Hi Bob!
I was referring to Chapter 11.3 Miller Input Compensation, second paragraph (2nd edition of your book). You are discussing Figure 11.4 which is the way Otala did his compensation (along with several lead-lag networks). You said "This kind of input compensation is not recommended for audio amplifiers."
I am aware of your excellent 1984 paper, having first read it that same year! This is a more refined approach and one I wanted to use in my last power amplifier (described in AX last year) but I was either too senile or too impatient to get simulation results showing any significant improvement over conventional Miller compensation. Or, perhaps this was due to the lowish open loop gain that I employed.
Bruce
Hi Bruce,
Thanks for pointing to that place in my book. It was probably misleading for me to put the Otala input compensation mention under a section called Miller Input compensation, since the Otala input compensation shown in Figure 11.4 is not Miller Input Compensation (MIC), but instead is just simple input compensation where a series R-C is placed across the differential inputs of the input stage. As explained, the simple input compensation is sub-optimal, partly because it can have detrimental effects on input-referred noise and input impedances, and does not provide the benefits of pole splitting.
Cheers,
Bob
Thanks for pointing to that place in my book. It was probably misleading for me to put the Otala input compensation mention under a section called Miller Input compensation, since the Otala input compensation shown in Figure 11.4 is not Miller Input Compensation (MIC), but instead is just simple input compensation where a series R-C is placed across the differential inputs of the input stage. As explained, the simple input compensation is sub-optimal, partly because it can have detrimental effects on input-referred noise and input impedances, and does not provide the benefits of pole splitting.
Cheers,
Bob
I tried it, and as great as it may be, it's not backward compatible with LTSpice, which I think is a big mistake....
I share your concern if by backward compatibility you mean that one should be able to take an existing asc file and directly run it on QSPICE. There are many of us out there that have hundreds if not thousands of simulation files that were done on LTspice and which we would like to use as-is or modified in QSPICE without needing to do re-entry of the schematic. Please confirm if my assumption about backward compatibility on your part is correct.
I am not a software person and have no idea if such compatibility could be achieved in practice or if a piece of translation software could be made to do it. In fact, I honestly don't know if an LTspice design can be directly ported to another SPICE simulator of the many that are out there. For example, can an LTspice simulation be ported to TINA?
Cheers,
Bob
I am not a software person and have no idea if such compatibility could be achieved in practice or if a piece of translation software could be made to do it. In fact, I honestly don't know if an LTspice design can be directly ported to another SPICE simulator of the many that are out there. For example, can an LTspice simulation be ported to TINA?
Cheers,
Bob
Yes, that's exactly what I'm missing in QSPICE - ability to import LTSpice files and libraries with models.
It is definitely possible to convert files from LTSpice to QSPICE - check out this Python library and scripts:
https://github.com/nunobrum/spicelib
developed by Nuno Brum. I was testing his prototype and I managed to import
some LTSpice simulations to QSPICE, however this solution is not 100% ready, and has problems with some symbols (E.g. op-amps).
But - it's a proof of concept that it can be done.
If this kind of function was built-in in QSPICE, it would definitely help it to gain some traction...
It is definitely possible to convert files from LTSpice to QSPICE - check out this Python library and scripts:
https://github.com/nunobrum/spicelib
developed by Nuno Brum. I was testing his prototype and I managed to import
some LTSpice simulations to QSPICE, however this solution is not 100% ready, and has problems with some symbols (E.g. op-amps).
But - it's a proof of concept that it can be done.
If this kind of function was built-in in QSPICE, it would definitely help it to gain some traction...
Having built, owned and repaired many different "Tiger" and SWTP power amps of various power levels in the 70s, frequency stability was a big issue with the "Plastic Tiger" of 1971 but thermal stability was HUGE problem with ALL of them. Lost count how many NPN/PNP output pairs I had to replace. None of the amps could survive hours of full power at a Frat party. ;-)
I had the 198 pre-amp, Plastic Tiger amp, 215 amp, Tiger .01 207 amp and Tiger Mark II amp.
The only amp that survived non-stop abuse was a clone I made of the 1st gen Quad 405 I made in 1977. Pre-Amp was my version of the Douglas Self Advanced Pre-Amp.
Still works today.
I had the 198 pre-amp, Plastic Tiger amp, 215 amp, Tiger .01 207 amp and Tiger Mark II amp.
The only amp that survived non-stop abuse was a clone I made of the 1st gen Quad 405 I made in 1977. Pre-Amp was my version of the Douglas Self Advanced Pre-Amp.
Still works today.
Attachments
Last edited:
Thanks for confirming my assumption about backward compatibility - we are on the same page.
Having a 3rd-party utility like you describe is a wonderful thing, and I am sure many people would love to have it.
Like most people, I'm guessing, these days virtually every single LTspice simulation I run has been copied and modified from a previous LTspice simulation.
I suspect that there might be some kind of legal or copyright problems if such a capability was built into QSPICE.
Cheers,
Bob
Having a 3rd-party utility like you describe is a wonderful thing, and I am sure many people would love to have it.
Like most people, I'm guessing, these days virtually every single LTspice simulation I run has been copied and modified from a previous LTspice simulation.
I suspect that there might be some kind of legal or copyright problems if such a capability was built into QSPICE.
Cheers,
Bob
I got a message from Mike Kiwanuka and he says that two of his amp designs are in the QSPICE demo folder.
I don't believe that there'd ever be a copyright problem by providing and import capability, or say compatibility with models.
The answer to that would be up to Mike. In any case, such capabilities would be nice.
Cheers,
Bob
Cheers,
Bob
With +-100 V rails I am not surprised these things went up in smoke. What was the designer thinking? If the goal was 700 W the solution was a bridge amp running at half that supply voltage.
A series output stage is similar, Ampzilla, Tigersaurus, big Leach, some SAE amps, and even fairly recently Bryston.
Why did Bryston do it with high voltage devices? Probably to make better use of the SOA curve.
Why did Bryston do it with high voltage devices? Probably to make better use of the SOA curve.
Series output stages do exactly the same thing for the SOA curve as bridging - providing the output stage behaves itself. With the fully class B NPN stage that PL used and is known to have local oscillations under some bias conditions that could be a tall order. Series stages with proven track records all used complementary pairs and ran at unity gain. In my experiments with super-Leach type output stages, a minimum bias is needed in the outer banks to avoid sinivets and output as well as front end decoupling is required. One might be able to make an all NPN stacked stage work with epitaxial 2N3773s or good Jap parts like the 2SD424, but it would need to run with output stage bias and require proper local decoupling. Might be fun (not) with that point to point output stage. It would give less grief with MJ15024/5 stacked 3x2. The full comp simply isn’t as finicky.
Interesting test with lateral power MOSFETS:
https://www.diyaudio.com/community/threads/lateral-mosfet-torture.111116/
Does anyone know if the Exicons are as rugged?
https://www.diyaudio.com/community/threads/lateral-mosfet-torture.111116/
Does anyone know if the Exicons are as rugged?
Probably. The “pop” that was heard after a few seconds was probably when the glass temperature of the overmold epoxy got exceeded. If that happens too rapidly pieces can go flying. Even with a slow cook the device can suffer mechanical damage long before significant electrical degradation in the die itself. Too bad the TO-3 versions aren’t expendable anymore, so no one is likely to test them to destruction. If one has any, they keep them to make amplifiers (or profit from someone who does). It would however, eliminate failure modes associated with the package.
Back in the 1980's I consulted in the design of the Crest 8001. It was a 3-rail class H design, full complementary. It was a monster, rated at 720 watts per channel into 8 ohms. It was widely used in pro audio where high power was needed, and was quite reliable. You are correct in that an all-NPN series design would be difficult to execute, and certainly not when good complementary devices are available. It is also difficult to obtain low crossover distortion with a quasi output because the Oliver criteria can usually not be met by biasing at an optimum value, due to the asymmetry of the top and bottom stages. The quasi stage can also be very difficult to stabilize.
The first solid state power amplifier I built was a clone of the Harman-Kardon Citation 12, circa 1971, and it used a Quasi output stage. It was a nice, well-regarded amplifier for the time. I think it did about 75 wpc. Later, when decent PNP power transistors were available, I redesigned it to be fully complementary.
Cheers,
Bob
Have you ever found a way to stabilize a QC triple when doing a switched rail class H? It seems to break into uncontrollable oscillation when the Vcc jumps, no matter what snubber is used or bypassing is used on the supplies. About the only thing I haven’t tried is putting a 100 uF cap directly on the switched bus itself. Which would be counterproductive, producing large charging currents and power draw that goes up dramatically with frequency. Not to mention that the cap itself would probably die in a week’s use. A full complementary stage gives no trouble whatsoever, if normal best practices are observed. If it just NEEDS that 100 uF cap right there (which seems to help ordinary amps using a PL output stage) then the idea is dead. Having the supplies themselves bypassed locally did nothing to help.
I have a huge stash of 2N3773’s and Je ne sais quoi NPNs. It would open up a lot of potential uses if I could ever crack that particular nut. Not every PA amplifier I put together needs especially low crossover distortion.
I have a huge stash of 2N3773’s and Je ne sais quoi NPNs. It would open up a lot of potential uses if I could ever crack that particular nut. Not every PA amplifier I put together needs especially low crossover distortion.
Have you noticed that the "quasi" side of the output stage is essentially CFP which are known
for stability problems? I would guess that the problem is in the local loop, in fact it has to be
since when you go to full comp there's no problem, where the collector of the driver feeds the
base of the output, and the collector of the output connects to the emitter of the driver. Parasitic
inductance and cap is probably playing a part. Something in that path or a Zobel or whatever is
needed there to stabilize that loop. Loop gain analysis should help.
Look at the Bryston and make it unity gain perhaps. Consider that the Bryston uses both PNP
and NPN outputs on both rails, just remove the PNPs and you have a quasi output stage. I've
not looked at the schematic but you could use power diodes to replace the PNP BE junctions or
just make the resistors on the driver for that side smaller.
for stability problems? I would guess that the problem is in the local loop, in fact it has to be
since when you go to full comp there's no problem, where the collector of the driver feeds the
base of the output, and the collector of the output connects to the emitter of the driver. Parasitic
inductance and cap is probably playing a part. Something in that path or a Zobel or whatever is
needed there to stabilize that loop. Loop gain analysis should help.
Look at the Bryston and make it unity gain perhaps. Consider that the Bryston uses both PNP
and NPN outputs on both rails, just remove the PNPs and you have a quasi output stage. I've
not looked at the schematic but you could use power diodes to replace the PNP BE junctions or
just make the resistors on the driver for that side smaller.
No, but I have never tried.
There has always been confusion about Class G and Class H in different countries. The Crest 8008 is referred to as Class H, where switched rails are not used. In this definition of class H, the rails for the intermediate output stage are created by an EF arrangement that is powered from an intermediate rail through a diode. The intermediate stage operates from that intermediate rail until its headroom falls below a certain point, at which time it becomes powered by the output stage of the next rail up, This is best described in my book "Designing Audio Power Amplifiers.
Cheers,
Bob
The old Crests are what I’d call “G”. It requires twice as many power transistors, as it hands off high dissipation duty from the lower to the upper bank at the voltage transitions. Both banks need to be capable of high dissipation. It can get complicated and expensive going to more than 2 rails. In the hard-switched variety, a hexfet turns ON as the VCE hits the transition point. I’ve aways called that class H, as in hard switched. G would mean “gradual”, as the upper bank’s output voltage gradually increases to keep the lower bank’s Vce low above the transition point. I’ve tried that with QC triples too - and guess what - the negative side still breaks into oscillation when it hands off from the commutation diode to the higher (negative) rail follower. Making the transition more “gradual” didn’t buy a thing.
It would tend to have a cleaner output spectrum in theory, assuming it’s well-behaved. There are crossover-distortion-like artifacts produced at hand-off, and the gradual hand-off reduces their amplitudes and the order of the distortion products produced. It is normally masked well by the music, and if turned up loud enough that you’re running off the high rail often enough for that distortion to add up to something, clipping is usually pretty severe already. At that point, the clipping behavior is more important than what is happening at rail hand-off.
It would tend to have a cleaner output spectrum in theory, assuming it’s well-behaved. There are crossover-distortion-like artifacts produced at hand-off, and the gradual hand-off reduces their amplitudes and the order of the distortion products produced. It is normally masked well by the music, and if turned up loud enough that you’re running off the high rail often enough for that distortion to add up to something, clipping is usually pretty severe already. At that point, the clipping behavior is more important than what is happening at rail hand-off.
Yes, the Quasi is basically a CFP configured as an emitter follower. They definitely need to be well-designed and, like any feedback circuit, need to be simulated for stability. A super-simple CFP with the NPN and PNP directly interconnected can have quite high loop gain, and that can lead to instability. BTW, the CFP is essentially a current feedback amplifier, easily seen if it is configured as a non-inverting gain stage by providing a feedback resistor between the collecto of the second transistor and the emitter of the first transistor, and setting the closed-loop gain with a shunt resistor to ground from the emitter of the first transistor. For improved stability, and emitter resistor can be placed in series with the emitter of the second transistor, thus reducing the loop gain of the CFP. A lag-lead series R-C in the collector circuit of the first transistor can also aid stability.
Cheers,
Bob