|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
172.98.67.57
In Reply to: RE: To Dither or Not to Dither? J. River/PonoMusic World posted by Tony Lauck on April 16, 2016 at 11:32:42
Tony Lauck:
"It should be possible to measure the distortion products added to low level test tones on the analog outputs of any of the SABRE chip based DACs. Most of these are pretty linear at low levels, so the distortion ought to show up out of the noise, especially when spectrum are computed over long averaging periods, which will reduce the residual noise of the DAC chip and output amplifiers.
You may need a low noise preamp to boost the output level of the DAC so that the noise and distortion products are well within the resolution of your ADC."
Sure but isn't that academic?
I'm absolutely happy to turn on dithering with bit-depth reduction to 16-bits and regularly use the noise-shaped MBIT+ algorithm in iZotope RX for 24 --> 16-bit conversions.
As for JRiver and others, for 24-bit playback I'm agnostic on this since I don't believe it makes an audible difference in any conventional situation (assuming the dithering is only affecting the lowest bit or two). On or off IMO makes no difference for an actual DAC and I have no reason to believe anyone can hear this.
-------
Archimago's Musings: A 'more objective' audiophile blog.
Follow Ups:
Here is what you said,
"Within 24 bits truncation is no problem since no DAC is capable of the last few bits."
This is about what the DAC does, not what you or someone else might or might not be able to hear.
I pointed out that this is likely to be false. Most sigma-delta DACs will respond to signal at the lower bits. The SABRE chips will respond down to at least 32 bits. If they did not respond then it would not be possible to measure differences at the 24 bit LSB, such as those provided by dither.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Wouldn't you say "this" is also about whether the information contained within the signal is still in a usable form? Just how far into the noise would you suggest a human can possibly hear? Just how quiet do you expect real world DAC implementations can actually get? I will believe people are hearing 24 bit resolutions out of real world DAC implementations when I see proof.
I made two separate posts in this thread. One post (not in this thread) confirmed that there were some people who reported hearing differences at 24 bit dither. This was about audibility. I was passing on what these people reported. I don't claim to hear these differences at 24 bit, by the way, but I have heard them at the sixteen bit level and decided that the safest approach was to always use TPDF dither. Using no dither or alternative dithers may be sometimes better, sometimes noticiably worse.
The other post was purely technical, and that's in this particular sub thread. I was rebutting Archimago's claim the DACs wouldn't resolve differences down to the lest significant bit in a way that can be measured. Since Archimago's supposedly a measurement guy, I threw this out as a challenge to him.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Measurement of not, my comment is that it's just not audible anyways; hence the statement to the OP that 24-bit dither is a take-it-or-leave it proposition.
The most contentious part to all this is the claim that some people can hear it... Would like to know who these people are.
-------
Archimago's Musings : A 'more objective' audiophile blog.
this is a predetermined statement about your belief and has nothing to do with other peoples' belief that they can hear a difference.
Whoever these folks are, assuming they can do what they claim, they are most definitely an edge case.
Tony Lauck trots them out from time to time, but he is always careful to disassociate himself from their claims.
We need not worry too much about TL. Let us instead look at reality.
Have you ever seen a Seattle Seahawks game? (For our international readers, the Seahawks are an NFL football team whose stadium is in Seattle, Washington, USA.) The Seahawks have set records for the sound levels created by crowd noise in their stadium. Their latest record was a sound level of 137.6 dB. Assuming the sound level at the field before the game was 0 db, (I know that's an unrealistic assumption) it would take about 23 bits of resolution to record a game in that stadium.
Let's think about how opposing teams play in that stadium. They are universally prepared to use a "silent count." That is, opposing teams fully expect that their healthy young teammates will not be able to hear their healthy young quarterbacks shouting signals at the top of their lungs. If the crowd is too loud, there is no point in even trying.
Doesn't this argue against the claims that we should be able to hear subtle differences even 24 bits down? These young, professional athletes are trying desperately to distinguish what another young, professional athlete is shouting at the top of his lungs yet are unable to do so.
How loudly do watches tick? If any of the linemen visiting Seattle are wearing a watch during the game, does anyone think he could hear it?
If he can't hear his watch, why would we expect to hear things at similar volume levels?
JE
"A difference which makes no difference is no difference at all." - William James
Your analysis does not take into account the necessity for headroom when making digital recordings. For classical music orchestral recordings this is typically 18 - 30 dB, comparing RMS levels with peak samples. In addition, for live recordings that are not normalized (just what came from the ADC) there is the need to keep a few extra dB of headroom that is otherwise wasted in case the musicians play louder than expected.
I have one recording that, were I to attempt to reproduce it at live volume, would result in audible dither noise. It is an uncompressed 24 bit recording of a Space Shuttle launch taken at the VIP booth. If adjusted so that the "ten, nine, eight,...." is at 70 dB at my listening position the launch SPL is 140 dB at 15 Hz. (According to Bob Katz who made the recording.)
There are other less extreme examples producing such high signal peaks, such as Arny Kruger's key jangling test. This has energy all the way up to 48 kHz on a 96/24 recording and for natural volume level requires a gain setting about 10 dB louder than I use for a Mahler symphony, even through the keys aren't loud. The peak to RMS ratio on this recording is 26 dB.
Also, note that the threshold of hearing at some frequencies is below 0 dB. It is right at the thermal motion of molecules against an ear drum.
Your arguments neglect many factors that can add up to 30 dB. Notice that in a live concert situation, one will hear a dynamic range of over 120 dB. The dynamic range of a full orchestral tutti to a solo violin can be 55 dB or more, however one then can listen to a note decay due to hall reverberations, right down to the listener's threshold of hearing.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
"Notice that in a live concert situation, one will hear a dynamic range of over 120 dB."
I've seen this claim made numerous times over the years. What exactly does it mean? Where does one need to sit in order to get this dynamic range? How large does the orchestra have to be in order to hit those volumes? How quiet does the venue have to be in order to realize dynamics like that? What size does the venue have to be?
In my experience, symphony orchestras typically perform in venues that seat hundreds of listeners. Those listeners all make noise: breathing, rustling clothing, rustling programs, quiet murmurings to one another and even the occasional snore from an over tired patron. Likewise, the orchestra itself makes breathing noise and rustles its sheet music on their stands. Every single one of those people gives off heat, requiring the venue to run some sort of air conditioning to cool the venue. As a result, I can safely say that I've never been in a venue with an effective noise floor of 0dB. A charitable estimate would be a noise floor of 20dB, reality is likely higher.
That means, to get a 120dB dynamic range, the orchestra must be capable of a 140dB output. Golly! Those are some boffo numbers! That 100 or so piece orchestra is able to beat out nearly 70,000 rabid Seahawks fans at full cry! But even then, would that sound volume be delivered to the entire venue? Doesn't sound attenuate with distance? Won't the seats in the back hear a diminished dynamic range? Or if they do not, won't the volumes up front need to be even higher?
So again, what exactly does that claim mean?
JE
"A difference which makes no difference is no difference at all." - William James
Peak SPLs measured in the front portion (e.g. row 10) of a concert hall during performance of Mahler symphonies have been measured at 122 dB. The quietest sound that one can hear is limited by the ambient noise, and if the hall has a quiet HVAC system the level in the critical range of 1 - 8 kHz will be very low. This is where the dynamic range requirement comes from.
If you want to experience the same sound as at a live concert you will want to be able to hear sounds like bows hitting music stands and page turns. If you are interested in a mid-fi experience, this may not be of any interest.
Most hall noise occurs at low frequencies and can be at moderately loud levels due to the F.M. curves. Audience noise, and other annoying events are not continuous and won't mask sonic effects when they are not present.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Where are these "24 bit" digital recordings even coming from?
The best commercially available ADC's I've come across at any price are claiming somewhere in the neighborhood of 120dB dynamic range, maybe a tiny bit higher in some cases. None are anywhere near the 24bit mark that I am aware of.
There is the same problem with playback...where are there DAC's claiming much better than around 130ish dB dynamic range capabilities, tops?
I could see ultra high end analog recording technologies maybe achieving this kind of performance but what digital equipment can do this?
These ADCs output 24 bits as a result of signal processing. The DSP creates low order bits, and generally quality of DSP will suffer if there isn't at least 48 bits of resolution in the signal processing. In addition, most recordings go through additional DSP in post production. The same situation happens during playback if a digital volume control is used in playback, whether done in the computer or in the DAC.
If dither is not used when bit depth is reduced (e.g. 48 down to 24) the result is signal dependent distortion. This is true even if the original analog signal is noisy. For example, if the ADC is only "good" to 22 bits, it might have noise at the 22 bit level added into the output. Or, there might even be distortion generated by the analog equipment when handling high level signals resulting in even greater errors. However, none of these errors will necessarily hide the errors involved when dither is omitted.
If you look at signal numbers that are applied to manufacturers' spec sheets and not how the mechanisms actually work you will not get a complete picture. The limitation in one dimension (noise or analog distortion) does not justify a limitation in another dimension. Each type of distortion can and will be perceived separately by critical listeners. In the case of 24 bit dither, most people would consider the benefit small and might not be willing to pay a large amount to eliminate what is a low level distortion. However, the cost of doing things correctly is essentially zero, being just a few instructions added to software.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
" The limitation in one dimension (noise or analog distortion) does not justify a limitation in another dimension."
They may not always be the same point on the graph but I would argue that noise level in a transmitted signal must always set a practical limit for making distortion measurements. And every signal has noise.
My question isn't whether to use dither or not. I'll always follow he recipes DSP experts suggest since I can't be bothered to think about that stuff.
It is regarding the practical limits of detecting it's benefit. I don't know at what level a computer could detect distortion on a signal
"They may not always be the same point on the graph but I would argue that noise level in a transmitted signal must always set a practical limit for making distortion measurements. And every signal has noise."
This is not necessarily the case. If for example the transmitted signal consists of a low level sine wave that is buried in the noise then it will be possible to pull out distortion created by non-linearities, such as those caused by lack of dither. This can be done by doing a spectrum analysis to average out the noise and bring into view the signal and its harmonics. The practical limit to this operation depends on the stability of the clocks involved and the patience of the experimenter. Clock stability can be reduced or eliminated if it is possible to synchronize the clocks or run them off a common source. It may also be necessary to temperature control your entire apparatus, lest it begin to operate as an unwanted thermometer. :-)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
from the Crookwood ADC page:
" With a good ADC, you can play and hear a sine wave, say at -20 or -30dB below the digital noise floor of the ADC."
I would have never guessed anywhere 20-30dB is possible. I wonder what the basis for this statement is, ie experimental evidence, wag, etc....
The depth you can go depends on the bandwidth over which the noise is spread. But these figures are approximately correct. Here the test would be to ABX several seconds of dither noise vs. several seconds of dither noise with a 1 kHz tone at the selected level. At the margin the tone is just barely perceptible by ear, as it fades in and fades out. If one is patient and listens carefully one can tell if it's present or not with some degree of reliability. If the tone is modulated with a random message using Morse code, I can decode the message reliably at about -15 dB, but the "bit error rate" is noticeable when the tone is further below the noise. At -20 dB I can tell something is happening, but I can't read the message. (This was at 44 kHz. More depth at higher sample rates.)If you capture the output of the signals on another computer using an ADC driven by a preamp that boosts the signal without adding more noise, ensuring that the ADC's own noise is not being tested, then you can use a very long FFT to go considerably lower than the broadband noise floor than you can do by straight listening. This is because the computer is averaging over a longer time period than your ear. What happens (assuming the clocks are stable) is that the noise is divided into more and more FFT bins as you increase the FFT size but the sine wave signal is always concentrated in only one bin, regardless of FFT size. (Here it may be on the edge and appear in two bins, this will mean each bin has half as much energy. Details depend on the FFT window. You can tweak this by changing the frequency of the test tone.)
All of these tests can be done with an audio editor that converts between bit depths with various dither algorithms and that has the ability to generate test tones and mix signals and that does FFT plots. I used Soundforge 10c for my tests.
Other fun and games include learning to hear the effects of various types of dither (or sample rates) on music. For more serious work, you have to write software that directly generates and processes the test signals involved and creates creates and analyzes WAV files that a sound card can use or generate.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Edits: 04/26/16
Kind of mind blowing to me to think that we/presumably other species can accomplish this much computational work with no apparent lag, ie we still get to hear everything we hear in real time. Nature still has tricks to teach us regarding the doing powerful information processing hardware the right way. I wont even try speculating on what processes our bodies must be using to accomplish this amazing amount of work.
"I could see ultra high end analog recording technologies maybe achieving this kind of performance..."
Just curious, but what would those analog recording technologies be?
TIA
JE
"A difference which makes no difference is no difference at all." - William James
I believe that really wide and high speed tape can do quite well in this regard, though I don't know what the theoretical and practical limits are.
Really wide and high speed tape can do extremely well, but only if someone else is paying for the blank media. :-)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
"I was passing on what these people reported."
In other words, your "evidence" is hearsay. Why would you waste our time with such silliness? Who cares what mystery listeners report? We are unable to quiz or question the folks making the assertion you say they made. Maybe you misunderstood them. Or maybe they misspoke. Or maybe they were flat out wrong. We will never know the truth because we were not able to question them ourselves.
"Since Archimago's supposedly a measurement guy, I threw this out as a challenge to him."
An even better challenge would be for you to post measurements of your own.
JE
"A difference which makes no difference is no difference at all." - William James
Audio Asylym is not a court of law and so there is no hearsay rule. In any event, this would not be hearsay, since the comments in question were sent to me via private email. These private emails (which I still have) would constitute evidence. I would also point out that the legal system is designed to resolve disputes peacefully and is not designed to discover scientific or engineering truth.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Sorry Tony, my bad. Now that I've gone back and reread your post I see that you were simply passing on the claims of others, and not saying that the claims they made were probative.
My eyes aren't what they used to be!
JE
"A difference which makes no difference is no difference at all." - William James
yes, but with the wrong tools and approach.
He ignores your facts or just pushes his agenda.
An unprovoked attack from a poster who seemingly thinks posting an opinion online turns it into fact.JE
What is going on tonight? It's as if the forum software is possessed.
"A difference which makes no difference is no difference at all." - William James
Edits: 04/18/16 04/18/16 04/18/16
enabled and listened to dither on hires files? It's something that I have wondered about but has been too lazy to go into it in a detailed way (which is necessary to draw 'right' conclusions)
What has been asbsent from the 'disbelievers's' posts has been an understanding of what dither does to 'linearise' the signal at the lowest levels.
Unlike analog signals, distortion rises in digital replay as the output is lowered towards the noise floor.
"Unlike analog signals, distortion rises in digital replay as the output is lowered towards the noise floor."
I have read this several places. The one I can remember is Jim Smith's, "Get Better Sound."
I'm not trying to boost my particular system, but I would like share something I've been thinking;
A good SET amp coasting along at 100 milliwatts output produces almost no distortion. A good modern compression driver/horn combo produces ten times less distortion than it's direct radiator counterpart.
It's possible that when you get your system to ultra low levels of distortion like that, is when you can really start to hear the low level details in a recording. Horn guys call it "micro detail," and I think it's the same exact thing. It's why horn guys love their turntables. Vinyl, or most any analog format, has plenty of low level detail. 16/44 does not. Is the detail covered by the higher noise floor, or is it a matter of quantity of data? Or does the greater data push the noise floor down by moving it up out of the audible band? Ai yai yai.
Not saying any good system won't let you hear the difference, but SET/horns may just make it a little plainer.
-------------------------------------------------------------
Big speakers and little amps blew my mind!
If you use properly dithered digital that is well implemented there should not be any more low level distortion than with analog gear. The key phrases are "properly dithered" and "well implemented".
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Really? Please show us how.
If so, why does JRiver sound different from Foobar from Teac Player from others at 24 and 32 bits?
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: