Bits Is/Ain't Bits: A Modest Proposal

[Fitzcaraldo215]

Maybe this has been done somewhere, but I have not encountered it. I am reasonably new to computer audio after many decades as both a high end audiophile and a computer professional.

We have two entrenched camps: bits is bits/bits ain't bits when received by the DAC unless major steps have been taken. Computer science everywhere since the Eniac has been predicated on the ability to engineer verifiably reliable data systems where absolute bit-level integrity is maintained. Or, if it cannot be maintained, the system identifies it and self corrects in various ways or else the system spits out an error message or halts completely.

The flip side is computer audiophiles claiming subtler forms of audible, undetected bit level corruption on the transmission path from digital storage devices to the computer and on to analog conversion inside the DAC. These claims seem to be based on theoretical, but unproven, arguments and mostly uncontrolled subjective listening. They suggest a number of possible sources: hard discs themselves (magnetic vs. solid state, 3.5" vs. 2.5"), computer RAM (the DDR3 Row Hammer), other computer hardware ( power supplies, USB cards, etc.), operating systems, operating system tweaks and optimization, player software, networks, communications protocols, cables, jitter, etc., etc.

It strikes me that the arguments by the Ain'ts are substantially weakened because they have no measured proof to back up the prevalence of bit flipping, including the extent to which it actually occurs, if at all. That is in contrast to the long history of established digital technology on which we and the world economy rely almost completely. Similar claims of hidden, undetected bit corruption also just do not seem to exist anywhere but in computer audio.

It strikes me as possible, though not necessarily easy, to build a computer program that would receive and record the audio stream coming from the playback computer configuration. That captured stream could be compared bit by bit to the original. The receiving test computer would replace the DAC, in other words, in a test computer linked to the playback computer configuration being tested, just as the DAC would be. Playback of selected musical passages could be performed and the exact bit by bit discrepancies, if any, could be reported by comparing the received and original files. Yes, there would be minor format differences (headers, metadata, etc.) between the original and the transmitted file, but the all important music signal should be identical, ideally. The playing computer hardware/software configuration, cables, etc. could be varied and comparative results obtained.

This could be a very powerful tool in proving/disproving once and for all to what extent the problem exists and what specific measures one should take to most effectively improve quality performance. It could serve as a valuable benchmark for a host of computer audio improvements. It does not deal with what goes on inside the DAC, but it would greatly help in identifying any and all problems upstream of that, to the extent they exist.

So, on behalf of all the agnostics out there like me, what are the chances of this? Difficult, perhaps, but not impossible, it strikes me. It might create countless long threads of new arguments about it, but it should end many other current threads of hostile arguments for or against bit flipping based on no concrete evidence that has been published to my knowledge.

I am aware of the Rightmark Audio Analyzer, as used by Archimago in his intriguiging blog. I have no experience with it. However, it seems to require analog input to the test PC via a sound card. It does not seem to measure external devices at the bit level, but rather in analog terms (THD, noise level, etc.) after a-d conversion. But, I might have thought that even this tool might have been used to further claims of "improvements" to bit perfection as claimed by many. I am not seeing those in my search of the web.

Sent from my iPad

[sbgk]

Maybe this has been done somewhere, but I have not encountered it. I am reasonably new to computer audio after many decades as both a high end audiophile and a computer professional.

We have two entrenched camps: bits is bits/bits ain't bits when received by the DAC unless major steps have been taken. Computer science everywhere since the Eniac has been predicated on the ability to engineer verifiably reliable data systems where absolute bit-level integrity is maintained. Or, if it cannot be maintained, the system identifies it and self corrects in various ways or else the system spits out an error message or halts completely.

The flip side is computer audiophiles claiming subtler forms of audible, undetected bit level corruption on the transmission path from digital storage devices to the computer and on to analog conversion inside the DAC. These claims seem to be based on theoretical, but unproven, arguments and mostly uncontrolled subjective listening. They suggest a number of possible sources: hard discs themselves (magnetic vs. solid state, 3.5" vs. 2.5"), computer RAM (the DDR3 Row Hammer), other computer hardware ( power supplies, USB cards, etc.), operating systems, operating system tweaks and optimization, player software, networks, communications protocols, cables, jitter, etc., etc.

It strikes me that the arguments by the Ain'ts are substantially weakened because they have no measured proof to back up the prevalence of bit flipping, including the extent to which it actually occurs, if at all. That is in contrast to the long history of established digital technology on which we and the world economy rely almost completely. Similar claims of hidden, undetected bit corruption also just do not seem to exist anywhere but in computer audio.

It strikes me as possible, though not necessarily easy, to build a computer program that would receive and record the audio stream coming from the playback computer configuration. That captured stream could be compared bit by bit to the original. The receiving test computer would replace the DAC, in other words, in a test computer linked to the playback computer configuration being tested, just as the DAC would be. Playback of selected musical passages could be performed and the exact bit by bit discrepancies, if any, could be reported by comparing the received and original files. Yes, there would be minor format differences (headers, metadata, etc.) between the original and the transmitted file, but the all important music signal should be identical, ideally. The playing computer hardware/software configuration, cables, etc. could be varied and comparative results obtained.

This could be a very powerful tool in proving/disproving once and for all to what extent the problem exists and what specific measures one should take to most effectively improve quality performance. It could serve as a valuable benchmark for a host of computer audio improvements. It does not deal with what goes on inside the DAC, but it would greatly help in identifying any and all problems upstream of that, to the extent they exist.

So, on behalf of all the agnostics out there like me, what are the chances of this? Difficult, perhaps, but not impossible, it strikes me. It might create countless long threads of new arguments about it, but it should end many other current threads of hostile arguments for or against bit flipping based on no concrete evidence that has been published to my knowledge.

I am aware of the Rightmark Audio Analyzer, as used by Archimago in his intriguiging blog. I have no experience with it. However, it seems to require analog input to the test PC via a sound card. It does not seem to measure external devices at the bit level, but rather in analog terms (THD, noise level, etc.) after a-d conversion. But, I might have thought that even this tool might have been used to further claims of "improvements" to bit perfection as claimed by many. I am not seeing those in my search of the web.

Sent from my iPad

 

don't think anyone is saying that the bits are corrupted and/or different at the dac, what seems to be observed is that the sound is different. Why, has yet to be answered. If measurements were found that correlated with the observed sound then an understanding might be obtained. Think people have said they have measured differences, but they haven't been adopted by the majority yet. Good luck.

[esldude]

Others and myself have done what is the equivalent of your proposal.

 

For nitty gritty details, I used a laptop feeding an Audiophileo USB/SPDIF converter. The output of that fed the SPDIF digital input on an M Audio Audiophile 24/96 sound card in a desktop computer. The digital input to the SPDIF port of the Maudio card was recorded with a sound editor. All digital no conversions to analog.

 

At one point of I had a 30 minute test file. I marked each end so everything could be lined up afterwards to chop off ends that didn't belong to the original file. I sent hours of this file over that and similar setups. Compared the result in two ways. One was to take the original file and recorded file. Invert one and mix them together. The result was nothing but a file full of zeroes. Nothing there no errors. I also did an MD5 hash of the original and recorded file. Hash values matched so again no errors.

 

Over time I used different laptops. Different desktops. Loaded computers with heavy tasks during the process. Had as much going on the USB ports as possible. Put a large gaming computer power supply outside the case and exposed different pieces of equipment to the raidiated noise by placing it next to it, and by running either the USB or an SPDIF extension past it. Used different playback software. None of it made any difference in the bits. No errors, not one bit wrong.

 

Even the people saying these things sound different admit it is not due to bit errors. That as a source of sound difference has been effectively eliminated.

[sandyk]

Even the people saying these things sound different admit it is not due to bit errors. That as a source of sound difference has been effectively eliminated.

 

The cause isn't due to bit errors, neither will recording the analogue output as a digital file of higher resolution (e.g. 24/96, show any measurable differences in the actual waveforms using DIGITAL means. I have tried that on many occasions, and found Sound Forge 9 inadequate for completely consistent results, even between some files with identical checksums that sound a little different.

I have also speculated that as it is the low level HF detail, ambience and wideband noise levels that appear to be different when listening. e.g. some female voices sounding very slightly higher pitched, perhaps due to less masking of very low level harmonics, that we may need to record at a much higher sampling and depth rate, perhaps using DACs with much higher precision and S/N than commonly used for PC measurements via a soundcard fed by a noisy internal SMPS.

We need to be able to measure the differences directly at the output of the DACs using analogue methods, but that is far from easy, and most current EEs have very little expertise in these areas.

We should remember too, that "Mr.USB" (Gordon Rankin) has also heard these differences via the output of USB DACs but is so far unable to explain what he is hearing .

Alex

[Allan F]

"Mr.USB" (Gordon Rankin)

 

For anyone unfamiliar with Mr. Rankin of Wavelength Audio, he was the one who developed the original code for digital playback via asynchronous USB.

[mayhem13]

For anyone unfamiliar with Mr. Rankin of Wavelength Audio, he was the one who developed the original code for digital playback via asynchronous USB.

 

Great.....woohoo......hoooRay......Another backwards step for Hifi. We're already on the precipice of a new USB implementation and a connector. Planned obsolescence works wonders for this industry..........and still able to build success AND avoid HDMI license fees.........of which USB is a joke when comparing capabilities.

[PeterSt]

It strikes me as possible, though not necessarily easy, to build a computer program that would receive and record the audio stream coming from the playback computer configuration. That captured stream could be compared bit by bit to the original.

 

As was already said - that has been done. I tend to say "obviously" because it is the first thing to do / check when the bits=bits is not trusted. It is only that this is not for everybody. But I have done it too.

 

And so I am afraid that instead of eliminating 1000s of threads, you just created another one.

[Fitzcaraldo215]

Fair enough, and thank you all. I am relieved to have this confirmed, not that I actually believed anything else myself. However, there are numerous others, generally neophytes who do insist that bits ain't bits, but they do not have the credibility of those who posted above. I will gladly refer those Ain'ts to this thread in future.

 

So, if bits is bits, as we agree, it could only be their timing, aka jitter, or else it is PC noise leaking into the DAC and affecting the conversion to analog. No other rational conclusions are possible, as far as I know. Jitter is measurable, as is noise, which brings us back to Rightmark and the Dunn J-Test. Why do I not see the results of these tests published alongside claims of "improved" sonics or at least claims that such and such changed the sound? Back to Archimago, who uses these tests routinely in his blog. He has debunked a number of myths to my satisfaction in so doing.

 

It also seems to be true that none of the listening tests I see upon which these claims are based are any better than the casual, uncontrolled golden ear reviews in hi fi mags. It is possible that changing certain things in the configuration prior to the DAC might change the sound, but it should not be too difficult to prove a change in jitter or noise, etc. occurred to support that. The tools for such proof do exist.

[PeterSt]

First response to

 

So, if bits is bits, as we agree, it could only be their timing, aka jitter, or else it is PC noise leaking into the DAC and affecting the conversion to analog. No other rational conclusions are possible, as far as I know.

 

I for one fully agree on that.

[alfe]

First response to

 

 

 

I for one fully agree on that.

 

As often I fully agree with you even calculating a bit depth for DSD:)

[PeterSt]

Then this :

 

Jitter is measurable, as is noise, which brings us back to Rightmark and the Dunn J-Test. Why do I not see the results of these tests published alongside claims of "improved" sonics or at least claims that such and such changed the sound?

 

Oh you do, if you know where to look. Could put up something of it myself, but not so interesting I think. This is more :

 

Back to Archimago, who uses these tests routinely in his blog. He has debunked a number of myths to my satisfaction in so doing.

 

No, he (and others) only proved that he doesn't know how to measure the ultra low detail of it in order. And how can you be satisfied, when doubting all yourself ? I mean, now you know that bits=bits your quest is over ? No changes to sound possible anywhere ? nah ...

So since I have still in mind where I posted it the other day, here :

 

http://www.computeraudiophile.com/f8-general-forum/ddr3-flipping-bits-software-without-accessing-them-memory-known-failure-mechanism-called-row-hammer-22815/index13.html#post388547

 

What you see in there (second picture) is the activity, measured at the analogue output of a DAC, implied by the movement of a running time cursor on the GUI of playback software. Happens exactly once per second, when that cursor progresses.

But since you mentioned that thread yourself, I guess you have already seen it.

 

Maybe it helps you - I don't know.

[PeterSt]

As often I fully agree with you even calculating a bit depth for DSD:)

 

But then you are special !

haha

[YashN]

The flip side is computer audiophiles claiming subtler forms of audible, undetected bit level corruption on the transmission path from digital storage devices to the computer and on to analog conversion inside the DAC.

 

You have serious misunderstandings of what one supposed 'camp' is saying...

 

There are changes in SQ in most systems even if the bits are bit-perfectly transmitted to the DAC receiver.

 

Nothing to do with corruption at all.

 

Read everything Swenson has shared, or you can also read the DDR3 thread more carefully (the thread has interesting exchanges by some of the most knowledgeable and experienced contributors to Computer Audiophilia) and goes much further than just the bit-flipping phenomenon.

[YashN]

No, he (and others) only proved that he doesn't know how to measure the ultra low detail of it in order.

 

Totally agree.

[YashN]

We need to be able to measure the differences directly at the output of the DACs using analogue methods, but that is far from easy, and most current EEs have very little expertise in these areas.

 

I figured this isn't easy to do at all as just the requirements for standardisation (room response, equipment, etc...) would be difficult to obtain or enforce. Some of the equipment isn't even available for modest budgets, so they would be out of reach for most of us. Maybe large manufacturers can do that easily.

 

On the other hand, people like John, Miska and PeterSt have done interesting and worthwhile measurements which provide clues to what happens in the various computer audiophile chains.

[Fitzcaraldo215]

Then this :

 

And how can you be satisfied, when doubting all yourself ? I mean, now you know that bits=bits your quest is over ? No changes to sound possible anywhere ? nah ...

So since I have still in mind where I posted it the other day, here :

 

http://www.computeraudiophile.com/f8-general-forum/ddr3-flipping-bits-software-without-accessing-them-memory-known-failure-mechanism-called-row-hammer-22815/index13.html#post388547

 

What you see in there (second picture) is the activity, measured at the analogue output of a DAC, implied by the movement of a running time cursor on the GUI of playback software. Happens exactly once per second, when that cursor progresses.

But since you mentioned that thread yourself, I guess you have already seen it.

 

Maybe it helps you - I don't know.

 

I do not doubt myself. My opening post only hinted at my own understanding, but it tried to leave the door open for alternative points of view. I never believed bits ain't bits. And, no, my quest is not over. As I said, and you agreed, there is still jitter and noise. But, I am still quite dissatisfied with the level of intelligibility of any measured proof that a particular approach might be "better".

 

Sorry, but your graph is not comprehendable or clear. One axis is unlabeled and I really do not know what the line and points on the graph are intending to portray. And, it includes a DAC of some origin in the test playback chain. Personally, I think DACs differ quite a bit, including in their jitter and noise susceptibility. The graph might point to jitter or timing issues, but you state it indicates dropped samples, which might be related. But, dropped samples is consistent with bits ain't bits. It is not clear where they were dropped. So, sorry, I am confused. It does not help me at all.

[Fitzcaraldo215]

You have serious misunderstandings of what one supposed 'camp' is saying...

 

There are changes in SQ in most systems even if the bits are bit-perfectly transmitted to the DAC receiver.

 

Nothing to do with corruption at all.

 

Read everything Swenson has shared, or you can also read the DDR3 thread more carefully (the thread has interesting exchanges by some of the most knowledgeable and experienced contributors to Computer Audiophilia) and goes much further than just the bit-flipping phenomenon.

 

Ok, we agreed bit corruption, as was implied by the Row Hammer paper, is a non-issue. Was that just a red herring? Sorry, but much of the thread is simply too esoteric and beyond my comprehension.

 

I cited jitter and noise as issues that can affect sound quality, even if the bits are perfect. Are there other causes?

 

I have read a number of Swenson papers. Again, I see no measured evidence that the theoretical issues he claims to exist are prevalent in well engineered DACs or computer configurations. I did see him also say that his measuring equipment was older and limited.

 

I do see dismissals here and elsewhere of standard tests, Dunn J-test, etc. which I do not believe are difficult or expensive. Instead, there is the suggestion that subtler or more esoteric measures are necessary, but they are too difficult or expensive, ignoring the accepted tests used by audio engineers.

[seeteeyou]

Master clock

Word clock

RFI inside/outside equipment

EMI inside/outside equipment

Grounding

Isolation

Temperature

Humidity

Internal cable length

Internal cable conductivity

Metal versus optical

5.01234V versus 4.98765V (steady supply of precise voltage)

Interaction between audio equipment and measuring equipment

 

Some of them should be obviously related to jitter / noise, and then we'll have to figure out the difference between correlation and causation etc.

 

Ultimately there's still this philosophical question - are we listening to music (emotional>logical) or listening to gears (logical>emotional) or a little bit of both?

[sandyk]

Master clock

Word clock

RFI inside/outside equipment

EMI inside/outside equipment

Grounding

Isolation

Temperature

Humidity

Internal cable length

Internal cable conductivity

Metal versus optical

5.01234V versus 4.98765V (steady supply of precise voltage)

Interaction between audio equipment and measuring equipment

 

Some of them should be obviously related to jitter / noise, and then we'll have to figure out the difference between correlation and causation etc.

 

Ultimately there's still this philosophical question - are we listening to music (emotional>logical) or listening to gears (logical>emotional) or a little bit of both?

 

Seeteeyou

You forgot internal cable shielding.(smile)

It can be readily demonstrated that the use of SATA3-6GB/S cables , no longer than necessary, can often result in an audible improvement over generic SATA cables of longer length that may be draped across the motherboard.

The SATA3 cables contain 2 separate screened cables, whereas the generic SATA cables contain 7 wires side by side.

If you have a resolving system , you don't need to pay big $ for Paul Pang SATA cables.

Try for example the shortest possible SATA3 6GB/s cables, (they are inexpensive, and 25cm may be adequate) and see if you can hear an improvement over the typical much longer generic SATA cables.

 

Alex

[PeterSt]

The graph might point to jitter or timing issues, but you state it indicates dropped samples, which might be related.

 

No idea where you read that. Most probably you combined two posts about different subjects and now somehow think such graph shows dropped samples. Maybe it can be my conclusion that you linked to that DDR3 thread but actually never read it ? Or maybe you did but it is all still a bit or byte beyond you ?

 

One axis is unlabeled and I really do not know what the line and points on the graph are intending to portray.

 

 

 

 

Y-axis (vertical for today ) shows the highest resolution possible of the level/amplitude, without over-zooming. It means that one screen pixel (vertically) occupies one bit value or 1 out of 65536 (plus and minus together) since this was about 16 bit data. Red line is the reference so all what's on that line is output as intended.

 

The deviations in the first half of the graph is to be regarded as unavoidable random noise, say caused by the ADC process (story is a bit longer).

 

Horizontally one pixel can show one sample, but it depends on the zooming. When the whole image of this would be something like 1000 pixels wide while 0.1 second shows, one pixel would represent 0.1/1000 = 0.0001 seconds = 1/10000 of a second = 100us = 5 samples for 44100 which sampling rate is in order here. Because 1 pixel represents 5 samples, obviously you see 1 out of 5 only and 4 you do not. Zoom more and they will pop up, but then the general picture gets lost (for this purpose, which is showing what the "activity" of a running time cursor implies and which total activity was to fit on the screen).

 

With some experience, and if you observe from some distance, you can see that there's a wave pattern of deviation, which can become "logical" when electrical behavior is observed; say that when first a deviation is upwards (too much plus) then logically a next is too much minus. This is how sampling as such works and how actually the sampling rate of the ADC is not high enough to create a more dense curve (see the right hand side where the impact gets less and the deviations get less steep so now that curve shows better). But this is not important, and actually convenient, because now we see the deviation better which always springs from something. In this case that cursor movement. In other words, we can try to dedicate what we see to the DAC/ADC relation (DAC creates the output, ADC reads it back into digital format) but what remains is that sudden activity, exactly when the cursor moves position.

 

If we roughly say that the maximum excursion we see in one direction is 50 pixels, then this deviation is this 50 on 32768 or 0.15%. Of course this is the maximum (we observe here) and it is clear that many deviations are a few pixels only, say 3, or 0.009%. The deviations occur almost from sample to sample and what you can see is that when averaged over more samples, it is fair to say that the average deviation will be zero. This is almost logic when we again observe the behavior as a function of time (mismatch).

 

Although I think it would be possible (to make it myself I mean), I don't think any general FFT means will be able to show this, just because the bins required have to be able to capture 1 sample. Logically bins as small as 44100 Hz (or 22050 I think) would be able to do that, but normally the bins will be much wider. And thus (think the averaging) you won't see a thing of this with the "nulling" tools available (they will be FFT based).

 

If we take this one step further, we can try to look at before mentioned time mismatch as jitter. But I don't say it really is, because electrical behavior only, could imply what we see already. To that regard I have other graphs which include the current required for for say such an excursion, and from that you would be able to see that what actually happens is a lag. That shows that the more current is needed, the more lag occurs. Interestingly (perhaps) I would call that jitter just the same, although I'd tend to say that it happens purely in the analogue domain. Regarding this, think about a speaker driver and the higher the transient (think of a woofer for your convenience) the longer it will last until the excurs to its full swing. Call that jittery between the lower and the higher transient (but far sought).

Same happens here but now implied in the 100% electrical domain (no physical movement in order).

 

Remember, what I just said happens always (think slew rate), but here it is implied by that moving cursor. So this is really at a micro level, hard to measure otherwise.

[Fitzcaraldo215]

Master clock

Word clock

RFI inside/outside equipment

EMI inside/outside equipment

Grounding

Isolation

Temperature

Humidity

Internal cable length

Internal cable conductivity

Metal versus optical

5.01234V versus 4.98765V (steady supply of precise voltage)

Interaction between audio equipment and measuring equipment

 

Some of them should be obviously related to jitter / noise, and then we'll have to figure out the difference between correlation and causation etc.

 

Ultimately there's still this philosophical question - are we listening to music (emotional>logical) or listening to gears (logical>emotional) or a little bit of both?

 

 

I am am sure that your list could go on an on with things that MIGHT change audio playback. But, I agree with you: we are listening to music. It therefore seems to me that we should be measuring the final result, which is audio coming out of a DAC at the end of a long and complex chain of digital processing and data transmission. There are accepted ways of measuring audio performance in audio engineering: frequency response, distortion, signal to noise, jitter spectra, etc., etc. Hardware or software configuration changes could be, rather should be, measured and evaluated based on those. For example, does a different USB, Ethernet, power, etc. cable make any difference in measured audio performance? Does a player software change make any measured difference? Does an OS tweak make a measurable difference in the sound? Does a CAPS PC, a Baetis, an Apple, etc. measure better than a plain vanilla PC?

 

As always in audio, there will be those who will say measurements do not tell the whole story. That is possibly true, depending on what audio performance parameter is measured. But, even so, what do the standard, accepted measurements actually say before we get into arguments about what they do not say?

 

I have a technical background, but I just want to be a consumer of computer audio, not an experimenter or developer. I just want to put my feet up and enjoy the music. Unlike many others, my whole music PC setup, including my DAC, network and NAS, is just a big music (and video) player to me. I am really, really delighted with how terrific it sounds. But, people are always wanting to sell me or have me try something "better". Ok, but why should I take precious time away from enjoyable listening to get into more DIY experiments with hardware and software? What is the proof of any tangible, overall audible benefit, other than anecdotal listening testimonials or attempts to argue at the bits and bytes level that there is some unmeasured, theoretical problem that can make a "big" difference to the sound. I say, show me the evidence in clear terms how that affects the sound coming out and, maybe, I will consider it.

[trithio]

don't think anyone is saying that the bits are corrupted and/or different at the dac, what seems to be observed is that the sound is different. Why, has yet to be answered. If measurements were found that correlated with the observed sound then an understanding might be obtained. Think people have said they have measured differences, but they haven't been adopted by the majority yet. Good luck.

 

There must be something in the air today, cause I can understand and almost fully agree with your post

Would still say that the first step is to actually prove that 'observed sound'. To my knowledge it is still on anecdote level. May be. May be not.

[trithio]

Others and myself have done what is the equivalent of your proposal.

 

For nitty gritty details, I used a laptop feeding an Audiophileo USB/SPDIF converter. The output of that fed the SPDIF digital input on an M Audio Audiophile 24/96 sound card in a desktop computer. The digital input to the SPDIF port of the Maudio card was recorded with a sound editor. All digital no conversions to analog.

 

At one point of I had a 30 minute test file. I marked each end so everything could be lined up afterwards to chop off ends that didn't belong to the original file. I sent hours of this file over that and similar setups. Compared the result in two ways. One was to take the original file and recorded file. Invert one and mix them together. The result was nothing but a file full of zeroes. Nothing there no errors. I also did an MD5 hash of the original and recorded file. Hash values matched so again no errors.

 

Over time I used different laptops. Different desktops. Loaded computers with heavy tasks during the process. Had as much going on the USB ports as possible. Put a large gaming computer power supply outside the case and exposed different pieces of equipment to the raidiated noise by placing it next to it, and by running either the USB or an SPDIF extension past it. Used different playback software. None of it made any difference in the bits. No errors, not one bit wrong.

 

Even the people saying these things sound different admit it is not due to bit errors. That as a source of sound difference has been effectively eliminated.

 

Digital loopback tests should be quite a staple in the DAC industry. Also some ASIO drivers come with useful debug/check features. That is, if you trust the producers. I think the mytek 192 comes with one. You can play with buffer sizes and they log the no of skipped samples. Usually zero.

 

As to your hardcore RFI/EMI tests I remember reading something similar a while ago. Guy tested USB cables. He tried lots of the same things and his only troublesome setup was something like: strip cable isolation until you have a portion of bare wires, put a cellphone on it and call that cellphone. Not exactly something one should worry about

[trithio]

Master clock

Word clock

RFI inside/outside equipment

EMI inside/outside equipment

Grounding

Isolation

Temperature

Humidity

Internal cable length

Internal cable conductivity

Metal versus optical

5.01234V versus 4.98765V (steady supply of precise voltage)

Interaction between audio equipment and measuring equipment

 

Some of them should be obviously related to jitter / noise, and then we'll have to figure out the difference between correlation and causation etc.

 

Ultimately there's still this philosophical question - are we listening to music (emotional>logical) or listening to gears (logical>emotional) or a little bit of both?

 

All very serious issues. For a DAC designer. Pretty much all that happens outside the PC so not sure how any of it will affect the bits=bits thing. I'd say not at all.

 

And if you mean 'ultimately' as in 'the last question one should care about', I fully agree . It's clearly always a combination of both. Different proportions for different people. And I dont see why would that question be so important and ultimate. Would you care to detail?

[YashN]

Ok, we agreed bit corruption, as was implied by the Row Hammer paper, is a non-issue. Was that just a red herring? Sorry, but much of the thread is simply too esoteric and beyond my comprehension.

 

Not a red herring at all - it's a proven phenomenon and there's a research paper just out on it. It's something to worry about *in addition* to everything else which happens in a computer audio setup.

 

The 'everything else' contains a lot of issues related to noise and jitter and the D/A process even if bit-perfection is observed from source to USB receiver of the DAC.

 

This is the crucial point that many who are too focused on just bit-perfection miss.

 

So, if you follow closely, the RH issue is something to worry about before everything else, because if the bits are already messed up with in the DRAM buffers, you can't have bit-perfection.

 

Now, suppose you do manage to have bit-perfection in the DRAM, then there are a lot of system interactions which affect SQ, many of which are counter-intuitive even for people schooled in IT, because they have nothing to do with bit corruption at all.

 

They have to do with what happens with energy consumption and ground plane noise and signal integrity and they ultimately affect the DAC chip and the DAC clock and hence the SQ.