20 QEX

A filter with wide-band matching cuts spuriousemissions, while avoiding problems caused by

out-of-band mismatch: reduced output, instability,poor linearity or efficiency. This filter suits

a 120-W amplifier for the HF bands.

By William E. Sabin, W0IYH1

1400 Harold Dr SECedar Rapids, IA 52403sabinw@mwci.net

Diplexer Filters for an HFMOSFET Power Amplifier

1Notes appear on page 26.

Filtering at the output of a solid-state linear HF SSB poweramplifier (PA) is an important

design problem for two reasons:• Strong harmonics must be attenu-

ated to acceptable levels.• Interactions and reflections be-

tween the filter and the power am-plifier affect the power level, effi-ciency, stability and linearity.

The third harmonic, in particular, istypically only 10 to 15 dB below thefundamental. In well-balanced push-pull amplifiers, the second harmonic istypically down 40 dB or can be im-proved to that level. Fig 1 shows a spec-

trum analyzer plot of a 120-W push-pullamplifier with no filtering. At the120-W level, the FCC presently re-quires a minimum of 40 dB, and 55 dBis a desirable (and sufficient) goal forour expected needxs in the future.

In addition to reducing harmonicproducts, the filter input shouldpresent the correct 50 +j0 Ω load resis-tance at the operating frequency, forwhich the power amplifier wasdesigned. The amplifier then has thedesired output level and linearity, asnormally determined by two-toneintermodulation tests. Fig 2 shows theworst-case two-tone products of my120-W, 1.8 through 29.7 MHz home-brew MOSFET power amplifier.

Another consideration is freedomfrom oscillations and significant regen-eration. Oscillations can be free run-

ning; or they can be triggered by thedesired signal, by switching the B+ offand on or adjusting the B+ and biaslevels up and down. There are manyinsidious ways for instabilities tooccur, but the ones that we will considerinvolve the wide-band (especially stop-band) impedance that the filter inputpresents to the transistors.

Fig 3 shows a typical power-ampli-fier output driving an LC low-pass fil-ter. Outside the passband, especiallyjust outside, the input impedance of thefilter is highly reactive. This reactancecan be transformed in complicatedways by the transformers, reactancesand transmission lines that lie betweenthe transistors and the filter input.

The impedance presented to thetransistors in this stop band can be ofsuch a high magnitude and so reactive

July/Aug 1999 21

Fig 1—Spectrum-analyzer plot of a 120-W amplifier with nolow-pass filtering. Fundamental frequency is 3.5 MHz and thesweep is 1.8 to 30 MHz, 10 dB/division.

Fig 2—Two-tone intermodulation products at 29.9 MHz, 120 WPEP.

Fig 3—Typical transistor power amplifier circuit arrangement with conventional transformers and feedback.

at certain frequencies that instabilityproblems are encouraged by feedbackmechanisms inside the transistors. Atthe higher frequencies, and particu-larly with MOSFETs designed forthese higher frequencies, problems canoccur, especially when the output-loadimpedance of the filter is moved arounda 2:1 SWR circle. One specific problemis that a harmonic of an in-band signalcan land on a frequency in the stopband where a response anomaly islocated. A spectrum analyzer with atracking generator, sweeping at vari-ous power levels, gate bias and drainvoltage, is a very valuable asset fordetecting these anomalies.

Also, my experimental experiencehas verified that the harmonics, espe-cially the third, are quite oftenreflected by the filter and returned tothe transistors at such an amplitudeand phase that intermodulation dis-tortion (IMD) products are degradedin unpredictable ways, and are there-fore difficult to specify.

The methods commonly used to sta-bilize the power amplifier are negativefeedback and resistive loading, asshown in Fig 3. As one example, theconventional output transformersget hot, and are therefore a constantresistive loading on fundamental andharmonic frequency products. My pre-

ferred design uses negative feedback,but uses a transmission line outputtransformer (1:4 impedance) that dis-sipates almost no power, and virtuallyruns at room temperature. This ap-proach is simple with MOSFETsoperating at 40 to 50 V dc and at the120-W level. That is, the peak-to-peakRF voltage is a sufficiently small frac-tion of the dc supply voltage that theclass-AB operation is highly linear.This assumes that linearity has ahigher priority than power efficiencyand maximum output.

Solid-state-power-amplifier designcan be a tough game (see Note 1) unlesswe copy a well-established design from

22 QEX

Fig 6—Same spectrum as Fig 1 except: (A) at output of low-pass filter, (B) at output of high-pass filter.

Fig 5—Computer simulation of the 80-meter diplexer.

a publication or kit. My experience withhomebrew power-amplifier design ef-forts has been that if the amplifieris stable with a broadband 50-Ω load,stability and linearity with filtering in-stalled are much easier to get with thediplexer filter method described in thisarticle. It also improves confidence thattransistors will not be zapped by a largeoscillation of some kind. The restrictionis that this filter method is most fea-sible in narrow-frequency bands, suchas the HF ham bands.2 The success ofthis method also requires that the sec-ond harmonic be reduced at least 40 dBprior to filtering by a well-balanced,push-pull amplifier, which I have foundeasy to achieve in a MOSFET amplifierusing matched-pair of MRF150 SSBtransistors.

The Diplexer FilterThe diplexer filter presents a load

impedance to the power amplifier thatis essentially 50 Ω, with a return loss(RL) of better than 25 dB (in principle),from dc to well beyond 50 MHz. Fig 4shows a filter of this type for 80 meters,and Fig 5 is an idealized computersimulation, using ARRL Radio De-signer, that shows its low-pass andhigh-pass frequency responses. A morerealistic discussion is presented later.The return loss within the 80-meterband is better than 35 dB, which is quitegood. The worst return loss is in theregion of the crossover frequency(5.45 MHz). Figs 6A and 6B show thesame spectrum as Fig 1 with the fre-quency components separated. Theharmonics are dissipated in the 50-Ω“dump” resistor. We also see that thisresistor dissipates a small amount of

Fig 4—Diplexer filter for the 80-meter band with exact values.

July/Aug 1999 23

the fundamental frequency, and thatthe low-band filter output containssmall amounts of the harmonics. Thisis also seen in Fig 5, where the high-pass response is down 22 dB at 4.0 MHz,and the low-pass is down 18 dB at7.0 MHz, and 40 dB at 10.5 MHz. This7.0 MHz item points out why we need apush-pull, balanced amplifier to getadequate attenuation of the second har-monic. Once that reduction is achieved,I have found it reliable.

Fig 5 also shows why the filter ap-plies to a narrow band such as 3.5 to4.0 MHz. For example, suppose thefundamental is moved to 5.0 MHz. Theamount of fundamental lost in thedump resistor would increase quite alot. If the fundamental is reduced to3.0 MHz, the second (6.0 MHz) andthird (9.0 MHz) harmonic reductionsmay not be good enough. However, wewill see that three diplexers cover the40/30 meter bands, the 17/15 meterbands, and the 12/10 meter bands. Wecan then cover all nine HF bands withsix diplexers.

Designing the DiplexerThe diplexer is derived from the low-

pass prototype of Fig 7 (upper part),which shows a zero-resistance voltagesource. This important point is treatedin the following paragraph. The filteris a five-element, low-pass filter with aseries-inductor input (also important),a 1.0-radian-per-second (0.1592 Hz)cutoff frequency, and a 1-Ω load resis-tor. This filter may be a Butterworth,Chebyshev or Bessel type. I chose the0.1 dB Chebyshev because of its steeperroll off. These normalized prototypeelement values are easily found in vari-ous tables3,4 and are shown on the fig-ure. The high-pass filter, Fig 7’s lowerpart, is found by:

1. Replacing a series L (low pass)with a series C (high pass) whose valueis 1/L, and

2. Replacing a shunt C (low pass)with a shunt L (high pass) whosevalue is 1/C.

The high-pass prototype values arealso shown.

When the two filters of Fig 7 are com-bined as illustrated, the input resis-tance at the crossover frequency, andall other frequencies, is close to 1.0 Ω,even though each filter is 3 dB down atthe crossover point. For the Bessel orButterworth, this would be almost ex-actly 1.0 Ω. For the Chebyshev, there isa slightly larger error; but if we multi-ply each of the LPF values by some con-stant and divide the HPF values by thesame constant, the return loss at the

Fig 7—Low-pass and high-pass prototype diplexer filter.

Fig 8—Diplexer PC-board layout.

crossover can be improved several deci-bels.5 I used the number 1.005, whichwas experimentally determined bysimulation for filters whose inductorshave a Q of 160. As we see from thereturn loss in Fig 5, the two filters ter-minate each other quite well at andnear the crossover frequency becausetheir input susceptances are complexconjugates. For this reason, the trans-fer characteristic of the diplexer ispretty much insensitive to the imped-ance of the generator,6 which works inour favor for a solid-state power ampli-fier whose dynamic output impedanceis not usually known or specified.

Having identified the low-pass(LPF) and high-pass (HPF) proto-types, the next step is to find the L andC values of the final LPF section:

LK L R

2 f

CK C

2 f R

LPP LP

co

LPP LP

co

=

=

( )

( )

π

π

(Eq 1)

where LP(LP) and CP(LP) are the proto-type LPF values in Fig 7, K = 1.005, R= 50 Ω, and fco is the cutoff (crossover)frequency in Fig 5 (5.45 MHz in thisexample). For the final HPF section:

LL R

2 f K

CC

2 f K R

HPP HP

co

HPP HP

co

=

=

( )

( )

π

π

(Eq 2)

Where LP(HP) and CP(HP) are theHPF prototype values in Fig 7. Animportant decision is the choice of fco(see Fig 5). It must be such that thedesired ham band is well within thepassband of the LPF. The response ofthe HPF should be down at least 20 dBso that the dump resistor does notwaste a lot of desired signal power, eg,1.2 W for a 120-W PA. The response ofthe LPF must be adequate at the sec-ond and third harmonics. Some experi-mentation, using Radio Designersimulation, is very helpful for this.

24 QEX

Fig 9—Photo of complete diplexer filter assembly for all nineHF amateur bands.

Note that the shape of Fig 5 is the sameat any part of the HF spectrum, andonly fco moves. I will suggest values offco for the six HF diplexers. Especiallynoteworthy in Fig 5 is the way that theLPF and HPF collaborate to maintainan almost perfect 50-Ω input resis-tance within the 3.5 to 4.0 MHz band.They do this by sending to the dumpresistor the small amount of powerthat would otherwise be “reflected” bythe LPF. Another item of interest isthat the inductors—especially thefirst—in the LPF may become parallelresonant at some high frequency, butthe HPF bypasses quite well any prob-lems that this might cause for the PA.

The five-element low-pass prototypewas selected as a compromise betweencomplexity, cost and performance. Theharmonic attenuation and the powerdissipated in the dump resistor, includ-ing a small amount of fundamental,have proved to be quite reasonable, inmy opinion. However, to accommodatethe 40- and 30-meter bands with asingle filter, a few extra watts of dissi-pation on 30 meters had to be accepted.

Diplexer ConstructionFig 8 shows the PC-board layout of

a diplexer filter. It is built on a2×83/8-inch one-sided PC board. Thecopper, shown as shaded areas, is onthe opposite side, and the componentsare all on the near side. The LPF is onthe right, and the HPF is on the left.The four resistors are 200-Ω, 5-W, 5%metal-oxide types that have low L andC and excellent stability. Twentywatts is overkill for these resistors, butI believe that it was a good decisionthat will provide plenty of safety mar-

Fig 10—Composite spectrum analyzer plot, 1.8 to 30 MHz, ofLPF response, HPF response and return loss. The scale is 5dB/div, and the filters are for the 40/30-meter band.

gin. The two relays are Radio Shack275-248 with 10-A contacts and mea-sured stray C and L values that areplenty small enough for this applica-tion. They work quite well, but shouldnot be “hot-switched” to assure longlife. Space is provided for two capaci-tors in parallel at each location so thatthe correct C values can be closely ap-proximated. The HPF and LPF groundsurfaces are separated on the boardand connected to the chassis to mini-mize cross talk, which can distort thefrequency response and the RF IN re-turn loss. The very short RF IN/OUTwires and the 12-V wires are broughtout through small holes in the chassis.

Fig 9 shows the complete assemblyof the experimental model. Each filteris mounted to the chassis with two3/8-inch lengths of aluminum anglestock and #4 hardware. To preventcross talk between filters, maintainthe distance between them as shown.The band-select toggle switches canbe replaced by programmable switch-ing that sources +12 V at 60 mA. TheRF INs and RF OUTs are joined to-gether underneath the chassis withshort lengths of 50-Ω miniature coax.Each of these lengths is grounded atboth ends to provide a uniform 50-ΩZ0. The 160-meter filter is closestto the BNC connectors, and the12/10-meter filter is at the far end.That way the 12/10-meter filter doesnot have open-circuited stubs ap-pended that could cause complica-tions. Because the coaxes are in seg-ments and have small values ofstray L and C at the filter connectionpoints, I found—experimentally—that a 10-pF capacitor across each

BNC connector improved the returnloss in the 10 to 60-MHz region. Eachfilter has an insertion loss of 0.2 to0.3 dB in its ham band(s). The set ofsix identical PC boards is availablefrom FAR Circuits.7

Test and TweakAs one might expect, the actual fil-

ters to do not conform exactly to the ide-alized computer example of Fig 5 be-cause of inaccuracies in coil and capaci-tor measurements, stray Ls and Cs,lead lengths, and so on. When the strayCs of the coils—especially the HPFcoils—are included in the simulation,the return loss looks much like Fig 10.That figure is a spectrum analyzerphoto that shows the composite LPF,HPF and return loss (RL) for the 40/30-meter filter. The problem is that the Lvalues must be correct at the crossoverfrequency fco, but their “effective” val-ues at much higher frequencies are alittle larger because of their stray Cs.So to get adequate return loss—andtherefore excellent LPF and HPF be-havior—some experimental tweakingwas necessary. To do this, I used thesetup in Fig 11A, which shows a high-quality spectrum analyzer with built-in tracking generator and a dual direc-tional coupler. Connecting lead A topoint B sets a reference at the top of thescreen. Connecting lead A to point Cthen registers the return loss, in deci-bels, on a 5-dB/div scale. The procedureis to tweak the Ls and Cs to get close tothe response in Fig 10. It is importantthat the load resistor be an accurate50 Ω up to 60 MHz.

The complete and detailed datasheets for the six filters that I have

July/Aug 1999 25

Fig 12—Two devices that help to test the diplexer filters. At the top is a 150-Wdummy load that is mentioned in the text. At the bottom is a high-impedance probefor testing the LPF and HPF filter outputs.

prepared8 give the exact measured Land C values and the coil details atwhich I arrived using the tweak proce-dure. Capacitors should be measuredwith a good digital C meter (betterthan 2% accuracy), and the meter read-ings should be within 2% of the desiredvalues. The 500-V dipped mica capaci-tors, types CMO5 and CMO6, are rec-ommended because of their very lowdissipation factor 1/Q and small size.Inductor values can be closely approxi-mated using the cores and windinginstructions given in the data-sheetpackage. It will most likely be neces-sary to do some additional spreadingor compressing of the toroid coil turnsto fine tune the return-loss values. Onthe 160-, 80- and 40/30-meter filters,small values of capacitance wereplaced across the output inductor (L5in Fig 8) that improved third-harmonicreduction by 4 to 5 dB, with no degra-dations in diplexer performance.

Observing return loss is a very sen-sitive indicator of coil adjustment. Thespectrum analyzer photos in the datasheets also indicate certain importantfrequencies (as indicated by “•”) thatcan be tweaked individually with asignal generator and a receiver asshown in the setup of Fig 11B. Thismethod could use the harmonics ofa 100-kHz or 1-MHz oscillator as asignal source. Establish a referenceusing point B. Then, connect to pointC and reduce the attenuator from25 dB to 0 dB in a ham band, or from20 dB to 0 dB otherwise and work toget the same reference level to thereceiver. The receiver must be inAGC-Off mode and must be operatinglinearly in that mode. Turn down theRF gain. Work back and forth over thefrequency range until the best resultis obtained. A little experience willshow that this goes fairly smoothly.

Fig 12 shows two homebrew devicesthat aid in the testing and tweakingprocedures. At the bottom is a high-impedance probe that connects to a50-Ω load. It is used to look at the LPFor HPF output, as shown in Figs 6 and10. The seven 470-Ω resistors in series,mounted as shown in mid-air, greatlyreduce stray capacitance to ground. Atthe coax end, place a 47-Ω resistor toground, and connect the coax to a 50-Ωmeasuring device. The probe responseis down 1 dB at 30 MHz. At the top ofFig 12 is a 50-Ω, 150-W load resistorconsisting of five 250-Ω carborundumresistors in parallel. It is used in placeof the four 5-W metal-oxide resistorswhen I want to sweep from 1.8 MHz toas high as 60 MHz at the full 120-W

Fig 11—(A) Spectrum analyzer and tracking generator to measure and adjustdiplexer return loss. (B) Signal generator and SSB receiver to measure adjustdiplexer return loss.

power to check for power-amplifierinstabilities. This device has beenvery helpful, and it reassures thatthere are no hidden problems.

The responses shown on the data

sheets, especially the return loss, havebeen verified to be sufficient for excel-lent and stable performance of myMOSFET PA. There is no need for per-fectionism in order to get a satisfactory

26 QEX

result, but a return loss of 25 dB in theham bands and 20 dB at other frequen-cies is a good goal to pursue. The HPFresponse should be down 20 dB or moreat the upper end of the ham band. Ifcarried to extremes, this tweak processcan be rather tedious, and it is not atall necessary. If the gain of the poweramplifier rolls off above 30 MHz, as itshould, this helps render tweaking ofthe 12/10-meter filter (the most diffi-cult one) much less critical, as shownon its data sheet.

That has been my experience with myhomebrew efforts, after a lot of experi-mentation using more conventionaland rather frustrating approaches,and it is the reason for this article. Dyeand Granberg (see Note 2) liked thisapproach when it was feasible. Thehome-basement-lab equipment that Iused is of unusual quality for its envi-ronment, but the results can be dupli-cated with simpler gear using the dataI have provided.

line on the World Wide Web at http://www.arrl.org/catalog/ .

2N. Dye and H.Granberg, Radio FrequencyTransistors, Principles and Applications,p 151, Butterworth-Heinemann, 1993.

3A.Williams and F. Taylor, Electronic FilterDesign Handbook, third edition, McGraw-Hill, 1995.

4A. Zverev, Handbook of Filter Synthesis,Wiley, 1967

5F. Methot, “Constant Impedance Bandpassand Diplexer Filters,” RF Design Maga-zine, November 1986, pp 104-109.

6J. E. Storer, Passive Network Synthesis,McGraw-Hill 1957, pp 168-170. The con-stant-resistance diplexer is derived from amodified Darlington synthesis procedure(no transformers). The required high-passsection can be synthesized if a voltagesource is assumed. This leads to the val-ues in Fig 7 of this article. Ideally, thetransfer properties are then independentof the actual generator resistance.

7A package of six PC boards is availablefrom FAR Circuits for $30 plus $2 ship-ping. Single boards are $7.50 each. Con-tact Far Circuits, 18N640 Field Ct,Dundee, IL 60118; tel/fax 847-836-9148;farcir@ais.net ; http://www.cl.ais.net/farcir/ . Orders by mail and phone only.

8You can download a package of filter valuesand coil-winding instructions from the ARRLWeb site (http://www.arrl.org/files /qex/ ).Look for the file SABIN799.ZIP.

Notes1W. E. Sabin and E. O. Schoenike, Editors,

Single Sideband Systems and Circuits(New York: McGraw-Hill, 1995) and HFRadio Systems and Circuits, (Tucker,Georgia: Noble Publishing http://www.noblepu.com , 1998); Chapter 12 by RodBlocksome. HF Radio Systems and Cir-cuits is also available as ARRL Order No.7253. ARRL publications are availablefrom your local ARRL dealer or directlyfrom the ARRL. Mail orders to Pub SalesDept, ARRL, 225 Main St, Newington, CT06111-1494. You can call us toll-free at tel888-277-5289; fax your order to 860-594-0303; or send e-mail to pubsales@arrl.org . Check out the full ARRL publications

ConclusionsThe diplexer filter is larger, more

expensive, a little more effort tobuild, and requires more tweakingthan the usual cookbook-type LPF.The test setup in Fig 11B and the de-vices in Fig 12 are simple, inexpen-sive and very helpful in tweakingthe filters and verifying their correctoperation prior to connecting tothe PA.

The motivation for the additionalwork is that it makes clean and spuri-ous-free performance of the transistorpower amplifier a lot easier to get.