Well the Ember differs in quite a few points:
- The power supply voltage is higher (48V vs 24V) allowing for a much higher output voltage swing.
- The output stage is a class A-B power op-amp instead of a MOSFET source follower
- The bias current circuit is not a fixed voltage with a varying resistor (SR) but instead it has sortof a fixed resistor with a varying voltage.
- auto bias insted of manual
- auto bias heater slector instead of jumpers
The higher voltage rail is obvious. More available voltage swing a s a result.
A (class-A) source follower wasn't an option for Ember as the current needed to be even higher than SR and the voltage is higher as well increasing the static power consumption.
MUCH bigger heatsinks and power supplies would be needed.
The OPA551 opamp can handle 60V and can deliver up to 250mA which is what we were looking for.
It also has 1x gain as it is used as a voltage follower. The input voltage = output voltage except with more current available.
No 2V G-S 'loss' as with MOSFETs either.
No class-A so no excessive heat when idle either.
Robust output stage with current limiter.
In the SR the bias voltage of transitor (Q2) is constant (1.8V, D4) and thus the emitter voltage drop across the resistor (P3 + R13) is constant.
A constant voltage across a resistor means a constant current.
When you change the resistance (P3) the voltage remains the same but the current changes.
This allows the tube to be set at the desired voltage (for optimal symmetrical voltage swing at the output)
R13 is there to limit the current should P3 be set at 0 Ohm.
because there are many tubes with many different anode currents the adjustment range needs to be wide.... hence the 50k pot.
The current can thus vary between 25uA (0.025mA) and 12mA.
That current limit is just there to protect the transistor really and about 1mA is about the max we need. A range of about 40x
Instead of varying the resistance (easy with a potmeter) you can also fix the resistance and vary the voltage across it.
However, to have as little as possible noise (from Q2) the voltage across the emitter resitor needs to be at least 0.5V
With a 500x adjustment range the voltage setting range (when using a fixed resistor) would run from 0.5V to 20V.
This 20V would severely limit the output swing, so not possible without a nifty trick.
Here is where D3,D4 and R23 come into play.... which is the nifty trick.
With a 1.8k resistor you need 1.8V to reach 1mA but for 25uA the voltage would be 0.005V which means NOISE !
The trick lies in a varying dynamic resistance of D3+D4 and R23.
With say 1V over the 18k the diodes do nothing (needs 1.2V) so the resistance is constant when the voltage ranges between 0V and 1.2V
With a voltage of say 1.5V the current through R22 is just slightly higher BUT the 0.3V across the R23 adds.
With 2.5V over R22 its current is 0.14mA + 0.72mA through R23 is almost 1mA already.
We want to limit the max current (for reliability) and thus the voltage over R23 is limited by Z3.
This limits R23 voltage to max (10V (Z3) - 0.6V (B-E Q2) - 1.2V (D3+D4) = 8.2V = 5mA.
So a voltage range at b-Q2 of 1V to 3.6V gives us an adjustment range of 25uA (0.025mA) to 1mA without the addition of too much noise.
A 2.5V range for a 40x current range.
The 25uA tubes by themselves are quite noisy and that noise is higher than that of the current source.
The autobias by itself is now easy.
Lower voltage across the R22 = lower current
= higher plate (Anode) voltageAs you can see they work the other way around. One higher means the other lower.
Feedback time.
For this feedback Z4 is used. It always drops 22V.
As the current source varies 1V to 3.6V this means the output stage voltage = anode voltage will thus vary between 23V and 25V.
So around 24V which is the halway point from 48V and thus maximum available voltage swing.
The small 1V variance is not important as clipping levels will not be reached and when they do the small percentage makes no difference.
Of course this is overall feedback which we do NOT want for the audio signal.
This is where R21 and C8 come in. These form a low pass filter of 0.05Hz (24 seconds) so audio is not fed back. Just the average DC output voltage via a long RC time.
Therefor the DC output bias voltage will always set itself (slowly) to around 24V.