Bits, ghosts, and metaphysics...

[ringenesherre]

My blog post (based on older discussions and providing a definite summary on the possibility of jitter in the digital recording-storage-playback chain) debunks the varying saturation theory when any transmission and/or RAM buffering is used:

 

Jitter in the digital recording-storage-playback chain. - Blogs - Computer Audiophile

[sandyk]

My blog post (based on older discussions and providing a definite summary on the possibility of jitter in the digital recording-storage-playback chain) debunks the varying saturation theory when any transmission and/or RAM buffering is used:

 

Jitter in the digital recording-storage-playback chain. - Blogs - Computer Audiophile

 

Here we go again, another self proclaimed expert like Archimago !

You may be a Computer Scientist, but it does NOT make you infallible in these matters either.

[Jud]

My blog post (based on older discussions and providing a definite summary on the possibility of jitter in the digital recording-storage-playback chain) debunks the varying saturation theory when any transmission and/or RAM buffering is used:

 

Jitter in the digital recording-storage-playback chain. - Blogs - Computer Audiophile

 

Thanks, Peter. I believe I recall that John Swenson was at least uncertain whether any effects of varying saturation could survive the journey into and out of RAM, as you, Tom (kumakuma), and I have all alluded to. You are quite certain it won't, and I'd say even as a layperson chances are quite good you're right. (On the other hand, were you certain there would be no variations in saturation level of the storage cells, or do I recall that incorrectly?)

 

Nevertheless, for thoroughness, let me go through the process of "freezing" jitter into storage locations. (And no, crenca, I am not some sleight of hand artist who is trying to sneak bit value changes past you while you're not looking.)

 

Imagine the analog waveform that is being interpreted as digits. Amplitude of the wave above the "zero crossing point" is evaluated as a 1; amplitude below zero crossing is evaluated as 0. If the analog waveform crosses from 1 to 0 or 0 to 1 a touch early or late, that's jitter.

 

Disc (and flash?) storage locations can be and are saturated to different levels depending on the strength (amplitude) of the waveform at the moment. These variations in saturation levels at each storage location trace the amplitude of the analog wave, in effect storing the time variation (slope) of the analog waveform, and thus any jitter (late or early zero crossing) that was part of it.

 

What Peter (ringenesherre) is saying is that when these stored values are transmitted from disc to RAM and from RAM to DAC, the variations in saturation, and thus any "stored" or "frozen" jitter, are lost.

[Don Hills]

Here we go again, another self proclaimed expert like Archimago !

You may be a Computer Scientist, but it does NOT make you infallible in these matters either.

 

We should note that your own stated qualifications don't put you in any position to judge his level of expertise...

[sandyk]

We should note that your own stated qualifications don't put you in any position to judge his level of expertise...

 

I have never claimed that they do. It should be obvious that there are areas of major disagreements for example in the USB

area among those who are well respected in that area, and none of them claim to know the whole story.

I put much more faith in people like John Swenson who actually do measurements WITHOUT an agenda (unlike Archimago) than those that fall back on theory.

[jabbr]

I'd say that while saturation levels may not survive read-out into RAM, that this process certainly may the a source of disc related noise which may have an audible effect.

 

One of the reasons I have no drive at all on my NAA which uses the iSCSI protocol. It's very hard for me to imagine such variances in saturation levels surviving not only the RAM cache but also the travel across optical Ethernet.

[livelistenlearn]

Imagine the analog waveform that is being interpreted as digits. Amplitude of the wave above the "zero crossing point" is evaluated as a 1; amplitude below zero crossing is evaluated as 0. If the analog waveform crosses from 1 to 0 or 0 to 1 a touch early or late, that's jitter.

 

Disc (and flash?) storage locations can be and are saturated to different levels depending on the strength (amplitude) of the waveform at the moment. These variations in saturation levels at each storage location trace the amplitude of the analog wave, in effect storing the time variation (slope) of the analog waveform, and thus any jitter (late or early zero crossing) that was part of it.

 

What Peter (ringenesherre) is saying is that when these stored values are transmitted from disc to RAM and from RAM to DAC, the variations in saturation, and thus any "stored" or "frozen" jitter, are lost.

 

does this theory imply that when one rips a music file from a source, say a CD, one would include some amount of ‘frozen/embedded jitter’ (a.k.a. small timing errors) alongside relevant waveform information as expressed in strings of 1s and 0s?

or, is it frozen ‘elsewhere’?

 

if so, iyo:

1. are present consumer-grade checksum processes (eg. shasum/md5 on a mac) not thorough/resolute enough to differentiate between amounts/levels of ‘frozen jitter’ in seemingly ‘bit-identical’ files from different ripping sources?
   
2. will subsequent copies/burns of the same music file contain (at least) the same or different quantities of ‘frozen jitter’?
   

 

 

[wgscott]

Maybe the jitter is embedded into the resource fork of OS X HFS+ filesystem files, which is why wav files sound better, since they don't have resource forks.

 

OK, I win the thread.

[livelistenlearn]

Maybe the jitter is embedded into the resource fork of OS X HFS+ filesystem files, which is why wav files sound better, since they don't have resource forks.

 

OK, I win the thread.

 

[kumakuma]

does this theory imply that when one rips a music file from a source, say a CD, one would include some amount of ‘frozen/embedded jitter’ (a.k.a. small timing errors) alongside relevant waveform information as expressed in strings of 1s and 0s?

or, is it frozen ‘elsewhere’?

 

if so, iyo:

1. are present consumer-grade checksum processes (eg. shasum/md5 on a mac) not thorough/resolute enough to differentiate between amounts/levels of ‘frozen jitter’ in seemingly ‘bit-identical’ files from different ripping sources?
   
2. will subsequent copies/burns of the same music file contain (at least) the same or different quantities of ‘frozen jitter’?
   

 

 

 

The theory is that "frozen jitter" is embedded in the surface of the media, not in the data itself. If this is the case, copying or reading the data into memory should cause such "frozen jitter" to disappear or perhaps even be replaced by different "frozen jitter". "Frozen jitter" should also not affect the checksum as programs for calculating checksums only look at the data, not the actual waveforms captured on the media.

[mmerrill99]

The theory is that "frozen jitter" is embedded in the surface of the media, not in the data itself. If this is the case, copying or reading the data into memory should cause such "frozen jitter" to disappear or perhaps even be replaced by different "frozen jitter". "Frozen jitter" should also not affect the checksum as programs for calculating checksums only look at the data, not the actual waveforms captured on the media.

"Frozen jitter" is not a good term for this - "media variation" or some such term is probably better.

So it's a case of the underlying media's variability in representing the digital information. Let's take the case of magnetic tape storage of 1s & 0s - it's not difficult to imagine that there can be variations in the tape or in the strength of signal recorded to the tape. Does this affect the digital information retrieved from the tape? No? But if I'm transcribing this information from one tape to another does the fact that there is some variation in difficulty in reading the digital bits off the tape have any effect on my writing of that same digital bit to a new tape? Yes, I would suggest it does as the difficulty of reading is affecting the writing through the shared power supply & ground noise variations that are occurring. If the storage of the digital bits is more consistent, less variable, will this reduce the transcription variability? Yes!

 

Translate all that to disk to disk copying, memory to memory copying & finally to the digital receiver that is located in the D/A converter where ground noise becomes an issue for the D to A process & can introduce variations in the analogue output.

 

Is the variability that existed on the initial digital data an exact replica of the variation that finally arrives at the D/A stage? Of course not, it will have been through many transcription processes, each one dealing with this variability in it's own way & possibly introducing new variability.

 

Does something like the Regen address this - it attempts to but can't do so completely

 

Why don't normal memory to memory copies not clean the data in the same way as the Regen? Good question & I suspect that the answer is because they are only concerned with the digital side of the transmission & as long as the noise side is within the noise budget for such processes, then all is good, let the noise ride along & become embedded in the recorded bit on the media, it makes no difference to the digital transmission.

 

It's the clash of requirements from two separate domains that is the cause of this mixup - the digital domain & the analog domain

[Jud]

This "frozen jitter," or "media variation" if you like, all depends on whether the storage cells in a given medium can store different levels of magnetic/electrical charge. I take it from reading John Swenson that at least a storage cell on a hard disk can do this. (Don't know about flash memory.) I take it from reading Peter's (ringenesherre's) blog post that memory locations in RAM cannot. As soon as a process of transmission and/or copying eliminates the ability to render different levels of magnetic/electrical charge to represent the signal (beyond the bare difference between 1s and 0s), any jitter/variation effect (it seems to me) must be eliminated. If that's true, at least this particular mechanism would be eliminated as a potential source for reported differences in playback of bit identical files through the same signal path.

[mmerrill99]

This "frozen jitter," or "media variation" if you like, all depends on whether the storage cells in a given medium can store different levels of magnetic/electrical charge. I take it from reading John Swenson that at least a storage cell on a hard disk can do this. (Don't know about flash memory.) I take it from reading Peter's (ringenesherre's) blog post that memory locations in RAM cannot. As soon as a process of transmission and/or copying eliminates the ability to render different levels of magnetic/electrical charge to represent the signal (beyond the bare difference between 1s and 0s), any jitter/variation effect (it seems to me) must be eliminated. If that's true, at least this particular mechanism would be eliminated as a potential source for reported differences in playback of bit identical files through the same signal path.

That may well be correct in the sense that no imprint of the original media variation remains in RAM but each step has the ability to add it's own variation to the transmission in the form of ground noise. If all concurrent processes could go to quiescence during a copy from RAM, then the ideal may apply but, as in all engineering, we are trying to deal with the real world tolerances, not ideals.

 

For instance, while RAM is being read from is it unusual for it to be written to also, not necessarily at exactly the same concurrent time but close enough in time that any current draw on one process could be in the recovery window of another process i.e ground noise of writing could affect ground noise during reading?

 

It really requires some very deep knowledge at the electrical level of the workings of PCs

[YashN]

This "frozen jitter," or "media variation" if you like, all depends on whether the storage cells in a given medium can store different levels of magnetic/electrical charge. I take it from reading John Swenson that at least a storage cell on a hard disk can do this. (Don't know about flash memory.) I take it from reading Peter's (ringenesherre's) blog post that memory locations in RAM cannot

 

In the case of DRAM, this is totally incorrect:

 

The storage is internally analogue: the values are stored in capacitors, which actually lose a little bit of voltage with time. In turn this requires periodic refreshes, and finally amplification to determine the digital representation.

 

The two threads 'DRAM Flipping Bits' and 'Beyond Bit-Perfect' have more info.

[YashN]

Here we go again, another self proclaimed expert like Archimago !

You may be a Computer Scientist, but it does NOT make you infallible in these matters either.

 

Indeed, it totally ignores the accompanying noise profile that the analogue parts of the circuitry has to work to determine the bits, thus completely missing out that in real-time playback, the bits don't need to be changed for a difference in sound to occur.

[wisnon]

Does Porzilli believe in the concept of embedded jitter? The Memory Player

 

The Memory Player is no ordinary digital audio player. The Memory Player is an intricate and powerful music server that accomplishes its goal through a comprehensive jitter erasure system, which:

 

1 - Copies music through re-reading from a CD, DVD or music file to a block of Fractionalized Memory,

 

2 - Eliminates digital volume control losses and other system losses by upsampling to 32 bit audio,

 

3 - Eliminates jitter-creating signal splitting by playing from a 64 bit sample,

 

4 - Removes jitter from the music data before you play it (something not possible when playing a moving disc or a file directly),

 

5 - Using the shortest signal path possible in a DAC, achieves a huge, deep image with air throughout the stage and around each instrument, seemingly unlimited dynamics and a sweet 'analog' tone, never before heard in digital playback.

 

The Memory Player's refined DAC design is a welcome departure from the high tech, circuitry-burdened devices now in use. D2D creates the clearest, cleanest, least colored and simplest path from the DAC's first moment of analog to the outputs of The Memory Player.

 

The Memory Player received over 20 rave reviews between 2006 and 2012, and was awarded "Most Wanted Component" by The Stereo Times for six years.

 

The next step in the process of creating analog quality from digital audio is known as IDEAS. Impulse Discharge of Events in Atemporal Space is explained under How The Memory Player Uses IDEAS.

 

More Ways Music Can Acquire Jitter

 

Musical information being played is always transferred or copied to a form of memory, from which it is played. Therefore, it could be said that all digital audio players are "memory players".

 

They vary according to how much music they hold on the memory, how much dissection that music is subjected to, and what, if anything, is done to clear the data of unwanted jitter. No matter what type of player, CD, file, download, stream/internet radio, music server or computer, data is played from some form of memory inside the player.

 

All digital audio players use small caches, or small pieces of memory, that dissect the incoming signal, and place portions of the music on the cache. They play it and hold it, as the cache awaits the next inflow of music data. By holding more and more music data, the cache compensates for any small delay or interruption of signal, because the cache may hold several seconds of music, giving the data time to arrive, so no silence occurs.

 

Memory experiences a great deal of corruption if left unattended. Electronic and physical obstructions (physical in the case of optical media) ultimately end up interrupting perfectly timed, pure music bit flow. All these obstructions cause jitter.

 

Obstructions can include local RF fields or radio itself, stray flux fields (magnetism and power supplies), dirt, scratches, noise, old data from other CDs played recently, internal operations of the unit itself, such as pre-fetching, DLLs, or ECC bits created to conceal dropped music bits.

 

All these must be cleaned off the cache memory, or the music data will experience time delays, known as jitter.

 

With nearly all those obstacles gone, the memory serves as an ideal palette for music bits alone.

 

The Memory Player, eliminating jitter through the IDEAS process, creates an almost perfect replica of the master on the Memory from which you'll play your music.

 

How The Memory Player Uses IDEAS

 

IDEAS is a complex process that attempts to replicate the master itself, through reading its index to determine the file type, and eliminating non-music data from the memory.

 

First, IDEAS prepares the space by cleaning it of any legacy (old) music data or any non-music data, as described above. Every bit on the Memory that is not music data is discharged. This is the Impulse Discharge.

 

The spaces where the data was in the past are called Events in Atemporal Space. If these spaces had bits in them, they would interrupt fresh data, forcing it to jump to the next available space. Even left empty, the bits will be scattered around these spaces.

 

The chart on the right may resemble a defragmentation screen. In truth, there is no relationship between IDEAS and defragmentation. Defragmentation compresses all the data together to increase the speed of a computer. IDEAS arranges the data for the lowest jitter, regardless of speed.

 

Determining that the space is empty, and then skipping it, takes more time than it would if the next bit been inside that space. When the bit is read late, that time delay results in jitter. Therefore, both the problematic bits and the empty spaces are discharged through the IDEAS process.

 

Once the IDEAS process is complete, music data can populate the Memory in a more natural configuration.

 

Now, music bits are in their original order, and since they are uninterrupted by vacant space, non-music bits, or old music bits, recorded music can mirror the master itself. Mirroring the master consists of not just capturing the quantity of bits, as is done in the nulling process used in studios, but capturing them in their original time.

 

The Memory Player can lower jitter to the femtosecond level, easily rivaling costly atomic clocking players, and 10 to 100 times lower than achieved by the finest conventional CD or high resolution file players.

 

Now that the recorded music has been cleared of unwanted jitter, the next stage in our challenge is reducing or eliminating further losses in playback, through D2D.

[ringenesherre]

Indeed, it totally ignores the accompanying noise profile that the analogue parts of the circuitry has to work to determine the bits, thus completely missing out that in real-time playback, the bits don't need to be changed for a difference in sound to occur.

 

Come on, give me a little more credit. Of course, I am aware of how DRAM stores bits - nowadays typically as a charge in a capacitor controlled by a single transistor - nicknamed 1T+1C. The physical proximity of the individual cells results in crosstalk phenomenae such as e.g. data dependent retention time. Here, retention time is a lower limit for the time that the charge will be kept well enough to avoid read errors. Under normal operating conditions, as far as I recall this time is at least 64 ms for DDR3. That means up to 16 refreshes per second can be necessary.

 

There are at least two reasons to doubt that noise patterns in the disk storage from the ripping process can influence the noise patterns in RAM significantly in a systematic way:

1) The DRAM is refreshed on the order of 1000 times during the playing of a single song. Each refresh diminishes existing noise patterns in favour of data-dependent noise patterns corresponding to a local minimal energy steady state.

2) The entire circuitry for accessing DRAMS is designed to charge the capacitors with a clear and rather consistent charge. The background is that error accumulation has to be avoided in order to prevent bit failures.

 

I'm not going into certain members attacks on my qualifications - I think we should stick to arguments and facts rather than ad hominrm tactics. Just to avoid misunderstandings, I will point out that a significant part of my professional life consists of reviewing and evaluating scientific publications and funding applications including on topics of covert channels in it security, software resilience for critical systems, and the discrepancies between real-world memory behavior and assumed memory models.

[ringenesherre]

Regarding the IDEAS stuff mentioned, there are some good points and a load of bullshit to find. For such a paranoid approach I'm amused not to find any measures that try to influence the refresh strategies of the DRAM modules ;-)

[YashN]

Come on, give me a little more credit. Of course, I am aware of how DRAM stores bits - nowadays typically as a charge in a capacitor controlled by a single transistor - nicknamed 1T+1C.

 

What credit? Point #2 in your blog post shows you still haven't got it system-wise, you're still talking about how the bits are the same and that it follows there shouldn't be any difference in sound.

[mmerrill99]

Regarding the IDEAS stuff mentioned, there are some good points and a load of bullshit to find. For such a paranoid approach I'm amused not to find any measures that try to influence the refresh strategies of the DRAM modules ;-)

Thanks for your understanding of how DRAM works - interesting

 

Can you tell us what aspects of IDEAS you consider to be bullshit & what you consider the good points?

[ringenesherre]

What credit? Point #2 in your blog post shows you still haven't got it system-wise, you're still talking about how the bits are the same and that it follows there shouldn't be any difference in sound.

 

No, I do not do what you claim. I am looking at it from a systemic view and as argued in my post above, loading a song into memory and then playing it from there practically eliminates the influence of any noise stored in the exact analog representation of the bits. If you do not understand what I am writing, do not make bold claims - just ask!

[jabbr]

No, I do not do what you claim. I am looking at it from a systemic view and as argued in my post above, loading a song into memory and then playing it from there practically eliminates the influence of any noise stored in the exact analog representation of the bits. If you do not understand what I am writing, do not make bold claims - just ask!

 

Agreed. The suggestion is that the ground plane carries the noise.

[YashN]

No, I do not do what you claim. I am looking at it from a systemic view and as argued in my post above, loading a song into memory and then playing it from there practically eliminates the influence of any noise stored in the exact analog representation of the bits. If you do not understand what I am writing, do not make bold claims - just ask!

 

It isn't the case: there are gray areas wherever analogue exists and all the circuitry around to determine the bits affect the ground plane, and also cause RFI/EMI.

[ringenesherre]

Thanks for your understanding of how DRAM works - interesting

 

Can you tell us what aspects of IDEAS you consider to be bullshit & what you consider the good points?

 

IDEAS = Impulse Discharge + Event Atemporal Space

 

Impulse Discharge = writing 0s to memory before loading audio data into that memory. harmless, but useless as noise rather depends on the neighbouring data than on old values AND there are many refreshes taking care of this.

 

Event Atemporal Space = load audio data into continous block of memory. Bullshit as DRAM is "Random Access Memory".

 

Regarding the Memory Player and the 5 steps of jitter erasure, 1 and 4 is what I would expect as benefits of playing from memory. 2 is improving headroom and potentially useful. 3 is bullshit as 32 bit operations are fast, even when they incur a 64 bit (8 banks x 1 byte) resd or write. 5 I will not comment on as I do not know their DAC design and I'm sure there are people more qualified than me in this respect.

[ringenesherre]

It isn't the case: there are gray areas wherever analogue exists and all the circuitry around to determine the bits affect the ground plane, and also cause RFI/EMI.

 

Yes, but once you have the audio data in memory, the noise will ver quickly be the same no matter what ground plane noise was generated e.g. when reading the hatddisk.