Transformer Output SET Line Stage



Some time ago I purchased a Lundahl LL1660/18mA transformer with the intent of building a transformer output line stage. Per Lundahl had provided a simple schematic diagram (Figure 1), which I breadboarded. It sounded pretty good but before I committed it to a final form I had a couple of questions to resolve. It all started simply enough – which was better, a voltage-regulated filament or a current-regulated filament?  I had also seen some discussions of using a fixed-bias instead of cathode-bias for a preamplifier. I had used both types of bias on occasion in power amplifiers, but I had never used fixed-bias in a preamplifier before. One problem that I have always had with resolving these kinds of questions is that you compare two things in similar but different situations, like “this amplifier has fixed-bias and that amplifier has cathode-bias.” I could never be sure if the differences that I heard were due to the bias mechanism or to some other, unrelated factor. Alternately I have taken a single amplifier and after listening to one configuration, modified it and listened to a different configuration.  The problem with this is the time-lapse between one listening and the next. Sometimes if a difference was gross I could easily detect it. However with more subtle changes I could never be sure. What I needed was a simple way I could isolate and quickly compare two things, which I accomplished with the circuit shown in 
Figure 2
. Although there are four tubes shown in the schematic, in reality there are four tube sockets but only one has a tube in it at any time. A good majority of the circuitry is common to all configurations and I can compare two things as quickly as I can move a tube from one socket to another. In position 1, the tube was cathode-bias with a current-regulated filament. Position 2 yielded a voltage-regulated filament with cathode-bias. When a tube was in position 3 it had a current-regulated filament and fixed-bias, while position 4 provided fixed-bias with a voltage-regulated filament. Figure 3 shows the regulators I used in all of the test beds. Early on I used a variable power supply to adjust for the slight plate voltage difference required between fixed-bias and cathode-bias. After a while I learned that such slight voltage differences were inaudible. 

Just about the time I finished this prototype the Piedmont Audio Society had a meeting focused on presenting preamplifiers that members had built. Although the prototype was a rat’s nest of wires and not the best-sounding unit there, members found the experiment very interesting. The general consensus was that current-regulated filaments sounded better than voltage-regulated filaments and cathode-bias was preferred over fixed-bias.

While the meeting answered a couple of questions, the members brought up many more:  Have you tried such-and-such a tube, or parafeed, or Ultrapath, or …?  Someone asked about how the regulated filament compared with old-fashioned unregulated AC filament. I had not thought of this; I just assumed that any regulated DC was superior to unregulated AC. While I had no intention of opening Pandora’s box and spending a lifetime researching options, some of the questions piqued my curiosity and there were a few questions of my own unanswered, for example, I wanted to experiment with the RC network on the output of the transformer.

The purpose of the RC network on the secondary of the transformer is to tame a peak in the frequency response of the LL1660 at around 80KHz. You can see the result of this peak on an oscilloscope as an overshoot or ringing on the leading edge of a 1 KHz square wave. The peak, which is caused by leakage inductance and distributed capacitance, is common in transformers. Design of the RC network is empirical, accomplished by substituting different values of resistance and capacitance while watching the resultant square wave on a scope. The objective is to find the largest value of resistance and the smallest value of capacitance that produces the best-looking square wave. The problem is – you don’t listen to square waves, and the best-looking square wave doesn’t necessarily correlate to the best-sounding music. So my process for determining the network is to start with the square wave to get in the ballpark and then go to listening to music while adjusting the RC values.  I found that I preferred a 2KW resistor and no capacitor at all. Adding a load on the secondary of a transformer will decrease the level of the signal but there is more than enough gain in the circuit so this is not a problem. We’ll come back to this network a couple of times in later articles of this series.  

I compared a regulated DC filament with an unregulated AC filament and was surprised by what I found. The sound of the unregulated AC filament was only subtly different from the sound of the current-regulated DC filament, and both were superior to the voltage-regulated DC filament. The current-regulated filament supply does have one other advantage – it protects the filament from over-current at startup, which can prolong the life of the filament.       

In a cathode-coupled configuration, the signal from the top of the transformer is returned directly to the cathode of the driver tube rather than through the power supply to ground and then through the bypassed cathode resistor. I believe that keeping the audio signal out of the power supply is one of the most important things in a good-sounding amplifier so I was ready to try out the cathode-coupled 
configuration. Figure 4 shows the modification that I made to the prototype for this test. I added a 400W resistor and 22mF capacitor that may be used either as an additional filter stage for the power supply or a cathode-coupled feed. With no signal flowing through the cathode resistor, when the capacitor is switched to cathode-coupled, the cathode bypass capacitor is shorted because it is not needed. In one view, the fixed bias configuration is always cathode-coupled because there is no cathode resistor. I threw the switch to the cathode-coupled position anticipating beautiful music and boy, was I disappointed. Instead of beautiful music, I heard a moderately loud hum. It took some probing and a lot of head scratching, but in retrospect it is clear what is happening. As well as returning the signal to the cathode, the capacitor feeds whatever noise is on the power rail directly into the cathode. The solution is to retain the cathode bypass capacitor. With the cathode bypass capacitor in place everything worked as I expected. The affect was not earth shaking, nor was it subtle. Rather it was a noticeable and positive effect that I recommend to anyone building a traditional transformer output preamplifier. Since the cathode-coupled capacitor is in the signal path, it should be of high quality. In the past I had tried poor quality electrolytic capacitors in the signal path and always regretted it. In this case I used a Black Gate.

Kevin Carter from K&K Audio, the Lundahl US distributor, told me that he knew of a couple of people who had used parallel triode sections of a 12AU7 to drive the LL1660 in a line stage. I decided to try it so I re-wired half of the sockets on the prototype to accept the 12AU7 for this experiment.  I’m sorry but the parallel 12AU7 was a clear loser – it lacked life and dynamics when compared to the 5687. Since it was time to listen to different tubes, I auditioned a couple of 5687s and a 7119 which all fell into the category of taste preference; I could have lived with any of them but in the final ballot I chose the Tung Sol NOS military 5687. I know that there are other tubes that I could have listened to, and later in the project I auditioned the ECC99 and 6H30. These are not all direct replacements for the 5687; you need to check the pin assignment and filament current on these tubes. 

Things were shaping up nicely: the sound of this line stage, while not the ultimate, could hold its own against most competition. I can recommend the design shown in Figure 5 to anyone who wants to build a traditional single-ended line stage. In my estimation its sound is clearly superior to any non-transformer-output line stage. Kevin Carter has what could be thought of a balanced version of this line stage in his system. It uses the LL1660 PP transformer driven by a pair of 5687s. This balanced version sounds somewhat better than the single ended version although the character of the sound is the same. Fernando Rodriguez had brought his Lynn Olsen design Raven line stage to the listening meeting and in my book this was “best in show.” The Raven is a balanced parafeed transformer output design. I speculate that the thing that makes the balanced line stage, and particularly the balanced parafeed line stage, so good is the separation of signal from power.