Hi Kyrill,
I dread these questions. They go to the core of the audio design paradigm, reveal my philosophies, and are very hard work to explain in logical terms. I design more by intuition than by pure logic (I am not an engineer!) and having to explain these difficult issues is a challenge for me.
I will present my POV in bullet points; it's a bit easier to follow. What I write is pretty heavy stuff!
#1 There are two schools of thought on audio; subjective v. objective. The disagreements are bitter, trenchant, and expose vested interest, intellectual snobbery, club membership, passionate belief, and marketing hype. It would seem one has to choose which camp to inhabit, a difficult choice, particularly as in life the best approach is often the middle course.
#2 If we choose objective, then on current convention the approach should be lowest possible distortion. This is a beguiling quest, as reduction of distortion to almost immeasureable levels is certainly possible and a most rewarding pursuit. But there appears to be limited correlation with the subjective listening experience, which I would define as the likes/dislikes of a group of a 100 non-technical audiophiles across an ownership of a couple of months on all musical genres. Wide ranging? Of course, that is the consumer, the designer must accommodate the consumer in all his diversity, just as in other technologies. There is promising work being done in this field, however, and in time this may bear fruit.
#3 If we choose subjective, then this means designing by hit and miss, lots of listening tests, not too much distortion analysis, and a general touchy feely/mystical approach much despised by engineers the world over, perhaps with good reason. However, another way of looking at this is designing according to engineering principles, but testing over a long period with subjective listening tests. This is how John Linsley Hood, Arthur Bailey, John Curl, Nelson Pass and many other successful designers do it.
#4 When one looks at the standard western chromatic scale, it is quite evident that it does not much favour odd order harmonic distortions. So 3rd, 5th, 7th, 9th etc tend to sound harsh, and unmusical. Even order distortion, however, is a different story, with 2nd and 4th sounding warm, intimate, if a little indistinct. Because of the transfer curves, push pull SS amplifiers tend to produce a spray of harmonics, frequently monotonically decreasing, from H2 through H3 to about H13. The odd order harmonics tend, for psychoacoustic reasons, to confer a hard, 'surgical' sound. Indeed, this has always been the main criticism of SS when compared to the more melifluous tube, which has a lower order transfer curve (exp. 1.5) and thus produces large quantities of H2 and H3 and some H4 and not much beyond.
#5 The feedback question is pivotal for the following reason. SS devices have very, very high transconductance, and thus in common emitter configuration yield very high gain. Feedback is necessary to bring this gain back to realistic levels. Tubes have much, much lower transconductance, and thus finite, measureable gain, typically in open loop around the 20 mark. Because of their more elongated, linear transfer regions, they are usually operated without global feedback, unlike their SS brethren. This means the distortion spectrum, reputedly of more significance than THD these days, is concentrated at the low order; mostly H2 and H3. It's high, around 2-3% on a good day, but low order, and largely musical. Add to this the fact that largish levels of low order distortion have a masking effect on the human ear (it's been estimated that the ear itself introduces around 30% H2 to all signals coming in), these levels effectively reduce the harsh effects of residual high, odd order distortion, and so you can see why tubes work so well. But the absence of feedback with tubes, pretty much mandatory in SS amplifiers, has problems, to do with bass response, drive, and frequency response.
#6 Global negative feedback reduces absolute levels of distortion, but is increasingly ineffective at the very high speeds at crossover. At the crossover point most of the high order artefacts are produced as devices switch on and off when one side of the output stage hands over to the other. Furthermore, GNFB tends to 'blenderise' lower distortion levels, morphing them into higher order distortions by intermodulation mechanisms, albeit at very low levels. The low levels are nevertheless objectionable in long listening sessions and usually show as 'listener fatigue'. When you realise that the masking effect of highish levels of H2 and H3 is lost with GNFB, you can see why it's a mixed blessing. There are other reasons too why GNFB is not ideal; chiefly in the way the error signal is extracted. This process itself can be inaccurate, leading to inappropriate correction.
OK, so that's the philosophy. Let's talk about the GK1.
Input stage is a discrete Opamp, with a long tailed pair, a common emitter voltage amplifier, a single ended output stage and heavy global feedback. With high levels of feedback, and a deliberately engineered level of H2 and H3, this has very low overall distortion largely because there is no push pull output stage. The power supplies help too; they are regulated, and have falling impedance with rising current draw. The fb factor, however, is high; with more than 60dB of feedback, the distortion levels are incredibly low, and the output impedance is a very low 30 ohms. This is ideal for driving a potentiometer, a considerable source of distortion (loss of resolution) in most amplifiers. The output from this first stage is exceptionally clean, with just sufficient H2/H3 to mask the nasties we normally associate with SS designs of this conventional nature. So, you see nothing unique, just a very careful balancing of operating points and dimensions.
The tube output stage is a self-biased cathode follower, using a variable mu tube. A plate loaded triode traditionally has high levels of distortion; the cathode follower around 15% of these levels according to the Radiotron Handbook by Fritz Langford Smith. Since we want a bit of H2/H3, we can easily achieve this by choosing a special kind of tube with variable mu, so that Vgk varies that bit more in operation than if we had a fixed mu tube. These tubes were used as automatic gain control devices in AM radios, and in recent years have been much in demand for Fairchild 670 tube limiters, used in the recording industry and to this day the standard for such devices. I use the 6ES8, and if you google Pendulum ES8 you will find something very interesting, no coincidence.
The operating point of the tube in the GK1 is chosen to give best sonics; this was a long quest, as many factors must balance. We want low order distortion in very small quantities, but we want strong, tight bass as well, one of the big problems for tube circuits.
I won't reveal all the details, but I will say that a great deal of the so called 'tube sound' is related to the tube's interaction with the power supply. This is why different designs using the same tubes sound quite different! For best effect on low level stages (and NOT transformer output stages, which are very different!), the supply should be high impedance, but should also be frequency dependent, ideally with low impedance at low frequencies and high impedance at high frequencies. This is done on the GK1 through some fairly trick circuitry. The result is that very little H2 and H3 is added by the tube at low frequencies, giving tight, punchy bass, but increasingly more is added to 'romanticise' the sound at higher frequencies.
Anyone who has heard the GK1 will agree that it is subtle; the effect creeps up on you as you listen more and more. The warmth of the midrange, and crystalline purity of the top end, with the wide image, the wonderful dynamics and very tight, strong bass is a revelation. None of this is terribly evident after ten minutes, but it really endears itself to you after a few weeks ownership. So how do we improve upon this for the new generation Aspen amps, the Lifeforce and the Soraya, which are specifically designed for very low distortion?
First off, use a teflon input cap. That really makes an impact by improving resolution.
Second, redesign the first stage for all single ended circuitry, so that any distortion will be principally even, and even only.
Third, redesign the tubestage for cathode follower operation but in half pentode, half triode. This is achieved with trick screen grid circuitry, and should give the best of both worlds.
And that, Kyrill, is the basis of my next generation Swift, which will use just this thinking.
If I ever get around to it......... In my semi-retirement and my advancing years, I'm starting to get a little tired!!
If this does not answer all your questions, do please send me an email, and don't worry about stepping on eggshells. I'm pretty confident of my products, and criticism is absolutely no problem. Two heads are always better than one.......
Cheers,
Hugh
