I have not found this issue addressed anywhere. Does the Red Book standard specify how a CD player should behave when it is given a signal that has large amplitude content in the 20 - 22.05 kHz range? It would never have occurred to me to pose this question until I read that the SACD standard restricts the amplitude of the audio signal to 1/2 full scale and specifies the noise-shaped frequency spectrum to some extent -- that is, there are PDM signals that are not compliant as SACD encodings. I'm wondering if there is anything comparable for CD's. Most of the time, I've seen CD format specs that list 20 - 20000 Hz for the frequency response. This suggests that one of the following may be true:
16/44.1 PCM signals that have large amplitude content over 20 kHz are not compliant with the CD standard, and the behavior of a CD player when faced with such an input is undefined;
a proper CD player must reproduce musical content in the 20 - 20000 Hz range within a few decibels, but has great leeway in how to handle content outside that band;
a CD player is expected to filter out most content outside the audio band; or
the standards are silent on this point.
I would like to know which is the case. The question interests me largely because theoretical discussions and tests of ADC's and DAC's often make use of test signals that contain nearly full-scale content at or near the Nyquist frequency, for example square waves, and I don't know if criticisms based on these test signals can be considered relevent to anything of importance.
First, all data needs to be limited in frequency to 1/2 the Nyquist limit or aliasing occurs. These filters are a normal part of a DAC, whether said DAC is part of a CD player or not. In practice, the type and extent of output filtering differs considerably. My soundcards record input and play back output all the way to the Nyquist limit, but I've seen claims that some DACs filter steeply above 20kHz.
The CD data is not constrained to any particular signal strength limitations because of this. I pretty sure that anything that can be encoded in 16 bits/44.1kHz is within the standard. However, hardware implementation has many variables. Some DACs do not behave properly with samples at or very near 0dB (of any and every frequency) but others have no problems at all with the same data. In some circumstances the analogue produced by digital data can be well above the value indicated. If those sample values are near enough to 0dB, the output can be above 0dB (up to about +6dB). Some DACS have adequate headroom to handle this, some will clip.
I have not found this issue addressed anywhere. Does the Red Book standard specify how a CD player should behave when it is given a signal that has large amplitude content in the 20 - 22.05 kHz range? It would never have occurred to me to pose this question until I read that the SACD standard restricts the amplitude of the audio signal to 1/2 full scale and specifies the noise-shaped frequency spectrum to some extent -- that is, there are PDM signals that are not compliant as SACD encodings.
That's because it's a stupid (IMO) non-linear (and potentially unstable) system which can go into self oscillation if those signal limits are exceeded.
There is no similar feature in CD, thankfully!
Cheers,
David.
I am not a technical person, but I want to know if different CD players will produce different lossless files from the same Audio CD input. (considering lossless files generated by same sw with same setup options).
Is my question correct or I am missing something?
Georges
16/44.1 PCM signals that have large amplitude content over 20 kHz are not compliant with the CD standard, and the behavior of a CD player when faced with such an input is undefined;
a proper CD player must reproduce musical content in the 20 - 20000 Hz range within a few decibels, but has great leeway in how to handle content outside that band;
a CD player is expected to filter out most content outside the audio band; or
the standards are silent on this point.
I don't know which of the above statements is the good one. I would vote for the 4th one, but this is just my feeling.
After all, digital-to-analog conversion is something that exists outside of the Compact Disc domain. Before the Compact Disc, ADC and DACs were already used in records manufacturing in order to replace the clusmy tape recorders with digital delay lines in order to drive the cutting head.
From a digital point of view, the CD standard allows you to write any PCM data. But in practice, a CD player will filter out DC offset. It is also possible to generate artificial wav files that are nearly un-burnable on CDR, like this one : http://perso.numericable.fr/laguill2/files/WeaksectorWav.zip (http://perso.numericable.fr/laguill2/files/WeaksectorWav.zip)
I have not found this issue addressed anywhere. Does the Red Book standard specify how a CD player should behave when it is given a signal that has large amplitude content in the 20 - 22.05 kHz range? It would never have occurred to me to pose this question until I read that the SACD standard restricts the amplitude of the audio signal to 1/2 full scale and specifies the noise-shaped frequency spectrum to some extent -- that is, there are PDM signals that are not compliant as SACD encodings.
That's because it's a stupid (IMO) non-linear (and potentially unstable) system which can go into self oscillation if those signal limits are exceeded.
There is no similar feature in CD, thankfully!
Cheers,
David.
I am interested in learning more about DSD encoding at a theoretical level. Are there any good resources you could recommend?
I am not a technical person, but I want to know if different CD players will produce different lossless files from the same Audio CD input. (considering lossless files generated by same sw with same setup options).
Assuming the readers are working correctly and the CD is undamaged, you should get identical results.
Pio2001: why is that WAV nearly unburnable?
Pio2001: why is that WAV nearly unburnable?
This is a wav file into which I inserted some so-called "weak sectors". These parts are not burned in an optimal way, and are most of the time unreadable. They produce read errors.
Weak sectors are data that usually confuse EFM encoders, that do not always optimize the DSV (balance between pits and lands), and produce too much pits or too much lands for optimal optical tracking. They are used in CD ROM copy protection, the manufacturing plant being capable of mastering them without problem.
The ones that I used here are strong ones, whose Pit/land balance can't be otpimized at all, even in professional mastering. The resulting CDR is at the limit of red book specification, because the pit/land imbalance reaches the maximum allowed value, and thus is extremely difficult to read.
This is my understanding of weak sectors. Feel free to correct me if i said something wrong.
I am interested in learning more about DSD encoding at a theoretical level. Are there any good resources you could recommend?
Some years ago I tried to get infos about that and to simulate PCM->DSD and DSD->PCM conversion processes. The important things I can remember are (sorry, no references):
- From an abstract DAC-perspective DSD and PCM are very similar (Convert sample values to voltages and apply a lowpass filter). But the idea of DSD is to trade high sample resolutions for very high sampling rates (1 bit/sample and 44100*64 samples/sec actually *). "PCM" is usually used to describe formats with lower sampling rates and higher bit/sample rates.
- When you want to create 1bit-signals it's very hard (likely impossible) to achieve all of the following 3 desirable system properties at the same time: (1) low noise power in the audible frequency range (2) linearity and (3) stability (no overload)
- Regarding the noise floor one tries to use noise shaping filters (you can do this for PCM as well). Due to the very low sample resolution (1bit/sample) the filter should have a rejection of at least 100 dB in the audible band. Special care has to be taken when designing these filters: You need to limit the maximum gain to 1.5 (3.52 dB) so one can still guarantee stability (assuming the folloring point 4 is also met). Together with the Gerzon/Craven "noise shaping theorem" this implies a narrow transition band which makes filter design difficult.
- The signal amplitude should not go beyond a certain fraction of full scale (like the OP said -- though I'm not sure about the actual faction anymore ... it may be 1/2) to keep the system from overloading
- Adding a little bit of dither helps reducing possible nonlinear distortions a little. Unfortunately "full" dithering can't be done because the system would overload.
*) "Wide-DSD" refers to 44100*64 samples/s and 8 bits/sample which could also be called "PCM at a very high samplinrate".
SG
Edit: Actually, there
is a paper I can link to. But this one doesn't deal directly with the additional requirements for the noise shaping filters and the signal level. However, it's a good read and it has very good arguments
against DSD:
Why 1-Bit Sigma-Delta Conversion is Unsuitable for High-Quality Applications (http://sjeng.org/ftp/SACD.pdf) by Lipshitz and Vanderkooy
You may also need to google the net / search this forum for
dithering and
noise shaping.
If I remember well, it is recommended that CD players filter out past 20kHz content, but i's not mandatory.
From a digital point of view, the CD standard allows you to write any PCM data. But in practice, a CD player will filter out DC offset. It is also possible to generate artificial wav files that are nearly un-burnable on CDR, like this one : http://perso.numericable.fr/laguill2/files/WeaksectorWav.zip (http://perso.numericable.fr/laguill2/files/WeaksectorWav.zip)
OT, but have you noticed low compression rations FLAC, APE and RAR have on this file, compared with simple ZIP routine?
Well, they're optimised for audio signals, which Pio's file probably does not much resemble; Zip instead is a general data compression.
Well, it's pure sine signal with some audio drops.
They should compress well, I think.
Yes, the sine part is highly compressible using Winzip. It detects the repetition and encodes it using only one symbol.
I'll never forget how I once compressed a wav with several minutes of silence.
Toto.wav -> Toto.zip : 99% compression ratio
Toto.zip -> Toto.zip.zip : 99% compression ratio
I added a few seconds of silence to the beginning of the signal. I pulled up the second sample of the silence to make a positive reference point. I wrote to CD-RW. I extracted from the CD. There was something funny happening. The extraction started out at speed, slowed down briefly, then speeded up again. However, the extraction was bit identical with the source.
Of course this might not be the result if tried repeatedly, but that possibly isn't interesting enough to make me find out.
Interesting. What burner did you use? The Plextor Premium and 716A could not produce a bit-identical copy of this file.
Here, the easy part is that the weak sectors are at constant level, and the read errors are interpolated. So the erroneous parts are always correct. But in spite of this, these two Plextors produced errors in the sinewave at the neighborhood of the weak parts.
This writer says PlexWriter 40/12/40A on the front of the drawer.