The profound asymmetry of some real acoustical musical instrument signals is eye-opening too. I had a nominal 100W-channel amp prototype, a class-D design which did not clip gently into that good night. It nonetheless performed well driving some moderately efficient speakers, and I was enjoying things until I used a direct-to-disc LP as the source, and heard some amazingly ugly splatter on certain brass instruments. I took a look with the oscilloscope and the waveforms told all, asymmetrical and with an astonishing peak-to-average ratio.
In most recordings there's been enough processing that these waveforms are often drastically altered, usually with the effect of diminishing the crest factor and as well making things more symmetrical. This makes the signals far less demanding of the amplifier.
In most recordings there's been enough processing that these waveforms are often drastically altered, usually with the effect of diminishing the crest factor and as well making things more symmetrical. This makes the signals far less demanding of the amplifier.
I think the idea was the number of charge carriers "present" as it were, not the net number required for current flow. But this would be a good calculation to do.
One of the misconceptions IMO about supposed "granularity" in devices used in electronics is the degree to which "adjacent" regions "see" each other. This led to some whoppers in the early days of JFETs for example, when credentialed people cited the "shot noise" in the channel current. Whitlock scolded me for mentioning John Linsley-Hood's use of the term "tunneling" to further describe JFET physics in the pinchoff region, I think as much or more out of loyalty to the writer (much appreciated by me as well, but this was I believe a terminological blunder that we don't need --- things are hard enough already).
I have been thinking about this a bit further and in disagreement with what I stated earlier, what Keith Johnson may have meant is that in tubes you have this cloud of electrons surrounding the cathode which have already been emitted, but can't bridge the grid potential yet. But, once called upon, there are electrons in abundance for a surge. With semiconductors this works a bit differently. Every hole or electron has, before it can be collected, to work it's way through the lattice. Carrier mobility might be more what Johnson meant to say, but I am just guessing.
Hi,
I got 114dB or maybe 115dB, but you are in the ballpark.
Point is we are looking at 20dB+ crestfactor. Recorded music is usually compressed down to 10....14dB...
Ciao T
I got 114dB or maybe 115dB, but you are in the ballpark.
Point is we are looking at 20dB+ crestfactor. Recorded music is usually compressed down to 10....14dB...
Ciao T
Ok, Thorsten, now the 12 dB global NFB makes much more sense. Once you explain things fully, it always makes more sense, but I have to prod you to explain a lot. We can't know what's in your head unless you tell us.
Whether it's 12 or 20 dB, the point is clear enough, and we are in full agreement there.
My reservations stand, but I realize they are mostly a product of my own inexperience in thinking and practicing along similar lines as you do. To be frank, I never really thought of it all the way you do. Once I try to follow your reasoning, it begins to make sense (to me, at least).
Whether it's 12 or 20 dB, the point is clear enough, and we are in full agreement there.
My reservations stand, but I realize they are mostly a product of my own inexperience in thinking and practicing along similar lines as you do. To be frank, I never really thought of it all the way you do. Once I try to follow your reasoning, it begins to make sense (to me, at least).
Dejan,
I did tell before. It is worth reading the qualifications I add (I rarely post anything unqualified). I just may not have made it quite as obvious.
Well, my reasoning (low high order etc.) is based on what I have learned about "good sound", which is not always in line with what some call "science" and equally often goes against "folklore".
To me we find a need for NFB if the natural parameters of the output stage (linearity, damping factor) are unacceptable or if we have an output stage with a large GM step (that is class B or AB).
And for example, overbias increases THD, but lowers high order HD over "optimum bias" (which is hence not optimum to me) and hence I specify it. Optimum bias for example will have already so much higher order products that it swamps other interactions.
My key concern is the distortion product multiplication (or "distortion of distortion") in the forward path (it is related but not the same as "re-entrant distortion" which needs looped systems).
In the end the way I approach things are from the "The Devil is in the details" positions. I find if the details are well considered the big picture takes care of itself. Perhaps this is the influence of my time earning a living with big financial computer systems. Accountants tend to say: "Take care of the pennies, the pounds and millions take care of themselves!".
Ciao T
I did tell before. It is worth reading the qualifications I add (I rarely post anything unqualified). I just may not have made it quite as obvious.
Well, my reasoning (low high order etc.) is based on what I have learned about "good sound", which is not always in line with what some call "science" and equally often goes against "folklore".
To me we find a need for NFB if the natural parameters of the output stage (linearity, damping factor) are unacceptable or if we have an output stage with a large GM step (that is class B or AB).
And for example, overbias increases THD, but lowers high order HD over "optimum bias" (which is hence not optimum to me) and hence I specify it. Optimum bias for example will have already so much higher order products that it swamps other interactions.
My key concern is the distortion product multiplication (or "distortion of distortion") in the forward path (it is related but not the same as "re-entrant distortion" which needs looped systems).
In the end the way I approach things are from the "The Devil is in the details" positions. I find if the details are well considered the big picture takes care of itself. Perhaps this is the influence of my time earning a living with big financial computer systems. Accountants tend to say: "Take care of the pennies, the pounds and millions take care of themselves!".
Ciao T
I went back and looked over my notes , taken from the 5th row , it was 114db peak , avg din 84 db -87 db. The 120db peak i measured in the much smaller practice hall and not the Symphony hall ...
Well, all the above graphs are from reputable manufacturers that know to adjust their AB quiescent current. And still THD grows exponential below 1W.
And using ohm's law you measure power? Strange... I always thought that ohm's law does something else:
Grows below 1w? Well that's another place Spice is useless. My models just get lower until they fall below the noise floor.
I still would love to see a lot more details on this measurement system that claims to measure harmonics at or below the noise floor. I also will put some amps on the bench and try to measure to as low a power as I can. Been a bit busy trying to get some cement work done.
An AP is a technical analyser. I recall that modern semantic expert systems with sophisticated "elastic" pattern recognition (based on spectrum vs time, basically) can identify different instruments in a recording, playing different or identical melodies at the same time (and even chords can be properly decomposed and associated to the instruments which played them). There was a scientific publication I saw about this recently, alas I can't seem to locate it now...
Seperating instruments in a massed string arrangement will sure be a very tough job but by no means technically impossible.
That's easy, as Jan mentioned already. Do thousands of syncronized time domain averages (before FFT'ing it), that is, average the sample values at the same positions for all the data blocks recorded. 30dB suppression of random noise and any other non-correlated signal (hum, notably) can be achieved, see my little experiment here : http://www.diyaudio.com/forums/solid-state/109147-geddes-distortion-measurements-6.html#post1337413
Last edited:
Agree with KSTR. And if you are unfortunate enough to use a signal frequency that happens to be correlated with hum, just change it a bit. This technique will also reveal generator distortion independently of generator noise.
And the lower limit on the improvement from signal averaging will be set by 1/f noise, and any drifts or other shifts during the signal averaging. But this can get you way down there. The advantage to having good sync information from the test generator is huge, as opposed to post-processing with autocorrelation etc. And with good ADCs and lots of memory and powerful number-crunching one should be able to do awfully well. We should encourage JA to get something along these lines. It would be very helpful to show what's going on rather than just saying thus and such is buried in the noise.
And the lower limit on the improvement from signal averaging will be set by 1/f noise, and any drifts or other shifts during the signal averaging. But this can get you way down there. The advantage to having good sync information from the test generator is huge, as opposed to post-processing with autocorrelation etc. And with good ADCs and lots of memory and powerful number-crunching one should be able to do awfully well. We should encourage JA to get something along these lines. It would be very helpful to show what's going on rather than just saying thus and such is buried in the noise.
Indeed so.
But there are strories and stories ...
Nothing ever sells a story better than when the storyteller puts his money where his mouth is.
Thorsten said what he said, and then he produced a schematic using what he spoke about. In my view, that's the way to go about things, I remember being impressed by Otala's initial texts in the IEEE precisely because he claimed what he claimed, and backed it up with his little 25 Wpc amp schematic, thus undenialbly proving most of what he said. I must have heard at least 20 amps made along his guidelines, and they did sound way better than most in those days.
Hi,
I am currently using Spice (still Tina, but with good models now) to play with the effect of Iq on high order distortion.
With +/-60V at modest load and a 0.3K/W heatsink with allowing 30K heatrise in the heatsink I can allow 100W dissipation or 833mA in total (drivers and outputs), so if my drivers are 2SJ162/2SK1056 at 133mA I can run 233mA per output pair.
So far it looks like more bias is more better. Doing this compared to theoretical "optimum" bias appreciably reduces higher order HD. I'll have to decide if I implement signal dependent bias, or not.
Then again, for rated power (150W/8R = 0dBFP) I get 0.12% THD at 50KHz and 0.066% at 1KHz, H2 dominant.
At around 1.5W (-20dBFP) I get 0.04% THD/50KHz with H2 dominant and H3 nearly 16dB down at optimum bias (26mV/120mA).
And at -60dBFP (dB below full power - so 0.15mW) I get 0.008% THD/50KHz H2 dominant.
Using Mosfet output stage with BJT drivers around triples low order HD over Mos Driver and BJT Output BTW, unsurprisingly, with no benefit in high order HD for the same Iq.
Ciao T
I am currently using Spice (still Tina, but with good models now) to play with the effect of Iq on high order distortion.
With +/-60V at modest load and a 0.3K/W heatsink with allowing 30K heatrise in the heatsink I can allow 100W dissipation or 833mA in total (drivers and outputs), so if my drivers are 2SJ162/2SK1056 at 133mA I can run 233mA per output pair.
So far it looks like more bias is more better. Doing this compared to theoretical "optimum" bias appreciably reduces higher order HD. I'll have to decide if I implement signal dependent bias, or not.
Then again, for rated power (150W/8R = 0dBFP) I get 0.12% THD at 50KHz and 0.066% at 1KHz, H2 dominant.
At around 1.5W (-20dBFP) I get 0.04% THD/50KHz with H2 dominant and H3 nearly 16dB down at optimum bias (26mV/120mA).
And at -60dBFP (dB below full power - so 0.15mW) I get 0.008% THD/50KHz H2 dominant.
Using Mosfet output stage with BJT drivers around triples low order HD over Mos Driver and BJT Output BTW, unsurprisingly, with no benefit in high order HD for the same Iq.
Ciao T
Hi Jan,
Nice experiment. Sorry to have missed it.
I have a question however. The ripple you see after applying the 'brick wall' filter in my opinion is not caused by an increase in level of (some) harmonics. It is just the waveform after low pass filtering. Don't you agree that is the missing harmonics above 15 kHz that smoothen the waveform?
regards, Bert
Nice experiment. Sorry to have missed it.
I have a question however. The ripple you see after applying the 'brick wall' filter in my opinion is not caused by an increase in level of (some) harmonics. It is just the waveform after low pass filtering. Don't you agree that is the missing harmonics above 15 kHz that smoothen the waveform?
regards, Bert
Hi Bert,
What I have trouble to get my head around is the following.
The filter freq response is flat within 0.1dB or better, then drops monotonously to -60dB at 19kHz.
Looking at the transient response on the 1kHz square wave, we see that ripple. How should one interprete/reconcile the ripple in the time domain with the flatness in the freq domain? I probably should have paid more attention at school
jan
- Status
- Not open for further replies.