Hey, Tourney - you were a recording engineer - have you got a well recorded cymbal strike "impulse" that you refer to, that we can check the spectrum it creates?
May I ask the honorable assembly, as a favor, to consider the following, completely disregarding their sympathies for their author. Thank you for this effort.
Let-us forget the Nyquist theories and stay at a practical level regarding digital, from an other angle.
Simple is beautiful.
We record musical signals of various forms by sampling them by point. These signals are limited in finesse by the performance of our microphones. The higher the number of points on the abscissa of the levels and the ordinate of time, the more the reproduction of the waveform will be faithful. Right ?
The same passionate debates, more friendly however, take place on photography forums, to determine the limits of the number of pixels of the digital captors required, the need for anti aliasing filters etc.
Including when zooming in a file (cropping).
The answer is simple, and the industry, that advocated 10 years ago that 12MPx was enough, is constantly increasing the number of pixels on professional cameras (>60Mpx).
The limit is easier to determine, here, because it seems easier for people to agree on visual perception than auditory: The useful limit is that after which an increase in the power of separation of the sensor brings no more improvement to the one, limited, of the lens (in constant progress).
A little more than twice this one, thank you Mr. Nyquist.
And anti aliasing filters are now useless. Winner play again, twice the bet.
In audio, good microphones go up to 40,000 Hz and super tweeters too.
On the level point of view, by the signal/noise ratio of the best mics preamps.
The acoustic noise is not to be considered, as it is part of what we want (or not) to record.
Everything else is a matter of personal perception.
Let-us forget the Nyquist theories and stay at a practical level regarding digital, from an other angle.
Simple is beautiful.
We record musical signals of various forms by sampling them by point. These signals are limited in finesse by the performance of our microphones. The higher the number of points on the abscissa of the levels and the ordinate of time, the more the reproduction of the waveform will be faithful. Right ?
The same passionate debates, more friendly however, take place on photography forums, to determine the limits of the number of pixels of the digital captors required, the need for anti aliasing filters etc.
Including when zooming in a file (cropping).
The answer is simple, and the industry, that advocated 10 years ago that 12MPx was enough, is constantly increasing the number of pixels on professional cameras (>60Mpx).
The limit is easier to determine, here, because it seems easier for people to agree on visual perception than auditory: The useful limit is that after which an increase in the power of separation of the sensor brings no more improvement to the one, limited, of the lens (in constant progress).
A little more than twice this one, thank you Mr. Nyquist.
And anti aliasing filters are now useless. Winner play again, twice the bet.
In audio, good microphones go up to 40,000 Hz and super tweeters too.
On the level point of view, by the signal/noise ratio of the best mics preamps.
The acoustic noise is not to be considered, as it is part of what we want (or not) to record.
Everything else is a matter of personal perception.
Last edited:
What would it take to make a 44.1Khz BW and x5-10 sampling rate?
Hardware-wise. adc/dac.
What can be done with existing parts? ...
DXD? DSD? Greater potential for higher accuracy?
THx-RNMarsh
Hardware-wise. adc/dac.
What can be done with existing parts? ...
DXD? DSD? Greater potential for higher accuracy?
THx-RNMarsh
Ultima,
Channeling Don Maclean are we?😀
We concur, a limited sample width makes perfect reproduction problematic, that is not new.
Post processing a data stream with say 16 or 32 deep sample math to calculate instantaneous value on the fly is better than a simple S/H, but not red book either.
Red book was not designed to perfectly reproduce instantaneous interchannel phase relationships at the 2 to 5 uSec level for music.
Unfortunately, humans can localize a vocal within a soundstage of multiple instruments and vocals.
Jn
Channeling Don Maclean are we?😀
We concur, a limited sample width makes perfect reproduction problematic, that is not new.
Post processing a data stream with say 16 or 32 deep sample math to calculate instantaneous value on the fly is better than a simple S/H, but not red book either.
Red book was not designed to perfectly reproduce instantaneous interchannel phase relationships at the 2 to 5 uSec level for music.
Unfortunately, humans can localize a vocal within a soundstage of multiple instruments and vocals.
Jn
Last edited:
How convenient, then you could propagate nonsense without any penalty.
How about forgetting the EM laws, then you could claim cables directionality, isn't it?
You tend to comment without actually reading the posts and understanding the content, rather you tend to pull out of context what you think will play well with the audience.
Unfortunate.
I hope for better of you in the future.
Jn
I am afraid you are not fitting in here. Partial understandings extrapolated to all conditions is more typical. 😉. !!!
Yes it is a bit much to do properly as that infinite bandwidth is hard to accomplish.
On the other end with some pieces of music running for many minutes to properly do a Fourier transform requires some very low frequencies, but not quite DC.
BTY I do not believe DC actually exists, as I am not immortal.
You did not, you barely admitted ("kinda" in #33062) that sampling at exactly 2x the sine frequency is violating Nyquist.
Nevertheless, you keep pushing the same zero crossing example, although you were several times told, by different people, that this example directly violates Nyquist.
And you keep propagating the BS that more than 2 samples per period are required to reconstruct a pure sine.
You seem to be unable to understand that a continuous sine wave, extending from -oo to +oo, is NOT the same as a time limited sine signal. The latter has an infinite frequency spectra, therefore two samples per period are not enough to reconstruct. You cannot exactly reconstruct such a time limited signal without taking an infinity of samples, or accepting some reconstruction accuracy compromise. The question is, where to put this accuracy limit - and here comes Audio, the existing body of knowledge is clearly showing that a brickwall filter at 20KHz *) is not audible and that all the discussion about pre-ringing in such a linear phase filter time response is pure BS **). Anyway, if you want to debate this, that's a different topic all together.
Your avoidance to make a clear point, lack of acknowledge when quoted relevant material ("What Nyquist doesn't say") and the attempts to shift the focus to other unrelated topics (like the "inter channel temporal information", and your avoidance tactic in specifying the relevance in the current context) speaks volumes about your willingness to admit you were caught in offside.
I can live without your penitence, but I need to remind you that "errare humanum est, sed perseverare diabolicum".
*) Either in the analog or digital domain, if that matters to you.
**) Nevertheless, the advances in DSP computing power allowed implementing such filters that are not necessary linear phase, therefore don't experience pre-ringing. Nobody claimed such a filter (and other types) is anywhere "better", but they are there to choose from in any modern audio (and not only) ADC/DAC.
Last edited:
44kHz bandwidth implies 88kHz or faster sampling.
5 x 88kHz = 440kHz, if that's what you meant?
DXD is only 352kHz. Digital eXtreme Definition - Wikipedia
Now i print arrows in my cables drums and i am a cable lifter designer ?
What the next, the inventor of a magical pyramid ?
Don't worry, this guy discredits himself with remarkable talent.
For my part, I do not read him any more.
Last edited:
A fine joke, however kinda private. Thank for the smile.
I plan to pass it, with your permission, on to my electricity supplier during the next blackout
Last edited:
No it isn't. This is from a side project, using only 480 point FFT's (10ms windows) you can resolve the phase drift of the two channels of a test LP and their difference to <0.1 degree.
The linear drift represents the phase relative to an exact 1kHz the wiggles are LP eccentricity. In the difference signal one can resolve the tiny change in tracking angle caused by the off center LP (45RPM in this case).
It is easy to show continuous delay of a signal by phase rotation (at arbitrary resolution) of its FFT and doing the inverse FFT.
EDIT - the data is 16/48
Attachments
Last edited:
I do, although lately is has been for laughs. I suspect at some point in time he will discontinue his obfuscations, diversions, and strawmen...however, I do not hold my breath. His run on posts, I find sometimes necessary to simply ignore and not comment on. Picking apart every sentence is too boring and unproductive. I like it when he posts technical stuff he has done, as opposed to the playing to the crowd antics of late where he views his standing with the crowd based on how much he can attack others.
Your parallel to optics is rather good.
I suspect that eventually, someone will do the immersion lithography thing with the sensors in situ and as part of a lens assembly. That will allow better lens measurement/fabrication, and will bump up pixel count just as it has taken semi's to the 7 nanometer regime (so far). I expect the nano center people are working on better bug juice to continue the scaling progress. I know they are working on transmissive coatings. What was really interesting is they have developed a quantum material that will not transmit infrared. I have no clue how that works.
A few years ago I had to help friends putting together a display of about 120 photos for a show. I had to learn photoshop very quickly, and it was actually fun seeing the algorithms I learned in the late 70's being used in the software of today.
Jn
Last edited:
BW = 44,1 => 88,2 ksps
OK - yes - your "sampling theorem" dictates not 2*BW to get to BW but 5 to 10x BW. So that would be 44,1*5 or 10.... 220 to 440 ksps.
Well you can get 4,36*BW for:
Roland RUBIX24 2-In/4-Out High-Resolution Interface: Amazon.co.uk: Musical Instruments
Sample size I believe is 24bits.
//
OK - yes - your "sampling theorem" dictates not 2*BW to get to BW but 5 to 10x BW. So that would be 44,1*5 or 10.... 220 to 440 ksps.
Well you can get 4,36*BW for:
Roland RUBIX24 2-In/4-Out High-Resolution Interface: Amazon.co.uk: Musical Instruments
Sample size I believe is 24bits.
//
Last edited:
- Status
- Not open for further replies.