SE+ amplifier
Once again, it is a small tube-based, single-ended headphone amplifier designed to drive the Sennheiser HD-580, Hd-600, and HD-650 (which I have just picked up). The HD-650s are not bad, although not a great leap from the HD58s0, more of an evolutionary improvement, but nonetheless a worthwhile improvement. To be honest, my Stax Lambda Pros have spoiled me and I have purposely not listened to them in over a year, so that my dynamic headphones do not sound the worse for it.

In the schematic above, we see two OpAmps, one working as an amplifier, the other as a virtual ground. The two 10k resistors define a voltage divider that splits the 9 volts in half and, thus, defines the “ground.” The 0.1µF capacitor may not be needed, but since most OpAmps suffer from a dissimilar PSRR figure for positive and negative power supply connections, the negative usually being about 20 dB worse than the positive, the capacitor added a little insurance.

The virtual ground is meant to be shared between channels and need not use an OpAmp, as a buffer IC would work just as well, if not better; this buffer could be the lowly LM310 or something hotter, such as, the LT1010CN8. Well, that’s how the plan started. I then wondered if each channel should get its own virtual ground. Two OpAmps or two buffers could have their outputs shorted together, with a 10-ohm resistor in series with each output. This would work, but the added resistors troubled me, so I wondered why not give each channel its own independent ground, as that would also allow easy phase reversals. Of course, headphones don’t come with two plugs, but splitting the wires and adding plugs would be easy enough.

Sandman's "error take off" scheme
Well, the next step was to look into each virtual ground doing something more that just splitting the 9 volts in half. It was then that I remembered a great article from the October 1974 issue of Wireless World, the king of electronic magazines. A.M. Sandman had revealed a fresh and radical approach to distortion elimination, which he labeled, “error take off.” Below is an example of his idea. (I believe that this article should be as well read as H.S. Black's feedback patents.)

In the schematic above, two views are presented. On the left side we see how an input signal would propagate through the circuit; on the right, distortion. The distortion side is the more interesting. Here we see 1.1 volts of distortion at the top amplifier’s output. This distortion is not present in the input signal, so the two feedback resistors voltage divide the distortion to one eleventh of it original value. The bottom amplifier then receives this 1/11 worth of error signal and amplifies it by elevenfold. The result is that the loudspeaker sees the signal from both amplifiers; but as the signal is equal in magnitude and in phase, the loudspeaker never experience any net current flow across its voice coil; thus, no sound. Brilliant, but expensive. Two amplifiers-worth of parts and beefed-upped power supply is quite a price to pay for reduced distortion. Still, the whole point of this journal is tube topology, not tube accountancy.

Mr. Sandman realized that this would be a common complaint, so he offered several variations on the error-take-off scheme. Consider this: even in the dirtiest amplifier, distortion is a small percentage of the output. So why use equally powerful amplifiers to vanquish distortion, when a smaller, less powerful, but cleaner amplifier can clean up the mess left by a bigger, dirtier amplifier. Below we see Sandman’s iterative circuit.

Each cascading amplifier receives the distortion from its preceding amplifier and works to cancel it at its output. Three amplifiers are shown, but many more could be used, each being less in power than the preceding. (Note the increasing in value series resistors at each output. This would imply that each succeeding amplifier would have to have an ever greater gain than the preceding.) By the way, this arrangement is somewhat similar to the Quad current dumping scheme, but not exactly.

Before the tube fanciers lose heart from seeing too much silicon, I should point out that Sandman’s error-take-off scheme is not limited to solid-state amplifiers—tube amplifiers could just as easily be used. Imagine a class-B, pentode-based, push-pull amplifier, one that sported four 6550s and a high-voltage power supply. Such an amplifier could easily put out 200 watts, 200 fairly-dirty watts. Now, imagine a class-A, triode-based, push-pull amplifier, one that sported two triode-connected EL34s and a medium-high-voltage power supply. Such an amplifier would put out only 10 watts, 10 fairly-clean watts. And, finally, imagine a class-A, triode-based, single-ended amplifier, one that sported one 12B4 and a lower-high-voltage power supply. Such an amplifier would only put out 1 watt, 1 clean watt. Each cascading amplifier would further refine the output signal. (I can hear the grumbling: What a tasteless idiot this guy is. Who would use 6550s, El34s, and 12B4s? Ok. Imagine four 211s, two 300Bs, and one 45.)

Some whining
The sad fact that I refuse to accept is that most tube fanciers are supremely indifferent to new topologies and circuitry understanding in general. Small, stupid, baby-steps variations on the grounded-cathode amplifier—such as using two cathode resistors in series, and then, several years later, two plate resistors in series—would make 99% of tube fanciers happy; all the rest just confuses them. Thus, the genius of the Ultrapath topology—you attach the power supply capacitor to the cathode, not ground—a technique requiring no understanding and certainly no formulas. Genius, pure genius.

For the remaining 1% of us, the readers of this journal, my recommendation is to seek out articles like Mr. Sandman’s and see what is common in the seemingly different. For example, why not build a Quad-like current-dumping amplifier with tubes? Or, why not look into US patent number 2748201 by McMillan for some ideas?

Back to Altoid tins
Well, what did I conclude was best for the little metal box? A few quick tests revealed that most OpAmps sound dreadful with only 4.5-volt power supply rails, that even rail-to-rail OpAmps do not work that much better with such a wimpy power supply. The Sennheisers might have made the situation worse, as they need a few volts of output to sound best (I’m sure that Grado headphones would work better in this setup.)

The best solution I found, other than using two 9-volt batteries, was to create a bridge amplifier out of two OpAmps. This arrangement effectively doubles the output-voltage and current swing over a single OpAmp and it was amazingly easy to implement, as shown below.

Notice how the unbalanced input signal is balanced at the two outputs and how the feedback loops do not require the usual coupling capacitor to ground.

Ok, so what’s wrong with this circuit? Each OpAmp now effectively sees a 150-ohm load, which is a bit much for most OpAmps. The NE5532s would work, but I hate the dead sound they make. The best solution might be to add a buffer stage to each output (I would use discrete transistors).