ZOTL and ferrites

[Nixie]

US patent 5,612,646 (PDF version) discusses Berning's ZOTL circuitry (there are actually commercial amplifiers), where instead of using a regular output transformer, the output tube is fed through a high frequency ferrite transformer by the load and MOSFET bridge. Begning claims significant improvements in performance as the high frequency transformer supposedly has little impact on audio frequencies (the switching frequency is suggested around 250 kHz).

A diyaudio member, hifizen, suggested to me that a better design would, instead of having a switching bridge in the signal path, use AM modulation, thus still allowing the use of a high frequency transformer instead of a regular OTL.

I'd like to get some opinions and discussion on the pros and cons of these two suggestions, as well as implementation caveats to watch out for (I actually intend to build one or the other, and have procured appropriate ferrite cores).

[smoking-amp]

Hi Nixie,

I think the AM modulation scheme will have a problem with demodulating the audio. If you use just diodes, then you have a single polarity only output. Could use a Mosfet bridge to get both polarities, but then you are back to the same parts count as the Berning scheme, and the zero crossings are at audio frequency instead of 250KHz.


A nice improvement on the Berning scheme can be had by going to a 2 phase arrangement with slowed down 250KHz edge transitions to avoid switching noise generation. (actually, no need to use such a high frequency with this appoach either) This requires doubling up the components, essentially two identical Berning units with the commutation diodes connected together at the tubes and the audio outputs connected together. Same LV power supplies. The switching of the two units is done 90 degrees out of phase using a set of Linear Tech ICs that control slew rates (both voltage slew and current slew controlled) to produce ultra low switching noise. The two overlapping phases allow one to stretch the transitions out with the diodes always selecting the phase at max voltage. No output filter (for HF suppression) required this way, and so no filter to interfere with feedback circuitry either.

One might also consider some other ripple free converter topologies such as the Cuk (pronounced shuk) converter. But more complicated.

Don

[Nixie]

I didn't quite follow the second part. The patent already has a dual circuit with the two outputs connected together in the second diagram.
Can you draw a sketch of what you mean?

[smoking-amp]

Hi Nixie,
The 2nd figure (actually labeled figure 1) is "doubled up" to handle the P-P case. For the two phase version, we double this again. Everything except the tubes, 70 & 90, and the power supplies, 28 & 88, get duplicated. Connect the two images together at the tubes and the power supplies, and the output, 44,46,42. The two osc in points , 61 & 62, would become the two 90 degree phase controls. Some simplifications are possible. For example, using center tapped primaries on the transformers, 50 & 51, allows cutting the number of transistors in half. (ie., CT config. versus full bridge config.) The output filter can now be just a "glitch" filter rather than a OSC freq filter, since there are no longer any "idle" crossover dead times at the OSC freq. in the output. (the 90 degree phases hand off control with overlap rather than dead time separation, so there are no longer any OSC freq. artifacts in the output)

Don

[Nixie]

Thanks! I'm probably half a year from starting work on this project, but I thought I'd get some discussion first (I've got a bit of a queue of DIY projects lined up here...)

[tubelab.com]

I did some experiments along these lines about 3 years ago. All that I can say is be prepared to offer up a lot of mosfets to the fire gods. I did manage to get sound, pretty good sound, but I got a lot more smoke. The thing would be playing along just fine, all fets were running cool, then bang, it would just blow up. I got tired of the smoke and gave up.

All of the cores that I used were robbed from computer power supplies. The circuit that I used seperated the power supply from the OPT. I was just working on a constant frequency OPT. The concept is outlined below. The best driver circuit used Microchip fet drivers in a floating arangement. I had a paper schematic, but I can't find it.

http://www.tubelab.com/SSopt.htm

[smoking-amp]

Hi Tubelab,

Interesting work. Did you have snubber circuits on the drive FETs to protect them from leakage inductance voltage spikes?

I like the Berning design better since it avoids the need for the HV transistors, and even eliminates the HV supply too. I could never understand how he got a patent on it though, since the "DC transformer" idea has been around for some time.

Another, maybe even better, way to do it is to use two Cuk converters (one for each output polarity like the Berning design) with the two tubes in the HV sides and the output in series with the LV power supplies and converters, just like the Berning design.

The advantage here is the Cuk converter can do ripple free conversion without the need for a 90 degree phase add-on circuit, or a low pass filter, to cover the dead-time slots. (one can still do the slew rate limited switching to avoid HF noise at the cost of a little efficiency) In addition, it allows variable impedance transformation by PWM control besides the basic turns ratio transformation. (it still is ripple free at less than 50% duty cycle) Imagine being able to "tune" your OT impedance! Only requires 2 LV transistors too for a P-P design. But does require considerable mastery of the HF transformer design, which is more complicated than usual.
I plan to build one of these this Spring. I would like to get a PC board made eventually, with the Linear Tech controller IC for slew rate limiting of the switching, ie. RF quiet operation.

Don

[Nixie]

Hmm, maybe you should start up a group order on the PCBs if the project looks promising.

With such variable impedance tuning, would it be possible to do it dynamically, so as to match perfectly the impedance curve of the load? That may be even better than the voltage-to-power amplifier mentioned here.

[smoking-amp]

Hi Nixie,

Hmmm, interesting idea on the dynamic impedance matching. I hadn't thought of that, but should be easy enough to do. The duty cycle, or PWM index, is just controlled by a voltage input to the PWM controller IC.

Trouble is that the audio signal is composed of multiple frequencies typically, and the speaker impedance varies with frequency. So there won't be an optimum impedance ratio to fit all the frequencies at once. Maybe for a multi speaker, frequency partitioned, system it would make sense to have it autotune.

If one used a constant power feedback system, as in the reference linked, it would likely skew the loudness of different frequencies playing simultaneously since it can only level the total power. One would need a powerful DSP chip to do frequency sensitive leveling in real time, where different frequency components would get scaled separately.

A "simpler" approach might be to just synthesize a fixed frequency selective filter with complementary characteristics to the attached speaker, and put it in the input signal path. That would work with an ordinary amplifier. Most practically done using a DSP chip again.

Don

[Nixie]

Well, one could argue that most distortion added by the analog chain can be dealt with by predistortion on the digital side.