Folded driver stage for the modular high end audio amplifier

This folded driver stage is based on my research for audio amplifier output stages. The folded driver is simple, performs reasonably well and from a thermal point of view, the driver transistors compensate the power transistors without requiring typical V_BE multiplier arrangements. I like to try this design and present a driver stage here for BJT output stages biased into class A.

This is the well-known basic Diamond Buffer circuit, a unity gain buffer often used in op-amp output stages.

This is a slightly improved variant of the circuit with reduced voltage swing across the driver transistors and lower distortion due to local feedback.

For driving multiple output stage transistors, the circuit could be modified like shown above and this is also the block diagram of the design presented here.

I don't call the driver presented here Diamond (anymore) since the circuit deviates considerably from the original design, but a lot of its original properties remain intact. After I spent some time evaluating the circuit, I found out that I reinvented the wheel when I stumbled across this exact circuit in Bob Cordell's book. Figure 11.16 (first edition of the book) shows exactly this circuit and he refers to is as folded driver. So I take over this name as well together with the small low pass filter from the output node to the common collector of the driver transistors. Since Bob mentions this circuit in his book and confirms that the basic theory behind is sound, I felt encouraged to move ahead with it. Bob proposed this circuit as driver for MOSFET output stages and it could work well in this application. In this case, the thermal compensation might no longer work and a bias spreader would be required.

The driver transistors are held at low and fixed V_CE. This reduces power dissipation and all kinds of non-linearity associated with varying V_CE of the driver transistors. Fast, low voltage and low power transistors can be used. However, since the driver transistors need to be mounted to the main radiator, the choice of transistors is somewhat limited.

Also, the collector-base capacitance of the driver transistors is constant due to constant V_CE and this presents a more constant capacitive load to the voltage amplification stage that is driving this folded driver stage. A standard dual emitter follower output stage has all transistors subject to very large V_CE voltage swing variations and therefore high influence from all kinds of V_CE dependent transistor parameter changes like early effect. Downside of this arrangement is that while V_CE is constant, this voltage is also very low. This is a difficult operating point for many transistors.

The constant current sources employed in this design are derived from my research I detailed here on my website. This type of CCS has very high PSRR, does not overshoot on transients and due to being cascoded, the CCS cascode transistors can be cooled on the main radiator without having influence on the standing current. As reference for the CCS, I used low voltage Zener diodes. This is not my preferred choice, but there are not too many options for this type of CCS for high current output applications like this one.

In any case, this driver module needs to be run from power supply rails that are higher than the output stage rails. Diodes D5 to D8 are for overload management. I don't expect D7 and D8 to ever conduct. Since they are held at constant voltage, they do no harm in any way. Diodes D5 and D6 are very important as they shunt excess input voltage to the output node. While this is not nice for the front-end driving the buffer, this avoids saturation of the buffer, which would not recover gracefully. This is just a fallback measure since all front-ends of the modular amplifier system are designed to avoid saturating themselves or subsequent stages.

R11 and C1 are optional and provide compensation of the driver to tame high frequency gain peaking. Resistor R12 and capacitor C2 further improve stability at high frequency. R12 needs to have low resistance to avoid excessive DC offset. For BJT output stages, R12 should be less than 10Ω. Capacitor C3 is optional and allows push-pull operation of the driver stage. The value of 470µF seems excessive. I prefer to reserve space for large capacitors and have the option to install smaller ones. This capacitor and the huge size is inspred by John Broskie's writing about Diamond buffers and improvement of Diamond buffers. I noticed that the ESR of this capacitor does not seem to be bad since this helped to tame oscillation that originated in the front end I debugged.

The bias of the output stage transistors is set by resistors R9, R10 and the standing current through the driver stage. This is a bit inconvenient as both resistors need to be of exact same value and resistor value tolerance also impacts DC offset. Biasing this arrangement into class AB with a BJT output stage is difficult with reasonable driver standing current. However, for an output stage biased into class A, the stability of the standing current is less critical. In a class A output stage, it does not really matter whether the standing current in the output stage is 3.9A or 4.1A. But in a class AB output stage, any deviation from the optimal bias point (the "Oliver" point) results in deviation of gm and causes more crossover distortion.

Due to the driver transistors V_CE being constant and the CCS being cascoded, the sensitivity regarding many transistor parameters is pretty low. This is a huge benefit in times when good transistors availability becomes increasingly problematic. I chose the TTC004B and TTA004B because they are convenient to mount due to being fully encapsulated. There might be more suitable transistors. For the CCS control transistors Q1, Q3, Q5 and Q7, any transistors may be used that allow sufficient I_C. Most important is that the CCS cascode transistors SOA fit the supply rail voltage and CCS current. The driver transistors Q9 and Q10 should have h_fe as high as possible.

Measurement at high current

DC conditions of the driver are pretty much as simulated. I measured 105mA standing current through the current sources and the voltage at the driver nodes to ground was 1.06V unloaded. I tested the driver module together with the four pair BJT output stage at +/25V power supplies, that dropped to +/- 22V being loaded. The test result was very surprising: The bias current reached almost 5A and therefore was much higher than anticipated (roughly 4A). The constant current sources surprised me being much more stable over temperature than I expected. I was also surprised that the small radiators HS1 to HS4 became very hot despite dissipating less than 1.5W each. And I was surprised how well the large radiator can handle roughly 200W of power dissipation reaching only 80°C. After some time of increasing bias current with temperature, there is an equilibrium. This indicates that the arrangement is thermally stable, but the delay of the control loop is very long.

Note that I recorded driver current and DC offset only at the beginning and at the end of the whole measurement procedure. R12 was 10Ω for this measurement setup.

Minimum current

Reducing the two 3R3 resistors R9 and R10 to zero Ohm would result in 1.3A bias current of the output stage. This means that there is roughly 1A/Ω added on top of 1.3A by R9 and R10. In order to reach roughly 4A OPS bias current as anticipated, 2Ω resistors should work fine and indeed, this sets the cold bias to 3.2A roughly.

Measurement at low current

During the first measurement setup I found the bias current in the output stage too high. Therefore I lowered resistors R9 and R10 to 2Ω and the bias current decreased considerably. The current reaches equilibrium at roughly 3.6A now and the radiator temperature at 70°C. This is enough bias current to support full class A operation from +/- 22V power supply rails even if the loudspeaker impedance drops to 6Ω. I noticed that equilibrium is reached much sooner than with the higher bias current. This is because the resistors R9 and R10 add a constant voltage to the output stage transistor bases, whereas transistors Q9 and Q10 add a temperature dependent voltage. Reducing the resistance has also reduced the percentage of constant bias spreading voltage.

DC offset has changed a little bit and this illustrates that the offset is affected by component tolerance. However, DC offset is pretty low and nothing to worry about. This can be either corrected by global negative feedback a DC servo. Setting R12 to 0Ω also helps to lower DC offset.

Summary

DC conditions differ from simulation a bit, but in general seem somewhat stable.

Testing AC performance is ongoing and interim results are promising.

The power supply rectifiers ran pretty hot during the test and 8A continuous current for a stereo amplifier might be too much stress. In case I really like to run this build in class A, I should drop in a power supply that uses active rectification to lower the rectifier power loss and dissipation.