CatSkin -- Class A2 SET with Secondary Flux Cancellation

BinaryMike

Pelagic EE
2017-07-19 016.JPG

CatSkin grew out of a fascination with MCM’s very low cost 555-7125 “30W” speaker line matching transformer, which was used successfully as an audio output transformer in a couple of earlier push-pull triode power amp projects (Perveant & Bonseki). Because of its excellent high-frequency characteristics, I began to wonder if this transformer could be shoehorned into an even less likely role; as the OPT in a five-watt single ended triode amplifier. MCM’s low-frequency spec for this item is 50Hz, which, by the power-proportional-to-frequency-squared rule, should extend down to 20Hz at 5W. MCM doesn’t disclose how the 50Hz cutoff was determined, but a little negative feedback can do a world of good in this department as long as the core doesn’t saturate. It’s not challenging to find power triodes with anode resistance in the 400~500Ω range, to work with the MCM transformer’s 1,000Ω nominal primary impedance, although this impedance reduces considerably due to insufficient inductance at the low end. The biggest problem is that single ended amps task the transformer with a formidable magnetic bias, which normally compels the use of gapped cores, and that’s akin to sharkbite on important aspects of transformer performance. One would think that cheap transformers are tragically unfit for this job.

Well, as my dad was fond of saying, there’s more than one way to skin a cat. I chose the outlandish solution documented here because I enjoy hacking my way through brambles and learning a thing or two along the way. In the end, I produced an amplifier with excellent sonic qualities and a bit more power than initially planned. A brief investigation of low-frequency distortion effects of DC bias in the 555-7125 transformer was published over at diyAudio in December of 2015. A follow-up description of CatSkin amplifier breadboard results was presented in June of 2016. My decision to publish the end product here was influenced mostly by the convivial atmosphere.

CatSkin isn’t a project for beginners. It does use inexpensive components, but it’s more complex than most SET amps in this power class and it lacks refinements that could improve reliability as well as temperature stability. At this juncture, CatSkin is more like an advanced experiment or early prototype than a production-ready design. Better output transformers could improve low-frequency performance, but would likely require major changes to the flux cancellation circuit and may not improve sonic results because the high-efficiency speakers normally used with SET amplifiers don’t often perform well below 40Hz. Advanced DIYers who are curious about the sonic impact of magnetic bias in OPTs could find a lot to interest them here.
 
Amplifier Circuit Description

A 6J6 twin triode with common cathode serves as a differential input stage. The cathode constant-current source is a standard complementary BJT type that puts 7.7mA through each side of the diff-amp. The 6J6 anodes see B+ through 57.6Ω load resistors, which develop transistor base bias and essentially clamp 6J6 anode voltages to the V+100 supply rail. In this way, Miller effect is eliminated and the 6J6 is forced to operate in a very linear pure transconductance regime, much like the lower section of a cascode pair. The input side anode directly drives the base of PNP transistor Q1 in a ‘multiplied transconductance’ arrangement (due to Frank Blöhbaum; see Linear Audio magazine). The feedback side anode resistor exists only to confirm symmetrical current flows. Q1 collector is loaded by a second CCS made with Q6 and Q8, then power-buffered by Sziklai pair Q3/Q2, an arrangement which yields extremely high voltage gain as well as high output drive capability. The buffered signal drives the final grid directly, but also reaches the diff-amp feedback side via R7. There is so much gain in this internal feedback loop that distortion due to transistor characteristics is driven into the noise floor and only ‘triode sound’ remains. The feedback network at V2B grid has two additional inputs, one to control power tube grid bias and one to establish 12dB of global NFB.

CatSkin’s output stage is a Russian 6P36S TV horizontal output tube, triode-strapped, running near 20W total dissipation at 100mA and 200V of B+. 20W is a bit over spec, but power tube specs for TV service are often understated relative to audio service. John Stewart, writing in the August 2004 issue of AudioXpress magazine, states that sweep tubes in audio service can typically dissipate 40% greater power relative to their TV service ratings. For class A2, this operating point puts maximum midrange audio power output near 10W. Grid drive peaks approach +16V at the onset of limiting. The driver circuit performs so well that there is no evidence of notching in the distortion waveform as grid current commences, which is at roughly 5W output. Because overdrive can easily take the final grid far above +16V, some form of drive limiting ought to be implemented. Without this feature, CatSkin demands diligence from the user at all times.

A third CCS, similar in structure to the others but made variable and with high power capability, is connected to the secondary side of the OPT in order to cancel magnetic flux induced by primary current and make the OPT behave as it would in an AC-only circuit. Current from this CCS is restricted to a portion of the secondary winding that is not connected to the load, so there is no DC offset voltage presented to the load. The CCS never affects CatSkin limiting behavior because its compliance range comfortably exceeds peak audio voltage at 10W output. This portion of the secondary winding is essentially unloaded for AC and was found to resonate strongly, hence the damping network of 27Ω and 220nF connected there.

Heatsinks for the CCS power transistors were procured from a surplus dealer online and turned out somewhat undersized for the job. To reduce total thermal resistance from junction to ambient, I mounted the transistors with thermally conductive grease only, and electrically isolated the heatsinks from the chassis. As a result, audio voltages up to about 30V peak are exposed to the user. Complete stabilization of power CCS current takes half an hour or more due to the long warmup time of the power transistors on their heatsinks. Thermal compensation of this circuit would be a significant improvement.

The calculated secondary bias current to null magnetic flux in the core is 623mA. I included a precision 1Ω sensing resistor to help measure and set this current accurately, but the 1.5Ω emitter resistor is probably sufficient for this purpose if it’s a close-tolerance device. Q10 base current adds only about 1% to emitter current. In addition, I found that calculated flux null current isn’t necessarily ideal. If I adjust CCS current for minimum THD at 100Hz and 1W, for instance, then it will be close to 700mA. This finding opens up a field of investigation that I’m not currently ready to pursue, but I’ll be most interested to hear of results from anyone with sufficient motivation. I retain no intellectual property rights herein.
 

Attachments

  • catskin-bb-c-Amp.pdf
    20.8 KB · Views: 59
Last edited:
Power Supply Description

CatSkin requires an awkward collection of operating voltages. I solved this problem by potting three low-cost toroid transformers into a single die-cast aluminum enclosure that is mounted atop the chassis by its lid attachment screws. Thermally conductive potting compound and over-specified toroids were used to insure that heat buildup wouldn’t be excessive. A twin topside enclosure houses the OPTs and low-voltage power supply circuits, freeing up space for high-voltage power components under the chassis.

The main 200V B+ rail derives from an 80VAC winding with a full-wave doubler rectifier and two additional R/C filter stages. The low value resistors and high value caps used here demolish all arguments for choke filtering. A tap on the voltage doubler is filtered to obtain the V+100 rail. The V-100 rail comes from a half-wave doubler connected to the center tap of the 80V main B+ winding.

V+28 power for the big CCS is produced by a full-wave CT rectifier on a 48VCT winding which also sources power for the low-current V-26 rail through the remaining two diodes in a chassis-mounted bridge rectifier. Because resistive filter power losses would be too great, the high-current V+28 rail is CLC filtered using a small surplus iron-core choke that measured about 13mH without bias current. In-circuit ripple reduction is much greater than expected, but anything over about 15mH should be sufficient to eliminate audio hum from this source. If your stash yields nothing suitable, then Triad’s C-56U looks like a good candidate for this chore. The V+28 rail is further filtered to remove all traces of ripple on its way to the bias adjust trimmers. This voltage is positive even though power tube grid bias is negative, because the feedback input to V2 is inverting relative to final grid drive. It was purposely left unregulated in order to minimize output stage bias current changes due to AC line voltage drift.

Snubbing networks of 1nF and 1K are applied to both rectifier windings, and were found to suppress the recovery spikes of commodity rectifiers so well they’re undetectable with a scope in the signal path. There’s nothing remarkable about the 6.3VAC heater supply aside from the 50mΩ resistor that was added to bring heater voltage precisely on target in my environment. The power supply schematic page includes a rear-panel ground terminal and a six-pin Mini-DIN connector for test access. I built a little test adaptor that plugs in here for convenient voltmeter hookup. If you build CatSkin on a bigger chassis, the trimmers and test points could be made accessible topside.
 

Attachments

  • CatSkin 1-0 PSU.pdf
    21.4 KB · Views: 40
Performance

Low-power frequency response is remarkably flat and phase linear. High-power low-frequency response is limited to about 40Hz at 5W. This calls MCM’s transformer specification into question, but of course the selling price was an obvious clue. At 1W, distortion kicks in around 15Hz. 1W high-frequency response is down 1dB at 30KHz and down 3dB at 55KHz. Here are the 100Hz and 10KHz squarewave results at 1W:

tek00000.png

tek00002.png

The distortion profile at 1KHz shows second harmonic roughly 40dB down at 1W and subsequent harmonics decaying rapidly, with only third, fourth and fifth visible above the noise. This measurement was taken prior to my last noise reduction efforts, so it’s conceivable that additional harmonics would be visible if the measurement was repeated:

CatSkinBB-1W.png

CatSkin is stable with test loads of 100nF, 470nF, and without load. Input sensitivity is 707mVrms for 5W output, or 1Vrms for 10W out. Wideband unweighted noise measures below 1mVrms at both speaker outputs. Damping factor hasn’t been measured as of this writing. Calculated output impedance is in the neighborhood of one ohm, which would put damping factor at eight.
 
Last edited:
More Photos

2017-07-19 018.JPG
Rear Panel

2017-07-19 006.JPG
Power CCS transistor mounting detail A

2017-07-19 012.JPG
Power CCS transistor mounting detail B. Rectangular brown shoulder washer is a standard TO-220 device mounting hardware item.

2017-06-12 018.JPG
First pour of potting compound

2017-06-12 011.JPG
Chassis top detail

2017-07-24 001.JPG
Chassis bottom detail A

2017-07-24 003.JPG
Chassis bottom detail B
 
Top notch stuff. Your build quality and engineering/design effort is absolutely first rate. I have also experimented with the MCM line matching transformers just a bit, mine with their 10W unit in push pull config as a replacement for output transformers in a EL84/6V6 PP AB1 stage. Worked surprisingly well down to 150 Hz or so.

Just curious why the Sziklai pair and not just a normal Darlington?
 
Excellent workmanship and documentation. Thank you for sharing your project. I will reread again when have more time but a lot of good information and ideas here to digest.
 
Very interesting way of making the transformer do something it was never meant to do. I'd heard of using dummy load resistors on half of a transformer to run single ended, basically relying on the resistor to simulate the missing tube. Can't say I've heard of using part of the secondary to cancel the flux out like this.
 
Just curious why the Sziklai pair and not just a normal Darlington?
The Sziklai pair generally has superior linearity, so I didn't consider using a Darlington. I did, however, sub in a MOSFET source follower at one point during development. No difference in performance was noticed at the time.
 
I'd heard of using dummy load resistors on half of a transformer to run single ended, basically relying on the resistor to simulate the missing tube.
The resistor method is going to severely cripple performance, but it's reasonable to make a single ended triode amp with a PP output transformer by using a pentode or MOSFET as CCS on one side of the primary. The triode will see 1/4 of the transformer's rated end-to-end primary impedance, of course.
 
Thanks for a beautifully presented project that would have been a featured article in the old 'Audio Amateur' magazine. The engineering is creative and the construction techniques excellent. I will return to both the schematic and the photos because I know I can learn from them for my own future projects.
 
I wrote one or two articles for The Audio Amateur back in the day. I was reminded of this when I unpacked an old Sony open-reel tape deck from storage yesterday and found barnacles on it --- actually parts of an optical sensor that was once connected to a Commodore computer to display tape run time in actual minutes and seconds! The interface circuit and software source code for this accessory were published long ago...
 
Top notch stuff. Your build quality and engineering/design effort is absolutely first rate.

I would like to echo Kevin on this point. Absolutely beautiful work, Mike. I marvel at your build precision, and I always learn so much from, and am inspired by, your threads. Kudos, sir.
 
Fascinating concept Mike! And superbly executed as usual. Your design makes for a very tidy way of dealing with with two of an SET's biggest performance issues: power and OPT limitations. Would using the FC circuit be a superior way to implement the use of a PP transformer in a SE circuit, or is the use of a CCS on the unused side of the primary winding just as effective? (6 to 1/2 doz of the other)

Dave
 
Would using the FC circuit be a superior way to implement the use of a PP transformer in a SE circuit, or is the use of a CCS on the unused side of the primary winding just as effective? (6 to 1/2 doz of the other)
Flux cancellation should be equally effective on any winding because it's DC, but I did encounter an unexpected issue with resonance in the unterminated (for AC) winding that I used in CatSkin. It was bad enough to exceed CCS compliance by a wide margin when tested with squarewaves at high power. The problem was largely remedied with an RC damping network in this case, but it makes me wonder what would happen with other transformers and other windings. I would strongly recommend the breadboard-first approach for anyone attempting variations on this concept. It's also worth noting that if damping networks are necessary to kill resonance, then it's pointless to expend effort on HF optimization of the CCS.
 
Last edited:
Mike, I just want to make sure I understand the concept of this flux cancellation. The CCS provides constant current through the upper half of the secondary winding; not changing with respect to the program signal, correct? So even at zero program signal, I assume the CCS still passes some (large-ish?) amount of current through the upper half of the winding? If that's true, then I assume the goal is to cancel the most flux at max power output?

In fact it occurs to me that the CCS approach is probably the only real way to do this--cuz if it wasn't a CCS, i.e, it was some sort of opposite phased AC signal as compared to the SE signal so as to minimize current needed to maximize flux cancellation at any signal level, well then it'd just be some sort of poor man's PP amp. :(.

So, if you used this approach on a PP output transformer used for SE duty, I assume you'd need quite a bit less current to achieve the same result if the CCS was employed on the unused half of the primary. i.e., you'd need to match the max current capable of being delivered at the primary to cancel equally the same level of flux?
 
The CCS provides constant current through the upper half of the secondary winding; not changing with respect to the program signal, correct? So even at zero program signal, I assume the CCS still passes some (large-ish?) amount of current through the upper half of the winding? If that's true, then I assume the goal is to cancel the most flux at max power output?

Sometimes it helps to visualize the CCS as a very high value resistor connected to a very high voltage source. It forces a constant current regardless of audio signal voltage excursions.

To cancel magnetic flux in the OPT core, the fundamental requirement is equal and opposite ampere-turn products. If you achieve this by using a center-tapped primary winding with CCS on one side, then CCS current must equal output tube bias current and CCS voltage compliance must accommodate the full peak-to-peak excursion of the output tube anode. A CCS is effectively an open circuit for AC, and if you don't use a CCS (or equivalent) in this role, then you're adding load to the amplifier.
 
Last edited:
Back
Top Bottom