The DVD/Dac thing

[wadia-miester]

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.

Mike,

you've preformed this very test yes?, level matched to 0.1db in an inert room?, with matching equipment in a total DBT enviroment?
I'm impressed Can you share your findings please
Yes indeed the Wadia digi master 2.4 filter does roll off the treble -1db @16khz and -3db@18khz Now if Andrew Thomas (did he/or still does he work for atc?) This interpolation curve is for Algorythym 'A'. 'B' and 'C' do not contain this roll off, though this is at the expense of time domain accuracy.(Less smearing than a traditional brick wall filter, by a factor of 10 (0.17m/s))
This is part of Wadia 'serious presence' and ultra tight focus of image presenation along with the bass. It may also explain why Wadia's arn't top of true 'hifi' buyers list, due to they precieved lack of top end air/space in favour of a much more prominent musical presenation
Filters 'B' & 'C' are about as flat to 20khz as it goes, though still with a far better time domain accuracy of around 1/2 millisecond to the best of the rest at 1.7 m/s. Both of these filter give a increase in 'air' if you like.
Oh and I can tell the difference between 'A' and 'B/C'.
To be honest trying to seperate filters B and C is some what tricky.
I listen to Filter 'A'
Hope that helps.

It seems Datty's been at it else where.....

05-09-04: A_thomas
WARNING AN ATC USER RESPONDS

I've heard Wilson WP's in a very expensive Levinson setup (390S, preamp, monoblocks) and it sounded very nice.. but it was four times as expensive as the active 50's I have... maybe it was the room, maybe it was the speakers, but I prefer the 50's.

As for using valve power amps with passive ATC's - I'd say forget it. These speakers love power, and you'd be looking at 100's of watts. Moreover, just about everyone agrees the ATC's should be active..

As for pre-amps, front ends, interconnect and mains leads (and power conditioners) your going to find that ATC don't play the Audiophile game. ATC recommend Belden interconnect, I don't think they recommend special mains treatment, and I beleive the opamps used in the active crossover would give most audiophiles apoplexy.

I've used my 50's with CD players and preamps on a spectrum of prices and the systems sonic signature is totally dominated by the room and speakers. I can recommend ATC's preamps (I happen to own both), but they are rare in the US and the cheaper CA2 is truly excellent for very little money.

I've run my 50's direct from a Benchmarkmedia DAC1 and they sound superb. It's possibly the best value for money system I've ever heard.And is not beaten by the Levinson 390S direct (which I also own..)

As to why no magazine has reviewed them. Well, widescreen review were very positive about the 150's. But truth be told ATC are a small company that in the last couple of years have become much more visible in the UK with the excellent active 10's, passive 7's and 12's - they've probably got their hands full over there

Moreover, they are a small company without advertizing revenue, so there is no "incentive" for review by the US magazines - the range has been well received by the UK press. The active 50's and SCA2 preamp were reviewed in HiFi+.

Finally, ATC have a solid reputation in the pro world - they don't need accolades from the highly subjective hifi press, or the desire to put up with the audiophile BS.

Here's a copy of the widescreen review article..

http://lib1.store.vip.sc5.yahoo.com/lib/transaudiodirect/65eratc.pdf

Finally, reading reviews is pointless. You need to hear a pair, no matter what any reviewer says (even someone with charisma like Art Dudly) it's your ears that matter. Also, 150's, or any speaker with that level of bass extension and power, needs to be tried at home, otherwise you might find yourself wanting to take a bath with an electric fire.

 

[michaelab]

Thorsten, you keep harping on about Leventhal's "Type 1" and "Type 2" errors but the conclusions you draw from his paper are incorrect. ABX tests are not setup to return any particular result. If the differences we are talking about are readily audible to sighted listeners (and they claim to be) then they should be large enough not to fall foul of type 2 errors. The differences that would fall into that category would have to be so miniscule that they certainly wouldn't be readily audible to sighted listeners and it's debatable whether they actually matter if even experienced listeners can only reliably detect them very infrequently.

Take a look at this site:

http://www.pcavtech.com/abx/abx_data.htm

Here you can see examples of ABX tests where small (but nevertheless audible) differences have been reliably detected in ABX test conditions. They show that some CDPs didn't sound the same and some amplifiers didn't sound the same.

It shows that you disregard the results of ABX tests at your peril. If you choose to ignore them you are simply choosing to ignore facts that you find uncomfortable.

merlin - I don't have any reason to setup such a null test. Like you, I don't need the hassle.

Michael.

 

[3DSonics]

Hi,

Thorsten, you keep harping on about Leventhal's "Type 1" and "Type 2" errors but the conclusions you draw from his paper are incorrect.

Am I now...

ABX tests are not setup to return any particular result.

If we posit that a test is done with an n of 16 and analysed with .05 Significance, what is the likelyhood that we will overlook small but audible differences? If it it is significantly higher than 50% I would say that test is set up to return null results. It is of little relevance if you deliberatly attempted to achieve a null result or if you are merely not very good at statistics.

If the differences we are talking about are readily audible to sighted listeners (and they claim to be) then they should be large enough not to fall foul of type 2 errors.

That is interesting twist of logic but irelevant. What is relevant is that if in reality the difference heard is small (regardless what anyone claimed) it will exist but not be detected. Things are at any extent considerably more difficult to detect outside the context of a system one knows well and many issues that effect such items as sound staging and realism are not ameanable to group tests but require single individuals in the "hot seat".

The differences that would fall into that category would have to be so miniscule that they certainly wouldn't be readily audible to sighted listeners

Hmmm. Why not? If a large scale listening test with (say) 100 Participants and tests in groups of 5 trials (to avoid stress reactions) resulted in an indication that a given item is audible, why would it then not be readily be audible to sighted listeners?

My problem is that a test with a small number of participants (possibly just 1) and with 16 Trials each is highly likely to show no audible difference unless we relax our significance level considerably in order to have "even stevens" between type & type 2 errors. And once we do this we find our sample size to small to allow any significant results to be pronounced, plus the number of trials in one row is by far too large. For myself personally I found that I loose concentration around trial 5 - 6, by the time it's past I'm usually guessing because I can't be arsed anymore.

At any extent, small scale listening tests are statistically insignificant, regardless of their results and large scale tests would require people who can listen well and analytically, many of them and to have them motivated to hear differences, yet they MUST not be interested in the outcome per se, as the person convinced all sounds the same will hear exqactly that in the blind test, while the person convinced there is a difference will hear one even if non is present.

Now, how and where do you find large enough numbers of good listeners who are willing and eager to hear differences but have no stake in what the difference or lack thereoff reveales?

Take a look at this site:

http://www.pcavtech.com/abx/abx_data.htm

I am well familiar with this site and have on previous occasoins criticiesed their tests, both in terms of implementation (hell, in one test where they tried to compare CD Players the output from the CDP was passed through the AD & DA conversion of a pretty poor quality DAT recorder "to match levels accuratly", no wonder all but the most gross differences where swamped out), including the statistics applied.

Even those test that show differences are best discarded, the tests that where done by that ABX crowd are simply not signifdicant in any way, plagued by poor implementation and inappropriate statistics. I choose to disregards their test en block as a result.

Anyone who where to publish equivalent data for clinical trials or who would seek to either have approved or withdrawn medication an the bassi of such resuts would be laughed of the face of the planet as Charlatan AND JUSTLY SO. I do not see any reason why I should extend more courtesy to charlatans in the Audio world than I would to those in other fields.

Bad science is and remains bad science.

It shows that you disregard the results of ABX tests at your peril. If you choose to ignore them you are simply choosing to ignore facts that you find uncomfortable.

On the contrary. It shows that ABX testing is actually a science and that most advocates of it are unaware of the implications and limits.

merlin - I don't have any reason to setup such a null test. Like you, I don't need the hassle.

Then would you care to upfront, before soliciting people to participate in your test, set out the test procedures, the test envoironment, the statistics and how you intend to combat attention span issues as well those I touched upon in terms of emotional involvement and interest in the tests outcome?

Will you "calibrate" your DBT Setup and the listeners by first testing known "positives" to ensure that your test will actually be able to at least distinguish effects that are known to be audible?

If not I will suggest that deliberatly or not you are highly likely to set up a flawed test. And to set up and conduct such a test whose results will be as inadmissable for any conclusions as those done by the ABCX crowd would indeed be a waste of everybodys time.

Ciao T

 

[julian2002]

when this thread is eventually locked i'm going to save it onto my hard drive as an insomnia cure.
cheers


julian.

 

[Saab]

It would be a shame to lock it,I think its becoming as powerful as the Matrix

 

[wadia-miester]

Nah Ju, this thread pales into insignificance compared to rewiring a Lancia Thema Turbo mate trust me on this

 

[BerylliumDust]

when this thread is eventually locked i'm going to save it onto my hard drive as an insomnia cure.
cheers


julian.

I'm with you brother... so much talk about something that will lead us to nothing:

Pure tone measurements!

However, Wm saved the day...

"time domain accuracy.(Less smearing than a traditional brick wall filter, by a factor of 10 (0.17m/s))"

"far better time domain accuracy of around 1/2 millisecond to the best of the rest at 1.7 m/s. "

Wm,

Could you please elaborate a little more on this subject... it is the first time I actually see these numbers...

How is that time domain accuracy measurement performed?

 

[oedipus]

I'd like to respond to all your points, but for tonight I'd like to work through just one:

I repeat, 95db A weighted THD & Noise is poor for a 16-Bit device.

Let's start again with the figure I quoted for the Sony - on CD:

Noise level (re 20 dBFS, A-wtd)........ 75.6 dB

Now we can convert that to 95.6 (re 0dB)

I must admit I'm glad you remember Julian Dunn's early work on jitter (I'm familiar with his work, we'll come back to that another time) Now Julian works for Audio Precision (who know a thing or two about how to measure DAC's) - I'll draw your attention to his TechNote 25:

The noise floor of most converters is fairly flat, [so these figures indicate the difference in results that might be quoted.] The A-weighting gives the lowest noise figure and is normally the figure quoted on the front page of a data sheet. Where the noise is fairly flat you can add 2.3 dB to an A-weighted noise figure to estimate the unweighted noise over the DC to 20 kHz band.

So, now we convert the a-weighted number above to an unweighted SNR of 93.3.

Now, we recall this basic formula: 6.02n +1.76dB which tells us the signal to noise ratio (or dynamic range) for an n bit system. We have a 16 bit CD so our best theoretical SNR is 98.1.

The test signals for the quoted measurements were dithered. We recall the results of Lipshitz and Vanderkooy, that rectangular dither causes noise floor modulation, and that Guassian has more noise, so we opt for dither with a triangular pdf. The triangular dither in the signal has a 4.76dB noise penalty, so our theoretical SNR is now (98.1-4.76) 93.34dB. From this we know what the noise level should be.

Comparing this with the measured result it appears, on paper, that the Sony unit seems to be getting perfect results and this is what prompted David Ranada to remark "The audio emerged in near pristine condition. Noise levels for Dolby Digital, CD, and SACD were extremely low (right at the theoretical limit for CDs) and some of the best we've ever measured at any price level."

The co-author of this review was Ken Pohlmann, who knows a thing or two about the Principles of Digital Audio, he did after all write the book on it.

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.

On the contrary, you seem to have (deliberately) chosen a player that is incompetantly designed - I'd guess that the difference between the A-weighted and unweighted numbers is power supply noise.

 

[michaelab]

If not I will suggest that deliberatly or not you are highly likely to set up a flawed test. And to set up and conduct such a test whose results will be as inadmissable for any conclusions

You and merlin, you're as bad as each other. You're rather like Victor Yanukovych who has lost the Ukraine elections but refuses to accept the result when the facts are staring him in the face . You're even making up your excuses in advance. That's really the reason I can't don't need the hassle of setting up a test because no matter how I do it some people simply won't accept the outcome

Michael.

 

[lAmBoY]

This all feels like a bad open university program from the early 80's (used to be screened on bbc2 at 4:30am).

 

[greg]

The cynic in me points out that it is in the interests of a company that as far as I recall don't make source components, to try and reverse source first type thoughts to get people to spend more on speakers.

Makes complete sense. If a person's wallet has a limit and upselling in a market which is seeking greater value is proving difficult, then its about securing the share of wallet. Trying to convince a buyer that 95% should be spent on an atc pre and a pair of active atc speakers makes sense for atc no dount as long as they could find the right £250 player as the source for demo's. Also perhaps free's some cash up as owners of £3K machines rush to downgrade their digital sources in order to fund atc purchases.

 

[Saab]

Also perhaps free's some cash up as owners of £3K machines rush to downgrade their digital sources in order to fund atc purchases

no there's a point,although it is worthy of the CIA,it certainly makes sense

 

[3DSonics]

Hi,

You and merlin, you're as bad as each other. You're rather like Victor Yanukovych who has lost the Ukraine elections but refuses to accept the result when the facts are staring him in the face .

That's a rather low blow. And as to WHO is the one missing the facts, that remains debatable.

I retain that it has been illustrated conclusivley that small scale blind listening tests do not allow statistic analysis to a level of significance that would give any confidence in the results as being reasonably freefrom both type 1 and type 2 statistical errors. That's a simple fact of life.

If you insist to still conduct such a test you must be aware that you are wasting your time unless you willing to accept a lot of abiguity, which in turn again makes the value of the test questionable.

Now despite these simple facts you persist in proposing such tests and you persist in proting the results of such tests as holding significant weight. It really rwaises the question whose head is in the sand here.

You're even making up your excuses in advance.

No excuses. I merely point out to you that such a test, unless implemented first of in a fashion that ensures statistically relevant information and secondly conducted to maximise the chance to discover any differences present (if indeed they are) the test is worthless, regardless of outcome.

That's really the reason I can't don't need the hassle of setting up a test because no matter how I do it some people simply won't accept the outcome

Now there is a copout. I am very happy to help you to set the various criteris such that the test can give relevant results. You only need to have at least around 20 Participants and you need to make sure that each of them has short and relaxed listening sessions of no more than 5 trials per session and that this happens in an acoustic/system envoironment they are familiar with.

This would suggest that you bring a suitable switchbox that is definitly sonically transparent with the option to transparently match levels to better +/-0.1db to peoples own houses, there conduct the tests preferably while including in each test a set of controls (which are NOT announced as controls to the listener of course) using a definitly audible change just to weed out the tinears and do that with 3 sets of 5 Trials for the actual test, in order tohave control data two runs of controls should be done as well, so overall 5 rounds of 5 trilas, should be doable with setup and all in around 2 Hours.

Repeat till you have 20 Listeners that showed significant audibility of the controls and then show that 300 Sets of Data procured show no significant correlation of the listeners reactions to the actual player player played (or in fact that theyr reactions do correlate to a statistically significant level) and you will have produced a test that actually has some value and meaning.

Most other approaches to testing, the absence of controls to ensure the test setup is actually transparent enough etc makes tests in my view invalid on procedural problems and to use a very small number of datapoints (< 100) to provide the test data leads to problems with the statistical significance.

Do you really dispute that the above contentions are valid?

Ciao T

 

[3DSonics]

Hi,

Noise level (re 20 dBFS, A-wtd)........ 75.6 dB

Now we can convert that to 95.6 (re 0dB)

Okay, I dropped off the .6 of a db.

I must admit I'm glad you remember Julian Dunn's early work on jitter

I am. I am also familiar with the noisefloor vs. frequency plots of DA Converters and his statement that the noisefloor is fairly flat actually is inaccurate, as can be seen from the plot I linked in earlier from stereophile. Practically all converters, but definitly those using some form of noiseshaping and Delta Sigma will show a significant rise in noise levels with frequency and it is exactly there where A-Wighting gets it's improvement in the case of DA Converter where no significant mains induced noise is present.

The test signals for the quoted measurements were dithered.

A fact you conveniently forgot to mention. Once dither is used the noise and distortion levels are largely equalised between players, regardless of perfomance beyond a certain point as in effect dither introduces a constant low level noise floor. As a result such measurements will only reveal extreme blunders in design.

HOWEVER, given the nature of the dither and the usual frequency dependent level of it (beacuse dither certainly does not create a "flat" noisefloor) I would still expect a better a-weighted result.

At any extent, the particular results you presented where not qualified as to their limits, so with you failing to supply these qualifications you have again merely illustrated how meaningles a few one dimensional numbers are, unless correctly qualified and correlated to what people actually hear.

Ciao T

 

[michaelab]

I am very happy to help you to set the various criteris such that the test can give relevant results. You only need to have at least around 20 Participants and you need to make sure that each of them has short and relaxed listening sessions of no more than 5 trials per session and that this happens in an acoustic/system envoironment they are familiar with.

This would suggest that you bring a suitable switchbox that is definitly sonically transparent with the option to transparently match levels to better +/-0.1db to peoples own houses, there conduct the tests preferably while including in each test a set of controls (which are NOT announced as controls to the listener of course) using a definitly audible change just to weed out the tinears and do that with 3 sets of 5 Trials for the actual test, in order tohave control data two runs of controls should be done as well, so overall 5 rounds of 5 trilas, should be doable with setup and all in around 2 Hours.

Repeat till you have 20 Listeners that showed significant audibility of the controls and then show that 300 Sets of Data procured show no significant correlation of the listeners reactions to the actual player player played (or in fact that theyr reactions do correlate to a statistically significant level) and you will have produced a test that actually has some value and meaning.

Most other approaches to testing, the absence of controls to ensure the test setup is actually transparent enough etc makes tests in my view invalid on procedural problems and to use a very small number of datapoints (< 100) to provide the test data leads to problems with the statistical significance.

Do you really dispute that the above contentions are valid?

No. I don't dispute them. Apologies about my Yanukovich comment.

I believe that (properly run) ABX testing is the only way to determine in an unbiased way what humans can and can't distinguish amongst audio components. AFAIK no one has come up with a better way. IMO enough of such tests have been run in the past to show that many things that are claimed to be audible (even claimed to make "huge" differences) are not actually audible or, at best, have an effect that's so subtle one has to question it's value.

Clearly I don't use such tests when choosing audio equipment. I choose based on many subjective criteria just like everyone else. However, at least I'm aware of what I'm really paying for and not hoodwinked into believing all the hyperbole that surrounds the hifi industry.

I'm bowing out of this thread now as I don't really feel there's any mileage in my continuing in it. I don't actually think there's much mileage in continuing the thread at all but I'll let it run for a bit more

Michael.

 

[3DSonics]

Hi,

Apologies about my Yanukovich comment.

Thank you.

I believe that (properly run) ABX testing is the only way to determine in an unbiased way what humans can and can't distinguish amongst audio components.

I agree with you. However my experience with such tests leads to me to place the barrier for "properly run" several floors above where you seem to accept it.

I once demonstrated the problem of ABX teststing to a strong advocate of the principle. I suggested we where hearing different cables, when in fact I was reversing the polarity of one channel only. The testsetup was not ideal either to be able to hear anything, but surely such a gross difference should have been audible? Yet our friend scored exactly random while two cable "believers" scored 100%. After I let him in on the little change of protocol I had omited to tell him upfront he was furious, claimed I had tricked him and did not converse with me for month.

What it proves is that if you include people in a test with an obvious bias as to the outcome theyr will not be usuable. It ultimatly places you in a position of requiring controls to eliminate such listeners from the data. The list of possible pitfalls is endless.

IMO enough of such tests have been run in the past to show that many things that are claimed to be audible (even claimed to make "huge" differences) are not actually audible or, at best, have an effect that's so subtle one has to question it's value.

IMHO these tests have been so severely falwed in implementation, process and subsequent analysis that theiir value is zero, below zero given the unscientific way the results of these tests are utilised by some.

There have been tests run that where blind and qualifiy under the rules of "properly run". They usually suggest that a lot of very low level stuff is audible, sadly non where specifically aimed at "voodoo" in HiFi, they invariably supported "serious" research into perceptual coding, watermarking and general hearing limits. Especially the watermarking research was interesting, as it seems to be not possible to devise a "robust" alteration of the signal which survives sample rate conversion of analogue re-recording and at the same time is reliably inaudible. Now that should give tome of the ABX crowd time for a pause....

Could we have some really seriosu studies we might find indeed that for example cheap DVD Players sound worse than well designed dedicated CD Players, however this would not be in the interest of the main players in the electronic inductry who actually would have the resources to do so., They are (on both sides of the debate) happy with the status quo.

Ciao T

 

[BerylliumDust]

I believe that (properly run) ABX testing is the only way to determine in an unbiased way what humans can and can't distinguish amongst audio components.

Michael.

Michael,

Allow me to disagree with you in that I think biased listening is NOT always a bad thing for equipment comparison (although I do believe in properly run blind tests). Actually, it can be utterly revealing...

 

[wolfgang]

What it proves is that if you include people in a test with an obvious bias as to the outcome theyr will not be usuable. It ultimatly places you in a position of requiring controls to eliminate such listeners from the data. The list of possible pitfalls is endless.

Interesting story. I presuame it is suppose to be extremely profound but very obvious is it not? If a person is biase then his listening score is most likely to be under suspect.

When comparing these flawed DBTs, did all the none blind tests have big enough participants to prevent these type 2 errors then?

 

[3DSonics]

Hi,

Interesting story. I presuame it is suppose to be extremely profound

No, it is to highlight the simple fact that if your listeners are biased a BD Test is much harder to get right.

Other experimenters have observed the opposite side of the coin where people convinced there was a difference scored one even though non was present.

This means that the typhical challenge to audiophiles "proof to me you can hear the difference" is also flawed.

When comparing these flawed DBTs, did all the none blind tests have big enough participants to prevent these type 2 errors then?

No, but the ABX Test advocates have already dimissed them entierly already, so they don't count.

Ciao T

 

[notaclue]

Other experimenters have observed the opposite side of the coin where people convinced there was a difference scored one even though non was present.

Yes. John Dunlavy (ex-speaker manufacturer) tells the tale of how he fooled people in a sighted cable test by showing them impressive looking speaker cables and pretending to hook them up. In fact, the same cable was used throughout the test.

The audiophiles present heard and described (often) big differences and improvements in the sound. But the sound never changed.

Personally, I used to believe in cables and tweaks etc. but now I do not. When I did believe, I heard differences between cables, tweaks etc. Now I don't believe, and I don't hear differences (I tried a couple of interconnects and speaker cables and despite previously hearing a difference, now I would honestly say they sound the same to me).

What puzzles me about the more extreme subjectivists is that they seem to discount the possibility that they are capable of imagining differences in audio equipment. I think it is a very real possibility. Human beings are much more complex and variable than copper interconnect cables. Well... most human beings are. Certainly, I believe I used to imagine changes in sound that were not really there.

That is why I believe we need some form of test whereby people are unaware of the identity of what they are listening to in order to verify if claimed differences in sound really do exist or not. I find it hard to accept that if things that measure the same do not sound the same that (as far as I know) there is not at least ONE person (on planet earth) that can reliably and repeatedly demonstrate this in a double blind test.

Just look at 'Record Breakers' to see what extraordinary things some people can do. Really amazing things. Surely passing a cable ABX test, for example, isn't that difficult if they sound different? Surely we could find at least one person who could do this? Or surely some other form of 'blind' test could be invented if the A-B/ABX ones are flawed?

That is what I want to see.