I am also a bit concerned about second harmonic distortion and wonder if some added circuitry (& admittedly added complexity) can neutralize more of it through inherent structure. Earlier Class A designs published by Pass like the A75 have more traditional topologies. But even a Class A push-pull output stage gets away from the elegant minimalism of Pass' Zen musings, and does add more stuff in the signal path. By way of overview, Zen are single stage, early Pass Aleph commercial amps are 3 stage, 2 in later Alephs, and 2 stages with inverting and non-inverting, positive and negative (4 way) output stage symmetry in the X amps. Alephs are single ended outputs, and X are sort of quad single ended. After reviewing two decades of Pass' designs, I'm leaning towards the A75 for some of the reasons described below. A75 is 2 to 2.5 stages, balanced, all class A, with push-pull outputs. A75 is fairly conventional, but has lower distortion and is still quite simple.
Class A amps need to dissipate a lot of heat and one of the biggest problems the Do-It-Yourself hobbyist amp builder faces is finding sources for chassis and especially the big heat sinks needed. John Inlow provides plans for an amp chassis with heat sinks made from aluminum bar stock for a truly DIY solution. At 0.55 degree C/W, this heat sinking is only suitable for smaller amps, so I'd probably use multiple 6 inch lengths of commercial Themalloy or Wakefield extrusions bolted to a chassis of Aluminum plate and bar. See also his link to SM0VPO's brief heat sink tutorial.
Chronologically, the original Zen design was a single MOSFET, and Son of Zen was a differential pair of power MOSFETs. While the original Zen had a constant current source, Son of Zen had simple resistors supplying the bias current. Zen was capacitively coupled on the input and output, and Son of Zen was DC coupled. Zen used modest feedback while Son of Zen used none.
Active current sources offer about four times the efficiency of passive resistors, and they also improve the rejection of power supply noise and fluctuations. They do however add another transistor into the circuit, which was objectionable to some minimalist-inspired readers. A differential input means balanced signals can be used to drive the amp, which makes for potentially cleaner signal transmission due to common mode noise rejection. Getting rid of the capacitors on input and output is a good thing since only horrible sounding electrolytics could be used in the large values required.
Hoffmann's design is at the same time a good reading of Pass' work and a classic, simple op amp with some feedback. It uses a differential input pair and a follower (two gain stages) with each stage biased by a constant current source. He also runs split power supplies so the inputs and outputs are direct coupled. Power output with 25 VDC rails is 35 watts into 8 ohms, half that into 4. Like most of Pass' recent commercial designs, it's a big, simple power opamp using International Rectifier HexFET Hybrid MOSFETS. I plan to build a pair, probably with higher-voltage, regulated supplies on each stage.
So I'm now thinking of starting with a Pass/Thagard A75, adding beefier and regulated power supplies, some audiophile type passive components such as Vishay bulk foil resistors and Rubycon Black Gate capacitors in most locations. Since the design is fairly minimalistic there aren't too many non-key locations to use "regular" Dale/Vishay RN55D resistors, Panasonic HFQ electrolytics, etc. though parts of the circuit that don't handle the signal directly or don't vary much might qualify. Another source of improvement could be lower noise precision voltage references to replace the zeners.
Like Zen and Aleph, the A75 is all Class A, but A75 has a balanced input and push-pull output. It also has a lot more power (150 watts and above into 4 Ohms). Being push pull, it's around 45% efficient versus 20% for single ended. Push pull also means the dominant second harmonic produced by all transistors' non-linearities is largely canceled out due to the complementary output. Yes this could make third and other odd order harmonics more noticeable, but with good matching and tweakable feedback levels, the objective distortion levels are well down and much lower than Pass single-ended designs.
Unlike Zen, A75 has what looks like a folded cascode driver stage, which would normally be thought of as the middle stage of a three stage design. But the folded cascode is arranged so that half of it acts as a current source and the gain and drive actually comes from the differential input transistors. In his introduction Dr. Thagard thus describes the design as having a compound single input stage, effectively resulting in a two stage amp. The design includes the option to run as essentially three stage by increasing the contributions of the folded cascode, resulting in more loop gain and greater negative feedback levels and lower distortion. Tuning out the folded cascode results in a simpler gain path and different sound. I would probably build A75 as designed, but with top quality parts. Would also build them as monoblocks with much larger toroidal transformers (say 1 kW Plitrons) and possibly higher supply voltages for more power. Only the filter caps would really care about higher voltage up to about +/- 100 Vdc. It would probably also be good to regulate the output stage power supply. The downside of more voltage and more bias is more power consumed and heat produced, meaning more waste and more heat sinking needed.
Here is a table comparing various Pass-related DIY Amplifier designs. I was finding it challenging to remember the variations so I though I would make this comparison chart to help sort them out.
Judging by the service manual schematic, the Adcom GFA-5802 looks like a dumbed-down relative to the A75. It's certainly based on a Pass design, with very similar ideas and even some identical component values. However the GFA-5802 is capacitor-coupled at the inputs, and it lacks some of the more sophisticated circuitry of the A75. They may share some sonic attributes but the A75 is very likely superior.
Nelson Pass responds to questions from DIYs and buyers in a guest forum at AV Science. There he describes differences in his commercial amps, offers some advice to DIY amp builders, and also clarifies his relationship with Adcom as: “Adcom has licensed the basic circuits employed in most of their amps and preamps, but I am not involved in construction and parts selection and so on.” Pass also contributes an occasional response at the DIYaudio.com forum where seeming legions of fans embrace his designs.
Wow! Found an excellent article on heatsinks at Rod Elliot's sound site. If the rest of the articles are as good, they are a huge contribution to on-line audio knowledge. MMM Metals appears to be a major manufactuerer of extruded, bonded, etc., heatsinks.
It is also possible to perform the modulation entirely in the digital domain. How about an amplifier that takes the digital signal in PCM from a CD or DVD and, without first entering the analog domain through a D/A converter, modulates it directly from PCM into PWM?
In essence the amplifer becomes the digital-to-analog converter. Attenuation should be done in the digital domain, not on the PCM side, but on the PWM side. (In other words, to attenuate the signal (i.e. turn down the volume) don't throw away bits on the PCM input, instead narrow the PWM pulse widths on the output.) This topology would eliminate a lot of analog problems, though of course it raises other challenges, such as proper filtering and feedback. Naturally the attenuation algorithm would be another key. It also creates a significant burden on jitter and clock cleanliness within the modulator if the results are to be good. I also suspect the PWM analog post-filter should deserve a close look. As we have seen in DACs, analog filtering is crucial and can not be a casual afterthought. Again, problems (challenges) get shifted into other areas. It would be interesting to experiment with this approach. It should have some good potential. Perhaps there are already some commercial examples? If not, perhaps there should be.
Such purely digital amplifiers already exist in sound reinforcement and very low-end consumer electronics, and one or two of high-end examples have been tried.
Some interesting Class-D documents: