Noise shaped CDs

[darmok]

I'd been under the impression that noise shaping for CD audio has been around for quite a long time and was being used pretty much universally on high dynamic range recordings. However, a quick look through my library finds very few CDs that actually show evidence of noise shaping. Sometimes I see what looks like noise shaping that's so faint that I wonder if it originated in a DSD recording that was then decimated to 44.1KHz for CD release; a lot of the time I don't see anything at all.

This is what I expect to see for noise shaping on a CD. Note how there's a significant rise in the noise floor starting at 17KHz up to 22KHz. This is evidence of the quantization noise being pushed out of the (easily) audible part of the spectrum, which results in a lower noise floor in the parts of the audible spectrum that we can more easily hear.

And this is what I see on a lot of CDs. Without noise shaping, the quantization noise is smeared evenly across the spectrum. The RMS level of this particular track is -32dB, and the peak level with the exception of the last 45 seconds is -12dB. For the first 2+ minutes the RMS level is -48dB and the peak level is -32dB! Since the noise floor of a 16-bit file without noise shaping is at best -96dBFS (and actually a little worse with triangular dither) that leaves only 48dB between the noise floor and RMS level for this stretch of the recording.

I have recordings from as recently as this year that don't seem to show any significant noise shaping. I haven't yet found a CD from before 2000 that shows visible noise shaping. Amusingly, the examples above are from the same CD which is a compilation of two recordings. The recording with noise shaping appears to have had it added after the fact, probably in the process of adjusting its level for the compilation CD. The original recording dates to 1987 and was recorded in either 14-bit or 16-bit PCM without noise shaping.

My question to everyone is: what's the earliest CD you can find that clearly has noise shaping? Can you see a pattern in when labels started using noise shaping? I definitely see it more recent releases from BIS (including the last RBCDs before they started issuing SACDs), Deutsche Grammophon, and Chandos. Naxos doesn't seem to be consistently using it, and among smaller labels I have a release on Alpha Classics from 2015 without noise shaping and one from 2018 with noise shaping.

 

[DonH56]

Noise shaping is a function of the ADC, and/or the DAC. It is not part of any CD standard I can recall. Delta-sigma ("noise shaping") ADCs and DACs did not become popular until the late 90's IIRC though they were around long before then. It took a while to iron out their quirks and get sufficient processing power to take advantage of them. They really took off by 2000 and after when refined architectures were developed and significant on-chip processing was made possible by advances in technology.

Delta-sigma converters: https://www.audiosciencereview.com/...igma-delta-digital-audio-converters-dac.1928/

 

[restorer-john]

what's the earliest CD you can find that clearly has noise shaping?

Pre-emphasis will also demonstrate significant rise (~9dB@20kHz) with a complementary de-emphasis on playback.

The very earliest CDs produced (1982/3) have such characteristics- not to be confused with noise shaping.

 

[staticV3]

@DonH56 I'm pretty sure that OP is referring to shaped dithering, rather than the noise shaping of delta sigma converters

 

[darmok]

Noise shaping is a function of the ADC, and/or the DAC. It is not part of any CD standard I can recall.

It's not necessarily a function of either of those. It can be part of the ADC, but it can also be used to shape quantization noise during conversion from a higher bit depth master to a CD quality 44.1/16 master.

Sony demonstrated their Super Bit Mapping technology in the early 90s (see this article from 1993 and an undated but likely contemporary whitepaper) but even a Sony Classical CD I have from 2016 doesn't show any sign of noise shaping. Hence the question.

 

[DonH56]

Pre-emphasis will also demonstrate significant rise (~9dB@20kHz) with a complementary de-emphasis on playback.

The very earliest CDs produced (1982/3) have such characteristics- not to be confused with noise shaping.

Excellent point. At Nyquist, the sampling (sinc) function has a 9.54 dB roll-off that was compensated with pre-emphasis. But since it compensated the built-in roll-off, I am not sure there was de-emphasis on playback? Unlike RIAA (etc.) Audio not my main field.

@DonH56 I'm pretty sure that OP is referring to shaped dithering, rather than the noise shaping of delta sigma converters

Ah, could well be, did not think of that. I do not remember when dither shaping became common; I thought it was pretty early in the CD days to make the noise floor more "analog" and get rid of (or at least greatly reduce) tones. That was probably around the time delta-sigma converters became more popular as tones were a big problem for early low-order modulators. At that time I was piddling with delta-sigmas and colored noise (dither), but for radar systems, not audio...

 

[darmok]

Pre-emphasis will also demonstrate significant rise (~9dB@20kHz) with a complementary de-emphasis on playback.

The very earliest CDs produced (1982/3) have such characteristics- not to be confused with noise shaping.

I haven't yet run across a CD in my collection that has pre-emphasis on it. I'm sure they're out there, but even the oldest CDs in my collection (purchased in the mid-to-late 90s) have IFPI codes which dates them to 1994 at the earliest.

 

[DonH56]

It's not necessarily a function of either of those. It can be part of the ADC, but it can also be used to shape quantization noise during conversion from a higher bit depth master to a CD quality 44.1/16 master.

Sony demonstrated their Super Bit Mapping technology in the early 90s (see this article from 1993 and an undated but likely contemporary whitepaper) but even a Sony Classical CD I have from 2016 doesn't show any sign of noise shaping. Hence the question.

Yes, @staticV3 just reminded me of that. Since my background is converter design I tend to think in those terms when people mention noise shaping. Sorry, I be shut up now.

 

[darmok]

Yes, @staticV3 just reminded me of that. Since my background is converter design I tend to think in those terms when people mention noise shaping. Sorry, I be shut up now.

No need to be sorry. I should have been more clear. The recordings with a small amount of noise above 17KHz I would consider as having noise either from the ADC or some PCM->DSD->back to PCM process, but noise shaped dither will create the constant smear of noise I showed in the first screenshot.

 

[restorer-john]

I haven't yet run across a CD in my collection that has pre-emphasis on it. I'm sure they're out there, but even the oldest CDs in my collection (purchased in the mid-to-late 90s) have IFPI codes which dates them to 1994 at the earliest.

I have plenty, but then again, not only have I got a bunch of industry test discs, but also was buying CDs from day one.

Pretty much all the early sample discs and the first several years of commercial releases, had pre-emphasised discs. Several of my players have relay operated de-emphasis circuits, triggered by the flag and you can hear the activation.

This 1983 Denon test disc has the emphasis marked for each and every track of the 99 on the disc.

De-emphasis has been performed in the digital domain, but sadly, many players made in the last 20 years don't perform accurate (if at all) de-emphasis.

 

[Blumlein 88]

I haven't yet run across a CD in my collection that has pre-emphasis on it. I'm sure they're out there, but even the oldest CDs in my collection (purchased in the mid-to-late 90s) have IFPI codes which dates them to 1994 at the earliest.

Noise shaped dither was available pretty early, but not always used.

Pre-emphasis is an option in redbook from the beginning, but I think I've only run across 3 or 4 in the few hundred CD's I have. I think it started a boost about 2 khz reaching +10 db at 20 khz. And of course the reverse on playback. Idea being that highs had much lower levels and early players had poor linearity so this kept the low level treble out of noise floor. Early players did the de-amphasis with analog filtering like a RIAA circuit on phono.

Pre-emphasis - Hydrogenaudio Knowledgebase

wiki.hydrogenaud.io

 

[mhardy6647]

This pre-emphasis thing just blew my mind.
Well, reading @Blumlein 88's link did, that is.

This sounds like arcane occult knowledge lost in the mists of time. Some CDs have it, some don't? Some do but claim not to? Wacky stuff!
I thought this digital stuff was meant to be deterministic.

 

[Mart68]

This pre-emphasis thing just blew my mind.
Well, reading @Blumlein 88's link did, that is.

This sounds like arcane occult knowledge lost in the mists of time. Some CDs have it, some don't? Some do but claim not to? Wacky stuff!
I thought this digital stuff was meant to be deterministic.

IME it's not the problem it's sometimes made out to be. I have a lot of early CDs and some must have pre-emphasis but I could not tell you which ones and frankly, I don't want to know.

Like with HDCD - one day I decided to look to see if I had any, and I found one - Neil Young 'Everyone Knows This Is Nowhere' - which I'd enjoyed for years with no problem.

Now I know it's HDCD and is not being decoded I hear it as bright.... (I do have an Oppo player that does decode HDCD but I don't use it in the main system).

 

[Blumlein 88]

IME it's not the problem it's sometimes made out to be. I have a lot of early CDs and some must have pre-emphasis but I could not tell you which ones and frankly, I don't want to know.

Like with HDCD - one day I decided to look to see if I had any, and I found one - Neil Young 'Everyone Knows This Is Nowhere' - which I'd enjoyed for years with no problem.

Now I know it's HDCD and is not being decoded I hear it as bright.... (I do have an Oppo player that does decode HDCD but I don't use it in the main system).

HDCD was made in a way that if not decoded what was available didn't have a frequency difference. It lacked some low level signals. Amir has commented about how Microsoft acquired HDCD rights and he thought it was audible, but when asked to test it blind never could hear a difference.

Pre-emphasis is pretty audible as rather bright. There are software players that detect it and de-emphasize it or have add ons that can. So I would consider un-reversed pre-emphasis a problem. You will easily detect it in a blind test, and will find it obnoxious depending upon the material encoded that way even sighted.

 

[Mart68]

HDCD was made in a way that if not decoded what was available didn't have a frequency difference. It lacked some low level signals. Amir has commented about how Microsoft acquired HDCD rights and he thought it was audible, but when asked to test it blind never could hear a difference.

Pre-emphasis is pretty audible as rather bright. There are software players that detect it and de-emphasize it or have add ons that can. So I would consider un-reversed pre-emphasis a problem. You will easily detect it in a blind test, and will find it obnoxious depending upon the material encoded that way even sighted.

my point was there's a real issue with what you know, aside from what you can actually physically hear. Once the idea is in your head...

Seems that de-emphasis applies -4.5dB at 5Khz rising to -9.5dB at 20Khz so I agree that will be an audible difference if not applied but I don't agree that it will necessarily sound bad, especially not to someone my age - might actually correct my HF loss!

 

[DVDdoug]

Since the noise floor of a 16-bit file without noise shaping is at best -96dBFS (and actually a little worse with triangular dither) that leaves only 48dB between the noise floor and RMS level for this stretch of the recording.

"No fair" comparing the 96dB peak ratio to the 48dB RMS ratio.

You're probably not going to hear it for two reasons. First, the noise is still -96 dB down from the peak so the only time you might hear it is if you crank-up the volume during the quiet part. and then you'll push your amp into distortion or you blow-out your speakers (or your ears) when the full-volume comes-along. Second, you've still got 48dB of masking (drowning-out).

...If you have a fade-to silence you get down to only a few bits of resolution but you don't hear any loss of quality because the quantization remains low and it gets to the point where you can't judge sound quality at low levels and then to the point where you can't hear anything.

 

[Blumlein 88]

my point was there's a real issue with what you know, aside from what you can actually physically hear. Once the idea is in your head...

Seems that de-emphasis applies -4.5dB at 5Khz rising to -9.5dB at 20Khz so I agree that will be an audible difference if not applied but I don't agree that it will necessarily sound bad, especially not to someone my age - might actually correct my HF loss!

But HDCD, decoded or not is not audible. Pre-emphasis un-decoded certainly is. Whether it helps this or that song or this or that person's hearing it is quite audibly noticeable.

 

[darmok]

"No fair" comparing the 96dB peak ratio to the 48dB RMS ratio.

You're probably not going to hear it for two reasons. First, the noise is still -96 dB down from the peak so the only time you might hear it is if you crank-up the volume during the quiet part. and then you'll push your amp into distortion or you blow-out your speakers (or your ears) when the full-volume comes-along. Second, you've still got 48dB of masking (drowning-out).

Yeah, that was a bit sloppy, but cranking up the volume during the quiet part is what I was getting at. The -48dB RMS measurement was over a two minute span and it's not a fade but a very quiet cadenza. It's absolutely reasonable to want to turn that up a bit knowing that you'll have to turn it down later. Even the loudest sections of the whole concerto are -18dB or so, with transient peaks that are clipping for a handful of samples. Again, it's absolutely reasonable to want to listen to the quiet parts at 60-70dB knowing that you're going to get blasted at 90-100dB for 10 or 15 seconds in the loud parts, or just to turn it up for a few seconds to catch a detail. None of this is going to blow out my amplifier either... getting the quiet part to 75dB RMS requires just 2V into something like Focal Elegias. The HE6se and Aeons are a different matter!

As far as the masking goes, at some point the degradation of SNR absolutely affects sound quality and I'd say this is well into that point. It's a cello concerto and during the quiet cadenzas there's virtually no signal below 1KHz, meaning you're trying to mask noise that's right in the audible peak with signal below 300Hz. As a sanity check, with the amplifier set to 2Vrms out and applying 48dB of attenuation to the signal, I can still hear the cadenza very quietly. Remove that attenuation and the hiss is clearly audible in the background.

This is precisely why I'm bringing up noise shaping. The noise shaped CDs I have sound significantly better than the non-noise-shaped counterparts and the ability to maximize the perceptual SNR for very high dynamic range recordings is no doubt the reason. This particular recording that I'm picking on is from 1990 and was likely recorded with a 16-bit ADC in the first place, so there's not much that can be done about it now. Newer recordings don't have that excuse.

 

[darmok]

BTW, lest you think I'm just picking on this recording, the most recent recording I have of this concerto is from 2019 and I'd be shocked if it was recorded at less than 24-bit. It's louder than the older BIS recording (-33dB for the first movement) and ambient noise from the venue is more clearly audible, but the hiss is also still clearly audible in the quiet cadenzas. Why is this still a problem on a CD released in the last few years?

 

[Ifrit]

Why is this still a problem on a CD released in the last few years?

Because the microphones, mic preamps, and ADC have analog circuitry that possess some levels of noise, and they usually use a lot of mics to record nowadays, so the noise levels multiply. The venue has air moving, and it is noisy. The lights at the venue have interference with recording equipment, and that adds more noise.
Unlikely the noise you hear is quantization, most likely digitized analog noise.
There isn't absolutely quiet recording out there, no matter noise shaped, dithered, undithered, "self-dithered".