CD audio is not good enough, CD Standard is bad quality
CD audio is not good enough, CD Standard is bad quality
May 1 2003, 04:29
Group: Members (Donating)
Joined: 7-April 03
From: Newark, CA
Member No.: 5871
Sound quality is a complex subject, and one that has been thrashed
out elsewhere time after time. Nonetheless I'll give it a bit of a
spin here, without getting all technical, in order to justify why I
think CD audio is not good enough.
When I complain about CD sound I am not doing so as some sort of
retrograde vinyl lover who can't change with the times. :-) I am
simply saying that the sound of the CD I am listening to has audible
problems and does not match what I expect the creators wanted. This
can be measured by how closely the CD replicates the master. Of
course in most cases I don't get to hear the master, so much is
guesswork. However, as a trained audio engineer I have *some* idea of
what is expected and can certainly compare CDs to the masters I
CDs improved on vinyl in many ways, notably in reduced noise floor,
phase artifacts, and crosstalk; ease of handling; and accurate
handling of low frequency stereo information. But CDs are inferior to
vinyl in frequency response and degradation characteristics. If a CD
gets a scratch you hear unlistenable white noise; if a record gets a
scratch you hear a DJ. ;-)
Going further, some may "prefer" the sound of vinyl precisely because
of the distortions it introduces. These include rounded signal peaks
and second-order harmonics, as well as the aforementioned phase
issues. All of these introduce a "warm" sound that is palatable to
many. Whether I like that sound or not, I prefer to hear what the
artist intended. If they wanted harmonics they could have used a
tube. And so on.
A bit more is in order about error correction. Most errors are
corrected by CD players, but this can produce tiny glitches of noise
that most people do not notice. I notice them. It's not that I have
better ears; once I point them out you can hear them as well. Of
course, the better the music reproduction system the more noticable
these are. (Though contrary to this, the better the CD player error
correction, the less you'll hear.) For most people with crappy
stereos it's not an issue.
I do not think that there is anything inherently wrong with digital
sound encoding, only that the 44.1KHz sampling rate and 16 bits per
sample are not good enough. Currently, studios use 96KHz and 24 (or
32) bits throughout the recording chain process, and must reduce this
down to consumer standards for replication. There's probably a good
reason why those people most highly trained in listening don't think
CD quality is good enough for recording. It's simply because their
ears tell them so.
You may be interested to know that the current CD standard was a
matter of much compromise between the American, European, and
Japanese manufacturers. I can remember reading some of the research
articles at the time (I was in university). The Japanese insisted
that 100KHz and 24-bit (if memory serves on the exact numbers) were
required for accurate reproduction. But the others argued that no-one
would hear the difference and it would reduce cost and time to market
if the lower standard was adopted. And so, unfortunately, it was.
Another big problem with many CDs is the terrible job of mastering.
Back in vinyl days you really had to know what you were doing to
adjust the master tape to the deficiencies of the medium. There were
relatively few mastering engineers, but they knew their job. Today
almost anyone thinks they can master, and so they do... badly.
So the problems with CD can be summarised as: insufficient frequency
response, insufficient resolution, poor mastering, nasty error
characteritics, and cases that break all the time. ;-)
MP3s inherit all of these except the bit about the cases.
May 1 2003, 18:06
Joined: 27-March 02
From: California, USA
Member No.: 1631
QUOTE (Joseph @ May 1 2003 - 07:32 AM)
> Generally speaking, humans can perceive frequencies around 20kHz.
The Nyquist numbers are pure theory and do not take into account implementation. For example, Nyquist requires a perfect low-pass filter for the digital-to-analogue conversion. Well, such a thing does not exist! Real filters are not perfect, but rather introduce
frequency, aliasing, and phase anomolies. Though techniques like oversampling can help, a sampling rate of 96KHz is the perfect solution, as it puts all of these distortions above the range of human hearing.
> I have also heard that 16bit resolution has a noise floor lower than that of a silent, empty recording studio. If this is true (sorry, I don't have hard evidence) then it is proof that 16bit is more than enough.
It is not true. The range of sounds from absolutely quiet (achieved only in an anechoic chamber) to hearing damage is 150dB. A more reasonable range to reproduce (from a recording studio to a loud
concert) is 130dB. 16-bit recordings reproduce a dynamic range of only 96dB, whereas 24-bit recordings reproduce 120dB. 16-bit looks rather limiting, doesn't it?
Here's another view: in music we want to listen to (not the overcompressed crap) the peaks are much higher than the average volume. In order to provide room for these peaks, most of the musical information must be restricted to about half of the available bits.
It follows that to avoid compromising the signal, we need a lot more dynamic range than 16-bit provides.
Any decent turntable can play any decently maintained vinyl record with an almost complete lack of background noise. Heck, even my mid-range Linn does a fine job. Traditional comments like these about bad vinyl quality come from people who have never heard a decent hi-fi in the first place.
> CDs are obviously digital recordings ... to the
> CD player, that means "no matter what I read, it's supposed to be either a 1 or a 0" ... introduce reading noise into a digital wave, and you still get a
> digital wave -- the player can still read the original sound through all the noise, because it can assume what is supposed to be read.
Wow, this is so wrong. If I am trying to read 10010011101001 and
there is a scratch and I get 10000000000000 then how exactly am I
supposed to recover the original?
True, Red Book audio uses interpolation to add redundancy to the
signal, but a big enough scratch and all is lost.
> Also, imperfect error correction in CD players won't
> into the result
Um, yes it will. It is clearly audible.
> ... at worst, it will read a 0 instead of a 1, which will result in a tiny pop that lasts about 1/(16 x 44100) of a second. Your brain cannot > perceive this.
Not even if there are thousands of them in a row?
There are other problems with CD reproduction as well, like jitter and single-bit distortion. The great thing about 96/24 recording is that many of the challenges of 44/16 go away. There is no need for dither, brick-wall filters, etc. Reproduction equipment can actually be simpler and yet achieve sonic excellence.
Need to add some facts to help clear up a lot of misunderstanding:
1) Nyquist does not require a perfect LPF. I have personally tested old, inexpensive (Technics) non-oversampling CD players with steep analog LPFs that can reproduce perfect 20kHz sinewaves with <0.01% distortion and reasonable phase shift. 96kHz is not necessary to reproduce high frequencies up to 20kHz accurately.
2) 24 bit dynamic range is 144dB, not 120dB (dynamic range is 6dB x # of bits). Of course you will not find A/D or D/A converters systems with more than 115dB dynamic range, so nobody has ever really recorded a true 24 bit signal that I am aware of. Are there microphone amplifiers with >144dB dynamic range? I suspect not.
In addition, proper mastering of a CD should include noise shaping which can increase the effective midrange dynamic range of a CD by 8 to 12dB (or more), giving an effective dynamic range of over 110dB from redbook CD. Not only that, but a properly noise shaped and dithered digital signal can reproduce audible signals below the noise floor (because the noise is not correlated with the signal it can be perceived), effectively increasing the resolution slightly further.
Assuming a well mastered CD with a dynamic range of 110dB, most recording chains and certianly most consumer playback gear could not reproduce this signal accurately enough to warrant even greater bit depth. For those that can reproduce it, a clean 110dB dynamic range should sound quite good without the need to transition to a completely different audio format.
16 bit does not look very limited for consumer distribution, does it?
3) LP playback with an almost complete lack of background noise? Yes, I've heard extremely high end vinyl reproduction and it sounds very nice, but complete lack of background noise it is not. Even extremely good phono gear can not typically achieve more than 80dB s/n ratio, and good gear in the 70dB range is not uncommon. All phono gear has a pretty steep high pass filter to remove strong low frequency resonance of the tonearm / cartridge system and turntable bearing rumble which could damage your equipment at worst and dramatically increase distortion at best. LPs are relatively noisy compared to good 16 bit digital audio.
4) There are several layers of error correction in the CD format, and uncorrectable errors are actually not very common (see posts above on the topic, they are accurate). Yes, a scratch can result in unrecoverable errors just like it does on an LP. But a reasonably well maintained CD should have no unrecoverable errors. The CD format uses eight to fourteen modulation and Reed-Solomon error correction systems and then uses interpolation and finally muting as last resorts. The system is very robust, and if you commonly enounter audible errors than I suggest you replace damaged CDs or invest in a good quality CD player. You can purchase calibrated test CDs that can measure the error correction capability of a CD player. My Philips CD player can fully correct (no errors, data fully recovered, no noise introduced) a 1.2mm scratch - that should be a rare level of damage, anything less is a non issue. In practice my experience is that error correction is not an issue for CD, so I find your claims to be unusual and suspect.
5) Jitter is time based errors - samples converted to the correct amplitude at the slightly wrong time. While jitter during recording can not be prevented, it can be eliminated during playback with a good buffer or PLL system. Jitter has been shown to be a non-issue (creating distortion well below the 16 bit dynamic range)with well designed playback systems. As a matter of fact, timing errors become more significant as the sampling rate increases so a 96kHz 24 bit system requires jitter several orders of magnitude below that required for a 44/16 system to prevent compromising the 24 bit dynamic range. And designing equipment for 144dB dynamic range is not simpler than designing for 110dB. I'm not sure what you mean by single-bit distortion, but if that refers to errors at the LSB level, then see item #4.
Bottom line - 96/24 may be a good tool for the studio but I do not see any evidence that it is beneficial to the consumer.
Was that a 1 or a 0?
|Lo-Fi Version||Time is now: 3rd September 2014 - 01:05|