co -- I must admit that I was surprised myself. It was yet another twist and turn that I did not foresee at the beginning of this project. What really surprised me, is that as the cause of the issue was being determined, I started scouring my resources to find any attention given to the matter of asymmetrical clipping due to unequal drive impedance from the previous stage, and could find nothing. I'm hardly saying it doesn't exist -- just that I haven't found it yet if it does.
So then I started looking at all the commercial UL designs that came to mind. The original UL Amplifier from Halfer and Keroes employs a driver stage that would prevent such behavior. Most of the early commercial and even DIY UL were all 6L6 based, and being largely Williamson oriented, employed driver stages as well. The bigger output tubes of the mid 50's were often directly driven by a cathode coupled inverter, which would also act to prevent such behavior. When the 6BQ5/EL84 arrived, use of the 6V6 disappeared almost overnight in high quality high fidelity equipment, with the vast majority of 6BQ5/EL84 designs not even being of UL design anyway. Off the top of my head, the small Heath and Dynaco amplifiers are the only commercial UL amplifiers I can think of using these tubes -- and they both use Cathodyne inverters. To that point then, I can't immediately think of any commercial design that in fact used a floating paraphase inverter that directly drives a UL output stage -- just the circuits offered in the Acro and Dyna transformer catalogs, that would have virtually all been built in DIY fashion, and would therefore have likely if not largely lacked any ability to observe, test, or diagnose the observation. And, given how few TO-310 and A-410 transformers are in circulation today -- compared to say the TO-300, TO-330, A-430, and A-431 transformers (all of which were used in commercial products) -- not many of the small catalog designs were even built.
Come to think of it, as I sit here typing and mulling over all of this, I can think of in fact one commercial product that does fit the description of this topic: The Acrosound Stereo 20, a mono version of which was detailed in an article entitled "Direct Coupled", as presented by Keroes in the March 1961 edition of Popular Electronics. Among it's many features, that amplifier does fit the topic of this discussion, employing a 12AX7 paraphase inverter driving an EL84/6BQ5 UL output stage. But lets face it, there were hardly many of these units in circulation to cause much of a stink about the matter, either. And, based on personal experience, these units had other far worse issues to deal with other than full power clipping, so again, the issue would have likely been masked.
As my memory comes up to speed this morning, I found that I actually have my own supporting evidence of this behavior with this very amplifier. There are a couple of well known AKs who absolutely say grace over their Acrosound Stereo 20 amplifiers, but not too long ago, sent me one out of frustration for numerous problems they were having with anything from overheated output tubes to high distortion and instability. One of the more obscure issues to address was the fact that the top output tube always clipped early in that amplifier! (exactly as noted with this project) As evidence of that fact, I have lifted a piece of my own correspondence back to the AKers regarding my explanation of the unequal clipping noted in that design:
"So the last performance anomaly of the design that I wanted to address (before finally getting to optimizing the HF response issues), is the fact that in the design as presented by Acro, V2 always clips first (or early) as the onset of clipping is approached. No doubt this was (in part) some of the reason for pursuing an AC Balance control. However, tests have shown that in this case, the uneven clipping is not a product of uneven drive, but of uneven impedance. Floating paraphase inverters can be adjusted for equal clipping -- which also coincides with minimum distortion as well -- rather easily with traditional designs. Of course, this design is anything but traditional, with the point being that V2 is driven by an incredibly high impedance from the starved operation of V1A. Now V1B also operates with starved operation, but: V1B does not have the additional cathode resistance of R6 to contend with, which is necessary for insertion of the NFB loop at this input. R6 supplies negative current feedback to V1A, which inherently raises its output impedance V1A over that of V1B. As well, and this is the biggest factor, V1B operates with considerable NFB around itself, which lowers its impedance quite significantly. With the impedance imbalance from the inverter then, as the output stage approaches clipping, V2 -- driven by a very high impedance -- will clip immediately at the first sign of any grid current, while V3 can be driven slightly into the first sign of grid current because of its lower drive impedance from V1B. With 25+ db of NFB in the global loop, it seeks to correct most of the effects of the imbalance, but cannot completely do so."
As implied in the text, the owners had tried (quite possibly at my own recommendation) to install an AC balance control to the Stereo 20 circuit to mitigate the unequal clipping themselves. In my response to them, I correctly identify the unequal drive impedance as being (what I now know to be) part of the asymmetrical clipping issue, but at the time, had not made the connection that in that particular design, it was in fact the unequal drive impedance when used in conjunction with a UL output stage. The more thorough investigation I did with the Magnavox project project here unearthed the relationship of unequal drive impedance and UL operation, which really puts the mystery to bed.
Obviously, there is more testing to be done, primarily centering on the question: how much drive imbalance can be tolerated in a UL output stage? Does it vary depending on the amount of screen FB applied (UL tap percent)? At what level of impedance does an imbalance no longer practically matter? These are things I will get to in time. At this point however, we do know that at least a Cathodyne inverter can be used with success (or at least insignificant concern) when directly driving a UL output stage (as all the Dynaco products show), and that pure pentode operation can tolerate quite a significant amount drive impedance imbalance, as the modified pentode based amplifiers offered with this project show not only equal clipping events, but also lowest THD when the drive is in fact adjusted for equal clipping. However, with the UL efforts of this project, it is worthy to note that when the AC balance control was adjust for minimum THD at a level of (say) 1 db below maximum power output, the amplifier then clearly displayed unequal clipping, and when adjusted to minimize the unequal clipping, the THD then rose considerably from the minimum distortion settings at all power levels.
So, there is more to do and chase down on this topic. But this does seem to be an issue that (apparently) hasn't been addressed in the literature by anyone I'm aware of -- either currently or back in the day -- which is further aggravated by the fact that the number of working examples of designs subject to this concern are extremely limited. Hafler almost certainly knew about it, as his commercial designs were clearly successful, and never based on a floating paraphase design -- except for his non-commercial catalog design that was quite likely simply a carryover from his days at Acrosound. The point is, if he did know about it, he never wrote about it to my knowledge.
Of note also, there is one last reference by Baldwin who worked for Acrosound, and published a article entitled "High-Power Performance With A Low-Power Amplifier" in the February 1958 issue of Radio-Electronics. His design also used a classic paraphase inverter that directly drives a pair of 6Y6 tubes operating in UL mode. He discusses the general use of the circuit with other output tubes, but makes no mention of any concerns for unequal clipping. Dollars to donuts his circuit displayed the same clipping characteristics -- but they are never mentioned in the article.
Dave