Jan/Feb 1999 1

R

L

R

A

David Sumner, K1ZZPublisherDoug Smith, KF6DX/7EditorRobert Schetgen, KU7GManaging EditorLori WeinbergAssistant EditorZack Lau, W1VTContributing Editor

Production DepartmentMark J. Wilson, K1ROPublications ManagerMichelle Bloom, WB1ENTProduction SupervisorSue FaganGraphic Design SupervisorDavid Pingree, N1NASTechnical IllustratorJoe SheaProduction Assistant

Advertising Information Contact:Brad Thomas, KC1EX, Advertising Manager

American Radio Relay League860-594-0207 direct860-594-0200 ARRL860-594-0259 fax

Circulation DepartmentDebra Jahnke, ManagerKathy Capodicasa, N1GZO, Deputy ManagerCathy Stepina, QEX Circulation

Offices225 Main St, Newington, CT 06111-1494 USATelephone: 860-594-0200Telex: 650215-5052 MCIFax: 860-594-0259 (24 hour direct line)e-mail:qex@arrl.org

Subscription rate for 6 issues:

In the US: ARRL Member $22,nonmember $34;

US, Canada and Mexico by First Class Mail:ARRL Member $35, nonmember $47;

Elsewhere by Surface Mail (4-8 week delivery):ARRL Member $27,nonmember $39;

Elsewhere by Airmail: ARRL Member $55,nonmember $67.

Members are asked to include their membershipcontrol number or a label from their QST wrapperwhen applying.

Jan/Feb 1999 QEX Advertising IndexAlpha Delta Communications: Cov IVAmerican Radio Relay League: 30, 60, 64, Cov IIICCT, Inc: 45Communications Specialists, Inc: 57

Crestone Technical Books: Cov IIHAL Communications Corp: 63SHF Microwave Parts Company: 8Tucson Amateur Packet Radio Corp: 63Z Domain Technologies, Inc: 41

Paul Wade, W1GHZ, shows howto build interdigital filters forthree bands from waveguideand tubing. See page 3.

Features3 Waveguide Interdigital Filters

By Paul Wade, W1GHZ (N1BWT)

9 Tune SSB AutomaticallyBy Robert Dick

19 The Need for Standard Application-ProgrammingInterfaces (APIs) in Amateur RadioBy Lawrence G. Dobranski, VA3LGD/VE3TVV

22 A Pair of 3CX800s for 6 MetersBy Dick Hanson, K5AND

28 A Compact Mobile TunerBy Patrick Wintheiser, W0OPW

31 Experiments with Phase-Noise MeasurementBy Jos F. M. van der List, PA0JOZ

42 A Temperature-Compensated DDS VFOBy Curtis Preuss, WB2V

46 Practical Application of Wind-Load Standards toYagi Antennas: Part 1By Stuart E. Bonney, K5PB

51 A Handy Coil WinderBy Bob Dildine, 7J1AFR/W6SFH

54 The Cheap SweepBy C. A. Hoover, K0VXM

56 No-Tune Transverter-Spur IndentificationBy Michael McKay, W4AZR

Columns62 Next Issue in QEX58 RF By Zack Lau, W1VT

In order to insure prompt delivery, we ask thatyou periodically check the address informationon your mailing label. If you find any inaccura-cies, please contact the Circulation Departmentimmediately. Thank you for your assistance.

QEX (ISSN: 0886-8093) is published bimonthlyin January 99, March 99, May 99, July 99,September 99, and November 99 by theAmerican Radio Relay League, 225 MainStreet, Newington CT 06111-1494. Subscrip-tion rate for 6 issues to ARRL members is $22;nonmembers $34. Other rates are listed below.Periodicals postage paid at Hartford CT and atadditional mailing offices.POSTMASTER: Form 3579 requested.Send address changes to: QEX, 225 Main St,Newington CT, 06111-1494Issue No. 192

Copyright © 1998 by the American Radio RelayLeague Inc. Material may be excerpted from QEXwithout prior permission provided that the originalcontributor is credited, and QEX is identified asthe source.

61 Letters to the Editor

2 QEX

THE AMERICAN RADIORELAY LEAGUEThe American Radio Relay League, Inc, is anoncommercial association of radio amateurs,organized for the promotion of interests in AmateurRadio communication and experimentation, forthe establishment of networks to providecommunications in the event of disasters or otheremergencies, for the advancement of radio artand of the public welfare, for the representationof the radio amateur in legislative matters, andfor the maintenance of fraternalism and a highstandard of conduct.

ARRL is an incorporated association withoutcapital stock chartered under the laws of thestate of Connecticut, and is an exempt organiza-tion under Section 501(c)(3) of the InternalRevenue Code of 1986. Its affairs are governedby a Board of Directors, whose voting membersare elected every two years by the generalmembership. The officers are elected orappointed by the Directors. The League isnoncommercial, and no one who could gainfinancially from the shaping of its affairs iseligible for membership on its Board.

“Of, by, and for the radio amateur, ”ARRLnumbers within its ranks the vast majority ofactive amateurs in the nation and has a proudhistory of achievement as the standard-bearer inamateur affairs.

A bona fide interest in Amateur Radio is theonly essential qualification of membership; anAmateur Radio license is not a prerequisite,although full voting membership is granted onlyto licensed amateurs in the US.

Membership inquiries and general corres-pondence should be addressed to theadministrative headquarters at 225 Main Street,Newington, CT 06111 USA.

Telephone: 860-594-0200Telex: 650215-5052 MCIMCIMAIL (electronic mail system) ID: 215-5052FAX: 860-594-0259 (24-hour direct line)

Officers

President: RODNEY STAFFORD, W6ROD5155 Shadow Estates, San Jose, CA 95135

Executive Vice President: DAVID SUMNER, K1ZZ

Purpose of QEX:

1) provide a medium for the exchange of ideasand information between Amateur Radioexperimenters

2) document advanced technical work in theAmateur Radio field

3) support efforts to advance the state of theAmateur Radio art

All correspondence concerning QEX should beaddressed to the American Radio Relay League,225 Main Street, Newington, CT 06111 USA.Envelopes containing manuscripts and corre-spondence for publication in QEX should bemarked: Editor, QEX.

Both theoretical and practical technical articlesare welcomed. Manuscripts should be typed anddoubled spaced. Please use the standard ARRLabbreviations found in recent editions of TheARRL Handbook. Photos should be glossy, blackand white positive prints of good definition andcontrast, and should be the same size or largerthan the size that is to appear in QEX.

Any opinions expressed in QEX are those ofthe authors, not necessarily those of the editor orthe League. While we attempt to ensure that allarticles are technically valid, authors areexpected to defend their own material. Productsmentioned in the text are included for yourinformation; no endorsement is implied. Theinformation is believed to be correct, but readersare cautioned to verify availability of the productbefore sending money to the vendor.

R

L

R

A Empirically Speaking

We amateurs have often had to beconcerned about our image with ourelected officials in Congress, ourfriends in the United Nations and thepublic. Concern has peaked aroundthose times when commercial or gov-ernmental interests have made agrab for some of our spectrum. We’vehad to continually refresh society’smemory about our history of service,especially our service during naturaldisasters and other emergencies.Perhaps our best publicity has comein times of need.

Satellite and cellular services, how-ever, are rapidly encroaching on thisarea, one of our primary raisons d’etre.In addition, digital-audio broadcast-ing is right around the corner. Inter-national broadcasters may soon needmore of our HF bandwidth. It mightnot be long before more folks are ask-ing: “What have you done for uslately?” We must increase the empha-sis on other unique aspects of our avo-cation. Of course, these will stillinclude public service, but our abilityto advance our art, to make the bestuse of our spectrum and to bring alongnew interest in purely technical pur-suits will be pivotal. Among otherthings, those are what fill the pages ofQEX.

I’ve heard from hams—some whoare ardent equipment builders orsoftware writers—that they want thebest forum for their constructionprojects or utility programs. Othersare saying they want more projects tobuild, but don’t want to bathe inhigher mathematics to do so. Stillothers are chiefly interested in keep-ing informed about new theory andtechnology. I think that many of usare intermediate to those views, butQEX will continue to have room forthem all. What is your perception?

W4AZR, examines things that hap-pen around those “no-tune” trans-verters, such as birdies and spuriousresponses, which may inspire you toget out that drill and iron.

Patrick Wintheiser, W0OPW, givesus a “Compact Mobile Tuner,” an ex-cellent project for getting more of theband from your antenna, whetheryou’re mobile or not. Winding coilsmay be good therapy for some of us,but Bob Dildine, 7J1AFR/W6SFH, isback with a way to reduce fatigue andincrease repeatability.

C. A. Hoover, K0VXM, shows howto put together his “Cheap Sweep”with parts gleaned from a junk box.This neat construction project helpsto tune UHF filters. For the moreambitious builder, Richard Hanson,K5AND, helps you assemble a pairof 3CX800s to get ready for more6-meter openings. Heck, with thisunit, make your own openings.

Curtis Preuss, WB2V, conquers thefrequency-versus-temperature prob-lem with his DDS compensationmethod. I have little doubt this tech-nique will find its way into manydesigns, if it hasn’t already. Fromacross the pond comes an article of in-terest especially to receiver designers.Designs with poor low-order IMD per-formance normally pack up in a hurrythere. Jos van der List, PA0JOZ, ana-lyzes phase-noise effects—from bothreceivers and transmitters—and pro-vides criteria and a fixture for mea-suring them accurately. This articlealso appears in two issues of theDutch magazine Electron.

Robert Dick shares his research onhow to “Tune SSB Automatically.”Yes, Virginia, it can be done—try it!Larry Dobranski, VA3LGD, reasonsthat it needn’t be so difficult to addsupport for new computer-controlledequipment every time something dif-ferent hits the market. I suspectthat’s right, and this is a subject wor-thy of some discussion. Stu Bonney,K5PB, looks at how well an antennawill survive that gale rolling in fromthe Gulf in the first part of his two.Zack Lau, W1VT, offers MicrowavePA tips in his RF column.—73, DougSmith, KF6DX, kf6dx@ arrl.org.

In This QEXOkay builders, put the coffee on

and warm up the iron—it’s construc-tion time! Paul Wade, N1BWT, hasfound one of the magical combina-tions of readily available parts that,together with a bit of drilling and sol-dering, produce useful results for youmicrowave fans. Michael McKay,

Jan/Feb 1999 3

Build sturdy, predictable microwave filters fromwaveguide. Here are designs for three bands:

1296, 2304 and 3456 MHz.

By Paul Wade, W1GHZ (N1BWT)

161 Center RdShirley, MA 01414wade@tiac.net

WaveguideInterdigital Filters

1Notes appear on page 8.

Most microwave transverters,especially the “no-tune”variety, need some additional

filtering to operate in locations with“RF pollution”—accessible mountain-tops are notoriously bad environments.Waveguide post filters provide supe-rior performance at 10 GHz1 and5760 MHz,2 but become large andheavy at lower frequencies. Inter-digital filters are excellent performersat the lower microwave frequencies,3

but the usual construction techniquesrequire some machining, mostlytedious tapping of threads in many

holes. One of the beauties of waveguidefilters is that the basic structure is ac-curately defined by the waveguide, soconstruction requires only drilling andsoldering. Since surplus waveguide isreasonably plentiful, I wondered if itcould be used to build interdigital fil-ters for the lower microwave bands. Aswe shall see, my experiments werequite successful.

Interdigital FiltersThe basic structure of an inter-

digital filter, shown in Fig 1, is a groupof coupled resonators in a metal hous-ing. Each resonator is an electrical λ/4long, but physically shortened bycapacitance at the open end. The reso-nators are interdigitated, with the po-sition of the open ends of the resonatorsalternating as depicted in Fig 1. (A

similar filter with all resonatorsaligned in the same direction is called acomb-line filter.) The coupling betweenresonators is controlled by their sepa-ration. Several methods are commonlyused to make input and output connec-tions, but a simple one is to use taps onthe input and output resonators.

The starting dimension for aninterdigital filter is the width of thehousing, which should be λ/4 at theoperating frequency. All of theother dimensions are interrelated—a change in one affects others—so thatempirical design of a filter would bedifficult and frustrating. Fortunately,computer programs are available todesign interdigital filters. A BASICprogram,4 by N6JH, appears in hamradio magazine. I translated thisinto PASCAL and compiled it. My

4 QEX

version, INTFIL.EXE, is available fordownloading at http://www.qsl.net/~n1bwt/intfil.zip. One 1296 MHzfilter that I built using this programwas carefully measured using anautomatic network analyzer and foundto match the predicted performance al-most perfectly with no tuning. Thisgave me confidence in the accuracy ofthe program.

Filter DesignThe first part of filter design is the

same for all types of filters—calcula-tion of coupling coefficients and otherparameters to achieve the desired per-formance. These are tabulated in TheARRL Handbook5 and other referencebooks6 for the most common types offilters: the Butterworth (maximally-flat) and the Chebyshev response,which trades some passband ripple(amplitude variation) for somewhatsteeper skirts at the passband edges.The tabulated parameters, gmn, are fora normalized prototype filter, so thatfurther calculations are required tofind actual component values for adesired frequency and impedance.

The second part of filter design is toconvert the normalized parametersinto component values or physical di-mensions. The calculations are quitetedious, so graphical solutions wereoften published7 before computerswere commonly available. Now thesecalculations are easily performed on aPC, allowing us to evaluate multiplefilter designs before choosing one tobuild.

Design of an interdigital filter be-gins with the choice of a requiredbandwidth. Simple filter programssuch as INTFIL are only reliable forbandwidths between about 1% and10% of the center frequency, and very-narrow-bandwidth filters are lossyand require tight tolerances in con-struction. Therefore, a 3% to 5% band-width is recommended as a good start-ing point. The next step is to decidehow steeply the skirts roll off at thepassband edges. For example, steeperskirts are required to reject an image

Table 1: Waveguide Dimensions for Interdigital Filters

Frequency

Waveguide Wide Dimension (λ/4, MHz)

WR-340 3.4″ 868WR-284 2.84″ 1039WR-229 2.29″ 1289WR-187 1.872″ 1577WR-159 1.59″ 1857WR-137 1.372″ 2152WR-112 1.12″ 2636WR-90 0.90″ 3280WR-75 0.75″ 3937WR-62 0.622″ 4747WR-50 0.51″ 5789Fig 1—Interdigital filter cross-section

sketch.

Fig 2—Performance of 1296-MHz filter—WR-229 waveguide.

Fig 3—Performance of 2304-MHz filter—WR-137 waveguide.

Jan/Feb 1999 5

Fig 4—Performance of 3456-MHz filter—WR-90 waveguide.

close to the operating frequency. Gen-erally, filters with steeper skirts re-quire more resonators and have moreloss, so a compromise may be in order.It is possible to calculate the numberof resonators required, but a few trialdesigns on the computer should yieldthe same result and provide some in-sight as well.

Waveguide Interdigital FiltersNow let’s design a filter in an avail-

able waveguide. As mentioned above,we should start with the width of theenclosure at λ/4 at the center fre-quency. This would be the large insidedimension of a waveguide used as theenclosure. Table 1 lists the insidedimensions for some commonly avail-able waveguides and the frequenciesfor which the wide dimensions are λ/4.The large dimension for WR-229 isλ/4 at 1289 MHz, so it is a logicalmaterial for a 1296 MHz filter. Simplydesigning a filter for the 1289 MHzcenter frequency with enough band-width to include 1296 MHz does thetrick. I chose a 50 MHz bandwidth andused INTFIL to calculate the rest ofthe dimensions for a resonator diam-eter of 3/8 inch. Since WR-229 is beingscrapped as 4 GHz telephone micro-wave links are decommissioned, I wasable to find all I could carry at a fleamarket for $5.

Once the filter was assembled, Ifound that the bandpass was slightlyabove 1296 MHz, a little higher thanthe design. This was easy to fix, how-ever: I drilled and tapped holes for

Fig 5—Waveguide interdigital filters for 1296, 2304 and 3456MHz.

tuning screws opposite the open endsof the resonators and inserted screwsto add capacitance and lower thefrequency. On the other hand if thefrequency ended up a bit low, then itwould be necessary to shorten theresonators slightly.

I first adjusted the screws for mini-mum insertion loss at 1296 MHz, thenreadjusted them for minimum SWR atboth ends. The second adjustment is abit more involved since each adjust-ment affects both ends and a few re-versals were needed. The final pass-band, shown in Fig 2, is about 58 MHzwide with an insertion loss less than0.5 dB at 1296 MHz.

Tuning is straightforward for thesefilters, with moderate bandwidth anda reasonable number of resonators.However, filters with very narrow orwide bandwidths, or with many reso-nators, require a more complex tuningprocedure. Dishal’s procedure8, 9 al-lows the tuning of one resonator at atime.

For other bands, no waveguide ex-actly matches λ/4, but there are somegood candidates that fall within about10% of the desired frequency. Theubiquitous X-band waveguide, WR-90,is close to λ/4 at 3456 MHz, whileWR-137 (used in 6-GHz microwavelinks) is close to 2304 MHz. For thesetwo, simply making a wide filter is notgood enough. The bandpass wouldinclude the commonly used LO fre-quencies for a 144-MHz IF. Thus, weneed a design procedure that can movethe center frequency slightly.

The design procedure that I usemakes two similar designs, one at thedesired frequency and one at the λ/4frequency waveguide, using the samepercentage bandwidth (bandwidth÷center frequency) for both designs.Using the same percentage bandwidthresults in two designs differing only inthe resonator lengths and tap posi-tions, and the difference is small be-cause the frequencies are close to-gether. Since the higher-frequencydesign also calls for the λ/4 distance tobe shorter, making the resonators thisshort would result in less capacitanceand an actual resonant frequencyhigher than desired. My compromiseis to split the difference between thetwo design lengths and make the reso-nator lengths halfway between thetwo designs.

I followed the above design proce-dure for two more filters, one for3456 MHz in WR-90 waveguide with108 MHz bandwidth and the other for2304 MHz in WR-137 waveguide with75 MHz bandwidth. Each design usesfour resonators with a Butterworthresponse. After fabrication, both fil-ters had passbands that included thedesign frequency, as shown in Figs 3and 4. Thus, they are usable with nofurther tuning. Any elective tuningwould optimize the input and outputSWR at the desired frequency.

To simplify testing, the effects of theend walls are minimized by locatingthem relatively far from the end reso-nators. The result is that leaving theend walls off during testing makes

6 QEX

little difference in performance. Ilocated the end walls one inch from theend resonators in the 1296 MHz filter,and could find only a slight perfor-mance difference with the end walls inplace. There was no detectable differ-ence for the higher frequency filters.Of course, the end walls should be in-stalled for operation; stray leakagecould otherwise negate the effect ofthe filter.

ConstructionThe three completed filters are

shown in Fig 5. Each resonator is at-tached by a screw through a narrowwall of the waveguide and the coaxialconnectors are mounted in a wide wallof the waveguide with short leads tothe tap points on the end resonators.

Resonator lengths and spacings arefairly critical, so accurate measure-ment is needed. The holes are bestmade with a drill press (see thesidebar “Tools for Interdigital FilterConstruction.” Start with a centerdrill or small drill bit to spot the hole,then follow with a drill bit of the de-sired diameter. The mounting holes inthe end of the resonators should betapped and countersunk slightly, socontact is made around the resonatorperimeter. For initial testing, I don’tsolder the input and output connec-tions, but rather make them slightlylong with a sharp point contacting thetap point on the resonators.

Using waveguide as the housingmakes the filters easy to build, andresults in a robust, stable filter, suit-able for rover operations. Some of myprevious experiments in filter con-struction using hobby brass and PCboard were less successful due tomechanical instability: Vibrations orthe weight of connecting coax cablesaffected their performance. One nota-bly bad filter was so unstable that thefrequency response would vary duringmeasurement.

After building and testing the filtersin Fig 5, I wondered if there was aneven simpler way to make these filters.Since the resonator length for the 3456MHz filter is just a hair’s breadth over3/4 inch, perhaps an ordinary 3/4-inch-long, 1/4-inch-diameter threaded stand-off could be used as a resonator. Theresonator spacings and the tap-pointdimensions are the critical ones, so Icalculated a filter with 75 MHz band-width using 1/4-inch-diameter resona-tors, then built it with threadedstandoffs. It took less than an hourto complete. A quick measurementshowed nearly 3 dB loss at 3456 MHz,

Table 2: Waveguide Interdigital Filter Examples

Waveguide WR-229 WR-137 WR-90 WR-90Target Frequency 1289 2304 3456 3456 MHzBandwidth 50 75 108 79 MHz

Resonator (Designed for waveguide λ/4)Diameter 0.375 0.25 0.1875 0.25 inchesEnd Length 1.983 1.099 0.732 0.727 inchesInterior Length 1.971 1.095 0.73 0.728 inchesTap Point 0.23 0.127 0.089 0.089 inchesλ/4 Frequency 1289 2155 3280 3280 MHzBandwidth same 70 100 75 MHz

Resonator (Designed for target frequency)Diameter ″ 0.25 0.1875 0.25 inchesEnd Length ″ 1.187 0.777 0.773 inchesInterior Length ″ 1.183 0.775 0.772 inchesTap Point ″ 0.136 0.093 0.095 inchesSpacing 1-2,3-4 1.47 0.864 0.58 0.653 inchesSpacing 2-3 1.632 0.951 0.636 0.709 inches

Compromise DimensionsResonatorDiameter same 0.25 0.1875 0.25* inchesEnd Length ″ 1.144 0.756 0.75* inchesInterior Length ″ 1.14 0.752 0.75* inchesTap Point ″ 0.132 0.09 0.092 inches

Loss, Calculated 0.2 0.3 0.4 0.6 dBLoss, Measured 0.45 1.25 0.8 2.3 dBBandwidth, Measured 58 70 100 68 MHzLO Frequency 1152 2160 3312 3312 MHzLO Rejection, Calculated –59 –47 –34 –45 dBLO Rejection, Measured –75 –49 –36 –49 dB* = threaded standoff

Fig 6—Performance of 3456 MHz filter built with threaded standoffs in WR-90waveguide (lower frequency response is after tuning).

Jan/Feb 1999 7

however, so I took it apart to add tun-ing screws to see if I could improve it.Careful tuning only reduced the loss to2.4 dB, versus 0.8 dB for the filter withmachined brass resonators. Fig 6shows the response before and aftertuning. The response is slightly higherin frequency before tuning, but other-wise there is little difference. I at-tribute the higher loss to three factors:• I arbitrarily designed this version

for a 75 MHz (2%) bandwidth, com-pared to a 108 MHz (3%) bandwidthfor the other 3456 MHz filter. Aspreviously mentioned, filters withnarrow bandwidths tend to be morelossy.

• The threaded standoffs are platedwith nickel, a lossy metal.

• The standoffs are chamfered at theends, so the contact area is smallerat the shorted end where currentsare highest.

PerformanceThe waveguide interdigital filters

exhibit excellent performance (asshown in Figs 2, 3, 4, and 6) with lowinsertion loss in the passband andhigh rejection of undesired frequen-cies. Steep skirts provide good rejec-tion of possible spurious signals, suchas LO leakage only 144 MHz awayfrom the operating frequency. LOrejection is much greater for the1296 MHz filter (–75 dB) than for theothers (–49 dB at 2304 MHz and–36 dB at 3456 MHz) because the rela-tive LO separation is much greater at1296 MHz. It’s 9% of the operating fre-quency at 1296 MHz, versus 6% at2304 MHz and 4% at 3456 MHz.

Table 2 lists the filter dimensionsand compares the measured perfor-mance with the design values, as cal-culated by INTFIL. The measured per-formance is quite close to the designvalues. The dimensions shown are forthe two designs for each filter, plus thecompromise values that I fabricated, toillustrate the design procedure.

The only performance flaw for thesefilters is poor harmonic rejection. Atfrequencies much higher than the op-erating frequency, the waveguide en-closure behaves as a waveguide ratherthan just an enclosure. This behavioris not unique to the waveguideinterdigital filters—all conductiveenclosures will propagate waveguidemodes at frequencies above the cutofffrequencies for the interior dimen-sions. Fig 7 shows the transmissioncharacteristics of the 3456 MHz filterfrom 50 MHz to 20 GHz. Out-of-bandrejection is excellent (> 70 dB) below

Tools for Interdigital Filter ConstructionI use a small metal lathe to trim the interdigital-filter resonators to length. A

lathe is the ideal tool for this work, but is a luxury for most hams. Some othertools are great for homebrewing, however, and inexpensive imports havemade them quite affordable. The two that I find almost indispensable are a drillpress and a dial caliper.

A drill press is a sturdy drill on a stand, with an adjustable table to hold thework. For the interdigital filters, it drills the holes square and true. The mount-ing holes in the resonator rods can be first drilled and then threaded, by chuck-ing a screw tap in the drill press and feeding it by hand (power off!), with therod held in a vise. Imported tabletop drill presses are available for less than$60.A, B I used and abused one of these constantly for 16 years before splurg-ing on a larger floor-mounted model; the old one is still in use for local “Elmer”sessions.

A dial caliper is a measuring instrument capable of resolving dimensions toa precision of 0.001 inches on a dial. Most of the dimensions in the interdigitalfilters, and for much microwave work, must be more precise than I can mea-sure with a ruler, so my dial caliper also sees constant use. I even scribedimensions directly with the caliper tips—a gross abuse of the tool that I justifyby its low replacement cost. Imported 6-inch dial calipers are available for lessthan $20,C, D a very modest investment compared to the alternative: eyestrainand frustration.

Everything else can be done with common hand tools—plus patience. Ahacksaw and file can cut the waveguide to length and trim the resonators,measuring frequently with the dial calipers. The holes are carefully marked,center-punched, and drilled to size with a drill press. A set of “number-sized”drills provides many more choices of hole size than ordinary fractional sizes,and is available at reasonable cost from any of the suppliers already men-tioned.—W1GHZ

A modest investment in tools—inexpensive, but not cheap—can add to thepleasures of homebrewing and improve the results.

SourcesAHarbor Freight Tools, 3491 Mission Oaks Blvd, Camarillo, CA 93011; tel 800-423-2567;http://www.harborfreight.comBGrizzly Industrial, Inc, 1821 Valencia St, Bellingham, WA 98226; tel 800-523-4777;http://www.grizzlyimports.comCMSC Industrial Supply Co, 151 Sunnyside Blvd, Plainview, NY 11803; tel 800-645-7270; http://www.mscdirect.com. The street address is for the corporate offices. Callfor a location near you.DEastern Tool & Supply, 149 Grand St, New York, NY 10013; tel 800-221-2679.

Fig 7—Wide-range performance (0 to 20 GHz) of 3456-MHz filter—WR-90waveguide.

8 QEX

the passband and good above the pass-band up to about 6 GHz. Above 6 GHz,various waveguide modes are propa-gated and limit the attenuation.

resulting in a robust, high-perfor-mance filter.

Notes1G. Elmore, N6GN, “A Simple and Effective

Filter for the 10-GHz Band,” QEX, July1987, pp 3-5.

2P. Wade, N1BWT, “A Dual Mixer for 5760MHz with Filter and Amplifier,” QEX,August 1995, pp 9-13.

3R. Fisher, W2CQH, “Interdigital BandpassFilters for Amateur VHF/UHF Applica-tions,” QST, March 1968, p 32.

4J. Hinshaw, N6JH, and S. Monemzadeh,“Computer-Aided Interdigital BandpassFilter Design,” ham radio, January 1985,pp 12-26.

5R. Dean Straw, N6BV, Editor, The ARRL

Handbook for Radio Amateurs, ARRL,1997. See pp 30.22 and following.

6G. Matthei, L. Young, and E. M. T. Jones,Microwave Filters, Impedance MatchingNetworks, and Coupling Structures,McGraw-Hill, 1968.

7W.S. Metcalf, “Graphs Speed Design ofInterdigital Filters,” Microwaves, February1967, pp 91-95.

8M. Dishal, “Alignment and Adjustment ofSynchronously Tuned Multiple-Resonator-Circuit Filters,” Proc. IRE, November1951, pp 1448-1455.

9G. Matthei, L. Young, and E. M. T. Jones,Microwave Filters, Impedance MatchingNetworks, and Coupling Structures,McGraw-Hill, 1968, pp 668-673.

ConclusionGood filters are important for opera-

tion in locations with RF pollution,and are recommended when no-tunetransverters are followed by broad-band power amplifiers. Interdigitalfilters offer excellent performance forthe lower microwave bands. The fil-ters are easily constructed in a wave-guide housing without extensivemachining and require little tuning,

Jan/Feb 1999 9

Need a new challenge? How about using DSPto accurately (within 2-3 Hz) tune SSB

transmissions in about five seconds? Here’san outline of the research and theory to do it.

By Robert Dick

13 Speer StSomerville, NJ 08876rdick@idt.net

Tune SSBAutomatically

Editor’s Note: This article covers a lotof ground quickly and assumes a fairlysophisticated knowledge of signal pro-cessing. The important message is thatyou can tune SSB automatically, andif you really want to get into it, obtaina copy of Reference 1. Those of you whotry this with a ham receiver, let usknow how it worked and the hard-ware/software you used. We can thenpublish a follow-up article.

It is possible for your personal com-puter to tune SSB voice signals usinga high-level language and standarddigital-signal-processing (DSP) tech-

niques. First you must tune to withinabout plus or minus one kilohertz,then your computer can finish the jobwith an accuracy of about plus or mi-nus three hertz. The tuning method isbased on the property of human speechthat most of the time it is periodic—and this period varies with time. Thisarticle describes the entire DSP pro-cess and gives code in C that does thesignal processing. To do the whole jobyou will need, in addition, a real-timeA/D and either a D/A or a way to letyour computer tune your receiver.

I discovered the method while doingUS Government sponsored researchtrying to remove voice-on-voice inter-ference. This was one of those caseswhere a byproduct of research wasmore successful than trying to accom-plish the main aim. Reference 1 re-

ports on my work, and contains codein FORTRAN that goes most of theway to estimating SSB mistuning. Italso contains code for SSB retuningvia DSP that I have improved on forthe present article.

Human speech is made up of twotypes of sounds: voiced and unvoiced,depending on whether or not thespeaker’s vocal cords are engaged whileproducing a certain sound. Unvoicedspeech is generally much lower in vol-ume than voiced speech and, in the3 kHz bandwidth of Amateur Radio,much of the unvoiced sound is filteredout. Voiced speech is nearly periodicover the short term (tens to hundreds ofmilliseconds), with a period that variesover time.

A periodic sound, when viewed inthe frequency domain, shows a series

10 QEX

of peaks (harmonics). The frequencydifference between neighboring peaksis the reciprocal of the time period ofrepetition. This range of peaks has onesignificant property for our purposes.If the range of peaks is extrapolateddownward in frequency there will al-ways be a resultant peak at zero fre-quency (dc). When the pitch changes,all the peaks shift position, except the(resultant) one at zero hertz. WhenSSB speech is mistuned, there will beone and only one (resultant) peak atsome non-zero frequency, which doesnot shift with time. This representsthe mistuning.

Fig 1 shows a “waterfall” plot givingan evolution of the (properly tuned)speech square-root-of-power spectraversus time. Each line shows a spec-trum of zero to 3200 Hz. A series ofpeaks is evident in most lines. Thereare no peaks at zero hertz on the left,of course, but if each “comb” of peakswere continued to the left, it wouldhave a tooth at zero hertz.

Note that each spectrum line in thewaterfall plot contains an envelopethat resembles a sinusoid. This essen-tial fact enables us to use the fast Fou-rier transform (FFT) to estimate theperiod and phase of the series of fre-quency-domain peaks. This estimatorFFT goes from the frequency domainback to the time domain. I call thisprocess—going from time to frequencyand back to time—complex correla-tion. The key to complex correlation iswhat is done in the frequency domainbetween the two FFTs: The squareroot of the power spectrum is taken,and the negative-frequency portion iszeroed. Then the inverse FFT is taken.Because the inverse FFT’s input wasnot symmetric, its output is complex.This is an essential part of the signalprocessing.

The complex correlation has a peakin its magnitude at the time intervalcorresponding to the period of repeti-tion of the speech waveform put into it.The phase at this magnitude-peak rep-resents something of the mistuning ofthe speech. If the speech is exactly prop-erly tuned the phase will be zero at thepeak and the complex correlation willbe purely real and positive.

If zero frequency is exactly halfwaybetween (resultant frequency-domain)peaks, the complex correlation will bepurely real and negative at its magni-tude-peak. Other conditions will resultin other complex phases at the peak.

Fig 2 shows another waterfall plot,this time of the magnitude of the com-plex correlation. All the horizontal

Fig 1—Square roots of speech-power spectra.

Fig 2—Magnitudes of speech “complex correlations” (see text for an explanationof this term).

Jan/Feb 1999 11

plots are normalized to have equalpower. On the left of the waterfall is avertical wavy line showing, by itsshape, the power in the waterfall. Theleftward position of each line segmentcorresponds to the power in the hori-zontal line. The maximum magnitudeof each (horizontal) complex-correla-tion line occurs at t = 0, on the left, butanother (local) maximum occurs at thespeech period. This waterfall plotshows the speech period peak waver-ing down the page. On certain lines,corresponding to unvoiced speech, thepeak disappears. Note however, thatthe magnitude-trace on the left showsthat at these times there is little powerin the speech.

After the complex correlation, a his-togram is incremented at each point inthe frequency domain where there wasa peak (or resultant peak) in the mag-nitude spectrum. The best value to usefor an increment is the squared magni-tude of the complex correlation at itsspeaker-pitch local peak. Because ofthe properties of voiced versus un-voiced speech, a common problem inspeech processing can simply be fi-nessed. This is the problem of distin-guishing between voiced and unvoicedspeech. Unvoiced speech is useless forestimating mistuning. However, ifwe use as a histogram increment thesquared magnitude of the complex-correlation peak, this increment will bedoubly small during unvoiced speech:First, because the total power is smallfor unvoiced speech. Second, since thisspeech is not periodic, the magnitude-squared of the complex-correlationpeak will be small, even relative to thetotal power, which itself is small.

Fig 3 shows the results of incre-menting a set of histograms. Using aset of histograms like this instead ofjust one histogram is not necessary tomy method, but here it illustrates theproperties of speech signals. Note thatthe complex correlation provides aseries of estimates of speaker pitch,along the way to estimating mistun-ing. Therefore, we may divide the his-togram into a set of subhistogramssorted by speaker pitch. Fig 3 resultedfrom 10 seconds of Amateur Radio SSBthat had not been exactly tuned. Zero-hertz mistuning (after only approxi-mate tuning) is represented by thevertical line in the center. Eachsubhistogram resulted from a 10-Hzrange of speaker pitch. The histo-grams are noisy, but they clearly showthe mistuning. Sighting up and downthe figure, we see that the histogrampeaks form slanting lines in most por-

tions of the spectrum. For example,one slanting-line of peaks cutsthrough zero Hz going from the lowerleft to the upper right. The next seriesof peaks to the right of this forms avertical line. This line shows the cor-rect estimate of the mistuning. To theright of the actual mistuning value,the peaks form a line from the lowerright to the upper left.

Thus, when a single histogram isformed, a single peak will stand out—if the frequency resolution is coarse.However, for maximum resolution inthe estimate, a fine resolution may beused. There will not in general in thiscase be a single big peak, but ratherthere will be a small cluster of (moreor less) large values in the histogrambordered on both sides by valleys. Forexample, in Fig 3, we see not only thatthe peaks line up at the actual mistun-ing, but also that there are valleys onboth sides of the actual mistuning.Therefore, looking for a concentrationin the histogram will allow finer reso-lution and a more accurate estimatethan looking for a single peak.

One method that works well is tofind the narrowest region of the histo-gram that contains (say) 80% of thetotal of increments to date. Then esti-mate the mistuning to be halfway be-tween the lower and upper limits ofthis region.

This is the basis of my method. Theprinciple of operation is straightfor-ward. First speech transmitted bySSB is roughly demodulated. As Fig 3

shows, mistuning of ±400 Hz can betolerated easily. Even greater mistun-ing is tolerable. The critical limit isreached (1) when too much of the sig-nal is excluded or (2) if the signal is notproperly filtered for anti-aliasing. Inthe latter case, the signal spectrum isfolded over onto itself and the “comb”structure is jumbled. Given proper fil-tering, I expect that mistunings up to+1 kHz can generally be tolerated.

Second, the roughly tuned signal isdigitized, preferably at a rate onlysomewhat greater than 6 kHz. Ofcourse, we must follow the samplingtheorem that requires us to sample atmore than twice the highest desiredfrequency. Conversely, this means wemust set the low-pass anti-aliasingfilter cutoff frequency at somewhatless than half the sampling rate. Inaddition, the signal should be high-pass filtered at about 300 Hz (accoupled). In particular, the signalmust have a very small dc component.

Next, successive sections of the digi-tized signal stream are selected. Eachsection is multiplied by a “window”that has a maximum in its middle andtapers to zero at each end. I use theraised-cosine window.

What length of window is suitable?Each window must have a number ofsamples that is an integral power oftwo. In addition, each window shouldpreferably contain three or four peri-ods of voiced speech. In the figuresshown above, I used a sampling rate of6400 Hz and a window of 512 samples

Fig 3—SSB speech pitch versus mistuning.

12 QEX

/ 6400 Hz = 80 milliseconds. This win-dow may have been a little long. I sug-gest a window somewhere in the rangeof 40 to 100 milliseconds.

The windows selected may havegaps between them. The duration ofthe gaps depends on how long it takesyour computer to perform a complexcorrelation and update a histogram. Iwould estimate that a rate of selectionas low as one window per second couldbe tolerated. What if you have a veryfast computer and need no gaps? Thenyou can let the windows overlap. How-ever, it is best that the windows notoverlap by more than 50%, so they arenot too redundant.

The SoftwareI have put together functions writ-

ten in C that do all the steps describedabove, and also functions that do SSBretuning within the computer. Theycompiled and ran successfully underBorland Turbo C (Version 1.5) forDOS, an old but serviceable compiler.I also compiled and ran some of thesoftware under Borland Turbo C++ forWindows. I used the fastest FFT high-level language computation methodthat I was aware of several years ago.I have since learned of Reference 3,which presents a faster FFT thanthose I used.

What I did requires a little explana-tion. A complex correlation requirestwo successive FFTs. Each FFT comesin two parts: one a series of so-called“butterfly” computations and the othera shuffling of the data. This shufflinghas the property that doing it twice isa no-op, the same as no shuffling at all.Since it is done in both FFTs, there isa way to avoid shuffling the data. Thisrequires two different FFT methods forthe two FFTs. I encoded the two meth-ods separately. The omitted data shuf-fling is represented in the “bitrev”function. Because the shuffling isomitted, the frequency domain isscrambled. Positive frequencies arerepresented by the even-numbered lo-cations. Those wishing to experimentto see what things look like in the fre-quency domain can apply “bitrev” tounscramble it and then plot it.

I have included a little function called“srss.” It approximately computes thesquare root of the sum of two squareswithout computing square roots. Theestimated approximation error is givenin the source code. It was determinedby means of a little calculus, but it canbe verified by experiment.

The module with the function “ccor”contains, in addition, a timing routine

Appendix 1: Parabolic Interpolation and Peak FindingHere is how to interpolate three equally spaced data points using a second-

order polynomial (ie, a quadratic or parabola).For x = –1, 0, 1, and y = ym, yz, and yp, we have the three points (–1, ym),

(0, yz), (1, yp).For polynomial coefficients cf0, cf1 and cf2, the formula

y cf cf x cf x= + +0 1 22 may be written

y cf x cf x cf= + + ×( )( )0 1 2

It is easy to show that

cf yz cfyp ym

cfyp ym

yz0 1 22 2= =

−( ) =+( )

, , –and

fits a curve to our three (x, y) points. This curve may be variously called a sec-ond-order curve, a quadratic or a parabola.

For ym < yz and yz > yp, the peak of this curve is at x = xm, where

xmyp ym

yz ym yp=

−( )− −

0 5

2

.

which is the direct formula for

xmcf

cf= − 1

22

This is a standard result of analytic geometry. You may have seen it as

xmb

aax bx c= − = + +

22for y

to estimate how fast your computercan perform complex correlations.(The preprocessor directive “#defineMAIN” enables this routine.) I use adigital watch to time from the “Enter”command until 1000 complex correla-tions are complete. My 486 DX2 PCwith a 66 MHz clock (running Win-dows) can do one “ccor” in about 25milliseconds for 256 points and about54 milliseconds for 512 points. UnderDOS, it takes 33 ms for 512 points. Itturns out that using brute-force meth-ods in “ccor” adds about 20% to thecomputing time. My 200 MHz PentiumMMX does 512 points in about 6 msunder DOS. A word of caution, if youwrite your own timing experiment:All-zero data gives an unrealisticallyoptimistic speed of computation. Mytiming programs use (admittedly non-sensical) nonzero data.

The module “timproc” puts thewhole process together and illustrateshow to call the various functions Iwrote. For timing purposes, I simu-lated an 8 kHz sampling rate, 512-point ccors, 8 ccors per second, andmistuning estimate once per second.On my 486 it takes 0.6 second of pro-cessing for each second of data. ThePentium takes 0.1 second. (Both num-bers are for DOS.) As computer speedincreases, calculation time may soonbe negligible.

The subroutine called “est” takes theoutput of ccor and estimates speaker

pitch and one possible mistuning. Ev-ery other possible mistuning is then amultiple of the pitch frequency awayfrom this first estimate. Most of thecode in est is the computation of para-bolic interpolation. You need this in-terpolation to get the maximum reso-lution of mistuning estimates. This isespecially necessary when tuninghigh-pitched speech. For high pitches,even a fairly large change in pitch re-sults in only a small shift of the ccorpeak. Further, the same mistuningdelta is more accurately measured atlow pitches than at high pitches. Anappendix shows the mathematics ofparabolic interpolation.

Updating and reading the histo-gram are the final stages of themistuning-estimator code. I have rep-resented them by key functions. Func-tion “inchist” increments a histogramwith data from an array (bins), wherethe number of array elements is nbins.Function “srchist” searches the histo-gram for the minimum-width portioncontaining some fraction (say 80%) ofthe sum of the increments to date. Theonly tricky thing about this code isbeing careful not to go beyond the endof the histogram. As you can see,“srchist” tests for this repeatedly.

The histogram should be zeroed atthe start of mistuning estimation andshould only be searched after at leastseveral increments have been added.You will probably need to adjust the

Jan/Feb 1999 13

Appendix 2: C Source Code

SOFTWARE.DOC for “Tune SSB Automatically.”This software is furnished as-is, without warranty of any kind. Do not attempt to create working software from thisdocument! It has been reformatted for publication. You can download workable source code from the ARRL athttp://www.arrl.org/files/. Look for the file SSBTUNE.ZIP. The file includes executable code for TI’s DSK3 DSP kit.Here is a list of the functions included:

Function File Descriptionaafilt aafilt.c Anti-alias filter for retuning.fir aafilt.c Finite Impulse Response filter.retune aaretun,c Retune a waveform block.

retune process.ccor ccor.c Complex correlation. Does one

window.main ccor.c Complex correlation timer.

ccor.exe DOS executable of ccor.c.fwfft ffts.c time-to-frequency FFT, minus

bit reversal shuffling.bitrev ffts.c Bit reversal data shuffling

(not used).srss ffts.c Square-root of sum-of-squares

approximation.rvfft ffts.c frequency-to-time FFT, minus

data shuffling and 1/n.hilb hilb.c Hilbert transform filter for

retuning.est histr.c Estimates pitch and one

possible mistuning.inchist inchist.c Increment histogram.srchist srchist.c Search histogram for mistuning.main timproc.c Time whole estimation and

timproc.exe DOS executable of timproc.c

/* AAFILT.C */#include <stdio.h>/*#define DEBUG*/#define NFILTS 7 /* Number of filters in filter bank */#define NTAPS 21 /* Number of taps in each fir filter */

float fir(float in, float *tr_coef, int n, float *history);

void aafilt(int remember, float rel_sft, int n_pts, float in[], float out[])/* * Anti-aliasing fir filters routine. Selects filter from filter bank, * and zeros filter memory, when remember==0. * * remember==0: Determine filter, zero its memory, proceed.

parameters in your estimator to oper-ate best with your sampling rate andrate of choosing windows. Generally,you’ll need about five seconds ofspeech to refine the tuning within twoor three hertz.

The method of retuning is standardfor DSP, and I present it here with noclaim of invention. Function “retune”runs the process. Given a waveform tobe retuned, it is first anti-alias filteredto prevent spectrum folding after theretuning. Function “aafilt” passes thewaveform through one of several high-pass FIR filters for a large downshift;through a band-pass filter for a smallshift; or one of several low-pass filtersfor a large upshift. The filters weredesigned using the “remez” programincluded with Reference 3. The func-tion “fir” is also from Reference 3.

A single-channel baseband wave-form always produces double-sideband modulation. One sidebandmust be removed to make SSB. To doso one attaches as an imaginary chan-nel, the negative of what is known asthe Hilbert transform of the wave-form. This is done using function“hilb.” Multiplying this complex wave-form by a complex exponential retunesit. Only one channel of the result needbe saved for output.

New Jersey: Prentice-Hall, 1975).3. P. M. Embree, C Algorithms for Real-

Time DSP (Upper Saddle River, NewJersey: Prentice-Hall PTR, 1995). Includesdiskette.

References1. R. J. Dick, “Cochannel Interference

Separation,” RADC-TR-80-365, Rome AirDevelopment Center, 1980. Available for$19.50 as document ADA096059 from theNational Technical Information Service,NTIS, US Department of Commerce,Springfield, VA 22161.

2. A. V. Oppenheim and R. W. Schafer, Digi-tal Signal Processing (Englewood Cliffs,

Robert Dick, while not a licensedradio amateur, has extensive educationand experience in communication anddigital signal processing (DSP). Heholds a Bachelor of Science in electricalengineering from MIT and a Master ofScience and doctorate from Cornell inthe mathematical theory of communica-tion. He has worked with delta modula-tion, sequential and Viterbi decoding,algebraic coding, HF channel simu-lation, radio “fingerprinting,” a decen-tralized radio network protocol andpattern classification, among other sub-jects. His most recent DSP work is inmultichannel adaptive filtering.

14 QEX

int i; float *hist_ptr, *hist1_ptr, *tr_coef_ptr; float out;

hist_ptr=history; hist1_ptr=hist_ptr; /* use for history update */ tr_coef_ptr=tr_coef; /* fir coefs must be symmetric or time reversed */

/* accumulate output */ out=*hist_ptr++ * (*tr_coef_ptr++); /* 1st contribution */ *hist1_ptr++ = *hist_ptr; /* update a term of history array */ out += (*hist_ptr++) * (*tr_coef_ptr++); /* 2nd contribution */ for(i=3; i<n; i+=2) { *hist1_ptr++ = *hist_ptr; out += (*hist_ptr++)*(*tr_coef_ptr++); /* ith contribution */ *hist1_ptr++ = *hist_ptr; out += (*hist_ptr++)*(*tr_coef_ptr++); /* (i+1)st contribution */ } *hist1_ptr=in; /* put input into history */ out += in * (*tr_coef_ptr); /* nth contribution */

return(out);}

/* AARETUN.C */#include <stdio.h>#include <math.h>#include “c:\turboc\ssb\hilb.c”#define SNGL_LEN 1024

void aafilt(int remember, float rel_shift, int n, float in[], float out[]);

void retune(int remember, float rel_shift, int n, float in[], float out[])/* * Anti-aliasing SSB retuner. * Use -1.<rel_shift<0.5. Fraction of sampling rate to shift. * Example: rel_shift==-0.5 inverts frequencies. * n is num points in and out. Need 15 < n <= SNGL_LEN. * Anti-alias filter, attach imag part, retune. * Calls aafilt for anti-aliasing. Calls hilb to null neg freqs. */{ static float inbuf[SNGL_LEN]; /* input anti-alias filtered */ static float rbuf[SNGL_LEN], hilbuf[SNGL_LEN], ibuf[SNGL_LEN]; static float oldrbuf[15]; /* static so preserved between calls */ /* hilbuf=Hilbert transform output buffer==minus imag part. * oldrbuf=old reals buffer for 15-sample delay to line up with * Hilbert transform output. */ float pi=3.14159265358979, twstr, twsti, turnr, turni; /* twst=unit-magnitude complex number with angle in radians * 2*pi*rel_shift=phase angle advance per sample * turn=repeatedly twisted unit-magnitude complex number. */ int idx; float tmp;

if(remember==0) { for(idx=0;idx<15;idx++) { oldrbuf[idx]=0.; } turnr=1.; turni=0.; } twstr=cos(2.0*pi*rel_shift); twsti=sin(2.0*pi*rel_shift);

/* prefilter to prevent postshift aliasing */ aafilt(remember, rel_shift, n, in, inbuf);

for(idx=0;idx<15;idx++) { rbuf[idx]=oldrbuf[idx]; oldrbuf[idx]=inbuf[n-15+idx]; } for(idx=0;idx<n-15;idx++) { rbuf[15+idx]=inbuf[idx]; }

* remember!=0: Use previously determined filter and its memory * rel_sft = Proposed frequency shift divided by sampling frequency * n_pts = Number of data points to anti-alias filter * in = n_pts data vector in * out out */{ static float filtmem[NTAPS-1]; /* filter init cond-memory */ static int ifilt; /* index to selected filter */ /* these are static so can be remembered between calls */

int idx;

static float aadat[NFILTS][NTAPS+1] = {{-0.12, /* hipass 0.16 rpl 0.163, stop 0.12 max gain 0.054 */0.0348, -0.0712, -0.0334, 0.0024, 0.0401, 0.0585, 0.0321,-0.0461, -0.1540, -0.2478, 0.7150, -0.2478, -0.1540, -0.0461,0.0321, 0.0585, 0.0401, 0.0024, -0.0334, -0.0712, 0.0348},

{-0.08, /* hipass 0.12 rpl 0.160, stop 0.08 max gain 0.053 */-0.0613, 0.0386, 0.0432, 0.0450, 0.0345, 0.0060, -0.0400,-0.0963, -0.1507, -0.1902, 0.7954, -0.1902, -0.1507, -0.0963,-0.0400, 0.0060, 0.0345, 0.0450, 0.0432, 0.0386, -0.0613},

{-0.04, /* hipass 0.08 rpl 0.150, stop 0.04 max gain 0.050 */0.0773, 0.0017, -0.0067, -0.0202, -0.0379, -0.0581, -0.0790,-0.0982, -0.1139, -0.1241, 0.8724, -0.1241, -0.1139, -0.0982,-0.0790, -0.0581, -0.0379, -0.0202, -0.0067, 0.0017, 0.0773},

{0.04, /* bandpass 0.04 to 0.46 rpl 0.160, stops 0.01 & 0.49 */-0.1105, 0., -0.0686, 0., -0.0832, 0., -0.0950,0., -0.1027, 0., 0.8946, 0., -0.1027, 0.,-0.0950, 0., -0.0832, 0., -0.0686, 0., -0.1105},

{0.08, /* lopass 0.42 rpl 0.150, stop 0.46 max gain 0.050 */0.0772, -0.0017, -0.0066, 0.0202, -0.0379, 0.0581, -0.0790,0.0983, -0.1139, 0.1240, 0.8724, 0.1240, -0.1139, 0.0983,-0.0790, 0.0581, -0.0379, 0.0202, -0.0066, -0.0017, 0.0772},

{0.12, /* lopass 0.38 rpl 0.160, stop 0.42 max gain 0.053 */-0.0613, -0.0387, 0.0431, -0.0450, 0.0346, -0.0060, -0.0401,0.0963, -0.1507, 0.1902, 0.7954, 0.1902, -0.1507, 0.0963,-0.0401, -0.0060, 0.0346, -0.0450, 0.0431, -0.0387, -0.0613},

{0.16, /* lopass 0.34 rpl 0.163, stop 0.38 max gain 0.054 */0.0348, 0.0712, -0.0334, -0.0024, 0.0401, -0.0585, 0.0321,0.0461, -0.1541, 0.2478, 0.7151, 0.2478, -0.1541, 0.0461,0.0321, -0.0585, 0.0401, -0.0024, -0.0334, 0.0712, 0.0348}};

if(remember==0) { for(idx=0; idx<NTAPS-1; idx++) { filtmem[idx]=0.; } for(ifilt=0; ifilt<NFILTS-1 && aadat[ifilt][0]<rel_sft; ifilt++); /* want range 0 <= ifilt < NFILTS */

}

#ifdef DEBUG printf(“Ifilt %d, relcut %g.\n”,ifilt,aadat[ifilt][0]);#endif

for(idx=0; idx<n_pts; idx++) { out[idx]=fir(in[idx], &aadat[ifilt][1], NTAPS, filtmem); }}

/* FIR filter routine */float fir(float in, float *tr_coef, int n, float *history)/* * finite impulse response (fir) filter computation, * adapted from Embree listing 4.1. * * in = one floating-point input * tr_coef = time-reversed or time-symmetric impulse response * n = number of terms of tr_coef. Must be odd and >=5. * history = n-1 size array of past inputs and/or initial conditions * * returns one floating-point output */{

Jan/Feb 1999 15

hilb(remember, n, inbuf, hilbuf); for(idx=0;idx<n;idx++) { /* Imaginary part is minus Hilbert transform of real part */ ibuf[idx]=-hilbuf[idx]; }

for(idx=0; idx<n; idx++) { /* Mult by complex exponential to shift freq. Save reals. */ out[idx]=rbuf[idx]*turnr - ibuf[idx]*turni; tmp=twstr*turnr - twsti*turni; turni=twstr*turni + twsti*turnr; turnr=tmp; }}

#ifdef MAINint main()

/* Timing test of retuner via 1000 point blocks */

{ int remember, nblks, idx, jdx; static float dat[SNGL_LEN], out[SNGL_LEN]; float rel_shift;

for(idx=0;idx<SNGL_LEN;idx+=32) { /* fill input data buffer with dummy data for test purposes */ for(jdx=0;jdx<32;jdx++) { if(idx+jdx<SNGL_LEN) { dat[idx+jdx]=15-jdx; } } }

nblks=1; while(nblks>0) { printf(“Enter no-resamp relative frequency shift.\n”); scanf(“%f”,&rel_shift); printf(“Enter number of 1000-point blocks to retune.\n”); scanf(“%d”,&nblks); remember=0; for(idx=0;idx<nblks;idx++) { retune(remember,rel_shift,1000,dat,out); remember=1; } printf(“Done.\n”); }}#endif

/* CCOR.C */#include<stdio.h>#include”c:\turboc\ssb\ffts.c” /* laptop version */#include”c:\turboc\ssb\histr.c” /* laptop version *//*#define MAIN*/

void ccor(int n,float dat[],float w[],float xr[],float xi[])

/* “Complex Correlation” t->f->t function */

/* int n; Must be a positive power of 2 *//* float dat[]; Input (time) waveform section *//* float w[]; Data Window (raised cosine) *//* float xr[], xi[]; Real & Imag data vectors */{ int idx;

/* Windowed data -> real vector, zeros -> imag vector */ for (idx = 0; idx < n; idx++) { xr[idx] = dat[idx]*w[idx]; xi[idx] = 0.0; } fwfft(n,xr,xi);/* Now in scrambled-index frequency domain *//* Form magnitudes at pos freqs, zeros elsewhere *//* Scrambled pos freq <=> even-numbered index */ xr[0]=0.0; /* DC is still zero index when scrambled */ xi[0]=0.0; xr[1]=0.0; xi[1]=0.0;

for (idx = 2; idx < n; idx = idx + 2) { xr[idx] = srss(xr[idx],xi[idx]); xi[idx] = 0.0; xr[idx+1] = 0.0; xi[idx+1] = 0.0; } rvfft(n,xr,xi);/* Now xr & xi are “complex correlation” */}

#ifdef MAIN

/* “Main” Routine */

int main()

/* Timing test of complex correlation plus mistuning estimator. */

{ int n, idx, jdx, iter, niter; float pi=3.14159265358979, smprt=6400., peak, pitch, mstn; static float dat[2048], w[2048], xr[2048], xi[2048];

for(idx=0;idx<2048;idx=idx+32) { for(jdx=0;jdx<32;jdx++)

dat[idx+jdx]=15-jdx; } n=1; while(n>0) { printf(“\n Enter fft size. “); scanf(“%d”,&n); for(idx=0;idx<n;idx++)

w[idx]=1.-cos((2.*pi*idx)/n); printf(“ Enter num ccor & est its. “); scanf(“%d”,&niter); for(iter=0;iter<niter;iter++) {

ccor(n,dat,w,xr,xi); est(n,xr,xi,smprt,&peak,&pitch,&mstn);

} printf(“ Done.”); printf(“\n Pitch %4.4f, mistun %4.4f”,pitch,mstn); printf(“\n Enter 0 to stop.”); scanf(“%d”,&n); } return 0;}#endif

/* FFTS.C */#include<math.h>

/* FWFTT Routine */void fwfft(int n,float xr[],float xi[])

/* t->f FFT omitting post-butterflies bit-reversal *//* From Oppenheim & Schaffer Fig. P6.5 page 332 */

/* int n; Must be a positive power of 2 *//* float xr[], xi[]; Real & Imag data vectors */{ float pi = 3.14159265358979; float tmpr, tmpi; /* Temporaries */ int idx, iptr, jptr, leap, jump; /* Indices */ float twstr, twsti, turnr, turni;

leap = n; while (leap > 1) { jump = leap>>1; /* Right shift 1 */ turnr = 1.0; turni = 0.0; twstr = cos(pi/jump); twsti = -sin(pi/jump); /* For t->f */ for (idx = 0; idx < jump; idx++)

16 QEX

{ for (iptr = idx; iptr < n; iptr = iptr + leap) { jptr = iptr + jump; tmpr = xr[iptr] - xr[jptr]; tmpi = xi[iptr] - xi[jptr]; xr[iptr] = xr[iptr] + xr[jptr]; xi[iptr] = xi[iptr] + xi[jptr]; xr[jptr] = tmpr*turnr - tmpi*turni; xi[jptr] = tmpr*turni + tmpi*turnr; } tmpr = turnr*twstr - turni*twsti; turni = turnr*twsti + turni*twstr; turnr = tmpr; } leap = jump; }

/* Bit-reversal data shuffling omittedfrom here */}

int bitrev(int n,float xr[],float xi[])

/* BIT-REVersal data shuffling From “Cochannel” page B-8 From Oppenheim & Schafer Fig. P6.5 page 332 Doing it twice is a no-op

int n; Must be a positive power of 2float xr[], xi[]; Data vectors */{ int nv2; /* n/2 */ int ibtrv, idx, iptr; /* ibtrv is “bit-reversed” index */ float tmpr, tmpi; /* Temporaries */

ibtrv = 0; nv2 = n/2; for (idx = 0; idx < n-1; idx++) { if (idx < ibtrv) { /* Swap buffer points */ tmpr = xr[ibtrv]; tmpi = xi[ibtrv]; xr[ibtrv] = xr[idx]; xi[ibtrv] = xi[idx]; xr[idx] = tmpr; xi[idx] = tmpi; } /* Increment the bit-reversed index */ iptr = nv2; while (ibtrv >= iptr) { ibtrv = ibtrv - iptr; iptr = iptr/2; } ibtrv = ibtrv + iptr; } return 0;}

/* RVFFT Routine */void rvfft(int n,float xr[],float xi[])

/* Reverse FFT (f->t) omitting bit-rev and div by n *//* Based on “Cochannel” CFFT pages B-8 to B-9 */

/* int n A positive power of 2 *//* float xr[], xi[] Real & Imag data vectors */{ float pi = 3.14159265358979; float tmpr, tmpi; /* Temporaries */ int idx, iptr, jptr, leap, jump; /* Indices */ float twstr, twsti, turnr, turni; /* Rotations */

/* Bit-reversal data shuffling omitted from here */ leap = 1; while (leap < n) { jump = leap; leap = leap*2; twstr = cos(pi/jump); twsti = sin(pi/jump); /* for f->t */ turnr = 1.0; turni = 0.0; for (idx = 0; idx < jump; idx++) { for (iptr = idx; iptr < n; iptr = iptr + leap) { jptr = iptr + jump; tmpr = xr[jptr]*turnr - xi[jptr]*turni; tmpi = xr[jptr]*turni + xi[jptr]*turnr; xr[jptr] = xr[iptr] - tmpr; xi[jptr] = xi[iptr] - tmpi; xr[iptr] = xr[iptr] + tmpr; xi[iptr] = xi[iptr] + tmpi; } tmpr = turnr*twstr - turni*twsti; turni = turnr*twsti + turni*twstr; turnr = tmpr; } }/* Dividing data by n omitted from here */}

/* HILB.C */#define BUFLEN 1054

void hilb(int remember, int n, float in[], float out[])/* remember==0 is signal to zero internal memory before proceeding * n is number of points in and out. Need 0<n<=BUFLEN-30 * in is input * out is output, delayed 15 points */{ static float buf[BUFLEN]; /* 30 pts old input then n pts new */ float tap[8] = {0.0205, 0.0213, 0.0326, 0.0488, 0.0730, 0.1140, 0.2040, 0.6338}; /* 31-tap filter, every second tap zero, antisymmetric * gain 0.976 to 1.024 in 0.03 to 0.47 sample rate */ int ipt, idx; float sum;

if(remember==0) { for(idx=0;idx<30;idx++) { buf[idx]=0.; } } for(idx=0;idx<n && idx<BUFLEN-30;idx++) { /* add new input data to buffer */ buf[idx+30]=in[idx]; } for(idx=0;idx<BUFLEN-30 && idx<n;idx++) { /* form one point of output */ sum=0.; for(ipt=0;ipt<8;ipt++) { /* add effects of two taps with same magnitude */ sum+=(buf[idx+16+2*ipt] -

/* SRSS Routine */float srss(float x,float y)

/* Square Root of Sum of Squares Approximator *//* Approx sqrt(x*x+y*y) without squareroot *//* RMS error 2.4 percent (32 dB down) */

{ float magx, magy, ans;

magx=fabs(x); magy=fabs(y); if (magx > magy) ans = 0.95*magx + 0.4*magy; else ans = 0.95*magy + 0.4*magx; return ans;}

Jan/Feb 1999 17

buf[idx+14-2*ipt])*tap[7-ipt]; } out[idx]=sum; } for(idx=0;idx<30 && idx+n<BUFLEN;idx++) { /* save last 30 points of input */ buf[idx]=buf[idx+n]; }}

/* HISTR.C */#include<math.h>

/* EST Routine */void est(int n,float xr[],float xi[],float smpfrq, float *ppeak,float *ppitch,float *pshift)/* Estimator of pitch and one (out of many possible) freq shift *//* See “Cochannel” page B-21 *//* n; Buffer size (power of 2) *//* xr[],xi[]; Real, imag buffers from ccor *//* smpfrq; Sampling frequency *//* peak; Peak power at speaker pitch *//* pitch; Estimated speaker pitch *//* shift; One possible frequency shift */{ float pi=3.14159265358979; float lolim,hilim; /* low, high limits for speaker_pitch search range */ int lidx,hidx; /* low, high index limits for tau search range */ float temp; int idx, idpk; /* index and index_to_peak */ float bsq,usq; /* below_square and upper_square */ float x; /* fractional spacing of interpolated peak, -1<=x<=1 */ float tau; /* sampling_frequency/speaker_pitch */ float cf1,cf2; /* coefficients for parabolic interpolation */ float partr,parti; /* real, imag, parts of ccor peak */ float angl; /* Shift_angle in radians versus speaker pitch */

lolim=50; /* Assumes speaker pitch >= 50 Hz */ hilim=250; /* Assumes speaker pitch <= 250 Hz */ lidx=smpfrq/hilim; /* hi pitch is low time delta */ if(lidx<4)lidx=4; hidx=smpfrq/lolim; if(hidx>n/2-2)hidx=n/2-2; *ppeak=0.; idpk=4; /* 2-18-98 */ for(idx=lidx;idx<=hidx;idx++) { temp=xr[idx]*xr[idx]+xi[idx]*xi[idx]; if(*ppeak<temp) { *ppeak=temp; idpk=idx; } } /* Find quadratic-interpolation peak */ bsq=xr[idpk-1]*xr[idpk-1]+xi[idpk-1]*xi[idpk-1]; usq=xr[idpk+1]*xr[idpk+1]+xi[idpk+1]*xi[idpk+1]; x=1.; if(*ppeak>usq) x=0.; if(bsq>=*ppeak) x=-1.; if(x==0.) x=0.5*(usq-bsq)/(2.* *ppeak-bsq-usq); tau=idpk+x; *ppitch=smpfrq/tau; /* Interpolate real and imag parts */ cf1=0.5*(xr[idpk+1]-xr[idpk-1]); cf2=0.5*(xr[idpk-1]+xr[idpk+1])-xr[idpk]; partr=xr[idpk]+x*(cf1+x*cf2); cf1=0.5*(xi[idpk+1]-xi[idpk-1]); cf2=0.5*(xi[idpk-1]+xi[idpk+1])-xi[idpk]; parti=xi[idpk]+x*(cf1+x*cf2); *ppeak=partr*partr+parti*parti; /* calculate 4-quadrant arctangent (-pi/2 to 3pi/2) */ if(partr>0.) angl=atan(parti/partr); if(partr==0.) {

if(parti>=0.) angl=0.5*pi; else angl=-0.5*pi; } if(partr<0.) angl=pi-atan(-parti/partr); *pshift=*ppitch*angl/(2.*pi);}

/* INCHIST.C *//* Increment Histogram */

void inchist(int nbins, float bins[], float hpitch, float hshift, float incr)

/* nbins Number of histogram bins *//* bins The array of histogram bins *//* hpitch Est spkr pitch in num of bins spanned *//* hshift One poss freq shift in histogram units *//* incr Amount to increment each selected bin */

/* For lowf=min freq of bin 0 (lowf may be < 0), *//* And hif=min freq of bin (nbins-1), *//* hpitch = pitch*nbins/(hif-lowf) and *//* hshift =(shift-lowf)*nbins/(hif-lowf). */

{ float fidx; /* Floating point histogram index */ int idx; /* Integer histogram index */

if(hpitch>=1.0 && 0.0<=hshift && hshift<nbins && incr>0.0) { fidx=hshift; while(fidx>=0.0) {

idx=fidx; bins[idx]=bins[idx]+incr; fidx=fidx-hpitch;

} fidx=hpitch+hshift; while(fidx<nbins) {

idx=fidx; bins[idx]=bins[idx]+incr; fidx=fidx+hpitch;

} }}

/* SRCHIST.C *//* Search Histogram */

void srchist(int nbins, float bins[], float thresh, int *pminwid, float *pctr)

/* nbins Number of histogram bins *//* bins The histogram bins themselves *//* thresh Threshold for sum of bins *//* minwid Min width-1 of interval with sum >= thresh *//* ctr Center of minwid bins interval *//* Note: Call using arguments &minwid and &ctr */

{ int lidx,hidx,oldlow,minlow; float sum;

*pminwid=nbins; lidx=hidx=oldlow=minlow=0; sum=bins[0]; /* 2-18-98 */

while(hidx<nbins) { while(sum<thresh && hidx<nbins) { /* Advance forward limit until sum>=thresh */

hidx++; if(hidx<nbins) sum=sum+bins[hidx];

} while(sum>=thresh && lidx<=hidx && hidx<nbins)

18 QEX

{ /* Advance rear limit until sum<thresh */ sum=sum-bins[lidx]; oldlow=lidx; lidx++;

} if(hidx<nbins) {

if(hidx-oldlow<*pminwid) { /* If intvl narrowest so far note it */ *pminwid=hidx-oldlow; minlow=oldlow; }

} } *pctr=minlow+*pminwid*0.5;}

/* TIMPROC.C * Time complete est-tune and retune process using dummy * speech data. Retune BEFORE tuning estimator for test * purposes. * Compile with AAFILT.C, AARETUN.C, CCOR.C, INCHIST.C, * and SRCHIST.C. */

#include <stdio.h>#include <math.h>

void retune(int rem,float shf,int n,float in[],float out[]);void ccor(int n,float in[],float win[],float xr[],float xi[]);void est(int n,float xr[],float xi[],float smf,

float *pk,float *ptch,float *shf);void inchist(int nb,float bs[],float hptch,float hsft,float incr);void srchist(int nb,float bs[],float thr,int *mwid,float *ctr);

#define SAMP_RATE 8000.0 /* Nominal sampling rate in Hz */#define FFT_N 512 /* FFT and IFFT block length */#define BLKS_TO_EST 8 /* Num input blks per mistune est */#define LOW_HIST_F -800.0 /* Low limit histogram frequency */#define HI_HIST_F 800.0 /* Highest hist bin bottom freq */#define HIST_N 1600 /* Number of histogram bins */#define SUM_PCT 80.0 /* Percent of sum of histogram increments

* to be in min width histogram interval */

#define HZTOBINS (HIST_N)/((HI_HIST_F)-(LOW_HIST_F)) /* Bins per Hz */#define RETUNE_N ((int)(SAMP_RATE)/(BLKS_TO_EST)) /* Retune blocklen */

/* “Main” Routine */main (){ static float in[RETUNE_N], retout[RETUNE_N]; /* in = admittedly nonsensical input data * retout = retuner output data */ static float win[FFT_N], fftr[FFT_N], ffti[FFT_N]; /* win = raised-cosine data window * fftr = fft and complex correlation real part * ffti = imag part */ static float histbins[HIST_N]; /* The bins of the power versus frequency histogram */ float act_shift, ccorppow, pitch, estshift, suminc;

/* act_shift = Actual sim mistuning freq shift in sim Hz * ccorppow = height of ccor (interpolated) peak mag-squared * pitch = value of ccor estimated speech pitch in Hz * estshift = ccor est of (one possible) freq shift in Hz * suminc = sum of hist increments since last his reset */ int idx, jdx, kdx, itau, nmloop, mloopx, minwid; /* minwid = num-1 hist bins in min width interval w sum >= thresh */ float ctr, shift_hz, f_delt; /* ctr = est in binwidths of mistuning minus LOW_HIST_F * shift_hz = ctr converted to Hz minus zero frequency * f_delt = 0.5 * (minwid+1) * (bin width in Hz) */

for(idx=0; idx<FFT_N; idx++) { /* Define raised-cosine window */ win[idx]=1.0-cos(2.0*3.1415926*(idx+0.5)/(FFT_N)); } nmloop=1; while(nmloop>0) { printf(“FFT pts %d, hist bin %g Hz wide, range %g to %g.\n”,

FFT_N, 1./(HZTOBINS), LOW_HIST_F,HI_HIST_F); printf(“Min width hist intvl %g pct incr fm reset.\n”,SUM_PCT); printf(“Enter retune freq Hz (samp rate %g Hz).\n”,SAMP_RATE); scanf(“%f”,&act_shift); printf(“Enter number of times thru (nom 1 sec) loop.\n”); scanf(“%d”,&nmloop); for(mloopx=0;mloopx<nmloop;mloopx++) {

for(idx=0;idx<HIST_N;idx++) { /* Clear histogram */ histbins[idx]=0.; } suminc=0.; /* Reset sum of histogram increments */ for(idx=0;idx<BLKS_TO_EST;idx++) { /* Vary dummy rep period (tau) from low to high */ itau=40 + 11*idx; /* tau increment shd be prime */ for(jdx=0;jdx<RETUNE_N;jdx+=itau) { /* sawtooth sims speech */ for(kdx=0;kdx<itau && jdx+kdx<RETUNE_N;kdx++) { in[jdx+kdx]=(1-(2.*kdx+1)/(float)itau); } }

retune(0, act_shift/(SAMP_RATE), RETUNE_N, in, retout); /* Skip first 64 points of retuner output */ ccor(FFT_N, &retout[64], win, fftr, ffti); est(FFT_N,fftr, ffti, SAMP_RATE,

&ccorppow, &pitch, &estshift); suminc+=ccorppow; inchist(HIST_N, histbins,

pitch*HZTOBINS, (estshift-LOW_HIST_F)*HZTOBINS, ccorppow);

}

srchist(HIST_N, histbins, (SUM_PCT*0.01)*suminc, &minwid, &ctr);

shift_hz=LOW_HIST_F+(ctr+0.5)/(HZTOBINS); f_delt=(minwid+1.0)/(2.0*HZTOBINS); /* minwid==0 for 1 bin */

} printf(“Done.\n”); printf(“Est shift %g Hz, +/- %g.\n”,shift_hz,f_delt); }

}

Jan/Feb 1999 19

Are you tired of requesting your favorite hamsoftware developer to support your favorite rig or device?

By Lawrence G. Dobranski, VA3LGD/VE3TVV

14 Sandhead TerraceNepean, Ontario, K2J 1L4va3lgd@amsat.org

The Need for StandardApplication-Programming

Interfaces (APIs) inAmateur Radio

The ProblemSome recent discussions on Internet

e-mail reflectors supporting some ofthe popular contesting and loggingsoftware reinforced the lack of pro-gramming standards within theAmateur Radio community. When anew radio or other product is released,this lack of standardization causes de-velopers to scramble; they mustmodify their programs to interface tothe new equipment. Many softwaredevelopers do not provide AmateurRadio software as their primary occu-pation. Finding the specifications, get-ting access to the equipment and veri-

fying the interfaces can be a continualand significant hardship.

The SolutionsTwo solutions exist for this problem.

The first is to have ham-radio equip-ment manufacturers develop stan-dard interfaces and command sets.Given the ham-radio equipment man-ufacturers’ inability to agree on thewiring of the microphone connector,the possibility of getting a standardcommand set developed, approved andused is unlikely.

The second solution is based on asimilar problem that has already beensolved in the software-developmentworld. Today’s operating systems pro-vide many features to application de-velopers. The features and servicesare usually accessed through an appli-

cation-programming interface (API).APIs provide a set of function callsthat application developers use. Thesoftware that implements these callsperforms the lower-level functions ofthe device or operating system.

What is an API?To understand what an API is, let’s

review how one works through anexample. Most contesting and loggingprograms use the computer inter-face on Amateur Radio transceivers.Through this interface, they read andset frequency, band and other infor-mation. To set the radio’s frequency,the program must convert the data toa numerical format the radio can un-derstand, format the data within anappropriate command structure andsend the result to the radio. [Acknow-

20 QEX

ledgment from the transceiver mayalso be required.—Ed.] Similar opera-tions are performed to set or readother rig data. Today, these programsmust support many different rigs frommany different manufacturers, eachwith its own command set and dataformat. The protocols may be quitedifferent, and the command sets mu-tually exclusive.

Instead of the application softwarecomposing the radio command di-rectly, it might call a standard soft-ware function instead—HAM_API_Rig_SetFrequency(x)—where x is thedesired frequency. This API call istranslated by a radio-specific libraryinto the radio’s command format.When new radios are released, a newradio-specific library is developed,rather than modifying the applicationsoftware. With the addition or updateof the new library, all existing appli-cations that use the HAM API wouldthen be able to interface to the radio.

Where in Amateur Radio WouldWe Use Them?

Amateur Radio is becoming increas-ingly computerized. In many of ourshacks, computers are interfaced toour rigs, TNCs, rotators, voice keyers,CW keyers, antenna switches, GPSs,etc. When software is developed to aidthe amateur, it must be built to sup-port specific equipment. If AmateurRadio APIs existed, then we wouldonly need to develop specific libraries.The application would no longer haveto be modified.

Table 1 lists a sample of the APIfunctions that could be developed foramateur use.

Two Sample APIsTo see the effectiveness of APIs,

Tables 2 and 3 define samples thatmight be used for rig control and in-terfacing to a PacketCluster. The styleused in the definition depicts the APIas a set of functions. It could be definedin terms of object-oriented-program-ming constructs as well.

How Do We Develop Them?If this approach to computer control

of amateur equipment is acceptable,interested amateurs must developworking groups to author the respec-tive APIs. These working groups coulddiscuss their development—using theInternet, for example—and authorthe various libraries. Once an API isdeveloped, it would not be considereda reference standard until two unre-lated applications use the API to con-trol two different devices. A test suiteis then developed to verify that futureAPI implementations meet the stan-dard. This conformance test ensuresthe user that the API implementationwill work with their application.

Before API standard developmentbegins, a standard naming conventionfor Amateur Radio APIs should bedeveloped. For example, a proposednaming convention is as follows:

HAM_API_xxx_yyyy(). xxx is theAPI name (ie, rig, tnc, rotor) and yyyyis the function name. Variables andconstants are named in a similar way.

Table 1Proposed API Classifications

Amplifier ControlAntenna SwitchCall book interfacesCW Contest KeyersDigital Voice KeyersGPS DataPacketClusterRig ControlRotor ControlSatellite TrackersTNC Control

Table 2A Sample API for Rig Control

Function DescriptionHAM_API_Rig_getName() Returns the rig name and model numberHAM_API_Rig_getCaps() Returns—in a standard data structure—information about the rig’s capabilities: mode,

frequency range, output power, etcHAM_API_Rig_selectRig() Sets the active rig for subsequent commands. Use when more than one rig is

controlled by the computerHAM_API_Rig_getSettings() Returns—in a standard data structure—current rig settings: frequency, mode, split, etcHAM_API_Rig_getFrequency() Returns the rig frequenciesHAM_API_Rig_setFrequency() Sets the rig frequencyHAM_API_Rig_setEventFunction() Sets the function to be executed if a rig generated event happens, ie, frequency

changed from rig’s front panelHAM_API_Rig_getEvent() Returns the event that caused the setEventFunction to be activatedHAM_API_Rig_setRIT() Sets the receive incremental tuning (RIT)HAM_API_Rig_setXIT() Sets the transmit incremental tuning (XIT)HAM_API_Rig_setMode() Sets the rig’s modeHAM_API_Rig_getMode() Get the rig’s mode

1Linux Application Development, by MichaelK. Johnson and Erik W. Troan, publishedby Addison Wesley Longman Inc, 1998.

Linux Application Development1 de-scribes Linux’s shared libraries andhow to implement them. Each HamAPI device-specific library would beimplemented as a shared library.

WindowsWin16 (the formal name for the Win

3.1X environment) and Win32 (Win 9Xand NT) provide support for run-timelibraries. In the Windows environ-ment, these libraries are called Dy-namic Link Libraries (DLLs). Theirfile name extension is “.dll”. A goodportion of the Windows operating sys-tem is implemented in DLLs. EachHam API device-specific library wouldbe implemented as a DLL.

DOSThe Microsoft DOS operating envi-

ronment presents an interestingchallenge when trying to implement

How Do We Implement Them?Linux

Linux and all UNIX derivatives sup-port run-time libraries. A chapter in

Jan/Feb 1999 21

standard libraries supporting the API.No one standard run-time-librarymodule has emerged. Standard link-ing libraries are used at compile andlink time, but often, no real run-timelibrary module exists. Instead, soft-ware developers have made use of thearchitecture of the Intel iAPX-86 fam-ily, for which DOS was developed.They use the same method by whichDOS communicates with underlyingbasic input/output system (BIOS)firmware and software; that is, viasoftware interrupts and terminate-and-stay-resident (TSR) techniques.

The Intel iAPX-86 architecture pro-vides support for up to 256 softwareinterrupts. Like their close cousins,hardware interrupts, software inter-rupts are invoked by asserting an in-terrupt request (IRQ). Instead of be-ing generated via hardware, softwareinterrupts are requested through soft-ware instructions. For example, DOSprovides a function for printing a char-acter to the standard output device. Itis invoked by the following fragmentof iAPX-86 assembly language code:mov al,32 ; move an ASCII 32 (space)

; Into the <AL> registermov al,dl ; in <DL> for DOS call

Table 3A Sample API for Interfacing to a PacketCluster

Functional DescriptionHAM_API_Packet_Cluster_login() Login in to the PacketClusterHAM_API_Packet_Cluster_setName() Set the operator’s nameHAM_API_Packet_Cluster_setQTH() Set the operator’s QTHHAM_API_Packet_Cluster_doSetCommand() Sends a set command to the PacketCluster. The command is contained

in a standard data structure that also contains the result of the commandHAM_API_Packet_Cluster_doDirCommand() Sends a Dir command to the PacketCluster. The command is contained

in a standard data structure that also contains the result of the commandHAM_API_Packet_Cluster_doShowCommand() Sends a Show command to the PacketCluster. The command is contained in

a standard data structure that also contains the result of the commandHAM_API_Packet_Cluster_delete() Sends a command to delete a mail messageHAM_API_Packet_Cluster_send() Sends a mail messageHAM_API_Packet_Cluster_announce() Sends an announcement. The type of announcement is contained in

standard data structureHAM_API_Packet_Cluster_quit() Sends the command to log off the nodeHAM_API_Packet_Cluster_dx() Announces a DX stationHAM_API_Packet_Cluster_reply() Reply to a read messageHAM_API_Packet_Cluster_talk() Enter talk modeHAM_API_Packet_Cluster_type() Enter a command to display a file. File contents are returned in the data structureHAM_API_Packet_Cluster_upload() Uploads a bulletin fileHAM_API_Packet_Cluster_wwv() Gets the solar fluxHAM_API_Packet_Cluster_read() Sends a command to read a message into the data structure.

mov ah,2 ; destination in standard outputint 21h ; execute the DOS library call

The HAM_API library can be devel-oped in a similar way. A suitable, un-used interrupt in the DOS architec-ture would have to be chosen. To allowportability across machines, thisvalue should be set by a SET commandat boot time. For example, the AX reg-ister pair would provide 256 differentAPI families, each with 256 differentfunctions. The API TSR would be adispatcher that loads and invokes spe-cific APIs as required, as configured bySET commands.

Once the API is completed, volun-teers would develop reference imple-mentations for use by application de-velopers. If these are successful, theworking group develops conformance-testing criteria to certify that APIimplementations meet the standard.The ARRL then publishes the standard.

After publication, the working groupconvenes to maintain the standard on aregular basis. As APIs are imple-mented, lessons will be learned andimprovements made in the functionsand descriptions.

Where Do We Go from Here?To ensure that this approach is suit-

able for Amateur Radio, discussion isneeded. Comments and observationsare needed from application develop-ers, product developers and amateurson the feasibility of this approach.Once agreement on its viability isreached, ARRL-sponsored workinggroups should be created to developthe respective API descriptions. Theseworking groups need not meet physi-cally to develop the standard, but canuse the Internet and Amateur Radiofor communications.

Lawrence G. Dobranski, VA3LGD,has a BS (with honors) in Engineering-Physics from Dalhousie University inHalifax, Nova Scotia, and a MS (Engi-neering) in Physics from Queen’s Uni-versity in Kingston, Ontario. He is pres-ently a Senior Consultant with theEXOCOM Group of Companies in Ot-tawa, Ontario, specializing in Informa-tion Technology Security. Lawrence hasbeen involved in the standards develop-ment of APIs for Information Technol-ogy Security Services. Lawrence’s inter-ests lie in the technical side of ham ra-dio. He operates mainly HF mobile, withsome dreams of serious contesting.

22 QEX

Do you want a little power for the magicband? This small box will do the trick!

By Dick Hanson, K5AND

7540 Williamsburg DrCumming, GA 30041k5and@prestige.netTel 770-844-7002 (eves)fax 770-889-8297

A Pair of 3CX800sfor 6 Meters

1Notes appear on page 26.

Editor’s Note: This article also appears in The Proceedingsof the 1998 Southeastern VHF Society Conference,published by the ARRL.

It’s that time of year when some of us begin thinkingabout the F2 layer, VHF DXpeditions and other fun stuffrelating to 6 meters—especially with the sunspot cycle onits way back up. While it is true that you can have a lot offun on this band with 10 W, more is most often better whenyou’re trying to make a contact 12,000 miles away as theband is fading, fading, fading....

As most of you know, there aren’t many commercialamplifiers being manufactured today for 50 MHz. Since younever know how many more sun-spot cycles you have, Idecided to build a legal-limit unit to suit my needs. Havinghad good luck with 3CX800s in other amplifier projects,and because these tubes are available at hamfests and fromdealers,1 they are my tube of choice.

Design Features• Two-part construction: RF deck separate from high-

voltage supply• Small package: 6×13×12 inches (HWD)

• T-network input presents good match to transceiver• Pi-L output network for good harmonic suppression• Pressurized anode compartment for efficient cooling• Transistor-keyed PTT line works with +12 V or ground to

transmit• Adjustable time delay on PTT line to allow for cathode

warm-up• Grid-trip circuit to protect grids• PTT inhibited without B+ present

Circuit Details and Mechanical ConstructionNot being an engineer, I depend on books, articles and

technical friends for guidance. I also fall back on 43 yearsof building experience: memories of what doesn’t work. Onenifty new design tool—at least for me—is software fromChuck Reichert, KD9JQ,2 which really streamlines thedesign process for triode amplifiers. Before I got this pack-age, called TAP2, my projects involved lots of trial anderror. Now things are much more predictable, and timeframes for completing projects are shortened. Life is good!

Notice below the values for the input network Q, plateload impedance and output Q. Also notice that the com-bined output capacitance of the tubes —about 12 pF—hasbeen subtracted from the total plate capacitance required.Tables 1, 2 and 3 (from the software report generator) showthe design parameters for various anode voltages. As youcan see from the reports, the plate load changes some withlower anode voltages.

After selecting your operating parameters, you canbegin the component-selection process and plan the chas-sis configuration.

Jan/Feb 1999 23

Circuit DetailsRefer to Figs 1 and 2. A coaxial relay (K4) at the input

maintains low SWR in the bypass mode. A tuned T inputnetwork matches transceiver outputs from 50 to 27 Ω. Avacuum plate-tuning capacitor is used in this design. Airvariables are also fine, but you need one that has littleminimum capacitance with a respectable working-voltage

Table 1SW.I.F.T. Enterprises©1989Triode Amplifier Program Version 1.2For Grounded Grid OperationCourtesy of KD9JQ(2) AB2 Biased 3CX800A7 @ 50.0 MHzRated for Forced Air

DC Plate Voltage (VP)* = 2800.0 VMax Plate Voltage = 2800.0 VPeak Plate Swing = 2550.0 VMin Plate Voltage = 250.0 VPlate Current Peak = 3.137 APlate Current dc = 1.056 AGrid Current dc = 0.056 ACath Current Peak = 3.398 ADesign Plate RL = 1625.6 ΩRL for Matching = 1654.6 ΩPlate Diss (PD)* = 956.9 WGrid Diss (PG)* = 3.0 WCathode Bias = 10.8 VPeak Grid Voltage = 34.7 VZin at Cathode = 26.8 ΩPin Drive (PEP) = 38.6 WPo @ Plate (PEP) = 2000.0 WPo to Load (PEP) = 2035.7 WDC Power Input = 2956.9 WEfficiency = 67.6 %Power Gain = 17.2 dBCath to Grid Cap = 52.00 pFPlate to Grid Cap = 12.20 pFConduction Angle = 190.4 Deg* Maximum allowable VP = 2500; PD = 1600 W; PG = 8 W

T Input NetworkRS 50.0 ΩL2 0.540 μHC1 42.688 pFL1 0.458 μHRFC 0.195 μHZin 24.1 –j 11.7 ΩQL 5.0F0 50.00 MHz

Pi-L Output NetworkRP 1654.6 ΩRFC 3.176 μHC1 14.075 pFL1 0.614 μHC2A + C2B 78. 889 pFRX 285.1 ΩL2 0.345 μHRL 50.0 ΩQL 12.0

Table 2SW.I.F.T. Enterprises©1989Triode Amplifier Program Version 1.2For Grounded Grid OperationCourtesy of KD9JQ(2) AB2 Biased 3CX800A7 @ 50.0 MHzRated for Forced Air

DC Plate Voltage (VP)* = 2400.0 VMax Plate Voltage = 2400.0 VPeak Plate Swing = 2150.0 VMin Plate Voltage = 250.0 VPlate Current Peak = 3.721 APlate Current dc = 1.253 AGrid Current dc = 0.072 ACath Current Peak = 4.043 ADesign Plate RL = 1155.6 ΩRL for Matching = 1181.6 ΩPlate Diss (PD)* = 1006.0 WGrid Diss (PG)* = 3.9 WCathode Bias = 8.3 VPeak Grid Voltage = 40.0 VZin @ Cathode = 23.9 ΩPin Drive (PEP) = 48.8 WPo @ Plate (PEP) = 2000.0 WPo to Load (PEP) = 2044.9 WDC Power Input = 3006.0 WEfficiency = 66.5 %Power Gain = 16.2 dBCath to Grid Cap = 52.00 pFPlate to Grid Cap = 12.20 pFConduction Angle = 190.4 Deg*Maximum allowable (VP) = 2500; (PD) = 1600 W; (PG) = 8 W

T Input NetworkRS 50.0 ΩL2 0.514 μHC1 45.831 pFL1 0.409 μHRFC 0.195 μHZin 22.0 –j 9.3 ΩQL 5.0

Pi-L Output NetworkRP 1181.6 ΩRFC 2.268 μHC1 24.595 pFL1 0.450 μHC2A + C2B 96.531 pFRX 240.4 ΩL2 0.311 μHRL 50.0 ΩQL 12.0

rating. A vacuum output relay (K5) provides QSK opera-tion, although an open-frame relay with the proper voltageand current ratings is okay if you don’t desire QSK.

Mechanical ConstructionFigs 3, 4, 5, 6, 7 and 8 show some construction details.

Some readers will no doubt recognize the chassis assem-

24 QEX

Table 3SW.I.F.T. Enterprises ©1989Triode Amplifier Program Version 1.2For Grounded Grid OperationCourtesy of KD9JQ(2) AB2 Biased 3CX800A7 @ 50.0 MHzRated for Forced Air

DC Plate Voltage (VP)* = 2250.0 VMax Plate Voltage = 2250.0 VPeak Plate Swing = 2000.0 VMin Plate Voltage = 250.0 VPlate Current Peak = 3.000 APlate Current dc = 1.010 AGrid Current dc = 0.055 ACath Current Peak = 3.250 ADesign Plate RL = 1333.3 ΩRL for Matching = 1361.1 ΩPlate Diss (PD)* = 772.1 WGrid Diss (PG)* = 2.6 WCathode Bias = 8.1 VPeak Grid Voltage = 33.5 VZin @ Cathode = 25.6 ΩPin Drive (PEP) = 33.8 WPO @ Plate (PEP) = 1500.0 WPO to Load (PEP) = 1531.2 WDC Power Input = 2272.1 WEfficiency = 66.0 %Power Gain = 16.6 dBCath to Grid Cap = 52.00 pFPlate to Grid Cap = 12.20 pFConduction Angle = 190.4 Deg*Maximum allowable (VP) = 2500; (PD) = 1600 W; (PG) = 8 W

T Input NetworkRS 50.0 ΩL2 0.530 μHC1 43.848 pFL1 0.438 μHRFC 0.195 μHZin 23.3 –j10.7 ΩQL 5.0

Pi-L Output NetworkRP 1361.1 ΩRFC 2.613 μHC1 19.742 pFL1 0.512 μHC2A + C2B 88.627 pFRX 258.2 ΩL2 0.325 μHRL 50.0 ΩQL 12.0

blies as modified Down East3 components. I have also hadexcellent results using chassis made by Charles Byers,K3IWK.4

I personally favor smaller, rather than larger, chassis.So the almost-ready-to-go units from Down East are attrac-tive. The only problem with chassis this small is where toput all the control stuff, the filament transformer, etc.Honestly, it would be easier with slightly more room.

Amplifier Control CircuitThe control board is the combination of several circuits that

I have used or seen in other projects. The board is now com-mercially available from FAR Circuits,5 and it provides aneat, compact layout (2×4 inches). This circuitry is designedto satisfy the various switching requirements of popular tri-odes used in modern amplifiers, while providing protectionagainst their worst enemies: loss of anode voltage, insuffi-cient warm-up time and excessive grid current.

Control Logic OperationAfter you switch the power on, nothing else happens until

time-delay relay K1 closes and high voltage is present. TheTR sequence can’t begin unless both of these conditions aremet. This is good, because you don’t want to apply driveuntil the cathode has reached its operating temperatureand B+ is present. If you apply even a small amount of driveto a triode without B+ present, the grid will draw lots ofcurrent, which could ruin the tube instantly—not a pleas-ant thought!

So, when you switch on the ac power, it starts the follow-ing sequence: The blower starts. The time-delay relay be-gins its cycle. The green AC ON LED illuminates. The B+and PTT lines are held open for a preset delay time (fourminutes for example). Then the delay relay, K1, closes,which applies 24 V dc to the various control transistors,and the AMP READY LED illuminates. The other set of K1contacts makes 120 V ac available to remotely control thehigh-voltage power supply.

If anode voltage is present, Q1 turns on when its basevoltage reaches 0.7 V. R6 sets the turn-on voltage; lower-ing the resistance of R6 raises the voltage necessary to turnon Q1. With the value shown in the schematic, the turn-onpoint is roughly 1400 V.

Q3 is one of two TR switching transistors. Grounding itsbase causes relays K3 through K5 to switch. Maximumcurrent drawn from the transceiver’s TR circuit is in the 7to 10 mA range. Q4, the other TR switching transistor,conducts when a nominal +12 V is applied to its base cir-cuit. This, likewise, causes the TR relays to switch as above.So both +12 V (active high) and grounding (active low) key-ing outputs can control the amplifier. Regardless ofwhether Q3 or Q4 is chosen to switch the amplifier, theTRANSMIT LED will light and the TR relays will switch.

Grid over-current protection is provided by Q5 and K2.R15 allows the trip current to be set in the range of 40 to120 mA, which should cover most tube combinations. Whenthe preset trip current is reached, Q5 conducts closing andlatching K2, the over-current relay. One set of the DPDTcontacts locks the relay on; the other set opens the +24-Vline to all relays so that the amplifier is taken off line. ATRIP LED also illuminates to indicate the over-currentcondition. Depressing normally closed switch, S3, resetsthe trip circuit.

AlignmentOne way I have used to adjust the trip value without

having to apply B+ or drive is as follows: Apply +5 Vthrough a 1-kΩ, current-limiting pot to the junction of M1and M2. Start with R15 at maximum resistance. Graduallydecrease R15—whilst keeping an eye on the grid current—until the circuit trips at the value you have selected. Youwill most likely need to reset the circuit several times (atleast) while you’re fooling around. Mine is set to trip at80 mA. During actual operation, you may notice that theLED begins to flicker as grid peaks approach the trip value.

Jan/Feb 1999 25

Hint: Watch the LED for an indication of too much drive,too little loading or both.

A handy tip for testing all of the above circuitry withoutB+ present is to temporarily connect +5 V—through a 1-kΩresistor—to the junction of R5 and R6. This provides thenecessary base voltage to turn on Q1, enabling the rest ofthe circuitry for testing purposes without high voltagelurking around.

The input and output circuits can be “fiddled” into rangein several ways. You can use a conventional grid-dipmeter—or you can use the new MFJ-259 do-almost-every-thing box. I will describe how to use the ’259 to tune boththe input and output circuits. This is done with the tubesplugged in, but with no voltages applied.

To tune the input, simply tack a 27-Ω carbon resistorfrom the cathode strap to ground; connect the ’259 to theinput. Then adjust the capacitor and both coils for near zeroreflected power at 50.1 or 50.2 MHz. Be sure to remove theresistor when you’re done!

To tune the output, tack a 1.6-kΩ resistor from the platestrap to ground, connect the ’259 to the output, and startadjusting things until you achieve minimum SWR. The coildimensions can be calculated from the standard formula:

La n

a b=

+

2 2

9 10(Eq 1)

or

nL a b

a=

+( )18 40(Eq 2)

where a equals the coil radius, n equals the number of turns

and b equals the length of the coil. (Remember that n × wirediameter ≤ b!—Ed.) The L4 coil shouldn’t need adjustment,but L3 will probably require some. It would be nice if thecoil resonated with the plate capacitor near maximumcapacitance, so that you can easily move up the band bysimply adjusting the plate-tuning vacuum capacitor to-ward minimum capacitance. This would also indicate thatthe Q is pretty close to the calculated value. Remove theplate-tuning resistor after your adjustments are complete.

Having built one or two amplifiers before, I confess acertain amount of wide-eyed wonder (and pleasure) at see-ing an amplifier work right off the bat with only the minoradjustments I’ve described. At 6 meters yet!

Power-Supply RecommendationsYou can use anode voltages ranging from 2200 to 2800 V.

Just beware that the plate load impedance changes withanode voltage. For example, this amplifier is designed for2800 V with a plate-load impedance of about 1600 Ω. If youchange the plate voltage to say, 2400 V, the plate load willchange to 1181 Ω, and the output-tank values will changeaccordingly. (Refer to Table 2.) You might want to considerusing the provided remote ac control to switch the high-voltage power supply from the amplifier. It’s very handyand one less thing to remember when you’re in a hurry.

Parting ThoughtsThis amplifier works just like an HF amplifier, so the

tune-up procedure is very conventional. For 1500 W out-put, the drive power will be in the neighborhood of 45 W.Grid current will be about 40 to 50 mA, and plate current

Fig 1—Schematic of the 6-meter amplifier RF deck. B+ is 2200 to 2800 V at 2 A. The coils of K4 and K5 are shown in Fig 2. Use1/4 W, 5%-tolerance carbon-composition or metal-film resistors unless otherwise specified. Equivalent parts may besubstituted.

L1—10 turns, #16 AWG, 1/2-inch diameter, 1-inch longL2—7 turns, #16 AWG, 1/2-inch diameter, 5/8-inch longL3—33/4 turns, 1/4-inch-diameter tubing,

13/4-inch diameter, 3 inches longL4—5 turns, #6 AWG, 1-inch diameter, 13/4-inch longRFC1, 3, 4—7 μHRFC2—40 turns, #22 AWG on 1/2-inch

diameter Teflon rod, winding is 13/8-inch long; alternative, 34 turns of #18 AWG on a 3/4-inch-diameter fiber- glass threaded form.RFC5—19 bifilar turns, #18 AWG on 5/8-inch-diameter form

26 QEX

Fig 2—Schematic of the 6-meter amplifier control circuits. The contacts of K4 and K5 are shown in Fig 1. All unmarked diodesare 1N4001s. LEDs do not have part numbers, but they are labeled with their function and color. Use 1/4 W, 5%-tolerancecarbon-composition or metal-film resistors unless otherwise specified. Equivalent parts may be substituted.

B—Dayton 4C440 blowerK1—120 V ac coil, time-delay (0 to 5 minute), such as Omron H3Y-2K2, K3—24 V dc, DPDT units such as PB KHAE-17D12-24, OEG SRET-203DP or various Radio Shack equivalents

K4—24 V dc operated coaxial relay, available from Allen Bond (see Note 1).K5—24 V dc operated vacuum relay, available from Allen Bond (see Note 1).

T1—120 V primary, 18 V, 1.5 A secondary, such as Hammond K166K18T2—120 V primary, 13.5 V, 3 A secondary

will be about 1.2 A. These conditions represent an effi-ciency of 65% or so, which is fairly respectable. I alwaysuse an ICE #426 low-pass filter on the output of my high-power, 50-MHz amplifiers. I will be happy to answer ques-tions by telephone, fax or e-mail.

Special thanks to Chuck Reichert, Steve Kostro at DownEast Microwave and Pat Stein at Command Technologies forsoftware, weird parts, ideas and general support. Projectslike this don’t happen without help and encouragement; Iwas fortunate to have had both.Notes1Tubes: There are so many used tubes available now at hamfests

that you should be able to buy “guaranteed” pulls for $150 to $225a piece. One source is Allen Bond, who can be reached atmgs@avana.net or at 770-973-6251.

2Triode software: Download from KD9JQ’s site www.imaxx.net/~kd9jq.

3Chassis: Down East Microwave, Steve Kostro; tel 908-996-3584;www.down-eastmicrowave.com.

4Chassis: Byers Chassis Kits, Charles Byers; tel 717-292-4901 6-9pm EST; k3iwk@juno.com.

5PC Board: FAR Circuits; tel 847-836-9148; www.cl.ais.net/farcir.6Obviously, some of the parts used are from flea markets, junk box,

etc but the schematic captions show additional sources.

Dick was first licensed as WN0UUU in 1954. He completed aBA at the University of Texas in 1966. Dick worked with themedical electronics division of Hewlett-Packard for 18 years.He was the medical manager for the southern region when heleft to purchase Southern Staircase in 1983. Since then, hehas been the President and CEO of Southern Staircase.

Dick enjoys building amplifiers and antennas. He alsolikes operating on 6 and 2 meters from rare countries dur-ing F2 propagation. He has been to most of the Caribbeanislands for DXpeditions.

Jan/Feb 1999 27

Fig 4—The front subpanel. The filament transformer is at thelower right. K1, K2 and K3 are visible. The filament resistor(0.6 Ω) is visible above the left-hand meter.

Fig 5—A close-up of the RF compartment. Starting at thelower left, we see C1. The white can is C11 (vacuum variable),which connects to C8 and C9 (type 858 blocking capacitors)via a metal plate. The large (silver plated) coils are L3 and L4.The darker, wire coil is RFC2 (on a threaded fiberglass form).The output vacuum relay (top left center) is labeled K5 on thechassis. The two 3CX800A7s are at right, with homebrewTeflon chimneys.

Fig 6—Control-circuits PC board. The coaxial input relay ison the chassis at the upper left of the PC board.

Fig 8—The amplifier rear panel. The unmarked banana jack isfor the B–. RF OUT is a Teflon-insulated SO-239. A multipinconnector for the ac, remote ac to high-voltage supply andblower connection is visible below the blower. Notice theblower plenum extension that permits the blower to clear themultipin plug. You can eliminate the plenum by orienting theblower 90° or 180° (ccw) from the position shown.

Fig 7—The cathode compartment shows RFC1 and RFC5.Rings of holes around the sockets allow air past the tubebases to the cooling fins.

Fig 3—A view of the amplifier front panel (without labels orpaint). There are three horizontal rows of controls andindicators: Top row (left to right): PLATE LOAD, GRID CURRENTmeter, PLATE TUNE and PLATE CURRENT meter. Second row(LEDs, left to right): AC ON, AMP IN/OUT, READY, TRANSMIT andGRID TRIP. Third row (switches, left to right): POWER, AMP IN/OUT and RESET.

28 QEX

Build this handy antenna tuner and tailor it toyour mobile station. A band-switched inductance

favors easy tuning while on the road.

By Patrick Wintheiser, W0OPW

12251 SE 59th St, #106Bellevue, WA 98006

A Compact Mobile Tuner

1Notes appear on page 30.

The Compact Mobile Tuner(CMT) is a very small, band-switched antenna tuner that is

intended for use with short verticalantennas. It can extend the bandwidthof a whip antenna so that you need notleave the car to make antenna adjust-ments on most HF bands. You can alsobuild it in a miniaturized form for QRPuse. Not only that, it is much moreefficient than you would expect in sucha small package. This small tuner eas-ily fits on the dash of a car and takeslittle space in a backpack. It is bandswitched so there is little hunting fora correct match. Just switch to thedesired band and adjust two variablecapacitors for the lowest SWR. Notice,however, that this tuner will not loada fence post in the desert! It is de-signed only for extending the usefulbandwidth of resonant antennas.

The need for this tuner arose lastwinter, when I went mobiling in theMidwest. My commercially made

T-network tuner is too bulky to be use-ful on the dashboard, and the band-changing adjustments became te-dious. In addition, I was never surewhether I was loading the antenna orthe tuner. In fact, I had my back-uplights flashing Morse code somewherein North Dakota!

I decided to see if it was possible tobuild a compact, band-switched tuner.The original design is by UlrichRhode,1 and I adapted from it freely.It is a low-pass L-network tuner for the80 through 10-meter bands. A typicalL-network design would use a roller in-ductor, which is monstrous for myneeds. Through some trial and error, Ireplaced the roller inductor with asmall toroid, and by adding a variablecapacitor in series with the output, Iwas able to extend the tuning range.The space and weight savings are wellworth the effort.

This tuner is capable of matchingalmost any vertical antenna that hasbeen resonated at one point in theband. It extends the tuning range of

short verticals beyond resonance, butonly so far. Using 75 meters as a worst-case example, I found that my mobileoperating bandwidth is now increasedby a factor of about four, before thelosses get excessive. In my case, that’sa tuning range of 160 kHz, comparedto 40 kHz without the tuner. On higherfrequencies, I can safely tune acrossan entire band. The tuning range oflonger verticals is much greater.

Please notice that 160 kHz is theuseful bandwidth with this tuner.This—or any—tuner may show a1:1 SWR over a much greater band-width, where tuner losses are muchgreater than the antenna’s radiationresistance. Because of this, theradiated power does drop off whenattempting to load highly reactive an-tennas beyond what Mother Natureallows. I’ve determined this with a verynoisy sodium-vapor lamp above mytown-house patio, located about 12 feetfrom the mobile antenna. It spews ra-diation across a very wide bandwidth,and I can use my S meter (on receive) tomonitor the power loss away from reso-nance. It is amazing to see the receive

Jan/Feb 1999 29

signal drop off 3 to 4 S-units eventhough the SWR bridge indicates a 1:1SWR. (My commercially made T-net-work tuner is even worse under theseconditions. A recent QEX2 articleshows why. It’s interesting to note thata T-network design can have seventimes the loss of a simple L-network.

Construction NotesBegin by winding the 4:1 toroid

transformer, T1, which steps the 50-Ωtransceiver output down to the typicalimpedance of a short vertical. (If yourshort HF vertical is a nice match for50 Ω all by itself, it is very lossy!—Ed.)I used 12 turns of RG-174 coax on anFT-114-43 ferrite toroid. The centerconductor is the primary and the shieldis the secondary. (The two windings areseries connected to form an autotrans-former, as shown in Fig 1.—Ed.)

After winding the core, tie the twoends of the coax together with somestring and liberally coat the windingwith glue to hold the turns in place.Let the glue dry overnight. Next, glueT1 on the back of the cabinet some-where out of the way.

Wind tuning inductor L1 with 22turns of #14 enameled copper wire.Glue L1 onto the bottom of the caseunder the band switch, S1. If you donot plan to use this tuner on 75 meters,you can eliminate L1 and use the ex-tra space to increase the size of L2, ifdesired.

L2 is an air-core inductor (81/2 turnsof #14 wire, about 11/2 inches long).This winding adds about 1.5 μH of in-ductance for the 40 to 10-meter range.Wind L2 using a plastic 35-mm filmcanister as a form: Tape some waxedpaper around the form and add bind-ing screws at each end of the form tohold the wires in place. Wind the turnson the form and then wrap 1/16--inchnylon braid between the wire turns toevenly space them. Use 5-Minute ep-oxy to secure the turns in place and letthe coil dry overnight. Remove thebinding screws in the morning; the coilshould easily slide from the form. Re-move the waxed paper, but you canleave the nylon braid in place, as partof the coil.

For the enclosure, I used a 71/2×41/4×23/8-inch plastic box from OceanState Electronics.3 If you use a metalenclosure, be sure to well insulate theinductors and capacitors from thecase: The shaft of C2 is “hot” and needsto be well insulated.

Examine the enclosure and planwhere to drill the holes for the coaxand connector (the back of the box), the

variable capacitors and band switch(the front). I did not use a connectorfor the line to the transceiver, choos-ing instead to hard-wire the coax di-rectly into the circuit. This eliminatesa connector from the box, and bymounting the input and output con-nections directly behind C1, keeps theground connections very short. ADPDT tuner-bypass switch is op-tional, but it’s handy when operatingnear resonance, or when you just wantto see what difference the tunermakes. C1 and C2 are heavy-duty 20to 400 pF units from Fair Radio Sales4

that will safely handle several hun-dred watts. Broadcast-receiver tuningcapacitors will also work fine, andthey’re much smaller. The physicalsize of C1 and C2 determines the sizeof the enclosure. Unfortunately, itseems that variable capacitors onlycome in two sizes; too big or too small!If you operate at higher power levels,use larger toroids for T1 and L1. S1can be anything from 1P6T to 1P12T,depending on how many taps youneed.

AdjustmentsIf you use the recommended compo-

nents, you should be able to install mytap points on L2 without any modifi-cations. Otherwise, attach an SWRbridge to the tuner’s input side and aresonant antenna to the output. Withlow power (less than 5 W), apply RF tothe tuner. Watch the SWR and—withC1 and C2 both 1/2 to 2/3 meshed—trydifferent tap positions on L1 or L2.When a minimum occurs on the SWR

meter, solder a jumper from the S1 tothe tap. I need taps for 75, 40, 30, 20and 17 to 10 meters on the switch.Although 40 and 30 meters can betuned on one switch position, it maybe better to provide separate band-switch positions. The very last switchposition covers 17 through 10 meters.This tuner could be dedicated to anindividual antenna without switch-ing, but I’ve found that it matchesevery vertical I could test. An addi-tional surprise is that it also tunesdipoles just fine!

Test each band by loading the an-tenna every 50 kHz. If the antennawill not load sufficiently across eachband, then move the tap up or downthe coil until the loading is satisfac-tory. Note that this L-network designwill not tune antennas very far fromresonance. If C2 is fully open withouta match, the antenna is too reactive forthis tuner. In that case, a T-networkdesign might be better. However, be-ware of losses in any tuner when theoperating frequency is far removedfrom antenna resonance!

An OptionalOutput-Current Meter

A simple output-current (propor-tional to the square root of power)meter is a useful addition to thistuner—or to any tuner for that mat-ter. (See Fig 2.) I added one to the CMTby routing the wire from the output toS2B through a T-37-2 powdered-irontoroid to form a one-loop primary. Thesecondary is about 20 turns of #28enameled wire. For the sensing cir-

Fig 1—A schematic of the Compact Mobile Tuner.

C1, C2—20-400 pF variable capacitorL1—22 turns of #14 enameled copper

wire (for 75 meters) on a T-130-6, orlarger, powdered-iron toroid core.

L2—81/2 turns of #14 bare copper wireon a 11/4-inch OD form, with taps atturn 1 (bottom, for 40 meters), turn 4(30 meters), turn 6 (20 meters), turn81/2 (top, for 17-10 meters). See text.

T1—4:1 balun using 12 turns of RG-174coax on a FT-114-43, or larger, ferritetoroid. Connect the output end of theinner conductor to the input end ofthe outer conductor.

S1—1P6T rotary switch.

30 QEX

cuit, add a germanium diode (1N34,1N60, etc) and a 0.01 mF bypass ca-pacitor to the output of the secondaryand connect it to a 100 kΩ potentiom-eter and a 0 to 200 mA meter outsidethe tuner case. (See Fig 2. Any sensi-tive meter up to about 1 mA shouldwork just fine.)

I calibrated the meter face by set-ting the pot for a full-scale (100%)reading with a 100 W transceiver out-put feeding a resonant antenna withthe tuner bypassed. The other per-centage marks were established withlesser power settings. This simplecircuit and meter indicates relativeoutput power. In essence, if outputcurrent increases while adjusting thetuner (with constant input power), theantenna match is improving. Since Ihave band switched my tuner and Iknow that I’m capable of good trans-ceiver match, I’ve eliminated the SWRbridge from the system and simplytune for maximum output current.

If the output current decreaseswhile adjusting the tuner (with con-stant input power), the tuner is deliv-ering less power to the load and dissi-pating the difference! In a series ofarticles for QST, Maxwell5 describedthe reasons for using antenna inputcurrent as a means of monitoring theconjugate match.

I ran a comparison test against mycommercial T-network tuner. After ad-justing the CMT for minimum SWR on20 meters, I set the pot for a full-scalereading. I then switched the tuner outof the line, connected the T-networktuner and adjusted it for minimumSWR. By observing the output current,I could see that the T-network tuneroutput current was only 50 to 75%,relative to the CMT output, dependingon the switch and capacitor settings.This dramatically proves to me theadvantages of monitoring output cur-rent when using any tuner. It alsoshows that a simple L-network is aboutas efficient as a tuner can be. Using thecurrent meter, I also discovered thatany tuner loses power in attempts toimprove the match of an antenna withan SWR of 1.5:1 or less!

On the RoadI took my new tuner on the road to

the Midwest, and it performed ex-tremely well. I deliberately omittedthe SWR bridge and tuned for a maxi-mum on the output-current meter.

The tuner easily works on 75 through17 meters and takes up little dash-board real estate. I received excellentsignal reports, including 579 fromON7GB, while I was operating CW—from South Dakota! I was able to givehim his 50th state for 30-meter WAS,which was a thrill for me.

Notes1Ulrich Rhode, KA2WEU, “Some Ideas on

Antenna Couplers,” QST, Dec 1974,pp 27-30.

2Kevin Schmidt, W9CF, “Estimating T-Net-work Losses on 80 and 160 Meters,” QEXJul 1996, pp 16-20.

3Ocean State Electronics, PO Box 1458,6 Industrial Dr, Westerly, RI 02891; tel401-596-3080, 800-866-6626, fax 401-596-3590; www.oselectronics.com.

4Fair Radio Sales Co Inc, PO Box 1105,

Fig 2—A schematic of the optional output-current meter. T2 is a transformer with aone-turn primary formed by the wire from S2 to the output cable, which passesthrough the center of a T-37-2 powdered-iron toroid core. The secondary is 20turns of #28 enameled wire.

1016 E Eureka St, Lima, OH 45804; tel419-227-6573, 419-223-2196, fax 419-227-1313; fairadio@wcoil.com; URLhttp://www2.wcoil.com/~fairadio/.

5M. Walter Maxwell, W2DU, “Another Lookat Reflections, Part 7,” QST, Aug 1976,pp 15-20.

First licensed as KN0OIW in 1957, inSt Peter, Minnesota (recently devas-tated by the tornado of March 29,1998). He has been employed as a com-puter systems engineer for 32 years.Pat only recently became actively in-volved in Amateur Radio again afterbeing off the air for 20 years. He is amember of The Mike and Key ARC inRenton, Washington, and also oper-ates from his summer lake cabin inMeeker County, Minnesota.

Jan/Feb 1999 31

Do you want to measure the phase noise of your latestoscillator, but don’t own a sophisticated spectrumanalyzer? Learn how to measure oscillator phase

noise with equipment you can afford!

By Jos F. M. van der List, PA0JOZ

Fluitekruid 202201 SM NoordwijkNetherlandsjvdrlist@gironet.nl

Experiments withPhase-Noise Measurement

This article, was previously publishedin Dutch. It appears in the May andJune 1992 issues of Electron maga-zine, which is published by VERON theDutch IARU Society.—Ed.

Phase NoiseAn ideal oscillator produces a clean

and unmodulated signal. The output isa pure sine wave without changes inamplitude or in the time between thezero-crossings. If we observe such a sig-nal on an ideal spectrum analyzer, wesee a single spectral line. (See Fig 1.)

Real oscillators are different, how-

ever. Both the amplitude and the timebetween the zero-crossings are proneto noisy variations. (See Fig 2.) On thesame ideal spectrum analyzer, we seethe main spectral line accompanied bysidebands caused by amplitude andphase modulation, the results of thenoisy variations. (See Fig 3.) I muststress here that the spectrum analyzer

is not adding to the sidebands causedby the amplitude modulation and thephase modulation. Both types ofmodulation cause sidebands to appearon either side of the carrier, and thespectrum analyzer can only measurethe amplitude of the results.

The local oscillator (LO) is used in areceiver to convert signals from onefrequency to another, as representedin Fig 4. It’s evident that the signals

Fig 1—The one single spectral line of anideal oscillator.

Fig 2—Exaggerated amplitude andphase noise of a non-ideal oscillator.

32 QEX

converted to the IF frequency may benoisy, although the input signals areclean. The noise on the IF signals iscaused by the sideband noise of theLO. To show that this is true, we haveto make a little excursion into modu-lation theory. Before doing so, let’shave a few words about amplitudevariations on the LO signal.

In a well-designed mixer, the ampli-tude of the LO signal is much strongerthan the strongest signal to be con-verted. Also, the mixer gain—or loss—will not depend very much on the os-cillator amplitude. (See Fig 5.) Theconversion loss of a normal, double-balanced diode mixer is shown as afunction of the oscillator power. Notethat the conversion loss does notchange very much with the oscillatorpower above a certain threshold. Fromthis, it is demonstrated that ampli-tude variations of the LO do not playan important role in a well-designedmixer and LO combination. The levelof the mixed signal is directly propor-tional to that of the input signal, anddoes not depend on the level of the LO.

Let’s return to Fig 4. In this figure,we can see the adverse effects of whatis called “reciprocal mixing.” This iswhere the noise sidebands of the LOmix with a strong, nearby signal toproduce noise in the IF. The noise de-

grades the signal-to-noise ratio (S/N)of any desired signal there. Even if theIF bandpass filter is an ideal “brick-wall” type, the receiver selectivitysuffers as a result of the reciprocalmixing effect. This is the reason for theattention paid to reciprocal mixingand to phase-noise reduction on oscil-lator signals.

The adverse effects of phase noiseplay their roles from the transmittingend, also. (See Fig 6.) In this example,the receiver is considered to be ideal,but one of the received signals hasnoise sidebands. When this signal isconverted to the IF, we have the sameproblem as in Fig 4. Now the S/N of theweaker signal is degraded by the noiseon the nearby, strong, undesired sig-nal. When our ideal receiver convertsall signals linearly to the IF, the noiseon the adjacent-channel signal causesinterference to the weaker, desiredsignal.

Frequency andPhase Modulation

To gain a better understanding ofthe deleterious effects caused by recip-rocal mixing, it is useful to have a lookat some of the principles of modula-tion, especially phase and frequencymodulation. Although we will ulti-mately deal with noise as a modula-

tion source, it is easier to begin byusing sinusoidal signals. This makesthe figures easier to understand andmakes the calculations easier.

The output signal of an oscillatorcan be written as:

U t A S t f S S( ) = ( ) ( ) + ( )[ ]cos 2π φ (Eq 1)where A is the amplitude of the oscil-lator. If A were a function of the modu-lating signal, we would be talkingabout amplitude modulation. Thistype of modulation is not discussedfurther here.

f is the frequency of the oscillator.When f is a function of the modulatingsignal, we are talking about frequencymodulation. The symbol φ is the rela-tive phase of the oscillator signal.From the fact that both f and φ deter-mine the argument of the cosine func-tion, it is obvious that frequency andphase modulation are close kin. Bothtypes of modulation are forms of“angle modulation.” In this format,the angle is made a function of themodulating signal.

For example: Suppose that thephase portion of the argument φ(S)=2000 π t, then:

U t A f t t( ) = +[ ]cos 2 2000π π (Eq 2)and this can be rewritten as:

U t A t f( ) = +( )[ ]cos 2 1000π (Eq 3)

Fig 3—Spectrum of a non-ideal oscillator.

Fig 4—Mixing in a receiver with a non-ideal oscillator.

Fig 5—Conversion loss of a double-balanced diode mixer as a function ofLO power.

Fig 6—The effect of reciprocal mixing: A weak input signal is masked by the phasenoise from a nearby strong signal.

Jan/Feb 1999 33

The frequency of the oscillator hasbecome 1000 Hz higher. In otherwords, a constant increase in the rateof change of phase is the same as ahigher frequency; a constant decreasein the rate of change of phase is thesame as a lower frequency. In math-ematical terms, we say that frequencyis the derivative of phase.

If we deal with sinusoidal informa-tion, it can be shown mathematically(for angle modulation) that the modu-lated signal can be represented as:

U t A f t m f tm( ) = + ( )[ ]cos sin2 2π π (Eq 4)

with fm the frequency of the modulat-ing signal, and m the so-called modu-lation index. For FM, the modulationindex is:

mpeak deviation=

⎛⎝⎜

⎞⎠⎟modulation frequency (Eq 5)

For PM, the modulation index is aconstant, and represents the peakphase deviation, in radians. The peakfrequency deviation of a phase-modu-lated signal is fm × m. For PM, andusing a constant-level modulationsource, the peak deviation is propor-tional to the modulating frequency.

For single-tone modulation, there isno way to tell the difference betweenfrequency and phase modulation. Asan example, I have recorded a spec-trum analyzer plot of an angle-modu-lated signal. (See Fig 7.) The carrierfrequency is 100 MHz and the modu-lating frequency is 2 kHz, as is evidentfrom the spacing of the sideband com-ponents. The modulation index is four,so the frequency deviation is 8 kHz andthe phase deviation is four radians.

Note that this signal has many side-band components. Theoretically, anangle-modulated signal has an infi-nite number of sidebands. In practice,happily, the strength of the compo-nents far from the carrier diminishesvery rapidly, and they disappear in thenoise. On the plot, the level of corre-sponding components at either side ofthe carrier is equal, and the spacing ofthe components is equal to the modu-lating frequency. The amplitude ofeach of the sidebands can be calcu-lated using so-called Bessel functions.

In Fig 8, these functions are plottedgraphically. When discussing phasenoise, we normally deal with phase-modulated carriers having low modu-lation indices, ie, below 0.5. For thislow index, the amplitude of the carrierand the first few sidebands can be esti-mated with sufficient accuracy fromthe following formulas: J0 = 1, whichmeans that the amplitude of the car-

rier is virtually equal to the amplitudeof the unmodulated carrier; J1 = m / 2,which means that the amplitude of thefirst pair of sidebands is well below theamplitude of the carrier (–12 dB for m= 0.5); J2 = m2 / 8 ( –30 dB for m = 0.5);J3 = m2 / 48, etc.

As an example, I have plotted thespectrum of an angle-modulated signalat 100 MHz as Fig 9. The modulationfrequency is 2 kHz again, but the fre-quency deviation is only 400 Hz. So, themodulation index is 0.2, and you can

verify that the above estimates of thesideband strengths are good. It is alsoevident that the occupied bandwidth ofthe signal is very much less than thatof the signal in Fig 7. One can say thatfor very low modulation indices, onlythe first sideband pair is significant. Inthis example, the second pair is morethan 20 dB below the first pair.

Application to Sideband NoiseAfter this somewhat theoretical

part, let’s go back to our subject. If we

Fig 7—Spectrum of an FM-modulated signal at 100 MHz; the modulation frequencyis 2 kHz, deviation is 8 kHz, modulation index is 4. The reference level of thespectrum analyzer is adjusted to the amplitude of the unmodulated carrier.

Fig 8—Graphic representation of the Bessel functions that are used to calculatethe amplitude of the carrier and the sidebands of an angle-modulated signal.

34 QEX

regard the signal in Fig 9 as the inputsignal of a receiver and mix it withan ideal LO, we get a signal at thedifference frequency—10.7 MHz, forexample—that retains the modulationcharacteristics. (See Figs 10 and 11.)The relationship between the carrierand the sideband components dependson the modulation, which is notchanged by the mixing process. Themixer in this example is performing afrequency translation: IF = RF – LO.The result is that, at the IF, one cannottell whether any FM or PM was causedby the input signal, or by the LO.

As I mentioned, these kinds of ex-amples are more easily understood ifwe use only sinusoidal modulation. Thesame processes, of course, are alsovalid for more complex modulationsources, such as noise. One thing hasto be clear when working with noise:The way we specify the difference be-tween the carrier and the sideband lev-els. With sinusoidal modulation, it iseasy to measure the difference betweenthe carrier and sidebands, since theyare each found at discrete frequencies.With noise, it is somewhat more com-plicated. The amount of noise that wemeasure depends on the measurementbandwidth—the power is directly pro-portional to the bandwidth.

Suppose we measure an oscillatorsignal at 100 MHz with a carrier level

of +10 dBm, or 10 mW. In a measure-ment bandwidth of 1 kHz, we measurea noise level of –50 dBm at 10 kHz fromthe carrier. This is 60 dB down. If werepeat the measurement with a band-width of 100 Hz, the result would be anoise power of –60 dBm, and that is70 dB below the carrier. Therefore, weusually normalize the results of noisepower measurements to a bandwidthof 1 Hz. In this example, the normal-ized noise-power level would be–80 dBm, and that is 90 dB below thecarrier level, thus –90 dBc/Hz. Ofcourse, it is not necessary to measurein a 1 Hz bandwidth, since the normal-ization is easily performed.

Now for one more example in whichall the previous information is used.Suppose we have a receiver with an IFof 9 MHz, a LO at 41 MHz, a preampli-fier at 50 MHz with a gain of 20 dB anda mixer with a conversion loss of 6 dB.(See Fig 12.) The LO has phase noiseof –100 dBc/Hz at 20 kHz from the car-rier. There are two signals at the in-put of the receiver, one at 50.000 MHzwith a level of –120 dBm, and the otherat 50.020 MHz with a level of –50 dBm.According to the IARU S-meter stan-dard for VHF, the first signal isapproximately S4 or S5, the secondsignal is S9 + 43 dB. The input signalat 50.000 MHz is converted to exactly9 MHz and it falls at the center of the

IF bandwidth. After the mixer, itslevel is –106 dBm (–120 dBm + 20 –6 dB). The second signal is convertedto 9.020 MHz, outside the passband ofthe IF filter; it has a level of –36 dBm.It seems that there is no problem. Theweak signal in the IF passband is re-ceived, and the strong signal outsidethe IF passband is rejected by the fil-ter. The difference between the twolevels is 70 dB, so it seems that anyregular IF filter (with an attenuationgreater than –70 dB at 20 kHz fromthe center) can do a fine job.

If we consider the reciprocal mixingeffect, the result is quite different. At20 kHz from the stronger signal, thenoise level is –100 dBc/Hz. In a 2.5 kHzbandwidth, this is 34 dB more, or–66 dBc / 2.5 kHz. This means that thestrong signal causes noise in the IFpassband at a level of –36 – 66 dBm =–102 dBm—4 dB stronger than theweak signal in the passband. Theweak signal is masked by the noisefrom the LO. From this example, it isvery clear that not only the selectivityof the IF filters determines the overallselectivity, but also the phase noise ofthe LO(s). Both transmitters and re-ceivers need to be considered.

Measuring Phase NoiseNow that we have shown the impor-

tance of phase noise, we can pose the

Fig 9—Spectrum of an FM-modulated signal at 100 MHz; the modulation frequencyis 2 kHz, deviation is 400 Hz, modulation index is 0.2. The reference level of thespectrum analyzer is adjusted to the amplitude of the unmodulated carrier.

Fig 10—FM modulated signal at 100MHz mixed to an IF of 10.7 MHz.

Fig 11—Unmodulated signal at 100 MHzmixed to an IF of 10.7 MHz by an FM-modulated LO.

Jan/Feb 1999 35

question: How should we measure thephase noise levels of our oscillators? Ipresent some of the possible methodsbelow.

First, because phase noise can beregarded as unwanted frequency mod-ulation, we could try to measure thedeviation. This is not very practical,because the deviation is liable to beonly a few hertz, and deviation metersare not suitable for these low levels.Furthermore, it would be difficult toget any information about the relationbetween the deviation and the modu-lating frequency.

Second, we could use a receiver or aspectrum analyzer to perform themeasurement. This might be the bestchoice for oscillators that are notvery good. Because of the phase-noiselevels of the LOs used in spectrumanalyzers, the sensitivity of the mea-surement is limited. Reciprocal mix-ing also occurs in spectrum analyzers!This method also measures the ampli-tude noise of the oscillator under test.Besides, not everyone has a spectrumanalyzer at home.

Alternatively, we could use the signalunder test as the LO in a receiver, anduse a very clean signal from a crystaloscillator as the receiver’s input signal.From the reciprocal mixing effects mea-sured, it is possible to calculate thephase-noise level of the oscillator. Thisarrangement works fine, and I use itwhen performing selectivity measure-ments on commercial receivers. Themethod is limited, however, by the IFfilter’s selectivity. This is not a problemwhen the whole receiver is being re-viewed, because the total selectivity iswhat matters. It does make things dif-ficult when the receiver is used as testequipment, and we are only interestedin the oscillator’s performance.

There are many more measuringmethods—such as those that usephase shifters and delay lines. Theyare not the most sensitive, althoughthey have certain advantages. Formore information about these meth-ods, see the References listed at theend of this article. The method I de-scribe is used in professional phase-noise measuring equipment. In fact,the test fixture I made is also a kind ofreceiver, with the carrier not con-verted to an IF but to 0 Hz. The fixturealso has a sensitive phase detector.

The block diagram is presented inFig 13. The blocks within the dottedlines are in the test jig. The compo-nents outside are external measuringand support equipment. The equip-ment needed includes an oscilloscope

and an ac voltmeter. At the input port,a double-balanced mixer is used as aphase detector. Two signals are phase-compared: The signal to be character-ized, and a reference signal from agood oscillator, preferably a crystaloscillator. The oscillators must be atthe same frequency. The loop circuitensures that the two oscillators arephase-locked in such a way that the dcoutput voltage of the phase detector iskept at 0 V. The switchable compo-nents in the loop provide sufficientcontrol to establish phase lock. Thegoal is to maintain a low loop band-width. In this way, any drift and low-frequency FM of the oscillators is com-pensated, but the higher-frequencycomponents of the phase noise appearat the output. In fact, we have made adirect-conversion receiver. The twosidebands of the tested oscillator areboth converted to a frequency rangestarting just above 10 to 50 Hz andreaching to some 100 kHz. Actually,the four sidebands from both oscilla-tors are converted to the audio range.(See Fig 14.)

At the phase detector’s output is alow-noise audio preamplifier and sometwin-T filters to suppress the hum thatwould otherwise overload the follow-

ing filters and amplifiers. Then wehave three filters/buffers in parallel.One of these is tuned to a nominal fre-quency of 1 kHz with a bandwidthof about 100 Hz. The next is tuned to10 kHz with a 1-kHz bandwidth, andthe last to 100 kHz with a 10-kHzbandwidth. The gains of the amplifiersafter the filters are adjusted duringcalibration, as described below. Thistest fixture makes it possible tomeasure phase-noise levels at threediscrete offsets (1, 10 and 100 kHz)from the carrier. Although this is notas nice as a plot from a spectrum ana-lyzer, it provides sufficient informa-tion for oscillator experiments.

The full schematic diagram is shownin Fig 15. An MD108 is used as thephase detector. An SBL-1 may be sub-stituted. The dc balance is achievedwith a small resistor network from the–12 V supply. This is something thatmust be adjustable, or determinedexperimentally with every individualfixture. The double-balanced mixer isterminated with 50 Ω—for RF—by theseries-connected 50-Ω resistor and the3.3-nF capacitor. The LC low-pass fil-ter keeps RF out of the audio circuits.The preamplifier with the BC149 (aEuropean low-noise audio transistor)

Fig 12—Block diagram of a 50-MHzreceiver.

Fig 13—Block diagram of the phase-noise test setup.

Fig 14—Both the lower and the uppersidebands are mixed to the frequencyrange from 0 to 100 kHz.

36 QEX

and the NE5534 amplifies the phasenoise and limits the frequency rangesomewhat. The two twin-T filters keep50 Hz and 100 Hz away from the fil-ters and amplifiers. They must beredesigned for 60 Hz and 120 Hz in theUS. These two notch filters are impor-tant, because often in test setups, os-cillators have some mains-inducedresidual FM. This FM can be strongenough to overload the amplifiers.On the output connector after theNE5534, one can connect a low-fre-quency spectrum analyzer. (A soundcard in a PC can be used for thispurpose.) At the time that I made thistest jig in 1990, such cards were notavailable.

There are three filters after thepreamplifier. The 1-kHz and 10-kHzfilters are active, with three identical,series-connected sections. The 100-kHzfilter is an inductively coupled band-pass filter. (I used two RF chokes.) Thegain of each amplifier is adjusted bymeans of the feedback resistors be-tween the output and the invertinginput of the operational amplifiers.This is done during initial calibration.

The circuit with the three OP27 op

amps is the loop amplifier and filterused to phase-lock the two oscillators.One of the oscillators must be elec-tronically tunable over a few-kilohertzrange. The tuning voltage is takenfrom the output of the last OP27.Switch S1 allows one to adjust the loopgain, S3 sets the loop time constant,and S2 adjusts the loop damping. Idon’t intend to cover basic PLL tech-niques in this article, and I supposethat knowledge of the required theoryis available to most people trying thiskind of thing. At the output of the firstOP27, an oscilloscope can be connectedto verify that the loop is behaving well.

The whole circuit is housed in a cast-aluminum box. The power supply is notin the box. Stray fields from the trans-former would interfere with the verysensitive measurements. Further cir-cuit-layout details and a PCB patternare not available. The test jig was con-structed using “ugly” (dead-bug) con-struction, and I never had time to give ita neat appearance. You must regard theschematic diagram as a starting pointor as a source of ideas. Design your owncircuit, and publish it [in QEX!—Ed.] ifit contains nice new ideas!

Regarding the external measuringinstruments: The oscilloscope musthave a sensitivity of 10 mV/div and abandwidth of at least 100 kHz. The acvoltmeter must have a best sensitivityof 1 mV full scale and a bandwidth ofat least 100 kHz.

Measuring with the Test FixtureAfter the setup is calibrated, mea-

surements can be made. (I will dealwith calibration later.) A measure-ment is performed as follows: Connectthe test oscillator and the referenceoscillator at the inputs. The power ofthe reference oscillator at the LO-portis not terribly critical. The power mustbe between +4 dBm and +13 dBm. Thelevel of the test oscillator at the RFport is critical. Ideally, the level of thisoscillator should equal the level usedduring calibration. Every decibel ofchange causes a decibel of change inthe measured composite noise levels. Icalibrated my setup at a level of 1 mW(0 dBm). The level at the RF port mustnot be very much higher than that,because the phase detector will be-come nonlinear above 0 dBm. If lessthan 0 dBm is used, the sensitivity of

Fig 15—Schematic diagram of the phase-noise test jig.

Jan/Feb 1999 37

the setup becomes progressively less.In some cases, however, it can be use-ful to deliberately use less inputpower. If the noise sidebands are verystrong and overload the preamplifierin the fixture, using less power bringsthe circuits back in their linear-oper-ating range. The tested oscillator’slevel must be set by an external at-tenuator. A word of caution: It is nec-essary that the two oscillators be wellbuffered. Otherwise, injection lockingis possible via the phase detector, andthat is an unwanted situation.

The output of the loop amplifier isconnected to a Varicap circuit in one ofthe oscillators. The oscilloscope is con-nected to the monitor output, and ifthere is a second channel, it is con-nected to the output of the loop ampli-fier. By adjusting S1, S2 and S3, theoscillators are forced to phase lock. Byadjusting potentiometer P1, the oper-ating range of the Varicap diode is set,while the average output of the phasedetector is kept at 0 V.

The ac voltmeter can be connectedsuccessively to the three outputs tomeasure the levels of the phase noise.The actual phase-noise levels can be

calculated using factors determinedduring calibration.

The levels determined this way are,in fact, the sum of the composite noiselevels of both oscillators. It is thereforeimportant to use a reference oscilla-tor—such as a crystal oscillator or aVCXO—that has much better noisecharacteristics than the oscillator be-ing characterized. If that is not pos-sible, one cannot find the actual noiselevels of the tested oscillator, but onecan determine that the tested oscillatoris at least as good as the measured level.

Now some words about the “art” ofphase noise measurement: The signallevels we are measuring are very low,sometimes below the microvolt level.Therefore, it is necessary to keep exter-nal influences out of the test setup. Usecables that are as short as possible. Besure that the coaxial connectors makegood contact with the cable, otherwisesome of the RF currents induced on theoutside of the shield may also flow onthe inside, causing interference. Thisis a well-known electromagnetic-com-patibility problem. Keep the setupaway from such interference sources asTVs and PC monitors. Do not try to

measure unshielded oscillators.A less-obvious problem is acoustical

noise. During some of my measure-ments of crystal oscillators, I noticeda correlation between variations inthe measured phase-noise level at1 kHz and the speech of a local hamtalking on 70 cm! When I investigated,I found that I could easily whistle atone near the oscillators that wouldproduce phase modulation well abovetheir normal noise levels. This makesyou wonder about transceivers withbuilt-in loudspeakers.

One more warning: I have noticedmany times during my experiments

Fig 16—Measurement of the series-resonant Q of a crystal.

38 QEX

that one must not perform phase-noisemeasurements shortly after havingsoldered parts in an oscillator. Duringthe first couple of hours after solder-ing in an oscillator circuit, the phase-noise level shows sudden outbursts,especially at 1 kHz from the carrier.My explanation for this is that minormechanical tension caused by the tem-perature changes must relieve itself,and this causes minor electricalchanges, which in turn cause thesenoise bursts. Maybe someone has abetter explanation?

Calibration of the Test FixtureAlthough relative measurements can

be performed without calibration, it isnice to have some idea about the abso-lute levels. Oscillators with absolutelycertain phase-noise behavior cannot be

bought, as far as I know. There are twogood calibration methods.

The first one uses a crystal oscilla-tor as the reference source, and aphase-modulated signal generator orVCXO as the RF source. The lattersource is modulated successively with1 kHz, 10 kHz and 100 kHz. Whileobserving the measurement on a spec-trum analyzer, the modulation level isadjusted until the first sidebandscome to –80 dBc. There will be only onepair of sidebands visible; the secondpair will be below the analyzer noiselevel. The power of the referencesource must be set to +10 dBm, thepower of the RF modulated RF sourceto 0 dBm. Phase lock is established,then each of the three feedback resis-tors in the output amplifiers is ad-justed to obtain exactly 1 V RMS from

Fig 17—Schematic diagram of the 50 MHz reference crystal oscillator.

each of the outputs. For this firstmethod, one needs a rather good spec-trum analyzer. Some of us are quitehappy to have such an instrument inour labs, but not everyone is so lucky.

The second method may be some-what easier to perform. If we take 0dBm as the RF input calibration leveland we want –80 dBc to be the level for1 V at the outputs, we can perform thecalibration using a single –74 dBm sig-nal at an appropriate offset frequency.Since during normal operation, bothnoise sidebands contribute to the totalpower, a 6-dB-stronger signal is usedhere. The calibration procedure isidentical to that of the first method.With an LO signal applied at +10 dBm,and the RF signal at –74 dBm placedat 1 kHz, 10 kHz and 100 kHz from theLO frequency, the calibration resis-

Jan/Feb 1999 39

tors are adjusted for 1 V output. A signal generator with acalibrated output is probably easier to get than a spectrumanalyzer.

I compared the calibration techniques, and for me, theyresulted in less than 2 dB difference. In my opinion, this isadequate for amateur purposes.

If we are using method two and a signal generator, wecan also easily determine the exact bandwidths of the threefilters. Carefully adjust the frequency of either oscillatorto both sides of the frequency where the unit was cali-brated, and note the two frequencies were the 1 V outputdrops to 0.707 V. These are the –3 dB points, and the dif-ference between the two is the 3-dB bandwidth of the filter.What we really need are the so-called noise bandwidths ofthe filters, but these can be estimated by multiplying the3-dB bandwidths by 1.1. Having measured the bandwidthsof the filters, we can calculate the correction factor for eachof the filters:

C B= ( )10 log (Eq 6)Where C is in decibels and B is in hertz. The conversion

factor translates the level measured in the respective band-widths to the normalized level, expressed in dBc/Hz. Forexample: if measuring a phase-noise level of 100 mV usingthe 1 kHz filter, and having determined that the noisebandwidth of that filter is 1100 Hz, the normalized phasenoise level is:– . – .80 20 30 4 130 4− − = dBc/Hz (Eq 7)

Finally, we must determine the noise floor of the wholesetup. To do this, we connect a 50-Ω load resistor to the RFport and a +10 dBm LO signal to the reference port. Thenwe measure the voltages from each of the three filter out-puts, and from these values, calculate the noise floor. Inmy case, it was approximately –162 dBc/Hz. If we are

measuring an oscillator, and the phase-noise levels hap-pen to be so good that they come close to the noise floor, wecan measure with less accuracy using the formula:

L BU

U

Un

l

n

= − ( ) −⎛⎝⎜

⎞⎠⎟

+⎛

⎝⎜⎞

⎠⎟−

⎡

⎣⎢⎢

⎤

⎦⎥⎥

– log log log80 10 201000

10 12

2 (Eq 8)

In this formula, to be used with measured phase noiselevels less than 10 dB above the noise floor:

L = phase-noise level, in dBc/HzB = noise bandwidth of the filter in useUn = noise-floor output level, in mVUl = measured phase-noise level, in mV

Fig 19—Comparison of the 50 MHzphase-noise levels of several oscillatorsand signal generators.

Table 1—Test Equipment

1. Crystal oscillator, PA0JOZ2. ADRET 7200A3. HP8640B4. HP8662A5. HP8642B6. Rohde and Schwarz SMG7. Rohde and Schwarz SMDU8. HP608E9. HP8657B10. LC Clapp oscillator, PA0JOZ

Fig 18—Block diagram of the phase-noise measurement of two crystal oscillators.

A Reference Crystal OscillatorThe phase-noise test jig was constructed to perform

measurements on LC oscillators around 50 MHz. There-fore, I needed a good reference source near this frequency.First, I determined the Qs of a batch of 50.4 MHz crystalsthat I had. The simple test setup is shown in Fig 16. I de-termined the –3 dB points of the response, and from that,calculated the series-resonant Q. The values ranged from20,000 to 30,000. Using the crystals with the highest Qs,two crystal oscillators were built.

The schematic diagram (Fig 17) was derived from anarticle about VHF crystal oscillators in the German maga-zine UKW Berichte. (This is translated to English as VHFCommunications magazine.—Ed.) The principle is that thelower FET works linearly and that amplitude limiting isdone in the upper FET. The crystal acts as the source by-pass and determines the oscillator frequency by its seriesresonance. If the crystal is removed from the circuit andreplaced with a capacitor of, say, 1 nF, the circuit can bemade to oscillate on approximately the correct frequency. If

40 QEX

the crystal is then put back in the cir-cuit, it will oscillate at the series reso-nance. The inductance in parallel withthe crystal is there to compensate forthe crystal’s parallel capacitance. Thisvalue can be determined simply, withthe crystal and the inductance isolatedfrom the circuit, using a grid-dipmeter.

So that the oscillator can be phaselocked by the test jig, a Varicap circuitin series with the crystal is used, en-abling a frequency variation of a fewkilohertz. The output buffer is a com-mon-base transistor stage. The inputimpedance of this buffer is very low—approximately 3.5 Ω—and thereforedoes not degrade the Q of the crystaltoo much. The bandwidth of the bufferis approximately 5 MHz, and it is ca-pable of delivering 32 mW. The powerat the output connector is +10 dBm.

The collector’s resonant circuit hasanother Varicap in parallel. It pro-vides us with a phase-modulation in-put. In my case, a signal between 100Hz and 100 kHz at a level of 1 V, causesPM sidebands at –60 dBc. This mustalso be calibrated of course, using aspectrum analyzer, to be useful. Thewhole circuit is housed in a tinned boxthat was soldered all around.

This may seem a rather complicatedway of making a crystal oscillator, butit really is a good circuit. The oscilla-tor frequency is determined solely bythe crystal, and does not depend on theother components as much as in sim-

Fig 20—Schematic diagram of a broadband buffer amplifier used in experiments with LC oscillators.

pler crystal oscillators. It is well buff-ered and the phase-noise behavior isexcellent.

Some Measurement ResultsFirst, I measured the two crystal oscil-

lators with the test setup shown in Fig18. The levels are plotted in Fig 19 withdots at 1 kHz, 10 kHz and 100 kHzmarked “1.” In fact, this is not the realphase-noise level of oscillator 1, since inthis measurement it is not certain whichof the two crystal oscillators is best. If weassume that the two oscillators producethe same amount of phase noise, then thereal phase noise level of each of the oscil-lators is 3 dB lower than the plottedvalues.

After that, I measured some signalgenerators available at the laboratorywhere I was employed at the time, us-ing one of the two crystal oscillators asthe reference signal. The results are inFig 19. In Fig 19, one of my own LC os-cillators is plotted, which shows thatit is possible to make a good oscillatoryourself.

50 Ω, has a high input impedance and avoltage gain of three. The currentthrough the complementary outputstage must be adjusted to approxi-mately 10 mA by altering the value ofthe resistor between the two diodes inthe base circuit. The bandwidth isabout 100 MHz and the isolation is bet-ter than 65 dB. With a 50-Ω source, thenoise figure is approximately 15 dB.

Using this buffer, no injection-lock-ing problems were experienced any-more. The buffer produces some dete-rioration in the high-frequency phase-noise levels, but this can only be noticedwhen using very good oscillators.

ConclusionI hope that this article gives a good

idea of how phase-noise measurementscan be performed with rather modestequipment. Realize that my experi-ments were done in 1990, so some of mycircuits are based on components that Ihad available at the time. If myworkload allows, I intend to designanother setup using a delay-line dis-criminator at some time in the future.Such a setup has the advantage that itrequires no reference source, althoughthe sensitivity may not be as good as inthis setup. If I succeed, I will report myexperiences in QEX!

References1. Reference Data for Engineers (Indianapo-

lis, Indiana, USA: H. W. Sams) 6th edition,1975.

2. Ulrich L. Rohde, KA2EUW, Digital PLL

A Broadband BufferDuring my experiments with LC

oscillators, I had some trouble with in-jection locking of the two oscillatorsconnected to the test jig. It became ob-vious that this was caused by insuffi-cient buffering of the oscillators. Itherefore designed a broadband bufferamplifier. It is shown in Fig 20. It is ca-pable of delivering some 100 mW into

Jan/Feb 1999 41

Frequency Synthesizers: Theory andDesign, (Englewood Cliffs, NJ: Prentice-Hall, 1983).

3. Spectrum Analysis, Amplitude and Fre-quency Modulation, Hewlett-PackardApplication Note 150-1, 1989.

4. Phase Noise Characterization of Micro-wave Oscillators, Phase Detector Method,Hewlett-Packard Product Note 11729B-1,1983.

5. The “Art” of Phase Noise Measurement,Hewlett-Packard RF and MicrowaveMeasurement Symposium and Exhibition,Dieter Scherer, May 1983.

6. Low Phase Noise Applications of theHP8662A and 8663A Synthesized Signal

Generators, Hewlett-Packard ApplicationNote 283-3, 1986.

7. Bernd Neubig, DK1AG, “An ExtremelyLow-Noise 96 MHz Crystal Oscillator forUHF/SHF Applications,” Part 1, VHF Com-munications, 1981/3, pp 135 to 143; Part2, 1981/4 pp 194 to 203. VHF Communi-cations back issues can be ordered fromKM Publications, 5 Ware Orchard, Barby,Nr.Rugby, CV23 8UF, UK.

8. Bernd Neubig, DK1AG, “A Modern Profes-sional Look at the Design of Stable, CrystalOscillators Working at High Frequencies,”Part-1, VHF Communications, 1991/1, pp35 to 42; Part 2, 1991/2, pp 74-79. See Ref-erence 7 for contact information.

Jos van der List obtained his Dutchclass-A license on his 18th birthday, 25years ago. He is married and has twoadopted children. His chief AmateurRadio interests are home-brew equip-ment and measurement techniques. Hehas build equipment for all bands from1.8 MHz through 10 GHz.

Jos has worked for many years intype-testing laboratories, both in tech-nical and managerial roles. He recentlygot a new job at the well-known Dutchresearch laboratory, TNO, where hedesigns radio front ends (0 to 10 GHz).

42 QEX

Use software compensation to stabilize WB2V’spopular DDS VFO within 0.5 ppm from 0 to 65°C.

By Curtis Preuss, WB2V

5150 Timberidge Ct SERochester, MN 55904wb2v@aol.com

A Temperature-Compensated DDS VFO

1Notes appear on page 45.

IntroductionOne of the challenges in the design

of oscillators is to reduce temperature-induced output frequency changes.Frequency drift versus temperaturecan be a particularly onerous problemin VFO designs. Various frequencysynthesis techniques can provide alow-drift alternative to the traditionalanalog VFO by using a more stablecrystal oscillator as a reference fre-quency. However, crystal oscillatorsare not immune to drift. For example,over a temperature range of 0 to 70°C,oscillators using an AT-cut crystalmay vary 20-40 parts per million

(ppm) in frequency. On the HF bands,this drift can cause errors of hundredsof hertz in the output frequencies.

Many techniques to control oscilla-tor drift have been developed. Thesetechniques include ovens, double ov-ens, specially cut crystals and a greatvariety of temperature-compensationtechniques. Temperature-compensa-tion techniques can be difficult foramateur experimenters. A tediousprocess of adjusting component valuesmay be required, and the results canbe disappointing.

Lately I’ve been experimenting witha temperature-compensation tech-nique that employs direct digital syn-thesis (DDS). In this technique, com-pensation is accomplished in softwarerather than hardware. With this com-pensation technique, I was able to dem-

onstrate a DDS VFO that had 0.5 ppmdrift over a temperature range of 0 to65°C. Fig 1 shows the measured VFOdrift with and without compensation.

A Little HistoryMy experiments employed a tem-

perature-compensation system thatwas invented1 in 1973. Fig 2 is a blockdiagram of that system. The systemworks by adjusting the control inputsof a “presettable” frequency divider.The adjustment is made a function oftemperature so that it cancels the driftin the reference oscillator, producinga divider output that is temperaturecompensated.

It’s probably safe to say that in theearly ’70s, this system required many

Jan/Feb 1999 43

parts. However, with present-day mi-crocontrollers and single-chip direct-digital synthesizers, the system can beimplemented using two off-the-shelfchips, along with a few other compo-nents. A microcontroller can be pro-grammed to calculate a frequencycontrol word “W” for a DDS chip, where“W” is a function of both temperatureand another external input.

A DDS chip is a convenient way toimplement the presettable frequencydivider portion of the system. An inter-esting example of using direct digitalsynthesis for temperature compensa-tion was published2 in 1989. It com-bined DDS with a novel temperature-sensing scheme,3 where the crystal it-self was used to sense temperature.This oscillator was reported4 to have astability of 0.03 ppm from –55 to 85°C.

The particular implementation formy experiments is shown in Fig 3.This system is very similar to a DDSVFO I’ve described previously.5 On thehardware side, the difference is theaddition of a resistor/thermistor volt-age divider and the substitution of amicrocontroller that contains an ana-log-to-digital converter (ADC). On thesoftware side, the code6 is changed sothat it adjusts the data sent to theDDS chip according to the tempera-ture indicated by the thermistor.

Test SetupIn order to collect frequency-versus-

temperature data, I built a small,insulated test chamber. Thermoelectricmodules7 (also known as Peltier-effectmodules) were used to pump heat in orout of the chamber. By controlling thecurrent through the thermoelectricmodules, the chamber temperaturecould be varied. My construction tech-niques limited the temperature rangeto 0 to 65°C.

The resolution was limited to one-degree steps by the control circuit—not state-of-the-art performance, butit saved many trips to the kitchen, andfreed up the refrigerator and oven fornormal use.

A crystal oscillator along with athermistor, buffer amplifier and volt-age regulator were placed in the ther-mal test chamber. The oscillator was acommon-base Colpitts using a fifth-overtone AT-cut crystal. A bead ther-mistor was held in place on the crystalusing heat-shrink tubing. The bufferamplifier was a pair of 74AC04 invert-ers wired in parallel. Two 78L05s wereused for voltage regulation, one for theoscillator and one for the buffer amp.The buffer amplifier output was ac

Fig 1—DDS VFO output drift with/without compensation.

Fig 2—An externally compensated oscillator, circa 1973.

Fig 3—A temperature-compensated DDS VFO.

coupled to a short run of coax. At theDDS-chip clock pin, the coax wasterminated with 100 Ω to ground and100 Ω to VDD.

The DDS chip and microcontrollerwere kept outside the thermal cham-ber to reduce the heat load. A DDSVFO—minus the microcontroller—

44 QEX

was mounted on a piece of perforatedboard along with a PIC16C74 micro-controller, LCD and shaft encoder.These were wire-wrapped together,with the PIC16C74 wired to replacethe original PIC16C54 on the DDSVFO. A switch was connected to one ofthe microcontroller inputs so that com-pensation could be turned on or off.The complete setup is shown in Fig 4.

Factors Affecting TemperatureCompensation

HysteresisHysteresis is a significant problem

for any temperature-compensationscheme. Here, hysteresis means thatthe oscillator frequency at a particu-lar temperature depends on what theprevious temperature was. A mea-surement of this effect can be seenin Fig 5. Data for the plot was taken in5° steps around the cycle 25°, 65°, 0°,20°C. After each step, 30 minutes wereallowed for everything to reach astable temperature.

Much of the literature on this sub-ject agrees on one thing: The exactcauses of hysteresis are not well un-derstood. Consequently, it is not pos-sible to control hysteresis very well.There are some oft-cited observations8

about hysteresis:• The crystal itself exhibits hysteresis.• Other oscillator components can

make the problem worse, oftendominating.

• The effect is larger over wider tem-perature swings.

Fig 4—A Photograph of the test setup, showing: thermalchamber, DDS VFO assembly, oscillator assembly and afrequency counter.

Fig 5—Measured DDS VFO hysteresis.

• The hysteresis of different crystalsmay have different “signs.” That is,the frequency for increasing tem-perature may be higher or lowerthan the frequency for decreasingtemperatures.

• The magnitude of the hysteresismay vary significantly from unit tounit.

Unit-to-Unit VariationsThe frequency-versus-temperature

characteristic of crystals may varygreatly from unit to unit. This meansthat in order to achieve good results,each oscillator must be individuallycharacterized. The compensation datafor one oscillator will not work well foranother. For example, Fig 6 comparestwo copies of the oscillator circuit Iused.

Numerical IssuesTemperature-compensation tech-

niques rely on a known relationshipbetween an oscillator’s frequencyand temperature. This relationship ismeasured, then stored in the micro-controller’s memory. The frequency-versus-temperature data can be storedas a table. Then interpolation can beused to calculate points between tableentries. Another storage method is tofit the measured data to a polynomialand store only the polynomial coeffi-cients. There are inevitable differ-ences, or residuals, between the mea-sured data and the stored representa-tion. These residuals can be reduced byusing higher-order polynomials or im-

proved interpolation techniques.9, 10

Since my experiments were carriedout over a limited temperature range,the problem of residuals was greatlyreduced. Crystals with an AT cuthave an S-shaped frequency-versus-temperature characteristic. Likewise,a plot of the voltage versus tempera-ture of a resistor-thermistor voltagedivider displays a complex curve. Overa limited temperature range, how-ever, the relationship of thermistorvoltage to frequency is nearly astraight line. This can be seen in Fig 6.

The resolution of the ADC is in play.The slope of the frequency drift versusthermistor voltage for oscillator B wasabout 12 ppm/V. Using an 8-bit ADCover a 5-V input range provides a reso-lution of about 20 mV. This translatesinto a frequency resolution of about0.23 ppm. For oscillator A, the limitwas about 0.33 ppm.

On the other hand, the resolution ofthe AD9850 DDS chip was not a sig-nificant factor, except at low outputfrequencies. With a 32-bit accumula-tor, and the reference oscillator at120 MHz, the DDS frequency resolu-tion is about 0.028 Hz. This is 0.1 ppmif the VFO output frequency is280 kHz, but at 5 MHz, the DDS reso-lution is 0.0056 ppm.

Thermal Time ConstantsThermal time constants can cause

more problems. Some time is requiredfor the resonating portion of the crys-tal to reach the same temperature asa thermistor on the outside of the crys-

Jan/Feb 1999 45

tal holder. Power-on warm up is onesituation where thermal time con-stants can be noticed; another isduring a rapid temperature change.

My frequency counter was too slowto directly measure the effects of ther-mal time constants, but I was able toobserve the warm-up drift by listeningto the VFO output on a receiver. Thesignal could be heard to shift back andforth a few hertz as the temperaturechanged, but it was difficult to distin-guish between the effects of thermaltime constants and the limited resolu-tion of the ADC.

ConclusionThese experiments verified the

effectiveness of adding temperaturecompensation to a DDS VFO. A signifi-cant reduction in frequency drift wasobtained with very little additionalhardware. However, time and effortwere required to measure the fre-quency-versus-temperature charac-teristic of the reference oscillator.

Software-compensation techniquesprovide a great deal of flexibility.They’re able to compensate for com-plex relationships between tempera-ture and frequency. For the limitedtemperature swing of 0 to 65°C, the os-cillator was characterized quiteclosely by linear interpolation. Forwider temperature swings, a more so-phisticated interpolation or curve-fitting technique should be used.

Good results were obtained using an8-bit ADC. A 10-bit converter may givebetter results, but at some point, hys-teresis and thermal time constants willlimit the stability that can be achieved.

Notes1US Patent 3,719,838: “Temperature

Compensating Digital System for Electro-mechanical Resonators,” Ralph Pedutoand Jan Willem L. Prak, issued March 6,1973, see claims 12-14.

2A. Benjaminson and S. Stallings, “A Micro-computer-Compensated Crystal OscillatorUsing a Dual-Mode Resonator,” Proceed-ings of the 43rd Frequency Control Sym-posium, 1989.

3S. Schodowski, “Resonator Self-Tempera-ture Sensing Using a Dual-Harmonic-Mode Crystal Oscillator,” Proceedings ofthe 43rd Frequency Control Symposium,1989.

4A. Benjaminson and B. Rose, “PerformanceTests on a MCXO Combining ASIC andHybrid Construction,” Proceedings of the45th Annual Symposium on FrequencyControl, 1991.

Fig 6—Drift versus thermistor voltage for oscillators A and B.

5C. Preuss, WB2V, “Building a Direct DigitalSynthesis VFO,” QEX, Jul 1997, pp 3-7.

6You can download the experimental codewith the new function from the ARRL Webhttp://www.arrl.org/files/. Look for DDS_VFO.ZIP.

7URL http://www.itiferrotec.com has a lotof technical information about thermoelec-tric modules.

8R. Filler, “Thermal Hysteresis in Quartz-Crystal Resonators and Oscillators,” Pro-ceedings of the 44th Frequency ControlSymposium, 1990.

9M. Frerking, “Crystal Oscillator Design andTemperature Compensation”, Van Nos-trand Reinhold Company, 1978, Chapter 10.

10R. Filler, “Frequency-Temperature Con-siderations for Digital Temperature Com-pensation,” Proceedings of the 44thFrequency Control Symposium, 1990.

46 QEX

By Stuart E. Bonney, K5PB

802 Melrose DrRichardson, TX 75080

Practical Application ofWind-Load Standards to

Yagi Antennas: Part 1

Whenever a fierce wind arises,or we start to think aboutnewer, bigger beam-antenna

systems, we are reminded that windplays a large part in antenna-systemsurvival. We need a reasonably simple—but accurate and consistent—way ofevaluating wind loads on antennas,masts and towers. We also need ameans to determine expected windvelocities in our own local areas. If weplan to use purchased antennas, weneed a way to evaluate their wind-related properties reliably. There is ahost of other practical issues as well,but these are fundamental.

Electronic Industries Association

(EIA) Standard 222 is often used as abasis for antenna wind-load analysisand specifications. Its more recentversions provide improved, more accu-rate and detailed methods that con-form with widely accepted practices.These later releases include data use-ful when estimating peak wind veloci-ties at a given location and installedantenna height. Although any re-quired permits and approvals are usu-ally governed by local building codes,the process can be simplified if ourmethods are compatible with thosecodes as well.

This standard has gone through sev-eral incarnations over many years.Probably the best-known version ofthese is RS-222-C,1 mainly because of

its longevity and wide circulation.Published in 1976, it is still oftenquoted, or its methods used, althoughit has been superseded by three laterreleases: EIA-222-D in 1987, EIA/TIA-222-E in 1991 and TIA/EIA-222-F2 in1996. Like an older standard forbridges, RS-222-C expressed windforce-versus-velocity in terms of pres-sure on a flat plate. This was the sourceof the widely published figures forwind pressure in pounds per squarefoot (psf): 70.7 mph = 20 psf; 86.6 mph= 30 psf; 100 mph = 40 psf and so on. Aswe will see shortly, these numbers arevalid only under a specific set of condi-tions, and not in the broad way thatthey have often been applied.

Effective with EIA-222-D, this stan-dard was made more consistent withother widely used structural codes,1Notes appear on page 50.

Is your antenna rated to survive local storms? Howmuch wind load does it put on your mast and tower? Basic

physics and a little engineering can provide answers. InPart 1, we learn how wind acts on flat plates and cylinders.

Jan/Feb 1999 47

such as the Uniform Building Code,the Standard Building Code and theBritish Code of Practice (CP3). Thisrelease included a basic expression fordynamic wind pressure, as well as ex-plicit drag coefficients for both flat andcylindrical shapes. It also includedheight and wind-gust factors. Releases222-E and 222-F also contain moredetailed, refined wind-velocity data bystate and county from NOAA andNational Weather Service sources.

Fluid dynamics and their applica-tion to antennas are not exactly com-monplace subjects. Our objective hereis to facilitate an understanding of theunderlying principles and offer practi-cal ways of applying such insight basedon the latest standard, EIA-222-F. Wewill also examine methods and defini-tions that (we hope) will help overcomesome of the confusion that has oftenresulted from shortcomings in the nowobsolete RS-222-C. The material pre-sented here does not address all of themany details relating to Yagi antennasystems, especially for large arraysused in severe environments. Thatwould require much more than just anarticle or two.

Basics of Wind and Wind LoadsAs a prelude to practical applica-

tions, and as a means of understandingwhere the numbers come from, let’sreview the basic principles of windloads on structures such as antennas.Wind is air in motion, and air has mass.Therefore, wind has energy that can bequantified. This energy is a function ofair mass density, m, and velocity, V, asexpressed in Eq 1 below.3

We can get the weight of one cubicfoot of air at standard temperature andpressure (STP) from handbooks, thendivide it by 32 to get its mass and applya factor to convert velocity in feet persecond to miles per hour. This yields anequivalent form expressed in Eq 2:P = 0.5 m V2 (Eq 1)or,P = 0.00256 V2 (Eq 2)

These are expressions for fluid dy-namic pressure. Eq 1 is a form ofBernoulli’s equation and is related to Ek = 0.5 m V2, the fundamental expres-sion for kinetic energy. Eq 2 can befound in many college-level physics andengineering textbooks, although thesymbol q is often used instead of thesymbol P. Later versions of EIA-222have adopted the symbol qz.

The two equations express the kineticenergy and resulting dynamic pressureof a wind stream at a given velocity, butthey do not express the net force on an

object in that stream. When wind en-counters an object, both the magnitudeof the resulting pressure and its gradi-ent around that object vary with objectshape and other factors.4

To elaborate, let us consider a flatplate of finite height, but greatlength—approaching infinite. Imaginethis plate is oriented broadside to on-coming wind, as shown in the cross-sectional view of Fig 1A. As the windencounters this object, the wind streamdivides and flows over its top andbottom edges, where the flow then be-comes turbulent. Due to the nature ofthis flow, not only does the wind exertdirect positive pressure on the plate’sfront side, but significant negativepressure also develops on its backside.

Fig 1B shows the approximate pres-sure distribution around this plate.Pressure on the front side is equal tothat expressed by Eq 2, but decreasestoward either edge. Negative pressure,or suction, on the backside actuallyreaches a value greater than the maxi-mum positive pressure on the front.The net effect of these pressures causesa plate of this particular shape to be-have as if it were subjected to twice theforce of the direct wind pressure alone.

The situation for a very long cylin-der is similar, but because of its morestreamlined shape, the airflow andpressure distribution are different.Both positive and negative pressurescan reach peak values similar to thoseon a flat plate, but pressure gradientsaround a cylinder are such that aver-age pressures are much lower. Theoryand experimental results described innumerous texts indicate the net forceto be about 60% of that on a flat plateof equal length and height.

Next, we must consider aspect ratio,which is the ratio of length to height(L/H) for a flat plate, or length to diam-eter (L/D) for a cylinder. When the as-pect ratio is finite, wind can flowaround the sides of an object as well asover its top and bottom. This reducesaverage pressures over the surface.

This reduction continues as aspect ra-tio decreases. For an aspect ratio of 1,the net force falls to a little more thanhalf that for an infinite L/D ratio. Theeffects of shape and aspect ratio arewhy equating wind velocity with a fixedpressure or force on objects can bemisleading.

Calculating actual wind forces on anobject would be greatly complicated bythese dependencies were it not for theintroduction of another factor. This isreferred to by aerodynamicists as dragcoefficient. You may encounter the termforce coefficient, but for our purposes,the terms are interchangeable. Table 1lists values for flat plates as well ascylinders. Drag is also influenced by therelative surface smoothness of an ob-ject, wind speed in relation to the speedof sound, and an object’s size in relationto wind velocity. Drag coefficientsshown are composites from severalsources and are applicable to typicalbeam antennas and the wind conditionsunder which antennas are used.

The final factor required to expressthe actual force on an antenna ele-ment—or an entire antenna or amast—is projected area. For any ob-ject, this is its “shadow area,” such asan object would cast on a nearby par-allel surface when illuminated by thedistant sun. For flat plates, it is sim-ply height times width; for cylinders,it is length times outside diameter.

We can now put all three factorstogether, resulting in the followingequation for objects that are broadsideto the wind:F = (P) (Cd) (A) (Eq 3)where

F = force in poundsP = dynamic wind pressure in

pounds per square footCd = drag or force coefficient

(dimensionless)A = projected area in square feet.By substituting the expression for P

from Eq 2 into Eq 3, we can find theactual force on an antenna-system

Fig 1—A shows wind flow around a flat plate. B shows pressure distributionaround the same plate.

48 QEX

component for any selected wind ve-locity. Keep in mind that this appliesto objects broadside to the wind, sub-jecting them to maximum force.F = 0.00256 (V 2) (Cd) (A) (Eq 4)

We would not need to go beyondEq 4, except for one important fact: Wedon’t always know what values to usefor Cd and A, especially for a pur-chased antenna. Some manufacturersdo not specify a value for A. Further, ifits value—or some equivalent—isspecified, but we don’t know its basis,or whether it already includes someunstated value for Cd, we cannot cal-culate actual wind loads with any de-gree of confidence or accuracy. It is avery desirable convenience for bothmanufacturers and purchasers to beable to specify or determine antennawind-load characteristics simply, ac-curately and uniformly.

A great deal of this uncertainty isattributable to ambiguity in RS-222-C,which was recognized and corrected insubsequent releases. We now take abrief look at this issue and relatedproblems with this standard. Closer ex-amination suggests that it containedsignificant lapses. It expresses windforce in psf on a flat plate as P = K V 2,where K = 0.004 and “includes a gustfactor and a drag factor for flat surfaces”that are, however, not quantified. Withregard to cylinders, 222-C was silent,except for stating, “In all cases, the pres-sure on cylindrical surfaces shall becomputed as being 2/3 of that specifiedfor flat surfaces.”

If we use the widely accepted windpressure coefficient 0.00256, and as-sume a gust factor of 25%, we can de-rive drag coefficients from the aboveequation. They are effectively about1.25 for flat plates and 0.8 for cylin-ders, both of which are much lowerthan generally accepted values. Onthe other hand, if K remains at 0.004and we assume it contains no gust fac-tor, effective drag coefficients areabout 1.6 for flat plates and 1.0 forcylinders. These values more closelyreflect actual antenna wind-load prop-erties. Thus, it can be concluded in apractical sense that 222-C does notinclude a gust factor, despite what itsays. Although 222-C has workedfairly well if users did not also dependon an implicit gust factor in determin-ing antenna survivability, it leavesmuch uncertainty over how wind sur-face areas are defined.

The Search for EffectiveWind Surface Areas

For many years, the term “effective

Table 1Composite Drag Coefficients (Cd) For Various Aspect Ratios

Aspect Ratio Cd Flat Plate Cd Cylinder∝ 2.0 1.2

100 1.8 1.140 1.6 1.010 1.4 0.84

5 1.2 0.721 1.16 0.70

Note: See Appendix for discussion.

Table 2Drag Coefficients from Various Sources

Source: Mechanical Engineering In Radar And Communications8

Aspect Ratio Correction0 to 4 0.64 to 8 0.78 to 40 0.8> 40 1.0Same table (credited to British CP3) appears in A. J. MacDonald, Wind LoadingOn Buildings,7 who also gives Cd of 1.0 for infinite cylinder (Reynolds number103 to 105).

Source: Standard Handbook for Mechanical Engineers9

Aspect Ratio Cd Flat Plate 1 1.16 4 1.17 8 1.2312.5 1.3425 1.5750 1.76∝ 2.0Cd = 1.11 listed for discs.

Source: Mechanical Engineering Handbook10

Aspect Ratio Cd Flat Plate 1 1.18 5 1.210 1.320 1.5Cd of 1.17 listed for discs. Reynolds number = 105, source cited: S. F. Hoerner

Source: EIA/TIA-222-E/F for “Appurtenances” (including antennas)2

Aspect Ratio Cd Flat plate Cd Cylinder≤ 7 1.4 0.8≥25 2.0 1.2

area” or something similar has beenused to specify an area that is usablefor calculating antenna wind loads.5

Unfortunately, the absence of accu-rate, universally understood and ap-plied definitions has rendered suchterms all but meaningless. To illus-trate, I have an old manual for a popu-lar four-element, 20-meter Yagi, whichspecifies the antenna’s effective areaas 3.9 square feet. A more recent cata-log lists 7.3 square feet for the very

same antenna. This is not a uniqueinstance. Another technical articleaddressing Yagi antenna wind loadsnotes that inconsistencies are commonand widespread.6

Some manufacturers and users of222-C have interpreted the 2/3 factorfor cylinders to mean that effectiveareas are 2/3 those of equivalent flatplates. Others have apparently ap-plied this factor to the specified valueof coefficient K and taken effective

Jan/Feb 1999 49

area to be the same as projected area.By assuming a gust-factor value, onecan also derive otherwise unspecifieddrag coefficients. However, each ofthese approaches produces a differentresult for effective area. We very rap-idly reach a point where it is impos-sible to determine where we standwithout a detailed description of themethod and assumptions used, whichusually are not readily available.

This discussion would be pointlessif not for the fact that 222-C’s dubious

Appendix: Practical Drag Coefficient Considerations

Whenever we design or evaluate an antenna element orboom for survivability, we want to know how much wind itwill tolerate and how much wind load it will accumulate ata given wind velocity. Actually, the two are interrelated, be-cause increasing diameters for increased strength alsomeans more wind load, resulting in more stress. Drag, ofcourse, is an integral part of all this, so we need to knowas accurately as we can just what drag coefficient to use.This is a different viewpoint from that of the structuralengineer, who is concerned with the practical aspects ofmeeting applicable codes and signing off on them. In thatcase, application of a particular drag coefficient is usuallymandated. The two viewpoints do not necessarily conflict,but they can result in different approaches.

Among the numerous technical sources I’ve re-searched, there is general agreement on basic Cd valuesof 2.0 for a flat plate of infinite aspect ratio, and 1.2 for acorresponding cylinder. There is less agreement on valuesfor aspect ratios of finite dimensions. Part of the reasonmay be the many variables affecting aerodynamic drag;they make analysis difficult, and experimental setups arevulnerable to measurement variations. Table 2 shows sev-eral examples of Cd values for various aspect ratios. Somesources list aspect ratios in groups, and include correctionfactors to be applied to basic Cd values.7, 8 This is clearlya shortcut, since the relationship between Cd and aspectratio is not a step function. Other sources also put aspectratios in groups, but list discrete Cd values for each, not-ing that intermediate values can be interpolated.9, 10

As shown in the table, the parts of 222-E/F dealing withattachments, such as antennas, list only two Cd values,each, for flat and cylindrical surfaces. One is for aspectratios of seven or less, the second for ratios of 25 or more.Values for ratios between these steps are to be interpo-lated. We should keep in mind that these standards weredeveloped primarily for towers, where failures can resultin total collapse. Safety and legal considerations, and adesire for substantial overload margins, justify a quite con-servative approach, which seems evident here.

Our goal of accuracy in selecting drag coefficients forantennas, especially in a design sense, is primarily toachieve the best balance between wind loading charac-teristics and survivability. Certainly, it is wise to allowreasonable margins for unexpected overloads, which wecan do by designing to a sufficiently high peak windvelocity. At the same time, we do not want to over-designor over-specify. That can lead to larger, heavier, morecostly construction and consequent higher loads onmasts, rotators and towers, resulting in a cascade effecton total system costs.

While much wind-tunnel test data exists for variousshapes, I’ve found none that deal with step-taperedcylinders. We know that the typical tapered element hasat least a partial path for wind to flow around the end of asegment where it tapers down to the next smaller seg-ment. When swages or inserts are used for greater step-down in tubing diameter, this end-flow path becomeslarger. Although this effect remains to be quantified andmay be relatively small, we can be reasonably sure itexists.

The Cd values listed earlier in Table 1 are compositesof data extracted from several sources, including thosediscussed above. Care was exercised to ensure theirvalidity for conditions normally applying to antennas, ie,Reynolds numbers from 103 to 105. (A note regardingReynolds number: It relates object dimensions, windvelocity, and air viscosity.) These values were then plot-ted graphically to aid in visualizing spreads and to deter-mine the best overall fit. Based on this data, the foregoingconsiderations and practical experience, it appears that areasonable value of Cd for cylindrical elements andbooms lies between 1.0 and 1.1. A value of 1.1 would beslightly conservative without being excessive. Again, ifyou must satisfy requirements of a building or structuralcode, it is best to use whatever figure that code specifies.

The values in Table 1 are not presented as absolutes;they should be considered a workable basis only untilbetter data becomes available. In passing, it is interest-ing to note that the calculated Cd for cylinders fromRS-222-C is 1.04 if no gust factor is assumed. This ex-plains why 222-C has worked reasonably well despite itslarge potential for confusion. Applying the 2/3 pressurefactor to K as specified, 0.004 (2/3) = 0.00267. Dividingthis by the basic dynamic wind pressure constant0.00256 yields 1.04, the effective Cd. The more usualform F = 0.00256 V 2 (Cd) (A) then emerges, but the stan-dard certainly is not clear as written.

Finally, there are a few points worth mentioning aboutthe mathematical model described here for wind loads. Forsimplicity of calculation without impairing basic accuracy,this model assumes that objects, such as antenna ele-ments, always remain fully broadside to the wind. In thereal world, elements deflect, and in high winds can shedwind loads amounting to several percentage points of cal-culated values, depending on deflection angles. Theseangles, which vary between root and tip, depend in turn onwind velocity and element construction. This effect contrib-utes to wind survival margins for elements but to a muchsmaller extent for booms, which are usually stiffer. Thus,the model is essentially conservative.

methodology often remains in use andcontinues to be a source of confusion.However, this is not a case of who’sright or wrong; it is regrettable if any-one chooses to view it in that light. Theproblems are within the document,and it seems well past time to retire it,especially since its successors providea basis for more consistent results.

Practical Solutionsand Their Results

Confusion and inconsistency con-

cerning effective area could be easilyended by broad adoption of a specificdefinition. The following, consistentwith current standards, is proposed:

Wind surface area (WSA) is definedas projected area (A) of an object ofinterest, multiplied by the drag or forcecoefficient (Cd) appropriate to thebasic shape and aspect ratio of thatobject.

Stated as an equation:

WSA = Cd (A) (Eq 5)

50 QEX

As discussed in the Appendix, a dragcoefficient of 1.1 appears to be—in apurely technical sense—accurate forcylindrical elements and booms of typi-cal beam antennas, but this is not theonly consideration. Building and struc-tural codes often mandate a value of1.2, and for manufactured antennas,this is the best figure to use. Substitut-ing Eq 5 into Eq 4, we have a practicaland uniform basis for determiningwind loads on antenna elements,booms, complete antennas or masts:F = 0.00256 V2 (WSA) (Eq 6)or, alternatively:F = V2 (WSA) / 390 (Eq 7)where:

F = force in poundsV = wind velocity in miles per hourWSA = wind surface area in square

feet.Note that this method results in

effective areas that are larger than wemay be accustomed to seeing. This is abyproduct of doing away with prob-lematic flat-plate-equivalent meth-ods. However, this newer method hasthe distinct advantage that all adjust-ments needed to account for differentshapes and aspect ratios are incor-porated into a single, net figure:WSA. The same dynamic-wind-pres-sure constant is used at all times, re-moving all ambi-guity.

Despite the increase in areas, netbroadside-wind-load forces computedin this way will not be greatly differ-ent in many cases than with the oldmethods. Among the reasons is thatthe pure dynamic wind pressure val-ues calculated with 222-D/E/F areonly about 65% of values commonlyassociated with 222-C for variouswind velocities. Further, based on thecross-flow principles described byWeber (see Ref. 6), wind loads for com-plete antennas, when rotated off thewind at angles of 30 to 50°, will belower. How to calculate wind loads fortypical antennas will be describedwith examples in Part 2 of this series.

We now arrive at the specificationsantenna manufacturers should pro-vide to users, or that you need to de-termine if building your own. First isa combined WSA figure for the ele-ments. Second is a separate WSA fig-ure for the boom. The reason you needboth is that the antenna configurationinfluences which case will representthe worst wind load. Normally, for HFbeams, the element load is the greater.For multielement VHF beams on longbooms, the boom often presents the

greatest wind load. If you plan tomount both types on one mast, maxi-mum net wind loads on the mast andtower may be produced by wind that iseither broadside to the elements ofboth antennas or broadside to thebooms. You need to be able to evaluateboth conditions.

The third item required is antennawind-survival rating expressed inmiles per hour of peak wind velocity.Survival rating refers to the wind ve-locity that, if exceeded, will result inpermanent, measurable deformationto a physical member of the antenna.Some manufacturers already providethis information. Those who do notshould begin. Actual computation ofwind-survival capability is a fairlycomplex process of stress analysis,best done with a computer. It is beyondthe intended scope of this series.

With WSA and wind-survival rat-ings in hand, you have the basic infor-mation you need to evaluate the physi-cal suitability of an antenna for yourlocation and installed height. You canthen proceed with installation plan-ning. Using the rated WSA and Eq 6 or7, the force that the mast and towermust handle can be easily found, oncethe local wind conditions are known.

The next hurdle you will normallyface in planning a new or upgradedantenna installation is site engineer-ing. By this we refer to the generalprocess of planning the installation,not to the services of professional engi-neers who, if needed, usually will haverequirements of their own. Includedare the applications of antenna heightand wind gust factors, determinationof peak design wind velocity appropri-ate to your locale and site topographyand practical selection of supportingmasts. Part 2 will explain how to do allof this based on 222-D/E/F criteria.

and methods have been more consis-tently and comprehensively imple-mented in EIA-222-D/E/F. There areseveral practical advantages to begained by broad adoption of these im-proved methods:

1. They provide antenna manufac-turers with a uniform, consistent, yetsimple method for specifying the wind-load properties of their products. This,in turn, helps to produce a level play-ing field for all.

2. They provide users of manufac-tured antennas with a simple meansof evaluating wind-load characteris-tics of these products, and of determin-ing their suitability for a particularinstallation.

3. While not a substitute for appli-cable building codes, these methodsare consistent with many of thesecodes, and can simplify the initialstages of application for local antennapermits and approvals.

ConclusionsMethods used to analyze wind loads

on antenna structures and to specifywind load areas still are often basedon, or derived from, the old EIA stan-dard RS-222-C. This standard andmethods based on it have been shownto have significant problems, prima-rily a consequence of ambiguity andabsent specifics in the document itself,along with the resulting lack of unifor-mity in application.

Here, we have briefly reviewed thebasic physical principles on whichwind-load analysis is based, as an aidto reader understanding and applica-tion. As we have seen, these principles

References1. EIA RS-222-C, Structural Standards for

Steel Antenna Towers and Antenna Sup-porting Structures (Washington, DC: Elec-tronic Industries Association, March 1976).

2. TIA/EIA-222-F, Structural Standards forSteel Antenna Towers and Antenna Sup-porting Structures (Washington, DC: Elec-tronic Industries Association, June 1996).

3. A. L. Sanford and J. M. Tanner (GeorgiaInstitute of Technology), Physics for Stu-dents of Science and Engineering (NewYork: Academic Press, 1985).

4. J. Bertin and M. Smith, Aerodynamics ForEngineers, Second Edition (New Jersey:Prentice Hall, 1989) p 85.

5. E. Fuller, W2FZJ, “The Application ofStress Analysis to Antenna Systems.”Ham Radio, Oct 1971, pp 23-31.

6. D. Weber, K5IU, “Determination of YagiWind Loads Using the ‘Cross Flow Prin-ciple’,” Communications Quarterly, Spring1993.

7. A. J. MacDonald (University ofEdinbergh), Wind Loading on Buildings(Toronto: Halstead Wiley, 1975) p 81.

8. C. J. Richards, Mechanical Engineering inRadar and Communications. (London:Van Nostrand Reinhold, 1969) p 125.

9. Marks’ Standard Handbook for Mechani-cal Engineers, 9th Edition (New York:McGraw Hill, 1987) Section 11, p 78.

10. M. Kutz, Mechanical Engineering Hand-book (New York: Wiley, 1986) p 1525.

Stuart E. Bonney, K5PB, has beenlicensed since 1952 and formerlyheld the calls W8JUV, W0DMR andW5PAQ. He is a retired engineer andmanager from Rockwell Collins, and isfounder and president of a communica-tions and computer services company.He is author of the book “AmateurSingle Sideband,” published by Collins,two other books on computer softwareand several articles.

Jan/Feb 1999 51

Before ferrites came on the scene, every ham wanteda coil winder. Some applications still require solenoid

coils with many turns. Here’s a winder small andinexpensive enough to rest on a shelf until it’s needed.

By Bob Dildine, 7J1AFR/W6SFH

Apt 203, 5-2-2 DenenchofuOta-Ku, Tokyo 145-0071, Japanbob_dildine@HP-Japan-om2.om.jpn.hp.com

A Handy Coil Winder

This small coil winder can bebuilt in a few evenings withnothing more than simple hand

tools. Constructed from readily avail-able parts, it features variable speedcontrol and a built-in turns counter.

For a recent project, I needed towind a transformer for a small, high-voltage switching power supply. Thesecondary consisted of about 500 turnsof #34 wire, wound on a small pot-corebobbin. I first attempted to wind thiscoil with a variable-speed electrichand drill. A simple turn counter wasrigged up using a small light bulb, aphotocell and the lid from a cat-foodcan. A small hole punched near the rimof the lid triggered a photocell con-nected to my frequency counter in itsaccumulator mode.

It was difficult to securely mount thedrill to the worktable, the speed control

was troublesome and it was tricky tosimultaneously watch the turns coun-ter and the flow of wire onto the bobbin.Although I managed to wind the coil,the results were less than satisfactory.About this time, I saw a simple coilwinder based on a toy motor, describedin one of the ham magazines here inJapan.1 That article was the inspira-tion for this coil winder.

ObjectivesI set out to build a coil winder that

could be used for small coils such aspot cores, transformers and air coilswith many turns of fine wire. The fol-lowing were the objectives for thisproject:• Inexpensive• Simple to build using readily-

available parts• Easy to control winding speed• Direction reversible• Built-in turn counter• Ruggedness

The MotorAfter investigating the toy motor

mentioned above, I decided to use amore substantial unit. At a local mo-tor shop, I found a 400-RPM gear mo-tor with a 6-mm shaft (about 1/4 inch)that could be powered from 12 V dc. Alow-voltage dc motor has the advan-tage that its speed can be easily con-trolled by varying the supply voltage.Its direction can also be reversed us-ing a DPDT switch or relay.

1CQ Ham Radio (a Japanese periodical),April 1998

The Turn CounterI originally planned to use my fre-

quency counter in the accumulatormode as the turn counter, rather than

Speed ControlThe speed control is shown in Fig 1.

It consists of a simple potentiometerconnected as a voltage divider, whichfeeds an NPN Darlington transistor inan emitter-follower configuration. Thisprovides plenty of current to power themotor and allows the use of a smallpotentiometer for the control. The mo-tor direction is governed by a DPDTtoggle switch.

52 QEX

build a dedicated counter circuit. How-ever, I found a little electromechanicalcounter at a local surplus shop thatturned out to be just right for this ap-plication. A disk about 5 cm (2 inches)in diameter was cut from a piece of thincopper-clad circuit board material, anda hole was punched near the rim. I sol-dered the disk to a 6 mm brass insertfrom an old plastic knob and fastenedthe insert to the motor shaft using itssetscrew. A Sharp 1A53 photo inter-rupter is mounted so that it triggerseach time the hole in the disk passedby. Each 1A53 contains a Schmitt-trig-ger inverter that causes the output togo low when the light beam is inter-rupted. Suitable substitutes can beobtained from Jameco (part number114091 or 114104) or Digi-Key (vari-ous). These substitute parts may haveonly an open-collector transistor out-put, which conducts when the lightbeam is not interrupted. If such a partis used, omit the 2N2222 inverter thatdrives the 555 pulse stretcher. Fig 1—Speed-control circuit.

Fig 2—Turn-counter circuit.

Refer to Fig 2. The output from thephoto interrupter drives a pulsestretcher made from a 555 timer. Themechanical counter is rated for 10

counts-per-second, so the pulse widthwas set to 50 ms. This gives a squarewave at the highest counter fre-quency. Without the pulse stretcher,

Jan/Feb 1999 53

Fig 3—The complete coil winder.

the pulses from the photo interrupterwere too narrow to trigger the counterat all but the lowest motor speeds. Anauxiliary pulse output was connectedto my frequency counter—set to its ac-cumulator mode—and the motor wasrun at full speed long enough to accu-mulate several thousand counts. Themechanical counter and the electroniccounter agreed precisely.

The Power SupplyThe motor and counter circuits are

powered from a small switching powersupply, found at one of the local sur-plus shops. This supply was an open-frame model, and it’s mounted so thatit’s difficult to accidentally contact anyhigh-voltage points. Because I ex-pected that the coil winder would onlybe used occasionally, it has no primarypower switch. A switch was installed,however, in the low-voltage circuit, toremove power from the motor duringset up. A bright LED connected to oneof the power-supply outputs showswhen the winder is plugged in!

Mechanical ConstructionThe winder was built on a 26×21 cm

(101/4×81/4 inches) flat aluminum plate,about 3 mm (1/8 inch) thick. The motoris securely mounted using angle brack-ets and braces, so that the shaft pointsto the right. The turn counter mountson an angle bracket just behind themotor shaft, so it is possible to see thecounter without taking your eyes offthe coil. The motor speed-control cir-cuitry is built on a small heat sink. Theturn-counter circuit is on piece of per-forated board, and mounted to the sideof the motor. The speed control and re-versing switch are mounted on a smallbracket on the left-front corner of thecoil-winder base. This allows speedcontrol with the left hand, while feed-ing wire with the right. “Southpaw”builders may wish to reverse the orien-tation of the motor and control panel.

A 6-mm shaft coupling joins the mo-tor shaft to a 6-mm bolt that holds thebobbin or coil being wound. It might bebetter to machine an adapter that con-nects the motor shaft to any of variousmachine screws, which could be used tohold coil forms of differing sizes. A pairof cones—drilled along their axes—would allow the coil form to be easilycentered on the machine screw. I don’thave access to the necessary machinetools to make these accessories, so Ijust carefully center the bobbin, andhold it in place with nuts and washers.

Using the WinderThe coil winder is easy to use. Fas-

ten the bobbin or coil form to the shaft,making sure that it is centered. Youcan check this by running the windera few turns while watching for eccen-tricity. It’s best to wind coils in thedirection that allows the wire to feedonto the top of the coil (rotating awayfrom you) so you can watch turns go onthe form. Don’t forget to zero the turnscounter before you start! Make thefirst few turns of any winding by hand,to be sure the wire starts properly.Then slowly increase the motor speeduntil you have a comfortable buildgoing. When the counter approachesthe desired number of turns, slowlydecrease the motor speed while keep-ing tension on the wire. Stop when thefinal turn count is reached. It’s best tohold the wire 30 cm or so (about a foot)back from the coil. This allows thewinding to accumulate in even rows.Fast-moving wire can burn your fin-gers, so use a glove or other protection.

Place the wire spool on a table orother support about a meter behindyou. I usually hang the spool on ascrewdriver that is clamped in a por-table vise. It’s important that the spoolcan rotate freely, especially if you areusing fine-gauge wire.

Some ImprovementsAfter using the coil winder a few

times, I thought of several improve-ments that would make it more useful:

My turn counter increments regard-less of the winding direction. If I re-verse the direction to take off a fewturns, the counter reads incorrectly.This could be solved by replacing thesimple mechanical counter with anup-down counter—either mechanicalor electronic—and by using an appro-priate direction-sensing circuit. Thiscircuit could simply be another photointerrupter, offset from the first one,including logic circuitry to sense thedirection. (Some photo interruptersare made for just this purpose.—Ed.)

A set of different-sized machinescrew shafts for various coil sizes,along with centering cones and theassociated adapters to the motor shaftwould be useful.

A foot-pedal speed control wouldkeep both hands free to apply thewinding. This could be nothing morethan a pedal attached to a speed-con-trol potentiometer.

SummaryThe coil winder has given good ser-

vice. It enabled me to wind a precisenumber of turns on the high-voltagetransformer for the switching powersupply. The convenience of a good coilwinder makes it easy to rewind coilsto change the number of turns, wiresize or other parameters.

54 QEX

Here’s an example of functional test equipment builtfrom components in the junk box. Perhaps you

have similar instruments hiding in yours!

By C. A. Hoover, K0VXM

1945 E Phillips CtMerritt Island, FL 32952

The Cheap Sweep

Having acquired a pair of beau-tiful six-pole adjustable filtersfrom an old cellular telephone,

I set myself to the task of using themon 902 MHz. I didn’t really want toabuse my solid-state final by pouringpower into an unknown load. Yet, I hadneither a signal generator nor a sweepgenerator of suitable characteristics.So, I came up with a sweeper usingjunk-box parts and equipment on hand.

The sweeper consists of a VCO, acontinuous rotation pot, a steppermotor and a stepper motor driver (SeePhoto A). A diode detector, a suitable12 V power supply and an oscilloscopewith X-Y display complete the setup.

Fig 1 is a block diagram of the system.Photo B shows the running system andPhoto C shows the trace of the filterunder test. Photo D is a full view of theentire test setup.

The VCO (again pulled from an

old cellular telephone) tunes from 890to 950 MHz and drives the device un-der test directly. Since the tune cur-rent is less than 1 mA, a pot of anyvalue from 100 Ω to 100 kΩ will suf-fice. I used a 1 k pot.

Fig 1—A block diagram of the test setup.

Jan/Feb 1999 55

The stepper-motor driver is a kit pur-chased from All Electronics.1 Othersuppliers sell similar kits; they are nothard to find. The stepper motor can beany motor that will drive the pot andmounts easily. The motor supplied withthe driver kit is fine. When coupling thestepper motor and the pot together, besure that the shafts rotate withoutbinding.

The diode detector is homebrewedfrom The ARRL Handbook.2 Keep theleads as short as possible. My detectorwas fabricated with composition resis-tors and disc capacitors; it is usable(barely) to 2 GHz.

The 12 V power supply should handleabout 2 A and be well regulated. Besure the voltage does not change whendisconnecting or stopping the steppermotor. Any change in voltage will cause

Photo A—The sweeper consists of a VCO, a continuousrotation pot, a stepper motor and a stepper motor driver. Photo B—The running system.

Photo C—A trace of the filter under test.

Photo D—A full view of the entire testsetup.

Once the trace is found, adjust the Xand Y positions and the motor speedfor best viewing. A motor speed ofabout 200 RPM seems to be optimumin my setup.

At this point one begins questioningthe use of a stepper motor as opposedto other possible methods of swingingthe VCO. Aside from having manystepper motors and a couple of steppermotor drivers on hand, I wanted theability to run the system manuallythrough its range. This makes for easy,relatively accurate frequency spotting.

I don’t expect that anyone will du-plicate my system exactly. However,my purpose is to demonstrate whatcan be done with a well-stocked junkbox and a little ingenuity.

the VCO frequency to change and thetrace on the ’scope to move.

The ’scope does not need to be at allfast. Any ’scope that has X-Y capabil-ity will do. Connect the output of thediode detector to the Y input and thewiper of the pot (which also goes to thetune pin of the VCO) to the X input.

Notes1All Electronics Corp, 14928 Oxnard St, PO

Box 567, Van Nuys, CA 91411; tel 800-826-5432, 818-904-0524, fax 818-781-2653;e-mail allcorp@allcorp.com; URL http://www.allcorp.com/. Catalog #SMKIT-2.

2R. Dean Straw, N6BV, Ed. The ARRL Hand-book (Newington: ARRL, 1998), Order No.1816. ARRL publications are available fromyour local ARRL dealer or directly from theARRL. Mail orders to Pub Sales Dept,ARRL, 225 Main St, Newington, CT 06111-1494. You can call us toll-free at tel 888-277-5289; fax your order to 860-594-0303;or send e-mail to pubsales@arrl.org.Check out the full ARRL publications line onthe World Wide Web at http://www.arrl.org/catalog.

56 QEX

No-tune transverters have encouraged many newcomersto the UHF and microwave bands. Yet a newcomer’s

receive system may be plagued by whistles and squeals.Here’s how to find the culprit and a few hints

to silence those birds.

By Michael McKay, W4AZR

2265 Windsor DrMerritt Island, FL 32952-5508

No-Tune Transverter-SpurIdentification

This article shows how to identifyspurs generated in a receiverlineup with one of the no-tune

transverters that are now popular.Although there is nothing very newhere, the analysis approach may helppeople just getting their feet wet in theUHF/Microwave game. The specificexample is from my own 33 cm station.

I use a Down East Microwave SHF902k transverter with a DEM 33LNAlocated at the 18-element loop Yagi.The 902k feeds a DEM 144-28DC,which in turn feeds my TS-940 HFtransceiver via the transverter socketon its rear panel. The arrangement is

shown in Fig 1. Notice that the TS-940has been modified to transmit outsidethe ham bands. This permits its use asa 25 to 30-MHz tunable IF. The 902kuses an oscillator injection frequencyof 759.000 MHz (derived from a 94.875MHz crystal oscillator) and the DEM144-28DC injection is at 118.000. Thebottom line is that 25 MHz relates toboth 143 MHz (118 + 25 = 143) and902 MHz (759 + 118 + 25 = 902). Whenthe TS-940 is at 30 MHz, the receivefrequency is 907 MHz; in an ideal re-ceiver one would hear nothing butwhite noise when tuning this range inabsence of a signal on the antenna.

With a 50 Ω termination on the 902kinput, I observed a loud carrier at afrequency of 25.500 MHz, another at27.874 MHz and still another at27.900 MHz. All had clean notes when

tuned for “zero beat.” The correspond-ing frequencies at the 2-meter levelare 143.500 MHz, 145.874 MHz, and145.900 MHz, respectively. I’ll refer tothese as spurs 1, 2 and 3. To under-stand the origins of these spurs, beginby listing the harmonics of the 2-meteroscillator and the base oscillator of the902k as in Table 1 (to the 10th order).

Now calculate the sum and differ-ence frequencies of the 2 meter fre-quency of spur 1 with each of theharmonics of the 2 meter oscillator. Forexample, 236.000 MHz plus and minus143.500 MHz gives 379.500 MHz and92.500 MHz. Then check to see if one ofthe results lies on or very near a fre-quency in the base-oscillator column.In this case, the sum frequency is thefourth harmonic of the base oscillator.This identifies the source of spur 1. In

Jan/Feb 1999 57

like fashion, spur 2 is caused by thesixth harmonic of the 2-meter oscilla-tor mixing with the ninth harmonicof the base oscillator. In numbers,we have: 853.875 MHz equals708.000 MHz plus 145.874 MHz.

Spur 3 is a bit different from the othertwo. It is the image response of theTS-940. Notice that the spur 3 fre-quency is 27.900 MHz. The first IF inthe TS-940 is 45.05 MHz and the oscil-lator is on the high side of the signal.Thus, the image response is at 118.000MHz (27.900 + 90.100 = 118.00). Al-though the signal is weak, this meansthe TS-940 hears the DEM 144-28DCoscillator. Because there is no crowdingon 33 cm at this time, this is no prob-lem, particularly since I (and the localshereabout) operate on 903.100 MHz. Ifyou’re in the neighborhood, look for usTuesday evenings at 7:30 local time.

So there will probably be spurs inyour nice new shiny UHF setup—isthis bad? The answer is no, becausethe spur signal can serve as a quickcheck of your receiver. At one time, Iverified that the 94.875 MHz oscilla-tor had started by listening to its fun-damental on an ordinary home-enter-tainment FM receiver. Now I simplytune down to spur 1 at 26.500 MHz(902.500 MHz on 33 cm) and verifythat all is well.

It’s interesting that—after an hourand a half—the oscillator in the DEM144-28DC sometimes develops a slightfrequency drift at about a 1 Hz rate.This is not detectable when workingFM, but it’s a disaster on SSB. Some-day I shall have to fix it, but the pointis the problem was found by listeningto the built-in spur.

The levels of spurs 1 and 2 can bereduced an S-unit or so by simply grasp-ing the +13.8 V dc feed line to the144-28DC tightly in hand. Since all themodules share an Astron RS-20M sup-ply, it is probable that better decoup-ling within the modules would improvethe matter. Although I have not tried ityet, I intend to use ferrite bead chokeson the internal +13.8 V wiring in con-junction with surface-mount bypasscapacitors. I’ve found that MMIC spurlevels wildly vary for the first five min-utes, until the MMICs achieve thermalequilibrium, at which point measure-ments are generally reproducible.

Yes, the identification of spurs can beautomated by use of a computer andsuch programs are available. For new-comers, however, I believe the approachjust described produces greater under-standing. It’s simple, you need onlybasic arithmetic and a scratchpad to getthe answers.

After discharge from the WW2 Navy,Mike went back to college, received aBA (1947) and an MA (1949), both inphysics. He then spent two years inTV-receiver circuit design. During theKorean War, Mike switched to defenseelectronics (for the next 38 years) work-ing mostly on microwave radar guid-ance of weapon systems. Mike wasnamed on 13 US patents in this period.First licensed in 1947 he has heldW1QVV, W8ERL, W2GRS and nowW4AZR. Current interests are weak-signal gear for UHF and above, a mod-est amount of RTTY and the 6 and12-meter bands.

Table 1—Oscillator Harmonics

Order 2 Meter Oscillator Base OscillatorHarmonics (MHz) Harmonics (MHz)

1 118.000 94.875 2 236.000 189.750 3 354.000 284.625 4 472.000 379.500 5 590.000 474.375 6 708.000 569.250 7 826.000 664.125 8 944.000 759.000 9 1062.000 853.87510 1180.000 948.750

Fig 1

58 QEX

By Zack Lau, W1VT

225 Main StNewington, CT 06111-1494zlau@arrl.org

RF

Building Solid-StateMicrowave Power Amplifiers

It’s been said that there are two dif-ferent approaches to building micro-wave projects. In Europe, they buildeverything from scratch, while in theUS many people re-engineer and con-vert the somewhat plentiful supply ofsurplus equipment. I’m going to dis-cuss an intermediate approach, takingsurplus gear and using the parts tobuild gear essentially from scratch.The idea is to combine the best of bothworlds: Get the performance and prac-ticality of gear built and designed withamateur techniques and the low costof surplus parts.

The reason for using surplus parts is

obvious—microwave parts can be ex-tremely expensive, particularly de-vices like power GaAs FETs. Some ofthe more-exotic parts, like 3 W, 10 GHzFETs, cost hundreds dollars—a bit toorich for most amateurs. Others, likeabsorptive rubber, may not cost a lot ofmoney, but are a real hassle to buy be-cause manufacturers aren’t set up for$10 orders from individuals. Thus, dis-assembling a surplus amplifier canprovide little parts that are a hassle toget, in addition to the obviously expen-sive pieces.

It isn’t always practical to modifysurplus equipment, however. Thejump to the nearest amateur band mayjust be too far for some circuits, suchas power splitters and combiners. Onthe other hand, the equipment mayuse tiny wires and single-layer capaci-tors that are easily destroyed whenhandled without specialized equip-ment. I’ve heard of circuit boards with

plated traces that vaporize and be-come unsolderable.1

Those combiners, even when theydon’t work, can be quite useful. Often,they indicate the equipment’s originaloperating frequency. The power-sup-ply bypassing may be helpful too.Typically, λ/4 stubs or radials areused. You can gauge frequency by thesize of the circuitry; as frequency in-creases, components get smaller.

Markings on transistors can also beuseful. Many manufacturers use a four-digit number that signifies the fre-quency range. Thus, 5964 correspondsto 5.9 to 6.4 GHz. Often, there is a suffixthat corresponds to the output power,in watts. Thus, 5964-3 is a 3-W device.These are impedance-matched transis-tors optimized for a narrow frequencyrange. The internal networks have alow-pass characteristic—they can often

1Notes appear on page 60.

Jan/Feb 1999 59

be retuned lower, but not higher, in fre-quency. Transistors with regular modelnumbers don’t have specific frequencyranges, although they yield less gain athigher frequencies. In addition, gainand other characteristics seem to beless predictable at higher frequencies.

Of course, the best way to identifydevices is with a data sheet from themanufacturer, but this isn’t alwayspractical. If you have a working am-plifier, you can often get the devicecurrents and voltages by making somemeasurements. Actually, the data-sheet is just a starting point if youwant to get as much power as possibleout of the device. Typically, people rundevices at 50% of the measured zero-gate-voltage drain current (IDSS). Of-ten, this current varies by 50% ormore, so you need to select deviceswith the highest current for the out-put stage. Be careful when measuringIDss—the drain voltage must be re-duced to avoid destroying the device.Otherwise, the transistor will dissi-pate far too much heat and burn up.

I strongly recommend you use a cur-rent-limited supply for each transistoryou are worried about destroying.With proper current and voltage regu-lation you can protect a transistor fromcommon accidents, such as shortingout the gate-bias supply. Active bias-ing protects transistors indirectly—with low-current transistors, the cur-rent-sensing resistor can act as acurrent-limiting device. With a powertransistor, there often isn’t enoughvoltage overhead to allow much cur-rent limiting. In addition, big powerresistors are very inefficient. You mayhave 100 Ω of resistance in an LNA,while 1 Ω is more typical in a powerapplication.

Contrary to popular belief, you don’talways want a lot of ground plane onthe topside of a double-sided circuitboard. At microwave frequencies, it isquite possible for the top and bottom“ground planes” to differ. Fig 1 showsan excellent example using a high-gain PF0011 33-cm hybrid module.The top ground plane provided a feed-back path resulting in an unstableamplifier. The cure was to add a coupleof screws to block the feedback path,as shown in Fig 2. While a groundingzealot may argue that there wasn’tenough grounding, a second version(shown in Fig 3) with no top ground foilbetween the input and output workedjust was well and required lesshardware. Admittedly, the problem ispartially due to the high gain of theamplifier—I’ve measured 46 dB of

Fig 1—A microwave amplifier with complete topside foil—oscillates.

Fig 2—Four bolts added near the lower board center connect the ground planes—no oscillation.

Table 1

Device Screw TorqueMGF K25M/K30M/K33M #0 pan head 2.0-2.5 kg-cmMGF-0905A #2 pan head 2.5-3.0 kg-cmMGF-C39V #4 pan head 5.0-6.0 kg-cm

Fig 3—Without the four bolts, but with the top foil removed between input andoutput connections—no oscillation.

gain! I’ve seen modules encased in atin plated shield—undoubtedly to re-duce unwanted feedback paths.

Instead, you really ought to designthe ground paths—how, exactly, is theground current going to go where youwant it? There are three critical paths:to the transistor, the coax connectionsand the matching elements.

The first item is what often makesRF transistors so expensive. The pack-age may be as expensive as the chipinside it! High-power transistor pack-ages are usually fastened to theground with screws. Mitsubishi warnsagainst using thermal grease—in-stead specifying that you use a veryflat surface (a surface finish of 32 μ

60 QEX

inch maximum with a camber of0.0003 inch maximum).2 They alsorecommend using a torque wrench tomount the device. Table 1 has somesuggested torque settings.

If you study the current path at thetransistor, a weak point may be thepath between the grounded flange ofthe transistor and the ground plane ofa circuit board. If you have a double-sided circuit board, what holds theground plane against the heat-sinksurface? The input and output leads ofmost transistors really aren’t meantfor that purpose; many are rather deli-cate strips of metal. Thus, some peopleuse conductive epoxy to attach the cir-cuit board to milled plates. This ap-proach is very expensive because con-ductive epoxy has a limited shelf life.Everyone I’ve talked with recom-mends using all the epoxy at once. Itslimited shelf life is even shorter onceopened. We can more cheaply usescrews to hold down the circuit boardat key points; I’ve gotten this to workwith 3 W, 10 GHz amplifiers. The bestapproach is to use an aluminum-backed circuit board, so the mountingsurface is the circuit ground plane.

I’m most familiar with the Rogersversion of this exotic circuit board,having purchased a sheet at a fleamarket nearly a decade ago. I only fig-ured out how to use it just recently.The easiest problem is etching thematerial without damaging the alumi-num. (A sodium persulfate etchantwon’t attack aluminum unless theproper catalyst is available, so it maynot be necessary to mask off the alu-minum.) According to a document onthe Kepro Web page,3 ordinary tablesalt can be used as a catalyst. Thus,handling the aluminum with sweatyhands may not be a good idea. Theworst case may be a poor maskingjob—where a little etchant is trappedagainst the aluminum. The concentra-tion of ions may be enough to start areaction displacing copper from usedetchant. Very good masking will berequired with ferric chloride etchant—its reaction with aluminum is quitevigorous.

A tougher problem is 6061 aluminumalloy, which has a tensile strengthof just 20 kpsi. While 6061-T6 with a35 kpsi tensile strength is easy to mill,20 kpsi 6061 is much softer and ma-chines poorly. I ended up using aSherline miniature milling machineand keeping the machining simple. Iopted for slots across the full width cir-cuit board, instead of the fancy pocketspeople normally cut.

When machining metal, it is essen-tial to fasten the work securely, whileminimizing distortion from the fasten-ers. Instead of using a vice, I makethreaded mounting holes an impor-tant part of the design. Not only dothese holes make the project easy toinstall in the final transmitter ortransverter, but also they can be usedduring the machining process. As abonus, they more closely duplicate thestresses found in the assembled unit,so unwanted warping is less likely.

A similar technique is to solder amachined brass block to the circuitboard. This works well if the circuitboard is just the right thickness—youmight just need to cut the block to sizeand add tapped mounting holes. It’s agood idea to mount the block to thechassis as well. This will allow heattransfer from the block to the chassis;there is usually enough thermal con-ductivity for small and medium powerdevices, up to a watt or two. Largerdevices may need a more-direct con-nection to a heat sink, however.

The next problem areas are the co-axial connectors. How do you get goodelectrical contact from the groundplane of a circuit board to the connec-tor? Again, soldering works well. I’veused brass plates that are soldered tothe ground plane. The connectors arethen attached to the plates withscrews. If the transistor mountingplate is a close fit, it may be necessaryto bevel it to accommodate the solderfillet. For testing purposes, one oftenneeds just the front and back plates—the side plates aren’t needed until af-ter the amplifier is working. They areuseful for shielding the amplifier.With thick metal-backed circuitboard, you need only bolt the connec-tors to the ground plane.

Matching elements can also be diffi-cult to ground. Often, you don’t knowexactly where to place them, thoughmost designs do end up with a lot ofcapacitance near the input and outputof the active device. Appropriatelyplaced screws normally work well,unless you guess wrong. A better solu-tion is to use matching techniques thatdon’t require actual ground connec-tions, such as bits of copper foil. Thisworks well at 10 GHz with 2.2 or2.5 εr board, but the foil gets too big at2.3 GHz. One solution is a high-dielec-tric-constant board, such as Rogers6010, with a constant of 10. Then,small bits of copper foil (just 50 or 100mils on a side) have a significanteffect on the tuning. I’ve successfullytuned 3 W, 6 GHz IMFETs down to

2304 MHz on 30-mil-thick Rogers6010 board.

Another approach is to make a thinmetal plate that fits between the cir-cuit board and the transistor mount-ing plate. This eliminates the need foran expensive milling machine. Sheetaluminum comes in a variety of sizes—if you can’t find the exact thicknessyou need try combining two sheets.There may be a complication when youtry to attach connectors with mount-ing screws—solid pieces of metal aregenerally required for drilling andtapping holes. Thus, it may be neces-sary to rotate the connector to obtainbetter connection points.

Notes1Keith Erickson, K0KE, “Tuning of Micro-

wave Stripline Amplifiers,” Proceedings ofthe Microwave Update ’87.

2Mitsubishi Technical Bulletin Recommen-dations for Mounting High-Power GaAsFET Package Devices, p 7.2.1.

3http://kepro.com/

Jan/Feb 1999 61

Letters to the EditorOf Fields, Near and Far, in“A Test for Ambient Noise”◊ Several times, I have started towrite a letter to QEX regardingAppendix 1 of the Sep/Oct ’98 article“A Test for Ambient Noise.” I am sur-prised this Appendix was published“as is,” although I guess there is muchmisconception over what is meant bythe far and near fields of an antenna.Some noted authors, including thosein Jasik’s Antenna Handbook, surehave not helped the matter.

Basically, the technique given inAppendix 1 is invalid. The distance(2 d 2)/λ has nothing to do with thenature of the fields around a basicdipole or loop antenna. It is an arbi-trary boundary (from ray optics) be-tween the Fresnel and Fraunhoferregions. That is where the rays froman aperture are nearly focused—if Irecall correctly, the phase error is lessthan λ/8.

In the case of phased arrays andreflector antennas, this distance isabout where the main beam gain ofthe antenna is pretty close to its gainat an infinite distance (the far-fieldgain). The main beam gain of an ap-erture at less than the (2 d 2)/λ dis-tance is less than the far-field value.Hence, to compute the power densitywithin the main beam (based onsimple spreading loss) at distancesless than (2 d 2) /λ , one has to reducethe far-field gain of the antenna.

In both the Fresnel and Fraunhoferregions, the field is generally a planewave. In the Appendix, the authorwas referring to the case where thereis not a plane wave, which is a totallydifferent concept. Here, we are refer-ring to the case where there is morethan one E field and/or one H fieldaround the antenna. In the case of ashort dipole, there are two E fields—an E field normal to the axis of thedipole, an E field parallel to the axisof the dipole and an H field concentricabout the dipole axis. The cross prod-uct of these E fields and H field istaken to form the Poynting vector,representing the propagation of en-ergy. The E field parallel to the axisof the dipole contributes to a Poyntingvector directed radially outward fromthe dipole axis; thus the power isradiated outward. The other E field

will produce Poynting vectors thatare opposite (in phase) on each side ofthe dipole, directed upward or down-ward along the axis of the dipole,canceling each other out. These fieldsdo not contribute to the outward ra-diation of energy, and they are some-times called “reactive fields.”

Apparently, the intent of theAppendix was to outline a method ofdetermining at what point the normalE field around a dipole or monopolewould become insignificant comparedto the E field parallel to the axis.Obviously, what is insignificant de-pends on your purpose—a level 10 or20 dB down might be satisfactory.The best way I know to determinethis distance is with a method-of-mo-ments-based model, such as is used inNEC, MININEC or similar codes.

By the way, the distance used inthe Fresnel-Fraunhofer boundaryequation is the actual dimension ofthe aperture, not the effective areacomputed from Eq 7 in the Appendix.The effective area is often only 40% to60% of the physical area for most re-flector-type antennas. I wonder ifQEX needs a tutorial on this subjectand its impact on RADHAZ calcula-tions?—Lee Garlock, KD4RE, 3163Plantation Pkwy, Fairfax, VA 22030◊ I think you’ve just given us one!Nevertheless, could not reality be de-scribed more simply? Try the follow-ing: When current flows in a wire,electric and magnetic fields are cre-ated around the wire. When the cur-rent ceases or reverses direction, thefields tend to collapse back onto thewire. Some of the fields’ energy,propagating away from the wire atvelocity c, escapes into free space.

It’s interesting to note that as theoperating frequency decreases, it be-comes easier for the fields to collapsebefore they can radiate away. At fre-quencies in the audio range, for ex-ample, almost all field energy isreturned to the antenna, making ra-diation efficiency very poor. Even so,the near-field intensity can be strong,hence the concern regarding humanexposure to fields from antennas andpower lines. The following is theauthor’s response.—Doug Smith,KF6DX, QEX Editor; dsmith@arrl.org

◊ I am grateful for your discussion ofnear and far fields. Because most ofus know an approximate value for ourantenna gain, I supplied a method toestimate the edge of the far field,based on antenna gain. In the Decem-ber 1987 issue of QEX, H. PaulShuch, N6TX, published a differentapproach in his paper “Far-Field Fal-lacy.” He saw the edge of the far fieldas the distance where the path loss is14 dB greater than the gain of twoantennas looking at each other. I willtake two antenna examples and cal-culate the edge of the far field usingmy method and that of Mr. Shuch.

1a. Two half-wave dipoles are oper-ating at 14 MHz. The edge of the farfield from Appendix 1 is:

The wavelength (λ) at 14 MHz =21.43 meters, the dipole gain is2.15 dB over isotropic, or G = 1.64 asa power ratio. This gives the apparentantenna area of:

A = [G λ2] / (4π)= 1.64 ( 21.432 / (4 × 3.14)= 59.9 m2

The apparent antenna diametersquared (D2) is:

D2 = 4 A / π= 4 × 59.9 / 3.14= 76.3 m2

The edge of the far field is:R = 2 D 2 / λ

= 2 × 76.3 / 21.43= 7.12 m or 23.4 ft

1b. The edge of far field from Requals the distance where path loss is14 dB greater than the gain of twoantennas:

The combined gain of two dipoles is2 × 2.15 dB = 4.30 dB. The specifiedpath loss is 4.30 + 14 = 18.30 dB. Thepath loss is:–27.56 + 20 log (f) + 20 log (R), or20 log (R) = path loss + 27.53 -20 log (f)

= 18.30 + 27.53 -22.92= 22.91 dB

R = 13.98 m or 45.86 ftWheref = frequency, in megahertzR = radius, in meters

2a. Basic Yagi with a gain of 7 dBabove a dipole and a simple dipole at14 MHz. The edge of the far field fromAppendix 1:

The wavelength (λ) at 14 MHz =21.43 meters. The Yagi gain is 9.15 dBover isotropic, or G = 8.22 as a powerratio. This gives the apparent an-tenna area of:

A = [G λ2] / (4π)= 8.22 × 21.432 / (4 × 3.14)= 300.4 m2

The apparent antenna diametersquared is:

62 QEX

D 2 = 4 A / π= 4 × 59.9 / 3.14= 382.5 m2

The edge of the far field is:R = 2 D 2 / λ

= 2 × 76.3 / 21.43= 35.70 m or 117.1 ft

2b. The edge of far field from Requals the distance where path loss is14 dB greater than the gain of twoantennas:

The combined gain of two antennasis 2.15 + 9.15 = 11.30 dB. The speci-fied path loss is:11.30 + 14 = 25.30 dBpath loss = –27.56 + 20 log (f) + 20 log(R), or20 log (R) = path loss + 27.53 -20 log (f)

= 25.30 + 27.53 -22.92=29.91 dB;

or R = 31.30 m or 102.7 ft.The definition of “far field” in this

application is somewhat academic.The goal is to find a generic method tocalculate the distance between anten-nas where the experimental noisemeasurement results represent an-tennas a great distance from eachother. One could test the experimen-tal results by first measuring at onedistance and then measuring again attwice the distance. Mr. Shuch sug-gested that we take his distance anddouble it for antenna gain measure-ments.—Pete Lefferson, K4POB, 61017th Ave N, Saint Petersburg, FL33710-7015; elecleff@cftnet.com

Signals, Samples and Stuff: A DSPTutorial◊ I was just reading Part 4 of DougSmith’s DSP series in the Sep/Oct ’98QEX. I could not follow a step in theequations.

In the DFFT analysis, I think thereis a typographical mistake in the sec-ond line of Eq 5. My calculations showthat there should be a k term in thefirst exponent, and there is an e miss-ing after the multiplication dot.

I did reach the same result, sincethe first e term reduces to 1, and thesecond term is Wn

–k.Thanks for putting this series to-

gether. DSP articles in ham radiomagazines are rare. We need more!—Pete Wyckoff, KA3WCA, 1080Taylorsville Rd, Washington Crossing,PA 18977, Member, Technical Staff,Personal Earth Station Hardware En-gineering, Hughes Network Systems,◊ You are correct! That one slipped byus at the last moment, although theresult of the derivation is accurate.Thanks for taking the time to write.—Doug Smith, KF7DX, QEX Editor

Educational Activity AwardNominations◊ The ARRL offers four awards to hamradio instructors and recruiters:

• Herb S. Brier Instructor of theYear (a nonpaid volunteer instructor)

• Professional Instructor of the Year(a paid, non-state licensed, instructor)

• Professional Educator of theYear (a professional teacher)

• Excellence in Recruiting (hamswho exemplify outstanding recruitingenthusiasm and technique).

Does someone you know (maybeyou) deserve this recognition? Startthe nomination process today. Accessour Web page at http://www.arrl.org/ead/award/ for informationabout the awards. You can downloadnomination forms at http://www.arrl.org/ead/award/application.html or call Jean at 860-594-0219 fora nomination form and informationon requirements and deadlines. Sendcompleted nomination forms to yourSection Manager by January 31. A jobwell done should be rewarded.

The Tetrode Boards◊ My article about tetrode power sup-plies (“Power and Protection for Mod-ern Tetrodes,” QEX, Oct 1997, pp15-26) has led to a commercial prod-uct called THE TETRODE BOARDS.This product is a new solution for thecontrol and protection of tetrodepower amplifiers. The circuits willwork with any transmitting tetrodefor amateur power levels, in anypower-supply grounding arrange-ment. Two small PC boards includeregulated power supplies for thescreen and control grids, screen andgrid current protection, T/R sequenc-

ing, ALC and relay supplies—that’salmost everything except the high-voltage supply and the tetrode!

The kit includes all the componentsfor the PC boards, a comprehensive32-page manual and full supportfrom the designer. Experienced con-structors can buy the bare boardsand manual. A special mains trans-former is also available, that connectsdirectly to the boards and provideseverything except the anode voltage.

For details, see http://www.ifwtech.demon.co.uk/g3sek or con-tact Down East Microwave Inc, tel908-996-3584. Outside the USA, con-tact Ian White, G3SEK at the Callbookaddress, or e-mail tetrode-boards@ifwtech.demon.co.uk—Ian White,G3SEK, ian@ifwtech .demon.co.uk

Multiple-Octave BidirectionalWire Antennas◊ I received about 10 e-mails regard-ing my Jul/Aug ’98 QEX article beforeI left IBM and cancelled my e-mailaddress there. Readers can now reachme at W7SX@aol.com.—Bob Zavrel,W7SX/0

Practical Hot-Guy-WireAntennas for Ham Radio◊ I’ve noticed an error in my Nov/DecQEX article. In Figure 13 B (page 55)the vertical dipole strung alongsidethe tower shows the bottom of the di-pole touching ground. While there isno connection dot there, I want tomake it clear that the dipole endshould be insulated from ground. Inparagraph 2 of page 53, “l /2” shouldbe “λ/2.”—Grant Bingeman, KM5KG,1908 Paris Ave, Plano, TX 75025;DrBingo@compuserve.com

Among other features in the nextissue of QEX, Rudy Severns, N6LF re-turns (!) as an author with some newand interesting ways of examining theproperties of full-size vertical anten-nas with counterpoise radials. Rudyhas put his real estate and timber toexcellent use—wait ’til you see how heerected his 160-meter vertical usingonly a “small” amount of base loading.Modeling and practice converge neatlyin this very engaging article.

Next Issue in QEXKen Beals, WK6F, puts the serial

control port of his HF transceiver togood use over a 10 GHz remote-con-trol link. Ken takes us over the sys-tem at the block diagram level,lightly, then provides some useful in-formation about how to get it all intoa box or two. Along the way, he dis-cusses the general requirements ofremotely controlled systems and illu-minates the many decisions facingdesigners. Complete schematics andPC board layouts are available forreaders serious about circumventingantenna restrictions and RF-expo-sure problems.

Jan/Feb 1999 63

64 QEX