• The move to the new server is done. There are some software and database maintenance updates in process. This has us passing the hat around to help out. We appreciate any donations. Seriously, even a dollar helps. The payment page may be found here - https://www.audiokarma.org/support.html

Please explain this criticism of CD audio

A

AintBaroque

Active Member
OK, so the guy at this site has a lot of info on linear turntables and other sound equipment, then I read this page:

http://www.soundfountain.com/amb/cd25years.html

and the author says this about number of samples per second and resolution of the sound as recorded on CDs (emphasis mine):

"The number of samples in the audio band is frequency dependent. The lowest frequencies are being measured many times more than the top frequencies. The frequencies from about 11.000 Hz. to 20.000 Hz. are sampled only 2 times.

There is another important trait of the CD format. The highest resolution of the sound is attained at the loudest level, that is at 0 dB recording level. 0 dB gives also the least distortion. At lower levels, the resolution is less and is not in line with that of the 0 dB level. The lower the level, the less the resolution. "

Is he correct - are higher frequencies only sampled a few times per second? And only at max loudness is the highest resolution achieved?

That means that quiet, higher frequency passages recorded to CDs will not sound very good.
 
Register to hide this ad
The author is displaying his ignorance of how digital audio works and/or his prejudice in favor of the more easily understood analog system. The sampling theorem teaches that a band-limited signal may be sampled, digitally encoded, and reconstructed without loss from the digital code as long as the sampling frequency is at least twice the highest frequency in the original signal. The word length used for each sample determines the signal-to-noise ratio. The CD system's 16-bit linear encoding puts the quantization noise 96 dB below the maximum signal level, which is more than 20 dB better than the performance of a first-rate phono system. Applying dither during production slightly reduces the signal-to-noise ratio, but it improves the sound by decorrelating the noise from the signal. Post-dither, the noise floor sounds like the white noise that you'd get from an extremely quiet analog tape deck. The CD system's sampling frequency of 44.1 kHz provides frequency response to just below 22.05 kHz, which exceeds the generally accepted upper limit of the human auditory system.

If you're up for some advanced math, Wikipedia's article on the sampling theorem provides a good background. Here's the link: http://en.wikipedia.org/wiki/Sampling_theorem.
 
Late

Dr Tinear said:
The author is displaying his ignorance of how digital audio works and/or his prejudice in favor of the more easily understood analog system. The sampling theorem teaches that a band-limited signal may be sampled, digitally encoded, and reconstructed without loss from the digital code as long as the sampling frequency is at least twice the highest frequency in the original signal. The word length used for each sample determines the signal-to-noise ratio. The CD system's 16-bit linear encoding puts the quantization noise 96 dB below the maximum signal level, which is more than 20 dB better than the performance of a first-rate phono system. Applying dither during production slightly reduces the signal-to-noise ratio, but it improves the sound by decorrelating the noise from the signal. Post-dither, the noise floor sounds like the white noise that you'd get from an extremely quiet analog tape deck. The CD system's sampling frequency of 44.1 kHz provides frequency response to just below 22.05 kHz, which exceeds the generally accepted upper limit of the human auditory system.

If you're up for some advanced math, Wikipedia's article on the sampling theorem provides a good background. Here's the link: http://en.wikipedia.org/wiki/Sampling_theorem.
Click to expand...

This is exactly what I was thinking too. I was too late.
 
kinda sorta

It is true that the higher the recorded frequency, the fewer samples for each cycle of that frequency. If the sampling rate is 44.1kHz, that would get you two samples per cycle for a recorded frequency of 22.05kHz.

For the recorded frequency of 11.025kHz, you'd get four samples per cycle or period. For a recorded frequency of 100Hz, you'd have 441 samples for that period.

Less resolution? Well the resolution is the step size coming out of the DAC. So, if the DAC puts out 2V rms, the 16 bit resolution equates to 2^16=
65536 steps or +/- 32768. That's like 61uV per bit. If you do look at that error voltage relative to the full scale output vs. a -20dB FS level, then yes, the distortion ratio is better, but the distortion performance at -20dB FS and lower is still very good. I remember that Sony put out a audio test CD and the -60, 70, and 80dB FS performance was still very good.
 
dusty42701 said:
I started researching this elsewhere and found that the dynamic range constitutes how many bits that the audio signal actually uses.
I was actually researching what "dither" was and found this article that makes better sense of the volume-to-resolution subject.
http://www.earlevel.com/main/1996/10/20/what-is-dither/
Click to expand...

I understand how dithering helps with repetitive signals, but I don't see how it can benefit non-repetitive signals (like music). Can someone explain how dithering affects non-repetitive signals? Every explaination I've read is based on averaging several samples of the signal while random noise is applied, which I can certainly see would improve the resolution. I've used this technique often in DAC hardware/software. But I've never seen a good descritption of how non-repetitive signals can be improved by dithering. If someone has a lucid description of it, I'd love to read it.

Terry
 
well music signals do have repetition to them.

Dither is there for low level signals. If the signal's amplitude is only a couple of bit 'high', you don't want the music to be modulationg the error (1/2 LSB). Dither is summed with the signal and is just background noise. The background noise is random and mixes with the music signal to make it a little higher at times and a little lower at times. The summed signal randomizes these 1/2 LSB errors and makes the low level signals fade smoothly with no chance having a fade with a grainy quality.

These low level signals are pretty much in the dirt. I, personally, have a hard time picking out a dithered recording amongst an undithered recording. And these are 16 bit recordings. I think for larger bit depths, dithering is not necessary.
 
Jon_Logan said:
well music signals do have repetition to them.

Dither is there for low level signals. If the signal's amplitude is only a couple of bit 'high', you don't want the music to be modulationg the error (1/2 LSB). Dither is summed with the signal and is just background noise. The background noise is random and mixes with the music signal to make it a little higher at times and a little lower at times. The summed signal randomizes these 1/2 LSB errors and makes the low level signals fade smoothly with no chance having a fade with a grainy quality.

These low level signals are pretty much in the dirt. I, personally, have a hard time picking out a dithered recording amongst an undithered recording. And these are 16 bit recordings. I think for larger bit depths, dithering is not necessary.
Click to expand...


I agree that dithering is not necessary for higher resolution systems, and I don't want to degrade this discussion into one about how much resolution is enough.
But it seems to me (unless I am missing something) that your description still requires that the same music signal is sampled several times with random noise mixed in and averaged to derive the "higher resolution" signal after dithering. But this means that the signal does not appreciably change for several samples. Is that a valid assumption with music?
I suppose there is some statistical analysis that shows that dithering will help with non-repetitive signals, but I've never seen it. I've only seen the analysis applied to repetitive signals.

Terry
 
The byline in the title sums it up: "hobbyists's views for hobbyists". He needs to refrain from explaining things beyond his grasp. I started to site examples, but there are too many to choose from.

It appears he is a serious listener to jazz and classical and enjoys that music on LP. I'm glad he is enjoying it. I own thousands of LP's and hundreds of CD's. I enjoy both. My quiet, higher frequency passages on my classical CD's sound just fine, especially without the inherent surface noise and clicks and pops that appear on even the most meticulously handled LP's.
 
Last edited:
I suppose there is some statistical analysis that shows that dithering will help with non-repetitive signals, but I've never seen it. I've only seen the analysis applied to repetitive signals
Click to expand...

I think the part that's being missed is that other forms of noise -- microphone hiss and such -- are far above the level that would be affected by sampling error.

Let's be generous and say that a good vinyl disk has a noise floor of -60 dB, and being generous, let's say you can detect music ten or twenty dB below that.

A CD has all that headroom and almost another 20 dB.
 
TerryS said:
I understand how dithering helps with repetitive signals, but I don't see how it can benefit non-repetitive signals (like music). Can someone explain how dithering affects non-repetitive signals? Every explaination I've read is based on averaging several samples of the signal while random noise is applied, which I can certainly see would improve the resolution. I've used this technique often in DAC hardware/software. But I've never seen a good descritption of how non-repetitive signals can be improved by dithering. If someone has a lucid description of it, I'd love to read it.

Terry
Click to expand...

Think of dither as raising the noise floor so that white noise, which is uncorrelated with the music, buries the quantization noise that is correlated with the music. This isn't a perfect description of how dither works, but it has the advantage of being intuitive.
 
There must be some similarity between dithering and anti-aliasing in displaying graphic images. It involves lowering the resolution, which is conceptually similar to adding noise.

In CD audio, the question, is the added noise perceptible. Would it ever raise the noise to a level comparable to that of tape or vinyl?
 
The dither added is summed with the desired signal, then sampled. So noise is added (or subtracted) from the music on a sample by sample basis. If there was a music signal that was sampled at a particular instant and it was A to D'd at a 3 bit amplitude, but 1 bit amplitude dither was added, that same sample might be 2 bits of amplitude or 4 bits of amplitude or the original 3 bits of amplitude. These varaiations in amplitude works just like analog noise. At the next sample period, the just music sample might be increased by a bit, or decreased by a bit, or left untouched. The dither is random If you subtracted the original music from a dithered recording and just listened to the dither output, you'd see that is does sound like white noise.

Dither is not there to reduce resolution, although I could see how some might think so. Dither is there to reduce the 'edgieness that is heard when only a few bits are used to to digitize a very low signal. Human ears are more sensitive to this 'granualar' kind of sound than the tiny amount of noise (dither) that is added to A to D conversions.

Anti-Aliasing is something different. Anti-Aliasing is a low-pass filter that is used to prevent signal higher than half the sampling frequency (Nyquist frequency) from being converted. If you try to digitize a frequency that is higher than the Nyquist frequency, you will get less than one sample per period. The resulting frequency will be an 'invented' new frequency. An aliased frequency.
 
I know it's different, but it is perceptually similar. You say dither removes edginess from audio, and anti-aliasing removes jaggies from images. Both functions perceptually smooth.
 
js1138 said:
I know it's different, but it is perceptually similar. You say dither removes edginess from audio, and anti-aliasing removes jaggies from images. Both functions perceptually smooth.
Click to expand...


If you want to use a video compression example, I would say that dither might breakup the sameness of a large field of one color, giving it a more natural, less cartoon-like look, while anti-aliasing might prevent a new color from being generated by an input color being out of range.
 
You must log in or register to reply here.
Back
Top Bottom