Original manuscript for what became "The Nearly Perfect Amplifier" (January, 1994 QST)


The Almost Perfect Amplifier

. After "Circuit Improvements for the Heath SB-220 Amplifier" was published in the November and December 1990 issues of QST , I began to receive letters and phone calls from amateur radio operators who were contemplating buying an HF amplifier. They wanted to buy an amplifier that needs no circuit improvements - that is, a perfect amplifier. What follows is a discussion of some design features that - in my opinion - would be present in a perfect, or near-perfect, amplifier.

Cathode Considerations

In a directly-heated type cathode, a ditungsten carbide layer on the hot tungsten (alloyed with about 1.5% thorium) filament wire emits electrons. In an indirectly-heated type cathode, the filament (a.k.a. heater) heats a nickel cylinder that is coated with strontium oxide and barium oxide. This coating is relatively frangible - but highly emissive.

Ditungsten carbide is commonly made by heating tungsten in an atmosphere of acetylene (C2H2) gas. Carbon atoms in the gas break their bonds with hydrogen atoms and bond with tungsten atoms to form ditungsten carbide on the surface of the filament wire. Since it is atomically linked to the underlying wire, the ditungsten carbide layer is very durable. During use, the process reverses. Ditungsten carbide gradually looses carbon and changes back to tungsten. Extra heat exponentially accelerates this process. The cathode is worn out when the carbon is mostly used up.

After they are worn out, large external-anode amplifier tubes are commonly "recarburized" with acetylene, vacuum pumped, and resealed. This restores full emission. Although it is possible to recarburize a 3-500Z, doing so is not economically feasible.

Each type of cathode has advantages and disadvantages. A nickel cylinder has much less inductance than a tungsten wire. Directly-heated cathodes are relatively poor performers above the low VHF range. Some indirectly-heated cathodes can perform satisfactorily at 2500MHz. A directly-heated cathode typically warms up in less than one second while few indirectly-heated cathodes can warm up safely in one minute - and three to five minutes is not uncommon. However, the major disadvantage of indirectly-heated cathode amplifier tubes is cost. In terms of dollars per watt, they are much more costly than 3-500Zs.

Cathodes deserve respect. Filament voltage and filament inrush current are areas for concern.

Filament Voltage

For optimum life from a directly-heated cathode, the filament voltage should be just above the voltage where PEP output begins to decrease. As the cathode ages, filament voltage needs to be increased gradually to restore full PEP. By using this technique commercial broadcasters typically achieve an operating life of 22,000 hours from directly-heated cathode amplifier tubes.

According to Eimac's *Care and Feeding of Power Grid Tubes*, "every 3% rise in directly-heated cathode filament voltage results in a "50% decrease in life due to carbon loss." Each additional 3% rise in filament voltage additionally decreases the life by half. If you are into math, cathode life is proportional to [E1/E2]23.4 - where E1 is the lowest filament voltage at which normal PEP output is realized - and E2 is the increased filament voltage.

It's easy to make the filament voltage adjustable when the filament is powered by its own transformer. All that's needed is a small rheostat in series with the primary. For dual voltage, dual primary transformers, a dual ganged rheostat is required. When the filament is powered by a winding on the HV transformer, making the filament voltage adjustable is more difficult since the rheostats must be connected to the LV secondary winding. For example, dual ganged 0.01-ohm, 30A rheostats are not to be found in your local Radio Shack.

Indirectly-heated cathodes can be permanently damaged by operating them below their rated minimum filament voltage. When operated above its maximum filament voltage rating, an indirectly-heated cathode quickly boils off emissive material. Errant emissive material is bad news when it lands on the grid.

For maximum cathode life in HF communications service, an indirectly-heated cathode should be operated at the rated minimum filament voltage. This can be accomplished best with a regulated DC supply. Set it once and forget about fluctuations in the electric mains voltage.

Filament Inrush Current

Directly-heated cathode filaments are commonly two vertical meshing helices (coils) of tungsten wire that are suspended by their ends. (see Sept. 1990 QST, p.15) The conductance of tungsten at room temperature is about 8.33 times the conductance at its normal operating temperature. Thus, the start-up current for a 15 ampere filament can be 125 amperes. Needless to say, 125 amperes makes for a dandy electromagnet.

In a high amplification tube such as the 3-500Z, the filament helices clear the grid cage by a matter of thousandths of an inch. If the position of the filament changes, a grid-to-filament short may result. Therefore, it is prudent to limit filament inrush current in order to minimize thermal and magnetic stresses.

Since the grid-to-cathode clearance in an indirectly-heated cathode is not affected by movement of the heater inside the rigid nickel cylinder, indirectly-heated cathodes are not affected by inrush current.

For many of its smaller directly-heated cathode amplifier tubes - such as the 3-400Z and 3-500Z - Eimac(tm) recommends that filament inrush current be limited to no more than double the normal current.

Measurements

I tested the filament inrush current in a popular (2) 3-500Z factory built MF/HF amplifier. Since operating filament current is 14.7A, the inrush current should not exceed 29.4A. I measured 34A of inrush current per filament. Eimac(tm) rates the filament voltage at 4.75V minimum to 5.25V maximum. The filament voltage measured 5.31V. With 4.8V instead of 5.31V, the useful life of the 3-500Zs would be about ten times longer.

I measured the filament voltage in a factory built single 3-500Z amplifier. The filament voltage was over 5.7V. At this voltage, the cathode would probably be worn out in 400 hours of operation - and this one was.

The 8877 has a filament voltage rating of 4.75V minimum to 5.25V maximum. For HF communications service, the optimum filament voltage is 4.75V. The Dentron DTR-2000 amplifier operates its 8877 filament at about 5.95V when the amplifier is operated at the U.S.-standard 120V/240V. Operating an 8877 at a filament voltage of 5.95V can only be recommended for those who have more money than brains.

A Simple AC Voltmeter

For measuring AC filament voltage, linearity at the low end of the meter scale is not important. With this in mind, designing an AC voltmeter using a DC meter movement is much easier than would otherwise be the case. All that's needed is a half wave rectifier using a Schottky diode, a capacitor and a few resistors. The meter can be the amplifier's multimeter. All that the operator needs is two marks on the meter scale. One for minimum V and one for maximum V.

Grid Protection

I have performed autopsies on too many kaput amplifier tubes that died in HF amplifiers. Some of these tubes had damaged grids - but the damage was the unique type that is caused by VHF or UHF current. Strangely, I have never found a grid that was damaged by excessive HF grid current. Perhaps this isn't so strange. I'm sure it's possible to roast a grid. Tuning up key-down for a couple of minutes on 10m with the load capacitor set for 40m comes to mind. This would result in very high grid current and almost no RF output. However, since most people - myself included - tune a grounded-grid amplifier for maximum RF - and maximum RF virtually coincides with normal grid current - very few people are likely to overheat a grid. Thus, complex electronic grid-protection circuits are unnecessary. A major disadvantage of electronic grid-protection circuits is they are not effective against the major source of grid damage - sudden, large bursts of VHF or UHF grid current. A more foolproof method of protecting the grid is a fuse or fuse resistor. Carbon film resistors make good grid fuses.

Glitch Protection

During a major glitch, the anode (plate) current meter is subjected to a current surge as the HV filter capacitors discharge. Such a current is typically several hundred peak amperes - not exactly courteous treatment for a 1A meter. However, the peak current will be much higher if a resistor is not used to limit the short circuit current that can be delivered by the HV filter capacitors. The current limiting resistor is placed in series with the positive output of the filter capacitors. A 10-ohm, 10W wirewound resistor is adequate for up to 3kV & 1A. Since the current limiting resistor will be dissipating many kW during a major glitch, it should be a rugged glass-coated (a.k.a. vitreous) type. If a glass-coated resistor opens during a major glitch, it won't be throwing large chunks of shrapnel around - like a rectangular ceramic cased resistor often does.

If the positive HV briefly arcs to chassis ground - due to lint, a tiny insect, an intermittent VHF parasitic, or an errant hair - the negative HV circuit will try to spike to several kilovolts negative. In the real world, this type of glitch is not an uncommon occurrence. Anything that gets in the way of the negative spike may be damaged. Since the grid current meter is normally connected between chassis ground and the negative HV circuit, the meter can be exposed to kilovolts at hundreds of amperes. I heard about one grid current meter in a homebrew amplifier that exploded. The glass landed on the floor.

The easiest way to protect a current meter is to connect a silicon rectifier diode across it or across its shunt resistor. Usually, only one diode {cathode band to meter negative} is needed in parallel with a meter.

It may take more than one diode to protect a meter shunt resistor. A silicon diode begins to conduct at a forward voltage of about 0.5V. To avoid affecting meter accuracy, the operating voltage per protection diode should not exceed 0.5V. For example, a 1-ohm shunt, at a reading of 1A full-scale, has 1V across it. Thus, two protection diodes would be needed to preserve meter accuracy. If the shunt resistor for a 1A full-scale meter is 1.5-ohm, three diodes are needed.

Protection diodes should not be petite. Big, ugly diodes with a peak current rating of 200a or more are best. I have seen smaller diodes - and the meter they were supposed to be protecting - literally blown away by a glitch. After some bitter experiences with lesser diodes, I began using the 1N5401. In small quantities, the 1N5401 costs about 20¢ each. It is rated at 200a for 8.3mS, 3A-RMS, and 100PIV. Other diodes from the 1N5400 family will work as well.

During an extremely high current surge, a glitch protection diode may short out - and by so doing still protect the precious parts. Replacing a shorted protection diode instead of a kaput meter is almost fun.

A brief HV flashover can damage an indirectly-heated cathode tube. Here's how: In many amplifiers, one side of the filament/heater is grounded. The cathode is connected to the negative HV circuit. If the negative HV spikes to several kilovolts, the cathode will arc to the grounded filament. At a minimum, this breaks down the insulation between the heater and the cathode. Sometimes the heater wire burns out - and sometimes the cathode arcs to the grounded grid. Either way the tube is kaput. There have been many 8877s and other indirectly-heated cathode amplifier tubes that died this way - all for lack of 60¢ worth of glitch protection diodes.

To prevent the negative HV circuit from spiking to several kilovolts, connect a string of 200a glitch protection diodes from the negative terminal on the HV filter capacitor to chassis. Each diode will limit the voltage across it to about 1.5v. Typically, three diodes are needed - thusly limiting the spike to about 4.5 volts. The diode polarity is: cathode band toward the negative HV. With one simple wiring change, the same string of diodes can also protect the grid I meter and the anode I meter. This dual protection technique is incorporated into the Adjustable Electronic Bias Switch in this article.

Power Supplies

Transformers

Virtually all transformers use paper to separate and insulate each layer of windings. Paper is hygroscopic - i.e., it absorbs water vapour from the air. The presence of water reduces the insulating ability of the paper. In time, insulation breakdown is likely. The solution is to pot the windings. Plastic resins are best. Petroleum tar is next best. Since potting fills up the air spaces in the windings - and air is a poor heat conductor - potting improves heat transfer, thereby reducing internal temperature and increasing MTBF. Potting adds very little to the initial cost of a transformer - and subtracts substantially from the long term cost.

Filters

Capacitor filter power supplies are the norm in amateur radio amplifiers. Light weight, relative ease of obtaining high DC potentials, good transient-current voltage regulation (a must for SSB), and cost are factors. Since choke and swinging-choke filters can not handle transient current loads, the only alternative for SSB use is a resonant-choke filter. It has advantages and trade-offs. A resonant-choke filter is tricky to tune, heavy, expensive, and it requires a much higher bleeder current than a capacitor filter requires. However, a resonant-choke filter demands only about one-sixth as much peak power from the electric mains as a capacitor filter demands. This means that for 120V operation, where power is much more limited than with 240V operation, a resonant-choke filter is clearly the best choice.

Rectifiers

The most frequent failure mode for HV power supply rectifiers is too much reverse-current. This problem can be virtually eliminated in full-wave rectifier circuits by making sure that the total PIV in each string of diodes exceeds the peak secondary voltage of the HV transformer by a comfortable margin. 50% sounds comfortable to me.

Modern solid-state rectifiers are made differently than they were 30 years ago. In those ancient days, rectifiers did not have uniform capacitance. In an attempt to help equalize the peak reverse currents in series connected rectifiers, a parallel resistor and capacitor was connected to each rectifier. It was felt that swamping 50 to 100pF with 10,000pF would help. In practice, the practice didn't work too well since the tolerance of the capacitors used was typically minus 20% to plus 80%. Another problem is that the resistors that were typically used - 470k-ohm, 0.5W - are rated at 250V absolute maximum. It is hardly safe to use one of these resistors with a 1000PIV rectifier. As a result, so-called "equalizing" did more unequalizing than anything else. After rectifier technology improved, people hung on to the old habit of using parallel resistors and capacitors.

There is a flaw in the logic behind using rectifier equalization. In any series circuit, the currents in all of the elements are exactly equal. Thus, when rectifiers are in series, the reverse-current burden is exactly the same for each rectifier. How is it that something which is already exactly equal needs to be "equalized"? However, series-connected rectifiers should always be of the same type. Mixing rectifiers with different junction capacitances could cause a problem.

There is one instance where equalizing resistors and capacitors are a good idea. Voltage spikes come in only two flavors - positive and negative. In a full-wave capacitor filter rectifier circuit, the energy from positive and negative voltage spikes are simply rectified and harmlessly stored in the filter capacitor. However, in a half-wave rectifier circuit only one polarity is rectified. A voltage spike of the other polarity can not be absorbed by the filter capacitor. Instead, the potentially-destructive spike appears across the rectifiers. Placing a capacitor across each rectifier helps to limit reverse spikes. A better solution is to connect a MOV across each half-wave rectifier - or use a full-wave rectifier circuit.

Electrolytic Capacitor Equalizing Resistors

1W carbon composition resistors have a maximum voltage rating of 350V. Voltage rating takes precedent over the wattage rating.

For example, it takes 469V to dissipate 1W in a 220k-ohm resistor. [E=(PR)0.5] In the past, some amplifier engineers decided that a 220k-ohm, 1W resistors would make good (and cheap) voltage equalizing resistors for 450V electrolytic capacitors. However, when a carbon composition resistor is operated above its maximum voltage rating it changes resistance - exactly what you don't want in a voltage divider. When their voltages are not properly equalized, electrolytic capacitor failure is likely. Even when they are operated within their voltage rating, carbon composition resistors change resistance with age. Thus, 2W carbon composition resistors, which are rated at 500V, are not the answer.

Metal oxide film [MOF] resistors are far superior to carbon composition resistors. A 3W, 100k-ohm MOF resistor makes an excellent equalizer resistor for 450V capacitors. Lower values of resistance create extra heat - something that electrolytic capacitors do not tolerate well.

Biasing

The operating bias in most amplifiers is not adjustable. A single Zener diode is typically used. The resulting zero-signal anode current (a.k.a. idling current) is seldom optimum. Adjustable bias would be nice. The solution: obtain the operating bias from a series string of forward biased rectifier diodes. By switching the number of diodes in and out with a rotary switch, the bias can be changed in approximately 0.7V increments.

Another area that could be improved on is the method of bias switching between receive and transmit. In this modern age there is no reason to be using a pokey, noisy mechanical relay to switch bias. An optoisolator coupled to a transistor switch can do this job better and cheaper. An electronic bias switch is more than fast enough to keep up with modern high speed RF relays.

Electronic bias switches that are RF-actuated create two problems. The amplifier tube switches between linear bias and non-linear bias during softly spoken syllables of speech. This causes choppy sounding audio and rotten splatter. These two problems are eliminated when the electronic bias switch is controlled by the current that passes through the RF relays' coils.

High Speed Relays

A conventional relay switches in 15mS to 25mS - doing so in a somewhat stentorian manner. Such relays have traditionally been used for RF and bias switching in HF amplifiers. This was acceptable when transceivers also used conventional relays. Currently manufactured transceivers are designed for AMTOR, QSK-telegraphy, and reasonably quiet SSB-VOX operation. Such transceivers switch quickly and quietly. Nipponese transceivers often use a Matsushita [Panasonic] NR-HD-12V reed relay to switch the RF output. Provided that the SWR isn't ridiculous, this relay can dependably switch 150W of RF. It works well in an HF amplifier when it is used to switch the input RF signal. This relay is available in the U.S. as a spare part for Kenwood and Yaesu radios. Yaesu's price is reportedly cheaper than Trio-Kenwood's.

Although Matsushita refuses to sell its NR-HD-12V relay to U.S. amateur radio equipment manufacturers, a U.S. reed relay manufacturer could probably come up with a similar relay.

Jennings and Kilovac manufacture high speed relays that will switch kilowatts of RF. The Jennings relay is the RJ-1A. Kilovac's relay is the HC-1. In large quantities, neither relay is terribly expensive. I have been using these two relays for many years to switch the output in my amplifiers. When mounted with silicone rubber, they are fairly quiet. When used with a speedup circuit, either relay can switch in under 2mS. I use a speedup/sequencing circuit that prevents them from being hot-switched. I have not had a relay fail in this circuit.

It makes very little sense to be currently building - or buying - an amplifier that switches in 25mS. If such an amplifier is used with a modern radio, it would be technically more correct to say "hot-switches in 25mS."

VHF Stability

On page 72, the 1926 Edition of the Radio Amateur's Handbook told us how to build an improved VHF parasitic suppressor - one that provides better VHF stability than ordinary parasitic suppressors. The logic was elementary. A suppressor is supposed to dampen a circuit. Since low Q is synonymous with high dampening, build a suppressor with low Q. Instead of using a conductor with a high VHF-Q - such as copper or silver - use a conductor with low VHF-Q, i.e., resistance wire. "The combination of both resistance and inductance is very effective in limiting parasitic oscillations to a negligible value of current."

After 1929, someone forgot to include this information in the Handbook. In those days, the oversight probably didn't matter very much. Electron tubes generally had poor amplification at VHF, so VHF instability was not much of an issue. During the ensuing decades, amateur radio operators and amplifier manufacturers got into the habit of using parasitic suppressors made from copper - or even worse - silver-plated copper. This was an easy habit to get into since copper and silver can be soldered easily and cheaply. Meanwhile, the VHF amplification of electron tubes kept improving. Modern tubes need 1926 - vintage low VHF-Q parasitic suppressors. Sure, it sounds crazy.

VHF parasitic oscillation can cause bandswitch arcing, tune capacitor arcing, and a large pulse of grid current. The pulse of grid current is so large that a powerful magnetic force is exerted between the grid and filament. In a 3-500Z, this force is capable of bending the hot tungsten filament helices - causing a filament-to-grid short.

Of course, there is a trade-off to using low VHF-Q suppressors. On the 10m band, they reduce amplifier output by roughly 0.08db. This should come as no surprise since anything that dampens VHF resonance is bound to have some effect at 29MHz. A VHF suppressor that does not get hot on 10m isn't doing its job.

Even though much has been published about VHF parasitic oscillation, amplifiers are still being built without the benefit of low VHF-Q suppressors.

Some amplifier manufacturers presently take a dim view of using low VHF-Q suppressors . Even if you have signs of parasitics, such as intermittent bandswitch arcing, they will void the warranty if you remove their high VHF-Q suppressors and install low VHF-Q suppressors. According to one of their customers, Ten-Tec claims that low VHF-Q suppressors wreck amplifiers.

There was a time when NASA wasn't too keen about improving the O-rings in the Shuttle booster rocket. Old habits are hard to break.

Step-Start

Large power supplies need something to soften the shock of start-up. A 10A DPST-NO or 10A DPDT relay and two approx. 25-ohm approx. 10W resistors are just about all that's needed to add a good step-start circuit to the average 1500W amplifier. The step-start circuit belongs in series with the main fuses or circuit breakers. This way the filaments will also enjoy the benefit of a gentle start-up.

Is More Gain Always Better?

Today, the more or less standard in transceiver output is 100W. There are amplifier tubes that can easily be ruined by 100W of drive. A good example is the 3CX800A7. Using 100W of drive will eventually strip flakes off of the cathode. The flakes lodge between the cathode and the grid cage - creating a short. Even a pair of 3CX800A7s are clearly over-driven by 100W. Doing so probably won't flake the cathodes but it can cause rotten splatter. The fix is simple: connect a 40-ohm resistor in series with each 3CX800A7 cathode. These [cathode RF negative feedback] resistors reduce gain. As a result, the amplifier won't be driven above its absolute maximum ratings - and into non-linearity - by a 100W transceiver.

Cathode RF negative feedback resistors are better than having a matched pair of 3CX800A7s -- the cathode currents automatically equalize themselves. And unlike all ALC circuits, cathode feedback resistors work instantaneously - eliminating ALC's generic flaw -- leading edge splatter on SSB. (Amplifier-to-transceiver ALC only works properly on constant signal level modes such as RTTY and FM.)

When a single 3-500Z is driven by 100W, it too splatters. Although the rated drive is around 55 watts, single 3-500Z amplifier manufacturers give the green light to driving their amplifiers to an anode current of up to 550mA - a feat that requires using about 100W of drive. On SSB, this produces distortion and splatter. Eimac(tm) rates the anode current at 400mA absolute maximum. What kind of an "amplifier engineer" ignores Eimac(tm)'s technical information? Certainly not the kind whose product I would consider buying.

It takes an approx. 25-ohm cathode feedback resistor to make a 3-500Z happy with 100W of drive. The resistor goes in series with the cathode coupling capacitor.

It would be nice if amplifiers were designed to be compatible with 100W of drive.

Adjustable Tuned Inputs

Modern solid-state output MF/HF transceivers use an untuned push-pull RF output stage. In order to meet FCC requirements on spurious emissions, passband LC filters are used. Such filters introduce inductive and capacitive reactance at various frequencies within their passbands. In other words the output Z of a modern transceiver is seldom 50 ±j0-ohms. This is of no consequence unless you happen to be driving a tuned input in a grounded-grid amplifier. In this case, the filter reactance interacts with the reactance of the input capacitor in the pi-network tuned input. The length of the coax between the filter and the tuned input affects the way the reactances interact. As a result, the input SWR can suffer. If the SWR is too high, the transceiver will automatically cut back on power.

For example, to obtain Eimac's recommended Q of 2, approximately 200pF of input capacitance is needed for a tuned input on the 10m band. In actual practice, however, a 50pF input capacitor may produce the best SWR with a particular model transceiver and a particular length of coax. A different model transceiver or a different length of coax may require a different value input capacitor.

It would be nice if an amplifier's input capacitors on the tuned inputs could be readily adjusted.

Summary

The number in [brackets] is the estimated difference in parts cost for the various design features.

Obviously there are other important elements in a good amplifier design. The elements discussed above are ones that are frequently overlooked. If you have any questions or comments, my new telephone number is 805-386-3734.

-- --