Oscillators and Synthesizers 10.1

Oscillators andSynthesizers

Chapter 10

Just say in public that oscillators are oneof the most important, fundamental build-ing blocks in radio technology and you willimmediately be interrupted by someonepointing out that tuned-RF (TRF) receiverscan be built without any form of oscillatorat all. This is certainly true, but it showshow some things can be taken for granted.What use is any receiver without signals toreceive? All intentionally transmitted sig-nals trace back to some sort of signal gen-erator—an oscillator or frequencysynthesizer. In contrast with the TRF re-ceivers just mentioned, a modern, all-mode, feature-laden MF/HF transceivermay contain in excess of a dozen RF oscil-lators and synthesizers, while a simpleQRP CW transmitter may consist of noth-ing more than a single oscillator. (Thischapter was written by David Stockton,GM4ZNX. Frederick J. Telewski,WA7TZY, also contributed to the Fre-quency Synthesizers section.)

In the 1980s, the main area of progress inthe performance of radio equipment wasthe recognition of receiver intermodulationas a major limit to our ability to communi-cate, with the consequent development ofreceiver front ends with improved abilityto handle large signals. So successful wasthis campaign that other areas of trans-ceiver performance now require similarattention. One indication of this is anyequipment review receiver dynamic rangemeasurement qualified by a phrase like“limited by oscillator phase noise.” A plotof a receiver’s effective selectivity canprovide another indication of work to bedone: An IF filter’s high-attenuation regionmay appear to be wider than the filter’spublished specifications would suggest—

almost as if the filter characteristic hasgrown sidebands! In fact, in a way, it has:This is the result of local-oscillator (LO) orsynthesizer phase noise spoiling thereceiver’s overall performance. Oscillatornoise is the prime candidate for the nextmajor assault on radio performance.

The sheer number of different oscillatorcircuits can be intimidating, but their greatdiversity is an illusion that evaporates oncetheir underlying pattern is seen. Almost allRF oscillators share one fundamental prin-ciple of operation: an amplifier and a filteroperate in a loop (Fig 10.1). There areplenty of filter types to choose from:

• LC• Quartz crystal and other piezoelectric

materials• Transmission line (stripline, microstrip,

troughline, open-wire, coax and so on)• Microwave cavities, YIG spheres, di-

electric resonators• Surface-acoustic-wave (SAW) devices

Should any new forms of filter be in-vented, it’s a safe guess that they will also

be applicable to oscillators. There is anequally large range of amplifiers to choosefrom:

• Vacuum tubes of all types• Bipolar junction transistors• Field effect transistors (JFET, MOSFET,

GaAsFET, in all their varieties)• Gunn diodes, tunnel diodes and other

negative-resistance generators

It seems superfluous to state that any-thing that can amplify can be used in anoscillator, because of the well-known pro-pensity of all prototype amplifiers to os-cillate! The choice of amplifier is widenedfurther by the option of using single- ormultiple-stage amplifiers and discrete de-vices versus integrated circuits. Multiplyall of these options with those of filterchoice and the resulting set of combina-tions is very large, but a long way fromcomplete. Then there are choices of howto couple the amplifier into the filter andthe filter into the amplifier. And then thereare choices to make in the filter section:Should it be tuned by variable capacitor,variable inductor or some form of slidingcavity or line?

Despite the number of combinationsthat are possible, a manageably small num-ber of types will cover all but very specialrequirements. Look at an oscillator circuitand “read” it: What form of filter—reso-nator—does it use? What form of ampli-fier? How have the amplifier’s input andoutput been coupled into the filter? How isthe filter tuned? These are simple, easilyanswered questions that put oscillatortypes into appropriate categories and makethem understandable. The questions them-selves may make more sense if we under-

Fig 10.1—Reduced to essentials, anoscillator consists of a filter and anamplifier operating in a feedback loop.

Chapter 10.pmd 7/31/2006, 11:34 AM1

10.2 Chapter 10

stand the mechanics of oscillation, inwhich resonance plays a major role.

HOW OSCILLATORS WORKMaintained Resonance

The pendulum, a good example of aresonator, has been known for millenniaand understood for centuries. It is closelyanalogous to an electronic resonator, asshown in Fig 10.2. The weighted end ofthe pendulum can store energy in two dif-ferent forms: The kinetic energy of itsmotion and the potential energy of it beingraised above its rest position. As it reachesits highest point at the extreme of a swing,its velocity is zero for an instant as itreverses direction. This means that it has,at that instant, no kinetic energy, but be-cause it is also raised above its rest posi-tion, it has some extra potential energy.

At the center of its swing, the pendulumis at its lowest point with respect to gravityand so has lost the extra potential energy.At the same time, however, it is moving atits highest speed and so has its greatestkinetic energy. Something interesting ishappening: The pendulum’s stored energyis continuously moving between potentialand kinetic forms. Looking at the pendu-lum at intermediate positions shows thatthis movement of energy is smooth. New-ton provided the keys to understanding this.It took his theory of gravity and laws ofmotion to explain the behavior of a simpleweight swinging on the end of a length ofstring and calculus to perform a quantita-tive mathematical analysis. Experimentshad shown the period of a pendulum to bevery stable and predictable. Apart from sideeffects of air drag and friction, the length ofthe period should not be affected by the

mass of the weight, nor by the amplitude ofthe swing.

A pendulum can be used for timingevents, but its usefulness is spoiled by theaction of drag or friction, which eventu-ally stops it. This problem was overcomeby the invention of the escapement, a partof a clock mechanism that senses theposition of the pendulum and applies asmall push in the right direction and at theright time to maintain the amplitude ofits swing or oscillation. The result is amechanical oscillator: The pendulum actsas the filter, the escapement acts as theamplifier and a weight system or wound-up spring powers the escapement.

Electrical oscillators are closely analo-gous to the pendulum, both in operation andin development. The voltage and current inthe tuned circuit—often called tank circuitbecause of its energy-storage ability—bothvary sinusoidally with time and are 90° outof phase. There are instants when the cur-rent is zero, so the energy stored in the in-ductor must be zero, but at the same timethe voltage across the capacitor is at itspeak, with all of the circuit’s energy storedin the electric field between the capacitor’splates. There are also instants when thevoltage is zero and the current is at a peak,with no energy in the capacitor. Then, allof the circuit’s energy is stored in theinductor’s magnetic field.

Just like the pendulum, the energystored in the electrical system is swingingsmoothly between two forms; electricfield and magnetic field. Also like thependulum, the tank circuit has losses. Itsconductors have resistance, and the capa-citor dielectric and inductor core are im-perfect. Leakage of electric and magneticfields also occurs, inducing currents inneighboring objects and just plain radiat-ing energy off into space as radio waves.The amplitudes of the oscillating voltageand current decrease steadily as a result.Early intentional radio transmissions, suchas those of Heinrich Hertz’s experiments,involved abruptly dumping energy into atuned circuit and letting it oscillate, orring, as shown in Fig 10.3. This was doneby applying a spark to the resonator.Hertz’s resonator was a gapped ring, agood choice for radiating as much ofthe energy as possible. Although thislooks very different from the LC tank ofFig 10.2, it has inductance distributedaround its length and capacitance distrib-uted across it and across its gap, as opposedto the lumped L and C values in Fig 10.2.The gapped ring therefore works just thesame as the LC tank in terms of oscillatingvoltages and currents. Like the pendulumand the LC tank, its period, and thereforethe frequency at which it oscillates, is in-

Fig 10.2—A resonator lies at the heart of every oscillatory mechanical andelectrical system. A mechanical resonator (here, a pendulum) and an electricalresonator (here, a tuned circuit consisting of L and C in parallel) share the samemechanism: the regular movement of energy between two forms—potential andkinetic in the pendulum, electric and magnetic in the tuned circuit. Both of theseresonators share another trait: Any oscillations induced in them eventually dieout because of losses—in the pendulum, due to drag and friction; in the tunedcircuit, due to the resistance, radiation and inductance. Note that the curvescorresponding to the pendulum’s displacement vs velocity and the tuned circuit’svoltage vs current, differ by one quarter of a cycle, or 90°.

Chapter 10.pmd 7/31/2006, 11:34 AM2

Oscillators and Synthesizers 10.3

der to check that those loops cannot os-cillate.) Fig 10.4C shows what happensto the amplitude of an oscillator if theloop gain is made a little higher or lowerthan one.

The loop gain has to be precisely one ifwe want a stable amplitude. Any inaccu-racy will cause the amplitude to grow toclipping or shrink to zero, making the oscil-lator useless. Better accuracy will onlyslow, not stop this process. Perfect preci-sion is clearly impossible, yet there areenough working oscillators in existenceto prove that we are missing somethingimportant. In an amplifier, nonlinearity is anuisance, leading to signal distortion andintermodulation, yet nonlinearity is whatmakes stable oscillation possible. All of thevacuum tubes and transistors used in oscil-lators tend to reduce their gain at highersignal levels. With such components, onlya tiny change in gain can shift the loop’soperation between amplitude growth andshrinkage. Oscillation stabilizes at thatlevel at which the gain of the active devicesets the loop gain at exactly one.

Another gain-stabilization techniqueinvolves biasing the device so that oncesome level is reached, the device starts toturn off over part of each cycle. At higherlevels, it cuts off over more of each cycle.This effect reduces the effective gain quitestrongly, stabilizing the amplitude. Thisbadly distorts the signal (true in most com-mon oscillator circuits) in the amplifyingdevice, but provided the amplifier islightly coupled to a high-Q resonant tank,the signal in the tank should not be badlydistorted.

Many radio amateurs now have someform of circuit-analysis software, usuallyrunning on a PC. Attempts to analyze oscil-lators by this means often fail by predict-ing growing or shrinking amplitudes, andoften no signal at all, in circuits that areknown to work. Computer analysis of os-cillator circuits can be done, but it requiresa sophisticated program with accurate,nonlinear, RF-valid models of the devicesused, to be able to predict operating am-plitude. Often even these programs needsome special tricks to get their modeledoscillators to start. Such software is likelyto be priced higher than most private userscan justify, and it still doesn’t replace theneed for the user to understand the circuit.With that understanding, some time, someparts and a little patience will do the job,unassisted.

Textbooks give plenty of coverage tothe frequency-determining mechanismsof oscillators, but the amplitude-deter-mining mechanism is rarely covered. Itis often not even mentioned. There is agood treatment in Clarke and Hess, Commu-

Fig 10.3—Stimulating a resonance, 1880s style. Shock-exciting a gapped ring withhigh voltage from a charged capacitor causes the ring to oscillate at its resonantfrequency. The result is a damped wave, each successive alternation of which isweaker than its predecessor because of resonator losses. Repetitively stimulatingthe ring produces trains of damped waves, but oscillation is not continuous.

dependent of the magnitude of its excita-tion.

Making a longer-lasting signal with theFig 10.3 arrangement merely involvesrepeating the sparks. The problem is that atruly continuous signal cannot be made thisway. The sparks cannot be applied oftenenough or always timed precisely enoughto guarantee that another spark re-excitesthe circuit at precisely the right instant. Thisarrangement amounts to a crude sparktransmitter, variations of which served asthe primary means of transmission for thefirst generation of radio amateurs. The useof damped waves is now entirely forbiddenby international treaty because of their greatimpurity. Damped waves look a lot like car-ignition waveforms and sound like car-ig-nition interference when received.

What we need is a continuous wave(CW) oscillation—a smooth, sinusoidalsignal of constant amplitude, withoutphase jumps, a “pure tone.” To get it, wemust add to our resonator an equivalent ofthe clock’s escapement—a means of syn-chronizing the application of energy and afast enough system to apply just enoughenergy every cycle to keep each cycle atthe same amplitude.

AmplificationA sample of the tank’s oscillation can be

extracted, amplified and reinserted. Thegain can be set to exactly compensate the

tank losses and perfectly maintain theoscillation. The amplifier usually need onlygive low gain, so active devices can be usedin oscillators not far below their unity(unity = 1) gain frequency. The amplifier’soutput must be lightly coupled into thetank—the aim is just to replace lost energy,not forcibly drive the tank. Similarly, theamplifier’s input should not heavily loadthe tank. It is a good idea to think of cou-pling networks rather than matching net-works in this application, because amatched impedance extracts the maximumavailable energy from a source, and thiswould certainly spoil an oscillator.

Fig 10.4A shows the block diagram ofan oscillator. Certain conditions must bemet for oscillation. The criteria that sepa-rate oscillator loops from stable loops areoften attributed to Barkhausen by thoseaiming to produce an oscillator and toNyquist by those aiming for amplifier sta-bility, although they boil down to thesame boundary. Fig 10.4B shows the loopbroken and a test signal inserted. (Theloop can be broken anywhere; the ampli-fier input just happens to be the easiestplace to do it.) The criterion for oscilla-tion says that at a frequency at which thephase shift around the loop is exactlyzero, the net gain around the loop mustequal or exceed unity (that is, one).(Later, when we design phase-lockedloops, we must revisit this concept in or-

Chapter 10.pmd 7/31/2006, 11:34 AM3

10.4 Chapter 10

nications Circuits: Analysis and Design.

Start-UpPerfect components don’t exist, but if

we could build an oscillator from them,we would naturally expect perfect per-formance. We would nonetheless be disap-pointed. We could assemble from ourperfect components an oscillator thatexactly met the criterion for oscillation,having slightly excessive gain that falls tothe correct amount at the target operatinglevel and so is capable of sustained, stableoscillation. But being capable of somethingis not the same as doing it, for there is an-other stable condition. If the amplifier inthe loop shown in Fig 10.4 has no inputsignal, and is perfect, it will give no output!No signal returns to the amplifier’s inputvia the resonator, and the result is a sus-tained and stable lack of oscillation. Some-thing is needed to start the oscillator.

This fits the pendulum-clock analogy:A wound-up clock is stable with its pendu-lum at rest, yet after a push the system willsustain oscillation. The mechanism thatdrives the pendulum is similar to a Class Camplifier: It does not act unless it is drivenby a signal that exceeds its threshold. Anelectrical oscillator based on a Class Camplifier can sometimes be kicked intoaction by the turn-on transient of its powersupply. The risk is that this may not al-ways happen, and also that should someexternal influence stop the oscillator, itwill not restart until someone notices theproblem and cycles the power. This can bevery inconvenient!

A real-life oscillator whose amplifierdoes not lose gain at low signal levels canself-start due to noise. Fig 10.5 shows anoscillator block diagram with the ampli-fier’s noise shown, for our convenience, asa second input that adds with the true input.The amplifier amplifies the noise. The reso-nator filters the output noise, and this sig-nal returns to the amplifier input. Theimportance of having slightly excessivegain until the oscillator reaches operatingamplitude is now obvious. If the loop gainis slightly above one, the recirculated noisemust, within the resonator’s bandwidth, belarger than its original level at the input.More noise is continually summed in as anoise-like signal continuously passesaround the loop, undergoing amplificationand filtering as it does. The level increases,causing the gain to reduce. Eventually, it

Fig 10.4—The bare-bones oscillator blockdiagram of Fig 10.1did not include twopractical essentials:Networks to couplepower in and out ofthe resonator (A).Breaking the loop,inserting a test signaland measuring theloop’s overall gain(B) allows us todetermine whetherthe system canoscillate, sustainoscillation or clip (C).

Fig 10.5—An oscillator with noise. Real-world amplifiers, no matter how quiet,generate some internal noise; thisallows real-world oscillators to self-start.

Chapter 10.pmd 7/31/2006, 11:34 AM4

Oscillators and Synthesizers 10.5

stabilizes at whatever level is necessary tomake the net loop gain equal to one.

So far, so good. The oscillator is runningat its proper level, but something seemsvery wrong. It is not making a proper sinewave; it is recirculating and filtering a noisesignal. It can also be thought of as a Qmultiplier with a controlled (high) gain,filtering a noise input and amplifying it toa set level. Narrow-band filtered noiseapproaches a true sine wave as the filter isnarrowed to zero width. What this means isthat we cannot make a true sine-wave sig-nal—all we can do is make narrow-bandfiltered noise as an approximation to one.A high-quality, low-noise oscillator ismerely one that does tighter filtering. Evena kick-started Class-C-amplifier oscillatorhas noise continuously entering its circu-lating signal, and so behaves similarly.

A small-signal gain greater than one isabsolutely critical for reliable starting, buthaving too much gain can make the finaloperating level unstable. Some oscillatorsare designed around limiting amplifiers tomake their operation predictable. AGC sys-tems have also been used, with an RFdetector and dc amplifier used to servo-con-trol the amplifier gain. It is notoriously dif-ficult to design reliable crystal oscillatorsthat can be published or mass-producedwithout having occasional individualsrefuse to start without some form of shock.

Mathematicians have been intrigued by“chaotic systems” where tiny changes ininitial conditions can yield large changes inoutcome. The most obvious example ismeteorology, but much of the necessarymath was developed in the study of oscilla-tor start-up, because it is a case of chaoticactivity in a simple system. The equationsthat describe oscillator start-up are similarto those used to generate many of the popu-lar, chaotic fractal images.

PHASE NOISEViewing an oscillator as a filtered-noise

generator is relatively modern. The olderapproach was to think of an oscillatormaking a true sine wave with an added,unwanted noise signal. These are just dif-ferent ways of visualizing the same thing:They are equally valid views, which areused interchangeably, depending whichbest makes some point clear. Thinking interms of the signal-plus-noise version, thenoise surrounds the carrier, looking likesidebands and so can also be considered tobe equivalent to random-noise FM and AMon the ideal sine-wave signal. This givesus a third viewpoint. Strangely, these noisesidebands are called phase noise. If weconsider the addition of a noise voltage toa sinusoidal voltage, we must take intoaccount the phase relationship. A phasor

Fig 10.6—At A, a phasor diagram of a clean (ideal) oscillator. Noise creates aregion of uncertainty in the vector’s length and position (B). AM noise varies thevector’s length; PM noise varies the vector’s relative angular position (C) Limitinga signal that includes AM and PM noise strips off the AM and leaves the PM (D).

Chapter 10.pmd 7/31/2006, 11:34 AM5

10.6 Chapter 10

diagram is the clearest way of illustratingthis. Fig 10.6A represents a clean sinewave as a rotating vector whose length isequal to the peak amplitude and whosefrequency is equal to the number of revo-lutions per second of its rotation. Movingthings are difficult to depict on paper, sophasor diagrams are usually drawn to showthe dominant signal as stationary, withother components drawn relative to this.

Noise contains components at many fre-quencies, so its phase with respect to thedominant, theoretically pure signal—the“carrier”—is random. Its amplitude is alsorandom. Noise can only be described instatistical terms because its voltage is con-stantly and randomly changing, yet it doeshave an average amplitude that can beexpressed in rms volts. Fig 10.6B showsnoise added to the carrier phasor, with thenoise represented as a fuzzy, uncertainregion in which the sum phasor wandersrandomly. The phase of the noise is uni-formly random—no direction is morelikely than any other—but the instanta-neous magnitude of the noise obeys a prob-ability distribution like that shown, highervalues being progressively rarer. Fig10.6C shows how the extremities of thenoise region can be considered as extremesof phase and amplitude variation from thenormal values of the carrier.

Phase modulation and frequency modu-lation are closely related. Phase is the in-tegral of frequency, so phase modulationresembles frequency modulation in whichdeviation decreases with increasing modu-lating frequency. Thus, there is no need totalk of “frequency noise” because phasenoise already covers it.

Fig 10.6C clearly shows AM noise asthe random variation of the length of thesum phasor, yet “amplitude noise” israrely discussed. The oscillator’s ampli-tude control mechanism acts to reduce theAM noise by a small amount, but the mainreason is that the output is often fed intosome form of limiter that strips off AMcomponents just as the limiting IF ampli-fier in an FM receiver removes any AM onincoming signals. The limiter can be obvi-ous, like a circuit to convert the signal tologic levels, or it can be implicit in someother function. A diode ring mixer may bedriven by a sine-wave LO of moderatepower, yet this signal drives the diodeshard on and hard off, approximatingsquare-wave switching. This is a form oflimiter, and it removes the effect of anyAM on the LO. Fig 10.6D shows the resultof passing a signal with noise through alimiting amplifier. For these reasons, AMnoise sidebands are rarely a problem inoscillators, and so are normally ignored.There is one subtle problem to beware of,

however. If a sine wave drives some formof switching circuit or limiter, and thethreshold is offset from the signal’s meanvoltage, any level changes will affect theexact switching times and cause phasejitter. In this way, AM sidebands are trans-lated into PM sidebands and pass throughthe limiter. This is usually called AM-to-PM conversion and is a classic problem oflimiters.

Effects of Phase NoiseYou would be excused for thinking that

phase noise is a recent discovery, but alloscillators have always produced it. Other

changes have elevated an unnoticed char-acteristic up to the status of a serious im-pairment. Increased crowding and powerlevels on the ham bands, allied with greaterexpectations of receiver performance as aresult of other improvements, have madephase noise more noticeable, but the big-gest factor has been the replacement ofVFOs in radios by frequency synthesizers.It is a major task to develop a synthesizerthat tunes in steps fine enough for SSB andCW while competing with the phase-noiseperformance of a reasonable-quality LCVFO. Many synthesizers have fallen farshort of that target. Phase noise is worse in

Fig 10.7—The effects of phase noise. At A, the relative phase-noise spectra ofseveral different signal-generation approaches. At B, how transmitted phase noisedegrades the weak-signal performance even of nearby receivers with phase-quietoscillators, raising the effective noise floor. What is perhaps most insidious aboutphase noise in Amateur Radio communication is that its presence in a receiver LOcan allow strong, clean transmitted signals to degrade the receiver’s noise floorjust as if the transmitted signals were dirty. This effect, reciprocal mixing, isshown at C.

Chapter 10.pmd 7/31/2006, 11:34 AM6

Oscillators and Synthesizers 10.7

higher-frequency oscillators and the trendtowards general-coverage, upconvertingstructures has required that local oscilla-tors operate at higher and higher frequen-cies. Fig 10.7A shows sketches of therelative phase-noise performance of someoscillators. The very high Q of the quartzcrystal in a crystal oscillator gives it thepotential for much lower phase noise thanLC oscillators. A medium-quality synthe-sizer has close-in phase noise performanceapproaching that of a crystal oscillator,while further from the carrier frequency, itdegrades to the performance of a modest-quality LC oscillator. There may be a smallbump in the noise spectrum at the boundaryof these two zones. A bad synthesizer canhave extensive noise sidebands extendingover many tens (hundreds in extreme cases)of kilohertz at high levels.

Phase noise on a transmitter’s localoscillators passes through its stages, isamplified and fed to the antenna along withthe intentional signal. The intentional sig-nal is thereby surrounded by a band ofnoise. This radiated noise exists in thesame proportion to the transmitter poweras the phase noise was to the oscillatorpower if it passes through no narrow-bandfiltering capable of limiting its bandwidth.This radiated noise makes for a noisierband. In bad cases, nearby stations can beunable to receive over many tens of kilo-hertz. Fig 10.7B illustrates the differencebetween clean and dirty transmitters.

The effects of receiver-LO phase noiseare more complicated, but at least itdoesn’t affect other stations’ reception.The process is called reciprocal mixing.

Reciprocal MixingThis is an effect that occurs in all mixers,

yet despite its name, reciprocal mixing isan LO, not a mixer, problem. Imagine thatthe outputs of two supposedly unmodu-lated signal generators are mixed togetherand the mixer output is fed into an FMreceiver. The receiver produces sounds,indicating that the resultant signal ismodulated nonetheless. Which signal gen-erator is responsible? This is a trick ques-tion, of course. A moment spent thinkingabout how a change in the frequency ofeither input signal affects the output sig-nal will show that FM or PM on eitherinput reaches the output. The best answerto the trick question is therefore “either orboth.”

The modulation on the mixer output isthe combined modulations of the inputs.This means that modulating a receiver’slocal oscillator is indistinguishable fromusing a clean LO and having the samemodulation present on all incoming sig-nals. (This is also true for AM provided

that a fully linear multiplier is used as themixer, but mixers are commonly driveninto switching, which strips any AM offthe LO signal. This is the chief reason whythe phase component of oscillator noise ismore important than any AM component.)

The word indistinguishable is importantin the preceding paragraph. It does not meanthat the incoming signals are themselvesmodulated, but that the signals in the re-ceiver IF and the noise in the IF, soundexactly as if they were. What really hap-pens is that the noise components of theLO are extra LO signals that are offset fromthe carrier frequency. Each of them mixesother signals that are appropriately offsetfrom the LO carrier into the receiver’s IF.Noise is the sum of an infinite number ofinfinitesimal components spread over arange of frequencies, so the signals it mixesinto the IF are spread into an infinite num-ber of small replicas, all at different fre-quencies. This amounts to scrambling theseother signals into noise. It is tedious to lookat the effects of receiver LO phase noisethis way. The concept of reciprocal mixinggives us an easier, alternative view that ismuch more digestible and produces identi-cal results.

A poor oscillator can have significantnoise sidebands extending out many tens ofkilohertz on either side of its carrier. This isthe same, as far as the signals in the re-ceiver IF are concerned, as if the LO wereclean and every signal entering the mixerhad these noise sidebands. Not only willthe wanted signal (and its noise sidebands)be received, but the noise sidebands addedby the LO to signals near, but outside, thereceiver’s IF passband will overlap it. Ifthe band is busy, each of many signalspresent will add its set of noise sidebands—and the effect is cumulative. This producesthe appearance of a high background-noiselevel on the band. Many hams tend to acceptthis, blaming “conditions.”

Hams now widely understand receptionproblems due to intermodulation, andalmost everyone knows to apply RF attenu-ation until the signal gets no clearer.Intermodulation is a nonlinear effect, andthe levels of the intermod products fall bygreater amounts than the reduction in theintermodulating signals. The net result isless signal, but with the intermodulationproducts dropped still further. This im-provement reaches a limit when moreattenuation pushes the desired signal tooclose to the receiver’s noise floor.

Reciprocal mixing is a linear process,and the mixer applies the same amount ofnoise “deviation” to incoming signals asthat present on the LO. Therefore the ratioof noise-sideband power to signal power isthe same for each signal, and the same as

that on the LO. Switching in RF attenua-tion reduces the power of signals enteringthe mixer, but the reciprocal mixing pro-cess still adds the noise sidebands at thesame relative power to each. Therefore, noreception improvement results. Other thanbuilding a quieter oscillator, the only wayof improving things is to use narrowpreselection to band-limit the receiver’sinput response and reduce the number ofincoming signals subject to reciprocalmixing. This reduces the number of timesthe phase noise sidebands get added intothe IF signal. Commercial trackingpreselectors—selective front-end circuitsthat tune in step with a radio’s bandchanges and tuning, are expensive, but onethat is manually tuned would make a mod-est-sized home-brew project and could alsohelp reduce intermodulation effects. Whenusing a good receiver with a linear frontend and a clean LO, amateurs accustomedto receivers with poor phase-noise perfor-mance report the impression is of a seem-ingly emptier band with gaps betweensignals—and then they begin to find read-able signals in some of the gaps.

Fig 10.7C shows how a noisy oscillatoraffects transmission and reception. Theeffects on reception are worst in Europe,on 40 m, at night. Visitors from NorthAmerica, and especially Asia, are usuallyshocked by the levels of background noise.In ITU Region 1, the Amateur Radio 40-mallocation is 7.0 to 7.1 MHz; above this,ultra-high-power broadcasters operate.The front-end filters in commercial hamgear are usually fixed band-pass designsthat cover the wider 40-m allocations inthe other regions. This allows huge signalsto reach the mixer and mix large levels ofLO phase noise into the IF. Operating col-located radios, on the same band, in amultioperator contest, requires linear frontends, preselection and state-of-the-artphase-noise performance. Outside of ama-teur circles, only warship operation ismore demanding, with kilowatt transmit-ters and receivers sharing antennas on thesame mast.

A Phase Noise DemonstrationHealthy curiosity demands some form

of demonstration so the scale of a problemcan be judged “by ear” before measure-ments are attempted. We need to be ableto measure the noise of an oscillator alone(to aid in the development of quieterones) and we also need to be able to mea-sure the phase noise of the oscillators in areceiver (a transmitter can be treated asan oscillator). Conveniently, a receivercontains most of the functions needed todemonstrate its own phase noise.

No mixer has perfect port-to-port iso-

Chapter 10.pmd 7/31/2006, 11:34 AM7

10.8 Chapter 10

lation, and some of its local-oscillator sig-nal leaks through into the IF. If we tunea general-coverage receiver, with itsantenna disconnected, to exactly 0 Hz, thelocal oscillator is exactly at the IF centerfrequency, and the receiver acts as if it istuned to a very strong unmodulated carrier.A typical mixer might give only 40 dB ofLO isolation and have an LO drive power ofat least 10 mW. If we tune away from0 Hz, the LO carrier tunes away from the IFcenter and out of the passband. The appar-ent signal level falls. Although this movesthe LO carrier out of the IF passband, someof its noise sidebands will not be, and thereceiver will respond to this energy as anincoming noise signal. To the receiver op-erator, this sounds like a rising noise flooras the receiver is tuned toward 0 Hz. To getgood noise floor at very low frequencies,some professional/military receivers, likethe Racal RA1772, use very carefully bal-anced mixers to get as much port-to-portisolation as possible, and they also mayswitch a crystal notch filter into the firstmixer’s LO feed.

This demonstration cannot be done ifthe receiver tunes amateur bands only. Asit is, most general-coverage radios inhibittuning in the LF or VLF region. It could besuggested by a cynic that how low manu-facturers allow you to tune is an indicationof how far they think their phase-noisesidebands could extend!

The majority of amateur transceiverswith general-coverage receivers are pro-grammed not to tune below 30 to 100 kHz,so means other than the “0 Hz” approachare needed to detect LO noise in theseradios. Because reciprocal mixing adds theLO’s sidebands to clean incoming signals,in the same proportion to the incomingcarrier as they exist with respect to the LOcarrier, all we need do is to apply a strong,clean signal wherever we want within thereceiver’s tuning range. This signal’s gen-erator must have lower phase noise thanthe radio being evaluated. A general-pur-pose signal generator is unlikely to be goodenough; a crystal oscillator is needed.

It’s appropriate to set the level into thereceiver to about that of a strong broadcastcarrier, say S9 + 40 dB. Set the receiver’smode to SSB or CW and tune around thetest signal, looking for an increasing noisefloor (higher hiss level) as you tune closertowards the signal, as shown in Fig 10.8.Switching in a narrow CW filter allowsyou to hear noise closer to the carrier thanis possible with an SSB filter. This is alsothe technique used to measure a receiver’seffective selectivity, and some equipmentreviewers kindly publish their plots in thisformat. QST reviews, done by the ARRLLab, often include the results of specific

phase-noise measurements.

Measuring Receiver Phase NoiseThere are several different ways of

measuring phase noise, offering differenttradeoffs between convenience, cost andeffort. Some methods suit oscillators inisolation, others suit them in-situ (in theirradios).

If you’re unfamiliar with noise measure-ments, the units involved may seemstrange. One reason for this is that a noisesignal’s power is spread over a frequencyrange, like an infinite number of infinitesi-mal sinusoidal components. This can bethought of as similar to painting a house.The area that a gallon of paint can coverdepends on how thinly it’s spread. If some-one asks how much paint was used on somepart of a wall, the answer would have to bein terms of paint volume per square foot.The wall can be considered to be an infi-nite number of points, each with an infini-tesimal amount of paint applied to it. Thequestion of what volume of paint has beenapplied at some specific point is unanswer-able. With noise, we must work in terms ofpower density, of watts per hertz. We there-fore express phase-noise level as a ratio ofthe carrier power to the noise’s power den-sity. Because of the large ratios involved,expression in decibels is convenient. It hasbeen a convention to use dBc to mean“decibels with respect to the carrier.”

For phase noise, we need to work interms of a standard bandwidth, and 1 Hz

is the obvious candidate. Even if the noiseis measured in a different bandwidth, itsequivalent power in 1 Hz can be easily cal-culated. A phase-noise level of –120 dBcin a 1-Hz bandwidth (often written as–120 dBc/Hz) translates into each hertz ofthe noise having a power of 10–12 of thecarrier power. In a bandwidth of 3 kHz,this would be 3000 times larger.

The most convenient way to measurephase noise is to buy and use a commercialphase noise test system. Such a systemusually contains a state-of-the-art, low-noise frequency synthesizer and a low-fre-quency spectrum analyzer, as well as somespecial hardware. Often, a second, DSP-based spectrum analyzer is included tospeed up and extend measurements veryclose to the carrier by using the Fast Fou-rier Transform (FFT). The whole systemis then controlled by a computer with pro-prietary software. With a good system likethis costing about $100,000, this is not apractical method for amateurs, although afew fortunate individuals have access tothem at work. These systems are also over-kill for our needs, because we are not par-ticularly interested in determining phase-noise levels very close to and very far fromthe carrier.

It’s possible to make respectable receiver-oscillator phase-noise measurements withless than $100 of parts and a multimeter.Although it’s time-consuming, the tech-nique is much more in keeping with theamateur spirit than using a $100k system!An ordinary multimeter will produceacceptable results, a meter capable of indi-cating “true rms” ac voltages is preferablebecause it can give correct readings on sinewaves and noise. Fig 10.9 shows the setup.Measurements can only be made around thefrequency of the crystal oscillator, so ifmore than one band is to be tested, crystalsmust be changed, or else a set of appropri-ate oscillators is needed. The oscillatorshould produce about +10 dBm (10 mW)and be very well shielded. (To this end, it’sadvisable to build the oscillator into a die-cast box and power it from internal batter-ies. A noticeable shielding improvementresults even from avoiding the use of anexternal power switch; a reed-relay elementinside the box can be positioned to connectthe battery when a small permanent magnetis placed against a marked place outside thebox.)

Likewise, great care must be taken withattenuator shielding. A total attenuationof around 140 dB is needed, and with somuch attenuation in line, signal leakagecan easily exceed the test signal thatreaches the receiver. It’s not necessary tobe able to switch-select all 140 dB of theattenuation, nor is this desirable, as

Fig 10.8—Tuning in a strong, cleancrystal-oscillator signal can allow youto hear your receiver’s relative phase-noise level. Listening with a narrow CWfilter switched in allows you to getbetter results closer to the carrier.

Chapter 10.pmd 7/31/2006, 11:34 AM8

Oscillators and Synthesizers 10.9

switches can leak. (The 1995 and oldereditions of this Handbook contained a stepattenuator that’s satisfactory.) All of theattenuators’ enclosure seams must be sol-dered. A pair of boxes with 30 dB of fixedattenuation each are needed to completethe set. With 140 dB of attenuation, coaxcable leakage is also a problem. The onlycountermeasure against this is to minimizeall cable lengths and to interconnect test-system modules with BNC plug-to-plugadapters (UG-491As) where possible.

Ideally, the receiver could simply betuned across the signal from the oscilla-tor and the response measured using itssignal-strength (S) meter. Unfortunately,receiver S meters are notoriously impre-cise, so an equivalent method is neededthat does not rely on the receiver’s AGCsystem.

The trick is not to measure the responseto a fixed level signal, but to measure thechanges in applied signal power needed togive a fixed response. Here is a step-by-step procedure based on that described byJohn Grebenkemper, KI6WX, in Marchand April 1988 QST:

1. Connect the equipment as shown inFig 10.9, but with the crystal oscillator off.Set the step attenuator to maximum attenu-ation. Set the receiver for SSB or CWreception with its narrowest available IFfilter selected. Switch out any internalpreamplifiers or RF attenuators. SelectAGC off, maximum AF and RF gain. Itmay be necessary to reduce the AF gain toensure the audio amplifier is at least 10 dBbelow its clipping point. The ac voltmeteror an oscilloscope on the AF output can beused to monitor this.

2. To measure noise, it is important toknow the bandwidth being measured. Atrue-RMS ac voltmeter measures thepower in the noise reaching it. To calcu-late the noise density, we need to divide bythe receiver’s noise bandwidth. Thereceiver’s –6-dB IF bandwidth can be usedas an approximation, but purists will wantto plot the top 20 dB of the receiver’s band-

width on linear scales and integrate thearea under it to find the width of a rect-angle of equal area and equal height. Thisaccounts properly for the noise in the skirtregions of the overall selectivity. (Thevery rectangular shape of common re-ceiver filters tends to minimize the errorof just taking the approximation.)

Switch on the test oscillator and set theattenuators to give an AF output above thenoise floor and below the clipping levelwith the receiver peaked on the signal.Tune the receiver off to each side to findthe frequencies at which the AF voltageis half that at the peak. The difference

between these is the receiver’s –6-dBbandwidth. High accuracy is not needed:25% error in the receiver bandwidth willonly cause a 1-dB error in the final result.The receiver’s published selectivity speci-fications will be close enough. The benefitof integration is greater if the receiver hasa very rounded, low-ringing or low-orderfilter.

3. Retune the receiver to the peak.Switch the oscillator off and note thenoise-floor voltage. Turn the oscillatorback on and adjust the attenuator to givean AF output voltage 1.41 times (3 dB)larger than the noise floor voltage. Thismeans that the noise power and the testsignal power at the AF output are equal—a value that’s often called the MDS (mini-mum discernible signal) of a receiver.Choosing a test-oscillator level at whichto do this test involves compromise.Higher levels give more accurate resultswhere the phase noise is high, but limit thelowest level of phase noise that can bemeasured because better receiver oscilla-tors require a greater input signal to pro-duce enough noise to get the chosen AF-output level. At some point, either we’vetaken all the attenuation out and our mea-surement range is limited by the test

Fig 10.9—Setup for measuring receiver-oscillator phase noise.

Table 10.1SSB Phase Noise of ICOM IC-745 Receiver SectionOscillator output power = –3 dBm (0.5 mW)Receiver bandwidth (Δf) = 1.8 kHzAudio noise voltage = –0.070 VAudio reference voltage (V

0) = 0.105 V

Reference attenuation (A0) = 121 dB

Offset Attenuation Audio Audio Ratio SSB PhaseFrequency (A1) (dB) V1 V2 V2/V1 Noise(kHz) (volts) (volts) (dBc/Hz) 4 35 0.102 0.122 1.20 –119 5 32 0.104 0.120 1.15 –122* 6 30 0.104 0.118 1.13 –124* 8 27 0.100 0.116 1.16 –127* 10 25 0.106 0.122 1.15 –129* 15 21 0.100 0.116 1.16 –133* 20 17 0.102 0.120 1.18 –137 25 14 0.102 0.122 1.20 –140 30 13 0.102 0.122 1.20 –141 40 10 0.104 0.124 1.19 –144 50 8 0.102 0.122 1.20 –146 60 6 0.104 0.124 1.19 –148 80 4 0.102 0.126 1.24 –150 100 3 0.102 0.126 1.24 –151 150 3 0.102 0.124 1.22 –151 200 0 0.104 –154 250 0 0.100 –154 300 0 0.98 –154 400 0 0.96 –154 500 0 0.96 –154 600 0 0.97 –154 800 0 0.96 –1541000 0 0.96 –154

*Asterisks indicate measurements possibly affected by receiver overload (see text).

Chapter 10.pmd 7/31/2006, 11:34 AM9

10.10 Chapter 10

oscillator’s available power, or we over-load the receiver’s front end, spoiling theresults.

Record the receiver frequency at thepeak, (f0), the attenuator setting (A0) andthe audio output voltage (V0). These arethe carrier measurements against which allthe noise measurements will be compared.

4. Now you must choose the offset fre-quencies—the spacings from the carrier—at which you wish to make measurements.The receiver’s skirt selectivity will limithow close to the carrier noise measure-ments can be made. (Any measurementsmade too close in are valid measurementsof the receiver selectivity, but because thesignal measured under these conditions issinusoidal and not noise like, the correc-tions for noise density and noise bandwidthare not appropriate.) It is difficult to decidewhere the filter skirt ends and the noisebegins, and what corrections to apply inthe region of doubt and uncertainty. A goodpractical approach is to listen to the audioand tune away from the carrier until youcan’t distinguish a tone in the noise. The

Fig A — ARRLLab phase-noisemeasurementsetup.

ear is superb at spotting sine tones buriedin noise, so this criterion, although subjec-tive, errs on the conservative side.

Tune the receiver to a frequency offsetfrom f0 by your first chosen offset andadjust the attenuators to get an audio out-put voltage as close as possible to V0.Record the total attenuation, A1 and theaudio output voltage, V1. The SSB phasenoise (qualified as SSB because we’remeasuring the phase noise on only one sideof the carrier, whereas some other meth-ods cannot segregate between upper andlower noise sidebands and measure theirsum, giving DSB phase noise) is now easyto calculate:

)BW(log10AA)f(L noise01 −=where

L(f) = SSB phase noise in dBc/HzBWnoise = receiver noise bandwidth,

Hz.5. It’s important to check for overload.

Decrease the attenuation by 3 dB, andrecord the new audio output voltage, V2. Ifall is well, the output voltage should in-

crease by 22% (1.8 dB); if the receiver isoperating nonlinearly, the increase will beless. (An 18% increase is still acceptablefor the overall accuracy we want.) RecordV2/V1 as a check: a ratio of 1.22:1 is ideal,and anything less than 1.18:1 indicates abad measurement.

If too many measurements are bad, youmay be overdriving the receiver’s AFamplifier, so try reducing the AF gain andstarting again back at Step 3. If this doesn’thelp, reducing the RF gain and startingagain at Step 3 should help if the compres-sion is occurring late in the IF stages.

6. Repeat Steps 4 and 5 at all the otheroffsets you wish to measure. If measure-ments are made at increments of about halfthe receiver’s bandwidth, any discrete(non-noise) spurs will be found. A notice-able tone in the audio can indicate the pres-ence of one of these. If it is well clear ofthe noise, the measurement is valid, butthe noise bandwidth correction should beignored, giving a result in dBc.

Table 10.1 shows the results for anICOM IC-745 as measured by KI6WX, and

Transmitter Phase-Noise Measurement in the ARRL LabHere is a brief description of the technique used in the

ARRL Lab to measure transmitter phase noise. Thesystem consists of a phase noise test set with low-noisevariable crystal oscillators for reference signals. The testset zero-beats the oscillator to the transmitter signal usingphase detectors on both signals. As shown in Fig A, weuse an attenuator after the transmitter to bring the signaldown to a suitable level for the phase noise test set. Thecrystal oscillator output is low level, so it does not requireattenuation.

The phase noise test set is an automated system runby a PC that steps through the test, setting parameters inthe spectrum analyzers and reading the data back fromthem, in addition to performing system calibration mea-surements at the start of the test. The computer screendisplays the transmitted phase-noise spectrum, which can

be printed or saved to a file on the hard drive. To test thebaseline phase noise of the system, two identical crystaloscillators are tested against each other. It is quiteimportant to be sure that the phase noise of the referencesource is lower than that of the signal under test.

A sample phase-noise plot for an amateur transceiver isshown in Fig B. It was produced with the test setup shownin Fig A. These plots do not necessarily reflect the phase-noise characteristics of all units of a particular model.

The reference level (the top horizontal line on the scalein the plot) represents 0 dBc/Hz. Because each verticaldivision represents 20 dB, the plot shows the noise levelbetween 0 dBc/Hz (the top horizontal line) and –160 dBc/Hz (the bottom horizontal line). The horizontal scale islogarithmic, with one decade per division (the first divisionshows noise from 100 Hz to 1000 Hz offset, whereas the

Chapter 10.pmd 7/31/2006, 11:34 AM10

Oscillators and Synthesizers 10.11

Fig 10.10 shows this data in graphic form.His oscillator power was only –3 dBm,which limited measurements to offsets lessthan 200 kHz. More power might haveallowed noise measurements to lower lev-els, although receiver overload places alimit on this. This is not important, becausethe real area of interest has been thoroughlycovered. When attempting phase-noisemeasurements at large offsets, rememberthat any front-end selectivity, before thefirst mixer, will limit the maximum offsetat which LO phase-noise measurement ispossible.

MEASURING OSCILLATOR ANDTRANSMITTER PHASE NOISE

Measuring the composite phase noise ofa receiver’s LO requires a clean test oscil-lator. Measuring the phase noise of an in-coming signal, whether from a singleoscillator or an entire transmitter, requiresthe use of a clean receiver, with lowerphase noise than the source under test. Thesidebar, “Transmitter Phase-Noise Mea-surement in the ARRL Lab,” details the

At first, this method—using a low-frequency spectrum analyzer and a low-phase-noise signal source—looks unneces-sarily elaborate. A growing number of radioamateurs have acquired good-quality spec-trum analyzers for their shacks since oldermodel Tektronix and Hewlett-Packard in-struments have started to appear on the sur-plus market at affordable prices. Theobvious question is, “Why not just use oneof these to view the signal and read phase-noise levels directly off the screen?” Re-ciprocal mixing is the problem. Very fewspectrum analyzers have clean enough localoscillators not to completely swamp thenoise being measured. Phase-noise mea-surements involve the measurement of low-level components very close to a largecarrier, and that carrier will mix the noisesidebands of the analyzer’s LO into its IF.Some way of notching out the carrier isneeded, so that the analyzer need onlyhandle the noise sidebands. A crystalfilter could be designed to do the job, butthis would be expensive, and one wouldbe needed for every different oscillator

Fig B — Sample phase noise plot for an amateur HF transceiver.

Fig 10.10—The SSB phase noise of anICOM IC-745 transceiver (serial number01528) as measured by KI6WX.

method used to measure composite noise(phase noise and amplitude noise, thepractical effects of which are indistin-guishable on the air) for QST ProductReviews. Although targeted at measuringhigh power signals from entire transmit-ters, this approach can be used to measurelower-level signals simply by changing theamount of input attenuation used.

last division shows noise from 100 kHz through 1 MHzoffset).

What Do the Phase-Noise Plots Mean?Although they are useful for comparing different radios,

plots can also be used to calculate the amount of interfer-ence you may receive from a nearby transmitter withknown phase-noise characteristics. An approximation isgiven by

AQRM = NL + 10 × log(BW)

whereAQRM = Interfering signal level, dBcNL = noise level on the receive frequency, dBcBW = receiver IF bandwidth, in Hz

For instance, if the noise level is –90 dBc/Hz and you

are using a 2.5-kHz SSB filter, the approximate interferingsignal will be –56 dBc. In other words, if the transmittedsignal is 20 dB over S9, and each S unit is 6 dB, theinterfering signal will be as strong as an S3 signal.

The measurements made in the ARRL Lab apply onlyto transmitted signals. It is reasonable to assume that thephase-noise characteristics of most transceivers aresimilar on transmit and receiver because the sameoscillators are generally used in local-oscillator (LO)chain.

In some cases, the receiver may have better phasenoise characteristics than the transmitter. Why thepossible difference? The most obvious reason is thatcircuits often perform less than optimally in strong RFfields, as anyone who has experienced RFI problems cantell you. A less obvious reason results from the way thatmany high-dynamic-range receivers work. To get gooddynamic range, a sharp crystal filter is often placedimmediately after the first mixer in the receive line. Thisfilter removes all but a small slice of spectrum for furthersignal processing. If the desired filtered signal is aproduct of mixing an incoming signal with a noisy oscilla-tor, signals far away from the desired one can end up inthis slice. Once this slice of spectrum is obtained,however, unwanted signals cannot be reintroduced, nomatter how noisy the oscillators used in further signalprocessing. As a result, some oscillators in receiversdon’t affect phase noise.

The difference between this situation and that intransmitters is that crystal filters are seldom used forreduction of phase noise in transmitting because of thehigh cost involved. Equipment designers have enoughtrouble getting smooth, click-free break-in operation intransceivers without having to worry about switchingcrystal filters in and out of circuits at 40-WPM keyingspeeds! — Zack Lau, W1VT, and Michael Tracy, KC1SX,ARRL Laboratory Engineers

10 2 10 3 10 4 10 5 10 6-160

-140

-120

-100

-80

-60

-40

-20

0

dBc/

Hz

Offset From Carrier (Hz)

Chapter 10.pmd 7/31/2006, 11:34 AM11

10.12 Chapter 10

frequency to be tested. The alternative is tobuild a direct-conversion receiver using aclean LO like the Hewlett-PackardHP8640B signal generator and spectrum-analyze its “audio” output with an audioanalyzer. This scheme mixes the carrier todc; the LF analyzer is then ac-coupled, andthis removes the carrier. The analyzer canbe made very sensitive without overload orreciprocal mixing being a problem. Theremaining problem is then keeping theLO—the HP8640B in this example—atexactly the carrier frequency. 8640s arebased on a shortened-UHF-cavity oscilla-tor and can drift a little. The oscillator undertest will also drift. The task is therefore tomake the 8640B track the oscillator undertest. For once we get something for free:The HP8640B’s FM input is dc coupled,and we can use this as an electronic fine-tuning input. As a further bonus, the8640B’s FM deviation control acts as asensitivity control for this input. We alsoget a phase detector for free, as the mixeroutput’s dc component depends on thephase relationship between the 8640B andthe signal under test (remember to usethe dc coupled port of a diode ring mixeras the output). Taken together, the systemincludes everything needed to create acrude phase-locked loop that will auto-matically track the input signal over a smallfrequency range. Fig 10.11 shows thearrangement.

The oscilloscope is not essential foroperation, but it is needed to adjust thesystem. With the loop unlocked (8640BFM input disconnected), tune the 8640 offthe signal frequency to give a beat atthe mixer output. Adjust the mixer drivelevels to get an undistorted sine wave onthe scope. This ensures that the mixer is

not being overdriven. While the loop isoff-tuned, adjust the beat to a frequencywithin the range of the LF spectrumanalyzer and use it to measure its level,“Ac” in dBm. This represents the carrierlevel and is used as the reference for thenoise measurements. Connect the FMinput of the signal generator, and switchon the generator’s dc FM facility. Try adeviation range of 10 kHz to start with.When you tune the signal generator towardthe input frequency, the scope will showthe falling beat frequency until the loopjumps into lock. Then it will display anoisy dc level. Fine tune to get a meanlevel of 0 V. (This is a very-low-band-width, very-low-gain loop. Stability is nota problem; careful loop design is notneeded. We actually want as slow a loopas possible; otherwise, the loop wouldtrack and cancel the slow components ofthe incoming signal’s phase noise, pre-venting their measurement.)

When you first take phase-noise plots,it’s a good idea to duplicate them at thegenerator’s next lower FM-deviationrange and check for any differences in thenoise level in the areas of interest. Reducethe FM deviation range until you find nofurther improvement. Insufficient FM de-viation range makes the loop’s lock rangenarrow, reducing the amount of drift it cancompensate. (It’s sometimes necessary tokeep gently trimming the generator’s finetune control.)

Set up the LF analyzer to show the noise.A sensitive range and 100-Hz resolutionbandwidth are appropriate. Measure thenoise level, “An” in dBm. We must nowcalculate the noise density that this repre-sents. Spectrum-analyzer filters are nor-mally Gaussian-shaped and bandwidth-

specified at their –3-dB points. To avoidusing integration to find their true-noisepower bandwidth, we can reasonably as-sume a value of 1.2 × BW. A spectrumanalyzer logarithmically compresses its IFsignal ahead of the detectors and averag-ing filter. This affects the statistical distri-bution of noise voltage and causes theanalyzer to read low by 2.5 dB. To pro-duce the same scale as the ARRL Labphotographs prior to May 2006 QST, theanalyzer reference level must be set to–60 dBc/Hz, which can be calculated as:

dBm 62.5BW)(1.2 log 10AA cref +×−= (2)where

Aref = analyzer reference level, dBmAc = carrier amplitude, dBm

This produces a scale of –60 dBc/Hz atthe top of the screen, falling to –140 dBc/Hz at the bottom. The frequency scale is 0to 20 kHz with a resolution bandwidth(BW in the above equation) of 100 Hz.This method combines the power of bothsidebands and so measures DSB phasenoise. To calculate the equivalent SSBphase noise, subtract 3 dB for noncoherentnoise (the general “hash” of phase noise)and subtract 6 dB for coherent, discretecomponents (that is, single-frequencyspurs). This can be done by setting thereference level 3 to 6 dB higher.

Low-Cost Phase Noise TestingAll that expensive equipment may seem

far beyond the means of the average Ama-teur Radio experimenter. With carefulshopping and a little more effort, alterna-tive equipment can be put together forpocket money. (All of the things needed—parts for a VXO, a surplus spectrum ana-lyzer and so on—have been seen on salecheap enough to total less than $100.) TheHP8640B is good and versatile, but for useat one oscillator frequency, you can builda VXO for a few dollars. It will only coverone oscillator frequency, but a VXO canprovide even better phase-noise perfor-mance than the 8640B.

As referenced at the end of this chapter,Pontius has also demonstrated that signal-source phase-noise measurements can beaccurately obtained without the aid of ex-pensive equipment.

OSCILLATOR CIRCUITS ANDCONSTRUCTIONLC Oscillator Circuits

The LC oscillators used in radio equip-ment are usually arranged as variablefrequency oscillators (VFOs). Tuning isachieved by either varying part of thecapacitance of the resonator or, less

Fig 10.11—Arrangement for measuring phase noise by directly converting thesignal under test to audio. The spectrum analyzer views the signal’s noisesidebands as audio; the signal’s carrier, converted to dc, provides a feedbacksignal to phase-lock the Hewlett-Packard HP8640B signal generator to the signalunder test.

Chapter 10.pmd 7/31/2006, 11:34 AM12

Oscillators and Synthesizers 10.13

have other problems though, and the VFOstill has much to offer in terms of signalcleanliness, cost and power consumption,making it attractive for home construction.No one VFO circuit has any overwhelm-ing advantage over any other—component quality, mechanical design andthe care taken in assembly are much moreimportant.

Fig 10.12 shows three popular oscilla-tor circuits stripped of any unessentialfrills so they can be more easily compared.The original Colpitts circuit (Fig 10.12A)is now often referred to as the parallel-tuned Colpitts because its series-tunedderivative (Fig 10.12B) has become com-mon. All three of these circuits use anamplifier with a voltage gain less thanunity, but large current gain. The N-chan-nel JFET source follower shown appearsto be the most popular choice nowadays.In the parallel-tuned Colpitts, C3 and C4are large values, perhaps 10 times largerthan typical values for C1 and C2. Thismeans that only a small fraction of the totaltank voltage is applied to the FET, and theFET can be considered to be only lightlycoupled into the tank. The FET is drivenby the sum of the voltages across C3 andC4, while it drives the voltage across C4alone. This means that the tank operates asa resonant, voltage-step-up transformercompensating for the less-than-unity-voltage-gain amplifier. The resonant cir-cuit consists of L, C1, C2, C3 and C4. Theresonant frequency can be calculated byusing the standard formulas for capacitorsin series and parallel to find the resultantcapacitance effectively connected acrossthe inductor, L, and then use the standardformula for LC resonance:

LC2

1f

π= (3)

wheref = frequency in hertzL = inductance in henriesC = capacitance in farads.

For a wide tuning range, C2 must be keptsmall to reduce the effect of C3 and C4swamping the variable capacitor C1.

The series-tuned Colpitts circuit worksin much the same way. The difference isthat the variable capacitor, C1, is posi-tioned so that it is well-protected frombeing swamped by the large values of C3and C4. In fact, small values of C3, C4would act to limit the tuning range. Fixedcapacitance, C2, is often added across C1to allow the tuning range to be reduced tothat required, without interfering with C3and C4, which set the amplifier coupling.The series-tuned Colpitts has a reputationfor better stability than the parallel-tuned

Fig 10.12—The Colpitts (A), series-tuned Colpitts (B) and Hartley (C) oscillatorcircuits. Rules of thumb: C3 and C4 at A and B should be equal and valued suchthat their Xc = 45 ΩΩΩΩΩ at the operating frequency; for C2 at A, Xc = 100 ΩΩΩΩΩ. For beststability, use C0G or NP0 units for all capacitors associated with the FETs’ gatesand sources. Depending on the FET chosen, the 1-kΩΩΩΩΩ source-bias-resistor valueshown may require adjustment for reliable starting.

commonly, by using a movable magneticcore to vary the inductance. Since the earlydays of radio, there has been a huge questfor the ideal, low-drift VFO. Amateurs andprofessionals have spent immense efforton this pursuit. A brief search of the litera-

ture reveals a large number of designs,many accompanied by claims of high sta-bility. The quest for stability has beensolved by the development of low-costfrequency synthesizers, which can givecrystal-controlled stability. Synthesizers

Chapter 10.pmd 7/31/2006, 11:34 AM13

10.14 Chapter 10

original. Note how C3 and C4 swamp thecapacitances of the amplifier in both ver-sions.

The Hartley is similar to the parallel-tuned Colpitts, but the amplifier source istapped up on the tank inductance insteadof the tank capacitance. A typical tapplacement is 10 to 20% of the total turnsup from the “cold” end of the inductor.(It’s usual to refer to the lowest-signal-voltage end of an inductor as cold and theother, with the highest signal voltage ashot.) C2 limits the tuning range as re-quired; C3 is reduced to the minimumvalue that allows reliable starting. This isnecessary because the Hartley’s lack of theColpitts’s capacitive divider would other-wise couple the FET’s capacitances to thetank more strongly than in the Colpitts,potentially affecting the circuit’s fre-quency stability.

In all three circuits, there is a 1 kΩ resis-tor in series with the source bias choke. Thisresistor does a number of desirable things.It spoils the Q of the inevitable low-fre-quency resonance of the choke with thetank tap circuit. It reduces tuning drift dueto choke impedance and winding capaci-tance variations. It also protects againstspurious oscillation due to internal chokeresonances. Less obviously, it acts to stabi-lize the loop gain of the built-in AGC actionof this oscillator. Stable operating condi-tions act to reduce frequency drift. Somevariations of these circuits may be foundwith added resistors providing a dc bias tostabilize the system quiescent current.More elaborate still are variations charac-terized by a constant-current source pro-viding bias. This can be driven from aseparate AGC detector system to give verytight level control. The gate-to-groundclamping diode (1N914 or similar) longused by radio amateurs as a means of avoid-ing gate-source conduction has been shownby Ulrich Rohde, KA2WEU, to degrade os-cillator phase-noise performance, and itsuse is virtually unknown in professionalcircles.

Fig 10.13 shows some more VFOs toillustrate the use of different devices. Thetriode Hartley shown includes permeabil-ity tuning, which has no sliding contactlike that to a capacitor’s rotor and can bemade reasonably linear by artfully spac-ing the coil turns. The slow-motion drivecan be done with a screw thread. The dis-advantage is that special care is needed toavoid backlash and eccentric rotation ofthe core. If a nonrotating core is used, theslides have to be carefully designed toprevent motion in unwanted directions.The Collins Radio Company made exten-sive use of tube-based permeability tun-ers, and a semiconductor version can still

Fig 10.13—Three more oscillator examples: at A, a triode-tube Hartley; at B,a bipolar junction transistor in a series-tuned Colpitts; at C, a dual-gateMOSFET Hartley.

be found in a number of Ten-Tec radios.Vacuum tubes cannot run as cool as

competitive semiconductor circuits, socare is needed to keep the tank circuitaway from the tube heat. In many amateurand commercial vacuum-tube oscillators,oscillation drives the tube into grid cur-

rent at the positive peaks, causing rectifi-cation and producing a negative grid bias.The oscillator thus runs in Class C, inwhich the conduction angle reduces as thesignal amplitude increases until the am-plitude stabilizes. As in the FET circuitsof Fig 10.12, better stability and phase-

Chapter 10.pmd 7/31/2006, 11:34 AM14

Oscillators and Synthesizers 10.15

noise performance can be achieved in avacuum-tube oscillator by moving it outof true Class C—that is, by including abypassed cathode-bias resistor (the resis-tance appropriate for Class A operation isa good starting value). A small number ofpeople still build vacuum-tube radiospartly to be different, partly for fun, butthe semiconductor long ago achieveddominance in VFOs.

The voltage regulator (VR) tube shownin Fig 10.13A has a potential drawback: Itis a gas-discharge device with a high strik-ing voltage at which it starts to conduct anda lower extinguishing voltage, at which itstops conducting. Between these extremeslies a region in which decreasing voltagetranslates to increasing current, whichimplies negative resistance. When the regu-lator strikes, it discharges any capacitanceconnected across it to the extinguishingvoltage. The capacitance then chargesthrough the source resistor until the tubestrikes again, and the process repeats. Thisrelaxation oscillation demonstrates hownegative resistance can cause oscillation.The oscillator translates the resultantsawtooth modulation of its power supplyinto frequency and amplitude variation.Because of VR tubes’ ability to supportrelaxation oscillation, a traditional rule ofthumb is to keep the capacitance directlyconnected across a VR tube to 0.1 μF orless. A value much lower than this canprovide sufficient bypassing in Fig 10.13Abecause the dropping resistor acts as adecoupler, increasing the bypass’s effec-tiveness.

There is a related effect calledsquegging, which can be loosely definedas oscillation on more than one frequencyat a time, but which may also manifest it-self as RF oscillation interrupted at an AFrate, as in a superregenerative detector. Oneform of squegging occurs when an oscilla-tor is fed from a power supply with a highsource impedance. The power supplycharges up the decoupling capacitor untiloscillation starts. The oscillator draws cur-rent and pulls down the capacitor voltage,starving itself of power until oscillationstops. The oscillator stops drawing current,and the decoupling capacitor then re-charges until oscillation restarts. The pro-cess, the low-frequency cycling of which isa form of relaxation oscillation, repeatsindefinitely. The oscillator output canclearly be seen to be pulse modulated if anoscilloscope is used to view it at a suitabletime-base setting. This fault is a well-known consequence of poor design in bat-tery-powered radios. As dry cells becomeexhausted, their internal resistance risesquickly and circuits they power can beginto misbehave. In audio stages, such misbe-

havior may manifest itself in the putt-puttsound of the slow relaxation oscillationcalled motorboating.

Compared to the frequently used JFET,bipolar transistors, Fig 10.13B, are rela-tively uncommon in oscillators becausetheir low input and output impedances aremore difficult to couple into a high-Q tankwithout excessively loading it. Bipolardevices do tend to give better sample-to-sample amplitude uniformity for a givenoscillator circuit, however, as JFETs of agiven type tend to vary more in their char-acteristics.

The dual-gate MOSFET, Fig 10.13C, isvery rarely seen in VFO circuits. It im-poses the cost of the components neededto bias and suppress VHF parasitic oscil-lation at the second gate, and its inabilityto generate its own AGC through gate-source conduction forces the addition of ablocking capacitor, resistor and diode atthe first gate for amplitude control.

VFO Components andConstructionTuning Capacitors and ReductionDrives

As most commercially made radios nowuse frequency synthesizers, it has becomeincreasingly difficult to find certain keycomponents needed to construct a goodVFO. The slow-motion drives, dials, gear-boxes, associated with names like Millen,National, Eddystone and Jackson are nolonger available. Similarly, the most suit-able silver-plated variable capacitors withball races at both ends, although still made,are not generally marketed and are expen-sive. Three approaches remain: Scavengesuitable parts from old equipment; usetuning diodes instead of variable capaci-tors—an approach that, if uncorrectedthrough phase locking, generally degradesstability and phase-noise performance; oruse two tuning capacitors, one with a ca-pacitance range 1/5 to 1/10

that of the other, in a bandset/bandspreadapproach.

Assembling a variable capacitor to achassis and its reduction drive to a frontpanel can result in backlash—an annoyingtuning effect in which rotating thecapacitor shaft deforms the chassis and/orpanel rather than tuning the capacitor. Oneway of minimizing this is to use the reduc-tion drive to support the capacitor, and usethe capacitor to support the oscillator cir-cuit board.

Fixed CapacitorsTraditionally, silver-mica fixed capaci-

tors have been used extensively in oscilla-tors, but their temperature coefficient is

not as low as can be achieved by othertypes, and some silver micas have beenknown to behave erratically. Polystyrenefilm has become a proven alternative. Onewarning is worth noting: Polystyrene ca-pacitors exhibit a permanent change invalue should they ever be exposed to tem-peratures much over 70°C; they do not re-turn to their old value on cooling.Particularly suitable for oscillator con-struction are the low-temperature-coeffi-cient ceramic capacitors, often describedas NP0 or C0G types. These names are ac-tually temperature-coefficient codes.Some ceramic capacitors are availablewith deliberate, controlled temperature co-efficients so that they can be used to com-pensate for other causes of frequency driftwith temperature. For example, the codeN750 denotes a part with a temperaturecoefficient of –750 parts per million perdegree Celsius. These parts are now some-what difficult to obtain, so other methodsare needed.

In a Colpitts circuit, the two large-valuecapacitors that form the voltage divider forthe active device still need careful selec-tion. It would be tempting to use any avail-able capacitor of the right value, becausethe effect of these components on thetank frequency is reduced by the propor-tions of the capacitance values in the cir-cuit. This reduction is not as great as thedifference between the temperature stabil-ity of an NP0 ceramic part and some of thelow-cost, decoupling-quality X7R-dielec-tric ceramic capacitors. It’s worth usinglow-temperature coefficient parts even inthe seemingly less-critical parts of a VFOcircuit—even as decouplers. Chasing thecause of temperature drift is more challeng-ing than fun. Buy critical components likehigh-stability capacitors from trustworthysources.

InductorsCeramic coil forms can give excellent

results, as can self-supporting air-woundcoils (Miniductor). If you use a magneticcore, make it powdered iron, never ferrite,and support it stably. Stable VFOs havebeen made using toroidal cores, but again,ferrite must be avoided. Micrometals mixnumber 6 has a low temperature coeffi-cient and works well in conjunction withNP0 ceramic capacitors. Coil forms inother materials have to be assessed on anindividual basis.

A material’s temperature stability willnot be apparent until you try it in an oscil-lator, but you can apply a quick test toidentify those nonmetallic materials thatare lossy enough to spoil a coil’s Q. Put asample of the coil-form material into amicrowave oven along with a glass of

Chapter 10.pmd 7/31/2006, 11:34 AM15

10.16 Chapter 10

water and cook it about 10 s on low power.Do not include any metal fittings or ferro-magnetic cores. Good materials will becompletely unaffected; poor ones willheat and may even melt, smoke, or burstinto flame. (This operation is a fire hazardif you try more than a tiny sample ofan unknown material. Observe yourexperiment continuously and do not leaveit unattended.)

Wes Hayward, W7ZOI, suggests anneal-ing toroidal VFO coils after winding. RoyLewallen, W7EL reports achieving suc-cess with this method by boiling his coilsin water and letting them cool in air.

Voltage RegulatorsVFO circuits are often run from locally

regulated power supplies, usually fromresistor/Zener diode combinations. Zenerdiodes have some idiosyncrasies thatcould spoil the oscillator. They are noisy,so decoupling is needed down to audiofrequencies to filter this out. Zener diodesare often run at much less than their speci-fied optimum bias current. Although thissaves power, it results in a lower outputvoltage than intended, and the diode’simpedance is much greater, increasing itssensitivity to variations in input voltage,output current and temperature. Somecommon Zener types may be designed torun at as much as 20 mA; check the datasheet for your diode family to find theoptimum current.

True Zener diodes are low-voltagedevices; above a couple of volts, so-calledZener diodes are actually avalanche types.The temperature coefficient of thesediodes depends on their breakdown volt-age and crosses through zero for diodesrated at about 5 V. If you intend to usenothing fancier than a common-varietyZener, designing the oscillator to run from5 V and using a 5.1-V Zener, will give youa free advantage in voltage-versus-tem-perature stability. There are some diodesavailable with especially low temperaturecoefficients, usually referred to as refer-ence or temperature-compensated diodes.These usually consist of a series pair ofdiodes designed to cancel each other’stemperature drift. Running at 7.5 mA, the1N829A gives 6.2 V ±5% and a tempera-ture coefficient of just ±5 parts per million(ppm) maximum per degree Celsius. Achange in bias current of 10% will shift thevoltage less than 7.5 mV, but this increasesrapidly for greater current variation. The1N821A is a lower-grade version, at ±100ppm/°C. The LM399 is a complex IC thatbehaves like a superb Zener at 6.95 V, ±0.3ppm/°C. There are also precision, low-power, three-terminal regulators designedto be used as voltage references, some of

which can provide enough current to run aVFO. There are comprehensive tables ofall these devices between pages 334 and337 of Horowitz and Hill, The Art of Elec-tronics, 2nd ed.

Oscillator DevicesThe 2N3819 FET, a classic from the

1960s, has proven to work well in VFOs,but, like the MPF102 also long-popularwith ham builders, it’s manufactured towide tolerances. Considering an oscil-lator’s importance in receiver stability,you should not hesitate to spend a bit moreon a better device. The 2N5484, 2N5485and 2N5486 are worth considering; to-gether, their transconductance ranges spanthat of the MPF102, but each is a better-controlled subset of that range. The2N5245 is a more recent device with bet-ter-than-average noise performance thatruns at low currents like the 2N3819. The2N4416/A, also available as the plastic-cased PN4416, is a low-noise device, de-signed for VHF/UHF amplifier use, whichhas been featured in a number of goodoscillators up to the VHF region. Its lowinternal capacitance contributes to lowfrequency drift. The J310 (plastic; themetal-cased U310 is similar) is anotherpopular JFET in oscillators.

The 2N5179 (plastic, PN5179 orMPS5179) is a bipolar transistor capableof good performance in oscillators up tothe top of the VHF region. Care is neededbecause its absolute-maximum collector-emitter voltage is only 12 V, and its col-lector current must not exceed 50 mA.Although these characteristics may seemto convey fragility, the 2N5179 is suffi-cient for circuits powered by stabilized6-V power supplies.

VHF-UHF devices are not really neces-sary in LC VFOs because such circuits arerarely used above 10 MHz. Absolute fre-quency stability is progressively harder toachieve with increasing frequency, sofree-running oscillators are rarely used togenerate VHF-UHF signals for radio com-munication. Instead, VHF-UHF radiosusually use voltage-tuned, phase-lockedoscillators in some form of synthesizer.Bipolar devices like the BFR90 andMRF901, with fTs in the 5-GHz region andmounted in stripline packages, are neededat UHF.

Integrated circuits have not been men-tioned until now because few specificRF-oscillator ICs exist. Some consumerICs—the NE602, for example—includethe active device(s) necessary for RF os-cillators, but often this is no more than asingle transistor fabricated on the chip.This works just as a single discrete devicewould, although using such on-chip de-

vices may result in poor isolation from therest of the circuits on the chip. There is onespecialized LC-oscillator IC: the MotorolaMC1648. This device has been made sincethe early 1970s and is a surviving memberof a long-obsolete family of emitter-coupled-logic (ECL) devices. Despite theMC1648’s antiquity, it has no real compe-tition and is still widely used in currentmilitary and commercial equipment. Mar-ket demand should force its continued pro-duction for some more years to come. Itscircuitry is complex for an oscillator, witha multitransistor oscillator cell controlledby a detector and amplifier in an on-chipALC system. The MC1648’s first problemis that the ECL families use only about a1-V swing between logic levels. Becausethe oscillator is made using the same ECL-optimized semiconductor manufacturingprocesses and circuit design techniques,this same limitation applies to the signal inthe oscillator tuned circuit. It is possible toimprove this situation by using a tapped ortransformer-coupled tank circuit to giveimproved Q, but this risks the occurrenceof the device’s second problem.

Periodically, semiconductor manufac-turers modernize their plants and scrap oldassembly lines used to make old products.Any surviving devices then must undergosome redesign to allow their productionby the new processes. One common resultof this is that devices are shrunk, whenpossible, to fit more onto a wafer. All thisincreases the fT of the transistors in thedevice, and such evolution has renderedtoday’s MC1648s capable of operation atmuch higher frequencies than the speci-fied 200-MHz limit. This allows higher-frequency use, but great care is needed inthe layout of circuits using it to preventspurious oscillation. A number of olddesigns using this part have neededreengineering because the newer partsgenerate spurious oscillations that the oldones didn’t, using PC-board traces as para-sitic tuned circuits.

The moral is that a UHF-capable devicerequires UHF-cognizant design and layouteven if the device will be used at far lowerfrequencies. Fig 10.14 shows the MC1648in a simple circuit and with a tapped reso-nator. These more complex circuits have agreater risk of presenting a stray resonancewithin the device’s operating range, risk-ing oscillation at an unwanted frequency.This device is not a prime choice for an HFVFO because the physical size of the vari-able capacitor and the inevitable leadlengths, combined with the need to tap-couple to get sufficient Q for good noiseperformance, makes spurious oscillationdifficult to avoid. The MC1648 is reallyintended for tuning-diode control in

Chapter 10.pmd 7/31/2006, 11:34 AM16

Oscillators and Synthesizers 10.17

quite tame. It’s still a relatively new part andmay not have been “improved” yet (so far, ithas progressed from the NE602 to theNE602A, the A version affording somewhathigher dynamic range than the originalNE602). It might be a good idea for anyonelaying out a board using one to take a littleextra care to keep PCB traces short in theoscillator section to build in some safety mar-gin so that the board can be used reliably inthe future. Experienced (read: “bitten”) pro-fessional designers know that their designsare going to be built for possibly more than 10years and have learned to make allowancesfor the progressive improvement of semicon-ductor manufacture.

Three High-Performance HF VFOsThe G3PDM Vackar VFO

The Vackar VFO shown in Fig 10.15was developed over 20 years ago by PeterMartin, G3PDM, for the Mark II versionof his high-dynamic-range receiver. Thiscan be found in the Radio Society of GreatBritain’s Radio Communication Hand-book, with some further comments on theoscillator in RSGB’s Amateur Radio Tech-niques. This is a prime example of an os-cillator that has been successfullyoptimized for maximum frequency stabil-ity. Not only does it work extremely well,but it still represents the highest stabilitythat can be achieved. Its developer com-mented on a number of points, which alsoapply to the construction of different VFOcircuits.

Fig 10.14—One of the few ICs ever designed solely for oscillator service, the ECLMotorola MC1648 (A) requires careful design to avoid VHF parasitics whenoperating at HF. Keeping its tank Q high is another challenge; B and C showmeans of coupling the IC’s low-impedance oscillator terminals to the tank bytapping up on the tank coil (B) or with a link (C).

Fig 10.15—G3PDM’s Vackar VFO has proved popular and successful for two decades. The MPF102 can be used as a substi-tute for the 2N3819. Generally, VFOs can be adapted to work at other frequencies (within the limits of the active device). Todo so, compute an adjustment factor: fold / fnew. Multiply the value of each frequency determining or feedback L or C by thefactor. As frequency increases, it may help to increase feedback even more than indicated by the factor.

phase-locked loops operating at VHF. Thisdifficulty is inherent in wideband devices,especially oscillator circuits connected totheir tank by a single “hot” terminal, wherethere is simply no isolation between theamplifier’s input and output paths. Any

resonance in the associated circuitry cancontrol the frequency of oscillation.

The popular NE602 mixer IC has a built-inoscillator and can be found in many publishedcircuits. This device has separate input andoutput pins to the tank and has proved to be

Chapter 10.pmd 7/31/2006, 11:34 AM17

10.18 Chapter 10

• Use a genuine Vackar circuit, withC1/(C4+C6) and C3/C2 = 6.

• Use a strong box, die-cast or milled fromsolid metal.

• Use a high-quality variable capacitor(double ball bearings, silver plated).

• Adjust feedback control C2, an air-dielectric trimmer, so the circuit justoscillates.

• Thoroughly clean all variable capaci-tors in an ultrasonic bath.

• Use an Oxley “Tempatrimmer,” a fixedcapacitor whose temperature coeffi-cient is variable over a wide range, foradjustable temperature compensation.(The “Thermatrimmer” is a lower cost,smaller range alternative.)

• C1, C3 and C6 are silver-mica typesglued to a solid support to reduce sensi-tivity to mechanical shock.

• The gate resistor is a 4.7-kΩ, 2-W car-bon-composition type to minimize self-heating.

• The buffer amplifier is essential.• Circuits using an added gate-ground

diode seem to suffer increased drift.• Its power supply should be well-regu-

lated. Make liberal use of decouplingcapacitors to prevent unintentionalfeedback via supply rails.

• Single-point grounding for the tank andFET is important. This usually meansusing one mounting screw of the tuningcapacitor.

• The inductor used a ceramic form witha powdered-iron core mounted on aspring-steel screw.

• Short leads and thick solid wire (#16 to#18 gauge) are essential for mechanicalstability.

Operating in the 6-MHz region, thisoscillator drifted 500 Hz during the firstminute as it warmed up and then drifted at2 Hz in 30 minutes. It must be stressed thatsuch performance does not indicate a won-derful circuit so much as care in construc-tion, skillful choice of components, artfulmechanical layout and diligence in adjust-ment. The 2-W gate resistor may seemstrange, but 20 years ago many compo-nents were not as good as they are today.A modern, low-inductance component oflow temperature coefficient and moremodest power rating should be fine.

The buffer amplifier is an important partof a good oscillator system as it serves toprevent the oscillator seeing any imped-ance changes in the circuit being driven.This could otherwise affect the frequency.

The Oxley Tempatrimmer used byG3PDM was a rare and expensive compo-nent, used commercially only in very high-quality equipment, and it may no longer bemade. An alternative temperature com-pensator can be constructed from currently

available components, as will be describedshortly. G3PDM also referred to his “stan-dard mallet test,” where a thump with awooden hammer produced an average shiftof 6 Hz, as an illustration of the benefits ofsolid mechanical design.

The K7HFD Low-Noise OscillatorThe other high performance oscillator

example, shown in Fig 10.16, was de-signed for low-noise performance byLinley Gumm, K7HFD, and appears onpage 126 of ARRL’s Solid State Designfor the Radio Amateur. Despite its publi-cation in the homebrewer’s bible, this cir-cuit seems to have been overlooked bymany builders. It uses no unusual compo-nents and looks simple, yet it is a subtleand sophisticated circuit. It represents theantithesis of G3PDM’s VFO: In the pur-suit of low noise sidebands, a number ofdesign choices have been made that willdegrade the stability of frequency overtemperature.

The effects of oscillator noise have al-ready been covered, and Fig 10.7 shows theeffect of limiting on the signal from a noisyoscillator. Because AM noise sidebandscan get translated into PM noise sidebandsby imperfect limiting, there is an advantageto stripping off the AM as early as possible,in the oscillator itself. An ALC system inthe oscillator will counteract and cancel

only the AM components within its band-width, but an oscillator based on a limiterwill do this over a broad bandwidth.K7HFD’s oscillator uses a differential pairof bipolar transistors as a limiting ampli-fier. The dc bias voltage at the bases and theresistor in the common emitter path toground establishes a controlled dc bias cur-rent. The ac voltage between the basesswitches this current between the two col-lectors. This applies a rectangular pulse ofcurrent into link winding L2, which drivesthe tank. The output impedance of the col-lector is high in both the current on andcurrent off states. Allied with the smallnumber of turns of the link winding, thispresents a very high impedance to the tankcircuit, which minimizes the damping ofthe tank Q. The input impedance of the lim-iter is also quite high and is applied acrossonly a one-turn tap of L1, which similarlyminimizes the effect on the tank Q. Theinput transistor base is driven into conduc-tion only on one peak of the tank wave-form. The output transformer has theinverse of the current pulse applied to it, sothe output is not a low distortion sine wave,although the output harmonics will not beas extensive as simple theory would sug-gest because the circuit’s high output im-pedance allows stray capacitances toattenuate high-frequency components. Thelow-frequency transistors used also act to

Fig 10.16—This low-noise oscillator design by K7HFD operates at an unusuallyhigh power level to achieve a high C/N (carrier-to-noise) ratio. Need other frequen-cies? See Fig 10.15 (caption) for a frequency-scaling technique.

Chapter 10.pmd 7/31/2006, 11:34 AM18

Oscillators and Synthesizers 10.19

reduce the harmonic power.With an output of +17 dBm, this is a

power oscillator, running with a large dcinput power, so appreciable heating resultsthat can cause temperature-induced drift.The circuit’s high-power operation is adeliberate ploy to create a high signal-to-noise ratio by having as high a signal poweras possible. This also reduces the problemof the oscillator’s broadband noise output.The limitation on the signal level in the tankis the transistors’ base-emitter-junctionbreakdown voltage. The circuit runs with afew volts peak-to-peak across the one-turntap, so the full tank is running at over 50 VP-P. The single easiest way to damage abipolar transistor is to reverse bias the base-emitter junction until it avalanches. Mostdevices are only rated to withstand 5 Vapplied this way, the current needed to dodamage is small, and very little power isneeded. If the avalanche current is limitedto less than that needed to perform immedi-ate destruction of the transistor, it is likelythat there will be some degradation of thedevice, a reduction in its bandwidth andgain along with an increase in its noise.These changes are irreversible and cumu-lative. Small, fast signal diodes havebreakdown voltages of over 30 V and lesscapacitance than the transistor bases, so onepossible experiment would be to try theeffect of adding a diode in series with thebase of each transistor and running the cir-cuit at even higher levels.

The amplitude must be controlled by thedrive current limit. The voltage on L2 mustnever allow the collector of the transistordriving it to go into saturation, if this hap-pens, the transistor presents a very lowimpedance to L2 and badly loads the tank,wrecking the Q and the noise performance.The circuit can be checked to verify themargin from saturation by probing the hotend of L2 and the emitter with an oscillo-scope. Another, less obvious, test is to varythe power-supply voltage and monitor theoutput power. While the circuit is undercurrent control, there is very little changein output power, but if the supply is lowenough to allow saturation, the outputpower will change significantly.

The use of the 2N3904 is interesting, asit is not normally associated with RF oscil-lators. It is a cheap, plain, general-purposetype more often used at dc or audio fre-quencies. There is evidence that suggestssome transistors that have good noise per-formance at RF have worse noise perfor-mance at low frequencies, and that thelow-frequency noise they create can modu-late an oscillator, creating noise sidebands.Experiments with low-noise audio transis-tors may be worthwhile, but many suchdevices have very low fT and high junction

capacitances. In the description of this cir-cuit in Solid State Design for the RadioAmateur, the results of a phase-noise testmade using a spectrum analyzer with acrystal filter as a preselector are given. Tenkilohertz out from the carrier, in a 3-kHzmeasurement bandwidth, the noise wasover 120 dB below the carrier level. Thistranslates into better than –120 – 10 log(3000), which equals –154.8 dBc/Hz, SSB.At this offset, –140 dBc is usually consid-ered to be excellent. This is state-of-the artperformance by today’s standards—in a1977 publication.

A JFET Hartley VFOFig 10.17 shows an 11.1-MHz version

of a VFO and buffer closely patterned afterthat used in 7-MHz transceiver designspublished by Roger Hayward, KA7EXM,and Wes Hayward, W7ZOI (The UglyWeekender) and Roy Lewallen, W7EL (theOptimized QRP Transceiver). In it, a2N5486 JFET Hartley oscillator drives thetwo-2N3904 buffer attributed to Lewallen.This version diverges from the originals inthat its JFET uses source bias (the bypassed910-Ω resistor) instead of a gate-clampingdiode and is powered from a low-current7-V regulator IC instead of a Zener diodeand dropping resistor. The 5-dB pad setsthe buffer’s output to a level appropriatefor “Level 7” (+7-dBm-LO) diode ringmixers.

The circuit shown was originally builtwith a gate-clamping diode, no source biasand a 3-dB output pad. Rebiasing theoscillator as shown increased its output by2 dB without degrading its frequency sta-bility (200 to 300-Hz drift at power up,stability within ±20 Hz thereafter at aconstant room temperature).

Temperature CompensationThe general principle for creating a

high-stability VFO is to use componentswith minimal temperature coefficients incircuits that are as insensitive as possibleto changes in components’ secondarycharacteristics. Even after careful mini-mization of the causes of temperature sen-sitivity, further improvement can still bedesirable. The traditional method was tosplit one of the capacitors in the tank sothat it could be apportioned between low-temperature-coefficient parts and partswith deliberate temperature dependency.Only a limited number of different, con-trolled temperature coefficients are avail-able, so the proportioning between lowcoefficient and controlled coefficient partswas varied to “dilute” the temperature sen-sitivity of a part more sensitive thanneeded. This was a tedious process, in-volving much trial and error, an undertak-

ing made more complicated by the diffi-culty of arranging means of heating andcooling the unit being compensated. (Hay-ward described such a means in December1993 QST.) As commercial and militaryequipment have been based on frequencysynthesizers for some time, supplies ofcapacitors with controlled temperaturesensitivity are drying up. An alternativeapproach is needed.

A temperature-compensated crystaloscillator (TCXO) is an improved-stabilityversion of a crystal oscillator that is usedwidely in industry. Instead of using con-trolled-temperature coefficient capacitors,most TCXOs use a network of thermistorsand normal resistors to control the bias of atuning diode. See Fig 10.18. Manufactur-ers measure the temperature vs frequencycharacteristic of sample oscillators, and usea computer program to calculate the opti-mum normal resistor values for production.This can reliably achieve at least a tenfoldimprovement in stability. We are not inter-ested in mass manufacture, but the idea ofa thermistor tuning a varactor is worthstealing. The parts involved are likely to beavailable for a long time.

Browsing through component suppliers’catalogs shows ready availability of 4.5- to5-kΩ-bead thermistors intended for tem-perature-compensation purposes, at lessthan a dollar each. Fig 10.19 shows a cir-cuit based on this form of temperature com-pensation. Fig 10.20 illustrates a practicalVCO using a tuning diode. Commonlyavailable thermistors have negative tem-perature coefficients, so as temperaturerises, the voltage at the counterclockwise(ccw) end of R8 increases, while that at theclockwise (cw) end drops. Somewherenear the center, there is no change. Increas-ing the voltage on the tuning diode de-creases its capacitance, so settings towardR8’s ccw end simulate a negative-tem-perature-coefficient capacitor; toward itsclockwise end, a positive-temperature-coefficient part. Choose R1 to pass 8.5 mAfrom whatever supply voltage is availableto the 6.2-V reference diode, D1. The1N821A/1N829A-family diode used has avery low temperature coefficient and needs7.5 mA bias for best performance; thebridge takes 1 mA. R7 and R8 shouldbe good-quality multiturn trimmers. D2and C1 are best determined by trial anderror. Practical components aren’t knownwell enough to rely on analytical models.Choose the least capacitance that providesenough compensation range. This reducesthe noise added to the oscillator. (It ispossible, though tedious, to solve forthe differential varactor voltage with re-spect to R2 and R5, via differential cal-culus and circuit theory. The equations

Chapter 10.pmd 7/31/2006, 11:34 AM19

10.20 Chapter 10

in Hayward’s 1993 article can then bemodified to accommodate the additionalcapacitors formed by D2 and C1.) Use asingle ground point near D2 to reduce theinfluence of ground currents from othercircuits. Use good-quality metal-film com-ponents for the circuit’s fixed resistors.

The circuit requires two adjustments,one at each of two different temperatures,and achieving them requires a stable fre-quency counter that can be kept far enoughfrom the radio so that the radio, not thecounter, is subjected to the temperatureextremes. (Using a receiver to listen to theoscillator under test can speed the adjust-ments.) After connecting the counter to theoscillator to be corrected, run the radiocontaining the oscillator and compensatorin a room-temperature, draft-free environ-

Fig 10.17—Incorporating ideas from KA2WEU, KA7EXM, W7ZOI and W7EL, the oscillator at A achieves excellent stability andoutput at 11.1 MHz without the use of a gate-clamping diode, as well as end-running the shrinking availability of reductiondrives through the use of bandset and bandspread capacitors. L1 consists of 10 turns of B & W #3041 Miniductor (#22 tinnedwire, 5/8 inch in diameter, 24 turns per inch). The source tap is 21/2 turns above ground; the tuning-capacitor taps are posi-tioned as necessary for bandset and bandspread ranges required. T1’s primary consists of 15 turns of #28 enameled wire onan FT-37-72 ferrite core; its secondary, 3 turns over the primary. B shows a system for adding fixed TR offset that can beapplied to any LC oscillator. The RF choke consists of 20 turns of #26 enameled wire on an FT-37-43 core. Need otherfrequencies? See Fig 10.15 (caption) for a frequency-scaling technique.

Fig 10.18—A modern voltage-controlled oscillator (VCO) uses the voltage-variablecharacteristic of a diode PN junction for tuning.

Chapter 10.pmd 7/31/2006, 11:34 AM20

Oscillators and Synthesizers 10.21

power, so it’s feasible, even advisable, tomount all of the oscillator components inan unventilated box. In the real world, tem-peratures change, and if the componentsbeing compensated and the componentsdoing the compensating have different ther-mal time constants, a change in tempera-ture can cause a temporary change infrequency until the slower componentshave caught up. One cure for this is to buildthe oscillator in a thick-walled metal boxthat’s slow to heat or cool, and so domi-nates and reduces the possible rate ofchange of temperature of the circuits in-side. This is sometimes called a cold oven.

Shielding and IsolationOscillators contain inductors running at

moderate power levels and so can radiatestrong enough signals to cause interfer-ence with other parts of a radio, or withother radios. Oscillators are also sensitiveto radiated signals. Effective shielding istherefore important. A VFO used to drivea power amplifier and antenna (to form asimple CW transmitter) can prove surpris-ingly difficult to shield well enough. Anyleakage of the power amplifier’s high-level signal back into the oscillator canaffect its frequency, resulting in a poortransmitted note. If the radio gear is in thestation antenna’s near field, sufficientshielding may be even more difficult. Thefollowing rules of thumb continue to serveham builders well:

• Use a complete metal box, with as fewholes as possible drilled in it, with goodcontact around surface(s) where itslid(s) fit(s) on.

• Use feedthrough capacitors on powerand control lines that pass in and out ofthe VFO enclosure, and on the transmit-ter or transceiver enclosure as well.

• Use buffer amplifier circuitry that am-plifies the signal by the desired amountand provide sufficient attenuation ofsignal energy flowing in the reverse di-rection. This is known as reverse isola-tion and is a frequently overlookedloophole in shielding. Figs 10.15 and10.17 include buffer circuitry of provenperformance. As another (and higher-cost) option, consider using a high-speed buffer-amplifier IC (such as theLM6321N by National Semiconductor,a part that combines the high input im-pedance of an op amp with the ability todrive 50-Ω loads directly up into theVHF range).

• Use a mixing-based frequency-genera-tion scheme instead of one that operatesstraight through or by means of multi-plication. Such a system’s oscillatorstages can operate on frequencies withno direct frequency relationship to its

Fig 10.20—A practical VCO. The tuning diodes are halves of a BB204 dual,common-cathode tuning diode (capacitance per section at 3 V, 39 pF) or equiva-lent. The ECG617, NTE617 and MV104 are suitable dual-diode substitutes, or usepairs of 1N5451s (39 pF at 4 V) or MV2109s (33 pF at 4 V).

Fig 10.19—Oscillator temperature compensation has become more difficultbecause of the scarcity of negative-temperature-coefficient capacitors. Thiscircuit, by GM4ZNX, uses a bridge containing two identical thermistors to steer atuning diode for drift correction. The 6.2-V Zener diode used (a 1N821A or1N829A) is a temperature-compensated part; just any 6.2-V Zener will not do.

ment until the oscillator’s frequencyreaches its stable operating temperaturerise over ambient. Lock its tuning, if pos-sible. Adjust R7 to balance the bridge. Thiscauses a drop of 0 V across R8, a conditionyou can reach by winding R8 back andforth across its range while slowly adjust-ing R7. When the bridge is balanced and0 V appears across R8, adjusting R8 causesno frequency shift. When you’ve foundthis R7 setting, leave it there, set R8 is tothe exact center of its range and record theoscillator frequency.

Run the radio in a hot environment and

allow its frequency to stabilize. Adjust R8to restore the frequency to the recordedvalue. The sensitivity of the oscillator totemperature should now be significantlyreduced between the temperatures atwhich you performed the adjustments.You will also have somewhat improvedthe oscillator’s stability outside this range.

For best results with any temperature-compensation scheme, it’s important togroup all the oscillator and compensatorcomponents in the same box, avoiding dif-ferences in airflow over components. Agood oscillator should not dissipate much

Chapter 10.pmd 7/31/2006, 11:34 AM21

10.22 Chapter 10

output frequency.• Use the time-tested technique of run-

ning your VFO at a subharmonic of theoutput signal desired—say, 3.5 MHz ina 7-MHz transmitter—and multiply itsoutput frequency in a suitably nonlinearstage for further amplification at thedesired frequency.

Quartz Crystals in OscillatorsBecause crystals afford Q values and

frequency stabilities that are orders ofmagnitude better than those achievablewith LC circuits, fixed-frequency oscilla-tors usually use quartz-crystal resonators.Master references for frequency countersand synthesizers are always based on crys-tal oscillators.

So glowing is the executive summary ofthe crystal’s reputation for stability thatnewcomers to radio experimentation natu-rally believe that the presence of a crystal inan oscillator will force oscillation at thefrequency stamped on the can. This impres-sion is usually revised after the first fewexperiences to the contrary! There is nosure-fire crystal oscillator circuit (althoughsome are better than others); reading andexperience soon provide a learner withplenty of anecdotes to the effect that:

• Some circuits have a reputation of be-ing temperamental, even to the point ofnot always starting.

• Crystals sometimes mysteriously oscil-late on unexpected frequencies.

Even crystal manufacturers have theseproblems, so don’t be discouraged frombuilding crystal oscillators. The occa-sional uncooperative oscillator is a nui-sance, not a disaster, and it just needs alittle individual attention. Knowing how acrystal behaves is the key to a cure.

Quartz and the Piezoelectric EffectQuartz is a crystalline material with a

regular atomic structure that can be dis-torted by the simple application of force.Remove the force, and the distorted struc-ture springs back to its original form withvery little energy loss. This property allowsacoustic waves—sound—to propagate rap-idly through quartz with very little attenu-ation, because the velocity of an acousticwave depends on the elasticity and density(mass/volume) of the medium throughwhich the wave travels.

If you heat a material, it expands. Heat-ing may cause other characteristics of amaterial to change—such as elasticity,which affects the speed of sound in thematerial. In quartz, however, expansionand change in the speed of sound are verysmall and tend to cancel, which means thatthe transit time for sound to pass through

a piece of quartz is very stable.The third property of this wonder mate-

rial is that it is piezoelectric. Apply anelectric field to a piece of quartz and thecrystal lattice distorts just as if a force hadbeen applied. The electric field applies aforce to electrical charges locked in thelattice structure. These charges are cap-tive and cannot move around in the latticeas they can in a semiconductor, for quartzis an insulator. A capacitor’s dielectricstores energy by creating physical distor-tion on an atomic or molecular scale. In apiezoelectric crystal’s lattice, the distor-tion affects the entire structure. In somepiezoelectric materials, this effect is suf-ficiently pronounced that special shapescan be made that bend visibly when a fieldis applied.

Consider a rod made of quartz. Anysound wave propagating along it eventu-ally hits an end, where there is a large andabrupt change in acoustic impedance. Justas when an RF wave hits the end of anunterminated transmission line, a strongreflection occurs. The rod’s other endsimilarly reflects the wave. At some fre-quency, the phase shift of a round trip willbe such that waves from successive roundtrips exactly coincide in phase and rein-force each other, dramatically increasingthe wave’s amplitude. This is resonance.

The passage of waves in opposite direc-tions forms a standing wave with antin-odes at the rod ends. Here we encounter aseeming ambiguity: not just one, but a fam-ily of different frequencies, causes stand-ing waves—a family fitting the pattern of1/2, 2/3, 5/2, 7/2 and so on, wavelengths intothe length of the rod. And this is the case:A quartz rod can resonate at anyand all of these frequencies.

The lowest of these frequencies, wherethe crystal is 1/2 wavelength long, is calledthe fundamental mode. The others arenamed the third, fifth, seventh and so on,overtones. There is a small phase-shift er-ror during reflection at the ends, whichcauses the frequencies of the overtonemodes to differ slightly from odd integermultiplies of the fundamental. Thus, acrystal’s third overtone is very close to,but not exactly, three times, its fundamen-tal frequency. Many people are confusedby overtones and harmonics. Harmonicsare additional signals at exact integermultiples of the fundamental frequency.Overtones are not signals at all; they areadditional resonances that can be exploitedif a circuit is configured to excite them.

The crystals we use most often resonatein the 1- to 30-MHz region and are of theAT cut, thickness shear type, althoughthese last two characteristics are rarelymentioned. A 15-MHz-fundamental crys-

tal of this type is about 0.15 mm thick.Because of the widespread use of repro-cessed war-surplus, pressure-mounted FT-243 crystals, you may think of crystals assmall rectangles on the order of a half inchin size. The crystals we commonly usetoday are discs, etched and/or doped totheir final dimensions, with metal elec-trodes deposited directly on the quartz. Acrystal’s diameter does not directly affectits frequency; diameters of 8 to15 mm aretypical.

AT cut is one of a number of possiblestandard designations for the orientationat which a crystal disc is sawed from theoriginal quartz crystal. The crystal latticeatomic structure is asymmetric, and theorientation of this with respect to the facesof the disc influences the crystal’s perfor-mance. Thickness shear is one of a numberof possible orientations of the crystal’smechanical vibration with respect to thedisc. In this case, the crystal vibratesperpendicularly to its thickness. This is noteasy to visualize, and diagrams don’t helpmuch, but Fig 10.21 is an attempt at illus-trating this. Place a moist bathroomsponge between the palms of your hands,move one hand up and down, and you’llsee thickness shear in action.

There is a limit to how thin a disc can bemade, given requirements of accuracy andprice. Traditionally, fundamental-modecrystals have been made up to 20 MHz,

Fig 10.21—Thickness-shear vibration ata crystal’s fundamental and thirdovertone (A); B shows how the moderncrystals commonly used by radioamateurs consist of etched quartzdiscs with electrodes deposited directlyon the crystal surface.

Chapter 10.pmd 7/31/2006, 11:34 AM22

Oscillators and Synthesizers 10.23

although 30 MHz is now common at amoderately raised price. Using techniquespioneered in the semiconductor industry,crystals have been made with a centralregion etched down to a thin membrane,surrounded by a thick ring for robustness.This approach can push fundamental reso-nances to over 100 MHz, but these aremore lab curiosities than parts for every-day use. The easy solution for higher fre-quencies is to use a nice, manufacturablythick crystal on an overtone mode. Allcrystals have all modes, so if you order a28.060-MHz, third-overtone unit for alittle QRP transmitter, you’ll get a crystalwith a fundamental resonance somewherenear 9.353333 MHz, but its manufacturerwill have adjusted the thickness to plantthe third overtone exactly on the orderedfrequency. An accomplished manufac-turer can do tricks with the flatness of thedisc faces to make the wanted overtonemode a little more active and the othermodes a little less active. (As some build-ers discover, however, this does not guar-antee that the wanted mode is the mostactive!)

Quartz’s piezoelectric property pro-vides a simple way of driving the crystalelectrically. Early crystals were placedbetween a pair of electrodes in a case. Thisgave amateurs the opportunity to buy sur-plus crystals, open them and grind them alittle to reduce their thickness, thus mov-ing them to higher frequencies. The fre-quency could be reduced very slightly byloading the face with extra mass, such asby blackening it with a soft pencil. Mod-ern crystals have metal electrodes depos-ited directly onto their surfaces (Fig10.21B), and such tricks no longer work.

The piezoelectric effect works bothways. Deformation of the crystal producesvoltage across its electrodes, so the me-chanical energy in the resonating crystalcan also be extracted electrically by thesame electrodes. Seen electrically, at theelectrodes, the mechanical resonanceslook like electrical resonances. Their Q isvery high. A Q of 10,000 would character-ize a poor crystal nowadays; 100,000 isoften reached by high-quality parts. Forcomparison, a Q of over 200 for an LCtank is considered good.

AccuracyA crystal’s frequency accuracy is as out-

standing as its Q. Several factors determinea crystal’s frequency accuracy. First, themanufacturer makes parts with certain tol-erances: ±200 ppm for a low-quality crys-tal for use as in a microprocessor clockoscillator, ±10 ppm for a good-quality partfor professional radio use. Anything muchbetter than this starts to get expensive! A

crystal’s resonant frequency is influencedby the impedance presented to its termi-nals, and manufacturers assume that once acrystal is brought within several parts permillion of the nominal frequency, its userwill perform fine adjustments electrically.

Second, a crystal ages after manufacture.Aging could give increasing or decreasingfrequency; whichever, a given crystal usu-ally keeps aging in the same direction.Aging is rapid at first and then slows down.Aging is influenced by the care in polishingthe surface of the crystal (time and money)and by its holder style. The cheapest holderis a soldered-together, two-part metal canwith glass bead insulation for the connec-tion pins. Soldering debris lands on thecrystal and affects its frequency. Alterna-tively, a two-part metal can be made withflanges that are pressed together until theyfuse, a process called cold-welding. This ismuch cleaner and improves aging ratesroughly fivefold compared to solderedcans. An all-glass case can be made in twoparts and fused together by heating in avacuum. The vacuum raises the Q, and thecleanliness results in aging that’s roughlyten times slower than that achievable witha soldered can. The best crystal holdersborrow from vacuum-tube assembly pro-cesses and have a getter, a highly reactivechemical substance that traps remaininggas molecules, but such crystals are usedonly for special purposes.

Third, temperature influences a crystal.A reasonable, professional quality partmight be specified to shift not more than±10 ppm over 0 to 70°C. An AT-cut crystalhas an S-shaped frequency-versus-tem-perature characteristic, which can be var-ied by slightly changing the crystal cut’sorientation. Fig 10.22 shows the generalshape and the effect of changing the cutangle by only a few seconds of arc. Noticehow all the curves converge at 25°C. Thisis because this temperature is normallychosen as the reference for specifying acrystal. The temperature stability specifi-

cation sets how accurate the manufacturermust make the cut. Better stability may beneeded for a crystal used as a receiver fre-quency standard, frequency counter clockand so on. A crystal’s temperature charac-teristic shows a little hysteresis. In otherwords, there’s a bit of offset to the curvedepending on whether temperature isincreasing or decreasing. This is usuallyof no consequence except in the highest-precision circuits.

It is the temperature of the quartz that isimportant, and as the usual holders forcrystals all give effective thermal insula-tion, only a couple of milliwatts dissipa-tion by the crystal itself can be toleratedbefore self-heating becomes troublesome.Because such heating occurs in the quartzitself and does not come from the sur-rounding environment, it defeats theeffects of temperature compensators andovens.

The techniques shown earlier for VFOfor temperature compensation can alsobe applied to crystal oscillators. An after-compensation drift of 1 ppm is routine and0.5 ppm is good. The result is a temper-ature-compensated crystal oscillator(TCXO). Recently, oscillators have ap-peared with built-in digital thermometers,microprocessors and ROM look-up tablescustomized on a unit-by-unit basis to con-trol a tuning diode via a digital-to-analogconverter (DAC) for temperature compen-sation. These digitally temperature-com-pensated oscillators (DTCXOs) can reach0.1 ppm over the temperature range. Withautomated production and adjustment,they promise to become the cheapest wayto achieve this level of stability.

Oscillators have long been placed intemperature-controlled ovens, which aretypically held at 80°C. Stability of severalparts per billion can be achieved over tem-perature, but this is a limited benefit asaging can easily dominate the accuracy.These are usually called oven-controlledcrystal oscillators (OCXOs).

Fourth, the crystal is influenced by theimpedance presented to it by the circuit inwhich it is used. This means that care isneeded to make the rest of an oscillatorcircuit stable, in terms of impedance andphase shift.

Gravity can slightly affect crystal reso-nance. Turning an oscillator upside downusually produces a small frequency shift,usually much less than 1 ppm; turning theoscillator back over reverses this. Thiseffect is quantified for the highest-qualityreference oscillators.

The Equivalent Circuit of a CrystalBecause a crystal is a passive, two-ter-

minal device, its electrical appearance is

Fig 10.22—Slight changes in a crystalcut’s orientation shift its frequency-versus-temperature curve.

Chapter 10.pmd 7/31/2006, 11:34 AM23

10.24 Chapter 10

that of an impedance that varies withfrequency. Fig 10.23A shows a very sim-plified sketch of the magnitude (phase isignored) of the impedance of a quartz crys-tal. The general trend of dropping imped-ance with increasing frequency impliescapacitance across the crystal. The sharpfall to a low value resembles a series-tunedtank, and the sharp peak resembles a paral-lel-tuned tank. These are referred to as se-ries and parallel resonances. Fig 10.23Bshows a simple circuit that will produce thisimpedance characteristic. The impedancelooks purely resistive at the exact centers ofboth resonances, and the region betweenthem has impedance increasing with fre-quency, which looks inductive.

C1 (sometimes called motional capa-citance, Cm, to distinguish it from thelumped capacitance it approximates) andL1 (motional inductance, Lm) create theseries resonance, and as C0 and R1 areboth fairly small, the impedance at thebottom of the dip is very close to R1. Atparallel resonance, L1 is resonating withC1 and C0 in series, hence the higher fre-quency. The impedance of the parallel tankis immense, the terminals are connected toa capacitive tap, which causes them to seeonly a small fraction of this, which is stilla very large impedance. The overtonesshould not be neglected, soFigs 10.23C and 10.23D include them.Each overtone has series and parallel reso-nances and so appears as a series tank inthe equivalent circuit. C0 again providesthe shifted parallel resonance.

This is still simplified, because real-lifecrystals have a number of spurious, un-wanted modes that add yet more reso-nances, as shown in Fig 10.23E. These arenot well controlled and may vary a lot even

Table 10.2Typical Equivalent Circuit Values for a Variety of Crystals

Crystal Type Series L Series C Series R Shunt C(pF) (Ω) (pF)

1-MHz fundamental 3.5 H 0.007 340 3.010-MHz fundamental 9.8 mH 0.026 7 6.330-MHz third overtone 14.9 mH 0.0018 27 6.2100-MHz fifth overtone 4.28 mH 0.0006 45 7.0

Fig 10.23—Exploring a crystal’simpedance (A) and equivalent circuit(B) through simplified diagrams. C andD extend the investigation to includeovertones; E, to spurious responsesnot easily predictable by theory orcontrollable through manufacture. Acrystal may oscillate on any of itsresonances under the right conditions.

between crystals made to the same speci-fication. Crystal manufacturers work hardto suppress these spurs and have evolved anumber of recipes for shaping crystals tominimize them. Just where they switchfrom one design to another varies frommanufacturer to manufacturer.

Always remember that the equivalentcircuit is just a representation of crystalbehavior and does not represent circuitcomponents actually present. Its only useis as an aid in designing and analyzingcircuits using crystals. Table 10.2 liststypical equivalent-circuit values for a va-riety of crystals. It is impossible to build acircuit with 0.026 to 0.0006-pF capacitors;such values would simply be swamped bystrays. Similarly, the inductor must have aQ orders of magnitude better than is prac-tically achievable, and impossibly lowstray C in its winding.

The values given in Table 10.2 are noth-ing more than rough guides. A crystal’sfrequency is tightly specified, but this stillallows inductance to be traded for capaci-tance. A good manufacturer could holdthese characteristics within a ±25% bandor could vary them over a 5:1 range byspecial design. Similarly marked partsfrom different sources vary widely inmotional inductance and capacitance.

Quartz is not the only material thatbehaves in this way, but it is the best. Reso-nators can be made out of lithium tantalateand a group of similar materials that havelower Q, allowing them to be pulled overa larger frequency range in VCXOs. Muchmore common, however, are ceramic reso-nators based on the technology of the well-known ceramic IF filters. These havemuch lower Q than quartz and much poorerfrequency precision. They serve mainly asclock resonators for cheap microproces-sor systems in which every last cent mustbe saved. A ceramic resonator could beused as the basis of a wide range, cheapVXO, but its frequency stability would notbe as good as a good LC VFO.

Crystal Oscillator CircuitsCrystal oscillator circuits are usually

categorized as series- or parallel-modetypes, depending on whether the crystal’s

Chapter 10.pmd 7/31/2006, 11:34 AM24

Oscillators and Synthesizers 10.25

low- or high-impedance resonance comesinto play at the operating frequency. Theseries mode is now the most common;parallel-mode operation was more oftenused with vacuum tubes. Fig 10.24 showsa basic series-mode oscillator. Somepeople would say that it is an overtonecircuit, used to run a crystal on one of itsovertones, but this is not necessarily true.The tank (L-C1-C2) tunes the collector ofthe common-base amplifier. C1 is largerthan C2, so the tank is tapped in a way thattransforms to a lower impedance, decreas-ing signal voltage, but increasing current.The current is fed back into the emitter viathe crystal. The common-base stage pro-vides a current gain of less than unity, sothe transformer in the form of the tappedtank is essential to give loop gain. Thereare two tuned circuits, the obvious collec-tor tank and the series-mode one “in” thecrystal. The tank kills the amplifier’s gainaway from its tuned frequency, and thecrystal will only pass current at the seriesresonant frequencies of its many modes.The tank resonance is much broader thanany of the crystal’s modes, so it can bethought of as the crystal setting the fre-quency, but the tank selecting which of thecrystal’s modes is active. The tank couldbe tuned to the crystal’s fundamental, orone of its overtones.

Fundamental oscillators can be builtwithout a tank quite successfully, but thereis always the occasional one that starts upon an overtone or spurious mode. Somesimple oscillators have been known tochange modes while running (an effecttriggered by changes in temperature or

loading) or to not always start in the samemode! A series-mode oscillator shouldpresent a low impedance to the crystal atthe operating frequency. In Fig 10.24, thetapped collector tank presents a trans-formed fraction of the 1-kΩ collector loadresistor to one end of the crystal, and theemitter presents a low impedance to theother. To build a practical oscillator fromthis circuit, choose an inductor with a re-actance of about 300 Ω at the wanted fre-quency and calculate C1 in series with C2to resonate with it. Choose C1 to be 3 to 4times larger than C2. The amplifier’s qui-escent (“idling”) current sets the gain andhence the operating level. This is not eas-ily calculable, but can be found by experi-ment. Too little quiescent current and theoscillator will not start reliably; too muchand the transistor can drive itself into satu-ration. If an oscilloscope is available, itcan be used to check the collector wave-form; otherwise, some form of RF voltme-ter can be used to allow the collectorvoltage to be set to 2 to 3 V RMS. 3.3 kΩwould be a suitable starting point for theemitter bias resistor. The transistor type isnot critical; 2N2222A or 2N3904 would

be fine up to 30 MHz; a 2N5179 wouldallow operation as an overtone oscillatorto over 100 MHz (because of the low col-lector voltage rating of the 2N5179, a sup-ply voltage lower than 12 V is required).The ferrite bead on the base gives someprotection against parasitic oscillation atUHF.

If the crystal is shorted, this circuitshould still oscillate. This gives an easyway of adjusting the tank; it is even betterto temporarily replace the crystal with asmall-value (tens of ohms) resistor tosimulate its equivalent series resistance(ESR), and adjust L until the circuit oscil-lates close to the wanted frequency. Thenrestore the crystal and set the quiescentcurrent. If a lot of these oscillators werebuilt, it would sometimes be necessary toadjust the current individually due to thedifferent equivalent series resistance ofindividual crystals. One variant of this cir-cuit has the emitter connected directly tothe C1/C2 junction, while the crystal is adecoupler for the transistor base (the ex-isting capacitor and ferrite bead not beingused). This works, but with a greater riskof parasitic oscillation.

Fig 10.24—A basic series-mode crystaloscillator. A 2N5179 can be used in thiscircuit if a lower supply voltage isused; see text. Fig 10.25—A Butler crystal oscillator.

Chapter 10.pmd 7/31/2006, 11:34 AM25

10.26 Chapter 10

We commonly want to trim a crystaloscillator’s frequency. While off-tuning thetank a little will pull the frequency slightly,too much detuning spoils the mode controland can stop oscillation (or worse, make thecircuit unreliable). The answer to this is toadd a trimmer capacitor, which will act aspart of the equivalent series tuned circuit,in series with the crystal. This will shift thefrequency in one way only, so the crystalfrequency must be respecified to allow thefrequency to be varying around the required

Fig 10.26—The crystal in the series-tuned Colpitts oscillator at A operates in its series-resonant mode. B shows KA2WEU’slow-noise version, which uses the crystal as a filter and features high harmonic suppression. The circuit at C builds on the Bversion by adding a common-base output amplifier and ALC loop.

value. It is common to specify a crystal’sfrequency with a standard load (30 pF iscommonly specified), so that the manufac-turer grinds the crystal such that the seriesresonance of the specified mode is accuratewhen measured with a capacitor of thisvalue in series. A 15- to 50-pF trimmer canbe used in series with the crystal to give finefrequency adjustment. Too little capaci-tance can stop oscillation or prevent reli-able starting. The Q of crystals is so highthat marginal oscillators can take several

seconds to start!This circuit can be improved by reduc-

ing the crystal’s driving impedance withan emitter follower as in Fig 10.25. This isthe Butler oscillator. Again the tank con-trols the mode to either force the wantedovertone or protect the fundamental mode.The tank need not be tapped because Q2provides current gain, although the circuitis sometimes seen with C split, driving Q2from a tap. The position between the emit-ters offers a good, low-impedance envi-

Chapter 10.pmd 7/31/2006, 11:34 AM26

Oscillators and Synthesizers 10.27

an overtone crystal has been used. To fur-ther help stability, the power dissipated inthe crystal is kept to about 50 μW. Thecommon-base stage is effectively drivenfrom a higher impedance than its own inputimpedance, under which conditions it givesa very low noise figure.

VXOsSome crystal oscillators have frequency

trimmers. If the trimmer is replaced by avariable capacitor as a front-panel control,we have a variable crystal oscillator(VXO): a crystal-based VFO with a nar-row tuning range, but good stability andnoise performance. VXOs are often usedin small, simple QRP transmitters to tunea few kilohertz around common callingfrequencies. Artful constructors, usingoptimized circuits and components, haveachieved 1000-ppm tuning ranges. Poor-quality “soft” crystals are more pullablethan high-Q ones. Overtone crystals arenot suited to VXOs. For frequencies be-yond the usual limit for fundamental modecrystals, use a fundamental unit and fre-quency multipliers.

ICOM and Mizuho made some 2-m SSBtransceivers based on multiplied VXOlocal oscillators. This system is simple andcan yield better performance than manyexpensive synthesized radios. SSB filtersare available at 9 or 10.7 MHz, to yieldsufficient image rejection with a singleconversion. Choice of VXO frequencydepends on whether the LO is to be aboveor below signal frequency and how muchmultiplication can be tolerated. Below8 MHz multiplier filtering is difficult.Above 15 MHz, the tuning range per crys-tal narrows. A 50-200 kHz range per crys-tal should work with a modern front-end

ronment to keep the crystal’s in-circuit Qhigh. R, in the emitter of Q1, is again selec-ted to give reliable oscillation. The circuithas been shown with a capacitive load forthe crystal, to suit a unit specified for a30-pF load. An alternative circuit to giveelectrical fine tuning is also shown. Thediodes across the tank act as limiters tostabilize the operating amplitude and limitthe power dissipated in the crystal by clip-ping the drive voltage to Q2. The tankshould be adjusted to peak at the operatingfrequency, not used to trim the frequency.The capacitance in series with the crystalis the proper frequency trimmer.

The Butler circuit works well, and hasbeen used in critical applications to 140 MHz(seventh-overtone crystal, 2N5179 transis-tor). Although the component count is high,the extra parts are cheap ones. Increasingthe capacitance in series with the crystalreduces the oscillation frequency but has aprogressively diminishing effect. Decreas-ing the capacitance pulls the frequencyhigher, to a point at which oscillation stops;before this point is reached, start-up willbecome unreliable. The possible amount ofadjustment, called pulling range, dependson the crystal; it can range from less thanten to several hundred parts per million.Overtone crystals have much less pullingrange than fundamental crystals on thesame frequency; the reduction in pulling isroughly proportional to the square of theovertone number.

Low-Noise Crystal OscillatorsFig 10.26A shows a crystal operating in

its series mode in a series-tuned Colpittscircuit. Because it does not include an LCtank to prevent operation on unwantedmodes, this circuit is intended for funda-mental mode operation only and relies onthat mode being the most active. If thecrystal is ordered for 30-pF loading, thefrequency trimming capacitor can be ad-justed to compensate for the loading of thecapacitive divider of the Colpitts circuit.An unloaded crystal without a trimmerwould operate slightly off the exact seriesresonant frequency in order to create aninductive impedance to resonate with thedivider capacitors. Ulrich Rohde,KA2WEU, in Fig 4-47 of his book DigitalPLL Frequency Synthesizers—Theory andDesign, published an elegant alternativemethod of extracting an output signal fromthis type of circuit, shown in Fig 10.26B.This taps off a signal from the current inthe crystal itself. This can be thought of asusing the crystal as a band-pass filter forthe oscillator output. The RF choke in theemitter keeps the emitter bias resistor fromloading the tank and degrading the Q. Inthis case (3-MHz operation), it has been

chosen to resonate close to 3 MHz with theparallel capacitor (510 pF) as a means offorcing operation on the wanted mode. The10-Ω resistor and the transformed loadimpedance will reduce the in-circuit Q ofthe crystal, so a further development sub-stituted a common base amplifier for theresistor and transformer. This is shown inFig 10.26C. The common-base amplifieris run at a large quiescent current to give avery low input impedance. Its collector istuned to give an output with low harmoniccontent and an emitter follower is used tobuffer this from the load. This oscillatorsports a simple ALC system, in which theamplified and rectified signal is used toreduce the bias voltage on the oscillatortransistor’s base. This circuit is describedas achieving a phase noise level of –168dBc/Hz a few kilohertz out from the car-rier. This may seem far beyond what mayever be needed, but frequency multiplica-tion to high frequencies, whether by clas-sic multipliers or by frequencysynthesizers, multiplies the deviation ofany FM/PM sidebands as well as the car-rier frequency. This means that phasenoise worsens by 20 dB for each tenfoldmultiplication of frequency. A clean crys-tal oscillator and a multiplier chain is stillthe best way of generating clean micro-wave signals for use with narrow-bandmodulation schemes.

It has already been mentioned that over-tone crystals are much harder to pull thanfundamental ones. This is another way ofsaying that overtone crystals are less influ-enced by their surrounding circuit, which ishelpful in a frequency-standard oscillatorlike this one. Even though 5 MHz is in themain range of fundamental-mode crystalsand this circuit will work well with them,

Fig 10.27—A wide-range variable-crystal oscillator (VXO).

Chapter 10.pmd 7/31/2006, 11:34 AM27

10.28 Chapter 10

design feeding a good 9-MHz IF, for acontest quality 2-m SSB Receiver.

The circuit in Fig 10.27 is a JFET VXOfrom Wes Hayward, W7ZOI, and DougDeMaw, W1FB, optimized for wide-rangepulling. Published in Solid State Designfor the Radio Amateur, many have beenbuilt and its ability to pull crystals as far aspossible has been proven. Ulrich Rohde,KA2WEU, has shown that the diodearrangement as used here to make signal-dependent negative bias for the gate con-fers a phase-noise disadvantage, butoscillators like this that pull crystals as faras possible need any available means tostabilize their amplitude and aid start-up.In this case, the noise penalty is worth pay-ing. This circuit can achieve a 2000-ppmtuning range with amenable crystals. If you

Fig 10.28—Using an inductor to “tuneout” C0 can increase a crystaloscillator’s pulling range.

have some overtone crystals in your junkbox whose fundamental frequency is closeto the wanted value, they are worth trying.

This sort of circuit doesn’t necessarilystop pulling at the extremes of the possibletuning range, sometimes the range is set bythe onset of undesirable behavior such asjumping mode or simply stopping oscillat-ing. L was a 16-μH slug-tuned inductor for10-MHz operation. It is important to mini-mize the stray and interwinding capaci-tance of L since this dilutes the range ofimpedance presented to the crystal.

One trick that can be used to aid thepulling range of oscillators is to tune outthe C0 of the equivalent circuit with anadded inductor. Fig 10.28 shows how. L ischosen to resonate with C0 for the indi-vidual crystal, turning it into a high-im-pedance parallel-tuned circuit. The Q ofthis circuit is orders of magnitude less thanthe Q of the true series resonance of thecrystal, so its tuning is much broader. Thevalue of C0 is usually just a few picofar-ads, so L has to be a fairly large valueconsidering the frequency it is resonatedat. This means that L has to have low straycapacitance or else it will self-resonate ata lower frequency. The tolerance on C0and the variations of the stray C of theinductor means that individual adjustmentis needed. This technique can also workwonders in crystal ladder filters.

Logic-Gate Crystal OscillatorsA 180° phase-shift network and an in-

verting amplifier can be used to make anoscillator. A single stage RC low-pass net-work cannot introduce more than 90° ofphase shift and only approaches that as thesignal becomes infinitely attenuated. Twostages can only approach 180° and thenhave immense attenuation. It takes threestages to give 180° and not destroy the loopgain. Fig 10.29A shows the basic form ofthe phase-shift oscillator; Fig 10.29B is anexample the phase-shift oscillator com-monly used in commercial Amateur Radiotransceivers as an audio sidetone oscilla-tor. The frequency-determining network ofan RC oscillator has a Q of less than one,which is a massive disadvantage comparedto an LC or crystal resonator. The Piercecrystal oscillator is a converted phase-shiftoscillator, with the crystal taking the placeof one series resistor, Fig 10.29C. At theexact series resonance, the crystal looksresistive, so by suitable choice of the ca-pacitor values, the circuit can be made tooscillate. The crystal has a far steeperphase/frequency relationship than the restof the network, so the crystal is the domi-nant controller of the frequency. ThePierce circuit is rarely seen in this fullform. Instead, a cut-down version has be-

come the most common circuit for crystal-clock oscillators for digital systems.Fig 10.29D shows this minimalist Pierce,using a logic inverter as the amplifier. Rbiasprovides dc negative feedback to bias thegate into its linear region. At first sight itappears that this arrangement should notoscillate, but the crystal is a resonator (nota simple resistor) and oscillation occursoffset from the series resonance, where thecrystal appears inductive, which makes upthe missing phase shift. This is one circuitthat cannot oscillate exactly on the crys-tal’s series resonance. This circuit is in-cluded in many microprocessors and otherdigital ICs that need a clock. It is also theusual circuit inside the miniature clockoscillator cans. It is not a very reliable cir-cuit, as operation is dependent on thecrystal’s equivalent series resistance andthe output impedance of the logic gate,occasionally the logic device or the crystalneeds to be changed to start oscillation,sometimes playing with the capacitor val-ues is necessary. It is doubtful whetherthese circuits are ever designed—values(the two capacitors) seem to be arrived atby experiment. Once going, the circuit isreliable, but its drift and noise are moder-ate, not good—acceptable for a clock oscil-lator. The commercial packaged oscillatorsare the same, but the manufacturers havehandled the production foibles on a batch-by-batch basis.

RC OscillatorsPlenty of RC oscillators are capable of

operating to several megahertz. Some ofthese are really constant-current-into-capacitor circuits, which are easier to makein silicon. Like the phase-shift oscillatorabove, the timing circuit Q is less than one,giving very poor noise performance that’sunsuitable even to the least demandingradio application. One example is theoscillator section of the CD4046 phase-locked-loop IC. This oscillator has poorstability over temperature, large batch-to-batch variation and a wide variation in itsvoltage-to-frequency relationship. It is notrecommended that this sort of oscillator isused at RF in radio systems. (The ’4046phase detector section is very useful, how-ever, as we’ll see later.) These oscillatorsare best suited to audio applications.

VHF AND UHF OSCILLATORSA traditional way to make signals at

higher frequencies is to make a signal ata lower frequency (where oscillators areeasier) and multiply it up to the wantedrange. Multipliers are still one of the easi-est ways of making a clean UHF/micro-wave signal. The design of a multiplierdepends on whether the multiplication

Chapter 10.pmd 7/31/2006, 11:34 AM28

Oscillators and Synthesizers 10.29

Fig 10.29—In a phase-shift oscillator (A) based on logic gates, a chain of RCnetworks—three or more—provide the feedback and phase shift necessary foroscillation, at the cost of low Q and considerable loop loss. Many commercialAmateur Radio transceivers have used a phase-lead oscillator similar to thatshown at B as a sidetone generator. Replacing one of the resistors in A with acrystal produces a Pierce oscillator (C), a cut-down version of which (D) hasbecome the most common clock oscillator configuration in digital systems.

Fig 10.30—Oscillators that use trans-mission-line segments as resonators.Such oscillators are more common thanmany of us may think, as Fig 10.31reveals.

factor is an odd or even number. For oddmultiplication, a Class-C biased amplifiercan be used to create a series of harmon-ics; a filter selects the one wanted. Foreven multiplication factors, a full-wave-rectifier arrangement of distorting de-vices can be used to create a series ofharmonics with strong even-order com-ponents, with a filter selecting the wantedcomponent. At higher frequencies, diode-based passive circuits are commonlyused. Oscillators using some of the LCcircuits already described can be used inthe VHF range. At UHF different ap-proaches become necessary.

Fig 10.30 shows a pair of oscillatorsbased on a resonant length of line. The first

one is a return to basics: a resonator, anamplifier and a pair of coupling loops. Theamplifier can be a single bipolar or FETdevice or one of the monolithic microwaveintegrated circuit (MMIC) amplifiers. Thesecond circuit is really a Hartley, and onewas made as a test oscillator for the 70-cmband from a 10-cm length of wire sus-pended 10 mm over an unetched PC boardas a ground plane, bent down and solderedat one end, with a trimmer at the other end.The FET was a BF981 dual-gate deviceused as a source follower.

No free-running oscillator will be stableenough on these bands except for use withwideband FM or video modulation andAFC at the receiver. Oscillators in this

range are almost invariably tuned withtuning diodes controlled by phase-locked-loop synthesizers, which are themselvescontrolled by a crystal oscillator.

There is one extremely common UHFoscillator that is rarely applied intention-ally. The answer to this riddle is a configu-ration that is sometimes deliberately builtas a useful wide-tuning oscillator cover-ing say, 500 MHz to 1 GHz—and is alsothe modus operandi of a very commonform of spurious VHF/UHF oscillation incircuitry intended to process lower-fre-quency signals! This oscillator has no gen-erally accepted name. It relies on thecreation of a small negative resistance inseries with a series resonant LC tank.Fig 10.31A shows the circuit in its sim-plest form. This circuit is well-suited toconstruction with printed-circuit induc-tors. Common FR4 glass-epoxy board islossy at these frequencies; better perfor-mance can be achieved by using the muchmore expensive glass-Teflon board. If youcan get surplus offcuts of this type of ma-terial, it has many uses at UHF and micro-wave, but it is difficult to use, as theadhesion between the copper and the sub-strate can be poor. A high-UHF transistorwith a 5-GHz fT like the BFR90 is suitable;the base inductor can be 30 mm of 1-mmtrace folded into a hairpin shape (induc-tance, less than 10 nH). Analyzing this cir-cuit using a comprehensive model of theUHF transistor reveals that the emitter pre-sents an impedance that is small, resistiveand negative to the outside world. If this islarge enough to more than cancel the

Chapter 10.pmd 7/31/2006, 11:34 AM29

10.30 Chapter 10

effective series resistance of the emittertank, oscillation will occur. Fig 10.31Bshows a very basic emitter-follower cir-cuit with some capacitance to ground onboth the input and output. If the capacitorshunting the input is a distance away fromthe transistor, the trace can look like aninductor—and small capacitors at audioand low RF can look like very gooddecouplers at hundreds of MHz. Thelength to the capacitor shunting the outputwill behave as a series resonator at a fre-quency where it is a 1/4 wavelength long.This circuit is the same as in Fig 10.31A;it, too, can oscillate. The semiconductormanufacturers have steadily improvedtheir small-signal transistors to give bettergain and bandwidth so that any transistorcircuit where all three electrodes findthemselves decoupled to ground at UHFmay oscillate at several hundred mega-hertz. The upshot of this is that there is nolonger any branch of electronics where RFdesign and layout techniques can be safelyignored. A circuit must not just be de-signed to do what it should do, it must alsobe designed so that it cannot do what itshould not do.

There are three ways of taming such acircuit; adding a small resistor, perhaps50 to 100 Ω in the collector lead, close tothe transistor, or adding a similar resistorin the base lead, or by fitting a ferrite beadover the base lead under the transistor. Theresistors can disturb dc conditions, de-pending on the circuit and its operatingcurrents. Ferrite beads have the advantagethat they can be easily added to existingequipment and have no effect at dc andlow frequencies. Beware of some electri-cally conductive ferrite materials that canshort transistor leads. If an HF oscillatoruses beads to prevent any risk of spurs(Fig 10.16), the beads should be anchoredwith a spot of adhesive to prevent movementthat can cause small frequency shifts. Ferritebeads of Fair-Rite no. 43 material are espe-cially suitable for this purpose; theyare specified in terms of impedance, notinductance. Ferrites at frequencies abovetheir normal usable range become very lossyand can make a lead look not inductive, butlike a few tens of ohms, resistive.

Microwave OscillatorsLow-noise microwave signals are still

best made by multiplying a very-low-noiseHF crystal oscillator, but there are a num-ber of oscillators that work directly atmicrowave frequencies. Such oscillatorscan be based on resonant lengths ofstripline or microstrip, and are simplyscaled-down versions of UHF oscillators,using microwave transistors and printedstriplines on a low-loss substrate, like alu-

Fig 10.31—High device gain at UHF andresonances in circuit board traces canresult in spurious oscillations even innon-RF equipment.

Fig 10.32—Evolution of the cavityresonator.

mina. Techniques for printing metal traceson substrates to form filters, couplers,matching networks and so on, have beenintensively developed over the past twodecades. Much of the professional micro-wave community has moved away fromwaveguide and now uses low-loss coaxialcable with a solid-copper shield—semi-rigid cable—to connect circuits made flaton ceramic or Teflon based substrates.Semiconductors are often bonded on asunpackaged chips, with their bond-wireconnections made directly to the traces onthe substrate. At lower microwave fre-quencies, they may be used in standardsurface-mount packages. From an Ama-teur Radio viewpoint, many of the pro-cesses involved are not feasible withoutaccess to specialized furnaces and materi-als. Using ordinary PC-board techniqueswith surface-mount components allowsthe construction of circuits up to 4 GHz orso. Above this, structures get smaller andaccuracy becomes critical; also PC-boardmaterials quickly become very lossy.

Older than stripline techniques and farmore amenable to home construction, cav-ity-based oscillators can give the highestpossible performance. The dielectric con-stant of the substrate causes stripline struc-tures to be much smaller than they wouldbe in free air, and the lowest-loss sub-strates tend to have very high dielectricconstants. Air is a very low-loss dielectricwith a dielectric constant of 1, so it giveshigh Q and does not force excessive min-iaturization. Fig 10.32 shows a series ofstructures used by G. R. Jessup, G6JP, toillustrate the evolution of a cavity from atank made of lumped components. Allcavities have a number of different modesof resonance, the orientation of the cur-rents and fields are shown in Fig 10.33.The cavity can take different shapes, butthat shown has proven to suppress un-wanted modes well. The gap need not becentral and is often right at the top. A screwcan be fitted through the top, protrudinginto the gap, to adjust the frequency.

To make an oscillator out of a cavity, anamplifier is needed. Gunn and tunneldiodes have regions in their characteris-tics where their current falls with increas-ing bias voltage. This is negative resis-tance. If such a device is mounted in a loopin a cavity and bias applied, the negativeresistance can more than cancel the effec-tive loss resistance of the cavity, causingoscillation. These diodes are capable ofoperating at extremely high frequenciesand were discovered long before transis-tors were developed that had any gain atmicrowave frequencies.

A Gunn-diode cavity oscillator is thebasis of many of the Doppler-radar

Chapter 10.pmd 7/31/2006, 11:34 AM30

Oscillators and Synthesizers 10.31

Fig 10.33—Currents and fields in acavity.

Fig 10.34—A Gunn diode oscillatoruses negative resistance and a cavityresonator to produce radio energy.

Fig 10.35—A transistor can alsodirectly excite a cavity resonator.

modules used to detect traffic or intruders.Fig 10.34 shows a common configuration.The coupling loop and coax output con-nector could be replaced with a simpleaperture to couple into waveguide or amixer cavity. Fig 10.35 shows a transistorcavity oscillator version using a modernmicrowave transistor. FET or bipolar de-vices can be used. The two coupling loops

are completed by the capacitance of thefeedthroughs.

The dielectric-resonator oscillator(DRO) may soon become the most com-mon microwave oscillator of all, as it isused in the downconverter of satellite TVreceivers. The dielectric resonator itselfis a ceramic cylinder, like a miniaturehockey puck, several millimeters indiameter. The ceramic has a very highdielectric constant, so the surface (whereceramic meets air in an abrupt mismatch)reflects electromagnetic waves and makesthe ceramic body act as a resonant cavity.It is mounted on a substrate and coupledto the active device of the oscillator bya stripline that runs past it. At 10 GHz,a FET made of gallium arsenide(GaAsFET), rather than silicon, is nor-mally used. The dielectric resonatorelements are made at frequencies appro-priate to mass applications like satelliteTV. The set-up charge to manufacturesmall quantities at special frequencies islikely to be prohibitive for the foreseeablefuture. The challenge with these devices isto devise new ways of using oscillators onindustry standard frequencies. Their chiefattraction is their low cost in large quanti-ties and compatibility with microwavestripline (microstrip) techniques. Fre-quency stability and Q are competitivewith good cavities, but are inferior to thatachievable with a crystal oscillator andchain of frequency multipliers. SatelliteTV downconverters need free runningoscillators with less than 1 MHz of drift at10 GHz.

There are a number of thermionic(vacuum-tube) microwave sources,klystrons, magnetrons and backwardswave oscillators (BWOs). Availabledevices are either very old or designedfor very high power.

The yttrium-iron garnet (YIG) oscilla-tor was developed for a wide tuning rangeas a solid-state replacement for the BWO,and many of them can be tuned over more

than an octave. They are complete, pack-aged units that appear to be a heavy blockof metal with low-frequency connectionsfor power supplies and tuning, and an SMAconnector for the RF output. Themanufacturer’s label usually states thetuning range and often the power supplyvoltages. This is very helpful because,with new units priced in the kilodollarrange, it is important to be able to identifysurplus units. The majority of YIGs are inthe 2- to 18-GHz region, although unitsdown to 500 MHz and up to 40 GHz areoccasionally found. In this part of the spec-trum there is no octave-tunable device thatcan equal their cleanliness and stability. A3-GHz unit drifting less than 1 kHz persecond gives an idea of typical stability.This seems very poor—until we realizethat this is 0.33 ppm per second. Neverthe-less, any YIG application involving nar-row-band modulation will usually requiresome form of frequency stabilizer.

Good quality, but elderly, RF spectrumanalyzers have found their way into theworkshops of a number of dedicated con-structors. A 0- to 1500-MHz analyzer usu-ally uses a 2- to 3.5-GHz YIG as its firstlocal oscillator, and its tuning circuits aredesigned around it. A reasonable under-standing of YIG oscillators will help introubleshooting and repair.

YIG spheres are resonant at a frequencycontrolled not only by their physical di-mensions, but also by any applied mag-netic field. A YIG sphere is carefullyoriented within a coupling loop connectedto a negative-resistance device and thewhole assembly is placed between thepoles of an electromagnet. Negative-resis-tance diodes have been used, but transistorcircuits are now common. The support forthe YIG sphere often contains a thermo-statically controlled heater to reduce tem-perature sensitivity.

The first problem with a magneticallytuned oscillator is that magnetic fields,especially at low frequencies, areextremely difficult to shield; the tuningwill be influenced by any local fields.Varying fields will cause frequency modu-lation. The magnetic core must be care-fully designed to be all-enclosing in anattempt at self-shielding and then one ormore nested mu-metal cans are fittedaround everything. It is still important tosite the unit away from obvious sources ofmagnetic noise, like power transformers.Cooling fans are less obvious sources offluctuating magnetic fields, since some are20 dB worse than a well designed200-W 50/60-Hz transformer.

The second problem is that the oscil-lator’s internal tuning coils need signifi-cant current from the power supply to

Chapter 10.pmd 7/31/2006, 11:34 AM31

10.32 Chapter 10

create strong fields. This can be eased byadding a permanent magnet as a fixed“bias” field, but the bias will shift as themagnet ages. The only solution is to havea coil with many turns, hence high induc-tance, which will require a high supplyvoltage to permit rapid tuning, thus in-creasing the power consumption. Theusual compromise is to have dual coils:One with many turns allows slow tuningover a wide range; a second with muchfewer turns allows fast tuning or FM overa limited range. The main coil can have asensitivity in the 20 MHz/mA range; andthe “FM” coil, perhaps 500 kHz/mA.

The frequency/current relationship canhave excellent linearity. Fig 10.36 showsthe construction of a YIG oscillator.

For some insight into present and futuretrends in oscillator applications, see thesidebar beginning on page 52.

Fig 10.36—A yttrium-iron-garnet (YIG)sphere serves as the resonator in thesweep oscillators used in manyspectrum analyzers.

Frequency Synthesizersceiver, cell phone, pager, AM/FM enter-tainment radio, scanner, television, HFcommunications equipment, or test equip-ment contains a synthesizer. Synthesis isthe technology that allows an easy inter-face with both computers and microproces-sors. It provides amateurs with manydesirable features, such as the feel of ananalog knob with 10-Hz frequency incre-ments, accuracy and stability determinedby a single precision crystal oscillator, fre-quency memories, and continuously vari-able splits. Now reduced in size to only afew small integrated circuits, frequencysynthesizers have also replaced the cum-bersome chains of frequency multipliersand filters in VHF, UHF and microwaveequipment, giving rise to many of thehighly portable communications deviceswe use today. Frequency synthesis has alsohad a major impact in lowering the cost of

modern equipment, as well as reducingmanufacturing complexity.

Frequency synthesizers have been cate-gorized in two general types, direct syn-thesizers (not to be confused with directdigital synthesis) and indirect synthesiz-ers. The architecture of the direct typesconsists mainly of multipliers, dividers,mixers, filters and copious amounts ofshielding. They are also cumbersome andexpensive, and have all but disappearedfrom the market. The indirect varietiesmake use of programmable dividers andphase-lock loops, thereby considerablyreducing their complexity. There are anumber of variations on the indirecttheme. They include programmable divi-sion, variable or dual modulus division,and fractional N. Other techniques thathave been developed since the direct/indirect classification are direct digital

Like many of our modern technologies,the origins of Frequency Synthesis can betraced back to WW II. The driving forcewas the desire for stable, rapidly switch-able and accurate frequency control tech-nology to meet the demands of narrow-band, frequency-agile HF communicationssystems without resorting to large banks ofswitched crystals. Early synthesizers werecumbersome and expensive, and thereforetheir use was limited to the most sophisti-cated communications systems. With thehelp of the same technologies that havetaken computers from “rooms” to the palmsof our hands, the role of frequency synthe-sis has gone far beyond its original purposeand has become one of the most enablingtechnologies in modern communicationsequipment.

Just about every communications devicemanufactured today, be it a handheld trans-

Chapter 10.pmd 7/31/2006, 11:34 AM32

Oscillators and Synthesizers 10.33

pares the relative timing of 5-V pulses(CMOS logic level) at 20 kHz. Inevitably,the phase detector output will containstrong components at 20 kHz and its har-monics in addition to the wanted “steering”signal. The loop filter must block theseunwanted signals; otherwise the loopamplifier will amplify them and apply themto the VCO, generating unwanted FM. Nofiltering can be perfect and the VCO is verysensitive, so most synthesizers have mea-surable sidebands spaced at the phasedetector operating frequency (and harmon-ics) away from the carrier. These are calledreference-frequency sidebands. (Exactlywhich is the reference frequency is ambigu-ous: Do we mean the frequency of the ref-erence oscillator, or the frequency appliedto the reference input of the phase detector?You must look carefully at context when-ever reference frequency is mentioned.)

The loop filter is not usually built as asingle block of circuitry, it is often madeup of three areas: Some components di-rectly after the phase detector, a shapedfrequency response in the loop amplifierand some more components between theloop amplifier and the VCO.

The PLL is like a feedback amplifier,although the “signal” around its loop is rep-resented by frequency in some places, byphase in others and by voltage in others.Like any feedback amplifier, there is therisk of instability and oscillations travelingaround the loop, which can be seen as mas-sive FM on the output and a strong ac signalon the tuning line to the VCO. The loop’sfiltering and gain has to be designed to pre-vent this.

The following example illustrates theloop action: Imagine that we shift the ref-erence oscillator by 1 Hz. The referenceapplied to the phase detector would shift1/M Hz, and the loop would respond byshifting the output frequency to produce amatching shift at the other phase detectorport, so the output is shifted by N/M Hz.Imagine now that we apply a very smallamount of FM to the reference oscillator.The amount of deviation will be amplifiedby N/M—but this is only true for lowmodulating frequencies. If the modulatingfrequency is increased, eventually the loopfilter starts to reduce the gain around theloop, the loop ceases to track the modula-tion and the deviation at the output falls.This is referred to as the closed-loop fre-quency response or closed-loop band-width. Poorly designed loops can havepoor closed-loop responses. For example,the gain can have large peaks above N/Mat some reference modulation frequencies,indicating marginal stability and exces-sively amplifying any noise at those off-sets from the carrier. The design of the

Fig 10.37—A basic phase-locked-loop (PLL) synthesizer acts to keep the divided-down signal from its voltage-controlled oscillator (VCO) phase-locked to thedivided-down signal from its reference oscillator. Fine tuning steps are thereforepossible without the complication of direct synthesis.

synthesis (DDS) and rate multiplier syn-thesis. Synthesizers used in equipmenttoday are usually a hybrid of the indirecttechnique and some of the latter develop-ments. There are entire textbooks devotedto treating these various methods in depth,and as such it is not practical to discusseach one in detail in this handbook. Wewill focus on the indirect approach, basedon phase-locked loops. This approachcame to dominate frequency synthesislong ago, and is still dominant today.

Phase-Locked LoopsTo understand the indirect synthesizer,

we need to understand the phase-lock loop.The principle of the phase-locked loop(PLL) synthesizer is very simple. An os-cillator can be built to cover the requiredfrequency range, so what is needed is asystem to keep its tuning correct. This isdone by continuously comparing the phaseof the oscillator to a stable reference, suchas a crystal oscillator and using the resultto steer the tuning. If the oscillator fre-quency is too high, the phase of its outputwill start to lead that of the reference byan increasing amount. This is detected andis used to decrease the frequency of theoscillator. Too low a frequency causes anincreasing phase lag, which is detected andused to increase the frequency of the oscil-lator. In this way, the oscillator is lockedinto a fixed-phase relationship with thereference, which means that their frequen-cies are held exactly equal.

The oscillator is now under control, butis locked to a fixed frequency. Fig 10.37shows the next step. The phase detector

does not simply compare the oscillatorfrequency with the reference oscillator;both signals have now been passed throughfrequency dividers. Advances in digitalintegrated circuits have made frequencydividers up to microwave frequencies(over 10 GHz) commonplace. The divideron the reference path has a fixed divisionfactor, but that in the VCO path is pro-grammable, the factor being entered digi-tally as a (usually) binary word. The phasedetector is now operating at a lower fre-quency, which is a submultiple of both thereference and output frequencies.

The phase detector, via the loop ampli-fier, steers the oscillator to keep both itsinputs equal in frequency. The referencefrequency, divided by M, is equal to theoutput frequency divided by N. The outputfrequency equals the reference frequency× N/M. N is programmable (and an inte-ger), so this synthesizer is capable of tun-ing in steps of Fref/M.

As an example, to make a 2-m radiocovering 144 to 148 MHz with a 10.7-MHzIF, we need a local oscillator covering154.7-158.7 MHz. If the channel spacingis 20 kHz, then Fref/M is 20 kHz, so N is7735 to 7935. There is a free choice of Frefor M, but division by a round binary num-ber is easiest. Crystals are readily avail-able for 10.24 MHz, and one of these(divided by 512) will give 20 kHz at lowcost. ICs are readily available containingmost of the circuitry necessary for the ref-erence oscillator (less the crystal), withthe programmable divider and the phasedetector.

The phase detector in this example com-

Chapter 10.pmd 7/31/2006, 11:34 AM33

10.34 Chapter 10

loop filtering and the gain around the loopsets the closed-loop performance.

In a single-loop synthesizer, there is atrade-off between how fine the step sizecan be versus the performance of the loop.A loop with a very fine step size runs thephase detector at a very low frequency, sothe loop bandwidth has to be kept very,very low to keep the reference-frequencysidebands low. Also, the reference oscil-lator usually has much better phase-noiseperformance than the VCO, and (withinthe loop bandwidth) the loop acts to op-pose the low-frequency components of theVCO phase-noise sidebands. This veryuseful cleanup activity is lost when loopshave to be narrowed in bandwidth to allownarrow steps. Low-bandwidth loopsare slow to respond—they exhibit overlylong settling times—to the changes in Nnecessary to change channels. Absolutelyeverything seems to become impossible inany attempt to get fine resolution from aphase-locked loop.

Single-loop synthesizers are okay forthe 2-m FM example for a number of rea-sons: Channel spacings in this band arenot too small, FM is not as critical of phasenoise as other modes, 2-m FM is rarelyused for weak-signal DXing and thechannelization involved does away withthe desirability of simulating fast, smooth,continuous tuning. None of these excusesapply to MF/HF radios, however, and waysto circumvent the problems are needed.

A clue was given earlier, by carefullyreferring to single-loop synthesizers. It’spossible to use several PLLs, or for thatmatter, some of the other mentioned tech-niques such as DDS, fractional N, vari-able- or dual-modulus etc, as componentsin a larger structure, dividing the fre-quency of one loop and adding it to thefrequency of another that has not been di-vided. This represents a form of hybridbetween the old direct synthesizer and thePLL. A form of PLL containing a mixer inplace of the programmable divider can beused to perform the frequency addition.

Let us continue our discussion of phase-lock loop synthesizers by examining therole of each of the component pieces of thesystem. They are, the VCO, the dividers,prescalers, the phase detector and the loop-compensation amplifier.

Voltage-Controlled OscillatorsVoltage-controlled oscillators are com-

monly referred to as VCOs. (Voltage-tuned oscillator, VTO, more accuratelydescribes the circuitry most commonlyused in VCOs, but tradition is tradition!An exception to this is the YIG oscillator,described earlier as sometimes used inUHF and microwave PLLs, which is cur-

rent-tuned.) In all the oscillators describedso far, except for permeability tuning, thefrequency is controlled or trimmed by avariable capacitor. These are modified forvoltage tuning by using a tuning diode. Theoscillator section of this chapter shows theschematic symbol and construction detailof a voltage-controlled or varactor (tun-ing) diode as is typically used in a VCO(see Fig 10.18).

When this oscillator is used in a PLL,the PLL adjusts the tuning voltage to com-pletely cancel any drift in the oscillator(provided it does not drift further than itcan be tuned) regardless of its cause, so nospecial care is needed to compensate a PLLVCO for temperature effects. Adjustingthe inductor core will not change the fre-quency; the PLL will adjust the tuningvoltage to hold the oscillator on frequency.This adjustment is important, though. It’sset so that the tuning voltage neither getstoo close to the maximum available, nortoo low for acceptable Q, as the PLL istuned across its full range.

In a high-performance, low noise syn-thesizer, the VCO may be replaced by abank of switched VCOs, each covering asection of the total range needed, eachVCO having better Q and lower noise be-cause the lossy diodes constitute only asmaller part of the total tank capacity thanthey would in a single, full-range oscilla-tor. Another method uses tuning diodes foronly part of the tuning range and switchesin other capacitors (or inductor taps) toextend the range. The performance ofwide-range VCOs can be improved byusing a large number of varactors in paral-lel in a high-C, low-L tank.

Programmable DividersDesigning your own programmable di-

vider is now a thing of the past, as com-plete systems have been made as singleICs for over 10 years. The only remainingreason to do so is when an ultra-low-noisesynthesizer is being designed.

It may seem strange to think of a digitalcounter as a source of random noise, butdigital circuits have propagation delayscaused by analog effects, such as thecharging of stray capacitances. Digitalcircuits are composed of the same sorts ofcomponents as analog circuits; there areno such things as digital and analog elec-trons! Signals are made of voltages andcurrents. The prime difference betweenanalog and digital electronics lies in thedifferent ways meaning is assigned to themagnitude of a signal. The differences incircuit design are consequences of this, notcauses.

Thermal noise in the components of a digi-tal circuit, added to the signal voltage will

slightly change the times at which thresh-olds are crossed. As a result, the output of adigital circuit will have picked up some tim-ing jitter. Jitter in one or both ofthe signals applied to a phase detector canbe viewed as low-level random phase modu-lation. Within the loop bandwidth, thephase detector steers the VCO so as to op-pose and cancel it, so phase jitter or noise isapplied to the VCO in order to cancel out thejitter added by the divider. This makes theVCO noisier. The noise performance of thehigh-speed CMOS logic normally used inprogrammable dividers is good. For ultra-low-noise dividers, ECL devices are some-times used.

As shown in Fig 10.38A, a program-mable divider consists of a loadablecounter and some control circuitry. Thecounter is designed to count downward.The programmed division factor is loadedinto the counter, and the incoming fre-quency clocks the downward counting.When the count reaches zero, the counteris quickly reloaded with the division factorbefore the next clock edge. The maximumfrequency of operation is limited by theminimum time needed to reliably performthe loading operation. One improvementis to have a circuit to recognize a state afew clock cycles before the zero count isreached, so the reload can be performed ata more leisurely pace during those cycles.This imposes a minimum division ratio,but that is rarely a problem. An outputpulse is given during the loading cycle. Inthis way a frequency is divided by thenumber loaded.

The maximum input frequency that canbe handled is set by the speed of availablelogic. The synchronous reset cycle forcesthe entire divider to be made of equallyfast logic. Just adding a fast prescaler—afixed-ratio divider ahead of the program-mable one—will increase the maximumfrequency that can be used, but it alsoscales up the loop step size. Equal divisionhas to be added to the reference input ofthe phase detector to restore the step size,reducing the phase detector’s operatingfrequency. The loop bandwidth has to bereduced to restore the filtering needed tosuppress reference frequency sidebands.This makes the loop much slower tochange frequency and degrades the noiseperformance.

Variable or Dual-ModulusPrescalers

A plain programmable divider for VHFuse would need to be built from ECLdevices. It would be expensive, hot andpower-hungry. The idea of a fast front endahead of a CMOS programmable dividerwould be perfect if these problems could

Chapter 10.pmd 7/31/2006, 11:34 AM34

Oscillators and Synthesizers 10.35

Fig 10.38—(A) Shows the mechanism ofa programmable frequency divider. (B)shows the function of a dual-modulusprescaler. The counter is reloaded withN when the count reaches 0 or 1,depending on the sequencer action.

be circumvented. Consider a fast divide-by-ten prescaler ahead of a programmabledivider where division by 947 is required.If the main divider is set to divide by 94,the overall division ratio is 940. Theprescaler goes through its cycle 94 timesand the main divider goes through its cycleonce, for every output pulse. If theprescaler is changed to divide by 11 for 7of its cycles for every cycle of the entiredivider system, the overall division ratiois now [(7 × 11) + (87 × 10)] = 947. At thecost of a more elaborate prescaler and theaddition of a slower programmable counterto control it, this prescaler does not multi-ply the step size and avoids all the prob-lems of fixed prescaling.

Fig 10.38B shows the general block dia-gram of a dual-modulus prescaled divider.The down counter controls the modulus ofthe prescaler. The numerical example, justgiven, used decimal arithmetic, althoughbinary is now usual. Each cycle of the sys-tem begins with the last output pulse hav-ing loaded the frequency control word,into both the main divider and theprescaler controller. If the part of the wordloaded into the prescaler controller is notzero, the prescaler is set to divide by 1greater than its normal ratio. Each cycle ofthe prescaler clocks the down counter.Eventually, it reaches zero, and two thingshappen: The counter is designed to freezeat zero (and it will remain frozen until it isnext reloaded) and the prescaler isswitched back to its normal ratio, at whichit will remain until the next reload. Oneway of visualizing this is to think of theprescaler as just being a divider of its nor-mal ratio, but with the ability to “steal” anumber of input pulses controlled by thedata loaded into its companion downcounter. Note that a dual-modulus pre-scaler system has a minimum divisionratio, needed to ensure there are enoughcycles of the prescaler to allow enoughinput pulses to be stolen.

Dual-modulus prescaler ICs are widelyused and widely available. Devices foruse to a few hundred megahertz are cheap,and devices in the 2.5-GHz region arecommonly available. Common prescalerIC division ratio pairs are: 8-9, 10-11, 16-17, 32-33, 64-65 and so on. Many ICscontaining programmable dividers areavailable in versions with and withoutbuilt-in prescaler controllers.

Chapter 10.pmd 7/31/2006, 11:34 AM35

10.36 Chapter 10

Phase DetectorsA phase detector (PSD) produces an

output voltage that depends on the phaserelationship between its two input signals.If two signals, in phase on exactly the samefrequency, are mixed together in a con-ventional diode-ring mixer with a dc-coupled output port, one of the products isdirect current (0 Hertz). If the phase rela-tionship between the signals changes, themixer’s dc output voltage changes. Withboth signals in phase, the output is at its

Fig 10.39—Simple phase detectors: a mixer (A), a sampler (B) and an exclusive-OR gate (C).

most positive; with the signals 180° out ofphase, the output is at its most negative.When the phase difference is 90° (the sig-nals are said to be in quadrature), the out-put is 0 V.

Applying sinusoidal signals to a phasedetector causes the detector’s output volt-age to vary sinusoidally with phase angle,as in Fig 10.39A. This nonlinearity is not aproblem, as the loop is usually arranged torun with phase differences close to 90°.What might seem to be a more serious com-plication is that the detector’s phase-volt-

age characteristic repeats every 180°, not360°. Two possible input phase differencescan therefore produce a given output volt-age. This turns out not to be a problem,because the two identical voltage points lieon opposing slopes on the detector’s out-put-voltage curve. One direction of slopegives positive feedback (making the loopunstable and driving the VCO away fromwhat otherwise would be the lock angle)over to the other slope, to the true and stablelock condition.

MD108, SBL-1 and other diode mixerscan be used as phase detectors, as can ac-tive mixers like the MC1496. Mixer manu-facturers make some parts (Mini-CircuitsRPD-1 and so on) that are specially opti-mized for phase-detector service. All thesedevices can make excellent, low-noisephase detectors but are not commonly usedin ham equipment. A high-speed sample-and-hold circuit, based on a Schottky-diode bridge, can form a very low-noisephase detector and is sometimes used inspecialized instruments. This is just a vari-ant on the basic mixer; it produces a simi-lar result, as shown in Fig 10.39B.

The most commonly used simple phasedetector is just a single exclusive-OR(XOR) logic gate. This circuit gives a logic1 output only when one of its inputs is atlogic 1; if both inputs are the same, theoutput is a logic 0. If inputs A and B in Fig10.39C are almost in phase, the output willbe low most of the time, and its averagefiltered value will be close to the logic 0level. If A and B are almost in oppositephase, the output will be high most of thetime, and its average voltage will be closeto logic 1. This circuit is very similar to themixer. In fact, the internal circuit of ECLXOR gates is the same transistor tree asfound in the MC1496 and similar mixers,with some added level shifting. Like theother simple phase detectors, it produces acyclic output, but because of the square-wave input signals, produces a triangularoutput signal. To achieve this circuit’s fulloutput-voltage range, it’s important thatthe reference and VCO signals appliedoperate at a 50% duty cycle.

Phase-Frequency DetectorsAll the simple phase detectors described

so far are really specialized mixers. If theloop is out of lock and the VCO is far offfrequency, such phase detectors give ahigh-frequency output midway betweenzero and maximum. This provides no in-formation to steer the VCO towards lock,so the loop remains unlocked. Varioussolutions to this problem, such as cruderelaxation oscillators that start up to sweepthe VCO tuning until lock is acquired, orlaboriously adjusted “pretune” systems in

Chapter 10.pmd 7/31/2006, 11:34 AM36

Oscillators and Synthesizers 10.37

which a DAC, driven from the dividercontrol data, is used to coarse tune theVCO to within locking range, have beenused in the past. Many of these solutionshave been superseded, although pretunesystems are still used in synthesizers thatmust change frequency very rapidly.

The phase-frequency detector is theusual solution to lock-acquisition prob-lems. It behaves like a simple phase detec-tor over an extended phase range, but itscharacteristic is not repetitive. Because itsoutput voltage stays high or low, depend-ing on which input is higher in frequency,this PSD can steer a loop towards lock fromanywhere in its tuning range. Fig 10.40shows the internal logic of the phase-fre-quency detector in the CD4046 PLL chip.(The CD4046 also contains an XOR PSD).When the phase of one input leads that ofthe other, one of the output MOSFETs ispulsed on with a duty cycle proportionalto the phase difference. Which MOSFETreceives drive depends on which input is

Fig 10.40—Input signals very far off frequency can confuse a simple phase-detector; a phase-frequency detector solves this problem.

leading. If both inputs are in phase, theoutput will include small pulses due tonoise, but their effects on the average out-put voltage will cancel. If one signal is ata higher frequency than the other, its phasewill lead by an increasing amount, and thedetector’s output will be held close to ei-ther VDD or ground, depending on whichinput signal is higher in frequency. To geta usable voltage output, the MOSFET out-puts can be terminated in a high-value re-sistor to VDD/2, but the CD4046’s outputstage was really designed to drive currentpulses into a capacitive load, with pulsesof one polarity charging the capacitor andpulses of the other discharging it. The ca-pacitor integrates the pulses. Simple phasedetectors are normally used to lock theirinputs in quadrature, but phase-frequencydetectors are used to lock their input sig-nals in phase.

Traditional textbooks and logic-designcourses give extensive coverage to avoid-ing race hazards caused by near-simulta-

neous signals racing through parallel pathsin a structure to control a single output.Avoiding such situations is important inmany circuits because the outcome isstrongly dependent on slight differencesin gate speed. The phase-frequency detec-tor is the one circuit whose entire functiondepends on its built-in ability to make bothits inputs race. Consequently, it’s not easyto home-brew a phase-frequency detectorfrom ordinary logic parts. Fortunately,they are available in IC form, usually com-bined with other functions. The MC4044is an old stand-alone TTL phase-frequencydetector; the MC12040 is an ECL deriva-tive, and the much faster and compatibleMCH 12140 can be used to over 600 MHz.The Hittite HMC403S8G can be used to1.3 GHz. CMOS versions can be found inthe CD4046 PLL chip and in almost allcurrent divider-PLL synthesizer chips.

The race-hazard tendency does causeone problem in phase-frequency detectors:A device’s delays and noise, rather than itsinput signals, control its phase-voltagecharacteristic in the zero-phase-differenceregion. This degrades the loop’s noise per-formance and makes its phase-to-voltagecoefficient uncertain and variable in asmall region, a “dead zone”—unfortu-nately, in the detector’s normal operatingrange! It’s therefore normal to bias opera-tion slightly away from exact phase equal-ity to avoid this problem. Fortunately, thenewer, faster phase detectors like theMCH 12140 and HMC403S8G tend tominimize this “dead zone” problem.

The Loop Compensation AmplifierPhase-locked loops have acquired a

troublesome reputation for a variety ofreasons. In the past, a number of commer-cially made radios have included poorlydesigned synthesizers that produced ex-cessive noise sidebands, or wouldn’t lockreliably. Some un-producible designs havebeen published for home construction thatcould not be made to work. Many experi-menters have toiled over an unstable loop,desperately trying anything to get it to lockstably. So the PLL has earned a shadyreputation. Because of all of this aversiontherapy, very few amateurs are now pre-pared to attempt to build a PLL.

Luckily, things are not as bad as theyused to be. The proliferation of synthesisand PLLs in contemporary equipmenthave lead to a number of excellent inte-grated circuits, as well as fine applicationnotes to support them.

The first task is to take action to ensurethat all previously discussed componentsof the loop are functioning properly andstably before attempting to close the loop.This means that the oscillators, dividers

Chapter 10.pmd 7/31/2006, 11:34 AM37

10.38 Chapter 10

and amplifiers must all be unconditionallystable. Any instability or squegging inthese components must be dealt with priorto attempting to close the loop.

The second task is to produce a closed-loop characteristic that best suits the ap-plication. Herein is where the greatestdifficulty frequently lies. The selection ofthe closed-loop bandwidth and phasemargin or peaking is usually one of greatcompromise and thought. The processfrequently begins with an educated ap-proximation that is then evaluated andmodified. There are many factors to con-sider, including the degree of referencefrequency suppression, switching speed,noise, microphonics and modulation, ifany. Once the optimum loop characteris-tic is established, the next problem is tomaintain it with respect to variations ofthe division resulting from the synthesis,and also gain variations associated withthe nonlinear tuning curve of the VCO.While not always required in amateur ap-plications, loops in sophisticated synthe-sizers frequently employ programmablemultiplying DACs to maintain constantloop bandwidths and phase margin. In anexample later in this chapter, we will ac-tually describe the process for establish-ing the loop bandwidth for a simplesynthesizer.

The third task is to design a loop com-pensation amplifier (filter) that will en-sure stable operation when the loop isclosed. The design of this compensationnetwork or filter requires knowledge of theVCO gain and linearity, the phase-detec-tor gain, and the effective division ratioand its variation in the loop. The foregoingrequirements imply the necessity for mea-surement and calculation to have any rea-sonable hope of producing a successfuloutcome.

Note carefully that we are designing aloop. Trial and error, intuitive componentchoice, or “reusing” a loop amplifier de-sign from a different synthesizer may leadto a low probability of success. All oscil-lators are loops, and any loop will oscil-late when certain conditions are met. Lookagain at the RC phase-shift oscillator fromwhich the “Pierce” crystal oscillator isderived (Fig 10.29A). Our PLL is a loop,containing an amplifier and a number ofRC sections. It has all the parts needed tomake an oscillator. If a loop is unstableand oscillates, there will be an oscillatoryvoltage superimposed on the VCO tuningvoltage. This will produce massive un-wanted FM on the output of the PLL,which is absolutely undesirable.

Before you turn to a less demandingchapter, consider this: It’s one of nature’sjokes that easy procedures often hide be-

hind a terror-inducing facade. The mathyou need to design a good PLL may lookweird when you see it for the first time andseem to involve some obviously impos-sible concepts, but it ends in a very simpleprocedure that allows you to calculate theresponse and stability of a loop. Profes-sional designers must handle these thingsdaily, and because they like differentialequations no more than anyone else does,they use a graphical method based on thework of the mathematician, Laplace, togenerate PLL designs. (We’ll leave proofsin the textbooks, where they belong! Inci-dentally, if you can get the math requiredfor PLL design under control, as a free giftyou will also be ready to do modern filtersynthesis and design beam antennas, sincethe math they require is essentially thesame.)

The easy solutions to PLL problems canonly be performed during the designphase. We must deliberately design loopswith sufficient stability safety margins sothat all foreseeable variations in compo-nent tolerances, from part-to-part, overtemperature and across the tuning range,cannot take the loop anywhere near thethreshold of oscillation. More than this, itis important to have adequate stabilitymargin because loops with lesser marginswill exhibit amplified phase-noise side-bands, and that is another PLL problemthat can be designed out.

At the beginning of this chapter (fol-lowing the text cite of Fig 10.4A), the cri-terion for stability/oscillation of a loopwas mentioned: Barkhausen’s for oscilla-tion or Nyquist’s for stability. This timeHarry Nyquist is our hero.

A PLL is a negative feedback system,with the feedback opposing the input, thisopposition or inversion around the loopamounts to a 180° phase shift. This is thefrequency-independent phase shift roundthe loop, but there are also frequency-dependent shifts that will add in. Thesewill inevitably give an increasing, laggingphase with increasing frequency. Eventu-ally an extra 180° phase shift will haveoccurred, giving a total of 360° around theloop at some frequency. We are in troubleif the gain around the loop has not droppedbelow unity (0 dB) by this frequency.

Note that we are not concerned only withthe loop operating frequency. Consider a30-MHz low-pass filter passing a 21-MHzsignal. Just because there is no signal at30 MHz, our filter is no less a 30-MHz cir-cuit. Here we use the concept of frequencyto describe “what would happen if a signalwas applied at that frequency.”

The next sections show us how to usethe graphical concepts of poles and zerosto calculate the gain and phase response

around a loop. The graphical concepts arequite useful in that they provide insightinto the effect of various componentchoices on the loop performance.

Poles and ZerosLet’s define some terms first. We all

think of frequency as stretching from zeroto infinity, but let’s imagine that additionalnumbers could be used. This is pure imagi-nation—we can’t make signals at complexvalues (with real and imaginary compo-nents) of hertz—but if we try using com-plex numbers for frequency, some“impossible” things happen. For example,take a simple RC low-pass network shownin Fig 10.41. The frequency response ofsuch a network is well known, but its phaseresponse is not so well known. The equa-tion shown for the transfer function, the“gain” of this simple circuit, is pretty stan-dard, although the j is included to make itcomplete and accurately describe the out-put in phase as well as magnitude (j is animpossible number, the square root of –1,usually called an imaginary number andused to represent a 90° phase shift).

Although the equation shown inFig 10.41 was constructed to show the be-havior of the circuit at real-world frequen-cies, there is nothing to stop us from usingit to explore how the circuit would behaveat impossible frequencies, just out of curi-osity. At one such frequency, the circuitgoes crazy. At f = –j 2πRC, the denomina-tor of the equation is zero, and the gain isinfinite! Infinite gain is a pretty amazingthing to achieve with a passive circuit—but because this can only happen at an im-possible frequency, it cannot happen in thereal world. This frequency is equal to thenetwork’s –3 dB frequency multiplied byj; it is called a pole. Other circuits, whichproduce real amplitude responses that startto rise with frequency, act similarly, buttheir crazy effect is called a zero—animaginary frequency at which gain goes tozero. The frequency of a circuit’s zero isagain j times its 3-dB frequency (+3 dB inthis case).

These things are clearly impossible, butwe can map all the poles and zeros of acomplex circuit and look at the patterns todetermine if the circuit possesses unwantedgain and/or phase bumps across a frequencyrange of interest. Poles are associated with3-dB roll-off frequencies and 45° phaselags, zeros are associated with 3-dB “roll-up” frequencies and 45° phase leads. Wecan plot the poles and zeros on a two-di-mensional chart of real and imaginary fre-quency. (The names of this chart and itsaxes are results of its origins in Laplacetransforms, which we don’t need to touchin this discussion.) The traditional names

Chapter 10.pmd 7/31/2006, 11:34 AM38

Oscillators and Synthesizers 10.39

Fig 10.41—A simple RC filter is a “pole.”

for the axes are used, but we can add labelswith clearer meanings.

What is a Pole?—A pole is associatedwith a bend in a frequency response plotwhere attenuation with increasing fre-quency increases by 6 dB per octave (20 dBper decade; an octave is a 2:1 frequencyratio, a decade is a 10:1 frequency ratio).

There are four ways to identify the ex-istence and frequency of a pole:

1. A downward bend in a gain vs fre-quency plot, the pole is at the –3 dB pointfor a single pole. If the bend is more than6 dB/octave, there must be multiple polesat this frequency.

2. A 90° change in a phase vs frequencyplot, where lag increases with frequency.The pole is at the point of 45° added lag.Multiple poles will add their lags, asabove.

3. On a circuit diagram, a single pole

looks like a simple RC low-pass filter. Thepole is at the –3 dB frequency (1/(2πRC)Hz). Any other circuit that gives the sameresponse will produce a pole at the samefrequency.

4. In an equation for the transfer func-tion of a circuit, a pole is a theoretical valueof frequency, which would make the equa-tion predict infinite gain. This is clearly im-possible, but as the value of frequency willeither be absolute zero, or will have animaginary component, it is impossible tomake a signal at a pole frequency.

What is a Zero?—A zero is the comple-ment of a pole. Each zero is associated withan upward bend of 6 dB per octave in a gainvs frequency plot. The zero is at the +3 dBpoint. Each zero is associated with a transi-tion on a phase-versus-frequency plot thatreduces the lag by 90°. The zero is at the 45°point of this S-shaped transition.

In math, it is a frequency at which thetransfer-function equation of a circuit pre-dicts zero gain. This is not impossible inreal life (unlike the pole), so zeros can befound at real-number frequencies as wellas complex-number frequencies.

On a circuit diagram, a pure zero wouldneed gain that increases with frequencyforevermore above the zero frequency.This implies active circuitry that wouldinevitably run out of gain at some fre-quency, which implies one or more polesup there. In real-world circuits, zeros areusually not found making gain go up, butrather in conjunction with a pole, giving again slope between two frequencies andflat gain beyond them. Real-world zerosare only found chaperoned by a greater orequal number of poles.

Consider a classic RC high-pass filter.The gain increases at 6 dB per octave from0 Hz (so there must be a zero at 0 Hz) andthen levels off at 1/(2πRC) Hz. This level-ing off is really a pole, it adds a 6 dB peroctave roll off to cancel the roll-up of thezero.

Poles and Zeros in the Loop Ampli-fier—The loop-amplifier circuit used in theexample loop has a blocking capacitor in itsfeedback path. This means there is no dcfeedback. At higher frequencies the reac-tance of this capacitor falls, increasing thefeedback and so reducing the gain. This isan integrator. The gain is immense at 0 Hz(dc) and falls at 6 dB per octave. This pointsto a pole at 0 Hz. This is true whatever thevalue of the series resistance feeding thesignal into the amplifier inverting input andwhatever the value of the feedback capaci-tor. The values of these components scalethe gain rather than shape it. The shape ofthe integrator gain is fixed at –6 dB peroctave, but these components allow us tomove it to achieve some wanted gain atsome wanted frequency.

At high frequencies, the feedback ca-pacitor in an ordinary integrator will havevery low reactance, giving the circuit verylow gain. In our loop amplifier, there is aresistor in series with the capacitor. Thislimits how low the gain can go, in otherwords the integrator’s downward (withincreasing frequency) slope is leveled off.This resistor and capacitor make a zero,and its +6 dB per octave slope cancels the–6 dB per octave slope created by the poleat 0 Hz.

Recipe for A PLL Pole-Zero DiagramWe can choose a well-tried loop-am-

plifier circuit and calculate values forits components to make it suit our loop.Fig 10.42 shows a common synthesizerloop arrangement. The op amp operates asan integrator, with R2 added to level off its

Chapter 10.pmd 7/31/2006, 11:34 AM39

10.40 Chapter 10

Fig 10.44—The loop’s resulting pole-zero “constellation.”

Fig 10.42—A common loop-amplifier/filter arrangement.

Fig 10.43—Loop-amplifier detail.

falling gain. An integrator converts a dcvoltage at its input to a ramping voltage atits output. Greater input voltages yieldfaster ramps. Reversing the input polarityreverses the direction of the ramp.

The system’s phase-frequency detector(connected to work in the right sense)steers the integrator to ramp in the direc-tion that tunes the VCO toward lock. Asthe VCO approaches lock, the phase de-tector reduces the drive to the integrator,and the ramping output slows and settleson the right voltage to give the exact,locked output frequency. Once lock isachieved, the phase detector outputs shortpulses that “nudge” the integrator to keepthe divided VCO frequency exactly lockedin phase with the reference. Now we’lltake a look around the entire loop and findthe poles and zeros of the circuits.

The integrator produces a –20 dB perdecade roll off from 0 Hz, so it has a pole at0 Hz. It also includes R2 to cancel this slope,which is the same as adding a rising slopethat exactly offsets the falling one. Thisimplies a zero at f = 1/(2πC1R2) Hz, as Fig10.43 shows. R3 and C2 make another poleat f = 1/(2πC2R3) Hz. A VCO usually in-cludes a series resistor that conveys thecontrol voltage to the tuning diode, whichalso is loaded by various capacitors. Thiscreates another pole.

The VCO generates frequency, whilethe phase detector responds to phase.Phase is the integral of frequency, so to-gether they act as another integrator andadd another pole at 0 Hz.

What About the Frequency Divi-der?—The frequency divider is like asimple attenuator in its effect on loop re-sponse. Imagine a signal generator that isswitched between 10 and 11 MHz on alter-nate Tuesdays. Imagine that we divide itsoutput by 10. The output of the dividerwill change between 1 and 1.1 MHz, still

on alternate Tuesdays. The divider dividesthe deviation of the frequency (or phase)modulated signal, but it cannot affect themodulating frequency.

A possible test signal passing around theexample loop is in the form of frequencymodulation as it passes through the di-vider, so it is simply attenuated. A divideby N circuit will give 20 log10 (N) dB ofattenuation (reduction of loop gain). Thiscompletes the loop. We can plot all thisinformation on one s-plane, Fig 10.44, aplot sometimes referred to as a PLLsystem’s pole-zero constellation.

Open-Loop Gain and PhaseJust putting together the characteristics

of the blocks around the loop, withoutallowing for the loop itself, gives us thesystem’s open-loop characteristic, whichis all we really need.

What Is Stability?—Here “open-loopresponse” means the gain around the loopthat would be experienced by a signal atsome frequency if such a signal were in-serted. We do not insert such signals in theactual circuit, but the concept of fre-quency-dependent gain is still valid. Weneed to ensure that there is insufficientgain, at all frequencies, to ensure that theloop cannot create a signal and begin os-cillating. More than this, we want a goodsafety margin to allow for componentvariations and because loops that are closeto instability perform poorly.

The loop gain and phase are doing inter-esting things on our plots at frequencies inthe AF part of the spectrum. This does notmean that there is visible operation at thosefrequencies. Imagine a bad, unstable loop.It will have a little too much loop gain atsome frequency, and unavoidable noise willbuild up until a strong signal is created. Theamplitude will increase until non-linearities limit it. A ’scope view of thetuning voltage input to the VCO will showa big signal, often in the hertz to tens ofkilohertz region, often large enough to drivethe loop amplifier to the limits of its outputswing, close to its supply voltages. As thisbig signal is applied to the tuning voltageinput to the VCO, it will modulate the VCOfrequency across a wide range. The outputwill look like that of a sweep generator ona spectrum analyzer. What is wrong? Howcan we fix this loop? Well, the problem maybe excessive loop gain, or improper looptime constants. Rather than work directlywith the loop time constants, it is far easierto work with the pole and zero frequenciesof the loop response. It’s just a differentview of the same things.

Now imagine a good, stable loop, withan adequate stability safety margin. Therewill be some activity around the loop: the

Chapter 10.pmd 7/31/2006, 11:34 AM40

Oscillators and Synthesizers 10.41

Fig 10.45—Calculating loop gain and phase characteristics from a pole-zerodiagram.

Fig 10.46—Pole-zero frequency-response equations capable of handling up to four poles and one zero.

⎥⎥

⎦

⎤

⎢⎢

⎣

⎡×

+++++

+++++=

20

gain freq unity10

12Z2f2

4P2f2

3P2f2

2P2f2

1P

2f2Z12

4P12

3P12

2P12

1Plog 20(dB) gain loop Open

⎟⎟⎠

⎞⎜⎜⎝

⎛⎟⎟⎠

⎞⎜⎜⎝

⎛⎟⎟⎠

⎞⎜⎜⎝

⎛⎟⎠⎞

⎜⎝⎛ −−−−−−−=

4P

f1tan3P

f1tan2P

f1tanZ

f1tan(lead) Phase

f is the frequency of the point to be characterized. P1, P2, P3, P4, Z are the frequencies of the poles and zero (all in the same units!)

PSD (phase detector) will demodulate thephase noise of the VCO and feed the de-modulated noise through the loop ampli-fier in such a way as to cancel the phasenoise. This is how a good PLL should giveless phase noise than the VCO alone wouldsuggest. In a good loop, this noise will besuch a small voltage that a ’scope will notshow it. In fact, connecting test equipmentin an attempt to measure it can usually addmore noise than is normally there.

Finally, imagine a poor loop that is onlyjust stable. With an inadequate safetymargin, the action will be like that of a Qmultiplier or a regenerative receiver closeto the point of oscillation: There will be anamplified noise peak at some frequency.This spoils the effect of the phase detectortrying to combat the VCO phase noise andgives the opposite effect. The output spec-trum will show prominent bumps of exag-gerated phase noise, as the excess noisefrequency modulates the VCO.

Now, let’s use the pole-zero diagram asa graphic tool to find the system open-loopgain and phase. As we do so, we need to

keep in mind that the frequency we havebeen discussing in designing a loop re-sponse is the frequency of a theoretical testsignal passing around the loop. In a realPLL, the loop signal exists in two forms: Assinusoidal voltage between the output ofthe phase detector and the input of the VCO,and as a sinusoidal modulation of the VCOfrequency in the remainder of the loop.

With our loop’s pole-zero diagram inhand, we can pick a frequency at which wewant to know the system’s open-loop gainand phase. We plot this value on thegraph’s vertical (Real Frequency) axis anddraw lines between it and each of the polesand zeros, as shown in Fig 10.45. Next,we measure the lengths of the lines and the“angles of elevation” of the lines. The loopgain is proportional to the product of thelengths of all the lines to zeros divided bythe product of all the lengths of the lines tothe poles. The phase shift around the loop(lagging phase equates to a positive phaseshift) is equal to the sum of the pole anglesminus the sum of the zero angles. We canrepeat this calculation for a number of dif-

ferent frequencies and draw graphs of theloop gain and phase versus frequency.

All the lines to poles and zeros are hypot-enuses of right triangles, so we can usePythagoras’s rule and the tangents of anglesto eliminate the need for scale drawings.Much tedious calculation is involved be-cause we need to repeat the whole businessfor each point on our open-loop responseplots. This much tedious calculation is anideal application for a computer.

The procedure we’ve followed so fargives only proportional changes in loopgain, so we need to calculate the loop gain’sabsolute value at some (chosen to be easy)frequency and then relate everything tothis. Let’s choose 1 Hz as our reference.(Note that it’s usual to express angles inradians, not degrees, in these calculationsand that this normally renders frequency inpeculiar units of radians per second. Wecan keep frequencies in hertz if we remem-ber to include factors of 2π in the rightplaces. A frequency of 1 Hz = 2π radians/second, because 2π radians = 360°.)

We must then calculate the loop’s pro-portional gain at 1 Hz from the pole-zerodiagram, so that the constant of proportioncan be found. For starters, we need a rea-sonable estimate of the VCO’s voltage-to-frequency gain—how much it changesfrequency per unit change of tuning volt-age. As we are primarily interested in sta-bility, we can just take this number as theslope, in Hertz per volt, at the steepest partof voltage-versus-frequency tuning charac-teristic, which is usually at the low-fre-quency end of the VCO tuning range. (Youcan characterize a VCO’s voltage-versus-frequency gain by varying the bias on itstuning diode with an adjustable power sup-ply and measuring its tuning characteristicwith a voltmeter and frequency counter.)

The loop divide-by-N stage divides ourtheoretical modulation—the tuning correc-tions provided through the phase detector,loop amplifier and the VCO tuning diode—by its programmed ratio. The worst casefor stability occurs at the divider’s lowestN value, where the divider’s “attenuation”

Chapter 10.pmd 7/31/2006, 11:34 AM41

10.42 Chapter 10

is least. The divider’s voltage-versus-fre-quency gain is therefore 1/N, which, indecibels, equates to –20 log (N).

The change from frequency to phase hasa voltage gain of one at the frequency of1 radian/second, 1/(2π), which equates to–16 dB, at 1 Hz. The phase detector willhave a specified “gain” in volts per radian.To finish off, we then calculate the gain ofthe loop amplifier, including its feedbacknetwork, at 1 Hz.

There is no need even to draw the pole-zero diagram. Fig 10.46 gives the equa-tions needed to compute the gain and phaseof a system with up to four poles and onezero. They can be extended to moresingularities and put into a simple com-puter program. Alternatively, you can typethem into your favorite spreadsheet andget printed plots. The computationalpower needed is trivial, and a listing to runthese equations on a Hewlett-PackardHP11C (or similar pocket calculator, usingRPN) is available from ARRL HQ for anSASE. (Contact the Technical DepartmentSecretary and ask for the 1995 HandbookPLL design program.)

Fig 10.47 shows the sort of gain- andphase-versus-frequency plots obtainedfrom a “recipe” loop design. We want toknow where the loop’s phase shift equals,or becomes more negative than, –180°.Added to the –180° shift inherent in theloop’s feedback (the polarity of whichmust be negative to ensure that the phasedetector drives the VCO toward lock in-stead of away from it), this will be the pointat which the loop itself oscillates—if anyloop gain remains at this point. The gameis to position the poles and zero, and setthe unit frequency gain, so that the loop

Fig 10.47—An open-loop gain/phaseplot.

gain falls to below 0 dB before the –180°line is crossed.

At the low-frequency end of the Fig 10.47plots, the two poles have had their effect,so the gain falls at 40 dB/decade, and thephase remains just infinitesimally on thesafe side of –180°. The next influence isthe zero, which throttles the gain’s fallback to 20 dB/decade and peels the phaseline away from the –180° line. The zerowould eventually bring the phase back to90° of lag, but the next pole bends it backbefore that and returns the gain slope backto a fall of 20 dB/decade. The last pole,that attributable to the VCO, pushes thephase over the –180° line.

It’s essential that there not be a polebetween the 0-Hz pair and the zero, or elsethe phase will cross the line early. In thisexample, the R3-C2 pole and the zero doall the work in setting this critical portionof the loop’s response. Their frequencyspacing should create a phase bump of 30to 45°, and their particular frequency po-sitions are constrained as a compromisebetween sufficient loop bandwidth andsufficient suppression of reference-fre-quency sidebands. We know the frequencyof the loop amplifier zero (that contrib-uted by R1 and C1), so all we need to donow is design the loop amplifier to exhibita frequency response such that the open-loop gain passes through 0 dB at a fre-quency close to that of the phase-plotbump.

This loop-design description may havebeen hard to follow, but a loop amplifiercan be designed in under 30 minutes witha pocket calculator and a little practice.The high output impedance of commonlyused CMOS phase detectors favors loopamplifiers based on FET-input op ampsand allows the use of high-impedance RCnetworks (large capacitors can thereforebe avoided in their design). LF356 andTL071 op amps have proven successful inloop-amplifier service and have noisecharacteristics suited to this environment.

To sum up, the recipe gives a provenpole zero pattern (listed in order of in-creasing frequency):

1. Two poles at 0 Hz (one is the integra-tor, the other is implicit in all PLs)

2. A zero controlled by the RC series inthe loop amplifier feedback path

3. A simple RC low-pass pole4. A second RC low-pass pole formed

by the resistor driving the capacitive loadof the tuning voltage inputs of the VCO

To design a loop: Design the VCO,divider, PSD and find their coefficients toget the loop gain at unit frequency (1 Hz),then, keeping the poles and zeros in theabove order, move them about until you

get the same 45° phase bump at a fre-quency close to your desired bandwidth.Work out how much loop-amplifier unit-frequency gain is needed to shift the gain-frequency-frequency plot so that 0-dBloop gain occurs at the center of the phasebump. Then calculate R and C values toposition the poles and zero and the feed-back-RC values to get the required loopamplifier gain at 1 Hz.

To Cheat—You can scale the exampleloop to other frequencies: Just take thereactance values at the zero and pole fre-quencies and use them to scale the compo-nents, you get the nice phase plot bump ata scaled frequency. Even so, you still mustdo the loop-gain design.

Don’t forget to do this for your lowestdivision factor, as this is usually the leaststable condition because there is lessattenuation. Also, VCOs are usually mostsensitive at the low end of their tuningrange.

Noise in Phase-Locked LoopsDifferences in Q usually make the

phase-noise sidebands of a loop’s refer-ence oscillator much smaller than those ofthe VCO. Within its loop bandwidth, a PLLacts to correct the phase-noise componentsof its VCO and impose those of the refer-ence. Dividing the reference oscillatorto produce the reference signal applied tothe phase detector also divides the devia-tion of the reference oscillator’s phase-noise sidebands, translating to a 20-dBreduction in phase noise per decade of di-vision, a factor of 20 log (M) dB, where Mis the reference divisor. Offsetting this,within its loop bandwidth the PLL acts asa frequency multiplier, and this multipliesthe deviation, again by 20 dB per decade,a factor of 20 log (N) dB, where N is theloop divider’s divisor. Overall, the refer-ence sidebands are increased by 20 log(N/M) dB. Noise in the dividers is, in ef-fect, present at the phase detector input,and so is increased by 20 log (N) dB. Simi-larly, op-amp noise can be calculated intoan equivalent phase value at the input tothe phase detector, and this can be in-creased by 20 log (N) to arrive at the effectit has on the output.

Phase noise can be introduced into aPLL by other means. Any amplifier stagesbetween the VCO and the circuits that fol-low it (such as the loop divider) will con-tribute some noise, as will microphoniceffects in loop- and reference-filter com-ponents (such as those due to the piezo-electric properties of ceramic capacitorsand the crystal filters sometimes used forreference-oscillator filtering). Noise onthe power supply to the system’s activecomponents can modulate the loop. The

Chapter 10.pmd 7/31/2006, 11:34 AM42

Oscillators and Synthesizers 10.43

Fig 10.48—A PLL’s open- and closed-loop phase-noise characteristics. The noisebumps at the crossover between open- and closed-loop characteristics are typicalof PLLs; the severity of the bumps reflects the quality of the system’s design.

fundamental and harmonics of thesystem’s ac line supply can be coupled intothe VCO directly or by means of groundloops.

Fig 10.48 shows the general shape ofthe PLL’s phase noise output. The dashedcurve shows the VCO’s noise performancewhen unlocked; the solid curve shows howmuch locking the VCO to a cleaner refer-ence improves its noise performance. Thetwo noise bumps are a classic characteris-tic of a phase-locked loop. If the loop ispoorly designed and has a low stabilitymargin, the bumps may be exaggerated—a sign of noise amplification due to anoverly peaky loop response.

Exaggerated noise bumps can also oc-cur if the loop bandwidth is less than opti-mum. Increasing the loop bandwidth insuch a case would widen the band overwhich the PLL acts, allowing it to do abetter job of purifying the VCO—but thismight cause other problems in a loop that’sdeliberately bandwidth-limited for bettersuppression of reference-frequency side-bands. Immense loop bandwidth is notdesirable, either: Farther away from thecarrier, the VCO may be so quiet on itsown that widening the loop would make itworse.

A BASIC DESIGN EXAMPLEIn this section, we will explore the ap-

plication of the design principles previ-ously covered, as well as some of thetradeoffs required in a practical design.We will also cover measurement tech-niques designed to give the builder confi-dence that the loop design goals have beenmet. Finally, we will cover some commontroubleshooting problems.

As our design example, a synthesizedlocal oscillator chain for a 10-GHz Trans-verter will be considered. Fig 10.49 is asimplified block diagram of a 10-GHzconverter. This example was chosen as itrepresents a departure from the traditionalmulti-stage multiply and filter approach.It permits realization of the oscillator sys-tem with two very simple loops and mini-mal RF hardware. It is also representativeof what is achievable with current hard-

Fig 10.49—A simplified block diagram of a local oscillator, for a 10-GHz converter.

ware, and can fit in a space of 2 to 3 squareinches. This example is intended as a vehi-cle to explore the loop design aspects andis not offered as a “construction project.”The additional detail required would bebeyond the scope of this chapter.

In this example, two synthesized fre-quencies, 10 GHz and 340 MHz, are re-quired. Since 10.368 GHz is one of thepopular traffic frequencies, we will mixthis with the 10-GHz LO to produce an IFof 368 MHz. The 368-MHz IF signal willbe subsequently mixed with a 340-MHzLO to produce a 28-MHz final IF, whichcan be fed into the 10-meter input of anyamateur transceiver. We will focus ourattention on the design of the 10-GHz syn-thesizer only. Once this is done, the sameprinciples may be applied to the 340-MHzsection. Our goal is to attempt to design alow-noise LO system (this implies mini-mum division) with a loop reference oscil-lator that is an integer multiple of 10 MHz.Using this technique will allow the en-tire system to be locked at a later time to a10-MHz standard for precise frequencycontrol.

Ideally, we would like to start with alow-noise 100-MHz crystal reference and a10-GHz oscillator divided by 100. It ishere that we are already confronted withour first practical design tradeoff. We areconsidering using a line of microwave inte-grated circuits made by Hittite Microwave.These include a selection of prescalers

Chapter 10.pmd 7/31/2006, 11:34 AM43

10.44 Chapter 10

operating to 12 GHz, a 5-bit counter that willrun to 2.2 GHz and a phase/frequency detec-tor that will run up to 1.3 GHz. The 5-bitcounter presents the first problem. Our idealscheme would be to use a divide-by-4prescaler and enter the 5-bit counter at2.5 GHz, subsequently dividing by 25 to100 MHz. The problem is that the 5-bitcounter is only rated to 2.2 GHz and not2.5 GHz. This forces us to use a divide-by-8 prescaler and enter the 5-bit divider at1.25 GHz. This is well in the range of thedivider, but we can no longer use an integerdivision (ie, 1.25 GHz/12.5 = 100 MHz) toget to 100 MHz. The next easiest option is toreduce the reference frequency to 50 MHzand to let the 5-bit divider run as a divide-by-25, giving us a total division ratio of 200.

Having already faced our first designtradeoff, a number of additional aspectsmust be considered to minimize the con-flicts in the design. First, this is a “static”synthesizer—That is, it will not be requiredto change frequency during operation. As aresult, switching time considerations areirrelevant, as are the implications that theswitching time would have had on the loopbandwidth. The effects of variable divisionratios are also eliminated, as well as anyproblems associated with nonlinear tuningof the VCOs. Second, the reference fre-quencies (50 MHz and 10 MHz) are verylarge with respect to any practically desir-able loop bandwidths. This makes the re-quirement of the loop filter to eliminatereference sidebands quite easy to achieve.In fact, reference sideband suppression willmore likely be a function of board layoutand shielding effectiveness rather than sup-pression in the loop filter. This also allowsplacement of the reference suppressionpoles far out in the pole zero constellation

and well outside the loop bandwidth (about10 × BW3 of the loop). This placement ofthe reference suppression poles will giveus the option of using a “type 2, 2nd order”loop approximation, which will greatlysimplify the calculation process. Third,based on all of the foregoing tradeoffs, theloop bandwidth can be chosen almost ex-clusively on the basis of noise performance.

Before we can proceed with the loopcalculations we need some additional in-formation: specifically VCO gain, VCOnoise performance, divider noise perfor-mance, phase detector gain, phase detectornoise performance and finally referencenoise performance. The phase detector anddivider information is available from speci-fication sheets. Now we need to select the50-MHz reference and the 10-GHz VCO.An excellent choice for a low noisereference is the one described by JohnStephensen on page 13 in Nov/Dec 1999QEX. The noise performance of this VCXOis in the order of –160 dBc/Hz at 10-kHzoffset at the fundamental frequency. For the10-GHz VCO, there are many possibilities,including cavity oscillators, YIGoscillators and dielectric resonator oscilla-tors. Always looking for parts that areeasily available, useable and economical,salvaging a dielectric resonator from aKu band LNB (see Jan-Feb 2002 QEX,p 3) appears promising. These high-Q os-cillators can be fitted with a varactor andtuned over a limited range with good re-sults. The tuning sensitivity of our dielec-tric resonator VCO is about 10 MHz pervolt and the phase noise at 10-kHz offset is–87dBc/Hz. We are now in a position toplot the phase noise of the reference, thedividers, the division effect (46 dB) and theVCO. The phase noise plots are shown in

Fig 10.50—Phase noise plots of the 10-GHz transverterdesign example.

Fig 10.51—Closed-loop frequency response of a second-order loop as a function of the loop’s natural frequency anddamping factor (ωωωωωn).

Fig 10.50. The noise of the reference isuniformly about 25 dB below the dividerand phase detector noise floor. This meansthat the noise of the phase detector anddividers will be the dominant contributionwithin the loop bandwidth. The cross overpoint is at approximately 9 kHz. At fre-quencies lower than 9 kHz, the loop willreduce the noise of the VCO (eg, at 1 kHzthe reduction will be about 20 dB). At fre-quencies above 9 kHz, the natural noise rolloff of the VCO will dominate. At this point,we can also see the effect of having chosena reference frequency of 50 MHz. Had webeen able to use a 100-MHz reference in-stead of 50 MHz and reduce the divisionfactor from 200 to 100, we could have putthe loop bandwidth at about 30 kHz andpicked up an additional 6-dB improvementin close-in noise performance. Neverthe-less, even with the 6-dB penalty, this willbe quite a respectable 10-GHz LO. It shouldbe apparent that a fair amount of thoughtand effort was required to be able to deter-mine an applicable bandwidth for this loop.

Now that we have some estimate of thenoise performance and the required loopbandwidth, it is time to design the loop.The second consideration cited above en-abling the application of the “type-2, 2ndorder” approximation is now going topay off. The math for this loop is reallyquite simple. There are two main vari-ables, the “natural frequency” of the loop,ωn, (omega-n in radians/second) and the“damping factor” D (delta is dimension-less). The closed-loop bandwidth of theloop is a function of ωn and D. Fig 10.51shows loop response as a function of ωnand D. Values of D less than 0.5 are notdesirable, as they tend towards instabilityand poor phase margin. A value for D of

Chapter 10.pmd 7/31/2006, 11:34 AM44

Oscillators and Synthesizers 10.45

0.707 is referred to as “critically damped”and is favored in applications where set-tling time is critical (see Gardner). For ourapplication, we will choose a D of 1.0. Thiswill give acceptable phase margin andfurther simplify the math. For a D of 1, ωn= (6.28 × BW3) / 2.48. Substituting the9-kHz loop bandwidth for BW3 yields22790 radians per second for ωn. Our loopfilter will take the basic form shown inFig 10.43. We can compute values for R1,R2 and C1 as follows.

)1CN/()KvcoKpd(1R 2n ×ω××=

where Kpd is the phase detector gain,1/6.28 volts/radian in this case, Kvco isthe VCO gain in radians/Hz, 6.28×107

radians/volt in this case, N is the divisionfactor, 200 in this case. C1 is the feedbackcapacitor value in Farads. To proceed, weneed to start with an estimate for C1. Prac-tically speaking, since odd values of ca-pacitors are more difficult to obtain, let ustry a value for C1 of 0.01 μF.

)1C/(D22R n ×ω×=

R1 computes as 9626 Ω and R2 is8775 Ω. Since we will be using a phase/frequency detector with differential out-puts, we will have to modify the circuit ofFig 10.43 to become a differential ampli-fier. We will also add a passive input filterto keep the inputs of the op amp from be-ing stressed by the very fast and shortpulses emanating from the phase/fre-quency detector. We will also add a pas-sive “hash filter” at the output of the opamp to limit the amount of out of bandnoise delivered to the VCO tuning port.Both of these filters will be designed for acutoff frequency of 90 kHz (ie,10 timesthe 9-kHz closed-loop bandwidth, as citedabove in the second consideration. Both ofthese filters will also aid in the rejection ofreference frequency sidebands. For theinput filter, we simply divide the inputresistor in half and add a capacitor toground. The value of this capacitor is de-termined as follows:

pF 7352C

)1R3BW8.62/(42C

=

××=

Using 680 or 750 pF will be adequate.The output filter is also quite simple, withone minor stipulation. Op Amps can exhi-bit stability problems when asked to drivea capacitive load through too low a valueof resistor. One very safe way around thisis not to use a value for R3 lower than therecommended minimum impedance forfull output of the amplifier. For manyamplifiers, this value is around 600 Ω. Forthe output filter, we compute C3 as fol-

Fig 10.52—A complete loop-compensation amplifier for the design example10-GHz transverter.

lows, for R3 = 600 Ω:

pF 29503C

)3BW8.623R/(13C

=

××=

Using 3000 pF will be adequate. Thiscompletes the computations for the loopcompensation amplifier. The completeamplifier is shown in Fig 10.52.

MEASUREMENTSOne of the first things we need to mea-

sure when designing a PLL is the VCOgain. The tools we will need include a volt-meter, some kind of frequency measuringdevice like a receiver or frequency counterand a source of clean variable DC voltage.The setup in Fig 10.53, containing one ormore 9-V batteries and a 10-turn, 10-kΩpot will due nicely. One simply varies thevoltage some amount and then records theassociated frequency change of the VCO.The gain of the VCO is then delta F overdelta V. The phase detector gain constantcan usually be found on the specificationsheets of the components selected.

Measuring closed loop bandwidth isslightly more complicated. The way it is

Fig 10.53—Clean, variable d-c voltagesource used to measure VCO gain in aPLL design.

commonly done in the laboratory is to re-place the reference oscillator with aDCFM-able (ie, a signal generator whoseFM port is DC coupled) signal generatorand then feed the tracking generator out-put of a low-frequency spectrum analyzerinto the DCFM-able generator while ob-serving the spectrum of the tuning voltageon the spectrum analyzer (see Fig 10.54).While this approach is quite straightfor-ward, few amateurs have access to or canafford the test equipment to do this. Todayhowever, thanks to the PC sound card-based spectrum analyzer and tracking gen-erator programs (software by Interflex forexample), amateurs can measure theclosed-loop response of loops that are lessthan 20 kHz in bandwidth for significantlyless than a king’s ransom in test equip-ment! Here’s how.

One approach is to build the referenceoscillator with some “built-in test equip-ment” (BITE) already in the design. ThisBITE takes the form of some means ofDCFM-ing the reference oscillator. Thisis one of the things that make JohnStephenson’s oscillator (previously men-tioned) attractive. The oscillator includes avaractor input for the purpose of phaselocking it to a high-stability low-frequencystandard oscillator. This same varactor tun-ing input takes the place of the aforemen-tioned DCFM-able signal generator. Nowall we need to complete our test setup issome batteries. First, we need to make surethe input of our sound card is AC coupled(when connected to the VCO tuning port,there will be a DC component present andthe sound card may not like this) and thatthe AC coupling is flat down to at least 10%of the natural frequency of the loop. If thesound card is not AC coupled, a capacitorwill be needed at the input to block the DCon the tuning voltage.

Chapter 10.pmd 7/31/2006, 11:34 AM45

10.46 Chapter 10

Fig 10.54—Common laboratory setup for measuring the closed-loop bandwidthof loops.

The next step is to determine the tuningsensitivity of the reference oscillator. Thisis done in the same manner as was done forthe VCO described above. Once this hasbeen established, we will set the audiooscillator for an output voltage that willproduce between 100 Hz and 1 kHz ofdeviation. We will also need to establishthat the output of the sound card is DCcoupled so that we can place a small bat-tery in series with the audio generator andthe tuning port of the reference oscillator.We can provide a DC return on the outputby putting a resistive 10-dB attenuator ofthe appropriate impedance on the outputof the sound card and passing the varactorbias through the attenuator to ground. Thisis required to bias the varactors and alsoassure that the audio voltage will not drivethe varactors into conduction. We can nowconnect the AC coupled input of the soundcard to the VCO tuning voltage and turn onthe loop.

Do not be surprised to see signal com-ponents at multiples of the power line fre-quency, as well as the signal from the audiooscillator. The presence of line frequencycomponents is an indication that the loopis doing its job and removing these com-ponents from the spectrum of the VCO.We can now “sweep” the audio oscillatorand, by plotting the amplitude of the audiooscillators response, determine the closedloop bandwidth of the PLL. In the case ofthe loop described above, (ωn = 22790, D= 1.0) we should expect to see slightly over1 dB of peaking at 0.7ωn and the 3-dBdown point should fall at 2.5ωn. In anyevent, if the peaking exceeds 3 dB, the loop

phase margin is growing dangerouslysmall and steps should be taken to improveit.

COMMON PROBLEMSHere are some frequently encountered

problems in PLL designs:

• The outputs of the phase detector areinverted. This results in the loop goingto one or the other rail. The loop cannotpossibly lock in this condition. Solution:Swap the phase detector outputs.

• The loop cannot comply with the tuningvoltage requirements of the VCO. If theloop runs out of tuning voltage beforethe required voltage for a lock isreached, the locked condition is notpossible. Solution: Re-center the VCOat a lower tuning voltage or increase therail voltages on the op amp.

• The loop is very noisy and the tuningvoltage is very low. The tuning voltageon the varactor diodes should not dropbelow the RF voltage swing in the oscil-lator tank circuit. Solution: Adjust theVCO so that the loop locks with a highertuning voltage.

PLL SYNTHESIZER ICSNow that we have learned how to deal

with the loop design aspects of an indirectsynthesizer, it is time to look at just a fewof the many PLL synthesizers availabletoday. Simple PLL ICs have been avail-able since the early 1970s, but most ofthese contain a crude, low-Q VCO and aphase detector. They were intended forgeneral use as tone detectors and demodu-

lators, not as major elements in communi-cation-quality frequency synthesizers.

One device well worth noting in thisgroup, however, is the CD4046, whichcontains a VCO and a pair of phase detec-tors. The CD4046’s VCO is useless for ourpurposes, but its phase-frequency detectoris quite good. Better yet, the CD4046 is alow-cost part and one of the cheapest waysof getting a good phase-frequency detec-tor. Its VCO-disable pin is a definite de-sign plus. Later CD4046 derivatives, the74HC4046 (CMOS input levels) and74HCT4046 (TTL compatible input lev-els) are usable to much higher frequenciesand seem to be more robust, but note thatthese versions are for +5 V supplies only,whereas the original CD4046 can be usedup to 15 V.

Since then, many more complex ICsspecifically intended for frequency syn-thesizers have been introduced by compa-nies like Motorola, National (NationalLMX series) and others as a result of thegrowth in popularity of wireless devices.They normally contain a programmabledivider, a phase-frequency detector and areference divider that usually allows asmall choice of division ratios. Usually,the buffer amplifier on these parts’ refer-ence input is arranged so that it can be usedas part of a simple crystal oscillator. Thisis adequate for modest frequency accu-racy, but an independent TCXO or OCXOis better.

Among Motorola’s more popular partsis the MC145151. The MC145151 bringsall of its division-control bits out to indi-vidual pins and needs no sequencing forcontrol. It is the best choice if only a fewoutput frequencies are needed, becausethey can be programmed via a diodematrix. The MC145151’s divide-by-Nrange is 3 to 16383, controlled by a 14-bitword in binary format. The reference canbe divided by 8, 128, 256, 512, 1024, 2048,2410 and 8192. It’s possible to operatethe MC145151 to 30 MHz, and its phase-frequency detector is similar to that in theCD4046, lacking the MOSFET “Tri-State”output, but with the added benefit of alock-detector output. The MC145152 is avariant that includes control circuitry foran external dual-modulus prescaler. Thechoice of external prescaler sets the maxi-mum operating frequency, as well as maxi-mum and minimum division ratios. Forsome unknown reason, the reference divi-sion choices differ from those of theMC145151: 8, 64, 128, 256, 512, 1024,1160 and 2048. There is also a Tri-Statephase-frequency detector output.

The MC145146 is simply an MC145152with its control data formatted as a 4-bitbus writing to eight addresses. This re-

Chapter 10.pmd 7/31/2006, 11:34 AM46

Oscillators and Synthesizers 10.47

duces the number of pins required and al-lows more data to be input, so the refer-ence divider is fully programmable from 3to 4095.

Multiloop SynthesizersThe trade-off between small step size

(resolution) and all other PLL perfor-mance parameters has already been men-tioned. The multiloop synthesizer, a directsynthesizer constructed from two or morePLL synthesizers, is one way to breakaway from this trade-off. Fig 10.55, theblock diagram of a five-loop synthesizer,reflects the complexity found in some pro-fessional receivers. All of its synthesisloops run with a 100-kHz step size, whichallows about a 10-kHz loop bandwidth.Such a system’s noise performance, set-tling time after a frequency change andreference-sideband suppression can all bevery good.

Cost concerns generally render suchelaboration beyond reach for consumer-grade equipment, so various cut-downversions, all of which involve trade-offs inperformance, have been used. For ex-ample, a three-loop machine can be madeby replacing the lower three loops with asingle loop that operates with a 1-kHz stepsize, accepting the slower frequency step-ping determined by the narrow loop band-width that this involves.

DIRECT DIGITAL SYNTHESISDirect digital synthesis (DDS) is cov-

ered in this Handbook’s DSP and Soft-

Fig 10.55—A five-loop synthesizer.

ware Radio Design chapter. DDS tech-nology is on the verge of covering all thelocal oscillator requirements of an HFtransceiver. Very economical chipsets thatcover partial requirements are availableand can be exploited with the help ofPLLs. For an example, see Fig 10.56 andTable 10.3. A single loop with a pro-grammable divider could be used in placeof the crystal-oscillator bank. The resultwould be very close to the approach usedin the latest generation of commercialAmateur Radio gear. That manufacturersadvertise them as using direct digital syn-thesis has given some people the impres-sion that their synthesizers consist entirelyof DDS. In fact, the DDS in modern hamtransceivers replaces the lower-sig-nificance “interpolation” loops in whatotherwise would be regular multiloop syn-thesizers.

This is not to say that the DDS in ourcurrent ham gear is not a great improve-ment over the size, complexity and cost ofwhat it replaces. Its random noise side-bands are usually excellent, and it canexecute fast, clean frequency changes. Thelatest devices do 32-bit phase arithmeticand so offer over a billion frequency steps,which translates into a frequency resolu-tion of a few millihertz with the usual clockrates for our applications—so the oldbattle for better resolution and low costhas already been won. Direct digital syn-thesis is not without a few problems, butfortunately, hybrid structures using PLLand DDS can allow one technique to com-

pensate for the weaknesses of the other.The prime weakness of the direct digital

synthesizer is its quantization noise. ADDS cannot construct a perfect sine wavebecause each sample it outputs must be thenearest available voltage level from theset its DAC can make. So a DDS’s outputwaveform is really a series of steps thatonly approximate a true sine wave.

We can view a DDS’s output as an idealsine wave plus an irregular “error” wave-form. The spectrum of the error waveformis the set of unwanted frequency compo-nents found on the output of the DDS.Quantization noise is not the only sourceof unwanted DAC outputs. DACs can alsogive large output spikes, called glitches,as they transit from one level to another. Insome DACs, glitches cause larger un-wanted components than quantization.These components are scattered over thefull frequency range passed by the DDS’slow-pass filter, and their frequencies shiftas the DDS tunes to different frequencies.At some frequencies, a number of compo-nents may coincide and form a single,larger component.

A summing loop PLL acts as a trackingfilter, with a bandwidth measured in kilo-hertz. Acting on a DDS’s output, a sum-ming loop passes only quantizationcomponents close to the carrier. A sys-tem designer seeking to minimize theDDS’s noise contribution must choosebetween using a loop bandwidth narrowenough to filter the DDS (thereby reduc-ing the loop’s ability to purify its VCO)or a more expensive low “glitch energy”DAC with more bits of resolution (allow-ing greater loop bandwidth and betterVCO-noise control).

Complex ICs containing most of a DDShave been on the market for over 15 years,steadily getting cheaper, adding func-tions and occasionally taking onboardthe functions of external parts (first thesine ROM, now the DAC as well). TheAnalog Devices AD7008 is an entire DDS(just add a low-pass filter…) on oneCMOS chip. It includes a 10-bit DAC(fast enough for the whole system to clockat 50 MHz) and some digital-modulationhardware.

For the home-brewing amateur, DDS’sfirst drawback is that these devices mustreceive their frequency data in binaryform, loaded via a serial port, and thisreally forces the use of a microprocessorsystem in the radio. DDS’s second draw-back for experimenters is that surface-mount packages are becoming the normfor DDS ICs. Offsetting this, the resultingsimplification of an entire synthesizer to afew ICs should cause more people to ex-periment with them.

Chapter 10.pmd 7/31/2006, 11:34 AM47

10.48 Chapter 10

Chapter 10.pmd 7/31/2006, 11:34 AM48

Oscillators and Synthesizers 10.49

Table 10.3Band-Specific Component Data for the Summing Loop in Fig 10.56

CrystalBand Output Range Frequency Cx(MHz) (MHz) (MHz) (pF) VCO Coil

1.5-2.0 10.5-11.0 16.0 0 20 turns on Toko 10K-series inductor form (≈4.29 μH)3.5-4.0 12.5-13.0 18.0 0 16 turns on Toko 10K-series inductor form (≈3.03 μH)7.0-7.5 16.0-16.5 21.5 0 12 turns on Toko 10K-series inductor form (≈1.85 μH)10.0-10.5 19.0-19.5 24.5 27 8 turns on Toko 10K-series inductor form (≈0.87 μH)14.0-14.5 23.0-23.5 28.5 56 81/2 turns white Toko S-18-series (≈0.435 μH)18.0-18.5 27.0-27.5 32.5 39 71/2 turns violet Toko S-18-series (≈0.375 μH)21.0-21.5 30.0-30.5 35.5 27 71/2 turns violet Toko S-18-series (≈0.350 μH)24.5-25.0 33.5-34.0 39.0 22 61/2 turns blue Toko S-18-series (≈0.300 μH)28.0-28.5 37.0-37.5 42.5 22 51/2 turns green Toko S-18-series (≈0.245 μH)28.5-29.0 37.5-38.0 43.0 22 51/2 turns green Toko S-18-series (≈0.239 μH)29.0-29.5 38.0-38.5 43.5 22 51/2 turns green Toko S-18-series (≈0.232 μH)29.5-30.0 38.5-39.0 44.0 22 51/2 turns green Toko S-18 series (≈0.227 μH)

The Toko 10K-series forms have four-section bobbins. The VCO-coil windings for 160 through 30 m are therefore split into fourequal sections (for example, 5 + 5 + 5 + 5 turns for the 160-m coil).

Fig 10.56—The G3ROO/GM4ZNXsumming-loop PLL phase-locks its VCOto the frequency difference between acrystal oscillator and 5.0- to 5.5-MHzVFO (these circuits are not shown; seetext). Table 10.3 lists the conversion-crystal frequency, VCO tuned-circuitpadding capacitance (Cx) and VCOinductor data required for each band.Any 5.0- to 5.5-MHz VFO capable atleast 2.5 mW (4 dBm) output with a 50-ΩΩΩΩΩload—1 V P-P—can drive the circuit’sVFO input.

Cx, C2, C3—NP0 or C0G ceramic, 10%or tighter tolerance.

C65, C66, C67—NP0, C0G or general-purpose ceramic, 10% tolerance ortighter.

D0A, D0B—BB204 dual, common-cathode tuning diode (capacitanceper section at 3 V, 39 pF) or equiva-lent. The ECG617, NTE617 andMV104 are suitable dual-diodesubstitutes, or use pairs of 1N5451s(39 pF at 4 V) or MV2109s (33 pF at4V).

D1—BA244 switching diode. The1N4152 is a suitable substitute.

L1—Variable inductor; see Table 10.3for value.

L13, L14—22-μμμμμH choke, 20% toleranceor better (Miller 70F225AI, 78F270J,8230-52, 9250-223; Mouser 43LQ225;Toko 144LY-220J [Digi-Key TK4232]suitable).

T1—6 bifilar turns of #28 enameled wireon an FT-37-72 ferrite toroid (≈≈≈≈≈30 μμμμμHper winding).

U2—74HC4046 or 74HCT4046 PLL IC(CD4046 unsuitable; see text).

Fractional-N SynthesisA single loop would be capable of any

step size if its programmable dividerweren’t tied to integer numbers. Somedesigners long ago tried to use such frac-tional-N values by switching the divisionratio between two integers, with a dutycycle that set the fractional part. Thiswhole process was synchronized with thedivider’s operation. Averaged over manycycles, the frequency really did come outas wanted, allowing interpolation betweenthe steps mandated by integer-N division.This approach was largely abandoned be-cause it added huge sidebands (at the frac-tional frequency and its harmonics) to theloop’s output. One such synthesizer de-sign—which has been applied in a largeamount of equipment and remains in use—uses a hybrid digital/analog system tocompute a sawtooth voltage waveform ofjust the right amplitude, frequency andphase that can be added to the VCO tuningvoltage to cancel the fractional-frequencyFM sidebands. This system is complex,however, and like all cancellation pro-cesses, it can never provide complete can-cellation. It is appreciably sensitive tochanges due to tolerances, aging and align-ment. Even when applied in a highly de-veloped form using many tight-tolerancecomponents, such a system cannot reduceits fractional frequency sidebands to alevel much below –70 to –80 dBc.

A new approach, which has been de-

scribed in a few articles in professionaland trade journals, does not try to cancelits sidebands at all. Its basic principle isdelightfully simple. A digital systemswitches the programmable divider of anormal single loop around a set of divisionvalues. This set of values has two proper-ties: First, its average value is controllablein very small steps, allowing fine interpo-lation of the integer steps of the loop; sec-ond, the FM it applies to the loop is huge,but the resultant sideband energy isstrongly concentrated in very-high-ordersidebands. The loop cannot track such fastFM and so filters off this modulation!

The key to this elegant approach is thatit deliberately shapes the spectrum of itsloop’s unwanted components such thatthey’re small at frequencies at which theloop will pass them and large at frequen-cies filterable by the loop. The result is areasonably clean output spectrum.

EXPLORING THE SYNTHESIZER INA COMMERCIAL MF/HFTRANSCEIVER

Few people would contemplate buildingan entire synthesized transceiver, but farmore will need to understand enough of acommercially built one to be able to fix it ormodify it. Choosing a radio to use as anexample was not too difficult. The ICOMIC-765 received high marks for its cleansynthesizer in its QST Product Review, soit certainly has a synthesizer worth exam-ining. We are grateful to ICOM America

Chapter 10.pmd 7/31/2006, 11:34 AM49

10.50 Chapter 10

Fig 10.57—Simplified (A) and detailed (B) block diagrams ofthe ICOM IC-765 frequency synthesizer.

Chapter 10.pmd 7/31/2006, 11:34 AM50

Oscillators and Synthesizers 10.51

for their permission to reprint the IC-765’sschematic in the discussion to follow.

Fig 10.57A shows a simplified block dia-gram of the IC-765’s synthesizer. It containsone DDS and two PLLs. Notice that the DDShas its own frequency-reference oscillator.DDSs usually use binary arithmetic in theirphase accumulators, so their step size isequal to their clock (reference) frequencydivided by a large, round, binary number.The latest 32-bit machines give a step size of1/(232) of their clock frequency, that is, aratio of 1/(4,294,967,296). This means thatif we want, say, a 10-Hz step size, we musthave a peculiar reference frequency so thatthe increment of the DDS is a submultiple of10 Hz. This is what ICOM designers chose.One alternative would be to use a convenientreference frequency (say 10 MHz) and ac-cept a strange synthesizer step size. A 32-bitmachine clocked at 10 MHz will give a stepsize of 0.002328 Hz. It would be simple tohave the radio’s microprocessor select thenearest of these very fine steps to the fre-quency set by the user, giving the user theappearance of a 10-Hz step size. An error of±1.2 millihertz is trivial compared to the ac-curacy of all usual reference oscillators.

Installing the IC-765’s optional high-stability 30-MHz TCXO does nothing toimprove the accuracy of the radio’s DDSsection, since it uses its own reference. Theeffect of the stability and accuracy of theDDS reference on the overall tuned fre-quency, however, is smaller than that ofthe main reference.

The DDS runs at a comfortably low fre-quency (0.5115 to 1.01149 MHz) to easethe demands placed on DAC settling time.The final loop needs a signal near 60 MHzas a summing input. Just mixing the DDSwith 60 MHz would require a complex fil-ter to reject the image. The IC-765 avoidsthis by using a summing loop. The sum-ming loop VCO must only tune from60.5115 to 61.01149 MHz, plus somemargin for aging and temperature, but careis needed in this circuit because, as in thehome-brew summing loop described ear-lier, the loop may latch up if this VCOgoes below 60 MHz.

The final loop is a normal PLL with aprogrammable divider, but instead theVCO is not fed directly into the divider—it is mixed down by the 60.5115- to61.01149-MHz signal first. The final loopuses a bank of four switched VCOs, eachone covering a one fourth of the system’sfull output range of 69.1115 to 99.01149

Fig 10.58—The IC-765 frequency synthe-sizer down to the component level. Thekey ICs in the radio’s DDS (IC1, IC2 andIC3, right) appear to be custom compo-nents made especially for this application.

Chapter 10.pmd 7/31/2006, 11:34 AM51

10.52 Chapter 10

MHz. This mix-down feeds an 8.5- to 38.5-MHz signal into the programmable divi-der. To remove unwanted mixer outputsover this range, three different, switchedfilters are needed before the signal isamplified to drive the divider.

Fig 10.57B shows the synthesizer ingreater detail. The programmable divider/PSD chip used (IC1, a TC9181P) will notoperate up to 38 MHz, so a prescaler (IC2,a TD6102P) has been added before itsinput. This is a fixed divide-by-four pre-scaler, which forces the phase detectorto be run at one-fourth of the step size, at125 kHz. (A dual-modulus prescaler wouldhave been very desirable here because itwould allow the PSD to run at 500 kHz,easing the trade-off between reference-fre-quency suppression and loop bandwidth.Unfortunately, the lowest division factornecessary is too low for the simple applica-tion of any of the common ICs withprescaler controllers.)

Fig 10.58 shows the full circuit diagramof the IC-765’s synthesizer. The DDS is ona small sub board, DDS unit (far right). IC1,an SC-1051, appears to be a custom IC,incorporating the summing loop’s phaseaccumulator and phase detector. TheDDS’s sine ROM is split into two parts:IC2, an SC-1052; and IC3, an SC-1053. IC4

and IC5, both 74HCT374s, are high-speed-CMOS flip-flops used as latches to mini-mize glitches by closely synchronizing all12 bits of data coming from the ROMs. TheDAC is simply a binary-weighted resistorladder network, R4, which relies on theCMOS latch outputs all switching exactlybetween the +5 V supply and ground.

Look at the output of the main loopphase detector (pin 17 of IC1 on the PLLunit), find Q5, Q6 and Q7 and look at theRC network around them. It’s the recipeloop design again! R3 creates the zero, but

notice C1 across it. This is a neat and eco-nomical way of creating one of the otherpoles. R6 and C7 create the final pole.

Fig 10.59 shows the phase noise of anIC-765 measured on a professional phase-noise-measurement system. The areasshown in Fig 10.48 can be seen clearly.The slope above 500 Hz is the phase noiseof the final VCO and exactly tracks theGood curve in KI6WX’s QST article onthe effects of phase noise. Below the noisepeak, the phase noise falls to levels 20 dBbetter than KI6WX’s Excellent curve. Thislow-noise area complements the narrowCW filter skirts and the narrow notch fil-ter.

Many technically inclined amateursmay be wondering what could be done toimprove the IC-765’s synthesizer, or todesign an improved one. The peak in theIC-765’s phase noise at 500 Hz is quiteprominent. It does not appear to be due toa marginally unstable loop design, butrather to a mismatch between the choice ofloop bandwidth and the noise performanceof the oscillators involved. The DDS’sresistor-based DAC is unlikely to performas well as a purpose-designed12-bit DAC, and it also lets power-supplynoise modulate the output. The system’sdesigners appear to have deliberately

Present and Future Trends in Oscillator ApplicationIn this chapter a wide variety of oscillator types and

frequency-synthesis schemes are discussed. Whichtechniques are the most important today and which oneswill likely become important in the future?

As we follow radio technology from the invention ofthe vacuum tube, we can see a continuous evolution inthe types of oscillators deemed most useful at any giventime. Broadly speaking, communications systemsstarted out with LC oscillators, and when the need forgreater frequency stability arose designers moved tocrystal oscillators. As the need to vary frequencybecame apparent, low-drift VFOs were developed toreplace the crystals. As higher frequencies began to beexploited, stability again became the dominant problem.Multi-conversion systems that used low-frequencystable VFOs with crystal oscillators were developed toestablish the desired high frequency stability. Todaywith stability, variability, and programmability all beingrequirements, frequency-synthesis techniques havebeen adopted as the norm.

When we look inside today’s synthesized transceiv-ers, we usually see only two types of oscillators. Thefirst is a temperature-compensated crystal oscillator(TCXO), whose main purpose is to set the frequencycalibration of the transceiver. The second type is the lowphase-noise voltage-controlled oscillator (VCO) used insynthesizer phase lock loops.

Today, formerly popular mechanically tuned VFOshave fallen by the wayside, with the exception perhapsbeing their use in QRP projects and nostalgia radios.Even this could change in the near future in QRP

projects, as very low power consumption synthesizersare already employed in cell phones and handhelds.

As we look to the future, we can see several emergingtrends. Audio DSP has been around for a number ofyears, and some of the more contemporary radios arenow employing IF DSP. As A-D/D-A technology ap-proaches 16 bits at sample rates exceeding 60 MHz, itbecomes feasible to produce an almost entirely DSP1.8 to 30-MHz transceiver. The only remaining analogRF sections will be input filtering and gain compensa-tion, and the output power amplifier and filtering. All ofthe traditional local oscillator synthesis hardware, IFfilters, IF amplifiers, mixers and demodulation will bedone in a DSP engine that resides behind an A-D/D-Aconverter pair. For the frequency ranges covered by theA-D/D-A pair, the synthesis process will be substantially“hidden.” It will likely look like the software for directdigital synthesizer (DDS) minus the DA converter. Whatkinds of oscillators will be required for this architectureand what parameters will be most important?

Again, two types of oscillators will be necessary. Atfirst glance, it would be reasonable to have a TCXO forfrequency accuracy, just as in today’s synthesizedradios. While this would be perfectly acceptable,another option now exists. Current cell phones are beingequipped with low-cost Global Positioning Satellite(GPS) chip sets to meet the federally mandated 911emergency location requirements. An appropriate GPSchip set could give amateur transceivers frequencyaccuracy on a par with the cesium frequency standardsemployed by the GPS system. An additional benefit

Fig 10.59—The phase noise of an ICOMIC-765 (serial no. 03077) as measuredby a Hewlett-Packard phase-noisemeasurement system.

Chapter 10.pmd 7/31/2006, 11:34 AM52

Oscillators and Synthesizers 10.53

reduced the loop bandwidth to better fil-ter DDS spurs. If we were free to increasethe cost and complexity of the unit, wecould buy a high-performance DAC de-signed for low DDS spurs and trade thisimprovement off against noise by in-creasing the loop bandwidth—redesign-ing the loop poles, zero and gain. To helpwith the loop bandwidth, and get a 12-dBimprovement in any noise from the mainloop’s phase detector, divider and loopamplifier, we could remove the fixedprescaler (IC2, the TD6102P) and oper-ate the PSD at 500 kHz. This would re-quire the design of a faster programmabledivider that can handle 38 MHz.

Finally, the IC-765’s VCOs could beimproved. Their tuned circuits could bechanged to low-L, high-C, multiple-tun-ing-diode types. If we added a higher-voltage power supply to allow higherdiode tuning voltages than the 5.6-V levelat which the design now operates (onlylittle current is needed), higher-Q tuningdiodes can be used—and we can avoidusing them at lower tuning voltages, wheretuning-diode Q degrades. This could re-duce the phase noise above the noise peakby several decibels. Changing the loopbandwidth would reduce the noise bump’sheight and move it farther from the carrier.

It’s important to keep all of these what-ifs in perspective: The IC-765’s synthe-sizer was one of the best available inamateur MF/HF transceivers when thischapter was written, and its complexity isabout half that of old multiloop, non-DDSexamples. Like any product of mass pro-duction, its design involved trade-offsbetween cost, performance and componentavailability. As better components be-come less expensive, we can expect excel-lent synthesizer designs like theIC-765’s to come down in price and evenbetter synthesizer designs to becomeaffordable in Amateur Radio gear.

BIBLIOGRAPHY ANDREFERENCESClarke and Hess, Communications Cir-

cuits; Analysis and Design (Addison-Wesley, 1971; ISBN 0-201-01040-2).Wide coverage of transistor circuit de-sign, including techniques suited to thedesign of integrated circuits. Its ageshows, but it is especially valuable forits good mathematical treatment of os-cillator circuits, covering both fre-quency- and amplitude-determiningmechanisms. Look for a copy at a uni-versity library or initiate an interlibraryloan.

J. Grebenkemper, “Phase Noise and ItsEffects on Amateur Communications,”Part 1, QST, Mar 1988, pp 14-20; Part2, Apr 1988, pp 22-25. Also see Feed-back, QST, May 1988, p 44. This mate-rial also appears in volumes 1 (1991)and 2 (1993) of The ARRL Radio Buyer’sSourcebook. Covers the effects of phasenoise and details the measurement tech-niques used in the ARRL Lab. Thosetechniques are also described in asidebar earlier in this chapter.

W. Hayward and D. DeMaw, Solid StateDesign for the Radio Amateur(Newington, CT: ARRL, 1986). A goodsourcebook of RF design ideas, withgood explanation of the reasoning be-hind design decisions.

“Spectrum Analysis...Random Noise Mea-surements,” Hewlett-Packard Appli-cation Note 150-4. Source of correc-tion factors for noise measurementsusing spectrum analyzers. HP notesshould be available through local HPdealers.

U. Rohde, Digital PLL Frequency Synthe-sizers (Englewood Cliffs, NJ: Prentice-Hall, Inc). Now available solely fromCompact Software, 201 McLean Blvd,Paterson, NJ 07504, this book containsthe textbook-standard mathematical

would be that the transceiver would now know itslocation anywhere in the world. This would be of valueto DXpeditioners and contesters!

The second oscillator would most likely take the form ofa very low phase noise voltage-controlled crystal oscilla-tor (VCXO). Some variant of the Butler oscillator wouldbe a likely candidate. This oscillator would be disciplinedby the output of the GPS chipset or the onboard fre-quency standard TCXO. The requirement for low phasenoise will remain very important, since any clock jitter inthe DSP engine or the A-D/D-A clocks will degrade thephase-noise performance of the transceiver.

The possibility of an all-DSP HF radio raises thequestion of whether or not frequency synthesizers, aswe now know them, will go the way of the VFO? For themost demanding applications, such as in signal genera-tors and test equipment, the answer is likely no. Thereason is that the most current synthesis techniquesoffer better performance than is possible with the basicDDS approach, although at additional cost. With respectto HF amateur radio equipment—where cost perfor-mance tradeoffs are important—the answer is likely yes.

Now that we have discussed the possibility of a DSPbased HF radio, what about VHF, UHF and microwaveamateur equipment? It is reasonable to assume that thedirect DSP approach could be extended up in frequencyas the A-D/D-A technologies improve in speed, how-ever, there are some limiting factors regarding phasenoise performance. The present method of using atunable IF, such as the HF transceiver and atransverter, is likely to remain the practical option for

some time to come. What we are likely to see is adifferent approach to the generation of the LO frequencyin transverters. Many of today’s transverters employ acrystal oscillator and multiplier chain to produce the LOinjection frequency. This architecture has some disad-vantages in that the crystal oscillator often has an offsetand drift that must be taken up in the HF transceiver.There is also the issue of spurious responses due to themultiplying process and adequacy of the attendantfilters. An alternative would be to use a simplified staticsynthesizer whose frequency reference is the same asthat used in the HF transceiver. This approach, imple-mented with a low-noise UHF or microwave VCO andphase-lock loop, could yield reasonable phase noise,greater frequency accuracy, fewer spurious responsesand smaller size as opposed to the crystal/multiplierchain technique. Again, we can see that the low-noiseVCO will be an important element in these designs.These UHF and microwave VCOs would likely employceramic or dielectric resonators like those now usedextensively in cell phones and TVRO front ends.

It is extremely likely that voltage-controlled oscillators,whether they are crystal, LC, ceramic or dielectricresonator types, will dominate the communicationselectronics landscape. Very low phase noise will be thecardinal performance specification for these VCOs.High-stability frequency-standard oscillators, while stillviable, may give way to GPS as a preferred means ofaccurate frequency deployment. Mechanically tunedVFOs have all but disappeared from modern equipmentand are likely to continue to diminish in application.

Chapter 10.pmd 7/31/2006, 11:34 AM53

10.54 Chapter 10

analyses of frequency synthesizers com-bined with unusually good insight intowhat makes a better synthesizer and a lotof practical circuits to entertain seriousconstructors. A good place to look forlow-noise circuits and techniques.

U. Rohde, “Key Components of ModernReceiver Design,” Part 1, QST, May1994, pp 29-32; Part 2, QST, Jun 1994,pp 27-31; Part 3, QST, Jul 1994, pp 42-45. Includes discussion on phase-noisereduction techniques in synthesizersand oscillators.

Pappenfus and others, Single Sideband

Circuits and Systems (McGraw-Hill). Abook on HF SSB transmitters, receiversand accoutrements by Rockwell-Collinsstaff. Contains chapters on synthesizersand frequency standards. The frequencysynthesizer chapter predates the rise ofthe DDS and other recent techniques,but good information about the effectsof synthesizer performance on commu-nications is spread throughout the book.

Motorola Applications Note AN-551,“Tuning Diode Design Techniques,” in-cluded in Motorola RF devices

databooks. A concise explanation ofvaractor diodes, their characteristicsand use.

I. Keyser, “An Easy to Set Up AmateurBand Synthesizer,” RADCOM (RSGB),Dec 1993, pp 33-36.

B.E. Pontius, “Measurement of SignalSource Phase Noise with Low-CostEquipment,” QEX, May-June 1998, pp38-49.

F.M. Gardner, Phase Lock Techniques(John Wiley and Sons, 1966, 1975;ISBN 0-471-29156-0).

Chapter 10.pmd 7/31/2006, 11:34 AM54