EuMW 2017 Workshop Swanson Intuitive Microwave Filter Design

Intuitive Microwave
Filter Design
with EM Simulation
Daniel G. Swanson, Jr.
DGS Associates, LLC
Boulder, CO
dan@dgsboulder.com
www.dgsboulder.com
Agenda
 Part 1: High Q Cavity Combline Filter Design
 Part 2: Microstrip Combline Filter Design
 In both cases the design flow is the same.
 If you can define a resonator and a coupling
mechanism between resonators, this simple
procedure works for all lumped and distributed
filters.
This material is a subset of a full day workshop

that can be scheduled by contacting NI AWR at
awr.training@ni.com or dan@dgsboulder.com.
The full day workshop includes material on
cross-coupled filters.
Intuitive Microwave Filter Design 2
Part 1: Cavity Combline Filters
N = 6 Inline, Cover Loaded, Combline Filter
 Single combline filters and combline multiplexers

can be found in many RF / microwave systems.
 Today we will introduce a simple design flow for narrowband
combline filters using NI AWR Analyst and Microwave Office.
 This material is suitable for the non-specialist who wants a
better understanding of narrowband filter design.

Combline Filter Examples
Combline Triplexer
www.orionmicrowave.com
www.dgsboulder.com
www.allenavionics.com

Combline Resonator Loading
Tuning screw Tuning screw
30° - 60° Resonator Resonator

Typical
Lumped Loading Resonator Loading Cover Loading
 We have resonators that are less than 90° long that

we resonate with capacitance off the end.
 Resonator loading is perhaps the most flexible.
 Lumped loading is used at higher frequencies.
 Cover loading is typically used at lower frequencies.

Input / Output Coupling Options
Metal
Disk
Tapped Resonator Inductive Loop Capacitive Probe
 Tapping into the resonator works over a broad

range of bandwidths and is quite common.
 Coupling with an inductive loop near the base
of the resonator is another option.
 Using a capacitive probe is a third option.

Combline Filter Design Flow
 Estimate order of filter and stopband rejection
 Build a model of the proposed resonator:
– Compute available unloaded Q
– Estimate insertion loss
 Build Kij design curve
 Build Qex design curve
 Build a model of complete filter and apply port tuning
 Use port tuning corrections to refine filter dimensions
 Do final simulation of filter with loss:
– Verify insertion loss in passband
– Verify rejection in stopbands
Wimax Filter Example
Center Frequency: f0 = 3440 MHz
Equal Ripple BW: BW = 70 MHz (add 10 MHz for temp)
Rejection: >30 dB @ f0 +/- 80 MHz
Insertion Loss: <1 dB at band edges
Return Loss: RL > 20 dB (should add margin)
Temperature Range: -30 to +70 deg C
Power Handling: < 20 dBm
Morten Hagensen, “Narrowband Microwave Bandpass

Filter Design by Coupling Matrix Synthesis,” Guided
Wave Technology, April 26, 2009.
www.gwtsoft.com/Misc/Hagensen3.pdf

Wimax Filter Example
www.gwtsoft.com/Misc/Hagensen3.pdf
Combline Filter Asymmetry or “Skewing”

Estimating Filter Order
Rejection (dB)  RtnLoss (dB)  6
N
20 log10 ( S  S 2  1)
Rejection  Stopband Insertion Loss
RtnLoss  Passband Return Loss
Reject Bandwidth
S
Filter Bandwidth
 Any simple formula that estimates filter order, N

assumes the filter is symmetrical.
 Our 2% bandwidth filter is almost symmetrical
and this estimate is probably good enough.
 For broader band combline filters, we may want
to generate a circuit theory model to get a better
estimate of stopband performance.
Estimating Filter Order
80 MHz
160 MHz
30  25  6
N  5.33
20 log10 (2  2  1)
2

Qc of Infinitely Long Coaxial Line
For a given dominant dimension D, maximum K and hence maximum realizable

Qc is achieved when D/d = 3.6, or  r Z 0 is about 77 ohms.
QK fD Collect K data from measured filters [1]

Resonator Design: Zo
Wave port defined
on top surface
Outer: 35 x 35 mm
Inner: 10 mm dia
Height: Don’t care
 Use Analyst as a 2D cross-section solver.

 79.9 ohms is close enough to ideal Zo.
 There are also analytical equations for coax with a
square outer conductor.
Resonator Design: Freq and Qu
6 mm
20 mm
10 mm
12 mm
35 mm
Resonator length = 50 deg
 Surface of box, resonator and screw assumed to be silver plated.

 Use 80% of ideal conductivity as a starting point.
 Use measured data from filters to adjust conductivity in the future.
 The quality of silver plating is very process dependent: different
vendors can get different results on different days.

Resonator Frequency

Estimating Qu with a Field-Solver
 For any given resonator geometry we can estimate
its Qu from a lightly coupled two port measurement.
  Td 10 IL / 20
Qu 
2 (10 IL / 20  1)
where IL  insertion loss in dB

Resonator Qu
5659

Chebyshev Lowpass Prototype
Chebyshev Lowpass Prototype: 0.044 dB ripple, 20 dB return loss, 1.22 VSWR
N g0 g1 g2 g3 g4 g5 g6 g7 g8 g9 g10  g1 - gN
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989
6 1.0000 0.9958 1.4131 1.8950 1.5505 1.7272 0.8147 1.2222 8.4011
7 1.0000 1.0097 1.4368 1.9414 1.6216 1.9414 1.4368 1.0097 1.0000 10.4028
8 1.0000 1.0189 1.4518 1.9682 1.6570 2.0252 1.6104 1.7744 0.8336 1.2222 12.3447
9 1.0000 1.0252 1.4618 1.9852 1.6772 2.0662 1.6772 1.9852 1.4618 1.0252 1.0000 14.3710
 N is the lowpass or bandpass filter order.

 The gi’s are frequency and impedance scaled values for a
lowpass filter with a cutoff frequency of  = 1 radian and a
return loss of 20 dB.
 Any given passband ripple / return loss level requires a
unique table.
 Other tables are available in the literature or the gi’s can
be computed.
Midband Insertion Loss
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989
6 1.0000 0.9958 1.4131 1.8950 1.5505 1.7272 0.8147 1.2222 8.4011
7 1.0000 1.0097 1.4368 1.9414 1.6216 1.9414 1.4368 1.0097 1.0000 10.4028
8 1.0000 1.0189 1.4518 1.9682 1.6570 2.0252 1.6104 1.7744 0.8336 1.2222 12.3447
9 1.0000 1.0252 1.4618 1.9852 1.6772 2.0662 1.6772 1.9852 1.4618 1.0252 1.0000 14.3710
N
4.343   g i  f 0
4.343  8.4011 3.44
Loss ( f 0 )  i 1
  0.28 dB
f  Qu 0.08  5659
 Qu is a little optimistic, at the high end of what is possible.

 Loss will be higher at the band edges.

Dishal’s Method
 As early as 1951, Milton Dishal [2] recognized that any narrow
band, lumped element or distributed bandpass filter could be
described by three fundamental variables:
– the synchronous tuning frequency, f0
– the couplings between adjacent resonators, Kr,r+1
– the singly loaded or external Q, Qex
 The Kij set the bandwidth of the filter and the Qex sets the
return loss level.
 For any narrowband filter (<10% bandwidth) we can compute
the required Kij and Qex from the Chebyshev lowpass prototype.
 The K and Q concept is universal and can be applied to any
lumped element or distributed filter topology or technology [4,5].

Definition of Kij and Qex
f 0  g 0  g1 g 0  g1
Qex  
f 2  f1 BW
( f 2  f1 ) BW
K ij  
f0 gi  g j gi  g j
f1  f 2 f 2  f1
f0  BW 
2 f0
f1 = bandpass filter lower equal ripple frequency
f2 = bandpass filter upper equal ripple frequency
f0 = bandpass filter center frequency
BW = percentage bandwidth
gi = prototype element value for element i
Note: Equations assume Qu is infinite.

Our Filter: N = 6, BW = 2.3%
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989
6 1.0000 0.9958 1.4131 1.8950 1.5505 1.7272 0.8147 1.2222 8.4011
7 1.0000 1.0097 1.4368 1.9414 1.6216 1.9414 1.4368 1.0097 1.0000 10.4028
8 1.0000 1.0189 1.4518 1.9682 1.6570 2.0252 1.6104 1.7744 0.8336 1.2222 12.3447
9 1.0000 1.0252 1.4618 1.9852 1.6772 2.0662 1.6772 1.9852 1.4618 1.0252 1.0000 14.3710
BW 0.023
K1, 2    0.0194
g1  g 2 0.9958 1.4131
BW 0.023
K 2,3    0.0141 Note:
g 2  g3 1.4131 1.8950 Resonator impedance
BW 0.023 is not a variable.
K 3, 4    0.0134
g3  g 4 1.8950 1.5505
g 0  g1 1.0  0.9958
Qex    43.3
BW 0.023
A Very Useful Spreadsheet

Computing Iris Widths and Tap Height
 Our resonator geometry is now fixed.
 We have enough Qu to meet the insertion loss goal.
 We have goals for the Kij’s and Qex
 Now we need to compute the iris widths and
the tap height.

Basic Two Resonator Analyst Project
 Distance between resonators is fixed
– Iris width controls coupling
– Some details ignored, like corner radii
 Lossless model
– Faster
– No corrections to Kij
 Make it parametric for future re-use
 Lumped ports for tuning in
our circuit simulator
– FEM mesh is not
perfectly symmetrical
– Faster than making
geometry changes in
the EM model
Lumped port

Extracting Coupling Coefficents
Hint: if you want initial turn

on to be better, adjust tuning
Loosely couple screw lengths to minimize
with transformers. tuning capacitors.
We want to force
synchronous tuning.
At resonance:
mag (im(Y (1,1)))  0
mag (im(Y (2,2)))  0

Extracting Coupling Coefficients
-30 dB min
f 2  f1
Coupling Coefficient   0.0279
f0
Coupling Bandwidth  f 2  f1  96 MHz

Dummy Elements Around The Iris
 There are many evanescent
modes in the iris region.
 The FEM mesher uses energy
balance to refine the mesh.
 The mesh may be too coarse
in the iris region for highest
accuracy.
 Add physical detail in the iris
region to force a finer mesh.
 Only important if you are
comparing this simulation
to measured hardware.

Coupling With and Without Dummies
0.025
+2.2%
No Coupling Screw / With Dummies
No Coupling Screw
0.020
Coupling Coefficient
+3.0%
0.015
+3.1%
0.010
+3.3% How significant is 3%?
0.005
9 10 11 12 13 14 15 16 17
Iris Width (mm)

Add Coupling Screw
 We can include a coupling screw in

our model set to a nominal depth.
 A longer screw increases coupling.

Coupling vs Screw Length & Iris Width
0.05
+99.7%
Coupling Screw Len = 10 mm

Coupling Screw Len = 5 mm
0.04 Coupling Screw Len = 0
+105.9%
All With Dummies
0.03 +115.8%
+20.8%
+22.8%
0.02 +129.7%
+25.8%
+31.3%
0.01 We can achieve at least
+/- 20% tuning around a
nominal 5 mm deep screw.
0
9 10 11 12 13 14 15 16 17
Iris Width (mm)

Coupling Curve For 2 mm Thick Wall
2nd order
2 mm wall
polynomial
6 mm screw
coefficients
5 mm deep
Iris Width  5.9  476.3  K  4103  K 2


2nd order
5 mm wall
polynomial
6 mm screw
coefficients
5 mm deep
Iris Width  8.3  604.1 K  7273  K 2


Coupling Coefficients vs Iris Thickness
0.030
Iris Thickness = 5 mm
0.025 Iris Thickness = 2 mm
0.020
0.015
0.010
Coupling is a function of iris

0.005
width, height and thickness.
0
9 10 11 12 13 14 15 16 17 18 19
Iris Width (mm)

Analyst Project for Qex
 Tune to center frequency at Port 2.

 Measure reflected group delay at Port 1.
 Adjust tuning screw length at start to minimize tuning cap.
 Tap height sets the return loss level of our filter.
 Make the model parametric for future re-use.

Port Tuned Reflected Delay
Tap_Height = 5 mm
2    f (GHz)  t d (nS)
Qex 
4
3.1416  3.44  8.054

2
 43.5

Qex Data Curve Fit in MathCAD
F(x) = a + bx + cx2
Tap Height  12.865  1.412  Delay  0.054  Delay2

Qex Data Curve Fit in Excel

Analyst Model of Complete Filter
 Fully parametric model
 About 2 hours to build model
 Solve time: 3 min 53 sec
 Quad core i-7 notebook
Jan 2016

Mesh Control and Convergence

Mesh Control and Convergence

Initial Simulation – No Tuning

Initial Simulation – No Tuning
This would be closer to ideal if we had

adjusted the coupling screw lengths in
the coupling coefficient and Qex projects.

Manual Tune of Resonators
 Only tune the resonators, not the couplings.

 We used symmetry to reduce the number of variables.
 We can tune this manually, don’t need an optimizer.
Manual Tune of Resonators

Full Port Tune with EQR_OPT_MWO
CCoup < 0 Increase coupling

CCoup > 0 Decrease coupling

Full Port Tune of Analyst Model
 EQR_OPT_MWO finds a perfect

equal ripple response.
 We are meeting our design goals.

Analyst Simulation with Loss

Computing Average Qu
27.3  f (GHz)  Td (nSec)

Qu 
Loss (dB)
27.3  3.44  16.66
Qu   5047
0.31

Cavity Combline Design Summary
 Dishal’s K and Q method leads us to a simple
design flow for narrowband filters.
 We can modernize the method by using Analyst to
build the Kij and Qex design curves that we need.
 We can then build a complete model of our filter in
Analyst, port tune it and get a very good prediction
of performance.
 These virtual prototypes in Analyst avoid the time
and expense of multiple hardware prototypes.
 Experience has shown that we can rely on the
Analyst filter model.

Part 2: Narrowband Microstrip Filters
 There are many topologies
we can choose from.
 The interdigital filter has
been very popular.
 But it has some design and
fabrication challenges.
D. G. Swanson, Jr., “Narrow-Band Microwave

Filter Design,” IEEE Microwave Magazine,
vol. 8, no. 5, pp. 105-114, Oct. 2007.

Microstrip Interdigital
y
+y metal pattern
misalignment
x
1 2 3 4 5  Resonators 1, 3, 5
get longer
 Resonators 2 & 4
get shorter
 The filter is badly
mistuned
 Standard metal pattern to via alignment spec is +/- 2 mils

 Interdigital at X-band requires something like +/- 0.2 mils

Microstrip Combline
 Microstrip interdigital topology
– Has been a workhorse for many years
– Very compact in terms of wavelengths
– Very sensitive to absolute via placement
– Very sensitive to alignment of metal pattern to vias
– Y-axis misalignment rapidly detunes filter
 Microstrip combline topology
– Has not been studied in detail
– Also very compact in terms of wavelengths
– True combline requires loading capacitors and extra vias
– Microstrip combline is not pure TEM, allows longer resonator
– All resonators are grounded at the same end
– Y-axis misalignment should only shift center frequency

Conventional Combline
 Resonators are typically 50

to 60 degrees long for
moderate bandwidths.
 For octave band filters
resonators may be in the
30 degree range.
30-80  For narrow band filters
deg resonators may be in the
70 to 80 degree range.
 Some form of capacitive
loading is used to achieve
resonance.
 If the medium is pure TEM,
90 degree long combline
resonators do not couple.

Microstrip Combline
 If we want to build a microstrip combline it is tempting to adopt the

conventional topology.
 But we need an extra set of vias for the capacitive loading.
 And we need to accurately realize the capacitive loading, possibly
with an interdigital capacitor structure to get significant loading.

10% BW Microstrip Combline
15 mil alumina
L = 94 mils 25 mil wide resonators
 What if we arbitrarily throw away the capacitive top loading?

 Our first assumption is that resonators will be close to 90 degrees
long and we may not get much coupling.
 This assumes the vias are ideal short circuits, which of course
they are not.
 It also assumes a pure TEM environment, which microstrip is not.
 In fact, we can port tune this structure to be a 10% bandwidth filter.
Port Tuned Response

10% BW Microstrip Combline
 After optimization, the printed

parts of the resonators are 73
to 77 degrees long, depending
on the assumed reference
plane for the vias.
 We have some capacitive
73-77
deg
loading due to the open end
fringing.
 And we have significant
loading due to the finite
inductance of the vias.
 There is also some mutual
inductance between the vias.
 Compared to the conventional
approach, this microstrip
combline is both bottom
loaded and top loaded.

Combline Filter Fabrication
 Absolute via placement is still

a problem with the combline
 Metalized slot replaces vias
 Misalignment variables
– Slot or pattern Y-axis shift
– Slot or pattern XY rotation
 Efficient thin-film process
 Applied Thin-Film Products
www.thinfilm.com
 EM modeling is simpler
and faster without vias
Metalized slot

Microstrip Combline Example
 N = 5 Microstrip Combline
 Center Frequency: 2.44 GHz
 Bandwidth: 244 MHz (10%)
 Insertion Loss: < 2 dB
 Return Loss: 20 dB (.044 dB ripple)

Design Flow
 Estimate order of filter and stopband rejection
 Choose waveguide channel dimensions
– Distributed filters couple to the waveguide channel
 Build model of proposed resonator (with loss)
– Compute available Qu
– Estimate insertion loss
 Build Kij design curve (no loss)
 Build Qex design curve (no loss)
 Build model of complete filter and apply port tuning
 Use port tuning corrections to refine filter dimensions
 Do final run of filter with loss turned on
– Verify insertion loss in passband
– Verify rejection in stopbands
Chebyshev Lowpass Prototype
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989
6 1.0000 0.9958 1.4131 1.8950 1.5505 1.7272 0.8147 1.2222 8.4011
7 1.0000 1.0097 1.4368 1.9414 1.6216 1.9414 1.4368 1.0097 1.0000 10.4028
8 1.0000 1.0189 1.4518 1.9682 1.6570 2.0252 1.6104 1.7744 0.8336 1.2222 12.3447
9 1.0000 1.0252 1.4618 1.9852 1.6772 2.0662 1.6772 1.9852 1.4618 1.0252 1.0000 14.3710
 N is the lowpass or bandpass filter order.

 The gi’s are frequency and impedance scaled values for a
lowpass filter with a cutoff frequency of  = 1 radian and a
return loss of 20 dB.
 Any given passband ripple / return loss level requires a
unique table.
 Other tables are available in the literature or the gi’s can
be computed.
Microstrip Qu
EMSight
AWRDE V13 25mil (.635mm) thick alumina
assumed r = 9.8
50mil by 435mil
(1.27mm by 11.05mm)
150 mil
25 mil
600 mil
Vertical via metal

Microstrip Qu – From Group Delay
230

Microstrip Qu
300
H = 25
H = 20 W=50, H=25
250 H = 15
Unloaded Q
W=40, H=20
200 W=25, H=25
W=30, H=15
W=20, H=20
150
W=15, H=15
100
0.5 1.0 1.5 2.0 2.5
Width / Height

Midband Insertion Loss
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989
6 1.0000 0.9958 1.4131 1.8950 1.5505 1.7272 0.8147 1.2222 8.4011
7 1.0000 1.0097 1.4368 1.9414 1.6216 1.9414 1.4368 1.0097 1.0000 10.4028
8 1.0000 1.0189 1.4518 1.9682 1.6570 2.0252 1.6104 1.7744 0.8336 1.2222 12.3447
9 1.0000 1.0252 1.4618 1.9852 1.6772 2.0662 1.6772 1.9852 1.4618 1.0252 1.0000 14.3710
N
4.343   g i  f 0
4.343  6.4989  2.44
Loss ( f 0 )  i 1
  1.23 dB
f  Qu 0.244  230
 Loss will be higher at the band edges.

Dishal’s Method
 As early as 1951, Milton Dishal [2] recognized that any narrow
band, lumped element or distributed bandpass filter could be
described by three fundamental variables:
– the synchronous tuning frequency, f0
– the couplings between adjacent resonators, Kr,r+1
– the singly loaded or external Q, Qex
 The Kij set the bandwidth of the filter and the Qex sets the
return loss level.
 For any narrowband filter (<10% bandwidth) we can compute
the required Kij and Qex from the Chebyshev lowpass prototype.
 The K and Q concept is universal and can be applied to any
lumped element or distributed filter topology or technology [4,5].

Definition of Kij and Qex
f 0  g 0  g1 g 0  g1
Qex  
f 2  f1 BW
( f 2  f1 ) BW
K ij  
f0 gi  g j gi  g j
f1  f 2 f 2  f1
f0  BW 
2 f0
f1 = bandpass filter lower equal ripple frequency
f2 = bandpass filter upper equal ripple frequency
f0 = bandpass filter center frequency
BW = percentage bandwidth
gi = prototype element value for element i
Note: Equations assume Qu is infinite.

Our Filter: N = 5, BW = 10%
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989
6 1.0000 0.9958 1.4131 1.8950 1.5505 1.7272 0.8147 1.2222 8.4011
7 1.0000 1.0097 1.4368 1.9414 1.6216 1.9414 1.4368 1.0097 1.0000 10.4028
8 1.0000 1.0189 1.4518 1.9682 1.6570 2.0252 1.6104 1.7744 0.8336 1.2222 12.3447
9 1.0000 1.0252 1.4618 1.9852 1.6772 2.0662 1.6772 1.9852 1.4618 1.0252 1.0000 14.3710
BW 0.1
K1, 2    0.08653
g1  g 2 0.9732 1.3723
BW 0.1
K 2,3    0.06357
g 2  g3 1.8032 1.3723
g 0  g1 1.0  0.9732
Qex    9.732
BW 0.1
Note: Resonator impedance is not a variable.
A Very Useful Spreadsheet

Computing Spacings and Tap Height
 Our resonator geometry is now fixed.
 We have enough Qu to meet the insertion loss goal.
 We have goals for the Kij’s and Qex
 Now we need to compute the spacings between
resonators and the tap height.

Computing Coupling Coefficients
 Compute coupling coefficient as function
of spacing between resonators.
 Lossless model
– Faster
– No corrections to Kij
 Via ports for tuning in our circuit simulator
– MoM mesh may not be
perfectly symmetrical
– Faster than making geometry
changes in the EM model
EMSight
AWRDE V11

Computing Coupling Coefficients
EMSight
AWRDE V13
150 mil
25 mil
600 mil
Vertical via metal

Loosely couple
with transformers.
We want to force
synchronous tuning.
At resonance:
mag (im(Y (1,1)))  0
mag (im(Y (2,2)))  0

-30 dB min
f 2  f1
Coupling Coefficient   0.0607
f0
Coupling Bandwidth  f 2  f1  148 MHz

Coupling Curve: Fit in Mathcad
150 mil
Spacing  0.1137  1.2903  K  3.9600  K 2
25 mil
600 mil

Coupling Curve: Fit in Excel

Computing Qex
 Tune to center frequency at Port 2.

 Measure reflected group delay at Port 1.
 Tap height sets the return loss level
of our filter.
 Note this resonator is longer than the
resonators used to compute couplings.

Port Tuned Reflected Delay
Tap_Height = 93 mils
2    f (GHz)  t d (nS)
Qex 
4
3.1416  2.44  2.605
Qex   9.98
2

Qex Curve: Fit in Mathcad
150 mil
4
Tap Height  0.2012  0.0158  Delay  4.498 10  Delay 2
25 mil
600 mil

Qex Curve: Fit in Excel

First Iteration Geometry
S1 = 31 mils
S2 = 47 mils
L1 = 442 mils
L2 = 437 mils
Tap Height = 97 mils

Default Meshing

First Iteration Response
4350 Unknowns 150 mil

AFS Sweep
15 Frequencies 25 mil
600 mil

Arbitrary Channel Height Change
100 mil
25 mil
600 mil

Internal Ports for Port Tuning
Internal nodes
External port
Tuning element
 Low error
 Very effective for
frequency tuning
 Limited to lumped
elements by the
transformer
 How do we tune
couplings?

Impact of Internal Ports

Port Tuning With Internal Ports
Custom symbol
“zero tuning” = +20 pH

Dummy
element Mutual couplings
M ij  K ij  Li  L j
Coupled Li  0
inductor array
 1  K ij  1
Mutual couplings
tune EM circuit
couplings
Add negative offset

inductors so coupled
L’s don’t go negative.

Port Tuning with EQR_OPT_MWO
 General purpose optimizers may work
fine for low order filters, but they can
be inefficient for more complex filters.
 EQR_OPT_MWO is a dedicated
optimizer for microwave filters.
 It finds an exact equal ripple response
with a very small number of iterations.
 It communicates with Microwave Office
via the COM interface.
 It works on any Chebyshev filter that
can be defined in Microwave Office.
 We can also use it to port tune
an S-parameter file from any
EM simulator.
www.dgsboulder.com

Second Iteration: Port Tuned
X X X X X
X EM simulation frequencies

What Do The Tunings Tell Us?
 Center resonator tuning is almost perfect
(remember “zero” is +20 pH)
 The outer resonators want to be longer
 The first and last gaps want to be smaller
 The inner gaps want to be larger
 Return loss tells us the tap position wants
to move down very slightly
 The resonator and coupling tunings
will interact
 The general strategy is to go after the
largest errors at each step
Next step: Resonators 1, 2, 4, 5 each one mil longer

Move tap down one mil
Third Iteration

Fourth Iteration
 Resonator tunings are all pretty close

 The first and last gaps want to be smaller
 The inner gaps want to be larger
 Return loss is perfect
Next step: First and last gaps one mil smaller

Inner gaps one mil larger

Fifth Iteration
 Coupling corrections are small and in the

numerical noise (note opposite signs)
 Resonator tunings have shifted
 We need less than a full one mil change
in resonator length and resonator
spacing.
Next step: Fine tune open ends

Fine tune couplings

Fine Tunings
Reso 1 Reso 2 Reso 3
Add and subtract bits of metal at the open

ends to fine tune the resonators.
We have to go back and

forth a little between
frequency and coupling
adjustments.
Adding or subtracting metal at the base of

the resonators fine tunes the coupling.

Final Tuning
 If we set the tunings to zero

and see very little movement
in the response we are done.
 Next step is to remove the
tuning ports and do a two
port analysis of the filter.

Two Port EM Simulation

Two Port EM Simulation

N = 7 Combline at X-Band
Note “patch tuning”
Via metal to ground
• No restriction on gaps with slot ground instead of vias

• Gaps will tend to be tighter than equivalent interdigital

Sonnet em Simulation
• EM simulation is simple and efficient

• Ignore air dielectric in the slot
Via metal to ground
• Ignore taper of slot

Nominal Response
Sonnet em Simulation
(loss and metal thickness)

Nominal +/- 1 mil Pattern Shift
• Mostly freq shift, some BW change

• Very little return loss detuning
• Critical alignment:
- Metal pattern to slot edge
- XY skew of pattern to edge

ATP Sample of Five
www.thinfilm.com

ATP Sample of Five

ATP #1, #3, #5, Sonnet Nom

ATP #1, #3, #5, Sonnet Nom
• Includes loss of IMC launchers

• Measure metal thickness
• Adjust conductivity

ATP #2, #4, Sonnet Nom +1 mil

Summary
 Dishal’s K and Q method leads us to a simple design flow
for narrowband filters.
 We can modernize the method by using EMSight to build
the Kij and Qex design curves that we need.
 We can then build a complete model of our filter in EMSight,
port tune it and get a very good prediction of performance.
 These virtual prototypes in our EM simulator avoid the
time and expense of multiple hardware prototypes.
 Experience has shown that we can rely on the
EM simulator models.

References
[1] R. Levy, R. Snyder and G. Matthaei, “Design of Microwave Filters,”
IEEE Trans. Microwave Theory Tech., vol. MTT-50, pp. 783-793, March
2002.
[2] M. Dishal, “Alignment and adjustment of synchronously tuned multiple
resonate circuit filters,” Proc IRE, vol. 30, pp. 1448-1455, Nov. 1951.
[3] M. Dishal, “A simple design procedure for small percentage bandwidth
round-rod interdigital filters, IEEE Trans. Microwave Theory Tech., vol.
MTT-13, pp. 696-698, Sept. 1965.
[4] J. Wong, “Microstrip tapped-line filter design,” IEEE Trans. Microwave
Theory Tech., vol. MTT-27, pp. 44-50, Jan. 1979.
[5] D. G. Swanson, Jr., “Narrow-Band Microwave Filter Design,” IEEE
Microwave Magazine, vol. 8, no. 5, pp. 105-114, Oct. 2007.
[6] D. G. Swanson, Jr., “Corrections to “Narrow-Band Microwave Filter
Design, “ IEEE Microwave Magazine, vol. 9, no. 1, p. 116, Feb. 2008.

EuMW 2017 Workshop Swanson Intuitive Microwave Filter Design

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

EuMW 2017 Workshop Swanson Intuitive Microwave Filter Design

Uploaded by

Copyright:

Available Formats

Intuitive Microwave

This material is a subset of a full day workshop

 Single combline filters and combline multiplexers

Intuitive Microwave Filter Design 3

Intuitive Microwave Filter Design 4

30° - 60° Resonator Resonator

Lumped Loading Resonator Loading Cover Loading

 We have resonators that are less than 90° long that

Intuitive Microwave Filter Design 5

Tapped Resonator Inductive Loop Capacitive Probe

 Tapping into the resonator works over a broad

Intuitive Microwave Filter Design 6

Morten Hagensen, “Narrowband Microwave Bandpass

Intuitive Microwave Filter Design 8

Intuitive Microwave Filter Design 10

 Any simple formula that estimates filter order, N

Intuitive Microwave Filter Design 12

For a given dominant dimension D, maximum K and hence maximum realizable

QK fD Collect K data from measured filters [1]

Intuitive Microwave Filter Design 13

 Use Analyst as a 2D cross-section solver.

Resonator length = 50 deg

 Surface of box, resonator and screw assumed to be silver plated.

Intuitive Microwave Filter Design 15

Intuitive Microwave Filter Design 16

Intuitive Microwave Filter Design 17

Intuitive Microwave Filter Design 18

 N is the lowpass or bandpass filter order.

 Qu is a little optimistic, at the high end of what is possible.

Intuitive Microwave Filter Design 20

Intuitive Microwave Filter Design 21

Intuitive Microwave Filter Design 22

Intuitive Microwave Filter Design 24

Intuitive Microwave Filter Design 25

Intuitive Microwave Filter Design 26

Hint: if you want initial turn

Intuitive Microwave Filter Design 27

Intuitive Microwave Filter Design 28

Intuitive Microwave Filter Design 29

Iris Width (mm)

 We can include a coupling screw in

Intuitive Microwave Filter Design 31

Coupling Screw Len = 10 mm

All With Dummies

Iris Width (mm)

Iris Width  5.9  476.3  K  4103  K 2

Intuitive Microwave Filter Design 33

Intuitive Microwave Filter Design 34

Iris Width  8.3  604.1 K  7273  K 2

Intuitive Microwave Filter Design 35

Intuitive Microwave Filter Design 36

Coupling is a function of iris

Iris Width (mm)

 Tune to center frequency at Port 2.

Intuitive Microwave Filter Design 38

Intuitive Microwave Filter Design 39

Tap Height  12.865  1.412  Delay  0.054  Delay2

Intuitive Microwave Filter Design 40

Intuitive Microwave Filter Design 41

Intuitive Microwave Filter Design 42

Intuitive Microwave Filter Design 43

Intuitive Microwave Filter Design 44

Intuitive Microwave Filter Design 45

This would be closer to ideal if we had