The DVD/Dac thing

We had a lengthy discussion about this in a long thread some time ago which concluded pretty definitively that error correction considerations aren't remotely relevant to sound quality - or at least, that was the inescapable conclusion you had to reach if you actually took the available evidence into consideration, rather than just making arbitrary a priori assertions. If you've got anything to add concerning uncorrectable error rates and interpolation that challenges that view, I'd certainly be very interested to hear it (am always up for learning new things ).

 

pete,
whilst i agree with your above statement in theory i'd still say that from practical experience (with data cd's at least) they are not always read error free first time hence the need for software to cater for error checking and re-reading. indeed this is a technical requirement for release on all of the software that i write. how frequently these errors occur and how much bearing this has on audio cd sound quality is debateable, and to be honest i really cannot be arsed debating it as it has no bearing on my setup.
i would think that a transport that was able to re-read incorrectly read data rather than just interpolating first time read data or just giving up and skipping would be preferable to a 'real time' read system - from an acuracy point of view at least. whether it would sound 'better' or is indeed worth stressing about (due to frequency of read errors unrecoverable by the hardware error correction) is another thing that i recon it's pointless to debate.
i'd also say that there is more to getting a true signal from the cd to the dac than just getting the info off the cd. but again it's something that doesn't effect me so idgaf.
cheers


julian

 

Pete,

Quite happy to domonstrate this, take a trip round, when your next in the area

 

The 'thing' that good speakers do WELL..is different from the 'thing' that good CD players do WELL. You can't say that one compensates [or complements] for the other. A good speaker, with huge headroom, linearity, massive bandwidth, whatever else, will not correct a loss of detail, ambience, phase info, bass speed, slam and depth etc.

I know this too well, since I'm stuck with a CD63SE driving a system I used to enjoy, until the bloody Densen CDP gets fixed

 

Hello Saab & Co
In my first submittal I mentioned that I think the amp is really the most important link in the reproduction chain. Reading a few comments on cheapy DVDs compared to dedicated CDPs..., it reminds me of Julian Hirsch's proclamation [back in the 70's] that all amps sound alike. Hogwash. I'm lucky enough to have picked up the Tabu Century (basically because of its phono 'sound') ands alot of readers here appear to own the NAD 300er sibling hereto. Why [if all amps sound the same]?
I just returned from a Christmas get-together at a freinds who recently purchased the B&O Beolab 5 active speakers. These are time-alligned (my wording) and seemed to present music well known to me at "about 7%" longer than I recalled from my own or other systems. Anyway, Thomas was playing radio at the party through a B&O (the mirrored one) tuner and the station sounds "terrible" to me even on my (h/k TU 9600). His system sounded terrible [because of the lousey input signal]. The day I first auditioned them, well, although the '5s' are dear in price, I thought they were well worth the investment - and we auditioned them with a Euro 150.--DVDplayer!
At the time I suggested Thomas to get a decent CD/DVD player, e.g., Micromega, Thule, et al others... and if he were to go DAC/Player, to use Gryphon's Guideline II (the red one - soldered shield to the earth/- from the signal input side ONLY) as digi-cable.
The performance was ever so disappointing, and with all those guests wondering why he paid Euro 15,000 (some figure like that)...

My son gave me a DVD player for Christmas. Mine (a Pioneer) a Metz DE81 does not play copied discs (I shun away always from 'burned' software - I want the artist to get his money). So in I popped my new Euro 150 DVDer into the video rack and shure enough - the picture (according to the press my present presents 'as good a picture as they come') was not as good as my Metz (maybe because I still watch, by choice, through a 50Hz and not a 100Hz TV - Loewe Profi Plus) and the sound did not even compare, i.e, subjectively I found it sounded "20%" worse than the Metz... The Metz goes into the cellar and my son feels good and it is more slim (hah! room for a Sony MD?!! - I only watch an occasional DVD movie - I like stereo sound (and I like the sound from "a good source" into the beolab 5s or as is, into my system).
I know how much hifi - hype (and shyster salespersons) has damaged our hobby. I personally can not agree though that "all sounds the same if the specs are the same."

 

Hi,

Hmmm. That is an interesting statement, but sadly in strong contradiction to:

1) The basic scientific that there is no way to provide proof beyound a reasonable doubt for the absolute absence of something, so your statement implying that all fatcors aother than those you listed are inaudible is clearly not correct.

2) The various observations (usually outside strict HiFi/Audio applications) of the discriminative ability of the human ear/brain system to "hear" things that are rather difficult to measure. As an example I might positi the ability of trained and experienced radio operators to reliably hears signals burried under 20 - 40db of noise, something that is not particulary easy to design a mecahanical/electronic decoding system for.

So sorry, it may be your opinion that "frequency response, noise, distortion, loudness and to a very limited extent phase, are all that can be heard.", however your opinion is far from established fact and not beyond VERY REASONABLE doubt.

Having thus disposed of the first basic fallacy stated lets proceed to the following ones.

I don't have to look again, I noted that this equates to 95db S/N in conventional terms (eg referenced to full scale). And allow me to repeat, -95db A WEIGHTED (eg with high and low frequencies where all the main spuria resides rolled off severely) S/N is piss poor for a 16 Bit Device. For a 16 Bit device an unwieghted S/N of 96db over a bandwidth of 20KHz should be observable, A-weighetd it should be between 10 - 15db improved on that level.

It may or may not show up in the THD+N Numbers, it will not show up if THD+N is already poor, but the fact that the jitter is burried below the noisefloor does not make it inaudible.

If the transfer function is sufficiently simple - yes. In the case of a DCP with a Digital filter, DAC and Analogue Stage the transfer function is sufficiently complex that siongle tone measurements usually result in results that are not sustained when measuring IMD.

Moreso if IMD is measured with more than 2 Tones, in which case the results often show a drastic change of the noisefloor (in the FFT) with signal level, bringing us to: NOISE FLOOR MODULATION.

It is very easy. If the IMD and THD result from the same direct transfer function one would expect IMD levels broadly similar (within the limits of experimental error) to THD levels. In reality IMD levels are suually significantly higher. Suitable basic IMD measurements are regulary published in a variety of magazines, so I save my writing.

Perhaps because of excessively high noise levls in the first place?

Obvioulsy. However serious problems are evident in the measured response you quoted and they may very well be severe enough to swamp out any others or indeed may be related to poor performance in exactly the areas mentioned (eg the poor S/N ratio may very well be the result of excessively high jitter levels).

Unless we actually measure some of the stuff that matters and make sure to correlate the measuremnents detail (eg noise/jitter/distortion spectruim) to the way it is actually percieved we are still unable to identify exactly the performance impact of a given problem.

You do not need to take the lid off, but you do need to apply yourself to an extensive detailed analysis of noise, jitter and distortion spectrae, preferably armed with the knowledge how they influence percieved sound.

After having found the unit wanting you may proceed to remove the lid and to track down the actual reasons for such behaviour.

Clearly what "measures well" differs very much between you and me.

Iagree that the required measurement equipment is rather expensive. I do not however feel that would be pointless, on the contrary, the currently commonly quoted single number measurements you place so much stock in have repeatedly been shown to be in fact pointless to the greatest degreee (most recently Earl Geddes on Distortion, others have worked on jitter audibility etc).

Hmmm. But are we talking about measurements that correlate with how distortion is percieved? It is easy to construct a signal with a distortion mechanism applied that results in 0.1% measured THD but is much more audbile as distorted than a signal with a distortion mechanism applied that leads to 9.6% measured THD.

So in that case the unit with 0.1% measured THD would be obviously EXTREMELY far from offering any fidelity, while the 9.6% measured THD unit would be barely distinguishable from the original.

Don't believe it? Take the test:

Sound Quality - Distortion Perception

Take the test and then repeat that statement.

Yup. BUT equally it has been noted that certain signals, distorted in the measured sense only very slightly are audibly distorted according to the listeners (see various results on the various perceptual coding and watermarking reasearch).

You mean modern day engineering is the craft of saving cost as much as possible while satisfying certain specific performance criteria which have been demonstrated as holding no reliable correlation with sound quality, while ignoring other performance criteria that have been shown to be audibly relevant but did not make the list.

I personally call that at the best (when I'm in a charitable mood, like now given it's X-Mas) stupid engineering.

I'd expect something like this on a suitable FFT analysis (uppermost trace, which is 16 Bit Data, dithered and thus already raises the noisefloor at least 3db above the theoretical limit:



BTW, the report to go with that picture is here:

Lavery Engineering DAC review

It does only if we assume the noisefloor is not signal correlated (which it usually however is) and we wish to know how audible the noise on it's own will be. It makes SOME sense for hum and hiss in amplification (but even there not really) but not if it is applied to measurements that mainly establish linearity and dynamic range.

Because using A-Weighting is in effect cheating. It tells us how audible the noise itself as noise will be, but nothing what various intermodulation processes will do to interfere with the signal and how audible that is (You may well asume that the results will be inaudible, but you may assume wrongly).

Hardly. Jitter in measured on the analogue output signal (example Miller analyser). Therefor it matters zip what is in the Box. Jitter is jitter and thsu if measured relevant, if we accept the extensive body of work that has illustrated jitter audibility.

Anyway, enough from me, it is not my job to make sure you (or indeed anyone else) are aware of how reality works, i would expect from someone who wishes to speak authoritatively to be aware of the current state of reasearch and it's implications. It seems obvious that you have some catching up to do.

Ciao T

 

Hi,

That is how it works WHEN the data read from the Disk has unrecoverable errors. I am aware of how the erroro correction system in CD works, however when the rather extremely basic error correction (much less than on any PC Hard Disc or Data CD) faisl to allow the data to be read correctly teh following happens:

1) for a limited time samples are interpolated
2) If a limit for interpolation is reached the player mutes

By the time you hear the crackeling mute ot skipping on CD's that are badly scratched you have already listend to loads of error concealement.

We may argue endlessly about how well error correction works or not.

The easiest way is to compare to an EAC Rip in "paranoid mode" and to weep as it becomes obvious how many Players (CD or otherwise) do not offer bit accurate digital side decoding, especially among the cheap ones.

I used to have an early CD Player that had an Error LED. Facinating to watch. On some CD's it would not blink once throughout the whole CD, on otther there would be constant flicker and occasionally quite bright flickering. It was also fun to note how the error light correltaed with sound quality.

Sadly modern decoder chipsets make accessing the error flag which is set whjen error correction fails and concealement (eg interpolation is used) is applied rather hard to pick up on.

Ciao T

 

A few of us would like to thank Thorsen, for bothering to put into words which we couldn't be arsed or have inclination too. Cheers Wm

 

indeed,I second that,marvellous post (this doesn't mean I understood it,but it was a bloody good read all the same and must have took some time,thanks)

 

I agree with Saab and WM a well written and excellently reasoned post from 3DSonics.

Technically it left me floundering a bit, but well written all the same.

I have 2 points to add;

1) From memory I recall reading that the error correction system draws more power than any other function in a CDP. Therefore when the error correction is in use, if the power supply is not substantial enough it draws power from other functions. Maybe this is why some players sound better than others. I also remember players that used to show error correction results, one player in particular used to give a % reading per track or per disc. It was a regular occurrence to have results in excess of 70 % of the disc when error correction was in use.

2) To be honest, I do not give a toss what a player (or any other piece of gear) measures like, if I like the sound of it is a good machine (to me), if I don't like the sound it's not a good machine (to me). I am buying a piece of equipment because it matches the likes and/or dislikes that I like/dislike, nothing more. I may not be correct in liking what I like, but it is my money. I have speakers and amps that others do not like, I do, they have been reviewed well and reviewed poorly, they have sold well and been commercial flops, it is the buyer's personal choice and how they sound in the room where they are in use.

At the end of the day if sound quality was the only factor, and there was a single absolute in that regard, there would only be one item to buy.

 

That phenomena is completely explaned by "Critical Bands" and is no miracle.

Well, after 150 years of research into hearing, those are the established facts...

AES Convention Papers 5890 and 5891 I presume. It's nice to see someone quoting properly conducted, bias controlled, Double Blind Tests!

Now, while you might think the inaudibilty of 9.6% THD vs the audibility of 0.1% THD is some sort of paradox, it isn't. This result is well known in psycho-acoustics. In fact Geddes references a very good introductory text by BCJ Moore, and it appears that he is using nothing more than simple frequency masking.

First off, those papers deal with relatively high levels of THD - with only 3 of 21 tests have a THD below 1% and (I'm quoting that directly from the paper) the primary focus of this study is to understand the perception of loudspeaker distortion. Here we are arguing about CD players with significantly lower distortion (and noise) of 0.003%.

Now, although the predictive value of THD is low in the context of high distortion, Geddes does not extrapolate that result to suggest (as you are doing) that low levels of distortion associated with CD players (for instance) are "pointless". He has not shown THD to be meaningless at low levels.. The problem with THD from an auditory masking perspective is that the audibility depends on the magnitude of each harmonic, but at low levels of THD, it doesn't really matter what the harmonic structure is because the harmonics fall below the threshold of hearing. This is why Geddes doesn't extrapolate his result downward.

What Geddes is really doing is trying to remove the inaudible (masked) parts of THD from the measurement and emphasize the harmonics which are not masked (and he's not the first to do this either).

Now, it's possible that your audience might still think Geddes's work is relevent to CD players. So, I'd like you to finish the following sentance for their benefit:

"The output stimuli were saved as 16 bit, 44.1 KHz .wav files. The sound was reproduced by a ....."

Bottom right paragraph on page 3 of AES convention paper 5891..

No, you've got this the wrong way around. You start with what people can really percieve (ie in a biased controlled DBT), and if they can prove that they can hear something you set about identifying and correcting the cause. You seem to be coming at this from the other direction: improving the measurements and then trying to correlate that with some "perception",

We can see this in the two biased experiments you've suggested to people: the first, of copying a CD and hearing it's improvement; and the second, the correlation of interpolation errors and flashing lights. The first one is a clear example of suggestion leading to experimenter expectation. The second is again suggestion of a different sort, where the flashing light is making you think the sound is worse. If you could, by hearing alone, detect the interpolation and press a button "in tune" with the unseen flashing of the error light then that would be an accomplishment (and indictaive of actual sonic problem rather than the result of a vivid imagination.)

And so it is with THD+N (and IM). If you have some published results of biased controlled tests which indicate that people can hear 0.003% THD+N then l would stop arguing with you. And the same goes for jitter audibilty, and noise modulation (for which the lowest published audible number I've seen is 1%).

You've brought up IM and jitter - but you have NOT shown them to provide any insight beyond THD+N, you are merely asserting that they do. The fact that Stereophile publishes these numbers doesn't lend any credence to them being audible.

The difficulty (for you) with jitter is that for it to be meaningful, you need to show that it produces an audible effect. If the jitter results in broadband noise, then that is very different to a 10th harmonic of a signal. In other words, you need to do what Geddes is trying to do: to turn a jitter measurement into something that makes sense in terms of audibility, but your task is complicated by the fact that diffent DAC architectures produce different distortions for the same amount of jitter.

This (and much of your other argument) revolves around the noise performance which you are judging to be poor based on your incorrect conversion of the quoted a-weighted number to an unweighted number by deducting 10-15dB...

 

Again impressive Datty Boy!

Now, tell us, why does just about everyone else on this forum, Pink Fish, Plus and others disagree with you?

And please give these people a little more credit than usual - even the most scientific forumites are happy to tell us that their ears and the maths do not correlate.

 

Hi,

I believe you are refering to the Laser Focus mechanism (electromagnet), not the error correction/conceaqlement mechanism.

I completely agree. The discussion largely revolves around assertations by the guy with the complex that as long as THD &N and Frequency response are identical everything "sounds the same", an assertation that many feel is inaccurate.

Ciao T

 

merlin - speak for yourself. I, for one, agree with Andy. I'm not going to take the liberty of speaking for others but I know I'm not the only one who agrees with him.

It would be quite straightforward to demonstrate in a DBT that neither you, nor WM, nor anyone else for that matter could reliably tell apart two competently designed CDPs/DVDPs whose measured differences are well below the limit of human audibility (that includes a great many CDPs and DVDPs from around £250 to 10 times that amount).

Michael.

 

so you can't tell a difference between say a $200 nos dac and £1900 dac 64 level matched same cables then mike?

 

Yes, I can. However, as has already been mentioned many times my NOS DAC measures quite differently to a DAC64 (considerably worse in fact). The measured differences are well within the limits of human audibility (as is the HF drop off of Wadia CDPs).

Michael.

 

Go on then Michael, put your money where your mouth is - set it up!

Find a £200 player that measures the same as my Marantz and I will happily take part. But it's up to you to organise it and fund it, I really don't need the hassle.

 

Hi,

I did not claim it was a miracle, I used it to illustrate how somesthing may be audible even if it cannot be measured reliably.

The Human Ear/Brain system is highly nonlinear and selective. It results in the fact that certain things are audible under conditions where we cannot employ (yet?) artificial means to accomplish the same feat with machinery.

I would suggest that if documented evidence exists that other issues are audible that would kinda sink your little pet theory, wouldn't it?

Extensive studies over the last 100 years have consistently revealed that the human ear is at least 10 times more sensitive to IM distortion than harmonic distortion.

A quick gogle search on "jitter audibility studies" reveales quite a few relevant results anyway.

Now if you promised to shut up and to stop engaging in counter factual argumentation if I provided such I'd even bother to dig such references out. But as this seems unlikely forgive me for not bothering.

Yes, that was what I referred to.

I never suggest that it was a paradox nor am I particulary unaware of this (I was well aware of the issues a decade before Mr. Geddes piublication). I merely pointed out that certain extremely large levels of Harmonic distortion may be percieved as actually leaving the signal UNDISTORTED in subjective terms, while small amounts of other types of harmonic distortion can result in drastically audible distortion.

In Mr. Geddes example files (I recommend downloading the and listening to them) the 0.1% THD file contians distortion so bad it in effects renders the music unlistenable. Now if of the type of distortion only 0.01% or even 0.003% would be present I would posit that we would still have an easily audible degradation and hence a claim that 0.003% Measured THD is reliably inaudible should not be made by anyone actually having a good working understanding of the matter.

Sadly extensive and credible blind studies (which means scratch out those by the ABX crowd, which regulary suffer from a lack of statistic significance due to the very low numbers of tries and participants involved) with regards to the audibility or not of any number od issues posited in audio.

With the absence of any reliable data to suggest that issues like jitter, low level but high order harmonic distortion and associated intermodulation as well as noisefloor modulation are reliably inaudible they remain on agenda, moreso if both the rather scant extant audibility research and anecdotal evidence suggest that these issues are definitly audible in certain conditions.

No, he does not. However I do.

Hmm, you CLAIM they are inaudible, there is a big difference between a baseless claim as you make and actual fact. If you have a reference to a study that does not show excessive flaws of any number of types (statistical, experimental execution) which allows you to conclude that an amount of 0.003% THD is reliably inaudible, why not provide the reference?

The problem with blnd listening tests is that the majority of them are poorely exexcuted and use too low numbers of participants to give statistically relevant results (in fact, most of them MUST RELIABLY RETURN null results irregardless of any audibility of the issue under test due to inaccurate/inappropriate statistics applied). To conduct tests that would provide results that are statistically relevant takes an amount of investment in time, money and effort that few people can expends and corporations who have the resources only apply if they percieve that there will be profit in the end. Given the absolutely low priority of High Quality music reproduction for all but a tiny minority of people the potential return is predictably low, so no-one bothers with the single exception of perceptual coding research, which BTW has yielded up quite some good leads.

No, I am coming from the angle that there is large body of evidence suggesting that current commonly taken measurements are not adequate or reliable indicators of sound quality.

How so? In fact, having done the same for some o my wifes CD's (she is terrible for getting them VERY scratched) the improvents where audible to her quite without prompting and without her knowing what I had done. If people wish to work on doing tests themselves it is obvious that they should conduct such test blind, I did not feel it particulary relevant to state.

Please note that I mentioned this as personal experience. And I stand by my observations. I used to work as sound engineer (live & recording) and a pretty good ability to HEAR differences when they occour is needed for that work (it sadly seems rare these days though). So I am rather certain of what I heard. For fun, I did a DBT once that a german guy wanted me to take. I provided him with EAC extracts of 3 pieces of music.

He burned a CD with some of them MP3 at varying datarates and some original Wav format. Unknown to our DBT advocat I had allowed me a control, by selecting only HDCD enecoded titles. At the time I myself had no HDCD Player but I did sit through the extremely stressing and annoying process of hearing the same excerpt 16 times and marked the excerpts that I felt as being the Original (unaltered) ones. I did note that even these copies showed a decrease in sound quality over my original CD's, but that they sounded the "least worst" of the lot.

I posted my results in the discussion group only to be told that I had gotten it all wrong. I then revealed my "control" and took the CD to a friend who had an HDCD equipped player. As it turned out I had correctly identified all but one of the original tracks, which could be easily identified by the HDCD indicator lighting up.

I promptly posted this result, only to be banned from this guys discussion board and to have the whole thread deleted.

At any extent, I was able to reliably note quite small degradations of sound quality, as most MP3 processed Exceprts had been lame'd at the highest datarate and no 128KBPS or lower examples had been provided.

If you had any such that indicated that such levels where reliably inaudible than I would. Given that neiter specifically existy we have live with the ambiguity of having no absolute proof. Personal experience and anecdotal eveidence suggest that two different pieces of equipment with a flat frequency response 20Hz-20KHz and 0.003% THD+N do not allways reliably sound the same. As a result I tend to suggest that a flat response and low THD are insufficient to correlate with the perception of sound, you obviously hold the opposite view, without any reliable evidence either. I guess we will have to wait for time to resolve the issue.

What I would like however from you is an explanation why 24Bit/96KHz recordings are reliably indentifable as superior to 16Bit/44.1KHz recordings, under blind conditions, if the levels of distortion in 16Bit Recording & Replay equipment are below the audibility treshold and if the frequency response with 20Hz-20KHz flat is already sufficiently wide?

Some notes on Jitter audibility are contained in this paper:

http://www.jitter.de/pdfextern/jitter92.pdf

Further comments on jitter including references to audibility limits:

Time Distortions Within Digital Audio Equipment Due to
Integrated Circuit Logic Induced Modulation Products"
AES Preprint Number: 3105 Convention: 91 1991-10
Authors: Edmund Meitner & Robert Gendron

Is the AES/EBU/SPDIF Digital Audio Interface Flawed?
AES Preprint Number: 3360
Author: Chris Dunn
Author: Malcolm O. J. Hawksford

I asserted that they do, bacause a large number of references exist that illustrate this to be true. Including AES Papers, including results from DB Testing in perceptual coding research.

I have no particluar interrest to do fundamerntal research to proove anything or to publish, so I'll leave that work to those who wish to do any of the above.

To me the extant body of work on the subject is sufficiently large, detailed, wellfounded to accept at least the possibility of audible differences resulting from mechnisms commonly called Jitter, noisefloor modulation which is really just a particular result of IMD anyway and so on.

I repeat, 95db A weighted THD & Noise is poor for a 16-Bit device. Note that the DVD Audio Specifications include 24 Bit audio. Dirt Cheap Pro-Audio gear (arguably with longer wordlength - but then DVD Audio is specified up to 24 Bit resolution anyway) manages much better.

A competently designed (if unexceptional IMHO) CD-Player from Arcam offers the following measured performance:

Noise (digital silence) <= -113dB ref 2.2Vrms (unweighted)
Dynamic range (1kHz, -60dB FS) * >= 93dB ref 2.2Vrms (unweighted)
* 16-bit, triangular pdf dithered data

A competently designed CD player from Musical Fidelity offers 96db S/N ratio unweighted and 105db A weighted, support9ing my assertation that applying A-Weighting to noise improves the figure buy around 10 - 15db, in that specific case the imporovement was 9db.

If we accept the 9db then the Sony Player you mentioned offers an unweighted S/N ratio of -86db or worse, that is nearly 14-Bit equivalent performance, not 16-Bit.

So, at the bottom line I still maintain that what measured performance you quote is piss poor for a 16Bit Device and awfull for a 20..24Bit device.

Ciao T

 

Hi,

It would be so indeed. All one needs to do is to design the test and statistical analysis the same way the ABX crowd does and then, as if by miracalous happenings all CD Players sound the same. All Cables sound the same. All Amplifiers sound the same.

Why?

Because the test is set up to return reliably a null result for any small to modest audible differences due excessive type 2 (the null hypothesis is inaccuratly accepted) Error potential which in turn is a direct result of applying a significance level of .05 to a small sample set, which is in fact too small to allow such a significance level without increasing the risk of type 2 statistical errors occuring to near certainty.

For more on that please consult:
Les Leventhal's
"Type 1 and Type 2 Errors in the Statistical Analysis of Listening Tests"
(JAES, Vol.34 No.6)

Untill and unless Blind test are analysed taking these issues into account we are well advised to simply disregard their results.

Ciao T