A friend of mine (who also frequented this thread) recently brought me a 339i with hum issue to look into. I spent some time doing measurement and modifications. The resulting improvement was pretty significant. I hope the analysis and mods here can help other head-fiers to troubleshoot similar problems. It was done in a hurry and no pictures were taken. Some pictures below are from a fellow Head-fier here, hopefully he/she wouldn't mind.
Initial measurement shows the audio band noise to be at 1.5mV (max vol) and 1.2mV (min vol), mostly power-line related, which is borderline as a high-impedance headphone amp, and is not acceptable as a pre-amp.
Three major factors contribute to the hum:
- Wire routing
- AC heater supply
- HV supply noise
- Wire routing
The 3 major routing problems are highlighted in the picture below. The green line is the input-output loop. The power trans is offset in the loop, and the leakage flux cannot cancel out very well. For the shielded wires, the designer chose to connect only one end of the shield to ground, trying to avoid a ground loop. But the loop is still there, and the induced noise is applied to the signal line now that the ground is disconnected.
The cyan line is the output loop. It also picks up stray magnetic field leaked from the trans, and added it directly to the output.
The red line is the AC heater supply line. The designer seems to know how to twist the AC heater supply lines properly from the trans to the LED board, then changed to straight parallel lines pretty far apart from each other, a clear no-no for routing AC heater supplies.
To fix the wiring issue, the input and output wires were re-arranged into a "Y" topology, minimizing the area of the loop . I had to use new input cables as the old one was glued to the chassis. Both ends of the shield are now connected to ground. The ground connection between the input and output at the rear RCA connector PCB are cut, but the left and right channels still share the same ground. The big filter capacitors were moved slightly to make room for the cables to pass in between.
The excess ground wires were removed (see red crosses). The two red circles indicate unwanted connections between the ground wire and the mounting screws at those locations. Insulating tubes can be added to the mounting standoff to avoid making contact to the wire. The output wire from the 6080 to the load resistors are replaced with shielded wires, with both ends of the shield grounded at the closest locations (dotted lines). Two small patches of copper are exposed on the load resistor PCB near the mounting screws for the ground connection.
A long and thick black wire is added from the common ground of the headphone connector to one of the chassis ground points.
2. AC heater supply
The AC heater supply is the secondary contributor to the hum issue. There is always some coupling from the heater to the cathode. It changes from tube to tube, and may change as tubes age. I prefer a regulated DC heater supply over 'lifted' and/or center-tapped AC heater. However, the heater voltage from the power trans doesn't provide enough headroom for a regulator, so I made a compromise and only changed the 6SJ7 heater to DC. The original heater voltage was too high (7V) and not good for tube life. Resistors are added to bring both heaters down to 6.3V.
3. HV supply filtering
The designer did a good job filtering the HV supply with multiple stages of RCs. The resulting noise at the 6080 anode was about 140uV. Considering that most of it would still make it to the output, I changed one resistor to an electronic filter (a.k.a. capacitor multiplier). This reduced the HV power supply noise to below 40uV. An IGBT was used because I had a few lying around. Most high-voltage MOSFET should work just fine.
The result of the above mods is that the audio-band noise reduced to about 60uV, a 26dB improvement in signal-to-noise ratio.
The designer should get credit for finding a nice combination between the 6AS7G/6080 and the 6SJ7 in an unconventional triode connection. We can see that their 2nd order harmonic canceled out quite well until more than 10Vrms of output voltage into a 300 ohm load. This unit looks to be optimized for 300 ohm load, as higher (100k) or lower load would significantly degrade the THD performance.
If we take 10V as the rated max output voltage, the 100dB signal to noise ratio quoted in the spec requires the noise level at about 100uV. This unit originally didn't meet the spec, but can now pass with sufficient margin.
Further improvements:
There are things we could do to further reduce the hum, to the 20uV range that works comfortably as a preamp. First would be a DC heater supply. If we connect two heater windings in series, there would be enough headroom for a regulated DC heater supply. The downside is that it would also add about 8W to the heat dissipation.
The second would be adding a copper shorting ring to the power trans to reduce the leakage flux. A significant part of the remaining hum comes from the leakage flux hitting the 6SJ7 tube directly. Using a metal tube helps a bit.