Suppose I want to play 1W average in a speaker of say 1W/88dB, which would reach my ears at maybe 3dB less, 85dB SPL.
For the amp not to clip, one could consider Peak to Average Power Ratio of music to be in worst case 16dB.
This would be the hard limit of before clipping.
More typical PAPR may be 12dB, but a couple of db margin would not hurt.
16db peaks would then hit the 40W level.
In an 8 ohm speaker this would be 50.6 v peak to peak. This would translate the 16 bit resolution (1:65535) an LSB level of 772uV.
At one Watt 2.82Vrms or 7.97V p-p 10,323 LSB's are used, or roughly 15.7% of the full scale range. If the musics quiet passages are 20dB lower, or 10mW, 0.282V rms, 797mV p-p, it uses 1,032 LSB's, which should be similar to 10-bit resolution. The weaker the signal, the worse it would sound.
The music material in this example has 36dB dynamic range, which may not be extreme.
For the weakest signals to have sufficient resolution much more than 16 bits is needed.
The Red Book spec seems a bit out of step with the times.
These guys thinks 90dB is more like it, which would make the case even worse for 16 bits.
https://www.sweetwater.com/insync/dynamic-range/
For the amp not to clip, one could consider Peak to Average Power Ratio of music to be in worst case 16dB.
This would be the hard limit of before clipping.
More typical PAPR may be 12dB, but a couple of db margin would not hurt.
16db peaks would then hit the 40W level.
In an 8 ohm speaker this would be 50.6 v peak to peak. This would translate the 16 bit resolution (1:65535) an LSB level of 772uV.
At one Watt 2.82Vrms or 7.97V p-p 10,323 LSB's are used, or roughly 15.7% of the full scale range. If the musics quiet passages are 20dB lower, or 10mW, 0.282V rms, 797mV p-p, it uses 1,032 LSB's, which should be similar to 10-bit resolution. The weaker the signal, the worse it would sound.
The music material in this example has 36dB dynamic range, which may not be extreme.
For the weakest signals to have sufficient resolution much more than 16 bits is needed.
The Red Book spec seems a bit out of step with the times.
These guys thinks 90dB is more like it, which would make the case even worse for 16 bits.
https://www.sweetwater.com/insync/dynamic-range/
The original source surely has dithering/noise-shaping before digitization/reduction to 16 bits, else it could be considered faulty. Yes, many of the early CDs could be considered faulty.
But for the most part (with "well made" sources), 16 bits is not nearly as bad as you might think.
This app note uses "straight" full-bandwidth noise, no "noise shaping" to (for audio purposes) move the noise out of the ear's most noise-sensitive range (which can add another 10 or 20dB to the perceived noise floor), but it demonstrates the basic principle:
"... only the fundamentals remain."
http://www.ti.com.cn/cn/lit/an/snoa232/snoa232.pdf
But for the most part (with "well made" sources), 16 bits is not nearly as bad as you might think.
This app note uses "straight" full-bandwidth noise, no "noise shaping" to (for audio purposes) move the noise out of the ear's most noise-sensitive range (which can add another 10 or 20dB to the perceived noise floor), but it demonstrates the basic principle:
"... only the fundamentals remain."
http://www.ti.com.cn/cn/lit/an/snoa232/snoa232.pdf
16/44 is a bit out of step with times, although it can be surprisingly good with careful mastering and the very best dacs. Problem is the there have been multiple attempts to introduce higher resolution formats that have failed in the marketplace. Why? Because record companies got greedy and wanted to charge a steep premium for increased resolution and institute strong copy protection at the same time. The desire for copy protection was understandable from the record company and artist perspective, but not a direction consumers were willing to move. They didn't want to pay extra to have their music on their various playback devices. It is a bit more complicated than what I said, but those were some factors. Besides all that, it turns out consumers are willing to forego sound quality for convenience, as we see from the wide acceptance of lossy compression. Anything else?
Last edited:
As a final delivery format 16 bit is plenty. As a working format 24 is useful, 32 bit float maybe more so.
A dynamic CD recording will have an RMS value about 22dB below full scale. A very dynamic recording about 30 dB. As mentioned above, it's not as bad as you think.
A dynamic CD recording will have an RMS value about 22dB below full scale. A very dynamic recording about 30 dB. As mentioned above, it's not as bad as you think.
not so bad
Music doesn't mean single sinewave. It has many frequencies. Large spectrum can become a "parent" for small signal, where 16-bit resolution isn't the absolute maximum. As long as your mastering is under 96kHz/24bit, a final delivery in 44k/16bit has more resolution than 16-bit even without dithering.
The attached pics are an example. Pic 1 is the original 44kHz/24bit. Pic 2 is truncation to 16bit with dithering. Pic 3 is simple truncation(without dither) to 16bit. Pic 3 has quantization noise but still has the original -110dBFS signal. If you don't have 1kHz@-20dBFS, -110dBFS signal of course fades away. I would say there are many "parents" who support small signal in real music.
Music doesn't mean single sinewave. It has many frequencies. Large spectrum can become a "parent" for small signal, where 16-bit resolution isn't the absolute maximum. As long as your mastering is under 96kHz/24bit, a final delivery in 44k/16bit has more resolution than 16-bit even without dithering.
The attached pics are an example. Pic 1 is the original 44kHz/24bit. Pic 2 is truncation to 16bit with dithering. Pic 3 is simple truncation(without dither) to 16bit. Pic 3 has quantization noise but still has the original -110dBFS signal. If you don't have 1kHz@-20dBFS, -110dBFS signal of course fades away. I would say there are many "parents" who support small signal in real music.
Attachments
In your example and assuming triangular dither with no noise shaping, the noise floor is 57.26 dB below the weakest sound in the recording (20*log10(1032)+10*log10(3/2)-10*log10(3)) and more than 93 dB below the loudest sound. What's so bad about that?
What's bad is it is easy for anyone mastering even 24/44 down to CD, that there is always some audible loss of low level details when going to 16/44, even with proper dither. One can work backwards from that to try to figure out with some numbers why it sounds worse, but the problem trying to do so is that old hearing research for 95% percent of the population using ABX test methods fails to make clear what almost anyone can hear given the opportunity. It is always sad to have to make a CD that sounds worse than the master, but they pretty much all do (maybe some exceptions could happen I guess, but sure wish it wasn't necessary).
Last edited:
Mark, you know you should stop quoting the '95%' thing. It's not a real number just some hand waving. 😛
I have a real hard time believing that going from 24 to 16 bit with proper dither loses anything given the low dynamic range of most recordings and the high noise floor of a domestic environment*. Given we have HD tracks now there must be some examples of recordings out there released on 24/48 and 16/44 where this can be shown to all. It would make a perfect session for the next burning amp.
*Note to all who have not seen the banter between myself and Mark I am not saying it's impossible, just I cannot get my head around it. Dithered 16 bit has a noise floor 120dB down which is below 0dBa for the levels most of us listen at.
I have a real hard time believing that going from 24 to 16 bit with proper dither loses anything given the low dynamic range of most recordings and the high noise floor of a domestic environment*. Given we have HD tracks now there must be some examples of recordings out there released on 24/48 and 16/44 where this can be shown to all. It would make a perfect session for the next burning amp.
*Note to all who have not seen the banter between myself and Mark I am not saying it's impossible, just I cannot get my head around it. Dithered 16 bit has a noise floor 120dB down which is below 0dBa for the levels most of us listen at.
From a technical point of view it is a pity to unnecessarily lose resolution, but I for one never heard the noise of a CD player. Even if I did, I doubt it would bother me if it's at such a low level. Of course that assumes that dithered quantization sounds like added noise, which is not necessarily true.
We did a listening test on this forum some time ago with roughly quantized music and finely quantized music with added noise, see High-order dither listening test . Mooly and PMA amazed me by hearing differences between the music with dithered quantization and with added noise, even after I made the probability distributions as similar as possible. Unfortunately I made a methodological error: the test did not make it clear whether or not Mooly and PMA would also have heard differences between two realizations of the same random process.
However, assuming the differences they heard were due to the fact that dithered quantization is not the same as added noise, neither of them expressed a preference for one or the other. At the rather high levels used in the test, they disliked both equally much.
We did a listening test on this forum some time ago with roughly quantized music and finely quantized music with added noise, see High-order dither listening test . Mooly and PMA amazed me by hearing differences between the music with dithered quantization and with added noise, even after I made the probability distributions as similar as possible. Unfortunately I made a methodological error: the test did not make it clear whether or not Mooly and PMA would also have heard differences between two realizations of the same random process.
However, assuming the differences they heard were due to the fact that dithered quantization is not the same as added noise, neither of them expressed a preference for one or the other. At the rather high levels used in the test, they disliked both equally much.
Bill, I understand the problem you mention with the numbers. Going back several years, when my daughter was a teenager she started playing guitar and singing. I wanted to record her at low cost, and looked around for a preamp and recording interface that should work. I knew good recording engineers advised against wasting money on what they considered junk, but I figured I was an engineer and I know if distortion was down below -100dB that distortion wasn't going to be a problem. So, I purchased a low cost interface made by a company that made clocks for studio references. They claimed the interface was low jitter. I was stunned at the awful sound quality, and I couldn't wrap my head around it. After suffering with trying to make do with it I eventually replaced it with something that was supposed to be pretty good, and mady be Apogee, no less. A Rosetta A/D. It wasn't quite as bad as the first thing, but it was still bad. Third try I purchased a Lynx II C, which was good enough although not perfect. I wasted a lot of time and money stupidly ignoring advise of people recording for a living and instead figured I knew how to calculate. There's more, but the point is after getting an expensive lesson, I learned that some of what is commonly believed to be accurate information about what people can hear is scoffed at by professional recording and mastering engineers. And, like it or not, they do know something electronics design engineers don't seem to have figured out. I don't think the problem is that engineers can't calculate, I think they have been given some incorrect information from faulty research. How could that happen? I don't know, I wasn't there at the time. But, it's wrong. At least some people, particularly with training and practice, can hear very low levels of distortion and also low level musical details. If and when I get you over here, you can see for yourself. You probably won't be able to hear everything I can hear, but you probably can hear more than you expect. I know I could be wrong, but chances are good with even a few minutes of coaching will start to change how you listen, including for blind listening. If we have time you can also see why ABX takes so much concentration to do over and over (in order to produce statistics) without getting confused.
In the meantime, I have to call it like I see it. It's either that or just shut up and ignore all the wrong thinking about distortion that goes on. I know there are problems on the other side too with audiophiles forming incorrect beliefs based on sighted listening. As I have said before, it is a very difficult area for researchers to work in.
In addition, I think I have said about all I have to say on this subject. Others please feel free to differ, I don't have any interest in arguing about it is all.
In the meantime, I have to call it like I see it. It's either that or just shut up and ignore all the wrong thinking about distortion that goes on. I know there are problems on the other side too with audiophiles forming incorrect beliefs based on sighted listening. As I have said before, it is a very difficult area for researchers to work in.
In addition, I think I have said about all I have to say on this subject. Others please feel free to differ, I don't have any interest in arguing about it is all.
Last edited:
Where does that 120 dB come from? 20 dB*log10(65536) + 10 dB*log10(3/2) - 10 dB*log10(3) ~= 93.32 dB. With A-weighting you may win some 10 dB*log10(22050 Hz/12.5 kHz) ~= 2.46 dB, but then you only end up with 95.78 dB(A).
Eh? Does not parse?
If recording and mastering engineers know so much why is the quality of released music generally so poor (serious question)? I often have to run screaming to listen to the redbird CD which is just 4 people, a ribbon mic and a recorder in a living room. That to me sounds 'right', live and natural which is what I want. But that is me and I realise I am not a normal music buyer.
I'd love to work out a test for ADCs that could be run to test this. Find out why the first units you tried were so awful and if their modern sucessors are still awful.
Marcel: look at any stereophile review measurements. example Chord Electronics Qutest D/A processor Measurements | Stereophile.com
It seems to me that you're mixing up A-weighted integral noise with noise in one DFT bin. You can make the noise in one DFT bin as low as you like by using very narrow bins.
The reason why many recordings sound terrible is commercial rather than technical: especially with popular music, severely compressed recordings sell better, or at least that's what the producers believe. Radio is even worse.
The most promising test of ADCs I've come across was basically running a signal (ie music) through them 50 times.
Even small aberrations become visible after being converted and reconverted 50 times and price has nothing to do with the accuracy of the convertors.
Sadly the site where the results were posted has disappeared a few years ago.
Even small aberrations become visible after being converted and reconverted 50 times and price has nothing to do with the accuracy of the convertors.
Sadly the site where the results were posted has disappeared a few years ago.
Sorry, just tried fix the confused language.
Regarding the quality of much modern pop music, many mastering engineers don't like what they are hired to do by producers and record companies. However, some mastering engineers pride themselves on being able to make CDs sound louder than anyone else can. Most of the 'good' mastering engineers really dislike the guys who enjoy making bad sounding CDs. Problem is that record companies know that loudness sells. Young listeners have been exposed to so much of that they they have learned to think over-limited, loud CDs sound 'heavy' and otherwise emotionally impactful. Some artists want their music 'competitively' loud, but still sounding good which is an rather contradictory set of desires. Mastering engineers are left to do their best to do what is asked of them, and you hear what we end up with. The only thing starting to turn the loudness wars around are the streaming music sites that have algorithms to adjust perceived loudness. Overly squashed recordings are actually penalized by being reduced in perceived loudness below less squashed recordings. Now mastering engineers are getting more requests to make recordings sound good for streaming, although as long as radio stations play music from CDs, what customers buy in stores are likely to be the more squashed radio competitive versions. Record companies and radio stations still know that listeners tend to stop at the loudest station with the strongest perceived signal, so the war isn't over yet.
More presenting information for comment. I look at those graphs and cannot see anything that suggests audibility in a room that has a human in (humans gurgle and groan and wheeze around the 20dBa mark. I didn't quote A weighted. I said the noise was 120dB down on the signal. Which is correct, if not the measure you were thinking of 🙂.
Point still stands that there is as yet no credible mechanism for why, in domestic replay a 24 bit master would sound any different from a well dithered 16bit downsample. Plenty of theories but nothing seems to have stood up to scrutiny so we are left with anecdotes, which is a sad state of affairs. There must be some way to move this forwards rather than continue to chase our tails.
Don't think it makes a difference in that case, myself. But some people like to listen with headphones, or in the near field in a quiet room. Don't know why they shouldn't be able to have a little bit better sound quality if they want it.
Whatever may be lacking with 16/44, at least if not overly squashed, it is far from the worst. I can't enjoy listening to satellite radio, or 128kbps streaming, they both just sound broken.
I listen to vinyl. I am not quite sure what that says about my tolerance for noise and distortion!
I am all for better quality if it really is and not just a cynical way to charge double for the same music!
I am all for better quality if it really is and not just a cynical way to charge double for the same music!
I think Panos assigned the hierarchy well.
There are plenty of people who hear faults with the high sample rates that they find even more objectionable than the flaws with 44.
I guess if one lives in an already microwave saturated environment then the noise of the higher resolution DACs is not noticeable? is it mainly those in suburbia who get to hear this? The admittedly few SACDs I have heard have this otherworldly kind of ambiance that sounds like neither the LP or the CD. Should there be that much of that kind of difference? Maybe this is what made it attractive to many folks? Never listened to high sample rate stuff when I had a computer setup. But then we are talking of similar noises to much greater level.
I listen to LP as my mainstay and there is nothing noisier than the phonograph even one taken to a very high state of tune. But all noise is not created equally. Some are more obnoxious than others, some are more distracting than others to individuals?.
We are lucky to have so many choices. Another reason why there will never be a single best audio component on this planet. Wish it was that simple.
There are plenty of people who hear faults with the high sample rates that they find even more objectionable than the flaws with 44.
I guess if one lives in an already microwave saturated environment then the noise of the higher resolution DACs is not noticeable? is it mainly those in suburbia who get to hear this? The admittedly few SACDs I have heard have this otherworldly kind of ambiance that sounds like neither the LP or the CD. Should there be that much of that kind of difference? Maybe this is what made it attractive to many folks? Never listened to high sample rate stuff when I had a computer setup. But then we are talking of similar noises to much greater level.
I listen to LP as my mainstay and there is nothing noisier than the phonograph even one taken to a very high state of tune. But all noise is not created equally. Some are more obnoxious than others, some are more distracting than others to individuals?.
We are lucky to have so many choices. Another reason why there will never be a single best audio component on this planet. Wish it was that simple.
- Status
- Not open for further replies.