This page attempts to explain the design of a SE tube amplifier from the initial concept to the finished product. I simplified the theory as much as possible. Some of the material presented here may be over the head of a beginner. That doesn’t stop you from building one!
Every design project must have some target,
When designing an amplifier, I start with the output stage. I get it working, then determine the drive and power requirements. Then I design a suitable driver stage. Any breadboarding is done using lab type power supplies at this time. After the complete amplifier is running, it is easy to determine its true power requirements. Then I design a power supply for the amplifier.
The Output Stage Design
The first major decision is the choice of output tubes. I really liked the sound of the 6AV5's, but they are not widely available, and their power handling capabilities varies wildly depending on manufacturer and vintage. Other sweep tubes were considered, but the same concerns apply to them as well, and some of them have become expensive. The obvious choice is to use a common low cost audio tube like a 6L6, KT-88 or an EL-34 strapped for triode operation. The ideal choice would be to design an amplifier that could accept any of these tubes without modifications. UL or pentode operation could be used for more power output. It was tentatively decided to try these output tubes.
Now that the preliminary output tube selection has been made, the next decision is what operating point to use. This is often not an easy decision. Many people opt to use the published conditions, or copy the bias circuitry from an existing amp. I needed to devise a method to test the chosen tubes in several configurations, find the optimum bias point, and determine the drive requirements. It was determined in a previous experiment that the Edcor and the Hammond low cost output transformers respond extremely well to cathode feedback. The Edcor transformers have taps for ultralinear operation. I needed to test a large combination of output stage configurations without designing and building a complete amplifier. Fortunately, I have done this before.
I wired external octal sockets and output transformers into a Tubelab SE amplifier. The Tubelab SE uses a 5842 and PowerDrive to generate a very good quality drive signal capable of pushing a 300B well into clipping. It is more than adequate for this task. I used a variable power supply, adjustable bias, and several different output transformers with 6L6's, EL-34's, KT-88's, 6550's and a few others. This test amp was set up to operate with or without cathode feedback, in triode, pentode, or UL operation. This yields a huge matrix of possible combinations. How do you decide how much to turn up the power supply, or the bias current without frying the tubes? How do you determine the proper load impedance to use. How much drive will I need?
One of the first things to consider when designing a SE amplifier is how hard to push the tubes. Each tube has a maximum power dissipation rating. You look this up, and then decide how close to this limit you want to operate. Many people use the 80% rule of thumb, some go more conservative, especially with expensive tubes. This is usually a safe bet, but some tubes can be safely operated in excess of their published ratings, while a rare few show obvious distress operating at the published maximum ratings and must be operated far below them.
I tend to test the tubes that I want to run in the circuit, and find the point where the tube shows distress. I tested the Chinese 6L6 at dissipations from 30 to 44 watts in 2 watt steps. The tube was allowed to operate for a few minutes at each power level. It was then carefully observed in a darkened room. There is a picture of this testing below. In this case the 6L6 was running at 44 watts (way over spec). The interesting information here is that the red glow on the plate was evenly distributed (no hot spots) and even though the plate was very red there is no glow on the screen grid wires. This tells me that I would have no problems running this tube at its published max of 30 watts (the fact that they are $8 helps). It also tells me that even though the plate is glowing, the screen grid has no problem with 400 volts.
In my usual style I had to go beyond the limits of safe operation. Here you can see the power supply maxed out. The voltmeter is pegged and the current meter is at the edge of the red zone. The digital meters show the tube current, 108 mA for one tube and 111 mA for the other at 400 volts, this is about 45 watts dissipation. I usually don't push tubes this hard, but Antique Electronics Supply stated in their ad for these 6L6 tubes "we couldn't blow up this tube no matter how hard we tried" so I had to try. I just looked and don't see these on their web site anymore, that is too bad (I have since found them on Ebay), I really like the sound of them. These tubes are still in the amp! See the pictures below.
I performed the same test with the JJ and Svetlana EL-34's, after finding one bad Svetlana EL-34 (one of the screen grid wires brightly glowed at 200 volts) the others were stable and showed a very slight redness (room lights off) at 35 watts. 27 watts should be no problem, but will work the tubes hard.
The problem child of the bunch turned out to be the Chinese KT-88's. These started to glow at 35 watts, and there were hot spots visible on the plates, not even heating. This will kill a tube relatively quickly. These tubes are specced for 35 watts Plate dissipation and 7 watts of screen dissipation. Here they are showing two pronounced red spots at 35 watts total plate and screen dissipation. This should not happen. I got these tubes from a vendor about 5 of 6 years ago because of customer returns. He decided not to carry them any more and sold me his entire stock cheap. I have been using them without issue in P-P amps. I decided to use them at 30 watts anyway and see what happens. One tube mysteriously died during casual listening. It lost its vacuum. There are no visible cracks or damage on the tube, it just wouldn't play anymore. The getter has been slowly turning white.
I did not torture test the Sovtek or GE 6550 tubes since I only have a few of the Sovtek's, and the GE's are to expensive to risk damage. Both of these performed flawlessly in the amp with no signs of distress in total darkness.
I decided that it would be acceptable to operate any of these tubes at dissipation levels between 25 and 30 total (plate and screen) watts. This is within the published maximum ratings of each tube, and far below my "tested safe limit". It is important to consider tube dissipation in all SE amplifiers, since this is usually the limiting factor when deciding how hard to push a tube. You must operate the tubes within the maximum voltage, cathode current and power dissipation limits for reasonable tube life. Now we have a dissipation level and we are bounded by a maximum of 450 volts (the upper limit for screen voltage on some 6L6 specs) and 150 mA (the maximum cathode current for an EL-34).
Once you decide on the dissipation that you want to run a tube at, then you need to figure out how to get there. Lets say you want to run a given tube at 30 watts. You could put 250 volts across it and draw 120 mA through it, that is 30 watts. You could use 300 volts and 100 mA or 400 volts and 75 mA. If the tube was rated for 450 volts you could use 450 volts and 66mA. Each of these sets of conditions will result in 30 watts dissipated in the tube (equal stress on the PLATE). Each should be capable of similar power output if the optimum load and drive conditions could be met. Each set of conditions will have an optimum load impedance and drive requirements that can be determined graphically by drawing a load line, or by a simulation program like the TubeCad SE amp calculator. TubeCad will draw the load lines, and let you play with the operating conditions and watch what happens. See the Simulations page. If you do this you will find that for low plate voltages the optimum drive conditions can not be met without driving the grid positive. This will limit the maximum power output for some tubes.
In other designs both SE and P-P, I have found more power to be available at higher voltages and higher load impedances, the higher load impedance tends to lower distortion and improve bass response. It usually sounds better too.
There is another point to consider. If each of these sets of conditions generate equal stress for the PLATE (dissipation), which set do you think would generate the most stress on the cathode? The job of the cathode is to push electrons out into the space charge cloud, to be sent on their way to the plate. More current = more electrons sucked out of the cathode! To operate the tube at less current, what do you do? You turn UP the plate voltage! I know that this sounds counterintuitive, but it is true. Find a ham radio operator who has built linear amplifiers using transmitting tubes and ask him what makes those old Eimac tubes live the longest. They will tell you to operate them at lots of voltage, 2.5 to 3.5 KV, and low current!
Now, I have simulation and load line data stating that I will get the most power at a relatively high plate voltage and load impedance, and most of the published operating conditions for any of these tubes seem to want to run them at a relatively low voltage (250 to 350 volts) and force a higher current swing by using a 2.2 to 3.6 k ohm load. I needed to investigate these and other possible operating points, load conditions, and drive requirements for each of the possible output tubes in a real amplifier. To do this I needed a test amp. I started with the modified Tubelab SE shown above. This allowed me to adjust everything by turning a the knobs and it provided an extremely clean drive signal. Even though the test amp used fixed bias and the final design will use cathode bias, the operating points and load impedance will be the same. I ran each tube from 250 volts to 400 volts (the max of my power supply) in 50 volt steps with several different output transformers. I measured the power and distortion in triode and UL mode. The results were compared to the values predicted from the graphical solutions. I also experimented with the TubeCad SE amp simulator. The results agree well, except for the Chinese KT-88. These tubes don't act like KT-88's at all. They still sound good though.
I ran the test amplifier with 2.5, 3.0 5.0 and 6.6 K ohm loads with 6L6's, EL-34's, KT-88's, 6550's and a few others to arrive at the combination that worked the best with all of the available tube types. I tried to find one set of conditions that allowed all of these tubes to work, afforded each tube the ability for its characteristic sound to come through, and sound great at the same time. This meant looking for a set of conditions that allowed for the SAME BIAS VOLTAGE, and output transformer to be used with all of the tubes. This turned out to be 400 volts or more. At lower voltages I could not get the tube current of the 6L6 to be any where near the tube current of the EL-34 for the same bias voltage. I also found far more dynamic sound by raising the supply voltage and lightening the load at the same time. I found best sound with a 5 K ohm load. Others have proposed similar operating conditions with a load as high as 10 K ohms on the EL-34.
I have already received a few e-mails asking why I chose an operating point of 450 volts for a beginner level amplifier. This is a valid concern. I chose a 440 to 460 volt range because it affords the best possible performance for this simple design. I understand the need for an amplifier that operates from a lower voltage supply, and I am working on a version that operates at 350 volts with a lower power output. I will post the low voltage version of this design as soon as it is thoroughly tested. Either version can be built on the same PC board.
OK, now we know which tubes we can use in the output stage, we know the desired voltage that we want across the tube (about 400 volts), the bias current (60 to 70 mA depending on tube type), and the load impedance (5 K ohms). We need to know how much bias voltage is required and the AC voltage required to drive the tube. TubeCad allows you to play with the bias, and even determine the cathode resistor. From the experiments I know that the bias voltage used during the testing sessions was 37 volts. I had a meter on the grid during testing. This agreed with the simulations. Now that we know the bias voltage and the tube current, the required cathode resistor can be calculated. 37 volts and 66 mA requires a resistance of 560 ohms. 560 ohms is also the value I settled on during simulation.
How much drive voltage do we need? We can easily estimate the worst case driving voltage by assuming a peak to peak voltage of twice the bias voltage. This means that we might need 74 volts peak to peak, or 26.2 volts rms, IF NO FEEDBACK IS USED. If we use cathode feedback from the secondary of the output transformer, we must add the total feedback voltage (about 10 volts rms) to the drive requirements. The simulations allow adjusting the drive level while watching the power and distortion. It shows a value of 38 volts to be the maximum drive before serious distortion sets in. This value is given in peak AC volts. This is 26.4 volts RMS, close to the computed values.
TubeCad was a great help in determining the initial conditions (starting point) for this amplifier. It saved several weeks of experimenting. However it has several limitations that must be overcome with a few weeks of experiments. Tubecad works in triode mode only it can't do UL or pentode mode. It also assumes that the amplifier does not use any feedback. There is also a list of expensive output transformer models. It does not have any models for low cost transformers. In short TubeCad gave me a working amplifier to start with. It is up to me to optimize it and refine it. Fortunately the DC operating conditions will not change, only the AC and drive levels.
Driver Stage Design
Several years ago I got a warehouse full of military surplus tubes, mostly from the WWII era. I went through the small tubes that I had large quantities of in search of a good tube for general purpose amplifiers and driver use. I found the 5842. The 5842 has been my driver tube of choice ever since, but my supply is getting very low, and they have gotten far too expensive. The 5842 can deliver 150 volts peak to peak with distortion below 1/2 % when operated with a CCS load. I needed to find a tube that works as good, but doesn't cost a small fortune. I would prefer a tube that is in current production so that it would be available world wide.
As with any other design task, I first gathered the requirements. I would prefer a single stage design. Bach in the warehouse, I have several boxes full of small tubes, both triodes and pentodes. There are some tubes that I have several thousand of like the 6AK5 that would make a good driver, but I am not sure that they are available world wide. I would prefer to use a tube that I have lots of. About a year ago, I started connecting a few samples of each of these tubes into a generic amplifier circuit and taking gain, distortion, frequency response, noise and other important measurements. I have compiled some of this information into a spread sheet. I will post it eventually. I used this information to narrow the list of possible tubes.
I need to deliver about 35 volts RMS, or 100 volts peak to peak from an input of 2 volts peak to peak. This requires a voltage gain of 50. I will assume that A2 operation will not be required, and no current amplification (PowerDrive) will be used. The load impedance is the grid resistor of the output stage in parallel with the grid capacitance of the output tube. This capacitance is given in the tube data sheets. You must add the G1 to everything capacitance (15.5 pf for EL-34) to the miller capacitance which is the G1 to plate capacitance multiplied by the triode wired Mu. (.6 pf * 8.3 for the EL-34) for a total of about 20 pf. I will add a few pf for socket and wiring capacitance and call it 30 pf. The grid resistor could be as low as 100K ohms.
I want the driver frequency response to extend to at least 50 KHz without any roll off or phase shift. Experience tells me that the typical 12AX7 type circuit could give me the gain, but would loose a lot of this gain driving a 100K ohm load, and not meet my frequency response goals with 30 pf in parallel with the 100K load. I would need a tube that is happy with 5 to 10 mA of current. This is why I picked the 5842 in the first place, it has good gain and noise with 10mA of current. This rules out the 12AX7 and the 6SL7, too bad, I have several hundred of each. I had a few voltage amplifier circuits set up on the Tubelab prototyping system for analysis. I went back and forth between real hardware and simulation several times before finding the right tube.
The Tube Cad voltage amplifier simulation program is of some help here, but it does not account for the capacitive load from the output stage, so its high frequency response simulations are a bit optimistic. It does give me the output impedance which I wanted below 10K ohms (hence the need for 10 mA of current) because I didn't want to use PowerDrive. Tube Cad even let me "invent" a tube that doesn't really exist to see if my expectations were possible. See the simulations page for details.
I found After extensive experimentation, I settled on the 12AT7 - ECC81. This tube works almost as well as the 5842 and is available world wide.