Diamond Buffer /w "bootstrapped power supply"

[hwfanatic]

I came across a small modification to diamond buffer.
http://www.battletonphoenix.co.uk/buffer.pdf

It basically sets Vce of the output pair to a somewhat constant 3 V. The author (Julian Satchell) claims this modification removes "residual distortion behaviour" of the diamond buffer and increases PSRR.

Could somebody please explain this in more detail ie. if this works and why exactly?

[Bigun]

Well it looks to me as if the only thing that's different from a 'normal' diamond buffer is the additional power devices added to the power rails of the output stage - i.e. Q2 and Q6. The rest of it looks conventional.

So, the way I see it, these devices in the power rails are rather like Cascoding the output devices. Q2 cascodes Q3, and Q6 cascodes Q5.

Looking at Q2....

The current source J2 provides a constant current through LED1 and LED2, which are in series. This creates a constant voltage drop across the LEDs - about 3V. The voltage at one end of the LED string is the output, therefore the voltage at the top of the LED string is equal to the output + 3V. This voltage is fed to the base of Q2. Therefore, the voltage at the base of Q2 follows the output voltage + 3V. Q2 is wired up as an emitter follower, the voltage at it's emitter will track up and down with the voltage on it's base. The emitter of Q2 sets the collector voltage on the output device Q3. Hence the collector voltage on Q3 is around 3V above the output voltage. Voila, the C-E voltage on Q3 is held at around 3V.

The bottom half of the circuit behaves in the same way.

I'd suggest 3V is a bit stingy, something around twice that would be my preference, which could be accomplished with an additional LED or two in the LED strings.

The advantage of this cascode approach is that with a constant voltage across the output devices they will have less distortion arising from the Early Effect - in which there is a dependency on the conduction through the output device on the voltage across C-E.

Another advantage is reduced power dissipation in the output device. However, the cascoding device, Q2 will in turn dissipate the remaining power and it will have to be fully heatsinked.

[hwfanatic]

Thank you for a very informative reply. I've looked up Early Effect. Exactly what was missing to understand.

How serious is distortion due to Early Effect? Audio wise. Truth is this is the only attempt to counter it I have seen so far.

I'd suggest 3V is a bit stingy, something around twice that would be my preference

Why? Does it depend on Q3 or in general? Also if Q3 was a BD139 type, which device would you go for Q2?

[AKSA]

The primary distortive mechanism in diamond buffers is the variation in Vbe across both drivers and outputs with signal excursion.

This variation is driven by current through each device rather more than voltage across the device (Early effect). Early effect is about three orders of magnitude less than variations in Vbe with collector current.

This circuit concentrates on a secondary DB distortion; it would be better to devise a circuit which keeps collector currents in drivers and outputs tracking with signal rather than trying to keep Vce constant. That way it should be possible to almost null the Vbe variations on the working devices.

But it's certainly a smart approach, credit where its due......

Cheers,

Hugh

[Elvee]

Originally Posted by AKSA

it would be better to devise a circuit which keeps collector currents in drivers and outputs tracking with signal rather than trying to keep Vce constant. That way it should be possible to almost null the Vbe variations on the working devices.

You dreamt it, I did it:

Halve the diamond, double the performance!!!

UniGaBuf, a follower cut out to be a leader

Anything else?

[hwfanatic]

Originally Posted by AKSA

Early effect is about three orders of magnitude less than variations in Vbe with collector current.

Puts things in perspective.

Originally Posted by AKSA

The primary distortive mechanism in diamond buffers is the variation in Vbe across both drivers and outputs with signal excursion.

The only way to reduce variation in Vbe without changing topology would be? Icreasing quiescent Ic?

[AKSA]

HW,

The distortion is compressive, that is, as the amplitude of the input increases, so too does the Vbe of the dominant output device, and so the output falls further behind the input, compressing the signal.

On a DB, the output device is driven from the emitter of a reversed gender device, ie, the upper npn output is driven from the emitter of a pnp, which is fed from a CCS from the same rail. That means that as the output base current increases, more of the output of the CCS is siphoned off to the output device, and LESS is passed onto the driver.

In turn, this means that as the Vbe of the output device increases, if anything the Vbe of the driver decreases. If you examine the topology carefully, and restricting ourselves to just the upper half of the output stage (lower half operation is identical, merely transgender), you see that the signal pass first to the driver base, then to the driver emitter/output base, then the output emitter. And to make it all worse, we have an emitter resistor sitting at the output emitter.

When you trace your voltages you see that this configuration is not optimal. We actually want the Vbe of the two devices - driver and output - to match each other, and then some more to null the voltage drop on the output device emitter resistor. If driver Vbe goes down while output Vbe goes up, then these events actually make things worse and deliver even more signal compression, which, incidentally, shows as odd order distortion. Very bad.

In short, you need a circuit which INCREASES driver Vbe at a greater rate than the output device. This would mean that current through the driver must increase faster - that is, on a greater slope - than current on the output.

Think about this - it's a thorny problem. How would you do it?

Cheers,

Hugh

[hwfanatic]

I feel at a disadvantage at this point. Wouldn't know where to start. Maybe you can suggest some reading material so I can familiarize myself with all the different buffer variants. In detail. Thanks in advance.

[Joachim Gerhard]

Hugh, do you think about a supply bootstrap or a driven cascode ?
The circuit could be remodeled a bit so that the diamond floats on the cascode.
Other then that i find the general idea very good. For distortion it may not do much but for speed. I found a trend that fast circuits sound better. OK the ear is not that fast so maybe it´s the settling time is what we hear. the hickups so to say.

[Joachim Gerhard]

For example the Leds could be connected to the emitters of the output transistors before the emitter resistors. OK, the bias conditions have to be adjusted for the new condition.