Some commonsense

[Paul R]

10 hours ago, pkane2001 said:

 

You're on the right track. If only there was a way to add zero samples to a signal without affecting it. If only it could be done.... 🤔

 

 

So? What’s interpolated about a recording at 24/192k?  That’s 192,000 honest information containing samples each and every second. Not 44,100 padded with zeros.

 

If you are referring to zero order hold, that does not mean a value of zero of course.  

 

Perhaps I am simply being dense here and not understanding what you are saying, and if so apologies. On the other paw, mathematically, perfect reconstruction of a band limited signal requires a doubly infinite set of values that can be interpolated into a continuous band limited function. As noted, we get practically perfect results from less than an infinite number of samples. Nothing anywhere says that those samples are zero padded. 

[Paul R]

35 minutes ago, mansr said:

All the recordings in my collection, even Götterdämmerung, are of finite length.

 

Which is very nice, but means there are things like phase errors in the reproduced signal. That of course, does not affect the frequency response at all, but is sure audible. 

 

I already pointed out out that the filters that can be used with high sample rate material can use a far less aggressive low-pass (anti aliasing) filter. One reason why DSD can typically sound better (in some ways) to most people than Redbook.

 

So again,  I can not rule out ultrasonic replication as being important, at least to some people. However, the benefits of high sample rates are more likely, as I said before, in the filters that a DAC uses.

 

If you want to argue that simple oversampling gives the same results, fine. That is a valid argument - I do not agree with it, but it is definitely something to consider. I would point out that different DACs, playing the same digital file, sent from the same equipment, over the same cables, can most certainly sound different. Filters are what I would point to as one reason why. 

[mansr]

2 minutes ago, Paul R said:

Which is very nice, but means there are things like phase errors in the reproduced signal. That of course, does not affect the frequency response at all, but is sure audible.

What the hell are you blithering about?

[pkane2001]

36 minutes ago, Paul R said:

Perhaps I am simply being dense here and not understanding what you are saying, and if so apologies. On the other paw, mathematically, perfect reconstruction of a band limited signal requires a doubly infinite set of values that can be interpolated into a continuous band limited function. As noted, we get practically perfect results from less than an infinite number of samples. Nothing anywhere says that those samples are zero padded. 

 

Zero  padding is a mathematical processing technique. Actual zero samples are not required in the signal, they are added during processing specifically to work with finite, non-periodic signals.

 

[Paul R]

1 hour ago, pkane2001 said:

 

Zero  padding is a mathematical processing technique. Actual zero samples are not required in the signal, they are added during processing specifically to work with finite, non-periodic signals.

 

Hi Paul - I thought all audio samples were periodic? 

[pkane2001]

5 minutes ago, Paul R said:

Hi Paul - I thought all audio samples were periodic? 

 

Not at the start or end of the track. Or when samples are processed in small chunks (as is needed for real-time convolution, for example). In other words, when the number of samples is not infinite

 

 

[Paul R]

35 minutes ago, pkane2001 said:

 

Not at the start or end of the track. Or when samples are processed in small chunks (as is needed for real-time convolution, for example). In other words, when the number of samples is not infinite

 

 

 

Eh? I am not following you there. While reconstruction from non-uniform sampling is quite possible, even a small sample set for convolution - of audio data - is periodic. In other words, the interval between samples is equal and consistent, so far as I know. 

 

 

[mansr]

7 minutes ago, Paul R said:

Eh? I am not following you there. While reconstruction from non-uniform sampling is quite possible, even a small sample set for convolution - of audio data - is periodic. In other words, the interval between samples is equal and consistent, so far as I know.

That is not what the word periodic means when describing a signal.

[pkane2001]

31 minutes ago, Paul R said:

 

Eh? I am not following you there. While reconstruction from non-uniform sampling is quite possible, even a small sample set for convolution - of audio data - is periodic. In other words, the interval between samples is equal and consistent, so far as I know. 

 

Paul, not sure what your math background is, but "common sense" is the wrong thing to use when discussing mathematical proofs and theorems. Math requires clear definitions with an exact understanding of terms.

[lucretius]

14 minutes ago, pkane2001 said:

 

Paul, not sure what your math background is, but "common sense" is the wrong thing to use when discussing mathematical proofs and theorems. Math requires clear definitions with an exact understanding of terms.

 

Yes. Common sense is good sense and sound judgment in practical matters, whereas math lacks empirical content.

[Paul R]

2 hours ago, pkane2001 said:

 

Paul, not sure what your math background is, but "common sense" is the wrong thing to use when discussing mathematical proofs and theorems. Math requires clear definitions with an exact understanding of terms.

 

Despite Mansr’s comment,  periodic is an exact and correct term here. The sampling frequency does not vary. As well, I specified what I meant much more precisely so there should have been little confusion. 

 

I was awarded my degree decades ago and like many things, perhaps some mathematical terms have been redefined.

 

I do not keep up with the field except for my own amusement. Such as working through Shannon’s proof just so I was sure I understood it. Also, studying and using HQPlayer and XXHighend is like a master class in practical engineering applications.  

 

And yes, I am giving people the benefit of  the doubt there, because I do not see zero value padding in digital files recorded here and not changed. It is possible I am not looking in the right places, or it is possible you are assuming something I missed. 

 

Precisely what do you mean then? In both a mathematical and “common” sense, I have no reason to expect to find zero padding consisting of samples with a value equal to zero in a 24/192k recording.  

 

By the way, almost all math is very much “common sense.”  As far as I can see, that applies to almost everything - including “what is the minimum sample rate here.”

 

[pkane2001]

1 hour ago, Paul R said:

By the way, almost all math is very much “common sense.”  As far as I can see, that applies to almost everything - including “what is the minimum sample rate here.”

 

If you could use common sense to come up with Shannon's Sampling theorem or the Fourier transform, I'd be very impressed!  Common sense only applies once you have a very solid background and understanding of the basic principles. Skimming a Wikipedia article is not sufficient.

 

1 hour ago, Paul R said:

And yes, I am giving people the benefit of  the doubt there, because I do not see zero value padding in digital files recorded here and not changed. It is possible I am not looking in the right places, or it is possible you are assuming something I missed. 

 

Zero padding occurs in processing when doing DFTs, such as convolutions, filtering, etc. It's internal to the algorithm inside the processing unit. It's needed to deal with finite sets of samples that do not represent an infinitely continuous function. I suggest some DSP-related reading if you want to understand this further. 

[mansr]

2 hours ago, Paul R said:

I have no reason to expect to find zero padding consisting of samples with a value equal to zero in a 24/192k recording.

Of course you wouldn't. Any sane person would stop the recording once the performance is over rather than include an infinite number of trailing zeros in the actual file.

[mansr]

2 hours ago, Paul R said:

By the way, almost all math is very much “common sense.”

Sure, if by "almost all" you mean algebra and basic calculus. After that, it gets a lot less intuitive.

[Ralf11]

one of the craziest things he's said yet

[Paul R]

23 minutes ago, pkane2001 said:

 

If you could use common sense to come up with Shannon's Sampling theorem or the Fourier transform, I'd be very impressed!  Common sense only applies once you have a very solid background and understanding of the basic principles. Skimming a Wikipedia article is not sufficient.

 

 

Zero padding occurs in processing when doing DFTs, such as convolutions, filtering, etc. It's internal to the algorithm inside the processing unit. It's needed to deal with finite sets of samples that do not represent an infinitely continuous function. I suggest some DSP-related reading if you want to understand this further. 

 

Wow - now you are accusing me of skimming Wikipedia. May I ask where you are getting your information from? 

 

A high quality SINC function does not need zero padding and works well with a short FFT.  Kinda basic DSP theory. At least it was when I studied it. 

But yes. If you are going to do longer convolutions, zero padding makes a handy storage space available. 

 

None of which will have zero value samples recorded in the digital file. You are talking about processing primarily for playback. In which case, as I said, the filters can be gentler and easier for high sample rate files. 

[pkane2001]

55 minutes ago, Paul R said:

 

Wow - now you are accusing me of skimming Wikipedia. May I ask where you are getting your information from? 

 

 

Not an accusation against you, just an example of what not to base "common sense" on.

[Ralf11]

Actually, I guess using a Lyapunov function to lasso an out of control, metastasizing nightmare on the complex plane before it blows up too much kind of IS common sense... 

 

 

[fas42]

2 hours ago, Paul R said:

None of which will have zero value samples recorded in the digital file. You are talking about processing primarily for playback. In which case, as I said, the filters can be gentler and easier for high sample rate files. 

 

What the filters do to the waveform should be separated from the electrical effects of circuitry doing the realtime filtering, at the time of playback - these are very distinct factors.

 

An excellent example of this was an ambitious rig I heard, whose CD player had switchable filtering - "Which filter do you prefer, as I go through them ...?" - "Umm, none of them are doing it for me, theres a slight unpleasantness to the sound which is a constant for each one; is it possible to completely switch off this filtering function?" - "Huhh, unusual request ... okay" - "Ahhh, that's it! The "musicality" is in the room now ... " 

[Paul R]

1 hour ago, pkane2001 said:

 

Not an accusation against you, just an example of what not to base "common sense" on.

 

Okay - apologies for jumping to conclusions then Paul. Guess I am a bit sensitive these days.

[Paul R]

51 minutes ago, Ralf11 said:

Actually, I guess using a Lyapunov function to lasso an out of control, metastasizing nightmare on the complex plane before it blows up too much kind of IS common sense... 

 

 

 

Packet queue delays are one of my professional nemesis, but I miss some of the humor in your post from ignorance. A very vague association to negative feedback systems is all that comes up. 

 

In English please, if possible with words of no more than three syllables? 

[Paul R]

3 hours ago, mansr said:

Sure, if by "almost all" you mean algebra and basic calculus. After that, it gets a lot less intuitive.

Admittedly, the Algebra does not really make sense until you grasp the Calculus. And then you encounter Linear Algebra, including that awesome matrix algebra that produces really cool spires in its graphs. 

 

But- I was taught that the set of  integers and their properties are pretty much at the root of all mathematical branches of study. Never really thought about it, just accepted that as an axiom. Of course, pretty much everything to do with integers is simply common sense. Sometimes incredibly obfuscated, but still...

 

. 

 

 

 

[Ralf11]

1 hour ago, Paul R said:

 

Packet queue delays are one of my professional nemesis, but I miss some of the humor in your post from ignorance. A very vague association to negative feedback systems is all that comes up. 

 

In English please, if possible with words of no more than three syllables? 

 

Complex Analysis - cowboy joke

[fas42]

1 hour ago, Paul R said:

But- I was taught that the set of  integers and their properties are pretty much at the root of all mathematical branches of study. Never really thought about it, just accepted that as an axiom. Of course, pretty much everything to do with integers is simply common sense. Sometimes incredibly obfuscated, but still...

 

.

 

Highly convoluted playing with maths may be necessary to understand, or distinguish what's happening between different examples of playback - but it's of zero use for actually evolving competent playback ...

 

So, rest easy on that front, .

[Le Concombre Masqué]

On 6/17/2019 at 3:27 PM, mansr said:

Any event occurring in the audible range (up to 20 kHz) is fully captured by a sample rate of 40 kHz or higher. If part of the event exceeds this bandwidth, that part is not audible. Put differently, a sample rate of 44.1 kHz is sufficient to fully capture the audible portion of any sound event.

Any event... how about "time-smearing" and minimum perceptible interaural time difference ?