Quiescent current balancing or auto bias?

In my Project 1 amplifier I implemented a balancing scheme that keeps the relative quiescent currents, through both halves of the push-pull, exactly the same. Earlier, I considered auto-bias schemes, such as the Broskie Auto-Bias Circuit or the TentLabs Negative Bias Supply. But in the end I opted for current balancing.

Quiescent current balancing is important

First, the boring background… skip to the next paragraph if you are not interested.

Many push-pull designs use a shared cathode resistor for both tubes. Matched tubes keep the quiescent currents a bit balanced. A bias balancing potentiometer does the rest. The disadvantage is that the tube’s characteristic changes over time and even changes when the tube heats up when the amplifier is used. That means you keep on adjusting every few months or so.

For my Project 1 amplifier, that is not what I wanted. I want the amplifier to be easy to operate and to need little maintenance while using it.

Menno van der Veen writes in his paper Secrets of Output Transformers (in summary):

For the low frequency response, the the primary inductance of the OPT Lp is important. The larger Lp the better the low frequency response of the transformer.

However, the larger you make Lp , the more sensitive the OPT becomes for an imbalance of the quiescent currents of the power tubes in a push-pull amplifier design. In practice this means: when you use high quality OPT’s with good bass response and a large primary inductance, you should pay special attention to carefully balancing the quiescent currents of the power tubes.

Bill Whitlock (of Jensen Transformers, Inc) further adds in his book Audio Transformers:

… magnetized cores will show significant even order distortion.

Menno demonstrates in his article Auto-bias largely improves base and micro detail reproduction and concludes:

This small difference current was able to DC-magnetize the core of the OPT and consequently the magnetic domains became extra bound to their position.  ….  If we apply the new auto-bias unit, then the quiescent currents are equal within 0.05 mA (or better) …. without any DC-magnetization of the core we can hear almost 20 dB deeper into the sound before any extra weakening of the micro details occur.

In summary, for a good low frequency response and good reproduction of micro details, I had to balance the quiescent of both tubes well.

Balancing or auto-bias?

Auto-bias maintains the quiescent current exact at the set value, nulling the DC magnetic field in the OPT. This is quite simple when the amplifier strictly stays in class-A. The challenge comes when the amplifier moves into class-AB. An asymmetrical voltage is seen across the (small) cathode current measurement resistor. In response the DC servo sees the average current rising and will strive to adjust it back to the set value. By doing so, the DC servo pushes the bias off its mark. The Broskie and TentLabs auto-bias address this by clipping the current sense signal. This keeps the bias on the mark.

Thinking about the challenge, I realized the goal is to precisely interlock the currents through both tubes, but the absolute current does not need to be maintained that strict. If the quiescent currents of both tubes rise or fall in concert, the average currents in the OPT still cancel out. The tubes themselves do not mind if their quiescent current varies a bit.

In addition, I wanted to use (a bit of) cathode bias. I gathered this would give a more dynamic and a softer overdrive. (My preference for my Project 1 amplifier.) A scheme using cathode bias and current balancing seems to fit my preference best. John Broskie (Tubecad) describes an interesting solution that is based on Alan Dower Blumlein’s historical garter circuit in his blog post 163.

Broskie: Blumlein came up with an auto-bias matching circuit that does not aim to set a specific idle current, but works to keep both output tubes conducting the same idle current; a current balancer, in other words.

The Blumlein circuit itself is quite inefficient, basically doubling up the total cathode resistors. Broskie’s approach uses more efficient approach with a pair of PNP transistors to measure the difference between the cathode voltages and pull up the bias voltage of the tube that draws too little current. At the same time the cathode resistor keeps the tube in track. Both tubes are constantly adjusting each others currents to match its own. Quite effective! As a bonus, this scheme works well in class-A but it works just as well when the amplifier moves into class-AB.

I tried the Broskie Cross-Refenced Auto Bias Circuit and it works pretty good.

Schematic showing quiescent current balancing between a pair of KT77 using PNP transistor difference amplifier

quiescent current balancing

This is how I solved it. For the explanation, refer to Broskie’s  blog post.

You’ll notice the negative Vbias injection point in the schema above. For the quiescent current balancing that is not necessary. However, I was unhappy about the amount of heat that is dissipated by the cathode resistors. I really did not want that heat production within the enclosure. But more so, I did not want the unnecessary loss of B+ voltage that cathode bias causes. Hence, I had to reduce the voltage drop across the cathode resistors (R53 and R63) and replace a portion of that voltage by a fixed negative bias source.

Hybrid fixed/cathode bias to the rescue

As said, I believe cathode bias causes too much heat and wastes B+ voltage. You lose some 23V and produce some 5 to 6W for both channels combined with cathode bias.

My approach is a hybrid fixed/cathode bias. I lower the cathode resistor for each tube (KT77) to 100Ω. At 60mA the voltage drop is 6V. The rest of the negative bias is provided by a negative supply. Result: only ~1.44W for both channels combined, one forth of 100% cathode bias.

This approach also now allows to separate the balancing of the relative quiescent currents between both tubes and the adjustment of the absolute quiescent current. I described this scheme below.

quiescent current balancing and regulation architecture (in ArchiMate)

Quiescent current balancing and regulation

This diagram shows both KT77 tubes are quiescent current balanced in real-time. That balancing is based on the schematic above. Furthermore, the controlled negative bias supply provides a negative voltage to regulate the average current (measured as an average for both tubes). The average current can be slowly adjusted in non-real-time as needed. So it actually works and sounds like a fixed bias, with a touch of cathode bias.

Thinking about it, how about regulating the dissipated power instead of the current? That would mean we also measure the B+ and aim for a fairly constant dissipated power. This power may deviate a few percent from its window before the controller adjusts it back. With regards to the dissipated power, the amplifier would be agnostic to the line voltage and thermal drift. Anyway, that is another post…

Negative power supply

I wanted the negative power supply to be simple and clean. A fairly large buffer cap and a potentiometer to adjust the voltage should do it. But, unlike a traditional fixed bias, which poses a minor load of the supply, the “balancing bias” scheme poses a variable load with some remnant analogue components. I wanted to provide the balancing bias with a clean and stable voltage. Furthermore, I did not want to see an influence between the channels. A buffer cap and a potentiometer is not ideal here. I needed a supply with a fairly low impedance. A shunt regulated supply is a good choice here. It gives a clean and stable voltage. And, a shunt regulated power supply has built-in overload protection.

Shut regulated bias suppmy

Shunt-regulated bias supply

This is it for now.

Next post I will discuss stability issues with fixed bias. A teaser: after a while I noticed the hard way that tubes driven hot with fixed (or mostly fixed) bias can cause a thermal runaway… In other words, over time the tube was biased hotter and hotter until the biasing got unstable and skyrocketed and a fuse was blown. It took a while for me to understand what was going on…. More next time…