High resolution audio. The science, or lack of...?

[jcbrum]

p.s. This is a similar argument to that used by some who say that analogue audio has infinite resolution whereas digital doesn't. It's not true. Analogue resolution is in practice limited by it's signal to noise ratio, and in most circumstances that's worse than the digital resolution available.

JC

 

[shadders]

An amplifier can resolve to a much lower resolution than -90dB.

In theory, a perfect mathematical model, of a perfect amplifier, in a perfect environment, can . . .

But in practice, in a real environment, a real amplifier will be limited by it's self noise and induced noise. That will be around the same level as -90 or probably worse. So the resolution would be lost in the noise, but more importantly you couldn't hear it anyway, because it's inaudible, because it's below your auditory threshold with human ears in a real environment.

JC

Hi,

Yes - in theory, if there was no amplifier noise then you would be able to see the clean signal, but practically, the signal is there with added noise.

My response was to fr0g who stated that this value is below the mute point. It is not below the mute point - there is no mute point in an amplifier - i think - please correct me if i am wrong.

Regards,

Shadders.

 

[jcbrum]

Ok - so you cannot hear at -90dB even if this is a difference error/distortion which equates to 0.003%.

At what value of S/N ratio (number of bits used) do you perceive (hear) the quantisation error, where dither removes this effect ?.

Is it 12bits, 14bits, 11bits ??.

In the example under discussion (fr0g's test), we are talking about the difference in perceived sound quality between 24bit wordlength and 16bit wordlength.

Frog has started with a 24bit file and re-sampled it to 16bit. This causes a quantisation error brought about by truncation (loss of 8 bits). This error is then corrected by applied dither, which restores the resolution but raises the noise floor. From the analysis seen it is likely that the dither was shaped so as to move the extra noise to a region above 17kHz where it's inaudible not only due to it's very low level, but also to a humanly insensitive frequency region.

So, in the application under discussion we are discussing the application of dither at the 16bit level.

JC

 

[shadders]

p.s. This is a similar argument to that used by some who say that analogue audio has infinite resolution whereas digital doesn't. It's not true. Analogue resolution is in practice limited by it's signal to noise ratio, and in most circumstances that's worse than the digital resolution available.

JC

Hi,

I disagree, in communications you can extract a signal which has a extremely low S/N.

An amplifier does have infinite resolution, but the system noise exceeds the lower resolution signals. If you analyse the amplifier output with the correct algorithm - you can determine the signal within the noise.

Regards,

Shadders.

 

[jcbrum]

p.s. This is a similar argument to that used by some who say that analogue audio has infinite resolution whereas digital doesn't. It's not true. Analogue resolution is in practice limited by it's signal to noise ratio, and in most circumstances that's worse than the digital resolution available.

JC

Hi,

I disagree, in communications you can extract a signal which has a extremely low S/N.

An amplifier does have infinite resolution, but the system noise exceeds the lower resolution signals. If you analyse the amplifier output with the correct algorithm - you can determine the signal within the noise.

Regards,

Shadders.

Maybe you can, but you can't hear it.

If something is audible you can almost certainly measure it.

But you can certainly measure things which you can't hear, because they're too quiet.

Is this level of technical discussion really worthwhile ?

JC

 

[shadders]

Ok - so you cannot hear at -90dB even if this is a difference error/distortion which equates to 0.003%.

At what value of S/N ratio (number of bits used) do you perceive (hear) the quantisation error, where dither removes this effect ?.

Is it 12bits, 14bits, 11bits ??.

In the example under discussion (fr0g's test), we are talking about the difference in perceived sound quality between 24bit wordlength and 16bit wordlength.

Frog has started with a 24bit file and re-sampled it to 16bit. This causes a quantisation error brought about by truncation (loss of 8 bits). This error is then corrected by applied dither, which restores the resolution but raises the noise floor. From the analysis seen it is likely that the dither was shaped so as to move the extra noise to a region above 17kHz where it's inaudible not only due to it's very low level, but also to a humanly insensitive frequency region.

So, in the application under discussion we are discussing the application of dither at the 16bit level.

JC

Hi,

You have not answered the question is presented based on your statement.

It doesn't matter what you call it, quantisation error, dither artifact, noise, distortion, difference error, whatever. If it's -90dB or distortion of 0.003%, whatever, I'm satisfied that it's below human perception for hearing in domestic audio replay, even though it might be measurable.

My questions :

Ok - so you cannot hear at -90dB even if this is a difference error/distortion which equates to 0.003%.

At what value of S/N ratio (number of bits used) do you perceive (hear) the quantisation error, where dither removes this effect ?.

Is it 12bits, 14bits, 11bits ??.

You have also stated :

From the analysis seen it is likely that the dither was shaped so as to move the extra noise to a region above 17kHz where it's inaudible not only due to it's very low level, but also to a humanly insensitive frequency region.

Are you stating that dither is noise shaping ?.

Are you stating that dither moves the quantisation noise to outside the audio band (above 17kHz as per fr0g's graph where this difference increased to -76dB) ?

Regards,

Shadders.

 

[jcbrum]

Yes - in theory, if there was no amplifier noise then you would be able to see the clean signal, but practically, the signal is there with added noise.

My response was to fr0g who stated that this value is below the mute point. It is not below the mute point - there is no mute point in an amplifier - i think - please correct me if i am wrong.

I think what fr0g meant by the 'mute' point was the level where the amplifier gain was reduced to zero.

The signal is still there, but the amplifier has no effect on it.

JC

 

[steve_1979]

...An amplifier does have infinite resolution, but the system noise exceeds the lower resolution signals. If you analyse the amplifier output with the correct algorithm - you can determine the signal within the noise.

Interesting, I didn't know that.

How does the algorithm work? How can it tell the difference between a music signal and the noise when the music signal is lower than the noise volume and it's not just random noise?

 

[fr0g]

Hi,

I disagree, in communications you can extract a signal which has a extremely low S/N.

An amplifier does have infinite resolution, but the system noise exceeds the lower resolution signals. If you analyse the amplifier output with the correct algorithm - you can determine the signal within the noise.

Regards,

Shadders.

But it isn't something you can "hear"

Anyway... As requested.

Spreadsheet 1

 

[jcbrum]

Are you stating that dither is noise shaping ?.

Are you stating that dither moves the quantisation noise to outside the audio band (above 17kHz as per fr0g's graph where this difference increased to -76dB) ?

In practice, yes, this is commonly done. It will be part of the algorithm employed by the software application used for the process.

I don't actually know what algorithm fr0g used, so I'm guessing, but it looks likely to me. I assume he used some competent proprietory application.

JC

 

[fr0g]

FYI, that list is from the first test.

This is for the recommended track

Spreadsheet 2

 

[Alec]

Fascinating.

 

[fr0g]

So, yes, Shadders, you are correct, it is not "zero". The -90dB seems to be what Audacity regards as zero, so I was mistaken in that claim

In any case it isn't audible. Not by me, not by you, not by anyone.

Here is Audacity's waveform plot for the difference between the HD track and downsampled version...

Essentially it is "nothing".

Zooming in 100% (which shows individual samples) also shows a completely flat waveform just like below.

 

[emperor's new clothes]

http://www.stereophile.com/content/jared-sacks-dsd-present-and-future

A man who knows his onions, starting from the seed

 

[shadders]

...An amplifier does have infinite resolution, but the system noise exceeds the lower resolution signals. If you analyse the amplifier output with the correct algorithm - you can determine the signal within the noise.

Interesting, I didn't know that.

How does the algorithm work? How can it tell the difference between a music signal and the noise when the music signal is lower than the noise volume and it's not just random noise?

Hi,

If you take the FFT of the signal - the spectrum will be present - this works for comms - with a specific modulation scheme, but will not be as evident for an audio signal.

Regards,

Shadders.

 

[Phileas]

http://www.stereophile.com/content/jared-sacks-dsd-present-and-future

A man who knows his onions, starting from the seed

"When you listen to PCM, you can literally hear it as a block of sound coming out of the speaker. That doesn't happen with DSD. "

:rofl:

 

[shadders]

Yes - in theory, if there was no amplifier noise then you would be able to see the clean signal, but practically, the signal is there with added noise.

My response was to fr0g who stated that this value is below the mute point. It is not below the mute point - there is no mute point in an amplifier - i think - please correct me if i am wrong.

I think what fr0g meant by the 'mute' point was the level where the amplifier gain was reduced to zero.

The signal is still there, but the amplifier has no effect on it.

JC

Hi,

The signal is there, the amplifier does process the signal, but the noise exceeds the level of the signal.

Regards,

Shadders.

 

[shadders]

FYI, that list is from the first test.

This is for the recommended track

Spreadsheet 2

Hi,

Thanks.

The values are varying across the audio band, rising from -119dB at 375Hz to -88dB at 18kHz.

Regards,

Shadders.

 

[busb]

...An amplifier does have infinite resolution, but the system noise exceeds the lower resolution signals. If you analyse the amplifier output with the correct algorithm - you can determine the signal within the noise.

Interesting, I didn't know that.

How does the.algorithm work? How does it tell the difference between a music signal and the noise when the signal volume is lower than the noise volume and it's not just random noise?

There's no great magic in signal recovery - average out (at the expense of time) the noise leaving the non-random signal, sharpen the bandwidth of the f range of interest will reduce noise power. Having said that, how usable the recovered signal is becomes a different matter - with audio, not very because it's non-periodic. As has been pointed out, analogue does not & cannot have infinite resolution. The whole point of digitising audio or other signals right at the front end is to fix the dynamic & f range & not degrade the signal during editing thus preserving the original. This is best done with greater bit depth & sampling rates than needed for the final output. The same applies to digital photography where three channels of 14 to 12bit data is combined & downconverted to three channels of 8bit data, giving the DR needed to produce a print or display on a screen. Those extra bits allow detail to be pulled out of shadow (LSBs) or highlights (MSBs). With visual media you have effectively 8bits per colour channel that can be "slid" up or down within those 14-12bits before hitting the "end-stops" of total black or white.

Another point with analogue is although the bandwidth is never infinite - you don't want a power amp with a 1MHz bandwidth because you risk parasitic oscillation that will be audible. Analogue tends to drop bandwidth fairly naturally (few dBs/octave). Brickwall filters are a necessary evil for A to D conversion - they distort phase relationships across the passband but digital audio works as well as it does by defining the bandwidth absolutely. By defining the bandwidth, you also directly define the slew-rate (rise time). The beauty of the N-S theorem is its elegance. I encourage anyone to understand the basics, if not the maths!

 

[jcbrum]

The values are varying across the audio band, rising from -119dB at 375Hz to -88dB at 18kHz.

I've forgotten which test result is which, but that looks like noise shaping to me.

Even if you turn your HIfi amplifier volume up to very loud indeed, and apply say +30dB of gain, it's still worst case nearly -60dB, which is still inaudible, especially with loud music playing.

It could be due to some response curve factor or other, but it's way below audible levels.

JC

 

[shadders]

So, yes, Shadders, you are correct, it is not "zero". The -90dB seems to be what Audacity regards as zero, so I was mistaken in that claim

In any case it isn't audible. Not by me, not by you, not by anyone.

Here is Audacity's waveform plot for the difference between the HD track and downsampled version...

Essentially it is "nothing".

Zooming in 100% (which shows individual samples) also shows a completely flat waveform just like below.

Hi,

Your table results indicate that there is something - a very small difference, which increases to -88dB whilst still in the stated audio range.

There has been statements by others that dither is important, which operates at the lowest end of the S/N range for a 16bit word depth, and it "apparently" impinges on the higher bits - and no evidence provided on this.

You have assumed that -90dB difference is not audible because you believe that a -90dB absolute signal is not audible. A -90dB signal may not be audible, but this is not evidence that the -90dB difference is not audible.

Regards,

Shadders.

 

[shadders]

As has been pointed out, analogue does not & cannot have infinite resolution.

Hi,

Can you explain why this is not the case ?. Thanks.

Regards,

Shadders.

 

[shadders]

The values are varying across the audio band, rising from -119dB at 375Hz to -88dB at 18kHz.

I've forgotten which test result is which, but that looks like noise shaping to me.

Even if you turn your HIfi amplifier volume up to very loud indeed, and apply say +30dB of gain, it's still worst case nearly -60dB, which is still inaudible, especially with loud music playing.

It could be due to some response curve factor or other, but it's way below audible levels.

JC

Hi,

I think this is the difference between the 96kHz and downsampled 44.1kHz - the 96kHz has extended bandwidth.

I was asking regarding whether you were stating dither is noise shaping. Thanks

Regards,

Shadders.

 

[andyjm]

As has been pointed out, analogue does not & cannot have infinite resolution.

Hi,

Can you explain why this is not the case ?. Thanks.

Regards,

Shadders.

All real world analogue systems have noise superimposed on the wanted signal. Suppose the wanted signal changes by a very small amount, but this small amount is less than the noise amplitude - it will be impossible to tell whether the signal has changed because of an intended change, or just because of noise. So the 'resolution' of an analogue system is a function of the noise inherent in the system. Hence why SNR (signal to noise ratio) has meaning for an analogue system but can also be derived from the resoultion of a digital system.

 

[altruistic.lemon]

http://www.stereophile.com/content/jared-sacks-dsd-present-and-future

A man who knows his onions, starting from the seed

Probably more interesting for the so called "experts" on this and the other site is this: http://www.audiostream.com/content/qa-andreas-koch . Since the guy was involved in the creation of SACD and works with DSD I presume he knows less than the ratpack