I don't think that was quite so, Dave. In the 70ies, everybody decared their goods as per the IHF standards, and they specify nominal power output into an 8 Ohm load, at or below nominal THD specs, 20-20.000 Hz. Typically:
Minimum of X Watts, into 8 Ohms, 20-20.000 Hz, with no more than Y % THD.
Yes, gate resistors are in place - this is not my design, I'm playing with another person's ideas, to see how far it can be pushed, where the weaknesses and room for improvements occur - it's a learning exercise, which may go somewhere, maybe not ...
Dejan, I'm pushing the output stage hard into low impedances, and even at 20kHz an individual gate will demand spikes of a number of mAs - get a few pairs in parallel and things start to get a bit serious.
Frank, on tests you can see that MOSFETs are capable of serious currents, the problem is that to me they simply do not sound as authoritative on bass lines as well selected BJTs do. They appear to be more polite than truthfull, they have bass, but it's somehow "rounded off" to my ears.
Dejan, this design is the conventional complementary configuration - matching amongst a particular type is the tricky part, where the design really gets interesting - I'm playing with a few ideas here.
I have had experience with MOSFETs, the Perreaux being a classic - its problems were all about power supplies, and I suppose you could call the symptoms a "politeness". I stiffened that supply, and it then maintained good impact in the bass, would deliver tremendous punch if that was in the recording.
My overall philosophy would be that any of the approaches can be made to work, if the weaknesses of the particular implementation are understood, and sufficient countering tweaking and re-engineering are applied - it then becomes a question as to whether the effort is worth it, in the specific instance.
I have had experience with MOSFETs, the Perreaux being a classic - its problems were all about power supplies, and I suppose you could call the symptoms a "politeness". I stiffened that supply, and it then maintained good impact in the bass, would deliver tremendous punch if that was in the recording.
My overall philosophy would be that any of the approaches can be made to work, if the weaknesses of the particular implementation are understood, and sufficient countering tweaking and re-engineering are applied - it then becomes a question as to whether the effort is worth it, in the specific instance.
IME, the old Perreaux's sound clean, clear and quite powerful.
However, to my ear they have always had the slight politeness that you mention, and more so, a subtle 'glassiness' for want of better term.
High 1/f noise corner might be the miscreant.
Dan.
The Quad clone needs a bit of woodwork so that will hold it up a bit. Last years baffle speaker has come back to life. Although I know it is a mistake I have gone down the conventional route. I am using 4 x 2 foot x 1 inch. The 1 inch is half MDF and half hardwood ply. The gluing done with 3 very large buckets of water and using my tiled floor as a level surface ( wood to spread the load ). Seems to be working. I would have prefered a low mass material. Too much like hard work and want to use them sometime before when I am 80. I borrowed an Elu router with hole cutting bar. How easy it is when having the right tools. 10 minutes per hole. The wood cost a fortune so hope it works. The idea of half and half seems better than I planned. The MDF helps the ply flatten. The other idea is to stop the ply having the resonance of a stiff material and the over damped MDF to be less so. Ply to the front as it looks nice and I suspect is the right place. EQ is + 16 dB 30 Hz. Driver 12 Lta. The looks is like mahogany.
I think it is simply Ron that causes FET's to lack slam. 3 devices will be like the old Quad 303 in terms of emitter resistance ( 1 x 2N3055 that is ). Ron if lucky is 1 ohm. If the bias is set low it helps a bit ( 20 mA ). The Fetlington is a reasonable idea. That is FET first. The difficult choice is what current suits the bipolar. My guess is make the feedforward resistor 0R33. The Bipolar resistor just a 5 A fuse. That is as it it will do 2 jobs if asked. The bipolar will be in class B/C. This is not the Quad 405. It is just a Fetlington. The beauty of the Fetlington is biasing is easy. Do it by ear and how hot.
One could use an alll N FET design using industrial FET's. I have some 10N50 in T03P. Also PSU FET's. These need Vbe biasing like bipolars and are in many ways the worse of all devices. However they are cheap and very fast. Ron < 0.1R is not unusual. SOA is usually a bit better than bipolar. Reading carefully few biopolars are > 3 amps real world sinewave. That is when the voltage rails high enough that's about the limit. Not very good really. Thank goodness real watts and music are vastly different.
One could use an alll N FET design using industrial FET's. I have some 10N50 in T03P. Also PSU FET's. These need Vbe biasing like bipolars and are in many ways the worse of all devices. However they are cheap and very fast. Ron < 0.1R is not unusual. SOA is usually a bit better than bipolar. Reading carefully few biopolars are > 3 amps real world sinewave. That is when the voltage rails high enough that's about the limit. Not very good really. Thank goodness real watts and music are vastly different.
Could be, Dan, could be. I never had a reasonably good source of MOSFETs, it was always a pain getting them, so I never experimented.
Wrong question put to a wrong man, Nigel never was one of the NFB glutton kings, he was of the investigative persuasion all along, he likes to learn what he has to in order to solve the problem. As old school as the RAF moustache.
Nige, before you proceed any further, please go to HiFi Engine - Owners and Service Manuals, register (it's free), and download the service schematics for my Harman/Kardon PA 2400 power amplifier, and look closely at their output stage construction. Never seen anything like it, but let me tell you, it certainly works well.
It might give you interesting ideas. That's what I'm hoping for here. Even if it's a BJT design.
I would also advise everyone else to do the same. Do not waste your time reinventing hot water, take a long good look at how others dealt with a certain problem. That site is a treasure cove of service schematics.
It might give you interesting ideas. That's what I'm hoping for here. Even if it's a BJT design.
I would also advise everyone else to do the same. Do not waste your time reinventing hot water, take a long good look at how others dealt with a certain problem. That site is a treasure cove of service schematics.
Last edited:
I've been forgetting for some days a bit of info I'd like to devulge, Nige, regarding your VAS. You used what is basiclly a current mirrored cascode stage. As you are well aware, I have been playing with that a lot lately, and with significant input from Demian Martin, who kindly offered help. I tried something almost exactly as you produced, but soon discovered that the version with buffers preceeding the VAS in a single differential mode (no cascodes) considtently produced a wide open loop response.
This did surprise me not just a bit, as I was conditioned to use cascodes for that, but however I turned or twisted it, this was not so, the buffer version did better. Finally, I shrugged off my own disbelief when I remembered that my Marantz 170 DC power amp uses something along buffer lines rather than cascodes, which are used in the input stage, using JFETs, to overcome the voltage limitation. In the final prototype, using four MJL 3281/1302 200W power devices, I got an open loop full power bandwidth up to a whopping point of 130 kHz. The damn thing almost needs no NFB (open loop full power THD 20-20.000 Hz is less than 0.4% into 4 Ohms), but I threw in my usual 20 dB just to sleep easy and got worst case THD 20-20,000 Hz of just 0.03% into 4 Ohms. Rise time was just under 1 uS, and slew rate worked out to some 300 V/uS.
So, my suggestion is that you try reworking your sample. Forget the current mirror in theIP stage, use the existing pair as cascodes to a pair of 2SK170 JFETs. They are still being produced, and can be bought from China for very reasonable money. Bias them at 5 mA per device and bias the VAS at around 20 mA per device, using 2SC3503/2SA1381, with local heat sinking, they will be dissipating 1.3-2 Watts. Look at their data curves and you'll see they need over 10 mA to get into their mot linear act. I tink dat ting gonna ROCK you, dude.
This did surprise me not just a bit, as I was conditioned to use cascodes for that, but however I turned or twisted it, this was not so, the buffer version did better. Finally, I shrugged off my own disbelief when I remembered that my Marantz 170 DC power amp uses something along buffer lines rather than cascodes, which are used in the input stage, using JFETs, to overcome the voltage limitation. In the final prototype, using four MJL 3281/1302 200W power devices, I got an open loop full power bandwidth up to a whopping point of 130 kHz. The damn thing almost needs no NFB (open loop full power THD 20-20.000 Hz is less than 0.4% into 4 Ohms), but I threw in my usual 20 dB just to sleep easy and got worst case THD 20-20,000 Hz of just 0.03% into 4 Ohms. Rise time was just under 1 uS, and slew rate worked out to some 300 V/uS.
So, my suggestion is that you try reworking your sample. Forget the current mirror in theIP stage, use the existing pair as cascodes to a pair of 2SK170 JFETs. They are still being produced, and can be bought from China for very reasonable money. Bias them at 5 mA per device and bias the VAS at around 20 mA per device, using 2SC3503/2SA1381, with local heat sinking, they will be dissipating 1.3-2 Watts. Look at their data curves and you'll see they need over 10 mA to get into their mot linear act. I tink dat ting gonna ROCK you, dude.
Last edited:
I think it would be more accurate to admit that many designers no longer perform transient and phase analysis. There has been a movement pushed by old minds that they are not needed - a fairly presumptuous error. There are also the unsophisticated minds in the business that fall the other way and begin to believe that there is something in electrical waveforms that can't be measured. That is the pseudoscientic anti-thesis of academic research and progress. Some of the Japanese companies did provide measurements along with their numerical specs of the products at various output levels. However, most others and even companies advertising on this site still don't. Instead thry throw on spec taken at 1kHz or tell their customers that specs don't matter. No, they don't matter if they are without any context or conditions.
Just a note, watch yourself around those TIM distortion articles, because they are laced with error. Not one of Otala's TIM papers can referenced from any of our local university libraries, because those articles (now hosted by the AES) are blatantly false. The papers were produced under the assumption that NFB induced TIM distortion, which was a critical error in judgement and easily disproven. TIM cannot exist in the absense of SID, as it would be the result of slew-rate limiting. The cause was miller effect from capacitive degeneration and stability compensation, and had absoluetely nothing to do with the magnitude of NFB at all. Any Prof. engineer worth his or her salt knows this, had it demonstrated, and can demonstrate it themselves for others.
TIM was a misnomer for a distortion that was already known prior to the 1970's, ie TID. But, Otala's incorrect information polluted the audio press without proper peer reviews. Transitory Intermodulation Distortion, TID, was the real name and it only affects equipment at RF frequencies, unless someone gets overzealous with their miller compensation values. In short, an audio amp with a bandwidth of 100kHz, or more, will not possess the requisite criteria to permit the existance of Transitory Intermodulation with multiple tones on any audio recordings. It literally isn't possible. Anyone that states otherwise is not a credible source, and any papers stating otherwise should be stricken from educational doctrine.
Last edited:
You are only partly right, because you forget that at the time the initial paper was published (1973, IEEE) amps WERE made much more differently than they are today. Too much reliance on NFB indeed can cause TIM problems, but once identified, the practice was corrected, threfore the accurate thing to say would be that much to the disdain of posh Otala bashers, his theories DID cause tangible improvement of design practices. Then the exact opposite movement was atsrted in the industry, people like Sanui, Kenwood, Pioneer, etc swnr extreme and started a slew rate wars. It was they, not Otala, who promoted the idea that slew rate was the new cure all, wich of course it is not, but it does go a long way in getting rid of TIM (which includes TID, IID, etc). So, like it or not, Otala and Lohstroh's paper had a tremendous impact overall and over time, no matter what the naysayers claim. Equally of course, over time, we did learn a thing or two, the most iportant of which, in my view, is that we need to look for a balanced dsign, one in which nothing is given precedence than something else, because it all has to work together to be just right. No single cure alls.
As for demonstration, in my viw, my H/K power amps which use only 12 dB of global NFB, will gladly take on any other one at any price, while you measure TIM however you like it. But then, I'd want you to measure peak power available into low impedances, like 2 Ohms with a -45 degree phase angle, let's see who does better. I did all that and my results are quite decisive - lower global NFB, and higher open loop bandiwdth amps win hands down.
BTW, Miller compensation, that's what? Not a form of feedback, albeit very local?Do you seriously propse that this is no longer a problem in the industry? Some still struggle with it, believe me, just look over some service schematics of modern amps, and it's not for lack of knowledge, but for lack of decent engineering, despite the glowing diplomas they can flash around.
And, BTW, it is not transitory but transient. Please quote sources prior to the Otala/Lohstroh article of 1973 where this phenomen was discussed at any length, I'd really like to see it or them.
Your claim that it affects only RF equipment is completely wrong, it will happen in anything where the input rate of exchange exceeds the output rate of exchange, i.e. where the amplified output cannot keep pace with fast changes at the input side. However, it's true that people sometimes overstimate and overdo the voltage slew rate, in general it is accepted that as long as your output slew rate is 1V/us for every PEAK volt of output, slew rate will no longer be a problem. So, to be able to write it off for a nominaly 100W/8 Ohms amp, 40 V/uS does the job nicely.
Your claim that it affects only RF equipment is completely wrong, it will happen in anything where the input rate of exchange exceeds the output rate of exchange, i.e. where the amplified output cannot keep pace with fast changes at the input side. However, it's true that people sometimes overstimate and overdo the voltage slew rate, in general it is accepted that as long as your output slew rate is 1V/us for every PEAK volt of output, slew rate will no longer be a problem. So, to be able to write it off for a nominaly 100W/8 Ohms amp, 40 V/uS does the job nicely.
Last edited:
Your replies are incorrect. Roddam, 1959, appeared to be the first to publish work on transitory intermodulation distortion. It could never affect any amplifier with a sufficient slew rate and bandwidth for its voltage gain, the two are insperable, as Miller compensation is the predominent determining force of bandwidth by controlling slew rates. The fact that you do not know what Miller compensation is underlines the cause of your misundstanding. Miller compensation is used as a phase advancement method to attenuate a devices frequency response as it approaces its upper pole. The side effect, and inherent method by which it works, of this compensation method is that it impedes slew rates.
You have some serious reading catchup to do in the meanwhile, and it won't all be doable in one sitting. There have been a host of peer reviewed papers to highlight the problems in Otala's work. Dr. Edward Cherry and Bob Cordell come to mind as some worth reading, and there are other better doctoral works on the subject that directly address and state that Otala's mathmatics, methods, and blame on feedback were incorrect. Otala's team were under the belief that high feedback caused their distortion, and they failed to note it was the stability compensation in parallel to the feedback that was limiting the slew rates. When they used less feedback, the closed loop bandwidth of the devices they were using was reduced to such extent that little miller compensation, if any was needed. This was not an answer, as low feedback resulted in high distortion and a narrow frequency range, albeit without any chance of SID or TID. Even after that, Otala failed for many years to make the connection that the distortion was the direct result of slew-limiting from over-compensation. I've even taken to task and tried to recreate it, to test the claims, and could not induce real TID below AM radio frequency range. You can read more about Transitory Intermodulation Distortion in microwave and senior signal processing journals. Sure, a poorly implemented mass amplifier can produce slew limited distortion, but that's out of context by some margin. I'm on a mobile device, but when I'm on PC I'll link some enlightning papers that support what I've written.
You have some serious reading catchup to do in the meanwhile, and it won't all be doable in one sitting. There have been a host of peer reviewed papers to highlight the problems in Otala's work. Dr. Edward Cherry and Bob Cordell come to mind as some worth reading, and there are other better doctoral works on the subject that directly address and state that Otala's mathmatics, methods, and blame on feedback were incorrect. Otala's team were under the belief that high feedback caused their distortion, and they failed to note it was the stability compensation in parallel to the feedback that was limiting the slew rates. When they used less feedback, the closed loop bandwidth of the devices they were using was reduced to such extent that little miller compensation, if any was needed. This was not an answer, as low feedback resulted in high distortion and a narrow frequency range, albeit without any chance of SID or TID. Even after that, Otala failed for many years to make the connection that the distortion was the direct result of slew-limiting from over-compensation. I've even taken to task and tried to recreate it, to test the claims, and could not induce real TID below AM radio frequency range. You can read more about Transitory Intermodulation Distortion in microwave and senior signal processing journals. Sure, a poorly implemented mass amplifier can produce slew limited distortion, but that's out of context by some margin. I'm on a mobile device, but when I'm on PC I'll link some enlightning papers that support what I've written.
Last edited:
The fact that you resort to personal insulta, such as suggesting that someone, in this case me, doesn't know what Miller compensation is, shows how well you do not want to accept anything but the "established" truths. You have not refuted any of my statements by any fact.
I lack your wisdom to say that an incorrectly compensated amp will produce TRANSIENT intermodulation is out of conetx for this topic, which is rečated to audio. As I said, take a look how quite a few modern amps are made and you might be suprised. Straight out of the 60ies, in this day and age. And you do not need to get to the AM range to have it, either, just base your production on the Miller cap, the one I know nothing of, as the key compensation, and use an input diff pair with a low bias current to improve your S/N ratio, and presto, you got it.
It may surprise you to learn that I too have been playing with that for the last 35 years or so, and outting together things even if I can't tell the difference ebtween a Miller cap and a diode, and no amount of fancy maths is going to disprove what I have experienced myself.
I see you are about to become a respected member of the Otala bashing club which is fairly strong here. As such, further communication on this topis is of no interest to me, I don't like fedayeen of any kind.
I lack your wisdom to say that an incorrectly compensated amp will produce TRANSIENT intermodulation is out of conetx for this topic, which is rečated to audio. As I said, take a look how quite a few modern amps are made and you might be suprised. Straight out of the 60ies, in this day and age. And you do not need to get to the AM range to have it, either, just base your production on the Miller cap, the one I know nothing of, as the key compensation, and use an input diff pair with a low bias current to improve your S/N ratio, and presto, you got it.
It may surprise you to learn that I too have been playing with that for the last 35 years or so, and outting together things even if I can't tell the difference ebtween a Miller cap and a diode, and no amount of fancy maths is going to disprove what I have experienced myself.
I see you are about to become a respected member of the Otala bashing club which is fairly strong here. As such, further communication on this topis is of no interest to me, I don't like fedayeen of any kind.
Last edited:
Why would anyone want to defend the commendum of misapprehensions and incorrect assumptions by Otala and his fellow colleages? That stuff is being culled from academia and will be available nowhere else than on audio sites, because it's patently incorrect. My problem isn't with yourself, Matti Otala, or anyone else, I only take issue with their false information.
Last edited:
- Status
- Not open for further replies.