Dithering is a Mathematical Process - NOT a psychoacoustic process.

[xr100]

Yes the end.

Same answer: I don't know what happened, as I am not the one who did the recording/editing/mastering, and don't want to blame anybody as the recordings sound fine to me.

What year was the CD released?

 

[bennetng]

What year was the CD released?

1991

 

[restorer-john]

Jerry Lettvin at MIT used that to demonstrate the issue a long time ago. I have a copy of SONY's first CD player test CD from 1982 or so, everything on it is undithered.

Is that a YEDs test disc or an early sample containing music/tones? (Interested which one you have) Emphasis was always marked on a per track basis for the test discs, but dithering was pretty much used on all recorded musical content demonstrations long prior to CD. The digitally generated signals of course had no dithering unless specifically annotated.

 

[scott wurcer]

Is that a YEDs test disc or an early sample containing music/tones? (Interested which one you have) Emphasis was always marked on a per track basis for the test discs, but dithering was pretty much used on all recorded musical content demonstrations long prior to CD. The digitally generated signals of course had no dithering unless specifically annotated.

I'd have to dig it out IIRC it was called Super Audio Check CD with a large booklet all in Japanese and not one of the YEDs series one of which I have somewhere. It was purchased at the Akihabara along with some of the first CD players. There was a full set of test tracks along with music samples, the CD started with a steam train starting right and leaving left just like some old LP demo disks. IIRC one of my CD players could not handle the 99 tracks.

 

[xr100]

1991

Ah, that may well explain it. Not sure of exactly what options were available (early CD masters were to U-Matic?) but presumably limited by the hardware available or used then. Recording should have an amply high noise floor anyway so not a catastrophe if not dithered at the final stage.

BTW, one way of doing the fade out then was to just to use the "AUTO FADE" facility on an SSL desk. In which case, or similar, quantization problems may well have been from the stereo mixdown, e.g. to DAT, given the "DDD" SPARS code.

Also, the SSL "AUTO FADE" would be recognised by anyone familiar with recorded music from the late 20th Century. It's a different curve to the standard ones in digital audio editors. (The Waves SSL bus compressor plug-in includes it.)

 

[xr100]

Is that a YEDs test disc or an early sample containing music/tones? (Interested which one you have) Emphasis was always marked on a per track basis for the test discs, but dithering was pretty much used on all recorded musical content demonstrations long prior to CD.

Don't think I ever came across a CD that used the pre-emphasis flag!

The digitally generated signals of course had no dithering unless specifically annotated.

A columnist in one of the UK hi-fi magazines loved to retell the story of how he once measured an undithered test tone at -90dBFS (or similar) on one of those test CDs through an early CD player and measured an incredibly high level of THD. IIRC, this was then used as justification for the "superiority" of analogue and vinyl.

He might just have well have measured tape without bias...

 

[restorer-john]

Don't think I ever came across a CD that used the pre-emphasis flag!

There's a lot out there, especially the early releases. IIRC, pretty much all the first 50 (CSR Japan) have pre-emphasis.

I'll have to dig out some of the very early ones and confirm. Years ago, I had a latch/buffer from the pre-emphasis flag out to an LED on one of my project players. Also there was a project on the early Philips chipsets to show error correction and interpolation via front panel LEDs so "we knew what was going on".

IIRC one of my CD players could not handle the 99 tracks.

One of my 99 track test discs actually has 100. A hidden track 0. Most machines won't see it at all. You have to have an early 1st gen that allows the manual control of the laser head (The Akai/Kyocera will do it) and you reverse it back (REW) before track 1 after it has started playing. The Kyocera 1st gen, doesn't really read the TOC and load the track list, so you can move the laser sled back and forth like a turntable arm. I actually like that concept, although the track search times are a little slower as it has to look for the track markers, not the running/elapsed time/frames like most machines.

A number of regular release CDs have hidden tracks buried in track 0. Maxwell's Embrya from 20 years ago was one I can think of. He always did something tricky with his albums. Urban Hang Suite had a massive 10 minute gap in the last track before a "secret" track played and one of his other CDs had an extra track which was "missing" in the TOC but listed on the album and was out of order.

 

[j_j]

Don't think I ever came across a CD that used the pre-emphasis flag!



A columnist in one of the UK hi-fi magazines loved to retell the story of how he once measured an undithered test tone at -90dBFS (or similar) on one of those test CDs through an early CD player and measured an incredibly high level of THD. IIRC, this was then used as justification for the "superiority" of analogue and vinyl.

He might just have well have measured tape without bias...

Or he may have simply seen all of the quantization noise, and not had any idea of what the quantization noise spectrum is. And obviously he never measured a recorded analog signal at -90dB/ref either.

 

[DonH56]

Don't think I ever came across a CD that used the pre-emphasis flag!

A number of early CDs used it before everyone realized how little it was needed and/or the market figured that out. Emphasis was a given for tape and vinyl so natural to include even if not really needed.

A columnist in one of the UK hi-fi magazines loved to retell the story of how he once measured an undithered test tone at -90dBFS (or similar) on one of those test CDs through an early CD player and measured an incredibly high level of THD. IIRC, this was then used as justification for the "superiority" of analogue and vinyl.

He might just have well have measured tape without bias...

Agreed. Did he measure a -90 dB signal on tape or LP record for comparison? As unworthy as looking at the unfiltered output of a DAC and claiming we can all here the "stairsteps" making the sound harsh...

 

[xr100]

Or he may have simply seen all of the quantization noise, and not had any idea of what the quantization noise spectrum is. And obviously he never measured a recorded analog signal at -90dB/ref either.

Did he measure a -90 dB signal on tape or LP record for comparison?

He would certainly have measured compact cassette and vinyl many times, although probably not at -90dBr. He was also aware a dithered test tone at -90dBFS would yield a very different measurement. I suspect it was a combination of some level of ignorance (i.e. not understanding that the system requires dither to work properly, aka "linearised," rather than it being a "cheat") combined with the "need" to come up with articles that were thought to be a "good read" for the target audience, i.e. ones that bashed digital and extolled the virtues (!) of vinyl.

To give some credit, the same author had talked about "hi-res" audio for years to overcome the "limitations" of CD-DA but, when it arrived in the form of the new optical disc digital audio formats, admitted that it didn't bring about the anticipated transformative change in sound quality.

A number of early CDs used it before everyone realized how little it was needed and/or the market figured that out. Emphasis was a given for tape and vinyl so natural to include even if not really needed.

If I'm not mistaken, the de-emphasis filters in early CD players were analogue? In which case it had the potential to do more harm than good.

I suppose it would be interesting to test some of these very early CD's. I don't think I've had any CD's dating earlier than about 1985.

 

[mansr]

If I'm not mistaken, the de-emphasis filters in early CD players were analogue?

Quite right. Here's the relevant part of the schematic for an early Philips CD player:

At left is the (current) output from the DAC chip. The input at the bottom is the emphasis flag from the CD decoder. The opamp serves as I/V converter and initial low-pass filter, the frequency response depending on the state of the emphasis indicator.

 

[j_j]

Quite right. Here's the relevant part of the schematic for an early Philips CD player:
View attachment 48506

At left is the (current) output from the DAC chip. The input at the bottom is the emphasis flag from the CD decoder. The opamp serves as I/V converter and initial low-pass filter, the frequency response depending on the state of the emphasis indicator.

Interesting. My first reaction is that that's going to have some high level linearity oddities. Maybe not too bad at 200mV levels.

 

[Serge Smirnoff]

I have calculated a sine wave of magnitude .4, rounded it, added uniform dither, rounded that, and added tpd, and rounded THAT, i.e. I've set a step size of '1' in the quantizer.

The routine that does this will follow in a comment.

The photo I'm attaching now has 4 lines. each line shows a signal. The four columns are
(signal) (signal spectrum) (error signal) and (error spectrum)

The four lines areView attachment 47956
undithered unquantized
undithered quantized - NOTE THE SIGNAL IS GONE. POOF, LOST, MISSING, GONE. Information is ***LOST***
uniform dithered quantized - Note, signal is back, but noise level changes with data value, so data modulates noise level.
TPD - Note, signal is back at same level, noise is 3dB higher, but there is no modulation of the noise level.

This shows, clearly, visibly, and without debate, that DITHERING PRESERVES INFORMATION.

That is all!

Thanks @j_j for your elaborated example. It really helps to clear up the dithering issue.

Dithering, in very basic sense, is adding noise for some purpose. In various areas of its use the purposes are different. It helps transmission of signals in fiber optics, reduces frequency peaks of electromagnetic emission in equipment; jitter is sometimes considered as temporal dither. In computer graphics it helps to use limited color palettes more efficiently by utilizing some features of human vision. In audio the dither is used quite similar as in computer graphics. It is added before quantization for changing structure/shape of quantization error. This serves two purposes:

(p1) For audio signals exceeding LSB it helps to convert audible distortion, caused by quantization, into less annoying random noise. It is pure psychoacoustic treatment of a signal because without the knowledge of our hearing it is not understandable why one type of error should be replaced with another one.

(p2) For audio signals below LSB, as in your example, it helps to preserve tonal components of a signal in the form of noise “modulated" by those components. Only in this case dithering can be considered as mathematical operation and only with the caveat: this operation has sense because sound processing in our hearing system starts with frequency analysis in basilar membrane, so we are very good in discerning tones buried in noise. If our hearing system would have developed in a way that, say, nerve firings would code waveform of a sound and the brain would compare the waveforms with the ones stored in memory, then such operation of dithering would not work for us. In other words, this “preserving of information" by means of dithering is successful thanks to our auditory system that is able to make use of its results.

Thus, in audio the dithering is adding noise for the purpose of improving auditory characteristics of sound signal and in this sense it is the psychoacoustic treatment/processing. In df-metric this is immediately understandable because dithering degrades waveform of a sound signal (worsens df levels) and the same time improves perception of that sound, which effectively means that the degradation was made for the purpose, using knowledge of hearing.

 

[j_j]

Serge, stop trying to justify linearization as a "psychoacoustic" thing. It's just not.

Dithering PRESERVES information. And that is using the term "information" formally, as in information rate.

 

[xr100]

. In computer graphics it helps to use limited color palettes more efficiently by utilizing some features of human vision

Quick Photoshop test. Save as GIF, 8 colours, Palette = Local (Perceptual), Forced = Primaries.

Dither (setting) = NONE

Dither (setting) = NOISE

Dither (setting) = ERROR DIFFUSION (75%)

This is a bit of an invalid example since Photoshop is selecting 8x 24-bit colours to form the "palette," and its "dither" settings are rather limited, and not least since it's imaging and not audio.

Nevertheless, I think it's plain to see that more INFORMATION is coded in the latter two.

 

[scott wurcer]

Thanks @j_j

(p2) For audio signals below LSB, as in your example, it helps to preserve tonal components of a signal in the form of noise “modulated" by those components. Only in this case dithering can be considered as mathematical operation and only with the caveat: this operation has sense because sound processing in our hearing system starts with frequency analysis in basilar membrane, so we are very good in discerning tones buried in noise. If our hearing system would have developed in a way that, say, nerve firings would code waveform of a sound and the brain would compare the waveforms with the ones stored in memory, then such operation of dithering would not work for us. In other words, this “preserving of information" by means of dithering is successful thanks to our auditory system that is able to make use of its results.

This is the same wrong idea that crops up all the time, there is no "modulation". Please read the ample literature on the subject. Below the LSB there is no tone buried in the noise, an ideal noiseless 16bit quantizer is blind below an LSB. BTW I find some software to be inconsistent on what is zero, that is some programs have a +-1/2 LSB offset at the origin.

 

[xr100]

In other words, this “preserving of information" by means of dithering is successful thanks to our auditory system that is able to make use of its results.

Spectrum of 1kHz sine at -120dBr, 64-bits Float:

Spectrum of 1kHz sine at -120dBr, 16-bits, Truncated:

Spectrum of 1kHz sine at -120dBr, 16-bits, Flat TPDF Dither:

The noise floor in the last example is at approx. -90dBr.

Thus, truncation to 16-bits per sample has resulted in total loss of information; dithering to 16-bits per sample has preserved it.

The information is encoded regardless of whether there is a "mechanism" available to "recover" it out of the noise floor. In this case, said "mechanism" is an FFT spectrum analyser.

 

[DonH56]

I'll have to go back to the 1980's and ask the radar systems that used dither how they felt it sounded, psychoacoustically speaking.

 

[restorer-john]

I suppose it would be interesting to test some of these very early CD's. I don't think I've had any CD's dating earlier than about 1985.

Here's a few samples showing the switched emphasis (for testing the de-emphasis accuracy) from my test disc collection, including the incredibly rare Philips 3/5/5a test disc set (alignment calibration and 5a defects test disc), the Pierre Verany 2 disc set (with calibrated defects for tracking EC testing etc), The Denons from 1982/3, CBS-CD1 etc.

Note the annotation for the emphasis flag as on or off for each of the 99 tracks in the booklet

Emphasis was of course a way of dealing with the early 14 bit D/As to give a decent HF performance.

Dither on the other hand, in the early digital recorders produced by NHK/Sony was produced by amplified noise (white) from a Zener diode.

 

[scott wurcer]

Interesting mine has all Japanese no English but the use of P.I.L.'s Under the House can not be a coincidence. I'll have to dig it out.