I believe that is the case. As far as I know Bruno, he is not religious about these things. You will also note that his arguments in his article about the F-word are not for a specific type of amp - they are valid for any amp type, any class.
jan
If one says something, and everybody misunderstands, then that someone was obviously not clear enough. Let me make amends.
I did mention a 1uF cap in testing power amp square waves at 10 kHz. However, I took that as only the most primtive, lowest level test, an entry point test, if you like. By no means did I assume that was that, pass that and you're home and dry.
For a full house test, I need an LRC mix much closer to representing a loudspeaker load, no single component "solution" will ever do. So, rather than approximate whatever, I use actual speaker crossovers, the good Lord knows that's easy enough with current simulators, which are my starting point.
Now, there will be somebody here who will start knocking simulators for being only simulators - but, that's what they are sold as, and they make no further claims to glory. I must also add that the one I am using is VERY conservative and utterly reliable in electrical terms, meaning that if it says something will work, it WILL work - not once has it cheated me over the years. Furthermore, it will consistenly show very conservative results; if it says it'll get to say 300 kHz, in real life it will probably get over 330 kHz, if it says THD of 0.1%, it will actually do say 0.08%, etc. This is good news to me, because the number of my prototypes needed before I can call it a day has been drastically cut do just 3...5, never more, depending on what it is.
And I understood from the start that it was just a tool which would reduce my design efforts, but that nothing is final or done unless it can do what it's supposed to do on real life models, hard wired. It has the first word, but my ears have the last word.
I did mention a 1uF cap in testing power amp square waves at 10 kHz. However, I took that as only the most primtive, lowest level test, an entry point test, if you like. By no means did I assume that was that, pass that and you're home and dry.
For a full house test, I need an LRC mix much closer to representing a loudspeaker load, no single component "solution" will ever do. So, rather than approximate whatever, I use actual speaker crossovers, the good Lord knows that's easy enough with current simulators, which are my starting point.
Now, there will be somebody here who will start knocking simulators for being only simulators - but, that's what they are sold as, and they make no further claims to glory. I must also add that the one I am using is VERY conservative and utterly reliable in electrical terms, meaning that if it says something will work, it WILL work - not once has it cheated me over the years. Furthermore, it will consistenly show very conservative results; if it says it'll get to say 300 kHz, in real life it will probably get over 330 kHz, if it says THD of 0.1%, it will actually do say 0.08%, etc. This is good news to me, because the number of my prototypes needed before I can call it a day has been drastically cut do just 3...5, never more, depending on what it is.
And I understood from the start that it was just a tool which would reduce my design efforts, but that nothing is final or done unless it can do what it's supposed to do on real life models, hard wired. It has the first word, but my ears have the last word.
I'm not so sure about that.
I've come across some really wild loads; retracing Otala's steps, I tried AR3a Improved, Yamaha NS-1000,etc, all the way to the notorious "amp killer", the Apogees.
The thing is that amps, the behavior of which with the 1uF cap proved better than with others, ALSO drove the said speakers more to much more convincingly than the others did.
So, while I do not claim that an amp which "fails" that oversimplified test cannot sound good, I am claiming that an amp which does pass that test has a better chance of sounding good with real world speakers than the one which doesn't pass the test.
Last edited:
I was actually trying to shoot for the other end of the spectrum, i.e. rather than just not dooming a design I was trying to look for ways of possibly taking it from good to great.
I assume that the 100μF/1μF/0.1μF are placed at EACH power device, and not just once per rail per board. Otherwisae the trace inductance would either degrade your transient response or increase the rail voltage disturbance amplitude, or some of both, not to mention the increased potential for high-frequency instability. (If any newer newbies than I ever read this thread: A good main power supply is a must. But unless the power rails are extremely short the PSU isn't sufficient for a great transient response and decoupling caps are needed, basically to act as small local power supplies, right at each point of load.)
Have you determined the decoupling capacitance value that would be needed for each output transistor for the worst-case transient current-slewing event, given a maximum allowed resulting power rail voltage disturbance, and the worst-case parasitic inductance that could be tolerated in that case, between the decoupling capacitance and the device?
Even just for an LM3886 to swing 5 Amps per microsecond for only two consecutive microseconds, and allowing for a spec of up to 0.1 V rail disturbance, with only one inch of round-trip trace length (say 15 nH, but would be more with bigger traces) between the main decoupling cap and each power pin and gnd, it would require about 200 uF minimum, if I recall correctly, implemented with three paralleled caps of 68μF each in order to lower the inductance enough to have the necessary speed. I realize that might not be a very realistic example for an LM3886. But hopefully it shows that it might be necessary to really make sure that what decoupling you are using, and exactly where, is actually sufficient.
And how do you decide that 1 uF and 0.1 uF are what is best, each time? Most people don't. And they're probably not.
Just food for thought.
Last edited:
Most of amplifiers are using a LR output network , so the vast
majority will drive a load with a 1 uF parasistic quite well.
Now , if the said output LR network is removed few will pass
such a test without oscillating or showing very large ringing,
particularly those who claim a high slew rate.
Some amps doesnt have the said network but generaly
they were designed accordingly , with reasonnable slew rate
and average but still good enoughTHD ratio.
majority will drive a load with a 1 uF parasistic quite well.
Now , if the said output LR network is removed few will pass
such a test without oscillating or showing very large ringing,
particularly those who claim a high slew rate.
Some amps doesnt have the said network but generaly
they were designed accordingly , with reasonnable slew rate
and average but still good enoughTHD ratio.
@gootee
I wasn't precise enough, sorry - yes, per output device.
Tom, did you know the very highly regarded Italian audio manufacturer, Galactron, used to install a 4,700 uF cap near each one of their MOSFET output devices?
By "near", I mean less than 1 inch away from the actual transistor.
Krell, in their FPB series from the mid-90ies until 2002, install "batteries" of no less than 3, and often 10 47uF caps, sprinkled around their voltage amplifying stages? And there are LOTS of them, I didn't count, but the voltage amplifying stage uses 112 transistors per channel.
I picked up "the habit" in the late 70ies, when an Italian designer, Mr Bartolomeo Aloia, created his version of a no reserve phono RIAA stage in conjucntion with "Suono", Italy's best selling audio mag at the time. His design had in turn 0.1 uF polyester (by Wima) and 10 uF tantalum caps (by Siemens) per each connection to both power supply lines.
Initially, I tried it without those caps and it was very good. Then I put the caps in, tried it again and this time, it was audiably better, more alive, with more ambience. That closed the case for me. If I don't believe my own ears, who will I believe?
I wasn't precise enough, sorry - yes, per output device.
Tom, did you know the very highly regarded Italian audio manufacturer, Galactron, used to install a 4,700 uF cap near each one of their MOSFET output devices?
By "near", I mean less than 1 inch away from the actual transistor.
Krell, in their FPB series from the mid-90ies until 2002, install "batteries" of no less than 3, and often 10 47uF caps, sprinkled around their voltage amplifying stages? And there are LOTS of them, I didn't count, but the voltage amplifying stage uses 112 transistors per channel.
I picked up "the habit" in the late 70ies, when an Italian designer, Mr Bartolomeo Aloia, created his version of a no reserve phono RIAA stage in conjucntion with "Suono", Italy's best selling audio mag at the time. His design had in turn 0.1 uF polyester (by Wima) and 10 uF tantalum caps (by Siemens) per each connection to both power supply lines.
Initially, I tried it without those caps and it was very good. Then I put the caps in, tried it again and this time, it was audiably better, more alive, with more ambience. That closed the case for me. If I don't believe my own ears, who will I believe?
My experience suggests that this is not always the case - especially to the ears of the system's owner.
For example, in many occasions getting more detail and removing the coloration of source/amplifier components (to the extent possible) causes deficiencies of the loudspeakers (and even loudspeaker/room combo) to become more evident.
To make things worse people are not always ready to accept that the aggressive highs they hear come from the jewel of their system, those brand new and expensive loudspeakers, and not from their new DAC or power amp.
Or that really their listening room and speaker positioning is not suited to those 4x 10'' drivers they paid $$$$ to get and show off - at least not when those drivers get enough power and actually move
I'm sure Jan will be happy to publish an article with the "correct" math.
As I said before, it's easy to talk the talk but harder to walk the walk.
Then again one might argue that if the HiEnd industry could afford bright scientist to do some solid R(esearch) before they actually get down to D(evelopment), we would see some real progress and not the recycling of 1970's technology.
But I guess most HiEnd companies can't afford employing such personnel nor do they have any real reason to invest in such (even in terms of outsourcing) since people keep buying that 1970's tech and are happy about it.
Thus, the fact that such aspiring bright minds wouldn't really be willing to work for a company of manufactory/workshop scale (aka 90% of the HiEnd companies out there) unless they REALLY love HiFi and are willing to "sacrifice" their careers doing something that'll get them no real money (as John Curls and others have often stated) or academic recognition, is irrelevant.
Regarding Bruno's preference, he's stated in numerous occasions that a properly built statement Class A amp will be untouchable by any Class D design.
But his definition of "proper" is not what we're used to seeing - esp. in the "less is more", "the 1st watt is the best", "feedback is bad" DIY world we live in.
An older Bruno's Class A design might give you some hints: ExtremA Reference Class A DIY amplifier
Last edited:
dvv, gootee,
I notice in one of the application guides for l-MOSFTES they are really specific to get both the gate resistor and tons of bypass right at they devices. I know in my digital logic days how critical it was to keep the bypass caps right at the legs. Some "bright" engineer wanted to save money and instead of a .001 at every chip, put a .01 at the end of the row. It was not cheap to lay out new boards ( 6 layer) and do it all over again. A class I tool from AMP in interconnection pretty much said you should put a bypass cap on anything that uses power and turns on or off. This is of source where the art of board layout gets hard.
Changing to the HEXFRED rectifier and new main caps, the ripple on the rails is about 25% less. About 12mV on the input side, 40 on the output side. No surprise. The shape of the little "burp" where the diode turns off looks a bit more rounded. I will try to look at the residual distortion today. I did keep a "before" graph.
I notice in one of the application guides for l-MOSFTES they are really specific to get both the gate resistor and tons of bypass right at they devices. I know in my digital logic days how critical it was to keep the bypass caps right at the legs. Some "bright" engineer wanted to save money and instead of a .001 at every chip, put a .01 at the end of the row. It was not cheap to lay out new boards ( 6 layer) and do it all over again. A class I tool from AMP in interconnection pretty much said you should put a bypass cap on anything that uses power and turns on or off. This is of source where the art of board layout gets hard.
Changing to the HEXFRED rectifier and new main caps, the ripple on the rails is about 25% less. About 12mV on the input side, 40 on the output side. No surprise. The shape of the little "burp" where the diode turns off looks a bit more rounded. I will try to look at the residual distortion today. I did keep a "before" graph.
Hi,
Why should I care?
If Bruno likes he can correct what he has written.
BTW, the effects are quite easily measured.
Sorry, I don't do '70's tech. It's either 1910's tech (Vacuum Tubes) or very recent.
And? Has that anything to do with the topic?
He is welcome to his definition. I know several of his designs quite intimately.
But let's recap.
For starters, up to now, not one Technocrat has provided proof positive that low THD is a valid design goal, in fact much evidence exists that show that comparably high distortion levels (of a certain spectrum) are in fact inaudible.
Secondly, Bruno's article elected to ignore significant factors that would have changed his analysis, meaning his analysis needs improvement. In fact the only value of the article I can see was that it attracted a range of criticism which ultimately gave a new insight (from BP), namely: "Negative feedback does not attenuate the harmonics of the open-loop nonlinearity, but of its inverse."
This should make it even more clear why disregarding distortion of distortion is a crucial fallacy.
Anyway, whoever may wish to believe BP's article is the ultimo ratio on the whole feedback subject is welcome to their beliefs.
Ciao T
Why should I care?
If Bruno likes he can correct what he has written.
BTW, the effects are quite easily measured.
Sorry, I don't do '70's tech. It's either 1910's tech (Vacuum Tubes) or very recent.
And? Has that anything to do with the topic?
He is welcome to his definition. I know several of his designs quite intimately.
But let's recap.
For starters, up to now, not one Technocrat has provided proof positive that low THD is a valid design goal, in fact much evidence exists that show that comparably high distortion levels (of a certain spectrum) are in fact inaudible.
Secondly, Bruno's article elected to ignore significant factors that would have changed his analysis, meaning his analysis needs improvement. In fact the only value of the article I can see was that it attracted a range of criticism which ultimately gave a new insight (from BP), namely: "Negative feedback does not attenuate the harmonics of the open-loop nonlinearity, but of its inverse."
This should make it even more clear why disregarding distortion of distortion is a crucial fallacy.
Anyway, whoever may wish to believe BP's article is the ultimo ratio on the whole feedback subject is welcome to their beliefs.
Ciao T
Presumably high THD = high IMD, and conversely low THD = low IMD..? It would take a very sophisticated (digital) box of tricks to provide harmonic distortion without intermodulation distortion. Are you therefore saying that high levels of some types of IMD are also inaudible? With what source material?
Power supply improvement for DH 120. Put in the Hexfred rectifier, new main caps of 2200uF instead of the 6600 and they have half the esr.
Referenced to 1.68v as 0DdB. 60Hz improvement of 6 dB, 120 Hz of 19 dB, 180 Hz of 13 dB. Drop in higher order in the 2 to 3 dB per harmonic. No residuals above 1K as was the original case. Noise floor below 60 Hz dropped by 5dB to -113.
So I am back to where I started from, confirming some basic power supply improvement.
I am happy it is an amplifier again. Hard to probe and work on the way it is stuffed in the case. I think I had best continue learning Spice and breadboard some of it rather than apply my bad vision and fat fingers to tight quarters.
Referenced to 1.68v as 0DdB. 60Hz improvement of 6 dB, 120 Hz of 19 dB, 180 Hz of 13 dB. Drop in higher order in the 2 to 3 dB per harmonic. No residuals above 1K as was the original case. Noise floor below 60 Hz dropped by 5dB to -113.
So I am back to where I started from, confirming some basic power supply improvement.
I am happy it is an amplifier again. Hard to probe and work on the way it is stuffed in the case. I think I had best continue learning Spice and breadboard some of it rather than apply my bad vision and fat fingers to tight quarters.
Hi,
Yes.
Yes.
Music, pure tones. SPL plays a role here. More HD/IMD is tolerated at higher levels.
Ciao T
Yes.
Yes.
Music, pure tones. SPL plays a role here. More HD/IMD is tolerated at higher levels.
Ciao T
He has shown that your arguments are in error, many pages back. Funny how memory deteriorates when you get older...
jan
Jan,
My memory must be indeed bad. I do NOT remember him providing numbers for "distortion of distortion" that allowed simple dismissal as unimportant, but please, what is the post where he does so?
Ciao T
My memory must be indeed bad. I do NOT remember him providing numbers for "distortion of distortion" that allowed simple dismissal as unimportant, but please, what is the post where he does so?
Ciao T
You know i can't recall ever hearing one ....
+10 , those that pass will go on to bigger and better things .....
- Status
- Not open for further replies.