[sdiy] CD sound quality evolution

Mike Bryant mbryant at futurehorizons.com
Tue Feb 16 12:21:28 CET 2021


We did a lot of tests on this in 2000 at Soundcraft to prove just how many bits we needed at the bottom end, so as to use the rest at the top end to give us an overload margin.  Everyone involved could hear at least 8.5 to 10dB below the noise level, and some about 14-15 dB down, which corresponds well with Brian's 18.5 bits.  This was done using the best at the time products from both AKM and Cirrus Logic, and there was no perceptible difference between the two so it was unlikely that people were noticing some distortion or other artefact and thinking it was the signal.

As for the test equipment, signals at greater than 20dB below the noise floor were easily visible in the noise on the Audio Precision, the gold standard for audio test equipment.

I'm afraid there's no associated AES paper on this work because company policy at the time was not to publish papers or patents, the company founder once saying "I was both surprised and delighted to be awarded an AES Fellowship, particularly because I've never presented any papers on my subject"


From: Synth-diy [mailto:synth-diy-bounces at synth-diy.org] On Behalf Of Randy Dawson
Sent: 16 February 2021 06:57
To: Brian Willoughby
Cc: synth-diy at synth-diy.org
Subject: Re: [sdiy] CD sound quality evolution

Thanks, Brian, for your great audio insight,  but I will stick with my original assertion that 16 bit is OK.
I offer a challenge you and AES to provide any thing about audio that is beyond this, and actually perceptible by even the Golden Ear customer.
I am a Senior member of IEEE, rdawson at ieee.org<mailto:rdawson at ieee.org>

From: Brian Willoughby <brianw at audiobanshee.com>
Sent: Monday, February 15, 2021 10:16 PM
To: Randy Dawson <rdawson16 at hotmail.com>
Cc: synth-diy at synth-diy.org <synth-diy at synth-diy.org>
Subject: Re: [sdiy] CD sound quality evolution

Human perception of sound does not work in plain black and white. Frequency response is not flat, sounds at one frequency can mask another; our brains are even so slow that sounds which occur later in time can actually mask sounds that occur earlier in time.

Bit depth is not a cliff where sounds completely disappear below -96 dBFS limit of 16-bit audio. It's way more complex than that. When comparing what's audible to the numbers, there's not a clear line between two distinct opposites.

On the one hand, the human brain can discern sounds well below whatever quantization depth is chosen. This is how 24-bit mastered CD actually sounds better than 16-bit mastering. The 16-bit noise is still all there - it's simply piled into frequency bands that our brains cannot detect quite so well.

On the other hand, the nature of the quantization itself determines whether it is audible. Correlated quantization noise is easily audible, and quite objectionable. Random dither almost disappears - not to test equipment, but to our perception - because we're constantly ignoring Gaussian noise sources in nature to discern quieter sounds that might be important.

AES experts have pointed out that the best audiophile listening setup can achieve only about 18.5 bits of signal to noise. I guess that puts it around 112 dB S/N.

So, yeah, depending upon the kinds of noise that will be picked up even with shielded cables, it is indeed still possible to hear better than 16-bit audio.

You'll never hear all 24 bits, though. The equipment will have more analog noise than quantization noise, although correlated quantization noise will still be audible despite the louder analog noise. Under the best conditions, though, the quantization noise can become inaudible and you're still left with less S/N than the 24-bit DAC handle theoretically produce. This is a blessing in disguise, because all 24-bit DAC chips cannot even hold a steady value due to the massive filter processing going on.

Brian Willoughby

On Feb 15, 2021, at 21:31, Randy Dawson wrote:
> You can't get lower than 16 bits in your home-lab environment with a 1V audio typical signal transport and shielded cables around the room you are working in.
> For 16 bits:
> 2 ^ 16 = 65536
> 1V / 65536 = .00001525878 V (LSB Voltage change)
> I challenge the Golden Ears to toggle this 16 bit LSB above; in their audio chain, from any source in any way, and claim it is audible, or measurable in any audio equipment.
> Randy
> rdawson at ieee.org<mailto:rdawson at ieee.org>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20210216/41658109/attachment.htm>

More information about the Synth-diy mailing list