Your opinions are sought on Audio Power Amplifier Design: 6th Edition. Douglas Self

Status
Not open for further replies.
The other approach is to prepare time domain distortion residual measurement with notch filter tuned at specified frequency, differential amplifier and oscilloscope. Then just turn on and turn off the sine wave and look if the distortion residual wave on the scope remains the same. It does not, and what you see is the changing bias with transistor temperature. It is almost impossible to build a thermal compensation that would be fast enough to follow and compensate sudden Vq/Iq thermal transients.

Pavel see attached, just what you mention. This is residual with two bias settings, +/- 4% of nominal (which is what I found a typical thermal shift due to signal dynamics).

jan
 

Attachments

  • jd v8 f17.jpg
    jd v8 f17.jpg
    645.5 KB · Views: 342
[...] I understand the superTIS is indeed fine, but it appears to be designed just for proving that TMC can, in a particular case, do better than TPC.

That's indeed what the title suggests (Where TMC really shines), but that was not the primary objective of that design. An equally applicable title could be: "How to get rid of a CMCL" or "How to tame a push-pull VAS". Anyhow, the primary ojective was: how to make a symmetrical front-end simpler wihout sacrificing the performance.

Overall, that's nothing outstanding in the superTIS performance, compared to a standard design with CMCL.
First, which standard design with CMCL? 🙄 Second, please read my chapter "What we have gained" and you will understand that the SuperTIS will outperform any other front-end in (almost) every aspect.

[..]The autobias output stage bias stability is very good, and as far as it doesn't start oscillating and subsequently latch up due to a fast input signal, the distortion performance is very good. It always happen only when the input signal is high (but still far away from clipping), even without any front end (input stage + VAS).
Hmm.. good news 🙂 and bad news.🙁

[..] I see your autobias (which is pretty much a high speed version of the Linear Technology LT1166) as just another error correction output stage, where the loop gain is created by a positive feedback network.
That's right, the bandwidth is much higher (10x?), but it doesn't rely on positive FB (like HEC), instead, just ordinary (negative) FB. Actually, it's a kind of reverse CFA configuration (sorry for the misnomer): inverting input tied to the base and non-inverting input tied to the emmiter. On the other hand, frequency compensation is the same as in Bob Cordell's HEC amp (i.e. feed-through) opposed to shunt compensation (less optimal!) in the LT1166.

If you could tame that transient instabilities in the output stage core, it would be worth exploring further, otherwise it will remain only of theoretical simulation interest...
Sure. I hope in collaboration with Paul.
That's all I know about, and please don't argue with me about these results, I'm only the messenger.
Understood.
[..]PS: I don't plan building any of your designs - I prefer to put my very limited resources in something else.
Of course. I wouldn't expect anything else. 😀

Cheers, E.
 
It is possible for this zero to move sufficiently low to make some difference if the transconductance of the IPS and the transconductance of the "VAS" are both simultaneously set to values that are atypical but feasible.

Otherwise said, of theoretical interest only. That's my point.

Regarding the TMC vs. TPC, give me a Blameless with an optimized TMC compensation and I'll give you a TPC version with the same loop gain and phase margin, within a very few %. Or the other way around. The fact that TMC can, in particular topologies (different from a Blameless) perform better doesn't prove that "TMC is better", I'm sure there are topologies where "TPC is better". From somebody that claims it understands the feedback theory, I would expect to think in terms of maximum feedback, rather than searching for odd corner cases that ultimately do not prove anything.
 
Last edited:
Hi Pavel,

Indeed, almost impossible, that is, as long as we are trying to do it the "thermal" way. So let's do it the "electrical" way, i.e. by means of an auto-bias circuit, which can be made way faster.

Cheers, E.

I agree Edmond, the 'thermal' way doesn't get that far. In my tests I went from heaksink-mounted TO-3 to 'air-mounted' TO-247 and the difference is very small - in each case you see the thermal delay from dissipation even to the diode on the ThermalTrak devices. Not sure I understand the exact mechanism though.

OTOH, with a few % transient shift in bias, should we worry anyway?

Jan
 
Fundamentally, in the Halcro, despite their sophistication, the input-stage (differential of Sziklai pairs, twice-cascoded, current mirror) and VAS (single-ended, cascoded) are of the same two-stages topology as the Blameless.

Absolutely true, but there are at least two more layers of complication: the bootstrapped power rails, and the error-correction output stage.

Please note that the Blameless configuration has three stages; it is the Lin that has two.

This VAS is a proof that a single-ended stage here does not constitute a limit in the search of extremely low distortion.

True enough, but I think this is already proven by the Blameless amplifier.

Again, anyone know what exactly happened to Halcro?
 
Now that I've seen the updated circuit, I have to say wow, there are quite a few parts. More things to go wrong. It's always nice to have a ridiculously low amount of distortion, and a great theoretical phase margin, but... past a certain point, once the distortion is below 0.1%, reliability, ability to blow up elegantly without taking the speakers with them, and phase margin are what I look for in a power amp. Also how it performs when over driven.

A sound company I once worked for (McCune in SF) favored the old Hafler DH220 poweramps because of how elegantly they blew up. They never damaged the speakers, and very few parts got damaged. Reliability was perhaps their top priority. I've still got 3 of those in my home systems.
 
Now that I've seen the updated circuit, I have to say wow, there are quite a few parts.

I'd be interested to see that circuit.

once the distortion is below 0.1%, reliability, ability to blow up elegantly without taking the speakers with them, and phase margin are what I look for in a power amp.

A sound company I once worked for (McCune in SF) favored the old Hafler DH220 poweramps because of how elegantly they blew up. They never damaged the speakers, and very few parts got damaged. Reliability was perhaps their top priority.

Surely any amplifier with a decent DC-offset-protect system ought to be able to do this?
 
Doug,
Black magic, no, but I'm still working on that voodoo science to extract the unmeasurable quality of scientifically Unachievable sound 🙂. I suppose I'm grappling with the concept that the lowest thd measurement does not neccessarily lead to the best sounding amplifier. Do you listen to a lot of music. Doug?Colin

Huge amounts. One of the main reasons I'm in the business.

I've recently enjoyed live: CSN, Leonard Cohen, & The Eagles. But maybe I'm giving my age away here.
 
My mistake. I assumed the rather complex circuit 2-3 pages back was your latest. Apparently it isn't. Sorry.

Back in the early 70's I built a few SWTPC Universal Tiger poweramps, which I thought sounded great until one of them completely evaporated the entire voicecoil of a 4 inch test speaker when it blew up... I've been a bit paranoid ever since. I was testing a preamp that was probably oscillating at a supersonic frequency. In those days my lab consisted of a volt/ohm meter and a few tools. A rich friend had a scope and generator but I only occasionally had access to that.
 
Forr said:
Fundamentally, in the Halcro, despite their sophistication, the input-stage (differential of Sziklai pairs, twice-cascoded, current mirror) and VAS (single-ended, cascoded) are of the same two-stages topology as the Blameless.

Absolutely true, but there are at least two more layers of complication: the bootstrapped power rails, and the error-correction output stage.
Please note that the Blameless configuration has three stages; it is the Lin that has two.

Of course. I should have written "the same topology for two first stages as the Blameless".

Forr said:
This VAS is a proof that a single-ended stage here does not constitute a limit in the search of extremely low distortion.

DouglasSelf said:
True enough, but I think this is already proven by the Blameless amplifier.

My thinking was that the HEC output stage having low distortion, traces of non-linearities could emerge from the VAS.
If a single ended VAS was chosen for a design as sophisticated as the Halcro, it means that the linearity of such a stage is adequate and that we can't expect much from more complicated schemes here.
Since Edition 6, I am not sure I will ever try any VAS else than a single ended one.
.
 
Huge amounts. One of the main reasons I'm in the business.

I've recently enjoyed live: CSN, Leonard Cohen, & The Eagles. But maybe I'm giving my age away here.

That is awesome Doug, absolutely a requirement for building audio gear, perhaps an even better requirement than proficiency with a probe 🙂.

Forr, never write off anything until you build it, you may just be surprised. Some very legendary amplifiers measure bad but sound ethereal 🙂. There are levels of distortion in instruments that far exceed even a .1% threshold, I may be more of a subjectivist but I also actively play a multitude of instruments Drums, guitar, bass and am subjected to many more that I'm constantly around. At the end of the day I am interested in something that sounds real, acoustic guitars that contain the body, image and harmonics along with spatial cues . Drums that have a proper sounding/feeling leading edge with proper sense of weight on a good recording. That's what I found with CFA, sound so clean you can clearly pick out every nuance in proper relation IMHO to the music and effortless sound at even the loudest volume 🙂.


Colin
 
Otherwise said, of theoretical interest only. That's my point.

Regarding the TMC vs. TPC, give me a Blameless with an optimized TMC compensation and I'll give you a TPC version with the same loop gain and phase margin, within a very few %. Or the other way around. The fact that TMC can, in particular topologies (different from a Blameless) perform better doesn't prove that "TMC is better", I'm sure there are topologies where "TPC is better". From somebody that claims it understands the feedback theory, I would expect to think in terms of maximum feedback, rather than searching for odd corner cases that ultimately do not prove anything.

:yawn: :Pinoc:

This has been discussed before and shown to be untrue. Give it a rest.
 
Fundamentally, in the Halcro, despite their sophistication, the input-stage (differential of Sziklai pairs, twice-cascoded, current mirror) and VAS (single-ended, cascoded) are of the same two-stages topology as the Blameless.
This VAS is a proof that a single-ended stage here does not constitute a limit in the search of extremely low distortion.

The Halcro THD isnt that impressive as figures such as those were already achieved back in 1989. Anone here familiar with Technics SAE series amps. I may be wrong but I think the technics has better figures, Ill try post some figures later.
 
Status
Not open for further replies.