Jeff Crawford, KØZR

19735 Lovella Country Ct., Purcellville, VA 20132; [email protected]

High-Power HF
Band-Pass Filter Design
K0ZR covers some of the considerations essential to a successful
high-power filter design in an example 20 m band pass filter.

The first step in a filter design for high placement for the adjacent contest bands, these next pages is to cover some of the
power and contesting applications, in my achieve acceptably low insertion loss, great considerations essential to a successful
opinion, is to leave nothing to chance or return loss, thus reaching the point it is ready high-power design, rather than a complete
hope. At the 1,500 watt level there is little to build. This is only the first step. Consider design of a 20 m band pass filter. There are
room for mistakes. the filter in Figure 1, which was designed a considerable number of design references
Let’s consider a filter with a dissipative with the “q k” method1, 2. Its response is available, some of which we will reference
insertion loss of 0.3 dB. Simple mathematics shown in Figure 2. It has a rather clean here to facilitate your design efforts.
shows that 100 W of that 1,500 W, will be passband, good return loss, and a smaller
dissipated in the filter. This heat will age parts count than a comparable N = 5 elliptic
filter. We discover, however, that at 1,500 W, Getting Ready
the components more rapidly, may shift the
some RF currents exceed a peak value of A certain minimum tool set is necessary
filter’s return loss as a function of duty cycle,
100 A. I do not believe your printed circuit to successfully design a filter. A filter design
and may unnecessarily cause thermal stresses
board will handle that current. package such as Elsie3 is invaluable, and is
to the filter, leading to possible premature
This example shows us that we need to available for free. A more traditional manual
failure. Heat is one of the enemies in high
consider alternate filter layouts. Knowledge approach4 is possible as well. A circuit
power filter design. The filter described here
of voltages, currents, parts values, core flux simulation program such as LTspice5 or
has about 0.1 dB insertion loss.
densities, and so on, are needed to avoid SIMetrix6 is essential for performing design
As a first step, one can carefully design
a possibly costly mishap. The intent over tradeoffs, especially in ascertaining expected
a filter, paying special attention to notch

R1 C4 C5
50 Ω 750 pF 750 pF

S21–dB
+
R3
+
V1 50 Ω
V2 C2 L2 C1 L1 C3 L3
AC 1 0 8900 pF 0.2 μH 8100 pF 0.2 μH 8900 pF 0.197 μH

S11–dB

R2
1 MΩ

Third-Order Top-Coupled Filter

QX1801-Crawford01

Figure 1 — Top-coupled filter.

QEX March/April 2018 3

 0 30
1 dB Chebyshev

–10
Attenuation (dB)

25
0.5 dB Chebyshev
–20
0.1 dB Chebyshev
20
–30

Qmin
–40
15
Gaussian to 6 dB
3.0 3.2 3.4 3.6 3.8 4.0 4.2
Frequency (MHz)
QX1801-Crawford02 10

Butterworth
Figure 2 — Response of the filter in Figure 1. 5

0.5°
Bessel
0
operational voltages and currents. Test
equipment to measure S21 and S11 is essential 2 4 6 8 10 12 14
as well. Some short-cut methods can be used Filter Order (n)
on simpler filter designs, however the filter of QX1801-Crawford03
the complexity described here does not lend
itself well to such an approach. Figure 3 — Minimum Qs required in low pass filters.

Discussion
The fundamentals covered here include Table 1
filter loss, associated voltages and currents, RF currents for 1,500 W and 50 W.
and different design concepts such as
impedance scaling of the filter, and use of Component Amps Component Amps Component Amps
powerful transform techniques, the Norton L1  8.4 L5  6.7 C1 11.8
C5  6.7 L2 11.8 L6 20.6
Transform in particular. C2  3.4 C6  9.6 L3  4.6
Minimizing Loss L7  2.7 C3  1.3 C7  7.8
As you delve deeper into general filter L4  3.0 L8  4.5 C4 14
C8 12.5 L9  6.7 C9  4.5
theory, you will encounter what are termed
“elemental g-values” for Butterworth and
Chebyshev filters. These “g-values”, upon 21 MHz. Additionally, heightened Qs in
impedance and frequency scaling, evolve Another important aspect of the design resonators11 will impact the accompanying
directly into the L and C values composing critical to loss is QBP, the Q of the pass band. voltages and currents, possibly further
a low pass filter. The low pass filter can In the case of the 20 m filter described here, complicating your design and component
then be transformed into a band pass filter w2 and w1 when multiplied by 2p, are 15.75 selection. The needed Q is given in Equation
by resonating capacitors with inductors, and 10.75 MHz, respectively, with a center (3).
and inductors with capacitors. There are frequency of 13.25 MHz. QBP is then, If the low pass minimum required Q were
multiple resources that describe how this is 25 for example, the inductor Q would have
accomplished, some of which are cited7, 8,= 9 13.25 (2)
QBP = 2.65 to be higher,
herein. 15.75 − 10.75
Your intuition may lead you to believe that Any given low-pass filter has minimum
the smaller the number of filter components, Q-values that each L and C must exceed to Q = 25 × 7 = 175 (4)
the lower the loss. This is not necessarily attain the desired passband shape, see Figure
true. It can be shown theoretically that, 3-8 in Williams10. For band pass filters, these This example shows why band pass filter
minimum Q values are multiplied by QBP. component selection can be more difficult
than for low pass or high pass filters because
4.34 w2 − w1 Had the designer of this filter opted for a
L =
wQ i
∑ gi where w w (1) narrow passband, such as 13 to 15 MHz, QBP of the QBP multiplying effect. Figure 3
o
would have been ~7 making the inductors illustrates the increasing minimum Q of low
that much more difficult, if not physically pass elements as the filter order increases,
L is the filter loss, with filter family as a parameter. The QBP
Q is the unloaded Q of resonators. impossible to build see Equation (3).
impact is precisely why the passband for the
There are cases where a more complex 20 m filter is a full 5‑MHz wide even though
filter, with higher filter order, actually has = Q QLP Minimum × QBP (3)
the 20 m band is 350 kHz in width.
lower insertion loss than a simpler filter.
It comes down to the calculated g-values QBP in the 20 m case is made as low Other Factors in Loss
for each implementation per Equation (1). as possible to offset this effect while still A familiar and often used expression for
attaining the desired rejection at 7 MHz and air-core coil inductance is,

4 QEX March/April 2018

 S21
dB
Table 2

50 Ω
R2
RF voltage at 1,500 W input.

0.6457 μH

C8 = 200 pF + 200 pF + 120 pF

Load, • L3, V L5, V C5, V L7, V

C1 = 150 pF + 150 pF + 27 pF

C4 = 150 pF + 150 pF + 68 pF
0.144 Ω

C7 = 91 pF + 91 pF + 33 pF

C9 = 47 pF + 47 pF + 39 pF
25 510 220 850 390

L9

R3
50 485 270 1,100 400

Compound Capacitors:

C6 = 150 pF + 150 pF
75 520 320 1,300 430
100 560 360 1,430 460
100 at 2 kW 645 - 1,600 -

C2 = 68 pF
C9
132 pF
0.116 Ω
R8

520.6 pF
r 2n2

C8
L=

0.69245 μH
9r + 10 s

1.6717 μH
L8

0.372 Ω

10 MΩ
where r is radius in inches, n is the

R12

R17
L7
number of turns and s is the coil length.
This is just a starting point and can be
rather inaccurate as the length to diameter

68 pF

68 pF
aspect ratio changes, frequency increases,

C11
C5
and wire size is varied. There is an optimum

C7
215.65 pF
range of coil aspect ratios which, when

Figure 4 — Completed schematic for 20 m band pass filter.

0.0435 Ω
R8

chosen, will heighten the available Q of the

coils. An internet-based tool12 employed in

295 pF
C6
this design uses modified Bessel functions
0.195 μH

C10
68 pF

C3
68 pF
for wire loss, and considers the length of
L6

the coil as a function of wavelength at the

0.02 Ω
frequency of operation. The coils used in the

R13
subject design have a diameter of 0.75 inches
and theoretical Qs of approximately 400.
1.5736 μH

0.135 μH

0.135 μH

0.135 μH
L5d
0.35 Ω

0.02 Ω

0.02 Ω
Voltages and Currents
L5b

L5c
R15

R14
L4

The opening example (Figure 1) served to

R7

emphasize the all-important consideration of

voltages and currents encountered in a filter.
The best means to assess these conditions
is through use of the circuit analysis tools 1.349 μH
0.135 μH

R16
0.02 Ω
L5a
C4
365 pF

previously mentioned. They are valuable for

0.3 Ω
R11
L3

“what if analyses”, such as, “what happens

to my RF currents if the VSWR were to be
2:1 instead of a nice, perfect 1:1?” Table 1
shows RF currents and Table 2 shows the
64.8 pF

RF voltage that result from this analysis for

324 pF

53.7 pF

53.7 pF
C3a

C3b

the 20 m filter shown in the schematic of

C2

C1

Figure 4.
At different power levels the currents are,

Power
10 MΩ
R5

I Pow Lev = I1500 W

L1
0.559 μH

0.5402 μH
L2

1500
0.12 Ω

R4
0.12 Ω
R6

For different load conditions, simulations

show approximately,

I= I 50 W × 1.3
AC 1 0

25 W
I100
= I 50 W × 0.70
V2

1 MΩ

W
R10

In constructing of this filter, capacitors

+

are placed in series when higher breakdown

QX1801-Crawford04

voltages are required, and similarly,

50 Ω
R1

capacitors are placed in parallel to increase

+

the net current capacity of a given capacitor.

These capacitor combinations are annotated
in Figure 4. Research revealed that CDV-16
V1
AC 2 0

capacitors can handle 5 A continuous current

S11
dB

at HF. This serves as a guideline.

QEX March/April 2018 5

 50 Ω
Improving Filter Realizability and
Performance
Although Figure 4 is the schematic of the
filter as constructed, it began with the Elsie
based schematic shown in Figure 5. There is

284.458 pF
quite a difference. A SIMetrix evaluation of

13.25 MHz
Figure 5 revealed excessively high currents

0.507213 μH
— more than 25 A — in resonator 4, the

7
178.7 nH inductor and 807 pF capacitor.
There are some techniques that can be used
to attack this problem. The first has already
been employed by widening the passband
to 5 MHz, and thus reducing QBP. This filter
was designed around an impedance of 50 W.

8.3823 MHz
0.979276 μH

368.137 pF
What if we designed it at 100 W and used
impedance transformers at the input and

6
output? This is an available option, but not
selected for the following reasons.
Resonator 3 becomes large, elevating
concerns about self-resonance in important
parts of the stop band. Capacitor C2 takes

20.944 MHz
0.391922 μH

147.334 pF
on decreasing values making the idea of
paralleling multiple capacitors troublesome.

Figure 5 — A 5-th order elliptic filter designed using Elsie.

5

Simple L-networks at the input and output do

not have sufficient bandwidth to comfortably
handle 5 MHz. Other techniques could
ameliorate this issue, but were not elected
here.

807.249 pF
We identified the use of Norton

13.25 MHz
Transforms, and their use is now briefly
0.178731 μH

outlined. Several references13 go into greater

4

detail for the interested reader. Norton

transformations appear in several forms and
are shown in the Table of Figure A. These
transformations allow different capacitor
and inductor arrangements to be replaced
equivalently with a different capacitor and 6.6398 MHz
3.14713 μH

182.582 pF

inductor arrangement accompanied by an

3

ideal transformer. Through use of a 1:n

Norton transformation near the input of
the filter and a n:1 complimentary Norton
transformation near the filter output, an
impedance transformation can be inserted
almost anywhere within the filter. The
26.441 MHz
0.790311 μH

45.8452 pF

wide bandwidth characteristics of the ideal

transformers are retained, as is not the case
2

for an L-network matching implementation

where filter impedance scaling is used.
The Table of Figure A addresses both
capacitors and inductors. In the case of a
series capacitor, row one of the Table, the
342.787 pF

equivalency using the pi-pad connected

13.25 MHz

capacitors and ideal transformer is used. Had

0.420906 μH
1

the series inductor been used, the pi-pad of

inductors and ideal transformer would be
used. These transforms are rather simply
derived from cascaded ABCD matrices for
QX1801-Crawford05

the two circuits we wish to equate, and the

relationships derived.
50 Ω

It bears repeating that this equivalency of

the Norton transform technique is superior
over bandwidth to a narrow-band solution of
an input and output L-network. As the filter’s
center frequency increases, L-networks can

6 QEX March/April 2018

 n:1 n:1
nL
C n1

Series L or C
C n–1
2
n(n–1)L n
L 1–n n C 1–n
n2

C
n:1
L/n
C3 = nC
or

L/(1–n) 1
L n(n–1) C2 = (1–n)C C1 = n(n–1)C
A series C or L with transformer

n:1 n:1
L(1–n) Ln(n–1) C C
1–n n(n–1)
or
nL C/n

Shunt C or L

n:1
L n–1
n L 1–n
n2 n n 2
C n–1 C 1–n
or L/n nC

QX1801-CrawfordA

Figure A — Table of multiple Norton Transforms.

be utilized with more success, not being negative values are handled in the same the negative components that would result
relegated to only capacitor networks — manner traditional inductors and capacitors from the second transformation. The extra
which for 1500 W run up costs — and the are combined. Ls and Cs in Figure 4 arise from the Norton
Norton Transform. The route to the final 20 m band pass Transform application. The two additional
An additional Norton Transform14, and filter design employed the Norton technique LC resonators (a) eliminate concerns about
the one used in this design, is that shown two times at two different locations within the otherwise floating node at this point, and
in Figure 6. This transform allows for an the filter. The first was required to alleviate (b) help equalize component values while
impedance step-up or step-down for a
parallel LC network with an ideal capacitor.
Capacitors C1, C2, and C3 in Figure 6 are
L1 = LSCS/C1 L2 = LSCS/C2 LS' = LS/n
derived from the simple algebraic expressions
in the row-2 column-3 entry of the Figure A
LS L′S
Table. The “n” is the transformer turns ratio.
There is a catch to the Norton approach,
however. Upon study of the relationships in n:1
the Table of Figure A, one finds that there
are always some resulting negative valued CS C3
components. Consequently, when using the L2 C2 C1 L1
Norton technique, other components must
be present to absorb these negative valued
Ls and Cs. One of the intermediate steps in
the 20 m filter design is shown in Figure 7 QX1801-Crawford06
where, indeed, there are negative component
values. Figures 8 and 9 show that these Figure 6 — An additional Norton Transform used in this design.

QEX March/April 2018 7

 QX1801-Crawford07
4.45 μH 0.5543 μH

10.744 μH –15.195 μH 0.3575 μH –1.892 μH 1.338 μH

129.1 pF 104.2 pF

53.47 pF –37.81 pF 404 pF –30.5 pF 43.15 pF

Figure 7 — One of the intermediate steps in the 20 m filter design is shown. Figures 8 and 9 show that these negative values are handled in
the same manner traditional inductors and capacitors are combined.

1
= 0.454 μH –37.81 + 404 – 30.5 = 335.7 pF
–1 1 1
15.195 + 0.3575 – 1.892

–37.81 pF 404 pF –30.5 pF

–15.195 μH 0.3575 μH –1.892 μH QX1801-Crawford09
QX1801-Crawford08

Figure 8 — Combining inductors that have negative values. Figure 9 — Combining capacitors that have negative values.

14.063 MHz 30.000 MHz

15.937 MHz
0
–94.2711

–10
> 30 dB RL @ 7 MHz
IL – 0.1 dB
–20

–54.45030
–30
Attenuation (dB)

–40

Loss 58 dB @ 7 MHz 68 dB @ 21 MHz

–50
–54.54457

–60

–70

–80

–90
0 5 10 15 20 25 30 35
REF A
Frequency (MHz)
QX1801-Crawford10

Figure 10 — Filter performance of the circuit in Figure 4 simulated with SIMetrix.

8 QEX March/April 2018

also diminishing currents and voltages in Figure 10 shows the filter performance,
many cases. An in-depth discussion of the The Filter Assembly
simulated by SIMetrix, of the circuit in
steps taken in the design is available at www. Figure 11 shows an image of the
Figure 4, and shown assembled in Figure
k0zr.com. filter assembly. The board dimension are
11. The insertion loss is theoretically about
approximately 5.5 inches by 11 inches. The
0.1 dB, and the passband return loss for
larger current-carrying inductors are wound
A Toroidal Transformer Solution the 20 m band is better than 30 dB. Table 3
with use #12 AWG thermaleze-coated wire.
Techniques of optimizing inductor Qs, show a summary of the insertion loss and
The several smaller inductors, including the
diminishing QBP, impedance scaling the filter return loss of the filter. Figure 12 shows the
toroid, use #14 AWG wire. All capacitors
directly, or manipulation through the use of S21 performance and Figure 13 shows the
are CDV16s available through Mouser.
Norton Transforms have been offered. There measured return loss.
These capacitors should safely handle 5 to
is yet another valuable technique presented 6 A each in the HF range. When needed,
to finalize the design. This technique is used multiple values are placed in parallel for
throughout the low-power W3NQN filter current sharing. Figure 14 is an image of the
designs15. Table 3 completed filter.
While many good characteristics resulted Insertion loss and return loss.
from the Norton Transformation application,
currents in the center resonator were Frequency, Insertion Return Filter Tuning
MHz Loss, dB Loss, dB This filter topology lends itself nicely
considered too high. To reduce the current
3.5 ~ 50 to final tuning. Each parallel LC resonator
to more acceptable levels, the W3NQN 7.0 65
technique of a multi-filar toroidal transformer frequency can be easily found using,
14.0 - 30
is used. The prevailing voltages were 14.1 0.15 1
sufficiently low such that core saturation was 14.35 - 28.4 f0 = .
of no concern. 21 58 2p LC
28 57 The resonators with resonant frequencies
Further study of Figure 4 shows that a
four-winding toroid (L5a, b, c, d) is used. The
transformer is a quadrifilar transformer, four
turns, making an impedance transformation
of 16 times. The toroids are two stacked
Amidon T-130-17cores. If one pulls apart
this assembly of four coupled inductors, as
shown in Figure 4, you see that indeed this
is an autotransformer composed of the four
different windings. Because there are four
windings and the composite resonator is
tapped just above the first one, an impedance
change of 42 results, thus changing the
inductor-capacitor currents from about
30 A to about 6.5 A. To maintain the same
original LC resonant frequency, the factor of
16 multiplies the effective inductance so the
accompanying capacitor must be reduced by
a factor of 16. One additional consideration
is that if the tap-point voltage were 270 V, the
voltage on the resonating capacitor will be
four times this value, or nearly 1.1 kV. Figure 11 — The filter assembly. [Jeff Crawford, KØZR, photo]

Figure 12 — The S22 performance of the filter. Figure 13 — Measured return loss of the filter.

QEX March/April 2018 9

 Figure 14 — The completed 20 m band pass filter. [Jeff Crawford, KØZR, photo]

outside the passband form the deep notches first welding project was a 63 foot free-standing
in the stop bands. Those resonators are tower, still standing after almost 40 years. Jeff
adjusted first, then the remaining resonators has designed and built assemblies to tip-over
are adjusted to optimize passband return his crank-up tower, a base for a quarter-wave
loss. It bears emphasizing the importance 80 m vertical, and a traveling hoist system in
of tuning the filter pass band by optimizing an out building. He enjoys design and analysis
return loss, not insertion loss. of RF and microwave systems. Jeff is an active
contester in the larger world-wide contests. He
Prior to assembly, each inductor was
has 304 DXCC entities confirmed on LOTW.
paralleled with a known capacitance and Professionally, he is employed by a government
adjusted to what should be the resonant think tank, specializing in RF and microwave
frequency for the “design-to” inductor value hardware and systems.
and known capacitor. This will save you
many headaches in your assembly and tuning
process.
Notes
1
R. W. Rhea, HF Filter Design and Computer
Summary Simulation, McGraw-Hill, Inc., 1995.
The filter was designed to achieve a 2
G. Matthaei, L. Young, and E. M. T. Jones,
minimum of 50 dB stop band attenuation, Microwave Filters, Impedance-Matching
while also taking advantage of the Networks, and Coupling Structures, Artech
House, 1980.
transmission zeros at 7 and 21 MHz. In 3
Elsie filter design and analysis,
operation at the 1500 W level, only inductors www.tonnesoftware.com/elsie.html.
L1 and L2 were elevated in temperature, and 4
A. B. Williams, Electronic Filter Design
Handbook, McGraw-Hill Book Company,
only slightly, after ten minutes of constantly 1981.
calling CQ. The cores were absolutely cold. 5
LTspice, www.linear.com/designtools/
The insertion loss is difficult to measure with software/.
the Rigol spectrum analyzer and tracking
6
SIMetrix, https://www.simetrix.co.uk/.
7
G. Matthaei, et al., op. cit.
generator. The insertion loss appears to 8
R. W Rhea, op.cit., pp. 169, 172, 174.
be about 0.1 dB. I will place a fan on the 9
J. Helszajn, Synthesis of Lumped Element,
backside of the filter so as to lessen concerns Distributed and Planar Filters, McGraw Hill,
about component heating. Component cost 1990.
10
A. B. Williams, op. cit.
for this 20 m filter is approximately $100. 11
W. H. Hayt and J. E. Kemmerly, Engineering
Circuit Analysis, Chapters 13-14, McGraw-
Jeff Crawford, KØZR, was licensed in 1969 Hill Book Company, 1978.
at age 15 with the call sign WAØZRT. He
12
Single-Layer Helical Round Wire Coil
upgraded to the Amateur Extra class in 1976, Inductor Calculator, hamwaves.com/
antennas/inductance.html.
and adopted call sign KØZR. He earned a B.S. 13
R. W Rhea, op. cit., G. Matthaei, et al., op.
in Zoology from the University of Nebraska, in cit., A. B. Williams, op. cit., J. Helszajn, op.
1975, a BSEE from the University of Nebraska cit.
in 1983, and an MSEE from the University of 14
A. Klappenberger, “Narrow Bandwidth
Southern California in 1988. Jeff is a member Elliptic Bandpass Filters with Low Parts
Value Spread,” RF Design, Dec. 1992.
of ARRL, Loudoun Amateur Radio Group, 15
W3NQN filters, https://www.array
Potomac Valley Radio Club, and CWOPs, His solutions.com/w3nqn-filters-info.

10 QEX March/April 2018