Design Critique Request

I came up with this design for an amplifier for self-constructed speaker boxes for the computer. This is a preliminary design that I have built. It seems to sound pretty good, as compared to this module which I have that's intended for use in car audio apps. That didn't sound so good (although for a long wave receiver I designed previously, it does OK) and this looked like a good excuse to try something a bit different. ;)

The first iteration of this project basically uses "transistors anonymous" that I get by the dozens from Radio Shack. At least that way, I have a good choice for selecting matched pairs for the differential (long tailed pairs) stages. I have some ideas for improvements (better initial pre-amp, another current source in the second LTP) for a newer version.

Anyway, I'd like to get a second opinion.

You can see the circuit schematic Here. (I didn't attach since it's pretty good size and I don't want to monbopolize too much screen.)

Good :D ?
So-so :apathic: ?
YYYYEEEECCCCHHHH!!!! :bawling: ?
 
Have you built it? Does it work?

Quasi-complimentary isn't bad, just not as common anymore. For such low power, transformer coupled, you should consider class A, using a current source. Bias might be more stable. Of course electrolitic caps in the signal path is usually not popular for high quality audio.
 
Have you built it? Does it work?

I have built a prototype (actually two) that I'm using for speaker boxes for the computer. This was mainly a "proof of concept" for an idea I had. They do work, and I was rather surprised at how well they seemed to work, despite the use of off the shelf components I got at Radio Shack. That would include the use of some "car replacement", "wide band" speakers, not the best.

Of course, I don't claim to be a master audiophile either.

I'm wondering if I can use the same circuit for a higher quality amp. Being more of an RF guy, I'm not real familiar with all the subtilties of audio design, and so I'm wondering if there are any defects in that circuit that I'm not seeing. (Other than the electrolytics in the signal path. That was a kludge as I was all out of the 1.0uF ceramic monolithics at the time. Since then I replaced them when the new ceramic monolithics finally arrived.)

Also need a better preamp since that's where all the gain comes from. Everything to the right of the gain control is unity gain.
 
Seems like a lot of trouble and a lot of silicon just to have unity voltage gain. The more transistors in the signal path the slower the action, i. e. slower slew rate and possible occilation. The least amount of transistors possible in the signal path is usually the best approach. Just because you are reproducing AF frequencies (sloooooow by your RF experience), the slew rate is an important factor. If only there was a single transistor that could do it all.:rolleyes: :D Since all the voltage gain is to come from a pre-amp stage, why not just use an emitter follower circuit to amplify current? On the other hand, of course if you want to increase the gain of the LTP (not considering feedback) perhaps a current mirror might help. Any idea of the open loop voltage gain?
 
I agree; there is enough loop gain here that you can do without the preamp and just modify the feedback for a gain of 20 or so. Which makes me wonder if the schematic truly reflects your prototype, it is amazing that you can get away with unity gain without any compensation?

Anyway, the use of a quasi-complementary output stage with the linearizing diode harks back at least 20 or 25 years, I think it came from Peter Walker, because of the unavailability of complementary power devices at that time.

That doesn't mean it is a bad good solution, it will surely sound OK, but in contemporary designs one would tend to go fully complementary.

Jan Didden
 
QSL man, message received?

This is a Ham (Radio Amateur) card.

Very old those things...now we have computers.

hehe

Carlos
 

Attachments

  • qsl card 2.jpg
    qsl card 2.jpg
    64 KB · Views: 298
OK, lots of good feedback ;) Thanks for taking a look-see at it. :)

Here is the design philosophy behind the circuit. It is obviously a derivative of that topology that you find in just about any text: differential input + high gain interstage + common collector output. Having the diff amp, I wondered what would happen if it drove another diff amp instead of just disconnecting one output. This seemed to open up the possibility of a second feedback path that would allow stability without having to add a 100pF or so capacitor across the collector-base junction of the interstage transistor. That never made much sense since the semiconductor manufacturers have gone to such lengths, devoted much research, towards the objective of giving us BJTs with nice, low junction capacitances that can operate reliably at frequencies up to UHF and beyond. After all, that 100pF capacitor can reflect an equivalent input capacitance measureable in microfarads (many). That just can't be a good thing so far as time delay is concerned. So I was seeing if that could be eliminated, and it seems to be the case. I get no hint of instability. This should not be a hard thing to do, considering that unconditionally stable RF amps aren't all that difficult to build. (Been there; done that.)

I agree; there is enough loop gain here that you can do without the preamp and just modify the feedback for a gain of 20 or so. Which makes me wonder if the schematic truly reflects your prototype, it is amazing that you can get away with unity gain without any compensation?

The schematic is accurate. Getting rid of that compensating capacitor was the main objective. I would expect that raising the closed loop gain would cause instability if not outright oscillation.

Anyway, the use of a quasi-complementary output stage with the linearizing diode harks back at least 20 or 25 years, I think it came from Peter Walker, because of the unavailability of complementary power devices at that time.

I'm not too worried about that. After all, the basic circuit itself harks back to the late 1940s :D (Nothing wrong in that. "New" just for the sake of being "new" is not necessarily a Good Thing. Check out Susan Parker's design: that harks back to 1910 or so. :D ) As for the quasi-complementary design itself, you'd naturally think that a full complementary would be better. However, I read that, when the matter was finally put to the test, the quasi-complementary actually performed better. (Probably from Doug Self). The "Sziklai Pair", after all, is a very similar topology.

Any idea of the open loop voltage gain?

98db(V) is what I estimate. Actually trying to measure it becomes problematic in that the signal generator I have (very basic unit) won't reliably go low enough to get output that's not clipped. So the open loop gain is definitely up there.

On the other hand, of course if you want to increase the gain of the LTP (not considering feedback) perhaps a current mirror might help.

I do believe it would.
 
Here is the design philosophy behind the circuit. It is obviously a derivative of that topology that you find in just about any text: differential input + high gain interstage + common collector output. Having the diff amp, I wondered what would happen if it drove another diff amp instead of just disconnecting one output. This seemed to open up the possibility of a second feedback path that would allow stability without having to add a 100pF or so capacitor across the collector-base junction of the interstage transistor...

To be perfectly honest I don't see how that would get rid of the Ccomp. A LTP front end driving a LTP current mirror terminated VAS/driver is definitely nothing new and in fact it is most often seen with two Ccomp (one for each side of the LTP), though a single one connected differentially can be used as well with some care. BTW how often do unconditionally stable RF amps have AC feedback applied across 3 stages? ;)

The real question is why use ner 100dB open loop gain to make a unity buffer and then use a single ended preamp that cannot by definition drive the output stage to maximum excursion?
Stability could have been achieved by limiting the VAS open loop gain by giving it a defined impedance to work into (instead of the variable impedance of the output stage). Ditto reducing input stage gain by degeneration in the emitters. If you wanted FET sound, using two FETs in the input LTP would work great, AND would reduce open loop gain too due to their lower gm.

Anyway, the use of a quasi-complementary output stage with the linearizing diode harks back at least 20 or 25 years, I think it came from Peter Walker, because of the unavailability of complementary power devices at that time.

More like 30-35 years ;) but that's beside the point. It is a good trick though ;)

I'm not too worried about that. After all, the basic circuit itself harks back to the late 1940s :D (Nothing wrong in that. "New" just for the sake of being "new" is not necessarily a Good Thing... However, I read that, when the matter was finally put to the test, the quasi-complementary actually performed better. (Probably from Doug Self). The "Sziklai Pair", after all, is a very similar topology.

Actualy, D. Self himself (no pun intended) gave a decided :( to the 'better audio from non-complements' artice (B. Ollsen, IIRC). I've read it myself and I would call it circumstantial (but of course, I'm no authority and heaven forbid I ever become one ;) ).
Sziklai Pair is HALF of the standard quasicomplementary topology, the other being a follower (though depending on drive arrangement the follower may reduce to a current source which dispenses with the problem of the upper half of the putput having low output impedance and the lower half having high output impedance).
The issue with traditional quasi-complementary designs is that the transfer characteristic is very different for a follower and a CFB pair (a.k.a. Sziklai Pair). Since each passes one half of the output waveform, even order distortion is introduced. Wether that is good or bad seems to remain in the subjective domain ;)

On the other hand, of course if you want to increase the gain of the LTP (not considering feedback) perhaps a current mirror might help.

Well, it's already there (though with a diode, which I call 'the Scrooge McDuck version' as it's less accurate and transistors (even diode connected ;) ) are really cheap these days, so no real $ saver...
 
To be perfectly honest I don't see how that would get rid of the Ccomp.

Well, it did. There are no Ccomp's in either one. I'm playing them right now (I'm Furry/Kwooky Womble, to be exact. Lots of quick, pulse-like sounds that ought to excite instability.) and I'm definitely not listening to a bunch of oscillation howls and/or damped oscillations. :)

A LTP front end driving a LTP current mirror terminated VAS/driver is definitely nothing new and in fact it is most often seen with two Ccomp (one for each side of the LTP), though a single one connected differentially can be used as well with some care.

Given the number of professionals working in the field, I would be as surprised to learn that no one else had thought of this as I would be to learn I just won the lottery. Although I haven't seen a design that got rid of that Ccomp. (Which brings up another question: is that thing included as a matter of tradition, regardless of whether it's needed or not? :confused: ) That doesn't mean that they don't exist, only that I haven't seen one yet.

The real question is why use ner 100dB open loop gain to make a unity buffer and then use a single ended preamp that cannot by definition drive the output stage to maximum excursion?

The single-ended JFET voltage amp was strictly a "quick and dirty" solution to the problem of getting the sound card's output up to a listenable level. The whole circuit is preliminary, and there are obviously areas that could use improvement. That's why I'm here; why I started the topic in the first place. ;) After all, op-amp voltage followers use even more open loop gain to achieve unity gain, don't they?

Actualy, D. Self himself (no pun intended) gave a decided to the 'better audio from non-complements' artice (B. Ollsen, IIRC). I've read it myself and I would call it circumstantial (but of course, I'm no authority and heaven forbid I ever become one ).

Could have been Ollsen. (It's been four months since I completed the prelim version I'm using now.) He seemed to make a good case for quasi-complementary, so that's why I went with it. Of course, this brings up yet another problem: separating advice that's valid from pure non-sense (of which there seems to be a considerable amount.) I don't recall Doug Self saying anything against the quasi-comp (although I haven't found anything he says in favor of it either.) I know it's counter-intuitive, but there are lots of other things in engineering that are counter-intuitive. :xeye:
 
QSL man, message received?

Message recieved. What I'm mainly into is Part 15 Longwave. I've been doing this for about ten years now. (Been interested in longwave since third grade.) I have some of my designs featured there. Getting really good longwave equipment isn't easy, so I design and build my own.

Before that, it was doing DX with these "kiddie" CB transcievers that advertise a range of 1/4 mile. I had a "Space Commander" base station. This unit featured a decent receiver, and a max legal power output on transmit (AM and CW). The final was connected to a pi-L network with fixed capacitors. I replaced these with variables and ran it into a ground plane. (It looked like the whole thing was designed for ease of modification.) That worked quite well indeed: QSOs of a good 1000 miles. We also had a guy in our neighborhood running a kilowatt linear on CB that was alleged to be a "Miles Prower" design, but you didn't hear that from me.

I've also done some high end Part 15 work as well.

73 destroyer X
 
Miles Prower said:
To be perfectly honest I don't see how that would get rid of the Ccomp.

Well, it did. There are no Ccomp's in either one. I'm playing them right now... and I'm definitely not listening to a bunch of oscillation howls and/or damped oscillations. :)


I have no issue or dispute with that. my point was that if your amp is stable wit CLG of 1, i cannot see how it wuld be less stable with CLG >1, which is simple to do adding one single resistor (or R and C in series if you want to keep DC gain = 1 to reduce offset), and which does not need a preamp any more.

Given the number of professionals working in the field, I would be as surprised to learn that no one else had thought of this as I would be to learn I just won the lottery. Although I haven't seen a design that got rid of that Ccomp. (Which brings up another question: is that thing included as a matter of tradition, regardless of whether it's needed or not? :confused: ) That doesn't mean that they don't exist, only that I haven't seen one yet.

As you cannot avoid capacitances, it just means that what there is, is enough to compensate the amp.
The question is repeatability. Commercial designs tend to play it safe and overcompensate rather than have an amp come back under waranty due to oscillation.
Can't see how using a LTP into LTP design solves stability problems on it's own, maybe you got lucky ;) and if you did, enjoy ;)

The single-ended JFET voltage amp was strictly a "quick and dirty" solution to the problem of getting the sound card's output up to a listenable level. The whole circuit is preliminary, and there are obviously areas that could use improvement. That's why I'm here; why I started the topic in the first place. ;) After all, op-amp voltage followers use even more open loop gain to achieve unity gain, don't they?

Yes, but I doubt you would build a follower and a gain block in front of it out of an OP-amp capable of being both at the same time. And, you hardly build OP-amp followers if that isn't really what you need ;) unless it works FAR better as a follower than an amp with gain. The point being, I think your 'follower' would work just as well if not better as an amp with gain.

Could have been Ollsen. (It's been four months since I completed the prelim version I'm using now.) He seemed to make a good case for quasi-complementary, so that's why I went with it.

At first glance it is attractive as it introduces even order distortion, which would be euphonic if only it did not persist up into higher harmonics. The drop of OLG at higher frequencies reduces the ability of the NFB to reduce higher order even harmonics, which symetry reduces automatically.
That being said, there are driving topologies that alow both upper and lower devices of the same kind with practically equal transfer characteristics, avoiding the above issues, but the standard quasi-complementary is not it.

Of course, this brings up yet another problem: separating advice that's valid from pure non-sense (of which there seems to be a considerable amount.) I don't recall Doug Self saying anything against the quasi-comp (although I haven't found anything he says in favor of it either.) I know it's counter-intuitive, but there are lots of other things in engineering that are counter-intuitive. :xeye:

True...
i don't think anyone has a definitive answer to your question. A good first approximation would be never to lose common sense and to read a lot, xperiment some, and decide for yourself, while remembering that nothing is set in stone.
 
Miles Prower said:
[snip] The schematic is accurate. Getting rid of that compensating capacitor was the main objective. I would expect that raising the closed loop gain would cause instability if not outright oscillation.
[snip]


Miles,

Not normally. The higher you make your closed loop gain the better the stability. That was the basis of my earlier remark - going to unity gain closed loop gain invites instability and oscillations. Why in your case it would be just the other way around, I have no idea. That's why I asked if the schematic really is accurate with the prototype, which you confirmed.

Jan Didden
 
Miles Prower said:
[snip]Could have been Ollsen. (It's been four months since I completed the prelim version I'm using now.) He seemed to make a good case for quasi-complementary, so that's why I went with it. Of course, this brings up yet another problem: separating advice that's valid from pure non-sense (of which there seems to be a considerable amount.) I don't recall Doug Self saying anything against the quasi-comp (although I haven't found anything he says in favor of it either.) I know it's counter-intuitive, but there are lots of other things in engineering that are counter-intuitive. :xeye:


Doug Self wrote a critique on Olsen's proposal called "Few compliments for non-complements" in EW (edit: Sept 1995*). His conclusion was, as noted above, that especially in the critical cross-over region, at low signal levels, the harmonic components were much higher and there were much more than compared to a real complementary pair, even with imperfect matching.
That doesn't mean that your amp isn't listenable, but it seems to me that a complementary circuit would be simpler, more elegant, better performer and maybe even cheaper. But, hey, it's your circuit!

Jan Didden

*quote:
"I have to conclude that the configuration appears to require a very high quiescent current for linear operation, and has onlya limited amount of negative feedback available to correct output stage distortions. Any deeper investigation would need to be encouraged by some promise that the Olsson configuration can deliver substantial benefits, and as far as my analysis goes, this does not seem likely".
 
That doesn't mean that your amp isn't listenable, but it seems to me that a complementary circuit would be simpler, more elegant, better performer and maybe even cheaper. But, hey, it's your circuit!

I didn't marry the thing! :bigeyes: There's no arguing with facts, so the next iteration will use full complementary. :) Although the current version is quite listenable (moreso than these amp modules I got that were originally intended for car stereos: those things are truly nasty) and I'll probably keep it around for those who want/need to go with the quasi-complementary topology. (After all, quasi-complementary was a kludge to get around the lack of availability of good PNP power BJTs.)