This pair of 50 SE amplifiers was built by me some 15 years ago. The circuit is very stable and during the years of service there wasn't much problem except one event 3 years ago. The problem was caused by one faulty 80 rectifier where the filament broke and fallen onto one of the plates. It caused a short-circuit from the high-voltage winding of the mains transformer to the ground. I was using a 3A fast blow fuse and the short-circuit current caused by the faulty rectifier wasn't high enough to blow the fuse. Consequently the transformer heated up and eventually blew up the winding. The transformer was carefully built by James Audio but apparently it can't withstand such high current for a long time. Ever since that event I became extremely cautious about safety. If you ever built a tube amplifier, ask yourself a question whether or not the amplifier will survive when a rectifier goes to short circuit. This is a common problem for the rectifiers with hanging filaments. And many people including me believe tubes with hanging filaments sound better.

The transformers were custom made by James Audio according to my requirements. Back that time I knew Mr. Wu of James Audio very well and that's why he was willing to make the transformers for me. So many years passed and I wasn't sure whether or not Mr. Wu would still remember me and would be willing to take custom orders hence I searched the web and looked for alternatives. Reading through some local forums I found Mr. Lo of Look-T audio has a good reputation in building transformers. I called up Mr. Lo and Mr. Lo agreed to build two mains transformers for me. The transformers Mr. Lo sent me were in exposed core style(non-potted). The EI cores were exposed and the mounting screws go through the iron cores to the chassis. In order to mount the newly built transformers, a 90mm by 70mm opening on both chassis were needed. The chassis were made of 1.5mm thick 304 stainless steel and weren't easy to cut. It took me like a whole week to get the cutting job done. There is a noticeable problem with Look-T's transformer: there is a vibration sound emitting from them when load is added. this doesn't happen with James transformers. I phoned Mr. Lo and he suspected the reason James transformers don't suffer this problem is because they are potted. I had a similar experience with such vibration noise when I worked on Music and Audio magazine 20 anniversary 6L6 power amplifier which also utilized transformers with exposed iron core. This problem is especially annoying because my seat is very close to the amplifiers. It would be less severe if I sit farther from the machines.

 After the Chinese New Year break in year 2016 I decided to fix the problems with the 50 amplifiers and I also want to build the 810 single ended power amplifier designed by Arthur Chang, the former chief editor of the Modern Audio magazine. I will need in a total of around 10 transformers and hence I decided to ring James up to see if I can order a batch. In the phone call I sadly realized Mr. Wu of James Audio forgot me completely. Nonetheless he agreed to build the transformers for me.

I don't get much chance to mess around with Arthur's gang since 2005. However, through rapid phone calls I learned that they found many interesting stuff over the years such as balanced choke connection, Dual Layer Capacitors, clamp-on plugs. They also found many new components which sound good. Also Arthur told me the transformer built by James Audio improved a lot in the past few years. I will try put these new toys on my 50 SE amplifiers.

The goal of the modification is:

  • Solve the hum issue and the transformer vibration issue
  • Use an interstage transformer to drive 50
  • Install a nice looking current meter

The first problem I encountered was the negative bias circuit for 50. The working voltage for 50 is relatively high at 450V. It also needs a negative bias of around -80V. If biased by a cathode resistor the supply voltage would be around 550V. This limits the selection of capacitors for the power supply. Electrolytic capacitors with a voltage rating beyond 450V are rare. It is even rare for electrolytic capacitors with a voltage of 500V and beyond. With this route we will need either serially connect two capacitors or we have to use other type of capacitors such as PP or oil can. These are all quite big in size. For this reason I decided to keep the HV under 500V and hence a fixed negative bias circuit is needed.

At a first thought building a negative voltage supply is not a big issue. Using a diode-bridge and capacitors to get a negative voltage. Through in a Zener diode if voltage accuracy is desired. However, if you think further these approaches all have various problems. For example, the internal impedance is too high and it's not trivial to adjust the voltage. Adjusting voltage? It's easy you might say. Just use a LM337 and the problem goes away. Unfortunately LM337 can only withstand a voltage difference 0f 40V and hence is not suitable for this task. The supply voltage from the mains transformer is 160VAC and the DC output voltage can be as high as 220V. Although it's possible to connect LM337 in a floating topology but the 40V adjustment range is still there. And it is also a safety risk operating LM337 at such high voltage. Hence I decided to look for different approaches.

I did quite an intensive search on the Internet but couldn't find much on negative voltage sources. Let alone one that works at such high voltage. I then decided to design my own circuit with LTSPIPCE. Here is what I have:

The circuit is very simple. The negative input voltage is first connected to a Zener Diode D1 and then followed by a Darlington PNP transistor Q1. The purpose of D1 is reducing the voltage to the allowed range for Q1. Without the protection components this circuit only needs like 5 components. The regulation performance is completely dictated by the hfe parameter of the Darlington transistor Q1. The hfe value of the transistor I selected has a minimum hfe value of 4000 at room temperature. This value will rise if the ambient temperature increases. According the the LTSPICE simulation results, the output ripple is around 0.01Vpp with an input ripple of 50Vpp. With a voltage source of 100Vpp connected at the output side with a 5k Ohms resistor serially connected the resulting ripple is around 0.1Vpp. Not too bad for a such simple circuit. If this module needs to share with the same winding with the power tubes more Zener diodes can be serially connected at the position of D1 to reduce the voltage further. D2 and R3 protection devices. D2 makes the voltage across the collector and the emitter of Q1 never go above 120V. R3 provides a biasing path for the base of Q1 in case R2 is not installed. These might look redundant but this is my design philosophy.

Now the circuit is designed, the next thing is to realize it. I only need two of them so making PCBs is out of question. I built them on two experimental boards with 0.8mm pure silver wires as signal and 1.5mm silver wires for ground. The built module is pictured below. The connection terminals on the right-hand side from bottom up are 160V-0-160V AC input, R1-D4, potentiometer R2 and output.

The reason making R1 and D3 removable is for swapping different Zener diodes and resistors for sound tuning. Also it is possible to replace D3 with a vacuum tube voltage reference such as OB2.

After the modules were completely built I hooked them up with my oscilloscope to check the noise floor. This is something LTSPICE short at. The measured result is very satisfying. I measured around 30mVpp connected and also 30mVpp without the modules connected. This means two things: The first is the noise floor is close or below 30mVpp. The second thing is I need a better scope.

During the time building the modules there was a small episode. One module had a higher noise floor and the output voltage changes much faster compared to the other module when the potentiometer is turned. For a normal module the change of the output voltage should be very slow because C3 needs to get charged/discharged. The quick response of the output voltage indicates something is wrong with C3. I measured it and the value is below 1uF which is very low compared to the nominal value of 470uF. I swapped C3 and everything worked as expected.