The word I used was transients.
As an example, I made reference to cymbals for an unique reason. The people not used to the real thing are not so sensitive to the losses of attacks of an instrument because they cannot compare. With cymbals, it is easy to feel the loss of overall level and presence.
But, in fact, the instrument where it is the most obvious is the piano.
Before the advent of digital, we were limited in dynamics, and these attacks were planned by the magnetic tape, if we wanted a sufficient overall level. I used a double trace oscilloscope (Before / After) to judge this. But the distortion of the magnetic tape at high levels somehow compensated for this loss, giving a feeling of volume.
In digital, if we wanted the attacks of the piano to pass correctly, the meter hardly moved (not enough average level).
To remedy this, many sound engineers used a limiter to plan them: Most of the recorded pianos are castrated.
The arrival of 24-96 remedied this. Both by the add of the dynamic and increasing the number of sampled points during very steep signals. (more luck to catch a peak in time)
It is not without reason that the 0dB was set between -12 and -18db, depending on the machine, in digital, and the resolution increased since the Red book.
As we are talking about audio, I'm not interested by controversies about Nyquist bode plots (we would fill a library with literature on the subject), but the practical results on our listening experience.
Last thing. I'm old. My frequency range has been considerably limited, if I test my ears with sinusoidal signals. Despite this, I'm still very sensible to the limitation of instant dynamic. Our ears don't work as some people imagine. look at how our vibrating lashes are implanted.
YouTube
YouTube (start at 1'30)
Last edited:
Mark,
I recommended paper and pencil to provide clearer visual. Excel would be just as good.
You are bogging down in pencil line and dot size? Really?
If you have excel, try what I said. Sample at 180 degree intervals, and move the offset. Watch the results.
Oh, don't worry about my work. In a single day, I was called to work transient analysis on PID systems, harmonic analysis of steady state waveform ripple, and get a laser interferometer vibration system to work, it samples for two minutes and displays ground motion (energy) at the 300 picometer level (well, after I fixed it). During a scan, we are not able to speak as the 15 ton granite block picks up our voice and trashes the measurement.
So I do indeed understand the strengths of huge windowed fft's as well as transient analysis. I fear many here are trying to fit me into a box labeled "bad, idiotic, stupid, arrogant, young engineer".
Well, I got news for you. I'm no longer young.🙁
Jn
I recommended paper and pencil to provide clearer visual. Excel would be just as good.
You are bogging down in pencil line and dot size? Really?
If you have excel, try what I said. Sample at 180 degree intervals, and move the offset. Watch the results.
Oh, don't worry about my work. In a single day, I was called to work transient analysis on PID systems, harmonic analysis of steady state waveform ripple, and get a laser interferometer vibration system to work, it samples for two minutes and displays ground motion (energy) at the 300 picometer level (well, after I fixed it). During a scan, we are not able to speak as the 15 ton granite block picks up our voice and trashes the measurement.
So I do indeed understand the strengths of huge windowed fft's as well as transient analysis. I fear many here are trying to fit me into a box labeled "bad, idiotic, stupid, arrogant, young engineer".
Well, I got news for you. I'm no longer young.🙁
Jn
Last edited:
Saw it, thanks.
I'll try to put up a spreadsheet in the next week or so detailing what I speak of. Things have been kinda hectic and I sometimes miss lunch as a result.
Jn
Edit: TT, I have never found any research correlating the aging induced hf sensitivity we all go through and our ability to localize. I've no idea if they track, or if localization degrades at a slower rate.
Last edited:
I already know you get a DC offset depending on phase. That's why sampling at 180 degrees is a violation of Nyquist.
if you go to 179.99 then the DC offset is gone, you can no longer plot a DC line through the points. Nor can you plot any other sine wave other than the sine wave you started with, including frequency, amplitude, and phase. You just have to carry out the calculations to enough decimal points in order to prove that since you started out with a case at 179.99 degrees that requires very exact calculations.
Last edited:
Well, that means in your own words that we were only wrong with 1 out of 5 points, right ??
Hans
Reading the last few pages of posts it seems to me much of the confusion and "talking past" is due to personal biases, as Mark mentions, including a form of blinkered thinking (?), presumptions of people's motivations rather than focusing on the posts, it's all good fun though, I don't believe those who say they aren't having a good time 😉
It was tossed up there as self depricating humor, a way of lightening the discussion.
It is not possible for me to tell if you got the humor or not.
If yes, our languages divide. If no, you have issues. Luckily, not my worry.
Jn
People get frustrated not knowing how to explain something that is rather nonintuitive to grok.
It may help to try the drawing experiment I mentioned. Even for the case you talk about when very close to Nyquist in frequency. If you expand the vertical scale of the graph where a set of sample points is located, with enough magnification of that area it should be evident that trying to draw a sine wave other than the original sine wave through a few points just doesn't work.
All the information needed to reproduce the wave later in time will have been captured. No need to sample for a long time in that case (in theory, of course).
If one were inclined to be critical of recent discussion the word 'stubborn' might come up, but not 'rigid.' 🙂
Last edited:
So, just by this single sentence, it declares that you don't understand the most basic in signal theory.
Sorry - it disqualifies you from giving anyone any advice within audio.
//
Sigh, verbage..I long for a pencil...😉
Not DC offset, temporal.
Plot a sine. Put a dot every 180 degrees starting at zero degrees. (2x).
Every point is zero amplitude. There are an infinite number of sines that those samples represent, as all you capture is zero crossing, no amplitude.
Now plot a twice amplitude same frequency sine but offset 45 degrees. If you look carefully, note the waveforms cross each other every 180 degrees. Sampling at those intersections cannot tell you which waveform it is, it is ambiguous. Insufficient data. Indeed, another infinite number of sines of various amplitudes and phase will fit those points.
As sampling rate increases, ambiguity decreases, dependent on window size.
Jn
Last edited:
Last I knew virtually all the audio D/A converters used a linear approximation (steps filtered) to reproduce each digital sample. In the correct application it should be the sinc function.
"The normalized sinc function is the Fourier transform of the rectangular function with no scaling. It is used in the concept of reconstructing a continuous bandlimited signal from uniformly spaced samples of that signal." (Wiki)
Then Harry Nyquist got lots of the credit even though the "Russians" really did the first work. Nyquist was working on data transfer and for his work the sampling frequency could be just double the highest data rate. However folks like Shannon did require the sampling frequency to be above double the signal frequency for other forms of signals.such as sine waves.
(I forget the Russians name he was at least 20 years ahead of the rest, but did not have the distribution of his paper advantage the Bell Labs guys did.
"The normalized sinc function is the Fourier transform of the rectangular function with no scaling. It is used in the concept of reconstructing a continuous bandlimited signal from uniformly spaced samples of that signal." (Wiki)
Then Harry Nyquist got lots of the credit even though the "Russians" really did the first work. Nyquist was working on data transfer and for his work the sampling frequency could be just double the highest data rate. However folks like Shannon did require the sampling frequency to be above double the signal frequency for other forms of signals.such as sine waves.
(I forget the Russians name he was at least 20 years ahead of the rest, but did not have the distribution of his paper advantage the Bell Labs guys did.
Last edited:
No problem, what you are listening to is BW limited anyway. All that's required for what you describe as instant dynamic is to have all your ducks in a row
His name was Peekup Andropov. He was the chauffeur for the car talk guys..😀
Have a happy new year Ed.
Jn
Have a happy new year Ed.
Jn
Last edited:
Sampling at zero crossings means sampling at exactly twice the frequency. This violates Nyquist, since the theorem says "greater" not "greater or equal".
Next point?
Yes, but you are speaking of ambiguity in a human mind trying to make sense of what original waveform looked like.
If data points are taken, then reconstruction is used to fill in the only possible solution to a curve connecting the dots, then it turns into a form a human mind can make sense of.
Absent reconstruction of a curve connecting the dots/points, a human mind needs many closely spaced points to 'see' a unique waveform. We can agree on that.
Its a limitation of the human mind not to be able to auto-visualize a unique curve (the only possible curve) that can connect the dots without violating Nyquist.
Last edited:
- Status
- Not open for further replies.