Monday, January 12, 2015

A Unity Coupled KT88 Amp with Plitron Transformers

I was always fascinated by the Unity Coupled output stage in old McIntosh tube amps. In many ways it is an incredibly elegant solution to many design problems in output transformers. Ever since I first studied the circuit I thought it would be really neat to build one myself, my way. 

Obviously, the most difficult design challenge to overcome is the large voltage drive requirement of the output stage. I got to wondering if it would be possible to build a Unity Coupled amp without a bootstrapped driver and if it would be possible to do that well enough that a global feedback loop wouldn't be necessary. Not that I have anything against a feedback loop, but negative feedback tends to linearize things at the cost of making overload behavior worse. If I can make an amp good enough without feedback I will do without it. So a few years ago I challenged myself to see if I could build such an amp without feedback, but I left my options open in case it was needed. Here was what I built:

I got the output transformers from Plitron and they cost a pretty penny but are really outstanding transformers. The impedance ratio is 4k:5 Ohms or 6.4k:8 Ohms. I was a little concerned about this at first. I don't like driving pentodes into saturation since that is really hard on the screen grid. I thought the impedance ratio was a bit high and actually corresponded a bit with the transformer designer and told him so. He thanked me for my input but stated that he would not be doing another design at a lower impedance ratio.

I eventually settled on making a floating screen regulator from a MOSFET source follower to reduce the screen voltage and make it so that the load line for the KT88s hits the knee of the pentode curves. I think that this is just a really healthy choice for the KT88s and will probably make them last a lot longer without any real sacrifice in output power. As far as I know I'm the first to use this solution on an amp like this. The simplified schematic doesn't show it but I added a current limit circuit on the source follower that will make the screen voltage sag if the screen draws too much current.

I have a source follower driver for the KT88s, not so much because they need it but because I wanted to present a very high input impedance to the 841s that develop the high voltage drive.

For the driver, I had some very unique requirements. I needed about 200Vrms (560Vpk-pk) of drive to take the output stage to clipping. That means I had to have a driver capable of dealing with an 800V or so B+ and be able to idle at 400V if I wanted some headroom. Most small tubes can't idle at 400V since it takes special cathode materials to withstand the ion bombardment that happens at these higher voltages. Looking at oxide coated cathode tubes only power tubes and a few selected types like 6SN7, 6BL7, could handle high idle voltages. I also discovered that thoriated tungsten filament tubes typically have higher idle voltage ratings since thoriated tungsten filaments are fairly resistant to ion bombardment.

At some point I came across the 841 which is a high impedance triode with mu = 30 that is meant for zero bias operation and needs to be operated in A2 to swing significant voltage. It has absurdly linear curves over a very wide voltage range, though:
I thought if it were driven with a source follower and loaded with a CCS it would do very well. I was not disappointed. This is the distortion performance at 230Vrms output running at 10mA and 400V idle:
Now you will notice a hump in the noise floor. I suspect that might be caused by the white hot filament in a high impedance tube but I really don't know that for sure.

Last of all is the input stage, which is pretty unremarkable. It is just a 6SN7 LTP with CCS tail operated in a very linear range. You may be wondering what the deal is with the pot. I knew that I would have to count on balance between 841s, which are made of pure unobtanium (I was able to acquire three sets for this amplifier with great effort). I added a pot so that I could attempt to introduce some imbalance in this stage to counteract any imbalance in the gains of the 841s.

All high voltage supplies are implemented as Maida-style regulators.

Anyway, it is time for pictures of the completed amp:





Kinda ugly inside. Oh well. I really love the glow of those 841s. They kind of steal the show from the KT88s.

Measurements:

I measured output impedance at ~1 Ohm. I'll look at doing some distortion measurements this weekend and update this page (hopefully).

As far as listening tests go, the amp just does everything effortlessly. It sounds great. The bass is strong. I can crank it to ear bleeding levels and it doesn't break a sweat. I'm pretty happy with the sound.

If I put my ear right in the tweeter I can hear a slight hiss, maybe something to do with that noise hump in the output of the 841s. I was hoping it would be dead quiet but apparently not. There is no hum or buzz though.

Oh, and I should thank Tom Christiansen for his excellent filament regulator boards and Guido Tent for the automatic bias board. I love that my unobtanium 841s have a soft start on the filaments and that I always have perfect balance on my KT88s. Both boards work well and are highly recommended.

Addendum 3/1/15:

More Measurements:

When I initially got this amp together, I flipped it right-side-up a bit prematurely. I was excited to have it finished and thought I would test distortion with it right-side-up.

I realized later that I had not adjusted the pots in the input stage plate loads to null out the gain imbalance in the driver tubes. The problem was that the amp was so heavy that every time I thought about flipping it over to make the adjustments, I just felt this overwhelming lazy feeling and flipped on the music instead. Besides, it sounded quite good despite making almost 1% distortion at 10W.

Anyway, this weekend I finally got back to it and I'm really excited with the distortion measurements with everything tuned up.

I measured:
0.025% @ 1W
0.15% @ 10W
0.22% @ 20W
0.24% @ 30W
clipping occurs at ~ 40W.

All were 2nd harmonic dominant, despite this being a push-pull amp. I could really get that 2nd really low by introducing imbalance in the phase splitter, but I never could get it under the 3rd.

All in all, I think these are pretty good figures for an open-loop amp.

Then I tested bandwidth. These output transformers are supposed to have ~500 kHz of total bandwidth per the manufacturer.

The amp was down 3dB at 120kHz. I'm guessing that the limiting factor in the amp is the 841 driver stage. I did some back-of-the-envelope calculations and it looks like since the 841 has such a high plate resistance, it only takes a 22pF load to put it 3dB down at 120kHz. An 801A would probably perform much better here in terms of bandwidth but they are too expensive and I don't really need to make an amp with half a megahertz of bandwidth.

I knew that would be the price of using a high-impedance triode as the driver, but the 841's curves sure are pretty. The other thing about the 841s is that I think they are the dominant noise source in the amplifier. I think those white-hot filaments combined with the high plate resistance aren't particularly good for low-noise design. This amp has a tiny bit of background hiss (with ear close to the tweeter), whereas other amps I have built have had none, and I used the same input stage I always use.

That said, I spent a bit of time listening after making the adjustments, which mostly just took out a bunch of 2nd harmonic distortion, and I think that modern pop or heavily electronic music definitely sounds better now, but simpler songs sound slightly less alive. I would say that songs with lots of electronic sounds and loud bass lines are now subjectively clearer.

My conclusion: The 2nd harmonic distortion can add a pleasant effect to sounds, but any non-linear transfer function, even if it only generates even harmonics, will generate IM products all over the place when you start mixing a lot of sounds together. Those IM products don't sound good and will muddy the sounds if you start mixing enough together.