300b lovers

One variant we did not try was a center-tapped inductor to load the 6SN7 plates, and direct coupling between the input and driver tubes. That requires a high B+ voltage for the driver, but that’s no problem when the input + driver have their own power supply that’s fully isolated from the output section power supply.

It’s a good question if this is better or worse than a special interstage transformer. The center-tapped inductor shares many of the design challenges of an interstage transformer without necessarily having any advantages. This falls into the category of "build it and see". A minor challenge with a 450 to 500 volt regulated B+ supply, but nothing impossible.

Another similar option, instead of a center-tapped inductor, are paired current sources and direct coupling between input and driver tubes. This has an ugly disadvantage that a small DC imbalance in the input tube turns into a 5 to 10 volt offset for the driver tubes, which is intolerable because the driver section is then grossly imbalanced with different operating points for each driver tube.

To prevent this, it would require something like an opto-isolated DC servo circuit to balance the plate voltages of the input tube. Doable but kind of nasty, adding a lot of pointless complexity that adds nothing directly to the sound ... basically, another point of failure. The alternative would be manual adjustments (with limited range) on each current source and a meter so the user could hand-adjust DC balance.

Alternatively, the center-tapped inductor, because it has a moderate DCR of a few hundred ohms, limits the amount of plate-to-plate DC imbalance to a volt or so. However ... small as that is, that’s more than an interstage transformer, where the output going to the driver grids is always perfectly balanced, with zero DC offset from grid-to-grid.

What the interstage transformer does is offload circuit complexity to the ingenuity of the transformer designer. The circuit schematic looks simple, but what’s going on inside that transformer is very complex, requiring sophisticated modeling tools to fully understand.

I am still puzzled why capacitors measure so well, yet are so audibly colored. And why on Earth do they require hundreds of hours of active operation, not just polarized but signal going through them, to finally stabilize sonically? What’s going on inside them? My only guess is a (very) slow electrochemical process that subtly alters the dielectric properties of the plastic film.

If you’ve been curious the kind of things two designers discuss with other, the previous posts give a simplified example. Sonics, different circuits and what they sound like, and reliability. It's also an example why collaborations between designers can result in designs that are better than any one person can do.

By the way, there’s a very intelligent discussion of the latest John Curl/Parasound amplifier design here:

Parasound Discussion

Note they are talking about fully isolated input+driver sections with 117 volt rails. That might sound insanely high for a transistor amp, but overshoots in input sections of transistor amps are a big deal, and these are very high-powered amps with serious voltages appearing in the feedback loop.

By the way, cascoding the input section is how you get both voltage resistance and a hundredfold reduction in Miller capacitance. Since Miller capacitance in transistors is grossly nonlinear, this is a very good idea. Tubes exhibit Miller capacitance too, but it is an order of magnitude lower, and it is stable and predictable instead of being nonlinear. There are cascode tube circuits as well, but they offer no improvement in linearity (unlike transistors), and are mostly seen in phono preamps and FM tuner input sections. In the tube universe, pentodes behave similarly to a pair of cascoded triodes, and are more commonly used when a cascode is called for.

The even-order distortion of the driver section is cancelled in the primary of the interstage transformer, reducing driver distortion even further.

@lynn_olson Actually if the circuit is fully balanced/differential from input to output, even orders are cancelled at every stage along the way. In this manner distortion is compounded less from stage to stage.

The result is the 3rd harmonic is the dominant distortion product rather than the 2nd. Many people do not realize that the ear treats the 3rd in much the same way as the 2nd; its the only odd ordered harmonic that is musical to the ear. The 3rd is very good as masking higher ordered harmonics.

BTW, any properly functioning analog tape machine will have a 3rd harmonic as its dominant distortion product as the tape approaches saturation.

Mathematically this type of distortion can be described as a 'cubic non-linearity' as opposed to the 'quadratic non-linearity' of an SET. As Daniel Cheever pointed out in his paper from 1989, its important that the harmonics fall off on an exponential curve. Both an SET and a fully balanced amplifier can do this (its part of the reason people regard SETs as musical despite their many failings). The advantage of a balanced circuit is harmonics fall off on a higher exponential rate so higher ordered harmonics are at a lower level than seen in an SET; its inherently lower distortion.

This allows the distortion signature to be innocuous.

The advantage is greater power output with lower distortion. So at any power level an SET can make, in a circuit using the same tubes the PP amp can have vastly lower distortion and so be smoother with greater detail, since distortion obscures detail. 

By the way, cascoding the input section is how you get both voltage resistance and a hundredfold reduction in Miller capacitance. Since Miller capacitance in transistors is grossly nonlinear, this is a very good idea. Tubes exhibit Miller capacitance too, but it is an order of magnitude lower, and it is stable and predictable instead of being nonlinear. There are cascode tube circuits as well, but they offer no improvement in linearity (unlike transistors), and are mostly seen in phono preamps and FM tuner input sections. In the tube universe, pentodes behave similarly to a pair of cascoded triodes, and are more commonly used when a cascode is called for.

We've been using a differential cascode circuit for decades. It has several advantages over pentode or cascade operation; one obvious one being a reduction in the need for a coupling capacitor. Differential circuits benefit from the devices (tubes in this case) having a lot of gain. That increases the differential effect so distortion cancellation is improved and noise is reduced. A differential cascode circuit can have a lot of gain.  

This circuit can have a very high CMRR even in a tube embodiment. Its linear enough open loop that you can run it without feedback (something you can't do with pentodes), but if you want to do it, its possible to operate it in ultra-linear mode, where the plate Voltage of the top tube is applied through a divider network to the grid of the same tube. You can do that with a pentode too, but you can't run a pentode zero feedback and the amount of feedback available in UL mode is limited. 

Since a cascode circuit is lower distortion, another advantage is it can be used in a circuit with feedback and result in less higher ordered harmonic generation than if a pentode is used.

Certainly a great benefit in a phono preamp where you have to overcome the wall of noise in the very first stage, and yes, cascodes are more linear than pentodes. The EF86 mike-preamp pentode so popular in the Fifties and Sixties is now very expensive and hard to get, so cascodes make more sense today.

Don’t need all that gain in a power amp in the absence of feedback, but if we ever needed feedback, yes, that’s a good way to get it.

I completely agree about the benign nature of low-order distortion. Most of all, it reduces nasty high-order IM distortion which is objectionable and obviously electronic-sounding. I also agree about the heavy 2nd-order distortion of SETs, which limits their effective dynamic range. Clever design techniques can mask the 2nd-order distortion (like coupling caps with complementary sonics) but the IM distortion remains, and is audible with symphonic, choral, and heavy rock music. (Music with a sparse spectra, like jazz quartets or chamber music, doesn't expose IM distortion, but music with a dense spectra turns IM distortion into a wall of noise that goes up and down with the music.)

Since transistors are notorious for high-order distortion, cascoded differential sections are about the only way to tame the things, while also getting rid of nonlinear Miller capacitance. The classic John Curl topology.