Continuing On:
The clipping occurs whether operating with cathode or fixed bias; the bottom half of a sine wave (1 kHz in this case) always clips early when the UL connection is made. This shot was taken as the top of the sine wave approaches the onset of clipping. However, a quick check with the pentode output stage connection shows this behavior does not occur in that mode, with clipping extremely well balanced in presentation when driven to clipping levels.
Back in UL mode, the NFB loop was opened so that a study could be made without that corrective element in place. The clipping remained. Further study revealed that the cause of the uneven clipping was due to one of the very circuit elements initially used to enhance the performance of the unit in the first place: convert the original Paraphase Inverter design into a Floating Paraphase design. It was found that when you bring:
1. A UL output stage,
2. Near full power clipping, and
3. A traditional Floating Paraphase inverter design that directly drives the output stage
..... All together in the same design, then early clipping on the bottom half of the waveform will result. You take out any of the three elements listed, and the clipping evens out on the tops and bottoms of the waveform. The problem was ultimately traced to the fact that a Floating Paraphase inverter does not present an equal drive impedance to the grids of the output tubes. More specifically, the impedance between the control grid of the lower output tube and ground is much lower than it is for that element in the upper tube -- this because of the NFB employed around the inverter section of the driver stage that turns the inverter into a floating type design. In pentode mode, that is not a problem. But when the UL connection is made, the lower impedance appearing at the control grid of the bottom output tube makes the screen grid more sensitive to the UL FB signal applied there, preventing that tube from clipping. In the upper tube however, the screen grid is not as sensitive to the UL signal, causing that tube to clip early relative to the bottom tube, and producing the display above.
This is perfectly normal behavior for pentode tubes: A pentode always produce more gain to a signal applied to its control grid when the impedance of the circuit powering the screen grid is made very low -- this because the low impedance prevents what would otherwise be a NFB signal from being injected into the grid were the impedance to remain high. This is the purpose of the screen bypass cap used in many small signal tube pentode stages. The scenario here is exactly the same, except that the roles are reversed. In this case, it is a signal being injected into the screen grid, with the impedance present at the control grid affecting how sensitive the screen grid is to a signal applied to that element.
What was determined then, is that optimum UL operation requires the drive impedance presented to the control grids to be equal, and for maximum effectiveness, as low as possible. As a quick check to see if this theory was correct then, please refer to the schematic provided for the modified amplifier when retaining use of the stock OPT. By attaching a 10 uF 450 volt cap between the bottom of the AC Balance control (where the 470K/3.3M resistors attach) and ground, the circuit immediately reverts to that of a standard Paraphase inverter design, with the AC Balance control still performing its duty quite nicely. This now creates nearly identical, albeit rather high impedance drive circuits to the output tubes, that could still be adjusted for a balanced drive. With the AC Balance control properly readjusted, that simple change resulted in this display at clipping:
Even clipping! So for UL operation, a Floating Paraphase inverter that directly drives the output stage is a problematic way to go. But what about the Split Load (Cathodyne) inverter? It too presents unequal drive impedance from its outputs. Since Dynaco used this type of inverter in all of it's commercial designs, and had it directly drive the output stage, shouldn't it create the same problem as well? Hafler/Laurent's solution to this problem was to always use the lowest practical plate and cathode load resistor values possible for the inverter stage, so that the effective impedance presented to both output tube control grids is low enough to prevent such behavior. The result is even clipping. But there's more.
From the data above, note that the low distortion quiescent current for UL operation came in at 46 mA per tube. But when the inverter was modified back to standard Paraphase operation, the optimum low distortion operating point was found to be 38 mA per tube, or almost exactly where the optimum pentode operating point was found to be (37 mA per tube). So the unequal drive impedance of the Floating Paraphase inverter not only upsets the dynamic balance of the output stage (at all power levels), but also upsets the optimum operating point as well. A couple of quick IM distortion tests were then run to see if they would supply further confirmation as to what was playing out. With the inverter modified to standard Paraphase operation, and the AC Balance control set for optimum, the following data was gathered:
1. With cathode bias, 8.2 watts equiv. power output was produced at 0.75% IMD (using 250Ω cathode resistor).
2. With fixed bias, 9.25 watts equiv. power output was produced at 0.50% IMD (38 mA quiescent current per tube).
So more power at less distortion with lower quiescent current was developed with both forms of bias when using the lowly Paraphase inverter circuit, which further validate the loss of performance when a UL output stage is directly teamed up with a Floating Paraphase design. This surely cannot bode well for the UL 6V6/6BQ5 circuits that both Acrosound and Dynaco show in their respective transformer catalogs, nor the design in Hafler's article, as all of these are of the type of design described here. No doubt however that the act of replacing an original pentode based OPT in a suitable commercial product with a premium UL transformer (as Hafler does in his article) would still produce significant performance gains.
But permanently reverting back to a standard Paraphase design in this scenario is not without its problems either: The adjustment of the AC balance control is quite sensitive in that mode, with even slight changes from optimum causing distortion to rise quite rapidly. Not everybody has a distortion test set handy to set such a control accurately. The topology of the ST-35 would seem to be the simplest and most effective approach then to directly driving small power tubes in UL from a phase inverter stage, without the problems noted previously. As I'm sure most will agree, such elegant solutions are a hallmark of Dynaco products.
For this project however, changing the phase inverter circuit to accommodate UL operation might be the last straw for some folks, as it would no longer represent anything resembling a Magnavox amplifier, but simply an ST-35 built into a Magnavox chassis, using 6V6 tubes. Fair criticism. But also the best solution for those committed to quality UL operation in the Magnavox chassis. Doubling back to pentode operation then, the following data was gathered:
1.
Cathode Biased Pentode Operation with 250Ω Bias Resistor:
THD @20 Hz = 3.5% at 9.68 watts. @1 kHz = 1.1% at 10.125 watts.
@20 KHZ = 1.65% at 8.20 watts.
IMD @ 10.34 watts equiv. power output = 2.0%
Cathode Biased Pentode Operation with 220Ω Bias Resistor:
THD @20 Hz = 2.8% at 9.9 watts. @1 kHz = 0.55% at 10.45 watts.
@20 KHZ = 1.10% at 8.82 watts.
IMD @ 10.60 watts equiv. power output = 1.10%
2.
Fixed Bias Operation:
THD @20 Hz = 3.0% at 10.60 watts. @1 kHz = 0.175% at 11.88 watts.
@20 KHZ = 0.61% at 11.28 watts.
IMD @ 11.47 watts equiv. power output = 0.58%
COMMENT: From a purely distortion standpoint, the big advantage to UL operation is THD produced on the low end at 20 Hz -- although, when pentode power output is reduced to equal that of its UL counter parts, the difference diminishes. And, at 25 Hz, pentode distortion is just under 1.0% at 11.52 watts, so distortion drops quickly as frequency rises above 20 Hz. With a power response of 10 watts RMS from 20 Hz to 20 kHz, and all distortion under 1% from 25 Hz to 20 kHz, pentode performance is hardly shabby, with many preferring the lower damping that connection provides that endeared them to the Magnavox sound in the first place. And, maintaining pentode operation of course helps to preserve some of the original Magnavox design as well.
Of course, there are other aspects of the amplifier that are not discussed here, yet deserve a quick mention:
1. Frequency Response: +/- 0 db from 20 Hz to 20 kHz, within in -0.5 db to 40 kHz.
2. Stability: Will tolerate any value of capacitive only output loading without bursting into oscillation.
3. 10 kHz square wave presentation:
4. And the performance as presented at the 4Ω tap:
It is performance like this that (in part) make these transformers far and away the best product available in their power and impedance class, whether for pentode or UL service. They are not inexpensive, but unlike the Raphaelite transformers, they are absolutely worth the cost of admission for the performance they provide. The performance figures quoted throughout this post are identical whether viewed at the 4Ω or 8Ω tap.
With that then, the amplifier has been all buttoned up (for now) in pentode mode, operating with fixed bias from a small bias supply installed earlier. As such, for a quality 6V6 amplifier of conventional design -- other than operating with fixed bias and only 12 db of NFB -- its performance is quite impressive, allowing it to rise considerably from it's basement dwelling location of Magnavox designs. A schematic will be presented in due course. For now then, those who are absolutely committed to operating these transformers in UL, the best recommendation would be to simply use the ST-35 inverter and FB/stability circuits, and operate the output stage with fixed bias. The results will be excellent.
Finally, for those interested, the following information was gathered with a 16Ω load presented to the 8Ω tap:
THD @20 Hz = 7.9 watts at 5.9%. @25 Hz = 8.41 watts at 2.5%. @1 kHz = 8.41 watts at .32%.
@20 KHZ = 9.0 watts at 0.93%.
IMD @ 8.50 watts equiv. power output = 1.0%.
A 10 kHz square wave presentation remains largely unchanged with a 16Ω load, as do response and stability characteristics:
All in all, not too shabby. A few final pics are presented:
Top side: The chassis controls now sport knobs.
Bottom Side: You can see the small bias supply added in the upper right corner of the chassis, powered by an appropriate transformer stolen from a discarded wall wart. This of course necessitated changing out all the Bias and DC Balance Controls for them to work properly with a fixed bias design:
And finally, a close up of the bias supply:
For now, that brings this Magnavox saga to an end. Time to enjoy some more listening!
Dave