Hybrid Headphone Amplifier
In blog 231, I described the Aikido SLPS headphone amplifier, which used a variation on split-load phase splitter to produce a low-noise tube headphone amplifier that could work over a large range of load impedances, unlike the SRPP or White cathode follower or SRCFPP circuits, which must be optimized to work into a single fixed load impedance. One downside to the Aikido SLPS headphone amplifier, however, is the need to use five triodes per channel, rather than the usual four triodes. (If audio triodes didn't usually come packaged two per tube envelope, this would not be so big a deal. Of course many single-triode tubes exist, such as the 6C4 or 6J5, but these tubes become rarer each day, as our audio tube palete shrinks.)

In addition, the split-load phase splitter's triode only gets 1/4th of the available B+ voltage. For example, if the B+ voltage is 160Vdc, then this triode would be allocated only 40Vdc of cathode-to-plate voltage—not much and, perhaps, too little, depending on the triode used. Of course the easy solution is to use a much higher B+ voltage; but the problem with easy solutions that they are so seldom easy. For example, as shown in the circuit below, the split-load phase splitter's triode sees a cathode-to-plate voltage of only 45Vdc and we might run into excessive output tube dissipation and we certainly would have to use smaller-valued power-supply capacitors, with a higher B+ voltage.

See SPICE simulation below

Let go over the circuit's operation, in light of the added parts. Each output transistor gets its own 2-ohm emitter resistor, which both linearizes and protects its transistor from dead shorts at the output and thermal runaway. The 1kµF bootstrap capacitor tightly couples the output signal to the phase splitter. If a positive going voltage pulse is forced on the output, this 1kµF capacitor relays the pulse to the top of the 300-ohm resistor; but the other end of the resistor is terminated by the 47 0hm resistor and the PNP transistor's emitter, so the 300-ohm resistor undergoes an increase in current conduction, which is then shared by the bottom 300-ohm resistor, both being in the same current path. In other words, most of the positive pulse appears at the bottom NPN transistor's base (300/347 worth), causing the transistor to greatly increase its conduction, pulling down the buffer's output. In turn, the top NPN transistor's base sees only 47/300 of the pulse, but its emitter and its emitter resistor see 100% of the positive pulse, which decreases the top transistor's conduction, causing the buffer's output to fall. In other words, both output transistors work equally to buck the imposed pulse.

Okay, why is there an output capacitor, when the output is only 10mV off from 0V? Safety and paranoia explain most of it. High-impedance headphones, say 300 ohms, might tolerate such a big DC offset, but I am not sure about all 16- or 32-ohm headphones. The workaround is to use an input coupling capacitor and a DC servo loop to keep the DC offset low.

The OpAmp runs off the same +/-15V rails and it compares the DC voltage at the output with ground, making the required adjustments to keep the two in line with each other. Well, since we added an OpAmp, why not just use an OpAMp to drive the headphones? The triangle circuit can be made with big power transistors in its output stage, either TO-3 or TO-247 devices, which means that an idle current of 500mA is possible, which, in turn, means that peaks of 1A are possible. Now, 1A into 8 ohms equals 8Vpk, which equals 4W! This may be flea power for big speakers, but is massive power for 32-ohm headphones. Could a triangle power buffer actually be built to drive 8-ohm loudspeakers, rather than headphones? Indeed, as the following schematic shows. Note the inclusion of the two constant-current sources. In addition, note the four output transistors and the heavy idle current. Such a buffer yields a truly single-ended harmonic structure.

Okay, so why should I use the triangle circuit instead of a diamond circuit, which can also be made with equally big power transistors and could also drive speakers? Good question. It seems that the big advantage the diamond circuit holds over the triangle circuit is that no bootstrap capacitor is needed, which is great if you are making ICs. And the big advantage the triangle circuit seems to offer is one fewer transistors. Well, the following graph from a SPICE simulation reveals the real answer.

Note that the only harmonic that the diamond circuit beats the triangle circuit at is the 2nd. Also note the single-ened aspect of the triangle harmonic structure, in contrast to the push-pull structure from the diamond. I know which I would prefer to listen to. Slam dunk, right? Well, no. At 1Vpk @1kHz, the battle isn't so decisive.

Note how the diamond's 2nd harmonic is substantially lower than the triangle's. Slam dunk? No. Note how the triangle's 3rd harmonic is substantially lower than the diamond's. The ear rather likes the 2nd harmonic, but it dislikes intensely the 3rd harmonic, as it sounds deaden, lifeless. So, given the choice between more 2nd harmonic and less 3rd, I always choose less 3rd, even when the total harmonic distortion is greater.

Considering that the common definition of an "audiophile" is someone who plays his music too dang loud, I am sure that most headphone-loving audiophiles would prefer listening to the triangle's vastly lower distortion at high levels.

 

Hybrid Triangle Circuit
The large-valued capacitor used to relay the output signal to the split-load phase splitter can be replaced by an NPN transistor or N-channel FET or a triode. The following schematic shows a hybrid triangle buffer. The triode's grid sees all the output signal, both AC and DC. In turn, its cathode follows its grid, which creates a varying current flow through both the top and bottom 200-ohm resistors. My first thought was that the triode's rp divided by its mu + 1 would define its output impedance at its cathode, which would have to be subtracted from the top 200-ohm resistor's value; but, no. In SPICE simulations the top resistor's value remained at 200 ohms. Why? The triode, which is configured in a cathode-follower topology, works into the 200-ohm cathode resistor that terminates into the PNP transistor's emitter, not ground, so the signal loss is much lower. Nonetheless, the PNP transistor's emitter resistor must be varied with the external load impedance; the lower the impedance, the lower the resistor value, with a floor value of 120 ohms with the output shorted to ground.

Since audio triodes most often come two to the tube envelope, we can use the extra triode as a high-impedance input buffer.

Of course, if we increased the bipolar rail voltages to +/-100Vdc, we could replace the two output transistors with triodes, making the buffer a closer approximation to pure-tube. Actually, +/40Vdc rails and a 6AS7 as the output tube might be fun.

Or we could replace both triodes with two transistors, which would require a voltage-dropping device, such as a zener or a diode string to give the base a higher voltage than the output.

 

Electrolytic+ +citylortcelE
Note the two electrolytic output capacitors, which are in series and have their positive leads attached together and biased by the +15V rail. Why? When the tube is cold and the triodes are not conducting, the constant-current source will pull the output down to the negative rail voltage. but the these output coupling capacitors will never be reversed biased, which is an absolute no-no with polarized capacitors. Beyond safety, such an arrangement sounds amazingly good, far better than a single electrolytic capacitor does.

I have no idea why this occurs; well, no idea that I would be willing to put into print. In a similar vein, I have found the sound from Panasonic non-polarized capacitors to be exceedingly fine. Perhaps, like transistors in a diamond configuration, electrolytic capacitors should always be be used in pairs. (I have found that even film capacitors sound better in parallel, with their ins and outs cross coupled.) Perhaps, the 100k polarizing resistor is not needed, but I have always found its inclusion a sonic good. (I remember reading that in radio work, the two-elecrolytic-capacitor-with-polarizing-resistor trick greatly increased the useful bandwidth of the capacitors, a very odd result, considering that it effectively doubles the capacitor's series inductance. Never forget that an electrolytic capacitor is an odd chemical-electro beast.)

Some enterprising reader should rush to the patent office and patent the following setup, wherein a grounded-cathode amplifier has its cathode resistor bypassed by two electrolytic capacitor in series, positives connecting, polarized by a two-resistor voltage divider. The name? Need I say it, The Ultra Bypass, of course.

For example, if 5Vdc appeared at the cathode and 16V electrolytic capacitors were used, the polarizing voltage should be about 10Vdc. (Several years ago, I received an interesting e-mail, wherein the reader asked what I thought of a bold new topology, which was said to redefine the very standards by which other tube circuits were judged and which, invariably, were found wanting. It was a grounded-cathode amplifier that held two cathode resistors in series. My answer was that it would better to use two plate resistors in series, as it reduced the voltage-induced resistor distortion. Well, this polarized-two-capacitor trick might actually be worth the fanfare.)

 

Aikido Line & Headphone Amplifier
Returning to tubes, the transitor-based triangle buffer could be used with a PS-6 power supply and an Aikido line stage.

The PS-6 power supply presents three output voltages: a low-voltage bipolar output (up to +/-50Vdc) and a voltage quintupled high-voltage B+ output voltage roughly equal to five times the bipolar rail voltage. For example, if the rail voltages are +/-30V, the high-voltage B+ output voltage will be about 147Vdc; if the rail voltages are +/-15V, the high-voltage B+ output voltage will be about 73Vdc.

Or we could use the hybrid version of the triangle buffer with an Aikido line stage.

Click on image to see closeup

This Aikido offers two ouputs, one for line use and one for driving headphones. If we wish to tap earlier in the signal chain, the following variation would be worth examining.

Click on image to see closeup