may be S21 Frequency Response
able to achieve a similar return loss, but the ilter will
be narrower due to the higher relection, as shown in
Figure 16.
Lossless
For many cases, ilter loss may be ignored, but for high- template Lossy filter
er-order ilters, it may be necessary to include the loss of response
each resonator in the model. Further, while many simula-
tors allow loss to be applied to ilter shapes, they do not
distribute the loss throughout the ilter. Thus, to properly
account for loss, it may be necessary to create a ilter
structure using lossy resonators with discrete coupling in
between.

To match a ilter’s return loss to a lossless ilter simula- S11 Time Domain Response
tion, it may be necessary to tune a lossy ilter primarily
from the S11 (input) side. The loss of the ilter will cause
the S22 time-domain response to differ from the S22 of a
lossless simulated ilter. Since the forward relection and
transmission (S11 and S21) are more important in most Lossless template

Lossy filter

Figure 16. Comparison of lossless and lossy ilters

16
More complex ilters

Cross-coupled ilters For ilters that have asymmetrical zeros as shown in

Figure 18, the resonators that are cross-coupled do not
Finally, many ilters are more complex than the tra- have the same frequency as the other resonators, so the
ditional all-pole ilters. Cavity-resonator ilters often dips in the time-domain response that correspond to
have “cross-coupling” that effectively adds one or more these resonators will not be minimized when viewed with
transmission zeros, similar to an elliptic-ilter response. the network analyzer’s center frequency set to the ilter’s
If these zeros, which create very narrow isolation regions center frequency. Tuning the resonators to a template
in the transmission response, are close to the ilter may not yield the correct response, because there is
passband edges, they can distort the time-domain ilter more than one setting of the tuning screw that can yield
response so that it no longer shows a deep null associat- the same amplitude response. Recall that when we were
ed with the resonator near the structure that creates the discussing the time-domain response of a simulated
zero. In general, the resonators that are not cross-cou- ilter, we found that tuning the resonator either too high
pled can still be tuned using the nulling technique or too low will both cause the dip to rise up from the
described earlier. But what about the cross-coupled minimum value. The setting is unique only when you are
resonators? tuning for a null. However, we can modify the time-do-
main ilter tuning technique to account for this.
Some ilters have transmission zeros that are symmet-
rical as shown in Figure 17; the response from the zeros
can be seen on both sides of the passband. These ilters
can usually be tuned with the methods previously de-
scribed. The symmetry of the zeros keeps the cross-cou-
pled resonators at approximately the same frequency S21
as the other resonators, so all of the resonators can be
tuned close to their proper values by tuning for deep S11
nulls in the time-domain response.

Some ine-tuning may be necessary, either by tuning in

the frequency domain, or by using the techniques de-
scribed in the next section.

Figure 18. Transmission and relection responses of a ilter with asym-

metrical transmission zeros

S21 Recall that the all-pole ilters we’ve been examining have
S11
resonators that are all tuned to the same frequency, with
the effects of coupling included. We set the network
analyzer’s center frequency to that frequency, and when
we look at the relection response in the time domain,
we get nulls corresponding to each resonator when that
resonator is set to the analyzer’s center frequency. For
ilters with asymmetrical responses, if we can determine
the correct frequency of the cross-coupled resonators,
we should be able to set the analyzer’s center frequen-
Figure 17. Transmission and relection responses of a ilter with symmet-
rical transmission zeros cy to that new value, and tune the dip corresponding
to the cross-coupled resonator to its minimum value to
properly tune the resonator. Now the challenge is how
to determine the correct frequency of the cross-coupled
resonators.

17
One way is to calculate the correct frequency mathemat- 3. Tune the coupling apertures to match the time-do-
ically based on the ilter design. Simulation tools can be main response to the template values.
very useful for doing this. 4. Go back and ine-tune the cross-coupled reso-
nators and any other resonators that need to be
An alternative method is to derive the information empir- tuned to a frequency other than the ilter’s center
ically using a “golden” or template ilter. You can set up frequency.
the analyzer for a frequency sweep on one channel and
the time-domain response on another channel. Identify Duplexers
the dip in the time-domain trace corresponding to the
cross-coupled resonator. Watch the change in this dip Tuning duplexers using the time domain can be a problem
as you slowly vary the center frequency of the analyzer’s if the passbands are too close together. If the passbands
sweep. You should see the dip reach a minimum when the are separated by at least one bandwidth, and you can
analyzer’s center frequency is set to the correct frequen- set up the analyzer for a span of at least two times the
cy for that resonator. Use this information to set up a new bandwidth without seeing the other ilter, you should be
instrument state for use in tuning that particular resona- able to tune the duplexer using the techniques described
tor. All of the resonators that are not cross-coupled will in this application note. If the passbands are closer than
probably still need to be tuned with the analyzer’s center one bandwidth apart, you will get interference from the
frequency set to the ilter’s center frequency. However, response of the other ilter, and you may not be able to
depending on the coupling, a cross-coupled resonator clearly distinguish the responses due to individual reso-
may also pull the frequency of its adjacent resonators nators in the time domain. In this case, you may be able
slightly off from the ilter’s center frequency, so you may to partially tune the ilter using time domain, but you will
need to ind the correct frequencies for some of the need some other method to complete the tuning.
neighboring resonators using this method as well.
Many duplexers have common elements (one or more
In general, cross-coupling will not have much impact resonators) in the antenna path that will form part of the
on tuning the coupling apertures, since the amount of response for both the Tx-Ant and the Ant-Rx paths. To
cross-coupling tends to be light and has minimal effect tune these resonators, it may be necessary to set their
on the peaks in the time-domain response corresponding frequencies to the center frequency between the Rx and
to the coupling apertures. Tx bands, instead of tuning them to the center frequency
of either passband.
For ilters with cross-coupled resonators, the recom-
mended order of tuning is: Both cross-coupled resonator ilters and duplexers are
more advanced topics that require more research. Further
1. Start out with the coupling screws pre-tuned (to reinement of time-domain ilter tuning techniques for
match the physical settings of a “golden” ilter), as dealing with such ilters is currently under development.
described in Example 2.
2. Set the analyzer’s center frequency to the ilter’s
center frequency and tune all of the resonators to
minimize the dips to get all of the resonators close
to the proper settings, ignoring the error for the
cross-coupled resonators for now.

18
Conclusion

While various techniques to simplify the process of ilter

tuning have been tried, until now, none have succeeded
fully because coupled-resonator ilters are inherently re-
sistant to techniques that cannot account for character-
istics such as coupling interaction. The method described
in this application note goes a long way toward solving
this problem. It allows coupling apertures to be tuned
to match any ilter shape within their tuning ranges, and
resonators to be adjusted to provide a perfectly matched
ilter, with interaction immediately seen and corrected.

While a better understanding of some types of ilters such

as cross-coupled ilters is needed, this technique already
shows enough promise in allowing ilters to be tuned
easily that the current trend to automate ilter tuning on
the production line may not be needed. Alternately, this
time-domain tuning may allow automation to become
practical for the irst time. It certainly makes it easier to
train inexperienced ilter tuners quickly. These attributes
alone make the technique worthy of implementation and
further study.

19
References

Ronald N. Bracewell, "The Fourier Transform and Its

Applications," second edition, McGraw-Hill, 1978

Williams and Taylor, "Electronic Filter Design Handbook,"

Chapter 5, McGraw-Hill, 1988

Blinchikoff and Zverev, John Wiley & Sons, "Filtering in

the Time and Frequency Domains," 1976

Joel Dunsmore, “Simplify Filter Tuning in the Time

Domain,” Microwaves and RF, March 1999, pp. 68-84

20
Summary: Hints for time-domain ilter tuning

– Set the center frequency of the network analyzer equal to the – If the ilter has tunable apertures, set the coupling screws
desired center frequency for the ilter.v approximately correct; for example, by adjusting them to the
– Set the frequency span to be 2 to 5 times the bandwidth of the same physical height as those on a “golden” ilter.
ilter. – Tune the resonators irst, adjusting for deepest dips in the
– Use 201 points in the sweep for a good compromise between time-domain trace. Start with the resonators at the input and
sweep speed and resolution. output sides and work towards the middle.
– Measure S11 on one channel and S22 on the other channel. – Tuning one resonator may cause the previous resonator to
If desired, 4-parameter display can be used to view both the become slightly untuned. In this case, go back and retune the
frequency- and time-domain responses at once. Viewing both previous resonator, then optimize the current resonator again.
domains while tuning may provide better insight for optimizing – Tune the coupling apertures from the input and output sides
the ilter’s response. irst and work towards the middle. After adjusting each cou-
– Select the bandpass time-domain transform. pling screw, readjust the resonators on each side to make the
– In the time domain, choose the start limit to be about one res- dips as low as possible.
onator’s delay on the minus side; approximately t = -(2/πBW). – If the ilter has cross-coupled resonators, ine-tune the
Choose a stop limit of about 2 to 3 times the full ilter’s delay; cross-coupled resonators to their correct frequencies.
approximately t = (2N+1)/(πBW), where N is the number of il- – Repeat the tuning process at least once to ine-tune, or as
ter sections (resonators) and BW is the ilter’s 3 dB bandwidth needed to achieve desired response.
in Hz.
– Use log magnitude format (dB), and set the reference position
to 10 (top of the graticule) and the reference value to 0 dB.

21
Appendix A: Understanding basic bandpass ilter design

Many bandpass ilters are designed by starting with a This design technique results in ilters that approximately
low-pass prototype that has the desired characteristics, retain the desired ilter shape. However, many narrow-
such as passband ripple, input return loss, or stop-band band (less than 10% bandwidth relative to the center
rejection. The values for the prototype low-pass ilter ele- frequency) bandpass ilters designed with this method
ments that are necessary to obtain these characteristics end up with LC elements that cannot be realized. For
may be found in most ilter design books (see References). these narrowband ilters, an alternative design technique
This prototype low-pass ilter can be transformed into a has been developed that uses coupled resonators as the
bandpass ilter by changing the inductors and capacitors main elements. With this technique, each resonator is
into LC circuits, with the center frequency of each LC tuned to the ilter’s center frequency, with the effects of
circuit at the desired bandpass ilter center frequency. the adjacent coupling elements included. The resonator’s
Figure 19 shows an example of a prototype 3-element center frequency is calculated by treating the adjacent
low-pass ilter with the corresponding bandpass ilter coupling capacitors as though they were shorted to
structure. The equations for calculating the values of the ground, so that the capacitances will be in parallel with
ilter elements are also found in most ilter design books. the capacitance in the resonator. Figure 20 shows the
bandpass ilter from Figure 19 transformed into its
equivalent coupled-resonator structure.
R1 LI

CI CII R1 3-pole Coupled Resonator Filter

R1 C12 C23

R1 R1
CII LII CI LI CII LII CIII LIII

R1
CI LI CIII LIII Figure 20. Equivalent 3-pole coupled resonator ilter

Figure 19. 3-element prototype low-pass ilter and corresponding band- A second aspect of the coupled-resonator design tech-
pass ilter nique is that any changes in ilter type and order affect
only the coupling factor between the resonator struc-
tures. Thus the ilter shape, bandwidth, ripple, and return
loss depend only on the coupling between resonator
sections, when the resonators have been properly tuned.
These ilters retain the shape factors of the prototype
low-pass ilter.

22
A circuit simulation program has been used to model For many ilters, the coupling factor changes only slowly
the response for the mathematically simple three- with frequency, so that the center frequency of the ilter
pole Butterworth low-pass ilter. Examining this ilter’s can be changed over a substantial range without chang-
response using the time-domain transform shows that ing the basic shape of the ilter. This is because the center
the characteristic nulls in the time-domain transform are frequency of the ilter is determined only by tuning the
indeed a consequence of the ilter design. Repeating this center frequency of each resonator.
simulation with a bandpass ilter shows that the band-
pass ilter has exactly the same time-domain relection An intuitive way to think about this is that the coupling of
impulse magnitude response as the low-pass prototype. other sections is what slightly pulls the center frequen-
Since the low-pass prototype’s impulse response has the cy of different resonators to move the poles about the
characteristic dips, and this ilter has optimal circuit ele- necessary amount to produce the desired ilter response.
ment values since it has no tunable components, we can So, if a tuning technique can assure that each resonator
conclude that the dips must also be present in a properly is properly tuned, the total ilter response will be correct.
tuned bandpass ilter.
In a simple cavity resonator ilter, all resonators have the
The actual values of the elements used in the resonator same center frequency, with the effects of the resonator
are of little consequence, except that they affect the coupling included in the calculation of resonator fre-
input and output impedances, so input and output cou- quency. This frequency is also the center frequency of
pling often include an impedance transformer to ensure a the ilter. However, this is not true for ilters with trans-
50-ohm match. mission zeroes, where cross-coupling between reso-
nators will cause the cross-coupled resonators to be at
These couplings can be capacitive, which is frequently a different center frequency than the other resonators.
the case in lumped-element ilters, or inductive (some- These cross-coupled resonators may pull the adjacent
times called magnetic or B ield coupling) which is often resonators slightly off from the center frequency of the
the case in cavity-tuned ilters. In the latter, the coupling ilter as well. Thus, in tuning these ilters, we need to
structure is an opening in the wall between sections that determine the correct center frequency of the cross-cou-
permits the circulating magnetic ields to couple. These pled resonators (and possibly some of the adjacent
openings or apertures can be made adjustable by nar- resonators), and tune those resonators for that frequency,
rowing the width of the opening, which reduces coupling, while tuning the remaining resonators to the ilter’s center
or adding a shorted tuning element, such as a machine frequency. A better understanding of using time-domain
screw, which increases coupling. ilter tuning for cross-coupled ilters is still needed, and
more research is being done on this topic.

23
Appendix B: Using time-domain in the network analyzer for ilter tuning

To understand how to set up the network analyzer for Looking at measurements of bandpass ilters with a broad
time-domain ilter-tuning measurements, it is helpful to frequency sweep causes the same problem as in a low-
review some basics of the time-domain transforms. pass TDR measurement: you see a near-total relection
at the input, and almost no other relections. A normal
Normal time-domain relectometers (TDRs) are inherent- network analyzer sweep of the bandpass ilter, perhaps
ly broadband and low-pass in nature. This means they over two or three times the ilter’s bandwidth, would be a
are only useful for measuring DC-coupled circuits. They narrow sweep and was previously thought to have insuf-
cannot be used for measuring bandpass ilters, since the icient resolution to determine any characteristics of the
ilters will appear to be almost totally relective. How- ilter. However, if the measurement is properly set up, the
ever, a special mode of the network analyzer time-do- resolution limitation does not apply in measuring ilters.
main transform called bandpass mode can be used on
band-limited devices. When a ilter is examined in the time domain, each ilter
section has substantially more delay than its physical
In this mode, the center frequency of the frequency sweep size would suggest. This is because the delay of a ilter
is effectively translated to DC, and the inverse Fourier is inversely proportional to its bandwidth. The narrower
transform is applied from minus one-half of the frequency the bandwidth, the longer the delay. For multiple-section
span to plus one-half of the span. This is important when ilters, the transmission delay is approximately N/πBW,
looking at a bandpass ilter with a frequency response where BW is the bandwidth in Hz and N is the number of
that is the same as a low-pass ilter response translated sections. Each section can be considered to add about
up in frequency to the center of the bandpass ilter. 1/N of the delay. Thus the relection delay of each section
is about 2/πBW, and the total delay for relection is about
The time-domain transform represents the return loss as 2N/πBW (twice as much as the transmission delay be-
a function of length through the device under test. For cause the signal must go through the ilter and back).
time-domain transforms to be useful, they must have
enough resolution to resolve the distinguishing char- If the frequency bandwidth used to sweep the ilters is at
acteristics of the network being measured. In general, least two times the ilter bandwidth, there will be sufi-
the resolution of a transform is inversely proportional cient resolution to discern the individual sections of the
to the frequency span, although in bandpass mode the ilter. The frequency span should not be too wide, or too
resolution is reduced by half because half the span is for much of the energy will be relected, and tuning sensitivi-
negative frequencies and half for positive frequencies. ty will be reduced. Depending upon the ilter, a frequency
span of two to ive times the ilter bandwidth can be used.

24
25 | Keysight Technologies | Simpliied Filter Tuning Using Time Domain - Application Note

myKeysight For more information on Keysight

Technologies’ products, applications or
www.keysight.com/find/mykeysight
services, please contact your local Keysight
A personalized view into the information most relevant to you.
office. The complete list is available at:
Three-Year Warranty www.keysight.com/find/contactus
www.keysight.com/find/ThreeYearWarranty
Keysight’s commitment to superior product quality and lower total cost Americas
of ownership. The only test and measurement company with three-year Canada (877) 894 4414
warranty standard on all instruments, worldwide. Brazil 55 11 3351 7010
Mexico 001 800 254 2440
Keysight Assurance Plans United States (800) 829 4444
www.keysight.com/find/AssurancePlans
Up to five years of protection and no budgetary surprises to ensure your Asia Paciic
instruments are operating to specification so you can rely on accurate Australia 1 800 629 485
measurements. China 800 810 0189
Hong Kong 800 938 693
www.keysight.com/quality India 1 800 112 929
Keysight Technologies, Inc. Japan 0120 (421) 345
DEKRA Certified ISO 9001:2008 Korea 080 769 0800
Quality Management System Malaysia 1 800 888 848
Singapore 1 800 375 8100
Keysight Channel Partners Taiwan 0800 047 866
www.keysight.com/find/channelpartners Other AP Countries (65) 6375 8100
Get the best of both worlds: Keysight’s measurement expertise and product
breadth, combined with channel partner convenience. Europe & Middle East
Austria 0800 001122
Belgium 0800 58580
Finland 0800 523252
This document was formerly known as application note number 1278-8.
France 0805 980333
www.keysight.com/find/assist Germany 0800 6270999
Ireland 1800 832700
Israel 1 809 343051
Italy 800 599100
Luxembourg +32 800 58580
Netherlands 0800 0233200
Russia 8800 5009286
Spain 0800 000154
Sweden 0200 882255
Switzerland 0800 805353
Opt. 1 (DE)
Opt. 2 (FR)
Opt. 3 (IT)
United Kingdom 0800 0260637

For other unlisted countries:

www.keysight.com/find/contactus
(BP-07-10-14)

This information is subject to change without notice.

© Keysight Technologies, 2000-2014
Published in USA, July 31, 2014
5968-5328E
www.keysight.com