Thank you for the observation, I wish more folks would pay attention to these occasional comments that reinforce how subjective the audio experience can be.
The key is to assess the audible importance of the errors, that is where we have some disagreement. As to FFT limits, the FFT is a mathematical process and its limitations are are easily quantified. They are often exploited to obfuscate the discussion at hand. Most folks that dwell on the limitations of the underlying mathematical processes are grasping at straws because they simply can not deal with having to understand what is going on.
John, did you or someone else test the ceramic C0g capacitors which are supposed to be accurate ?
In general which caps exhibit the lowest (non-linear) distortions ? I'm talking about all types of caps. not just the big values ones (like the ones used for filtering or audio signals coupling).
I suggest an innocent misinterpretation of data from primitive tools in this case. A linear time invariant system that shows second harmonic distortion from a synthesized waveform but not sine waves is an extraordinary result. The traditional DA model has no distortion creating mechanisms. Capacitors with a voltage coefficient of capacitance are another matter, to be rigorous these effects must be separated. It is not fair to present results without taking the care to persure all avenues of well known effects.
In this case we were looking for the effects of DA that might be added by INTERSTAGE CAPACITORS. Now, what sort of value would that be? 1pf? Not at audio, 100pf? Maybe, from a condenser microphone, and over 100 Meg Ohm load. 1000pf, not very practical. 0.01uf possible, but again not likely. Guess what, we are just about out of range of the COG or NPO caps. 0.1uf? Here is a value that CAN be used for both coupling (hi Z), power supply bypassing, or EQ. And guess what? We measured a couple of those, and got results.
However, it was easiest and the most realistic to measure 1- 10uf or caps that could be representative for AC coupling between stages, or model the RC feedback combination so popular at the time, before servos were used as often, and by many amateurs today.
Now, why did we chose a 20ms pulse? We could do the same test with 100ms or 1ms pulses, AND we did, but IF we went too long in the pulse, it was difficult to see what was going on, because the OFF TIME was so long. You had to use a storage scope to do it right, and I did not have one, at the moment. What about faster pulses? It would work, AND you could also see the inductive component of the cap, but it did not give as much error as 20ms, and we were trying to show: REASONABLE WORST CASE
However, we did compare the SAME CAP at 450 ohms load and 50K, or a difference of about 100+, and we got an error difference of maybe 63, or so, and this shows that loading or increasing cap value, let us say 10uf to 1000uf would probably do the same thing, but it is not unmeasurable, and still about 100 times above the RESIDUAL of the test itself,with the same set-up, testing two nearly identical highest quality caps.
However, it was easiest and the most realistic to measure 1- 10uf or caps that could be representative for AC coupling between stages, or model the RC feedback combination so popular at the time, before servos were used as often, and by many amateurs today.
Now, why did we chose a 20ms pulse? We could do the same test with 100ms or 1ms pulses, AND we did, but IF we went too long in the pulse, it was difficult to see what was going on, because the OFF TIME was so long. You had to use a storage scope to do it right, and I did not have one, at the moment. What about faster pulses? It would work, AND you could also see the inductive component of the cap, but it did not give as much error as 20ms, and we were trying to show: REASONABLE WORST CASE
However, we did compare the SAME CAP at 450 ohms load and 50K, or a difference of about 100+, and we got an error difference of maybe 63, or so, and this shows that loading or increasing cap value, let us say 10uf to 1000uf would probably do the same thing, but it is not unmeasurable, and still about 100 times above the RESIDUAL of the test itself,with the same set-up, testing two nearly identical highest quality caps.
Let me correct the place of the invited talk -- DSP^x conference in san jose and a famous man - Scott Wurcer - was the next speaker after me. Note that the inquiry into cap sonics, originally, began because people had replaced caps and said notice differences.... the difference was described to me. I found the model of DA to be closest to what was described as the audible effects. A thickening of the sound for high DA caps as well as obscuring 'inner' detail et al. These people knew nothing about electronics... just keen listeners. I got curious.
Other than this -- there is no new claims of discovery about the science then or now. What is new is how linear distortion can sound non-linear and that is the connection I was disussing amoungst other issues over looked and assumed at the DSP^x talk. And, the reason for the strange way of putting it here. The test I used was to attempt to connect its results with what was actually described as hearing. The test used pulses more like music transients and that is why it is a better test and its correlation was excellent.... unlike sine wave thd results which leaves one to draw a blank. It was during the Q&A period that F. Toole asked about doing an FFT on the spurious DA signal (which is just another view of things but a better match to what we hear). I told him I did do such tests etc and the data was second harmonic mostly.
I can not cover the past all over again... as someone said -- time is running out. All was discussed in detail with John C. and W.Jung and others at the time and a long time before the invitation. Check the Proceedings (1995) and then get back to me. I am adding this in here to help make a more complete picture and the history of the capacitor issue. Thx - Richard (Dick) Marsh
Last edited:
Well, ever onward! IF people think that high value ceramics or electrolytic caps are A-ok for their audio projects, then go to it. Yet, it is not impractical to use better parts, or no coupling caps at all. All you have to do is to add servos. Are servos a perfect solution? No, it is possible that they can contribute something to the sound IF they are made with distinctly non-linear components, and their bandwidth (operating range) comes too close to the audio bandwidth. It is best to never use a servo for a high pass filter. I have tried it, and failed. You probably will too. This is because the servo is WORKING along with the audio and adding to it, rather than JUST CONTROLLING the DC offset. If you ever have a servo and you can see AC rapidly changing across it, you are too high and you have to lower your time constant (RC) that is usually 1 meg and 1uF making a RC time constant of 1. This seems to work almost always. Don't try to compromise, it is fairly easy to do, at a reasonable price and in a reasonable space. I have found Mylar caps to work OK, but don't use anything much cheaper.
Now, some here will NOT use servos, and they may be on to something. I have seen DC coupled power amps and preamps made with very close matching over temperature, and they do NOT need a servo, but at what price? A very high price in time and effort, and that is why I use servos in almost all my designs, including the CTC Blowtorch. Works for me!
Now, some here will NOT use servos, and they may be on to something. I have seen DC coupled power amps and preamps made with very close matching over temperature, and they do NOT need a servo, but at what price? A very high price in time and effort, and that is why I use servos in almost all my designs, including the CTC Blowtorch. Works for me!
"A linear time invariant system that shows second harmonic distortion from a synthesized waveform but not sine waves is an extraordinary result."
The AP shows no distortion from a suitably sized electrolytic cap. That said, I think some caps are very bad news and I wrote a few comments about my experiences in my e-Amp article - e.g. singing ceramics on one of the prototypes.
If you follow the good advice about sizing electrolytics correctly (from Cyril Bateman), avoid using ceramics (other than NPO/COG) in audio, steer clear of polyester and mylar and one or two others (from Dick Marsh and Walt Jung's article) you should be ok.
Hell, I just realized that doesn't leave much choice ;-)
The AP shows no distortion from a suitably sized electrolytic cap. That said, I think some caps are very bad news and I wrote a few comments about my experiences in my e-Amp article - e.g. singing ceramics on one of the prototypes.
If you follow the good advice about sizing electrolytics correctly (from Cyril Bateman), avoid using ceramics (other than NPO/COG) in audio, steer clear of polyester and mylar and one or two others (from Dick Marsh and Walt Jung's article) you should be ok.
Hell, I just realized that doesn't leave much choice ;-)
Those who measure harmonics ONLY will not find much, except in tantalum and ceramic.
DA does NOT measure like distortion, but it is a deviation from the IDEAL waveform that is normally expected from simple models. However, apparently virtually every electrolytic cap has SOME distortion, if you look hard enough, over a broad range of frequencies.
DA does NOT measure like distortion, but it is a deviation from the IDEAL waveform that is normally expected from simple models. However, apparently virtually every electrolytic cap has SOME distortion, if you look hard enough, over a broad range of frequencies.
Maybe the question as to the effect electrolytics have on sound quality should be posed differently, if we cannot measure distortion through them.
What audible changes are possible that might alter the perceived sound quality that would not be detectable using the usual THD test?
Let's exclude frequency response on the assumption that the cap value is chosen such that these effects lie well beyond hearing limits, say 3 octaves or more.
What audible changes are possible that might alter the perceived sound quality that would not be detectable using the usual THD test?
Let's exclude frequency response on the assumption that the cap value is chosen such that these effects lie well beyond hearing limits, say 3 octaves or more.
It is good that DA does not measure like distortion, because it is not distortion. Which "ideal" waveform? The one assuming that all capacitors are just that: perfect capacitors? Who "expects" that?john curl said:
If DA really can be heard, then that means that we are sensitive to very tiny deviations in frequency response. If DA apparently produces second harmonic from any waveform then there is a problem with either the measurements or their interpretation as DA cannot do this. Far more likely that DA is a marker for some other dielectric problem. Can we look for that, and stop all this nonsense about DA itself?
Without specific reference to DA, my own (non-scientific) experiments have led me to conclude that very small, but broadband, frequency and/or timing (phase and impulse) responses are indeed audible. It seeems as though the information correlation between the channels is instrumental in revealing such very small response variations. Okay, I'm going to open myself to ridicule, because likely few will believe the following anecdotal report. I could hardly believe it myself. I recently was experimenting with electrically mixing minute amounts of crosstalk (don't ask why
) between the stereo channels. By minute, I mean, a ratio of crossfeed at around -110dB. Not only could I hear an effect at that level, but the effect was was immediately obvious. What was more disconcerting is that I could also hear small changes in the crossfeed around that -110dB level.Because of that experience, I've come to suspect that, perhaps, the differences some of us hear in cables and interconnects, and various 'voodoo' tweaks, stem from minute alterations in system frequency and/or timing response.
Last edited:
- Status
- Not open for further replies.