Dithering is a Mathematical Process - NOT a psychoacoustic process.

[xr100]

It's not a full review @xr100 !

You can use white noise or a multitone to give a pretty decent FR.

Sorry, in a particularly grumpy mood at the moment, should have added "do you have" and "please" etc. ;-)

 

[mansr]

It's Philips first 14 bit D/A converter (2xTDA-1540S lowest grade epoxy packs) strapped onto their first 4x OS filter in the cheapest CD player they could bring to market in 1985.

Also compare apples with apples. It's 16/44 not 24/44 which all ASRs DACs get tested with- a format that doesn't really represent the majority of files or physical content does it?

And it beat its spec of 0.005% by a decent margin. And what is your A/D's contribution BTW?

Don't get me wrong, the Philips CD-150 was a horrible cheap nasty thing, but let's put its performance into perspective- it's actually quite good.

The CD150 is basically a cost-reduced CD100. All the important parts are the same. Out of curiosity, I picked one up off eBay for a few quid. The CD100 sells for around £1000. Insanity.

As for the graphs, 16/44 can be a lot better. As you say, though, this was probably reasonable in 1985.

 

[mansr]

Frequency response? Or no suitable test tracks?

I can post a more comprehensive set of measurements tomorrow if you're interested.

 

[restorer-john]

Can you elaborate, please?

IIRC, the early players weren't too hot in terms of having properly engineered/specified power supplies/isolation between sections. Performance could depend on how hard the transport has to work...

The machine concerned was an extreme budget player. By 1985, CD had enjoyed an initial blip of sales, mainly by audiophiles and some "tech" early adopters. Sales were poor and the format was needing a boost or it would fail.

All the major brands scurried to produce low cost machines. Sony didn't cheapen their products too much, but released a lower cost version of the CDP-101 and had the buzz of portables and car CD players to help it. Sony went upmarket when everyone else went down as they had the initial sales to offset their R&D and were big enough to absorb losses for a few years. Even their first portable, the D-50 was loss making for 2 years. Apparently it was sold at half its actual cost for that period. Sony was doing particularly well in Japan too.

Philips still hadn't got the TDA-1541 (16 bit D/A) ready, refused to use the Burr Brown PCM-56 (which was new), so they pushed out a bunch of cheap players with nasty plastic moulded chassis, a massively cheapened CDM (laser mech) and their old lower grade TDA-1540 (wider tolerance) and virtually no user functionality other than the basics. I think Magnavox/Philips/Nordmende etc all got the cheap Philips machines.

In Japan, CDC/Teac/, Toshiba/Onkyo/NAD, Sharp, Panasonic/Technics, Akai, Pioneer etc all had machines hitting market with incredibly low prices and full of "features". They also had better D/A converters, but in the interests of keeping the price low, they only used one D/A and time-shared the output. Within a few short years, all machines had two (or more) D/As, one for each channel and interchannel delays were a thing of the past.

Anyway, enough of the history.

 

[j_j]

Philips still hadn't got the TDA-1541 (16 bit D/A) ready, refused to use the Burr Brown PCM-56 (which was new), so they pushed out a bunch of cheap players with nasty plastic moulded chassis, a massively cheapened CDM (laser mech) and their old lower grade TDA-1540 (wider tolerance) and virtually no user functionality other than the basics. I think Magnavox/Philips/Nordmende etc all got the cheap Philips machines.

I had the Magnavox (Maggotbox) 2020SL. The chip it used for oversampling did not round correct in the oversampler. This is detailed out there on the web somewhere, and as a result, it had bogus low-level linearity. It blew a power supply regulator about a month into its life. After 8 weeks at NAP in Seacaucus NJ, it came back, played half a track and emitted smoke. Back it went again. TWELVE weeks later, it was returned, this time it worked for about a year and a half, until one channel just went dead.

At that point I gave it to a college student who may have fixed it, or may have used it for a practice discus, I don't know. Said college student was warned of its history.

In a USENET test of something like 8 different CD players, it was the only one to fail (which is to say, it WAS identified) in a very, very primitive ABX test in a living room. I know, I'm the one who did 9/10 and 8/10 on it. The problem was, of all things, somehow in the imaging side of things. That can be found somewhere on net.audio or rec.audio.(somethingorother) a long long time ago. I tried to find the report on google search and failed.

 

[restorer-john]

After 8 weeks at NAP in Seacaucus NJ, it came back, played half a track and emitted smoke.

Such a cool story. I can add another equally frustrating one.

Dad (I was only about 17yo) umm'ed and aah'ed and finally purchased an Akai CD-D1. A magnificent machine using twin TDA-1540Ds and the Philips 1st gen OS chipset, alongside a completely discrete tracking and focus front end. Vertical machine, entirely built by Kyocera even before the home market release of the Sony CDP-101. (7 months earlier)

Sounded beautiful, but it had random "blips" which drove him crazy. Akai couldn't fix it- their technicians had no idea what they were dealing with and eventually replaced it with a new model (CD-A7) at no extra cost. Never forgot that drama. He gave the CD-A7 to me for my 21st birthday and moved on.

Years later, I bought up a few 1st gen Akai CD-D1s and spent an entire day (no kidding) aligning and stetting up a single unit. Would play anything perfectly. Six months later, wouldn't play a thing. Gorgeous machine, but a pain in the a##.

Anyway, I have dug one out of my storeroom just for fun. Will see what it can do- or not do. Anyone want to see it?

Cheers.

 

[xr100]

The CD100 sells for around £1000. Insanity.

I suppose it might look cool in a clear perspex cabinet; that's about all it's good for. Horrifically crude user interface/"display."

 

[mansr]

Dad (I was only about 17yo) umm'ed and aah'ed and finally purchased an Akai CD-D1. A magnificent machine using twin TDA-1540Ds and the Philips 1st gen OS chipset, alongside a completely discrete tracking and focus front end.

So it was basically a clone of the CD100 then?

 

[Serge Smirnoff]

Really now? Seriously? Now we're saying THAT?

Here's a narrowband noise signal, with the same plotting as the original sine signal.

Yes, one could do wider band noise, but I'm pretty sure that the statistical approach to detection will require endless explanations.
View attachment 48732

Oh, look, there's the noise signal in the output. Yes, it's noisier, but it's still detectable and this is purely by analytic methods.

For sure I should have said "Non-tonal components (noise-like) can be saved for human hearing by this method with much less success." But your example with two tonal components is a bad representative of the signal I meant - non-tonal (noise-like).

And yes, I'm serious, may be too peremptory sometimes, but I'm sure you can cope with this.

 

[JohnYang1997]

For sure I should have said "Non-tonal components (noise-like) can be saved for human hearing by this method with much less success." But your example with two tonal components is a bad representative of the signal I meant - non-tonal (noise-like).

And yes, I'm serious, may be too peremptory sometimes, but I'm sure you can cope with this.

Or just wide band noise...

 

[Serge Smirnoff]

Or just wide band noise...

May be the tail from some percussion ...

 

[Serge Smirnoff]

I had to Google the above terms and it seems that their origin is in structuralism/post-structuralism/linguistics?

Leaving aside their validity, I am all for broader and cross-disciplinary views but it is clear that you are mixing categories and getting into an epistemological mess here.

Probably, they could be used to look at, say, a "bitcrushed" sound in a recording (as discussed in the other thread.)

What is being discussed here, though, is the application of dither to linearise quantisation, as set out in the ("narrowly" (!) defined) discipline of information theory. No "broader" perspective, valid or otherwise, is needed.

And, I thought the objective of the "DF Metric" was to identify audio equipment and/or processes which yield perceptual transparency?

Thanks to this discussion the origin of my terminological confusion is pretty clear now (for me at least).

When I took from time to time in the past some experiments with dithering I always used normal signals (above LSB), which I used in other experiments with df levels. With such signals the dithering before quantization shows all features of psychoacoustic processing as it degrades accuracy of the d+q operation (worsens df level of a signal) and the same time improves perception of that signal (by changing the character/type of distortion). I never used signals below LSB for the experiments. But, right you are, exactly this case (p2 in the post #113) was the main reason for applying dither in audio, while perceptual improvements in above-LSB case (p1 in the post #113) was just a nice bonus. And indeed below LSB the dither can be considered as mathematical operation that linearizes behavior of quantization operation and thus preserves some info from below-LSB region of a signal. In this case the dithering can be really considered as formal/math operation. So, classification of dithering depends on what case - p1 or p2 - is considered.

There is another reason for the confusion. It is my expanded interpreting of some terms from the audio science. This happens because I used very basic/general assumptions developing df-metric and the whole approach to audio quality measurements. For me reproduced sound is fully determined by the signal that codes it. In other words an audio signal carries all information that is perceived during audition. And this is the point, where the distinction between different layers/aspects of information turns out to be very helpful for further analysis.

Assume we have a signal of unknown origin - a black signal (output of black box). It can be anything - price, temperature, coordinate of some movement, index of happiness in population ... we just do not know and can not obtain that info. This is a signal on syntactic level without a knowledge of its meaning/semantics. Now if some processing/operation of a signal has sense only taking into account that the signal is intended to be perceived/listened (requires knowledge of the semantic level) then such processing/operation is psychoacoustic by its nature. In this sense amplification, resampling, quantization ... of a signal are not psychoacoustic operations; equalization, dynamic compression, removing of inaudible components ... are. According to this criterion the dithering before quantization is psychoacoustic operation for signals above LSB (does anybody apply dither, say, before rounding financial data, which is a type of quantization?) and formal/math operation for below-LSB signals (used in many areas; radar systems as an example).

Taking all this into account I can not uniquely define the dithering as psychoacoustic processing/treatment of an audio signal and will consider it twofold in this sense. I think the issue is closed for me now. Probably this is the case where discussion of the issue is more important/helpful than the result.

A bit more of creative but rather off topic prose.

Such extended view of audio processing/measurements is also the reason for my broader view of audio distortion. If reproduced sound is fully determined by the signal that codes it, then perceived distortion of the sound is fully determined by the signal distortion. The distortion then can be analyzed further as caused by linear or non-linear process with all differences, mathematical and perceptual, between these cases. Distortion is always the distortion of something. Having more general approach the distortion refers to the whole signal, having the more narrow definition it refers only to magnitude spectra of a signal, which seems to me less grounded because linear distortion is also audible, just to less extend and more in regards to spatial perception of sound. I do not see this as enough reason for not calling it a distortion. So, I reserve the right for myself to use more general definition of this term regardless of commonly accepted in audio science more narrow definition (hence supporting the general definition of distortion given in wiki). Though I promise not to insist on my point of view and where important - to note/mark that I use the broader term.

The last but not the least. The extended view of audio processing/measurements not only allows creation of clear axiomatic system of meanings, concepts and terms in audio science. It helps to develop grounded and reliable approach to measurements of perceived audio quality. This approach is a bit unusual and can be considered by someone as radical or even ridiculous (I mean you, j_j). But I'm pretty sure that sooner or later it will be accepted by audio engineers. I will state it in the different thread, where I already described/explained main concepts of df-metric. It will look more logical there and a bit offtop here. Give me a few hours.

 

[SIY]

For sure I should have said "Non-tonal components (noise-like) can be saved for human hearing by this method with much less success." But your example with two tonal components is a bad representative of the signal I meant - non-tonal (noise-like).

This is not correct either.

 

[scott wurcer]

This approach is a bit unusual and can be considered by someone as radical or even ridiculous (I mean you, j_j). But I'm pretty sure that sooner or later it will be accepted by audio engineers. I will state it in the different thread, where I already described/explained main concepts of df-metric. It will look more logical there and a bit offtop here. Give me a few hours.

This is the crux of the issue, you are wed to your idea of a new metric and are not responding in a rational manner to peer review.

 

[Serge Smirnoff]

This is the crux of the issue, you are wed to your idea of a new metric and are not responding in a rational manner to peer review.

I'm wed to another idea, df-metric is only an instrument for its realization.
I think I'm quite rational in responding to the objections, as best as I can.

 

[mansr]

I can post a more comprehensive set of measurements tomorrow if you're interested.

Done: https://www.audiosciencereview.com/forum/index.php?threads/philips-cd150-measurements.11296/

 

[xr100]

And indeed below LSB the dither can be considered as mathematical operation that linearizes behavior of quantization operation and thus preserves some info from below-LSB region of a signal. In this case the dithering can be really considered as formal/math operation. So, classification of dithering depends on what case - p1 or p2 - is considered.

Thanks for your response. A partial reply follows...

Dither also linearises signals of amplitude ABOVE "LSB region."

For example, here is a 1kHz sine wave at -40dBr, truncated to 16-bits: (rescaled so that the 1kHz sine wave is at 0dBr in the plots.)

(N.B. Sample rate = 96kHz.)

And the same, only this time dithered and noise shaped: (iZotope Ozone, Dither=Strong, Noise Shaping=Max.)

Note that the "floor" of the FFT plots have been set to -100dBr.

"Error" for the truncated case is ~-56dB, for the dithered and noise shaped, ~-30dB.

However, if a brickwall LP filter at 20kHz is applied, the truncated case is ~-64dB, whereas the dithered and noise shaped is ~-79dB.

Thus, even the most basic consideration of the signal's spectrum, vis-à-vis human auditory perception, gives an entirely different picture.

It's also clear from the plots that truncated example has yielded odd-order distortion, with components at frequencies where hearing is highly sensitive; whereas, the dithered and noise shaped example, within the floor of -100dBr, has no noise throughout most of the audible range, and what little there is, is on the fringe.

> (does anybody apply dither, say, before rounding financial data, which is a type of quantization?)

Not a domain that I have much, if any, familiarity with. However, I don't think the "signal" could be considered the information that is lost through rounding, but rather the signal is the "trend;" a moving average filter being a basic example of a method to extract the "signal" out of "noisy" movements in price. IOW the quantization error is below the "noise" anyway.

 

[SIY]

Not a domain that I have much, if any, familiarity with. However, I don't think the "signal" could be considered the information that is lost through rounding, but rather the signal is the "trend;" a moving average filter being a basic example of a method to extract the "signal" out of "noisy" movements in price. IOW the quantization error is below the "noise" anyway.

It does work universally. My first exposure to dithering was in the 1970s in capturing photoacoustic interferograms, decidedly nonperiodic.

 

[j_j]

For sure I should have said "Non-tonal components (noise-like) can be saved for human hearing by this method with much less success." But your example with two tonal components is a bad representative of the signal I meant - non-tonal (noise-like).

And yes, I'm serious, may be too peremptory sometimes, but I'm sure you can cope with this.

No, my signal does not have two tonal components. It is narrowband noise.

 

[j_j]

Or just wide band noise...

Using statistical methods (like your cell phone does unless it's turned off) you can detect the presence of 'excess' wideband signals in a dithered signal, when the wideband signal (which is indeed noise-like, and can be "just noise" t00) well enough to transmit data.