Yeah, but you were never an Internet/i-Pod/MP3 kid, either.
Heck Brad, you used to READ for your info!
Sorry. I know I quoted your post but I wasn't trying to single you out, or anything.
And from what you have said in the quoted section above, you are doing an exquisite job, indeed.
I mainly just wanted to make another attempt at drawing attention to the transient response.
Not a problem, Gootee.
Thank you for the "exquisite", but I beg to differ, as I see it, all I'm trying to do is get it right the way it should be, and I am asking a lot from it.
Before you receive, you must first give. Receiving 270 Watts in impulses into 2 Ohms requires a lot you must give it for it to be able to swing that kind of transient power. To boot, I was never happy with IEC's definition of an impulse of 20 mS, to me that should be be more like 100 mS. Anyway, 100 mS can't hurt.
The net rersult is that some people just don't believe it's nominally a 50W per side amp into 8 Ohms, they think I'm kidding them. And in a sense, I probably am, because the current stage is fed by +/- 42V lines, so short term bursts of some 90W/8 Ohms are possible.
When I wrote this to James Bongiorno about 10 years ago, he wrote back and said he was glad to hear it because that's how it should be done and isn't for commercial reasons only. And I've taken his words to heart, for that and all other comments he gave me over the years. He was dead right all the way.
Transient behavior courtesy of Dr Matti Otala and all the crazy ideas he has put in my head. I'd love to buy that man a beer - a keg of it.
If peope will forgive . SVF . Having tweaked my SVF to be OK I wondered if anyone has just simply added multistage filtering rather than using the Rosen design and it's variants ? Is it diminishing returns ? My instinct says no . A design I handed to another some years back of a 10 section Chebishev he tweaked to - 120 db ( 50 Hz ) starting from a square-wave in . I had about - 70 dB with no real output difference between 48 an 64 Hz . He put in a detector and made it 50 and 60Hz adjustable with AGC . An FET AGC I used was a noticeable source of distortion . I don't care if the output varies as I have a very good bench oscillator for that . It is hopeless on distortion , excellent on level .
There is a place called Watership down ( as in song Bright Eyes ) . Our factory was there . We took the local "no hope kids " to work for us ( Beavis and Butthead's ) . All took drugs I suspect . I learned to make things simple with them in mind . More expensive sometimes that way . They had no idea who I was and sat me down to train me one day . After two hours I was asked who I was . Your designer I answered . They said why are we training you ? I said because you never asked me who I was . From then on I never had to work hard to train them . They were all excellent kids . I begged them not to take drugs, I think they did stop . Alas the company went offtrack and not due to financial problems . 23 of 24 got jobs nearly instantly when no one else was ! Number 24 was given work by a friend of mine but it seems the drugs got to him . Best work I ever did . Watership Down is very middle class . No one suspects " no hope kids " live there . Headley Tyres are great if you live in the area .
There is a place called Watership down ( as in song Bright Eyes ) . Our factory was there . We took the local "no hope kids " to work for us ( Beavis and Butthead's ) . All took drugs I suspect . I learned to make things simple with them in mind . More expensive sometimes that way . They had no idea who I was and sat me down to train me one day . After two hours I was asked who I was . Your designer I answered . They said why are we training you ? I said because you never asked me who I was . From then on I never had to work hard to train them . They were all excellent kids . I begged them not to take drugs, I think they did stop . Alas the company went offtrack and not due to financial problems . 23 of 24 got jobs nearly instantly when no one else was ! Number 24 was given work by a friend of mine but it seems the drugs got to him . Best work I ever did . Watership Down is very middle class . No one suspects " no hope kids " live there . Headley Tyres are great if you live in the area .
Last edited:
Just this morning, I had to tell my 27 year old son that since I am no longer 27 years old, I do NOT know everything. And I harbour some doubts.
When I was 16 years old, my father didn't know anything. By the time I turned 21, I was amazed how much he'd learnt in just 5 years.
I think this man was said to be the last person who knew everything ( nice ) . Forgive me for saying it so crudely .
Thomas Young (scientist) - Wikipedia, the free encyclopedia
Do you have steam railways in NZ Don ?
Thomas Young (scientist) - Wikipedia, the free encyclopedia
Do you have steam railways in NZ Don ?
Steam locos were withdrawn from service here over 50 years ago. There are a few in the hands of preservation societies. They run main line excursions occasionally. There's also a private line: http://en.wikipedia.org/wiki/Kingston_Flyer_(train)
dvv,
Well, to me, what you described, even only in terms of decoupling, excited one of my "resonant peaks". It's one of my "pet peeves", or something. I got infatuated with decoupling about a year or two ago and tried to revive some more of my former KSAs ("knowledge, skills, and abilities") in order to try to do "a deep dive" into the realm of calculating decoupling requirements, i.e. from current vs. time requirements to cap values and their connection inductances (i.e. lengths), paralleling, and all of that, with an eye toward DIY-available construction techniques, too, but also learning a lot about the way it's done for high-speed digital stuff with multi-layer PCBs.
I came up with some good stuff, I think, but still haven't quite "tied it all together", maybe because of a series of significant interruptions in my "real life".
But anyway, I have harped on decoupling, quite a lot, in several threads around here, with mixed results (which tends to prove, yet again, that being able to communicate as well as the Madison Avenue types is often more important than knowing the answer or having the best idea).
So it was simply very gratifying to read that someone has really taken it all to heart, and is applying it, with gusto, and has been doing so for a long time!
I also was very glad to hear that you keep the large peak capability available, above the rated max output power, as it should be done. It's amazing how many people rate systems as if a single sine wave will be the signal. What if several peaks coincide? (Or is that impossible to get through mastering or whatever?) But also, if you need a transient to make a large, sharp edge, at the speakers, then maybe somewhere, internally, maybe in the driver of the output transistors, or even in the output stage itself, or the feedback, instead of the square edge you want as the result, you might see a very large overshoot peak, followed by a similar undershoot, to "force" the edge from the power stage to get it right. It reminds me of one of the first ideas I had about optimal control theory (as opposed to classical control theory), where instead of setting a set-point at the input and waiting for the system to respond, while concentrating on having a nice Q = 0.5 (or phase margin = 76 degrees) to make the response go as quickly as possible but without overshoot, instead you stand on the accelerator pedal until the last second and then stand on the brake pedal, and it's "perfect".
One of my hypotheses, about decoupling, was/is that if it's not implemented well-enough, which probably by definition means for the higher-frequency stuff, then the transient response accuracy could vary with frequency, at low-enough (but probably still quite high in the audio range) frequencies that, essentially, the Fourier components of an edge would no longer be correctly aligned, in time (or phase angle), and the edge would be distorted, and would probably look like there was ringing too. My hypothesis for that case is that the inaccurate transient response, with frequency-dependent inaccuracies, would degrade some subtle higher-frequency-dependent "timing cues" and thereby degrade the quality of the soundstage imaging (and probably other aspects of sound quality, as well).
Part of my reasoning was that since it seems to have become clear that extremely low THD (and other steady-state-type metrics) does not necessarily correlate with achieving "profound high fidelity", and much of music is "transients", anyway, then the difference between a good amplifier and a great amplifier could lie (to some extent at least, and I'm thinking "large extent") in the transient response's accuracy. It probably can't be some very large difference, after all. So maybe it's in the last few percent (or less) of getting the transient response "just right".
And, now that I think about it, by "transient response", I think we would possibly have to go significantly beyond a simple step response or impulse response criteria and testing/measurements. I know that the theory says that the impulse response completely defines the system's dynamic behavior and is equivalent to the transfer function, in information content. But in the real world, maybe things aren't that linear. So it seems like we might need to try to worry about every possible pair of initial and final commanded states.
Sorry. It's getting late and I was just "thinking out loud" (and not trying to imply that I have thought through it all very well). Probably much is wrong in what I wrote. But maybe it will spark a thought in someone.
Decoupling is interesting, because at first it can seem that it has two sort-of opposite driving ideas. The high-frequency part needs to deal with fast rise-times, probably with the worst case of supplying the current for the max slew rate from zero to the rail. And we can figure out what minimum capacitance and what maximum tolerable inductance can accomplish what is needed (and we usually even have to parallel caps to try to lower the overall inductance). And it seems that shorter rise-times are the biggest problem to deal with, there. But then we see that for slower rise times, we need even more capacitance, because we have to supply the same current for a longer time. Following a rise-time or frequency progression that goes from fast to slow, we get a continuum of increasing capacitance AND allowable inductance (= conductor length plus lead-spacing, mostly), that leads us all the way to the reservoir capacitance.
It's also funny how so many people have such a voltage-centric view of the power supply. The current is where the action is. The current from the caps IS THE MUSIC. And the ripple isn't 120 Hz, except in textbooks, or with constant DC signals. The ripple voltage is (mostly) the integral of the music current. Luckily we have charging pulses fairly often!
Anyway, I did a bunch of math to come up with equations for things like the minimum capacitance required to supply the current for signals OTHER than constant DC. It's surprising how complex it becomes, even just for a sine wave, to take into account the difference between the charging pulse frequency and the signal frequency, et al. But, I wanted to be able to truly KNOW, in advance, what the lowest frequency would be, that a particular reservoir + decoupling capacitance could produce, without clipping, or "running out of charge before the next charging pulse", for a given amplifier. It's not easy and I still don't have it quite exactly right for the case where the signal frequency is lower than the AC Mains frequency.
But we don't have to worry about any of that, if we are willing to accept a little overkill in capacitance. We can simply assume that the worst-case signal will be a constant DC at the PEAK (not RMS) sine level that is implied by the RATED maximum output power. THEN we can handle any signal at all (within the rated max power spec, at least). I guess that if we leave headroom for peaks, as we should, then we should increase the capacitance requirements accordingly!
Time for bed. Sorry to have blathered-on for so long, again!
Your work is inspiring.
Cheers,
Tom
P.S. Nigel's (and many others') work and ideas are also inspiring. I just don't usually know enough to comment without embarrassing myself or contributing only noise. But I'm learning a lot (or trying, at least).
Last edited:
My belief is the single ended VAS is the least understood aspect of design . True it works . It is like a dog chasing it's tail to get it to work . More current in , bigger VAS Cdom . Eventually you win . Double VAS is easier as source and sink are similar . Tweak for HF balance and all is well . My recent adventure in single VAS has led me to believe it is true . I must build a better oscillator to be sure .
Do people feel a 1 to 1 ratio between LTP and VAS resistance is desirable ? I have seen it the other way so that 10 K feeds into VAS base at 2K5 ( 4 : 1 ) . I usually use 1:3 if I can .
Can excessive decoupling if using high grade caps be worse ? If a DAC perhaps close to DAC and HF ability count ? Ceramic 0.1 uF from 0 V to Vcc is best ( turn your CD PCB over and solder to SMD ) ? The fact ceramic is a bad word might not always be true ? Not talking NPO/COG .
Do people feel a 1 to 1 ratio between LTP and VAS resistance is desirable ? I have seen it the other way so that 10 K feeds into VAS base at 2K5 ( 4 : 1 ) . I usually use 1:3 if I can .
Can excessive decoupling if using high grade caps be worse ? If a DAC perhaps close to DAC and HF ability count ? Ceramic 0.1 uF from 0 V to Vcc is best ( turn your CD PCB over and solder to SMD ) ? The fact ceramic is a bad word might not always be true ? Not talking NPO/COG .
Last edited:
These seem to have been forgotten . Denis Morecroft often had interesting ideas . Simple idea to slit the foils to reduce eddy currents .
DNM Products - Capacitors (overview)
DNM Products - Capacitors (overview)
Not forgotten, Nige, it's just that the idea never really took off.
I tried them on several occasions, and for the life of me, I couldn't tell any difference between them and other quality caps.
But I could tell the difference in purchasing price outside the UK. So why spend more for something I cannot hear?
I tried them on several occasions, and for the life of me, I couldn't tell any difference between them and other quality caps.
But I could tell the difference in purchasing price outside the UK. So why spend more for something I cannot hear?
I suspect the company that makes them learned a lot . BBH I think ? Mission used them I think ?
I use a clone of DNM speaker cable . Telephone cable you might call it . Dennis said to me years ago " how much current can 79 strand cable carry ( QED ) " ? I said 500 A instantaneous . He said I was the only one to understand that . I guess at 50 A without it getting too hot continuous ( 50 C free air ) . Dennis always said thin / solid cable sounds best . When we bi or tri wire it is a point of view worth trying . Bass whatever you like . Mid 0.6 mm dia . Top 0.25 mm dia . For inter connects pick up internal wire is good ( AT 95 ) . Use 300 ohm type construction .
I use a clone of DNM speaker cable . Telephone cable you might call it . Dennis said to me years ago " how much current can 79 strand cable carry ( QED ) " ? I said 500 A instantaneous . He said I was the only one to understand that . I guess at 50 A without it getting too hot continuous ( 50 C free air ) . Dennis always said thin / solid cable sounds best . When we bi or tri wire it is a point of view worth trying . Bass whatever you like . Mid 0.6 mm dia . Top 0.25 mm dia . For inter connects pick up internal wire is good ( AT 95 ) . Use 300 ohm type construction .
Hi Tom,
Well, you’ve just about said it all, so rather than discuss what we obviously agree on, let me tell you how I like to do it. This as a result from 40+ years in this hobby, and 30 years or so of avidly looking at how other people did it, thinking (for what’s that worth) and analyzing, and in the end, trying it out live.
Current starvation.
My enemy No.1! I have on many occasions found this to be the prime (though often not the only) reason why amp A sounds better than amp B – A has a bigger and better power supply. By a stroke of luck, I started looking at this when I owned AR 5 and later AR 94 speakers. The 94s are the world’s first ever 2.5 way speakers, as the two bass drivers are different. One has a larger spool and works in parallel with the second driver only up to about 85 Hz, and is gradually cut off with a gentle 6 dB/oct slope thereafter via a simple inductor. Despite much care, they are still not an easy load to drive, even if they are not as notorious as say Apogees or Dahlquists of this world.
Some amps almost choked on them, some didn’t seem to mind. As a rule, those with better power supplies fared better to much better than those with seedy PSUs. No surprise there, but it’s still interesting to switch off the ’scope and just listen. No problem in hearing the difference.
Most fortunately for me, a very good friends of 35 years owns the Apogees, so I can do my final tests on them, and trust me, they are the acid test. When my nominally 50 WRMS amp at room volume starts to sound like his nominally 200 WRMS amp, I reckon I’m doing something right.
Transient response.
Now, there’s a big, large field. The obvious aspect is how well does it respond to suddenly changing signals, but not so obvious is how well does it not only rise, but also recover from this, i.e. its settling time. On that, Keith Johnson of Spectral has been an inspiring example which I have tried hard to follow. To cut a long story short, my view is that fast rise times are literally only one half of the story, while the settling times are the second one. Missing out on that is like telling a story for which you have no end.
Eventually, I’ve settled on my VAS being of cascode persuasion, with a fast recovery diode collector to base of the lower transistor. I have also settled on BF 720/721 being my lower transistors and spent a lot of time experimenting with them, in order to really get to know them. Eventually, I’ve discovered that they team up with 2N 5551/5401 very well indeed, and with MPSA 56/06 even better. Just keep them at around 5-6 mA and you have an exceptionally linear both device and stage.
Obviously, there must be a number of similar combinations out there, and the point is that one needs to test it to some extremes to be really sure and to know what they can and what they cannot do well. The assumption that a transistor is just a transistor is like saying a car is just a car – and we all know what folly that assumption is (Rolls-Royce and the Mini are the same?)
Open Loop Power Bandwidth.
I am sure quite a few folks here could produce the math to conclusively prove that one does not need a wide open loop power bandwidth for good sound, but nevertheless to my ears, just such amps somehow always have an edge on their different cousins – much along the Harman/Kardon way. I’m not saying they got it just right every time, that wouldn’t be quite true, but by and large they do manage to climb up the hill oftheir respective price range more often than most others. And from top to bottom of their range, they use wide bandwidth designs with global NFB at 26 dB worst case (in the mid-80ies) to 12 dB (Citation series, 680 integrated, etc). To this day, I have never heard of an H/K amp becoming unstable, no matter what. And the dynamic peak power they are typically capable of is very, very good.
So 50 kHz is the highest I really take care about. If it’s good at 50 kHz, I am able to conclude that I have no dynamic problems to speak of – and that’s open loop.
I would only remind you that while most of my stuff has a closed loop full power bandwidth in excess of 300 kHz, I still use an input filter which has its -3 dB point at 200 kHz.
Sizes.
I take an opportunistic view of my power transformers. I simply look at what sort of dynamic power I expect of them, say 270 W/2 Ohms, then pick out a transformer capable of that and multiply its rating by 1.41. So, 270 x 1.41 = 380.7 VA, nearest standard value 400 VA.
Anything but science, yet that simple rule has never once let me down. Although, to be honest, as the supply line voltage rises, this rule can be relaxed a bit. For a nominally 100 W/8 Ohm amp, I plan to use „just“ 500 VA transformers per channel. That’s at least double of what the industry uses under similar circumstances.
Trick (and sometimes Treat)
A nice habit I picked up from Otala’s amp in 1974 is to add to the usual smaller cap values in parallel placed on the PCBs (100uF // 3.3 uF // 220 nF) is a series 1 Ohm resistor with a xxx nF cap from supply line to ground. A Zobel, if you like. Over the years, I found that this little stunt can clean up the treble range like you wouldn’t believe. However, to the best of my knowledge, there is no way to calculate the value of C because it depends on everything preceeding it. It’s simply a heueristic process, although experience helps, but it might be anything from 220 to 680 nF. Just listen and change values. Very good as an add-on after the fact, too.
Using that AND 2,200 uF caps as near to the output transistors as physically possible will most likely give you all the speed you could ever want. A 2,200 uF cap should be capable of supplying 2 A peak easily, and more important virtually instantaneously, which is what we are hardly likely to ever need in normal room use. The residual inductance of those caps is also rather small, assuming you are using quality caps.
That's about it, I probably missed something, but not the heatsinks - a simple rule, use as big and efficient as you can possibly install and damn the price.
Well, you’ve just about said it all, so rather than discuss what we obviously agree on, let me tell you how I like to do it. This as a result from 40+ years in this hobby, and 30 years or so of avidly looking at how other people did it, thinking (for what’s that worth) and analyzing, and in the end, trying it out live.
Current starvation.
My enemy No.1! I have on many occasions found this to be the prime (though often not the only) reason why amp A sounds better than amp B – A has a bigger and better power supply. By a stroke of luck, I started looking at this when I owned AR 5 and later AR 94 speakers. The 94s are the world’s first ever 2.5 way speakers, as the two bass drivers are different. One has a larger spool and works in parallel with the second driver only up to about 85 Hz, and is gradually cut off with a gentle 6 dB/oct slope thereafter via a simple inductor. Despite much care, they are still not an easy load to drive, even if they are not as notorious as say Apogees or Dahlquists of this world.
Some amps almost choked on them, some didn’t seem to mind. As a rule, those with better power supplies fared better to much better than those with seedy PSUs. No surprise there, but it’s still interesting to switch off the ’scope and just listen. No problem in hearing the difference.
Most fortunately for me, a very good friends of 35 years owns the Apogees, so I can do my final tests on them, and trust me, they are the acid test. When my nominally 50 WRMS amp at room volume starts to sound like his nominally 200 WRMS amp, I reckon I’m doing something right.
Transient response.
Now, there’s a big, large field. The obvious aspect is how well does it respond to suddenly changing signals, but not so obvious is how well does it not only rise, but also recover from this, i.e. its settling time. On that, Keith Johnson of Spectral has been an inspiring example which I have tried hard to follow. To cut a long story short, my view is that fast rise times are literally only one half of the story, while the settling times are the second one. Missing out on that is like telling a story for which you have no end.
Eventually, I’ve settled on my VAS being of cascode persuasion, with a fast recovery diode collector to base of the lower transistor. I have also settled on BF 720/721 being my lower transistors and spent a lot of time experimenting with them, in order to really get to know them. Eventually, I’ve discovered that they team up with 2N 5551/5401 very well indeed, and with MPSA 56/06 even better. Just keep them at around 5-6 mA and you have an exceptionally linear both device and stage.
Obviously, there must be a number of similar combinations out there, and the point is that one needs to test it to some extremes to be really sure and to know what they can and what they cannot do well. The assumption that a transistor is just a transistor is like saying a car is just a car – and we all know what folly that assumption is (Rolls-Royce and the Mini are the same?)
Open Loop Power Bandwidth.
I am sure quite a few folks here could produce the math to conclusively prove that one does not need a wide open loop power bandwidth for good sound, but nevertheless to my ears, just such amps somehow always have an edge on their different cousins – much along the Harman/Kardon way. I’m not saying they got it just right every time, that wouldn’t be quite true, but by and large they do manage to climb up the hill oftheir respective price range more often than most others. And from top to bottom of their range, they use wide bandwidth designs with global NFB at 26 dB worst case (in the mid-80ies) to 12 dB (Citation series, 680 integrated, etc). To this day, I have never heard of an H/K amp becoming unstable, no matter what. And the dynamic peak power they are typically capable of is very, very good.
So 50 kHz is the highest I really take care about. If it’s good at 50 kHz, I am able to conclude that I have no dynamic problems to speak of – and that’s open loop.
I would only remind you that while most of my stuff has a closed loop full power bandwidth in excess of 300 kHz, I still use an input filter which has its -3 dB point at 200 kHz.
Sizes.
I take an opportunistic view of my power transformers. I simply look at what sort of dynamic power I expect of them, say 270 W/2 Ohms, then pick out a transformer capable of that and multiply its rating by 1.41. So, 270 x 1.41 = 380.7 VA, nearest standard value 400 VA.
Anything but science, yet that simple rule has never once let me down. Although, to be honest, as the supply line voltage rises, this rule can be relaxed a bit. For a nominally 100 W/8 Ohm amp, I plan to use „just“ 500 VA transformers per channel. That’s at least double of what the industry uses under similar circumstances.
Trick (and sometimes Treat)
A nice habit I picked up from Otala’s amp in 1974 is to add to the usual smaller cap values in parallel placed on the PCBs (100uF // 3.3 uF // 220 nF) is a series 1 Ohm resistor with a xxx nF cap from supply line to ground. A Zobel, if you like. Over the years, I found that this little stunt can clean up the treble range like you wouldn’t believe. However, to the best of my knowledge, there is no way to calculate the value of C because it depends on everything preceeding it. It’s simply a heueristic process, although experience helps, but it might be anything from 220 to 680 nF. Just listen and change values. Very good as an add-on after the fact, too.
Using that AND 2,200 uF caps as near to the output transistors as physically possible will most likely give you all the speed you could ever want. A 2,200 uF cap should be capable of supplying 2 A peak easily, and more important virtually instantaneously, which is what we are hardly likely to ever need in normal room use. The residual inductance of those caps is also rather small, assuming you are using quality caps.
That's about it, I probably missed something, but not the heatsinks - a simple rule, use as big and efficient as you can possibly install and damn the price.
Sorry, but I fail to grasp the benefits of thin cables, just as much as I have never heard (yet) a solid core cable which would outdo better stranded cables.
BTW, my van den Hul Hybrid 352 cables are 2 x 256 strands. Does that mean I can pass over 1,000 A of instantaneous current? Jesus, Wayne should love that!
When you say settling times do you include getting outputs to switch off faster ? I like FET's for that as they do not have the same mechanisms . Double VAS helps make source and sink symmetrical which is a way of looking at a similar problem . It also doubles slew rates whilst keeping simplicity and lower DC offset .
Do you remember when you first used the diode on the cascode ?
Thin cable is more like the internal of the speaker coil or output transformer if valves . Nordost is a good version .
A few answers here
http://www.bcae1.com/wire.htm
Do you remember when you first used the diode on the cascode ?
Thin cable is more like the internal of the speaker coil or output transformer if valves . Nordost is a good version .
A few answers here
http://www.bcae1.com/wire.htm
Last edited:
Of course I do Nige, I do all my final measurements at the output of the entire amp.
I do, we elephants have a long memory. It was in 1979, less than a year after I saw it done on that German made amp, LAS Mega 1. Much to my delight, there were technical comments with the original review, from which I learnt MUCH.
Two tails I have since then is that speedup diode for 1/10 of the original settling time and the protection circuits with a programmable delay.
Nige, you for one would have been delighted with some of German industry's work from those days. Some really smart people there, believe me.
Will look it over, thanks.
- Status
- Not open for further replies.