PCM 768kHz! - 24bit - FREE TRACK from Carmen Gomes Inc.'s Ray! Album, Is this overkill or does it make sense?

[microstrip]

I wouldn't say its more complex to use - its certainly a lot easier. The technology and speed of the hardware has transformed what is now possible. I am setting up a orchestral recording with organ for Wayne Marshall next week, and we doing 7 channels at DXD 352.8 kHz 24 bit. An ordinary Macintosh Laptop that is 5 years old can easily accommodate such a bandwidth that would have been unimaginable 20 years ago.

I am only addressing the technical aspects, nothing else. Can yo tell us of all the technical details of the data flow from the microphone analog signal to the DXD?

 

[Jake Purches]

Only theoretically - I don't have access to the manufacturer's techniques. There are interfaces that include mic preamps that then digitise the signal and send it via USB or Thunderbolt like the Steinberg systems. Then there are others like Merging Horus that send the digital signal via CAT45. What is more common is what happens next when it enters a DAW - Digital Audio Workstation software, Protools, Reaper, etc. Usually the A to D converters spit out a 24 bit word length at what ever resolution you want - from 44.1 kHz to the more common 48 kHz, 88.2, 96, 176.4, 192, and now 352.8 and 384 kHz. Newer ones are going 32 bit. Not for more quality, but so there is no requirement to have a level control anymore. The 32 bit space is huge and captures anything a mic can through at it. (I am slightly dubious about this mind.. need to check it out). I usually record in 352.8 kHz. Why? because it decimates to 44.1 kHz evenly, carries the same audio benefits of 384 with almost 8% smaller file sizes and transfer requirements. I can explain more thoroughly the workings of a CD player though, because I learned that.

 

[oldvinyl]

This is nonsense. There are no stairsteps in digital. Rather, the sampling points are just that, points, and a perfect analog waveform connects them at the output of a D/A converter. At least, in any scenario where the sampling rate is at the minimum twice the highest frequency of an, emphasis, *bandwidth limited* signal. This, according to the Shannon-Nyquist theorem, which also holds for the Redbook CD format.

Thus, the reproduced waveforms will look the same in all three formats, left to right. No stairsteps with "errors".

My recollection from communications theory classes and EE work is that the Nyquist frequency only guarantees that the reproduced signal will not be aliased. The stair step is the instantaneous voltage at the DAC output. This is also instantaneously reproduced as the sum of the sine waves that represent the step (the Fourier series). Each step is a set of waveforms and harmonics. While the original frequency is reproduced - so are harmonics. The low pass output takes care of the higher frequency noise, but not the lower frequency. In other words, a 10 kHz output would have 10/20/30/40... kHz components in various magnitudes that sum to the "step" difference from the last sample. The upper order harmonics are filtered out. But a 5 kHz signal would have 10/15/20/15... kHz components in various magnitudes that sum to the "step" difference from the last sample. Some of these are not filtered out. In other words Nyquist predicts the frequency but not necessarily the correct shape of the reproduced signal.

As the bit depth decreases, the quantization error would then result in errors in the frequency of the signal being sampled - if the sampling overlaps with the zero transition. The samples could misinterpret the frequency and/or the magnitude since the various sine waves have a lot of overlap near zero.

Nyquist is great for the frequency - but not the shape of a signal. Acoustic music will have a lot of sine waves. Some electronic instruments (synthesizers or distortion generators) will have square or triangular waves (or something more akin to these). The Nyquist frequency of a since wave, square wave and triangle wave are the same. These require higher sampling rates to sort out the differences.

My understanding is that digital will introduce artifacts in the outputs that are artifacts of quantization sampling. How well they are dealt with seems to affect the reproduced sound.

 

[Al M.]

and anyway trying to get more than 21 bits dynamic range in the real world is a challenge anyway.

We definitely can agree on that one.

 

[microstrip]

My recollection from communications theory classes and EE work is that the Nyquist frequency only guarantees that the reproduced signal will not be aliased. The stair step is the instantaneous voltage at the DAC output. This is also instantaneously reproduced as the sum of the sine waves that represent the step (the Fourier series). Each step is a set of waveforms and harmonics. While the original frequency is reproduced - so are harmonics. The low pass output takes care of the higher frequency noise, but not the lower frequency. In other words, a 10 kHz output would have 10/20/30/40... kHz components in various magnitudes that sum to the "step" difference from the last sample. The upper order harmonics are filtered out. But a 5 kHz signal would have 10/15/20/15... kHz components in various magnitudes that sum to the "step" difference from the last sample. Some of these are not filtered out. In other words Nyquist predicts the frequency but not necessarily the correct shape of the reproduced signal.

As the bit depth decreases, the quantization error would then result in errors in the frequency of the signal being sampled - if the sampling overlaps with the zero transition. The samples could misinterpret the frequency and/or the magnitude since the various sine waves have a lot of overlap near zero.

Nyquist is great for the frequency - but not the shape of a signal. Acoustic music will have a lot of sine waves. Some electronic instruments (synthesizers or distortion generators) will have square or triangular waves (or something more akin to these). The Nyquist frequency of a since wave, square wave and triangle wave are the same. These require higher sampling rates to sort out the differences.

My understanding is that digital will introduce artifacts in the outputs that are artifacts of quantization sampling. How well they are dealt with seems to affect the reproduced sound.

My apologies, IMHO you manage to make something simple confusing and incorrect ... Nyquist tells us that we can rebuilt a signal from samples if we sample it at twice the maximum frequency, something that can be done today very easily, even without severe slope filters.

Quantization error is nowadays so small that it becomes negligible, considering the recording dynamic range and noise. Square and triangular waves are just a supeposition of sinusoidal waves. Nothing special here. Try listening to square waves or triangular synthesized waves with harmonics above 22 kHz suppressed and using all of them up to 96 kHz. IMHO you will not be able to distinguish them.

IMHO the artifacts of digital recording and playback are due to errors in the implementation, not to the theoretical limits. As digital hardware and software improves it is becoming more free of any artifacts. People also forget that mixing, equalization and mastering is carried in the digital domain. All these processes are mathematical operations performed in the digital data and can generate artifacts - it is why some digital recordings sound very poor. However, considering the sound quality of some modern recordings, it seems it is not a limit to digital performance any more.

 

[microstrip]

Only theoretically - I don't have access to the manufacturer's techniques. There are interfaces that include mic preamps that then digitise the signal and send it via USB or Thunderbolt like the Steinberg systems. Then there are others like Merging Horus that send the digital signal via CAT45. What is more common is what happens next when it enters a DAW - Digital Audio Workstation software, Protools, Reaper, etc. Usually the A to D converters spit out a 24 bit word length at what ever resolution you want - from 44.1 kHz to the more common 48 kHz, 88.2, 96, 176.4, 192, and now 352.8 and 384 kHz. Newer ones are going 32 bit. Not for more quality, but so there is no requirement to have a level control anymore. The 32 bit space is huge and captures anything a mic can through at it. (I am slightly dubious about this mind.. need to check it out). I usually record in 352.8 kHz. Why? because it decimates to 44.1 kHz evenly, carries the same audio benefits of 384 with almost 8% smaller file sizes and transfer requirements. I can explain more thoroughly the workings of a CD player though, because I learned that.

Exactly my point. We refer to DXD and PCM 768 and have a very limited idea of the process and sampling used in the microphones and the intermediate processes. We need such knowledge (and much more) to discuss technically these resolutions. And yes, the 32 bit space is a new world ...

I could easily read and interpret all the technical specs of my Studer A80, including the scape flutter measuring techniques. But when I read, for example, the Merging Technologies Horus and Hapi specifications I can't see any real reason why a 384k sampling rate should sound better than 192 kHz - the only difference is that the -1dB point goes from 75 to 64.7 kHz!
(surely naif readers like me would expect something like form 150 kHz to 75 kHz!)
See https://www.merging.com/products/interfaces/specifications

 

[Al M.]

Square and triangular waves are just a supeposition of sinusoidal waves. Nothing special here. Try listening to square waves or triangular synthesized waves with harmonics above 22 kHz suppressed and using all of them up to 96 kHz. IMHO you will not be able to distinguish them.

Yes, this old post (#22 in the thread) was an eye opener for me at the time.

 

[Jake Purches]

My recollection from communications theory classes and EE work is that the Nyquist frequency only guarantees that the reproduced signal will not be aliased. The stair step is the instantaneous voltage at the DAC output. This is also instantaneously reproduced as the sum of the sine waves that represent the step (the Fourier series). Each step is a set of waveforms and harmonics. While the original frequency is reproduced - so are harmonics. The low pass output takes care of the higher frequency noise, but not the lower frequency. In other words, a 10 kHz output would have 10/20/30/40... kHz components in various magnitudes that sum to the "step" difference from the last sample. The upper order harmonics are filtered out. But a 5 kHz signal would have 10/15/20/15... kHz components in various magnitudes that sum to the "step" difference from the last sample. Some of these are not filtered out. In other words Nyquist predicts the frequency but not necessarily the correct shape of the reproduced signal.

As the bit depth decreases, the quantization error would then result in errors in the frequency of the signal being sampled - if the sampling overlaps with the zero transition. The samples could misinterpret the frequency and/or the magnitude since the various sine waves have a lot of overlap near zero.

Nyquist is great for the frequency - but not the shape of a signal. Acoustic music will have a lot of sine waves. Some electronic instruments (synthesizers or distortion generators) will have square or triangular waves (or something more akin to these). The Nyquist frequency of a since wave, square wave and triangle wave are the same. These require higher sampling rates to sort out the differences.

My understanding is that digital will introduce artifacts in the outputs that are artifacts of quantization sampling. How well they are dealt with seems to affect the reproduced sound.

Thanks for articulating that nicely Oldvinyl.