Damn right, Thorsten. I have heard only a few amplifiers with little or no local degeneration which sounded good, but a whole truckful of those which sounded from bad to worse.
Oddly enough, one of the "good guys" was a German design from circa 1978 (or 1979?), from a company called Linear Audio Systems (LAS). I'll see if I can dig up the priginal schematic, I used to stare at it admiring the simplicity and outrageous thuinking behind it. I don't know who actually designed it, but I'd love to buy him a beer; if that man satyed in the business, I would love to know how far he has gone by now.
I nicked their protection scheme, experimented with it for years, and still use it to this day, you saw it in my schematics. One of very few circuits I can honestly say I know everything of importance about it.
Wahab's concept is his own, and it's pointless trying to dissuade him from it, nor is there any reason to do so.
The problem is, I own two integrated amps with low global feedback, where "low" means 17 dB for H/K 6550 (and it's a SEPP design!) and 12 dB for H/K 680. Thus, I have personal hands-on experience with such designs, as well as quite a few others as well - by now, it should be obvious to one and all that I hold Harman/Kardon in very high regard, and low global NFB is one of their cornerstone philosophies. I'm not saying they are the best ever, but I am saying that they are, design wise, a very consistent and serious company. Whether you like or hate their sound, that's a matter of personal preference, and ultimately, not all of their products sound exactly the same.
But I become very cautious each and every time someone has a different view which serves to criticize others, but offer no practical proof of their own ideas. You think you can do better? Fine, let's see what you've done, perhaps we can all learn something from it.
Perhaps I'm too inquisitive, but I am not concenrned with theory only. Theory without practical confirmation is meaningless, it should not exist for its own sake, but as a guide towards new and presumably better solutions.
As an example, I admit to feeling a bit challenged by your amp shown here. Because that's so, I will take the time and trouble to actually make it, with only one minor change, I will add a fourth pair of output devices. I can do that easily, since unlike you, I have no existing basis, I am starting from scratch, and I have a box full of Fischer heat sinks, 300x40x100 mm, about 50+ kilos of them. It can serve as another take on your work, but the key point is, I want to hear it play music. My faith in its musical qualities is strong.
And anyway, The Force is strong in my family.
(Sorry about that, a few days ago I purchased a Blue Ray version of Star Wars, The Complete Saga, and boy oh boy, am I enjoying it, or what! Such fine memories it conjurs up!)
I m not talking of reducing degeneration and maximize the OLG
but to find the good balance between all theses parameters
to yield unconditionnal stability even with 2uF load , lower
possible THD and IMD as well as excellent harmonic distribution,
not counting better PSRR and DF..
As a clue , using your last schematic with almost no modification:
Attachments
Which sim is this??? Is is some form of SPICE? SPICE is an acronym for Simulation Program with Integrated Circuit Emphasis, which makes me think that, if it adds anything at all (and I'm doubtful it adds anything not already in the models, though they might include lead capacitance and inductance), it's unlikely that it would add some "typical" capacitance between all nodes (!) based on what printed circuit boards might have.
You're of course free to add any capacitor between any two nodes, and this might even be useful depending on your layout, frequencies and impedances involved, whether it's on a PCB or IC. I could imagine some back-annotation where a layout (with netlist tied back to the simulation schematic) is scanned with board thickness and material specified, and the resulting capacitances are added to the schematic. But does anything actually do this for either a PCB or an IC layout? I recall hearing of software that calculates impedance of a trace at microwave frequencies, but that's specific to RF design.
But the idea that a simulator would add such things automatically "just because all circuits have 'em" is a little scary. It seems it would too often be a bad guess on the simulator's part. It sure seems easier to add capacitors than to go through menus and help screens to take out invisible ones.
On the subject of stability...
Why dont people doing DIY for personal use take advantage being able to measure the speaker load (over frequency) to be driven?
My designs quickly get over-constrained and being able to slightly relax one constraint for the sake of improving another aspect is always welcome.
I'm not suggesting having the amp break into oscillation if not loaded or when driving "typical" speakers.
Wahab, the transient response plot you present looks good to me especially if the 1uF case is supposed to show stable but not necessarily optimal response. What criterion do you for phase margin when subjected to the extreme load capacitances?
Thanks
-Antonio
Why dont people doing DIY for personal use take advantage being able to measure the speaker load (over frequency) to be driven?
My designs quickly get over-constrained and being able to slightly relax one constraint for the sake of improving another aspect is always welcome.
I'm not suggesting having the amp break into oscillation if not loaded or when driving "typical" speakers.
Wahab, the transient response plot you present looks good to me especially if the 1uF case is supposed to show stable but not necessarily optimal response. What criterion do you for phase margin when subjected to the extreme load capacitances?
Thanks
-Antonio
Ben, to be perfectly honest, I have no idea what the simulator I am using (NI Multisim Pro 10.3) does or does not assume. I assume that it assumes simply because its results are so incredibly close to what I later get from a living model, and if there is an error, it errs on the ideal side, being very conservative. This suits me just fine.
But the proof is in the pudding - it's a simple fact that over the last 6 years, and around 30 models made in it, it never once failed me in its predictions.
Wahab, if I read it correctly, in your graph of say 1 kHz distortion, with a peak output of 40 V, the distortion is around 2 mV, or 86 dB down, or at 0.005%. This is "unrealized potential"?
I would also like to remind you that in his explanation, Thorsten DID say in several places that something needed more work, thus clearly indicating that this was still a rough prototype, and not the final form. Knowing him for many a year, I am aware that this means he is simply occupied with other issues, but being very pedantic (sometimes to a fault), he WILL get back to it and do what needs to be done.
In the end, you or I may not agree with what he's done, but I feel certain it will perform better than it does at this time. I have never known Thorsten to leave anything he starts unfinished. This is why I will make a version of his amp, if he provides the complete schematic once he feels it's ready.
If however you post a schematic of one of your complete works, rest assured I will give it proper attention as well. I believe I am not alone in saying that the proof is in the pudding, the actual living model, not merely in simulations, although we all start there.
I can't speak for others, but my default design and test load is 4 Ohms, not 8. If it can stand its ground with 2 Ohms clean, or 4 Ohms in parallel with 1 uF, I will then consider it ready to be made as a model. I expect its distortion to rise as the load deteriorates, but at no point can they rise over 0.1% peak value.
This may not appear to be impressive, but then the very idea of somebody sitting and listening to full power sine waves is hardly realistic, isn't it? If you want no clipping, you would need to lower the volume by at least 3 dB, and 6 dB would be more realistic.
Also, I use load models of my own loudspeakers, the actual circuits. This is of limited value only, as it overlooks what happens after the speaker heats up and changes its properties, but it's still more realistic than a plain dummy resistor load.
You mention the DIY people. I find that that too many times, it's a matter of personal promotion and preceived glory to put out whatever as soon as possible, without REALLY testing it.
My first model is doomed from the outset, because its final fate will be to be burned out - I need to know its actual, real life absolute limits. So, throw out the overvoltage/overcurrent protection circuits and pump him stupid until he gives up the ghost. But then, there's no more guessing, no more arguing, there are simply FACTS you adjust your protection circuitry for. As a result of this perhaps fatalistic approach, beside the material loss, is that over the years, NOT ONE of my amps has ever burnt down or gone dead on its operator, a statistic I am rather proud of (not that I've made an awful lot of amps, something like 12 or 13).
Thus, in full agreement with you, I would like to enlarge your argument - it's not only testing for pure 8 Ohms as a workload, but also it seems very important to LEAVE OUT any form of common sense protection circuitry, which can be executed in many different ways. Some (insert your favorite swear work here) idiot launched the thesis that protection circuits deteriorate the sound, so it's best to leave them out altogether - same as with tone controls on preamps. As a result, I have seen at least 30 amps blown to bits with their "protection" fuses intact.
Only incompenently designed and adjusted protection circuits deteriorate the sound, same for tone controls, and the bad rep they garner comes from flashy on the outside, junk on the inside Japanese mostly products from the 70ies. They used highly strung up amps, made so for commercial reasons only, which were pushed to their limits with even 6 Ohms, let alone 4 and less Ohms; that was the Lab Era of audio, when products were made or failed in the lab stage. On the opposing side, speaker X was thought of as a very difficult load, only partly because it may really have been difficult, but mostly because amps were not made to drive actual, real world loads.
At least, not in Japan. In Germany, it as a different story altogether, simply because their amps were made to DIN (Germany Industry standards, 45500 for audio) standards, which assumed a 4 Ohm load as default. Thus, in terms of watt delivery capability on paper, they were way behind the Japanese, but in real life, at least 70% of the German serious production of the day sounded better than 70% of the Japanese production in your room. Even your average Grundig (considered Low Fi forever, despite some stunning work they produced here and there) sounded WAY better than your average Pioneer, or some such.
Thus, the DIY arena seems to have adopted as standard the really worst of it out there, although there have been really remarkable projects here and there (eg. Hafler in its time with their DIY kits, Erno Borbelly, Nelson Pass, etc).
And here's where Nelson Pass' philosophy backfires (although through no fault of Nelson). His single-transistor-do-it-all approach has created many bastards in his name, but without his knowledge in the background, and the otherwise sound KISS principle has been vulgarized to the extreme. Yes, keep it simple, but no more than you can without stressing its operating parameters. Too simple is just as bad as too complex. Simple often creates real world drive problems.
The simulator itself doesn't assume anything. It's just a bunch of mathematical equations that are solved.
What you CAN do is add parasitics to models. For instance, you can define a cap as a capacitor of say 1uF, but a more sophisticated model could include a 10meg parallel resistance to model the losses, and a 4nH series L to model the connection wires. And you can really get carried away here, with ESR, temperature effects etc.
The simulator wouldn't care less, it just churns through the equations.
So it's not the sim that assumes anything, its the model designer.
Many opamp models for instance include 2pF cap on the input terminals to model parasitics, but that's just an assumption. If your PCB causes more or less than that 2pF obviously your results will deviate from the real world results.
jan didden
First things i look for is sufficient phase/gain margin to drive highly
capacitive loads , at least 2uF or even more, unconditionnal stability
whatever the load being a plus.
Generaly used numbers are 26db and 90° , although 120° is better,
one never knows.
Of course this will translate in not so impressive perfs on the THD/IMD
as well as in the slew rate fronts , still , good enough numbers are
easy to obtain with very few components.
This will mandate either heavy degenerations or compensation or boths,
depending on the designs preferences.
This is the modded version simulation.
The schematic he published , if ever he manage to stabilize it
enough , will produce 20dB higher THD , not counting an erratic
behaviour on capacitive loads.
The main mistake is that a cascoded VAS cant be compensated
the way he wants to , as the VAS local loop ,comprising the common emitter
and common base VAS transistors, is so fast that the input stage is unable to transmit the signal derivative back to the VAS input , as its slowness will integrate the said signal and render any phase lead compensation unfonctionnal.
(Bold and red by DVV)
In which case I am left to conclude that somebody out there made a number of such assumptions, in addition to allowing me to add and/or modify them at will for each and every component.
I can think of no other way for it to be absolutely right in 30 out of 30 cases, some of which were relatively complex.
That's exactly the case. If you would open a typical opamp model file you can see the values and often they include the model definitions of the transistors and diodes used as well. See example attached for AD815.
There's nothing to stop you to modify any parameter if you know what you are doing.
I'm not sure what you mean by being right 30 out of 30.
It is without doubt that the simulation cannot include all real world effects because you cannot include all of them in the models. But, for simple cases and low freq ranges like audio, it may well be that the sim results look identical with the real world results within the limited measurements used.
jan
Hi,
Middeling. What I like about it is that it uses many of the same keyboard shortcuts etc. as P-Spice. Also, the discrete Parts it has seem more complex models than standard spice (but I ain't no Spice Expert, I just use the darn Stuff).
I find LT-Spice infuriating, everything works different and I lack the time.
I will be spending in Beijing this week (I hate these monster flights) and when I'm off work I will probably not join the customary sport of the local Gentry (the Lucky Journey Cat hunting the White Tiger) every evening, given the quality of TV Programming I should finally have the time to overcome my LT-Spice aversion.
Many here use it and there are lots of posted circuits, including the Transistor Models and it accepts standard Spice.
Personally I just throw in Generators into the supply to check PSRR and intermodulation.
In principle you can add generators with LRC additions to simulate leakage inductance, primary inductance, parasitic capacitance and winding resistance of the transformer and the transformed mains impedance and then just add your diodes and the capacitors (include the main parasitics) and other gubbins.
Ciao T
Middeling. What I like about it is that it uses many of the same keyboard shortcuts etc. as P-Spice. Also, the discrete Parts it has seem more complex models than standard spice (but I ain't no Spice Expert, I just use the darn Stuff).
I find LT-Spice infuriating, everything works different and I lack the time.
I will be spending in Beijing this week (I hate these monster flights) and when I'm off work I will probably not join the customary sport of the local Gentry (the Lucky Journey Cat hunting the White Tiger) every evening, given the quality of TV Programming I should finally have the time to overcome my LT-Spice aversion.
Many here use it and there are lots of posted circuits, including the Transistor Models and it accepts standard Spice.
Personally I just throw in Generators into the supply to check PSRR and intermodulation.
In principle you can add generators with LRC additions to simulate leakage inductance, primary inductance, parasitic capacitance and winding resistance of the transformer and the transformed mains impedance and then just add your diodes and the capacitors (include the main parasitics) and other gubbins.
Ciao T
Hi,
The basic amplifier will have a "build out network", actually, to be precise a tapped aircore inductor that will act as bidirectional 300KHz lowpass (a Ham Radio style one, not audio, mainly to protect the Amplifier from RFI ingress via the output), however I have not drawn that in, as I call it "housekeeping", together with overload protection, speaker protection and so on.
Seeing the 100nF curve I think the Amplifier is overcompensated as is, and I can loosen up the compensation, once the build out impedance / output filter are added.
I a not interested in improving Total Hopless Disintegration, Interfaith Malarkey Disintegration or Duping Factor beyond what is shown. As I know how to design power supplies with low noise I do not necessarily view high Powah Supper Replication Reason as a desirable quality, however the Amplifier circuit as shown should do pretty horrorshow anyway.
Any of these parameters are already sufficiently past "good and bad" that any further improvements are meaningless.
The only thing that would interest me, is how to reduce upper harmonics more and/or how to keep the performance the design is capable of while removing looped feedback completely.
Almost no modification means "with modification". It would be polite to state what the modifications are.
Also, you may want to check your analysis parameters, those overly broadened spectral lines are an indication something is not horrorshow and you may have to align windowing, sample number and time-steps to get a clear view.
Past that -90dB H2 for 33V and 8R @ 1KHz (0.003%) look already way too good for my taste. The 0.03% for 33V and 8R @ 10KHz are still way good. It indicates an increase in negative loop feedback that is unwanted.
What I would lime to see is feedback lowered and adjusted so we get 0.1% H2 for 33V/8R with both 1KHz and 10KHz.
Of course, those are MY design goals.
Ciao T
The basic amplifier will have a "build out network", actually, to be precise a tapped aircore inductor that will act as bidirectional 300KHz lowpass (a Ham Radio style one, not audio, mainly to protect the Amplifier from RFI ingress via the output), however I have not drawn that in, as I call it "housekeeping", together with overload protection, speaker protection and so on.
Seeing the 100nF curve I think the Amplifier is overcompensated as is, and I can loosen up the compensation, once the build out impedance / output filter are added.
I a not interested in improving Total Hopless Disintegration, Interfaith Malarkey Disintegration or Duping Factor beyond what is shown. As I know how to design power supplies with low noise I do not necessarily view high Powah Supper Replication Reason as a desirable quality, however the Amplifier circuit as shown should do pretty horrorshow anyway.
Any of these parameters are already sufficiently past "good and bad" that any further improvements are meaningless.
The only thing that would interest me, is how to reduce upper harmonics more and/or how to keep the performance the design is capable of while removing looped feedback completely.
Almost no modification means "with modification". It would be polite to state what the modifications are.
Also, you may want to check your analysis parameters, those overly broadened spectral lines are an indication something is not horrorshow and you may have to align windowing, sample number and time-steps to get a clear view.
Past that -90dB H2 for 33V and 8R @ 1KHz (0.003%) look already way too good for my taste. The 0.03% for 33V and 8R @ 10KHz are still way good. It indicates an increase in negative loop feedback that is unwanted.
What I would lime to see is feedback lowered and adjusted so we get 0.1% H2 for 33V/8R with both 1KHz and 10KHz.
Of course, those are MY design goals.
Ciao T
Indeed, and there are many Spice transformer models that include secondary effects like coupling factor, leakage L, parasitic cap etc.
Of course you must set these values for the xformer you are planning to use and those values might be difficult to obtain.
jan
Hi,
I did mention before that I expect having to apply some Miller compensation. However I want to see what is possible with minmal Miller compensation.
Well, good thing we have some capacitive load on the VAS to slow it right down, so the IPS can keep up then, innit? Also, I doubt that the VAS is actually "faster" than the IPS.
As my Tina-Ti circuit has all "wrong" parts and parasitics play a great role in all of that I simply currently leave it at "stable 10KHz squarewave into 8Ohm".
Ciao T
I did mention before that I expect having to apply some Miller compensation. However I want to see what is possible with minmal Miller compensation.
Well, good thing we have some capacitive load on the VAS to slow it right down, so the IPS can keep up then, innit? Also, I doubt that the VAS is actually "faster" than the IPS.
As my Tina-Ti circuit has all "wrong" parts and parasitics play a great role in all of that I simply currently leave it at "stable 10KHz squarewave into 8Ohm".
Ciao T
Hi,
If you have a reliable AC voltmeter past 10KHz, a 'scope and a generator maybe up to 1MHz you can measure of these easily (see RDH, which has information on this).
The inductances can be measured using a suitable signal via a known value resistor quite accurately, as can be coupling factors and DC resistances. Capacitances can be reverse estimated by measuring the main resonance (scope and generator) as we already know inductance and frequency.
Extreme precision in these values is likely not beneficial, all we need is a decent ballpark.
Ciao T
If you have a reliable AC voltmeter past 10KHz, a 'scope and a generator maybe up to 1MHz you can measure of these easily (see RDH, which has information on this).
The inductances can be measured using a suitable signal via a known value resistor quite accurately, as can be coupling factors and DC resistances. Capacitances can be reverse estimated by measuring the main resonance (scope and generator) as we already know inductance and frequency.
Extreme precision in these values is likely not beneficial, all we need is a decent ballpark.
Ciao T
Thanks for your insight.
Wahab,
The cascode does look into the MOSFET driver stage.
Since you have a model it might be interesting to see the transient response with the lead capacitors removed.
Thanks
-Antonio
pure uF capacitance without significant ESR is a silly test load - nearly nonexistant in the wild - ESL are driven thru step up xmfr with series leakage inductance, winding R that decouples the panel C from the amp way below frequencies that stability problems occur
typically the 1-100nF range is where actual stability problems with Cload occur - loudspeaker cable C usually lies in the lower decade of that range
typically the 1-100nF range is where actual stability problems with Cload occur - loudspeaker cable C usually lies in the lower decade of that range
- Status
- Not open for further replies.