Improvement in VAS

[adiaz]

Has anyone tested improvements in VAS?

I've always had doubts. What produces better sound? a transistor in cascodo mode or one transistor in darlington mode. See schematics images, figure C or D.

If has tried this, could tell his experience?.

When I mean better sound, I think it is very important slew-rate, distortion and linearity.

I read the schematic C do not often make great improvement in VAS, and usually a single transistor usually do the job well.

The schematic D, I think improves the slew-rate, linearity and distortion, I'm right?

Thanks.

 

[jxdking]

In my experience, C is better than D.

 

[tvrgeek]

I would agree C is a big improvement. Self discussed this at great length in Self on Audio.

C is not Darlington, it is an EF. Cascode only helps if the load is buffered.

A single transistor does not have sufficient gain to be used for feedback and does not isolate itself from the varying impedance load. Toss them into SPICE to see what I mean. We are taking orders of magnitude between a single and "C".

Two pole or TMC will also give some big improvements.

 

[sgrossklass]

As so often, it would depend on the ifs and buts - particularly the kind of devices you have available, closed-loop gain and input stage current.
C loads the input stage less, but the second transistor must be a good VAS type for best performance. It may also introduce significant amounts of common-mode distortion unless the emitter resistor is replaced by a current source, though this isn't as much of a problem at the 20+ dB gains of speaker power amps (a headphone amp would be a different story).
D has the usual benefits of a cascode (i.e. you can combine a high-hFE lower transistor with a higher Va, higher power upper one), but is likely to require more base current.
Both approaches can also be combined, but obviously, the more transistors, the more likely you are to get into trouble with parasitic oscillation.

I am pretty sure that the clipping behavior of the various types has been discussed here in the past.

BTW - .bmp? Now that's a throwback to the '90s. I would suggest going to the trouble of converting to PNG, which shrinks these to about 11-12K each (and would have them immediately visible here).

 

[adiaz]

As so often, it would depend on the ifs and buts - particularly the kind of devices you have available, closed-loop gain and input stage current.
C loads the input stage less, but the second transistor must be a good VAS type for best performance. It may also introduce significant amounts of common-mode distortion unless the emitter resistor is replaced by a current source, though this isn't as much of a problem at the 20+ dB gains of speaker power amps (a headphone amp would be a different story).
D has the usual benefits of a cascode (i.e. you can combine a high-hFE lower transistor with a higher Va, higher power upper one), but is likely to require more base current.
Both approaches can also be combined, but obviously, the more transistors, the more likely you are to get into trouble with parasitic oscillation.

I am pretty sure that the clipping behavior of the various types has been discussed here in the past.

BTW - .bmp? Now that's a throwback to the '90s. I would suggest going to the trouble of converting to PNG, which shrinks these to about 11-12K each (and would have them immediately visible here).

Is true, put schemas in png. I'm talking in power amplifiers that will drive power lateral mosfets.

I thought that was the best option D (mode cascode).

I am carefully reading their reviews.

 

[CBS240]

You are talking about the VAS. Voltage amplifier stage is not meant to drive power devices, mosfet or BJT. This is why you must have a driver stage.

Ignoring the type of global loop compensation used, the buffered VAS (first pic) will provide the IPS with a higher Z load which makes it happy.🙂 The cascode VAS will have a higher output Z than the buffered VAS. Cascode VAS required lots of driver current gain from the driver stage because it has a much higher output Z. However, linearity of the cascode (common base) VAS is much better, just like a common grid triode amplfier. It depends on the surrounding circuitry as to which one will provide better results as a whole. For every action there is a reaction. For every pro in a particular circuit there is a con. Personally I like the cascode VAS, but there are drawbacks. It requires a larger voltage to the VAS circuit in order to provide full output voltage swing to the OPS. Combining the two circuits, a buffered cascode VAS, is not a bad solution either but there is still the penalty of driver current gain to suffice.....such as a tripple output EF. This is where a local error correction circuit within the output stage might come in handy because it has within the output stage it's own feedback (feedforward) nested loop surrounding the output stage. Besides, most of the non-linear distortion comes from the output devices. Generally adding an extra stage of 'small signal' devices within the global loop is of little consequence of propagation delay in an audio amplifier when considering the improved linearity of the VAS.😉

 

[BV]

Combination of both (with Hawksford cascode), works great. Compensation not showed.

 

[Vostro]

Hi BV theres been discussion about 3 transistor VAS like in your post and the problems of getting it stable.

Can you please show how you would compensate this arrangement, and have you found this arrangement to be reliable and stable yourself.

Also in picture you use a 1mA current source, is this the typical current you use for this application ?
Thanks

 

[BV]

I am using TMC compensation, no problems with stability in realization. LED current is not critical and 0,5-1mA are sufficient . But current source is a must for good PSRR. Or use 2 resistors in series with midpoint blocking to -Ub, (e.g 2x22k, 47u).

 

[AndrewT]

D.Self tells us that both are very good.

It seems he chooses the EF because it uses fewer transistors, when a buffer is required after the cascode.

If one has already committed to a pre-driver output stage, then both EF and cascode use the same number of transistors.

 

[BV]

Example in attachments.

 

[John_wr]

Type C may have problems with local parasitic oscillation. You need to have a carefully laid out PCB, use the correct transistors, then do testing with a scope. I believe D.Self always uses emitter degeneration resistors and current limit protection with this circuit in real applications. (Maybe some other tricks as well to ensure stability)

Personally I prefer not use this circuit presented as "Type C" due to my concern about parasitic oscillation. I would rather connect the two transistors like a regular Darlington - where both the collectors are tied together as the output of the VAS. Even wired as a regular Darlington you can still get some advantage in reducing the non-linear C-E capacitance if you choose appropriate transistors for the driver device (with sufficient voltage rating). You still need an emitter degeneration resistor and current limit protection for the vas in a practical design.

 

[BV]

Type C may have problems with local parasitic oscillation. You need to have a carefully laid out PCB, use the correct transistors, then do testing with a scope. I believe D.Self always uses emitter degeneration resistors and current limit protection with this circuit in real applications.

I used this arrangement in hundreds amps, all works reliable . Of course final values must be be verified by test on real amplifier and by measurement, also PCB topology is very important.
And current limit (for VAS) is implemented (D1, D2, D3) in posted schematic.Try it in simulation..

 

[John_wr]

I used this arrangement in hundreds amps, all works reliable . Of course final values must be be verified by test on real amplifier and by measurement, also PCB topology is very important.
And current limit (for VAS) is implemented (D1, D2, D3) in posted schematic.Try it in simulation..

The comments I made were in response to the "Type C" in the original post. Your design for the VAS and current limiting looks nice though!

 

[Bigun]

My favourite so far is a combined Singleton input - EF bootstrap - VAS, with 3 devices all working together and with an integral two-pole compensation scheme. There is an EF buffer in front of the VAS, but it also provides a directly coupled bootstrap to the collector of the input device which greatly increases linearity. The TPC is also combined with the EF buffer.


I haven't seen this before so I claim the invention "Ingram's front end" 😀

See here: http://www.diyaudio.com/forums/solid-state/245619-tgm8-amplifier-based-rod-elliot-p3a-2.html

 

[vzaichenko]

I used this topology quite a number of times in different designs (see attached). Very successfully. In this particular case compensation is simple Miller - very light one. In some cases I used TMC. With the right compensation never had oscillation issues.

In case you nevertheless face oscillation in such schematic (due to bad pcb layout, etc.), there is a simple and effective way of addressing it (published by Bonsai):
(Cascode Oscillation in Audio Amplifiers)

The front EF ensures low load on IPS, cascode with mid-power transistor allows rather high rails voltage and controls the drivers well, also allowing some good high hfe transistor in VAS, taking all the power and heat away from it.

Cheers,
Valery

 

[vzaichenko]

My favourite so far is a combined Singleton input - EF bootstrap - VAS, with 3 devices all working together and with an integral two-pole compensation scheme. There is an EF buffer in front of the VAS, but it also provides a directly coupled bootstrap to the collector of the input device which greatly increases linearity. The TPC is also combined with the EF buffer.


I haven't seen this before so I claim the invention "Ingram's front end" 😀

See here: http://www.diyaudio.com/forums/solid-state/245619-tgm8-amplifier-based-rod-elliot-p3a-2.html

Very elegant one! 😎🙂

 

[adiaz]

Type C may have problems with local parasitic oscillation. You need to have a carefully laid out PCB, use the correct transistors, then do testing with a scope. I believe D.Self always uses emitter degeneration resistors and current limit protection with this circuit in real applications. (Maybe some other tricks as well to ensure stability)

Personally I prefer not use this circuit presented as "Type C" due to my concern about parasitic oscillation. I would rather connect the two transistors like a regular Darlington - where both the collectors are tied together as the output of the VAS. Even wired as a regular Darlington you can still get some advantage in reducing the non-linear C-E capacitance if you choose appropriate transistors for the driver device (with sufficient voltage rating). You still need an emitter degeneration resistor and current limit protection for the vas in a practical design.

Interesting response.
Have you tried using a single darlington transistor in VAS?
Works better sound than a single transistor?
That there are models darlington transistor with 100v Vce?

I think mostly of commentary option "C" is preferred that the cascode transistor.

I prefer not to make use of both circuits. I just need a little improvement to the option of a single transistor in the VAS.

I've also seen using a single mosfet transistor in VAS? Experiences? I think it's worse than a bipolar transistor ...

Thanks to all.

 

[forr]

My favourite so far is a combined Singleton input - EF bootstrap - VAS, with 3 devices all working together and with an integral two-pole compensation scheme. There is an EF buffer in front of the VAS, but it also provides a directly coupled bootstrap to the collector of the input device which greatly increases linearity. The TPC is also combined with the EF buffer.
I haven't seen this before so I claim the invention "Ingram's front end" 😀

Interesting. I am intrigued by Q3 collector connected to R7.
Can you tell more about the performances of the circuit ?

 

[jcx]

seems like a bad idea - a EF buffered VAS is good - provides isolation of output Miller, AC gnd of drive EF Collector also minimizes its Miller C effect

can't see any reason for actual Darlington connection or involving the EF buffer Collector parasitics in the compensation network - just AC gnd it