18/20/24-bit recording with atrac, mp3 &c.

[dex Otaku]

(This is related to the thread about the MD deck with 20-bit playback.)

Why do some MDs claim 20+bit recording and playback?

Lossy formats such as mp3, atrac, &c. use various transform methods to convert data from linear PCM [time-domain] to the frequency-domain. Fourier transforms, discrete cosine and modified cosine transforms, &c. tend to be the most common wasy of doing this.

What this means, in English, is that once the original PCM data is compressed, it doesn't even have a bit-resolution per se - at least, not in the conventional PCM sense.

While the sampling rate has to be locked [or resampling done] between all stages of encoding/decoding, the bit depth is actually kind of irrelevant. As long as the codec knows how to handle more bits, what you'll end up with is a slightly more accurate encoding. And when I say slightly, I'd opine that it's so slight as to be marginal, but still maybe measurable.

Likewise, when decoding you can do so to any bit depth, with higher bit depths getting marginally better accuracy [as long as they're properly dithered adn noise-shaped] from the decoder.

This does not mean that you can decode something at a higher resolution and get better results than what originally went into the encoder, however.

GIGO - Garbage In, Garbage Out; if you rip a CD [16 bit audio] and then decode it as 20-bit, all you're doing is getting a more accurate decoding of the compressed format it's in. You can NOT get something better than the original. It's LOSSY, remember?

In the end, I personally consider this a huge non-issue; since 95% of source material out there now is quantised to 16-bits, the differences that higher-resolution playback equipment can make are negligible at best.

It would be better to concern ourselves with proper dither &c.

[RandomHajile2]

well i kind of dissagree, on a premastered album, such as Mariahs 1995 Daydream, this is a 20bit Super Bit Mapped dithered down on to Gold CD, but on the MiniDisc version it truly is 20bits.

i have a 30es deck, and listening to certain premasters or recordings of my pro-ject deck is fantastic (with its 4 filters)

i am interested in this, i want to get a soundcard and capture a 16bit out put and compare it to a 20bit out put from my deck, as you can set the digtial out put word length!

[dex Otaku]

The MD version is ATRAC. It may be ATRAC coded from 20-bit PCM, but it's still ATRAC, in which case the bit resolution is still basically irrelevant.

Compressed audio does not have a bit resolution in the traditional sense [i.e. that of PCM]. Period.

What you have is a marginally more accurate ATRAC encoding than one would get from a 16-bit source. Saying that "on the MD version it truly is 20 bit" is a misnomer. ATRAC itself is neither 8, nor 16, nor 18, nor 20, nor 24, nor 2-billion bit audio. It's ATRAC. Period.

Comparing 16- vs. 20-bit output .. well, you will have the longer wordlength from your ES deck. You will get 20-bit audio. But it's only as good as the source, which in this case is ATRAC, not the 20-bit master.

Again: yes, you'll have slightly more accurate decoding, but it's still only as good as the source - ATRAC.

[aeriyn]

Thank you, dex. T.T No one wanted to listen to me on this subject.

[RandomHajile2]

the thing is, if ATRAC aint even 16bit, what the hell is Dolby AC-3/DTS/AAC/WMP and other compression formats???

DTS is 20bits right?

its compressed at a 4;1 ratio comparable to MD.

look up the interview with the sony 3es engineers, they say its 20bits all the way.

take digi beta for an example, the video is 10bit compressed 2;1 via sonys CORE compression, so its not that lossy, but the SDI output is still 250megabytes plus, cos its been reconstucted.

so i guess even DVD-audio aint really 24bit as its compressed, or for that matter even SA-CD is compressed if you didnt know???

there always is a bit depth, in audio or video.

[dex Otaku]

Bits [as used here] refers to the quantisation depth of linear PCM recording.

Most lossily-compressed formats rely on converting PCM to a completely different domain - from that of time [samples per second, linear PCM] to frequency [intensity @ pitch, in bands and sub-bands, as used by basically all lossy formats].

When the engineers say it's 20 bits, what they meaning is that the native resolution of the compression format is sufficiently high that it can reproduce 20-bit dynamic range when converted back to PCM.

I can't speak as to CORE compression, I'll have to read up on that one. It sounds to me like it might use something similar to ADPCM or non-linear quantising.

DVD audio has multiple formats, from AC3 [which is a lossy compression method quite different from the rest, but still uses methods related to the rest] to DTS [which while it has been updated is one of the oldest formats still being commercially used, coming from around the same time as the original ATRAC], PCM of various bit depths and channel configurations, and MLP [meridian lossless packing].

SA-CD uses Sony's bitstream digital, which is a lossless, uncompressed format unrelated to PCM, based on another known as pulse width modulation. It technically also has no bit-depth in the sense we're talking about [it is a 1-bit system].

There is always a bit depth in PCM recording.

There is also always a bit depth on A/D conversion for compression to lossy formats - however, the lossyformat itself, the format being stored in, does not have a bit depth in the traditional sense, because it is NOT PCM.

edit:

Looking for info on CORE audio and digi beta. Can't find a reference to CORE audio anywhere on Google so far. Digi beta's latest incarnation lists its specs [taken directly from a Sony product info PDF]:

48 kHz (synchronized with video)

20 bits/sample

Frequency response (0 dB at 1kHz): 20 Hz to 20 kHz +0.5/-1.0 dB, Dynamic range (at 1 kHz, emphasis ON): More than 95 dB,

Distortion (at 1 kHz, emphasis ON, reference level): Less than 0.05%,

Cross talk (at 1 kHz, between any two channels): Less than -80 dB,Wow & flutter: Below measurable level

I've come across the same spec on every info page re: digi beta, which would seem consistent with what I remember about at least one of my contractors' [CTV] use of them from about 1995. The consistency in format was important at the time because D1 through D3, Hi8, BetaSP, digiBeta, and basically all videotape formats being used all conformed to the 48kHz, 16-bit standard of "analogue" mode on DAT, which was later extended to 18 and 20 bit.

Where can I find this info?

[aeriyn]

Oi, this guy can't get it through his head.

Hey, what dex is saying is true.

Only pulse code modulation formats have bit depths. It is true that DVD-A, which is a type of PCM and NOT compressed btw, is 24/192. Super Audio CD is not PCM at all, but DSD (Direct Stream Digital) and doesn't have a bit depth because... well, it's not PCM.

If it's not pulse code modulation, it doesn't have a bit depth.

Thank you. :sleep:

[dex Otaku]

DVD-A is a multiple-format standard, actually, encompassing PCM [anywhere from mono 16-bit 44.1kHz to 6-channel 24/48 to stereo 24/192, different channel combinations at different rates and bit depths being possible as long as they don't exceed available bandwidth], MLP [lossless compression under more or less the same constraints], AC3 [stereo to 6.1-channel , but poorly suited to music mainly because the audio industry has no universal mastering standards as the film industry does, and part of AC3 encoding is the requirement of following Dolby's standard which includes telling the codec what the peak level of the center channel is, which determines playback levels of all channels by declaring what the normal headroom is compared to the dialogue level standard for theatres], DTS [usually just 5.1], and likely still MP2 [the standard used mostly in Europe for DAB and DVD].

[RandomHajile2]

core is the video compression, i didnt mention it was the audio, you just asumed, the audio is uncompressed.

i know what dolby ac-3 is, and i know the bit rate is better on dvds then it is on film prints,

i know DTS is 20bit, and there was an updated 24bit dts but theres like no software.

i know a lot about audio formats and video formats, i know the reason why its 44.1khz is because of a converted sony beta VCR.

why does it seem like you post is sarcastic???

like i said, if its digital, there is bit depth, what the hell does it get converted to from its lossy form??? what gets sent to the internal DACs???

and by the way, if you look up SAcd properly you will find its compress lossless 2-1

and i know its a 1bit sample running 64 times faster then cd (or is it 8)

i hav a SCD-1 (even though it lyes and is not a true balanced deck!)

[dex Otaku]

core is the video compression, i didnt mention it was the audio, you just asumed, the audio is uncompressed.

We were talking about audio compression, not video compression. Hence the assumption.

i know DTS is 20bit, and there was an updated 24bit dts but theres like no software.

Sorry. Check here: http://www.dtsonline.com/media/uploads/pdf.../whitepaper.pdf

DTS is in fact a different type of sub-band coder from those such as MP3. Still, it is a sub-band coder nonetheless. End result: when they say it's 20 or 24 bit, what they're referring to is the encoder's ability to handle that resolution of PCM input stream natively, without requantising it. The stored DTS stream itself, being compressed, does not have a bit depth in the traditional sense.

i know a lot about audio formats and video formats, i know the reason why its 44.1khz is because of a converted sony beta VCR.

I don't doubt that you do. That doesn't mean that every single thing you know is correct, or that there's nothing left to learn, though. The same applies for myself, of course.

why does it seem like you post is sarcastic???

Because your resistance to understanding the fact that lossily compressed audio is not PCM, and therefore does not actually have a bit depth in the sense that PCM does, is frustrating the hell out of me.

My apologies. I know it's not necessary, and is in fact a childish thing to do.

If you preferred, however, you could take your reasoning over to HydrogenAudio, where posting it would immediately have 50-75 people jumping on you to tell you why you're wrong.

like i said, if its digital, there is bit depth, what the hell does it get converted to from its lossy form??? what gets sent to the internal DACs???

No. No, no, no. If it's PCM, there's bit depth. Other stream formats like DSD and every lossily-compressed one that I'm aware of don't have a bit depth [or, if you prefer, with PWM the depth is 1 bit - but since it's a different modulation system, the question is moot, PWM does not have a bit depth, only a bandwidth].

And what does it get converted to? PCM, usually. The thing is - and this is one of the great advantages of lossy formats in general - you can take the compressed stream and decode to whatever bit depth you want, because the stream itself has only a bandwidth [data rate], not a bit depth.

and by the way, if you look up SAcd properly you will find its compressed losslessly 2-1

and i know its a 1bit sample running 64 times faster then cd (or is it 8)

Yes it is. I meant in the sense of lossy compression [which, technically, is also incorporated into the SACD standard, see http://www.superaudio-cd.com/technology_ex.../whitepaper.pdf ] but in any case, yes - DSD is losslessly compressed from a PWM stream at 64 times the sampling rate of CD [with 1/16th the number of bits of course], or 2.8224 MHz [4 times the data bandwidth of CD audio for stereo, if I'm not mistaken].

So, let me rephrase the whole thing from the start, then:

PCM has a bit depth.

Lossy formats [and PWM] have a bit rate, or signal bandwidth.

In the case of PWM, it could be argued that it has a bit depth of 1 bit, but this is really a misnomer in my opinion. A depth of 1 implies no depth, for one thing. For another, it is called pulse width modulation for a reason.

In the case of [most] lossy formats, there is only a bit rate, independent of depth. The higher the depth of the PCM signal it was coded from, the higher the accuracy of the encoding. The higher the depth of the decoded PCM signal, the higher the accuracy of its decoding. However, it being lossy, the comparison is moot - regardless of what came in, what comes out will be lower in overall resolution.

You could say that [most] lossily-compressed formats, though not by definition having a bit depth in the sense that PCM does, they do have a reference bit depth - that being the depth of the original stream they were encoded from. The actual compressed stream itself does not have a bit depth, though.

Interesting point though: as has been proven with ATRAC's newest incarnations, the compressed format itself can actually exceed 16-bit PCM in terms of dynamic range - probably because it is free of bit depth constraints. Of course, the source would have to be that good for this to be true.

Compressed data from a 16-bit source can be decoded with higher accuracy to, say, 20 or 24 bit PCM. However - you can't exceed the resolution of what went in. Ever. Period. You can't make something out of nothing. This is the equivalent of blowing up a 100*100 JPEG to 600*600 and saying it's a higher resolution picture. It's not higher in resolution - it's just bigger.

In the case of, for example, DTS - which also does a lot of funky processing to load the available bandwidth better, this can mean that in certain senses it is in fact exceeding the quality of the input signal. The thing is, it's still lossy. I would compare DTS in some senses to dbx bandwidth compansion in analogue recording; you are in fact loading the tape better in terms of signal balance and dynamic range, but you are also introducing artifacts into the signal at the same time.

So, again:

PCM = bit depth.

Lossy = bit rate, independent of depth.

Last thing: I kept saying 'most' lossily-compressed formats, because there are some that do simple bit reduction or requantise audio to nonlinear PCM [most commonly 10 or 12 bit] that do in fact have bit depths, but then, they aren't sub-band encoders. Some sub-band encoders, such as DTS, does use these methods within them [ADPCM] though.

edit:

Incidentally - 44.1kHz was chosen because the predominantly available digital audio mastering recorders that existed when CD was brought about were Sony's PCM-1630s, based on their U-Matic 3/4" tape format, which was [originally] used for broadcast-quality video recording. See http://en.wikipedia.org/wiki/Compact_disc . I have actually used U-Matic equipment, and at least got to load a PCM-1630 once for backing up tracks from a Sony DASH recorder.

[RandomHajile2]

so this means nothing? http://www.minidisc.org/mj_ja3es.html

so theres me all this time with my ja3es and mds-ja30es (both black) thinking when i record from my vinyl, it was digitising/quantizing @ 44.1khz at 20bits PCM, compressing 4.7 to 1, and on playback RECONFIGURED back to 44.1khz @20bits PCM and sent to the D/A.

lying monkeys. are they wrong?

i know of the u-matic audio conversion, there was one for the betamax aswell, about the same time beta changed its logo, and i know its 3/4 inch, i work in tv and have dubbed/archived loads of vacume sucking 2", dirty 1", and 8track umatic hi and low band, to d1/digi beta.

[dex Otaku]

so theres me all this time with my ja3es and mds-ja30es (both black) thinking when i record from my vinyl, it was digitising/quantizing @ 44.1khz at 20bits PCM, compressing 4.7 to 1, and on playback RECONFIGURED back to 44.1khz @20bits PCM and sent to the D/A.

lying monkeys. are they wrong?

No, they're not wrong. And neither are you - but read what you've said. The audio is quantised as 44.1kHz 20-bit PCM, then encoded to ATRAC [at which point it has no bit depth in the traditional sense], which is then decoded back to 44.1kHz 20-bit PCM.

What goes in is PCM. What comes out is PCM. These have bit-depths.

The ATRAC audio that is stored does not, per se. And, as I said, you can decode the compressed audio to whatever bit-depth you want, but you cannot exceed the resolution of either what came in or what the limitations of the CODEC impose.

Furthermore, if you actually -read- the document you linked to, they explain in excruciating detail all of the reasons for what I've been saying all along.

I think the whole issue here is that you and I are actually agreeing with each other, with the exception of one really minor detail. [The agreement being that higher resolution in = higher encoding accuracy, logically follow by higher resolution decoding = higher decoding accuracy; the disagreement being over only one thing, really: the compressed format itself does not have a bit depth in the traditional (i.e. PCM) sense, though you would argue that whatever the bit depth of the signal being encoded is, is the bit depth of what's being recorded, which is incorrect.]

It's been sort-of fun but maybe we can lay this to rest. Agree to disagree, or something. Because I'm sure there are more useful conversations we could be having than to keep rehashing one single point over and over ad nauseum.

i know of the u-matic audio conversion, there was one for the betamax aswell, about the same time beta changed its logo, and i know its 3/4 inch, i work in tv and have dubbed/archived loads of vacume sucking 2", dirty 1", and 8track umatic hi and low band, to d1/digi beta.

I hated U-Matic when I was in college. The year I started they'd replaced all the 3/4" equipment with Hi8 camera backs and VTRs. Some things we still had to use 3/4" for though, and I always found it a pain in the arse [mainly because the equipment was old and fairly unreliable].

My father and his friends/colleagues started the local cable co-operative here in the mid-1970s. He'd occasionally bring home one of the studio U-Matic VTRs so we could watch a movie they'd taped off one or another satellite feed.

I suppose it's no wonder I've ended up doing the stuff I do/have done.

[Guest NRen2k5]

Interesting point though: as has been proven with ATRAC's newest incarnations, the compressed format itself can actually exceed 16-bit PCM in terms of dynamic range - probably because it is free of bit depth constraints. Of course, the source would have to be that good for this to be true.

I think this is particularly nice with MP3. So much music these days is mastered to be as loud as possible, which means that it is always just kissing 0dB. When encoded to a lossy format like MP3 the quantization will make some peaks exceed 0dB. This will cause the signal to clip when decoded back to PCM for playback. Luckily, since MP3 and the like have a wider dynamic range than PCM (maye infinite... ?), the clipped samples aren't really lost, and you can just adjust the gain values in the MP3 so that the output will be under 0dB and won't clip.

[vova]

If I understood the preceding correctly, "compressed" music is based on completely different principles of digitally recording sound compared to PCM. Would it then be possible to bypass PCM altogether, ie master directly to ATRAC and convert ATRAC to analg in the player? And would it not be possible for ATRAC or any other lossy format theoretically to deliver sound quality better than PCM, which is limited by quantization frequency and bit-depth?

[Guest NRen2k5]

so this means nothing?  http://www.minidisc.org/mj_ja3es.html

In addition to what has already been said, I don't personally like that article very much.They say that ATRAC is high-quality but they only say how they have improved it over time and not how it compares with anything else other than CD and they focus almost exclusively on dynamic range which is not the most important thing ever.

[Guest NRen2k5]

If I understood the preceding correctly, "compressed" music is based on completely different principles of digitally recording sound compared to PCM. Would it then be possible to bypass PCM altogether, ie master directly to ATRAC and convert ATRAC to analg in the player? And would it not be possible for ATRAC or any other lossy format theoretically to deliver sound quality better than PCM, which is limited by quantization frequency and bit-depth?

That's an interesting proposition. It may actually be somehow possible, since after all the encoder is looking for frequency content and not the actual sampling points in the PCM data.

[Edit]Nah... the encoder would never be able to look at the right part of the signal at the right time.[/Edit]

[aeriyn]

Well, that could be a problem because D/A converters don't understand ATRAC. :laugh:

[vova]

That is assuming that players first covert atrac to pcm than run it thru d/a. But why can't they use a direct atrac/analogue converter?

[Hyena]

That is assuming that players first covert atrac to pcm than run it thru d/a. But why can't they use a direct atrac/analogue converter?

The assumption is correct. MD recorders decode first and then run it to a D/A. The same goes for any analog signal coming into the recorder. (A/D->Encoder)

They could make a direct ATRAC/Analog converter, but it would be a misnomer, since quantisization has to take place before encoding does. (You'd have a ADC in there somewhere. ) ATRAC compresses digital audio signals, hence the reason being.

[vova]

which I guess brings us back to square 1 - since most of ATRAC recordings are home-made from 16 bit sources they would be by definition inferior to the original. However, if the label were to master the MDs from the original 24-bit tapes, conceivably they may sound better than their CD brethren. If the MD player has a hi-res decoder that is

[ProAudioGuy]

Dex's responses are spot on regarding the subject of bit-depth and these different types of formats. Bit depth is a PCM concept, but we tend to apply it to non-PCM formats (such as DSD) as well, mostly as a point of reference. What is the bit-depth equivalent of DSD (SACD) for example? Well, somewhere well north of 24bits. But then, that -is- misleading, because DSD works so differently from PCM.

The more important point Dex makes though is not driven home enough. Lossy formats such as MP3, ACC, and yes, ATRAC, all depend on conversion to a linear stream (PCM) first. I say linear in the sense of the time domain, if you want to get technical. So the conversion depth of the A/D and D/A converters becomes signficant. If your A/D front end on an MD unit is 16 bits, then you have a limitation right there.

Now, you say, I'm using the digital input of my MD, so conversion isn't an issue. No? What is the input, native ATRAC? No, it is likely a S/PDIF connection passing at the most 20bit PCM audio through (you don't see a lot of 24 bit stuff on a S/PDIF connection folks). So again, conversion depth -somewhere- along the line has become a factor.

So my MDS-JB940 has 20bit 44.1khz converters. I will get up-to (theoretically) 20bits of resolution on the other end of my ATRAC recording. You say wait, it is ATRAC compressed, how can you re-resolve 20bits? Well, ATRAC supports > 16 bit dynamic range, and that is largely what we are talking about. Then again, the noise floor in ATRAC recordings is probably less signficant than the ATRAC version and the encoding algorithm used.

An interesting subject with many, many dimensions. Remember to consider, as Dex says over and over, the entire signal chain. And remember that in the majority of pop music recordings, at least some of the content has been through 16/44.1 PCM conversion (effects processors, digital recorders, digital mixing desks, etc. etc.).

Happy MD'ing everyone!