So I’ve spent the last week looking at the data from the initial effort. I’ve retaken some data, analyzed the performance and setup, poured over data sheets and characteristics curves, listened to some music at a lot of different operating points, and I’ve drawn some initial conclusions. And I must confess I’ve gotten some interesting surprises.

This optimization effort has been decidedly different from the previous two for the 6V6 and 6L6. I think I did not fully appreciate the impact that the structural differences between Beam Power Tubes and Screen Suppressor Pentodes would have on the UL operation. Normally I look for very neutral color in a UL power stage. But after listening to several hours of various music, I have concluded that this power stage is anything but neutral. I also think I teased out a critical flaw in the topology I’m testing.

A short note on “Ultra Liner” topologies

It is important to understand, that the “UL” sets of plate characteristics which can be found on the internet are not like the normal sets of plate characteristics curves published in vacuum tube data sheets. The published plate characteristics are static data, meaning that they represent the operation of the tube with DC voltages applied, not with AC signals. Now normally this doesn’t matter because the shift in the characteristics curves and operating points, in either pentode or triode mode, with the application of signal are relatively small. In most cases they can even be mostly ignored.

However, the case of UL is fundamentally different. What we call ultra linear operation assumes that a portion of the AC signal on the plate shows up on the screen grid as well. This is something that doesn’t happen at DC. So the “UL characteristics” one finds are generally mathematical constructions for dynamic tube operation assuming this plate-to-screen coupling. However, the DC operating points for UL operation generally fall on the triode mode plate characteristics. This is because in a UL topology at DC (i.e. zero signal) the screen voltage is actually very close to the plate voltage (usually just a few volts higher due to differences in IR losses in the output transformer primary at different taps).

So in truth, the application of the “ultra linear” topology really shifts the dynamic plate characteristic curves (of constant grid voltage) to a more linear distribution across the load line. The percentage numbers cited (e.g. 25%, 40%, 43%, etc.) are the amount of plate signal being applied to the screen grid. The limits being 100% (i.e. triode mode where the screen grid sees all the plate signal) and 0% (i.e. pentode mode where the screen grid sees none of the plate signal or strictly constant DC). The screen signal is identical in shape (i.e. frequency content and phase) to the plate signal. Because the screen grid is a control electrode, but has a negative relationship with the overall plate current, this is a case of controlled negative feedback.

Back to the tested topology.

The transformer I’m using in my testing is an Edcor GXSE10-5K with an 8Ω secondary. This transformer has a 40% tap for “UL” operation. What this means in operation is that for every 10V of AC signal appearing at the plate (i.e. the plate to cathode AC voltage) then the screen will see 4V of AC signal (i.e. the screen to cathode AC voltage) from the transformer UL tap. This will tend to shift the control grid reference voltages (from those shown on the triode curves) along the load line to the right (i.e. lower plate voltages and higher plate currents) as the signal level increases.

And after looking at the data this is where my first surprise happened. With this transformer the tube was going into compression relatively early. It was definitely an improvement over triode mode (i.e. 100%) in terms of power and distortion, but not really close enough to pentode mode (i.e. 0%) to really linearize those control points and drive distortion down. Basically, the screen grid was getting too much feedback. It is my estimation that the EL84 would greatly benefit from a lower percentage UL tap transformer. I believe that a 20% to 25% tap would do a much better job of linearizing the EL84 UL topology.

So this was my major discovery. I am still going to publish the writeup and all the data for this study. However, I am going to caveat it with the understanding that the points are suboptimal from a tap percentage perspective.

Now 40% and 43% UL transformers are ubiquitous from most transformer suppliers. Tap values of 20% and 25% are much harder to come by. So for anyone wishing to use an available 40% tap UL transformer, this data should really help out. But I am going to be on the lookout for an equivalent output transformer with a 20% or 25% tap to collect some more data.

As always, questions and comments are welcome.