Depends on what you mean by "averaging." It does in the same sense that tape bias "requires averaging."
Seriously, read the Lipshitz papers.
Averaging, for instance, as in the Daqarta link you posted:
Daqarta - Dither: Noise + Averaging = Resolution
I am sure there are many clever ways to reduce the impact of this kind of processing, but it still doesn't make the performance equal to doing it properly where real music is concerned
The effect is well illustrated by Terry's words a few posts ago:
in other words, the nature of the processing needs to be tailored to the specifics of the music, in order to minimize the degradation! I don't find this surprising, it is characteristic of trying to squeeze the best out of a skimped-resource system.
Daqarta - Dither: Noise + Averaging = Resolution
I am sure there are many clever ways to reduce the impact of this kind of processing, but it still doesn't make the performance equal to doing it properly where real music is concerned
The effect is well illustrated by Terry's words a few posts ago:
in other words, the nature of the processing needs to be tailored to the specifics of the music, in order to minimize the degradation! I don't find this surprising, it is characteristic of trying to squeeze the best out of a skimped-resource system.
Again, you've avoided the question of the bias signal on the analog tapes of recordings that you have found satisfying.
The effects of dither on resolution of a CD were clearly shown in the Stereophile review I posted (along with almost all of their other CD player reviews). FAR better performance when it comes to resolution than excellent analog systems. And the Lavry link that Terry so kindly posted. But most of all, the Lipshitz papers, which no-one besides me seems to have bothered to read.
I'm reminded of Max Planck's famous dictum:
"A new scientific truth does not triumph by convincing its opponents and making them see the light, but because the its opponents eventually die, and a new generation grows up that is familiar with it.”
When you have some data showing the audibility of a 16/44 A/D-D/A inserted into an analog signal chain, I'll be most interested to see that.
Well, no one seems to want to discuss the differential (if any) between listening to live mics and recorded mics (your choice of recording method)??
As far as noise on a recording - it is "random" noise when it happens, but once it is recorded it is determinate noise, yes? So, it might be instructive to look at the spectrum (or listen to it)??
_-_-bear
As far as noise on a recording - it is "random" noise when it happens, but once it is recorded it is determinate noise, yes? So, it might be instructive to look at the spectrum (or listen to it)??
_-_-bear
Where do you get this nonsense, seriously? Is there a textbook which I can have a peek at which dishes this stuff out?
I favour the latter explanation - why do you go for the former?
Interesting how the pitch is queered - comparing individual, cherry picked LPs with a format. I'd say that the best that can be said here is that your LP playback system makes a fool of your CD playback.
Sy,
Not to pick on you, but there are some issues that always seem to get confused.
Number of bits might mean the limit to resolution, but not always. For example an 8 bit converter should have 256 unique non-overlapping step sizes. Many 24 bit converters do not do this as do some of the older 16 bit units. The original CD's were only nine bit linear because to get the 16 bits they staged a nine bit unit with a seven bit device.
44 is the sampling rate. But it does not include how accurate the clock or even the sample time can be. In some early products the clock was divided by a non-binary number and although you got the right number of cycles per second the width was not uniform.
There are many approaches to conversion. The ratiometric or single bit converters were first used in the 1940's. A simple clamped signal and an accurate clock made a very serviceable converter. After some different designs the R-2R became a very widely used design. These designs of course work differently and have different effects on the signal being processed.
As all converters use analog input and output it is important to note that there are more issues than "is the analog as good as the digital?" When you charge a capacitor from a voltage source through a resistance the curve is an inverted exponential curve. Same for your current limited inductor, but of course you rarely see them in modern low level filters. So when you use a low pass filter to clean up your step jumps it is not a true linear average. This also is why timing errors are of greater importance than it would seem at first glance.
When you over sample, the sampling should be uniform in time. This gives us a linear average. If everything is working.
When you add something (noise, dither or a deliberate funny waveform) to the input of a converter properly it changes things. If you are oversampling you can with some care get increased resolution depending on a few issues. Some work has been done on this but there is still more to do. If you are at the Nyquist limit you may prefer the results over the non added version. This work has been done pretty thoroughly but there may still be a surprise or two to come.
What should be obvious is that the distortion from any added signal process is very hard to predict because the step size error at the lowest resolution is not constant at 1/2 of a bit but varies throughout the entire range.
So YES I can make a 16/44 chain that will sound bad to almost anyone. When I read some of the experiments using equipment that I know is not really as claimed or when the math shown does not understand the importance of how the filtering is done, some room for inquiry is left.
ES
Bear,
Long ago I tried binaural recording. In theory it should be possible to use a mold of your head to make a recording that to you would be the same as being there, when played through matching headphones.
I never got this to work for me. Small mics and headphones were never good enough never mind the electronics.
Of course the obvious problem is even a mold of my head is too big to fit through most doors!
ES
Ed, I'm certain you can. I probably could, too.
But what remains standing is that, using standard, competently designed 16/44 systems (of which there are many, many), there is still NO data showing audibility when inserted into an analog signal chain. The idea that the resolution of a true 16 bit system (of which there are many, many, probably the majority) is somehow less than analog systems like tape and vinyl is just flat-out incorrect.
Where are you going with this - in what sense is the output of a low pass filter not a 'true linear average' ? Because the settling to the zero order hold DAC output is exponential so the target voltage is never in fact reached, or for some other reason? Are you suggesting there's an inherent non-linearity here?
Please do flesh out the relationship of DAC setttling time to timing errors if you can, its something that's been a bit of a fuzzy area for me for a while.
SY, you're confusing the debate. I called dithering a second-rate solution compared to proper high resolution recording. In no place did I say that dithering wasn't clever, nor that it might fail to improve CD-format recording. But as Terry's evidence shows, the mechanism is programme dependent - Dithering is not the same as high resolution.
For the other question of LP vs 16/44.1, I have compared analogue LPs to 16/44.1 in one repeatable fashion: Digitally recorded & mastered LPs. I have hundreds of these (they're very cheap, and often have interesting pieces on them). Many are EMI pressings manufactured in the same plant as the last of the analogue versions. I do listening tests with a second opinion (first in weighting) - one of the longest serving sopranos at Armonico:
Armonico Consort - Home
She is completely uninterested in records, hifi or the like - she's a acoustic performer. But when I play digital LPs, she has consistently ranked them at the bottom, mostly due to a lack of life or vibrancy; and though the standard of analogue LPs is not at all even, they usually show a realism that convinces her, and me. Yes, we have experimented with SRA & VTA on these LPs, and will often return to within 0.5mm of the arm position she preferred previously.
Whether these faults in these records are down to word size, sampling rate or something else, I don't know. But the effect is not subtle, and I would not believe that tinkering around the edges will be enough to put it right.
CD shows a lot of the same traits to me. I know that the players are flawed, and can be improved greatly, but what I would really be interested in comparing is 16/44.1 against real high resolution - how high do you need to go before you stop hearing improvement?
Anecdotes of uncontrolled tests with poor validity aren't exactly data.
And PLEASE read the Lipshitz papers. You keep repeating the same incorrect statement regarding resolution. If you're just interested in mantras, that's fine- just man up and say, "My mind is made up and I don't really want to understand the issue." Nothing wrong with that and it's certainly more honest. But saying something wrong again and again does not make it right.
And PLEASE read the Lipshitz papers. You keep repeating the same incorrect statement regarding resolution. If you're just interested in mantras, that's fine- just man up and say, "My mind is made up and I don't really want to understand the issue." Nothing wrong with that and it's certainly more honest. But saying something wrong again and again does not make it right.
Reports of real listening may not constitute DATA but are no less real for all of that.
I am not repeating anything!! Perhaps you should clarify what you think Dithered 16/44.1 is actually equivalent to in resolution terms, so that we can debate the same thing, not confused nonsense.
I am only aware of one test where a 16/44 A/D-D/A unit was inserted (or not) in a playback chain with an effort at test control. That was where Ivor Tiefenbrunn of Linn fame IIRC failed to hear it.
One test participant (who happened to be Lipshitz) WAS able to reliably identify the unit because after some time he realised that there was a slight difference in background noise with or without the digital set.
But those who listened to the music only could not identify what was what.
The test is documented on the Boston Audio Society (BAS) website, look under archive I think.
jd
One test participant (who happened to be Lipshitz) WAS able to reliably identify the unit because after some time he realised that there was a slight difference in background noise with or without the digital set.
But those who listened to the music only could not identify what was what.
The test is documented on the Boston Audio Society (BAS) website, look under archive I think.
jd
Correct, but that 'data' is only valid for that person, that situation, that equipment, his mood etc. It's worthless for anyone else.
Come to think of it, it is worse than worthless, because it biases people to a certain outcome for their own listening.
jd
John,
I don't think he disregarded your measurements and mathematics as such, he only didn't believe the effects were audible, which was an unfounded assumption anyway.
jd
nobody is actually literally recording 16/44.1 today - oversampling, "24 bit" ADC and signal processing are much cheaper than 20 KHz analog anti-alias filters
broadband S/N >110 dB and differential linearity way down in the noise is the norm for top monlithic audio ADC
target rate brickwall filtering, decimation and bit depth reduction with dither is only used at the final stage of the production process
people should also ckeck out Lukin's dither clips - he provides 8 bit audio clips to exagerate quantization noise artifacts and show various dither's improvement in audible resolution and noise shaping's reduction in dither noise perception
Homepage of Alexey Lukin
broadband S/N >110 dB and differential linearity way down in the noise is the norm for top monlithic audio ADC
target rate brickwall filtering, decimation and bit depth reduction with dither is only used at the final stage of the production process
people should also ckeck out Lukin's dither clips - he provides 8 bit audio clips to exagerate quantization noise artifacts and show various dither's improvement in audible resolution and noise shaping's reduction in dither noise perception
Homepage of Alexey Lukin
- Status
- Not open for further replies.