[Riboge]

Quote:

Originally Posted by n4k33n /img/forum/go_quote.gif The problem is that some usb reciever chips on dacs can only accept 16 bit audio. I think this is the case with the pcm2902

It appears to me from searching around in our forums and elsewhere that the only DACs that can process 24/96 via USB are the new Benchmark DAC1 and the Empirical Audio USB to SPDIF/12s converters. I own the Stello 220 MkII and have been able to find no evidence that it can do so. This seems confirmed by the fact that the SPDIF out of my Audigy 2zs card sounds better than USB with this DAC in my system, especially with 24/96 music. The ranking of USB as better can only be, if ever, with those that can take 24/96 when playing 24/96 recordings which are inreasingly becoming available, e.g., from Highdefinitiontapetransfers and Linn Records.

 

[LawnGnome]

Quote:

Originally Posted by MaloS /img/forum/go_quote.gif Well in reality digital is not 100% right or wrong, I have knowledge of transistor-level logic design and could bring up quite a few examples of how it is not that way, but the different story is that I doubt either one of the mediums causes any noticeable issues (I am working with 192 khz sample rate, much higher than needed for sampling a signal functionality of which caps out at 22 khz), that concern is mainly inferior here. Video is a different story...but I point this out for general knowledge. But anyways, thanks for pointing out that USB uses the processor, I might switch to coax soon then. I do play games and don't want to waste valuable clocks on my C2D E6750

Nobody is saying there are not error's, thats why there is error detection/protection.

Digital Data IS either right or wrong.

What is of more concern is if there is extra signals on the line other than what is meant to be sent, from interference. As this can affect performance of the hardware downstream. This only matters here because it is being played back in real time.

But in the end, either the data is correct or incorrect.

 

[Nebby]

^Do you mean to say that there is no chance that the data can be partially incorrect?

 

[MaloS]

Quote:

Originally Posted by LawnGnome /img/forum/go_quote.gif Nobody is saying there are not error's, thats why there is error detection/protection. Digital Data IS either right or wrong. What is of more concern is if there is extra signals on the line other than what is meant to be sent, from interference. As this can affect performance of the hardware downstream. This only matters here because it is being played back in real time. But in the end, either the data is correct or incorrect.

Well I am going with assumption that when we sent a 1, incorrect would be classified as 0. There is a region that can be classified as neither, and the results coming out of that can get very interesting - since D/A conversion is not exactly working by clicking 'if 1 - play back this frequency', and on top of that, the chip will SOMEHOW respond to this non-#, which will produce results other than what is considered correct response to the data by specifications.
On top of this, 1 is classified as for example 3 to 3.5 V. For example if we feed this into a gate that has a buffer function - depending on the gate we could still get a variation in resulting output. There is a possibility that if we feed 3.2V, we will get 3.4V, when in reality we are hoping to get only 3.5V when feeding the range of 1 into a buffer.
Now why does my question of preferable medium come in - ? Well, suppose you have a cable, with certain impedance. It will invariably take some power from the signal transfer, and if our output Voltage from the PC was a 1, but right on the edge - it can fall into the non-defined voltage when received by the chip - and results at that point can be questionable.
That is not all - even more interesting scenario is known as response time of a gate. If we have an inverter (feed a 1 get back a 0 and vice versa), there is a certain amount of time it will take the gate to respond, since when switching from 1 to 0 it has to go through the voltage where the digit is unidentified by our layout - and all the stages of the DA conversion have to respond to that (which also has effects on fidelity). With not so quality medium this can be especially adverse if the change we are undergoing is not so great (from margin of 0 to margin of 1 will not have very fast response as opposed to going from the edges - 0V to 3.5V). If the medium, such as USB, is somehow lossy (meaning losing voltage) - there is going to be fidelity loss in the data WITHOUT actual wrong data being sent/received just because things will slow down.

Now all of these things occur at very high speeds, so it obviously should not be noticeable, just like mp3 compression, which at its most basic methods removes frequencies that are not audible to human ear...eh? eh? Get where I am going with this? There is absolutely no need for interference to create problems, but I am trying to figure out which one of the 3 mediums can produce the least probability of it...might end up being the best idea just to get a quality cable for each and test it all back to back...


As you can see - digital actually has a whole lot of characteristics you seem to expect from analog Computer does not know what 1 or 0 is, its just a bunch of designations that are consistently refined so we, humans, can perceive as digital data.

Oh, and another addition. This is what Emperical Audio is basically working with - since such differences can cause problems when working with data real time - one of their products has a neat effect to accept the signal from USB or coax or optical - and store it on new memory, filtering it and setting the voltages as close to extremes as possible - and then feeding the refreshed, neat information to the DAC.

 

[slwiser]

Quote:

Originally Posted by Iron_Dreamer /img/forum/go_quote.gif Well I don't think the differences should be much, if at all with a well-designed DAC. But if it doesn't handle jitter properly, this is the order I'd gather based on what I've heard over the years: 1) USB (DAC does USB to I2S directly) 2) Coax (if used with 75-ohm cable and BNC's) (if soundcard's coax output is transformer coupled) 3) USB (DAC does USB to SPDIF to RCA) 4) Coax (if used with cable/connectors of incorrect impedance) (if soundcard's coax output is transformer coupled) 5) Optical 6) Coax (if used with 75-ohm cable and BNC's) (if soundcard's coax output is NOT transformer coupled) 7) Coax (if used with cable/connectors of incorrect impedance) (if soundcard's coax output is NOT transformer coupled) Of course, #1 is very rare these days. USB is still generally better (lower jitter) than optical, though, and often better than coax, since a lot of soundcards don't have properly implemented coax outputs, which can result in a ground loop among other problems. And of course, most people use improper coax cables (i.e. with RCA connectors).

Where would you put a HagUSB with XLR connections into a XLR input DAC? The DAC is the Lavry DA10 receiving a SPDIF signal, 110 ohm connection and cable.

 

[LawnGnome]

Quote:

Originally Posted by MaloS /img/forum/go_quote.gif Well I am going with assumption that when we sent a 1, incorrect would be classified as 0. There is a region that can be classified as neither, and the results coming out of that can get very interesting - since D/A conversion is not exactly working by clicking 'if 1 - play back this frequency', and on top of that, the chip will SOMEHOW respond to this non-#, which will produce results other than what is considered correct response to the data by specifications. On top of this, 1 is classified as for example 3 to 3.5 V. For example if we feed this into a gate that has a buffer function - depending on the gate we could still get a variation in resulting output. There is a possibility that if we feed 3.2V, we will get 3.4V, when in reality we are hoping to get only 3.5V when feeding the range of 1 into a buffer. Now why does my question of preferable medium come in - ? Well, suppose you have a cable, with certain impedance. It will invariably take some power from the signal transfer, and if our output Voltage from the PC was a 1, but right on the edge - it can fall into the non-defined voltage when received by the chip - and results at that point can be questionable. That is not all - even more interesting scenario is known as response time of a gate. If we have an inverter (feed a 1 get back a 0 and vice versa), there is a certain amount of time it will take the gate to respond, since when switching from 1 to 0 it has to go through the voltage where the digit is unidentified by our layout - and all the stages of the DA conversion have to respond to that (which also has effects on fidelity). With not so quality medium this can be especially adverse if the change we are undergoing is not so great (from margin of 0 to margin of 1 will not have very fast response as opposed to going from the edges - 0V to 3.5V). If the medium, such as USB, is somehow lossy (meaning losing voltage) - there is going to be fidelity loss in the data WITHOUT actual wrong data being sent/received just because things will slow down. Now all of these things occur at very high speeds, so it obviously should not be noticeable, just like mp3 compression, which at its most basic methods removes frequencies that are not audible to human ear...eh? eh? Get where I am going with this? There is absolutely no need for interference to create problems, but I am trying to figure out which one of the 3 mediums can produce the least probability of it...might end up being the best idea just to get a quality cable for each and test it all back to back... As you can see - digital actually has a whole lot of characteristics you seem to expect from analog Computer does not know what 1 or 0 is, its just a bunch of designations that are consistently refined so we, humans, can perceive as digital data. Oh, and another addition. This is what Emperical Audio is basically working with - since such differences can cause problems when working with data real time - one of their products has a neat effect to accept the signal from USB or coax or optical - and store it on new memory, filtering it and setting the voltages as close to extremes as possible - and then feeding the refreshed, neat information to the DAC.

You do realize a failure to properly transmit the data intact is still incorrect data.

SO, it is STILL either correct or incorrect.

Either the data received is the same as the data sent, or it isn't.

If there is enough interference that the data is read incorrectly, the data is incorrect.

If the interference isn't enough to cause the data to be misread, then it is correct.


It very truly is black and white.


Also, just so you know, there is either a 1 or a 0, either a pulse is read or it isn't. So if a pulse is not recognized, it is considered a 0. There is no "neither". It is one or the other.

 

[Nebby]

Ah, so you're talking about the actual stream. I think that's where the mixup was. The conditions I personally consider to be the gray area is when false-positives and false-negatives start showing up. From a theoretical standpoint, yes WE know that the data wasn't correctly read, but the question is....does the device know?

disclaimer: I am not too technically informed about the exact workings of the process, so please excuse me if I'm talking gibberish

 

[error401]

Quote:

Originally Posted by LawnGnome /img/forum/go_quote.gif Also, just so you know, there is either a 1 or a 0, either a pulse is read or it isn't. So if a pulse is not recognized, it is considered a 0. There is no "neither". It is one or the other.

Well it's not quite that clearcut. Logic circuits usually have an indeterminate region between 'this is definitely 1' and 'this is definitely 0' where the value is undefined. Within this region bad things (tm) can happen regarding oscillating values if the input is near a switching threshold (eg. voltage turns fet switch on, fet switch heats up due to current flow and turn-on voltage increases, and thus turns off again and so on a few million times a second). Many logic circuits that expect external or noisy inputs use Schmitt triggers to eliminate these effects, but in general there is definitely a no-mans-land between values. This is one reason that when working with digital you try and keep signal edges as square as possible, to minimize the time spent in the undefined region which can cause serious issues.

In general though, your point is sound. The data is either transmitted properly, or it's corrupted. There is no middle ground in that respect.

 

[error401]

Quote:

Originally Posted by Nebby /img/forum/go_quote.gif Ah, so you're talking about the actual stream. I think that's where the mixup was. The conditions I personally consider to be the gray area is when false-positives and false-negatives start showing up. From a theoretical standpoint, yes WE know that the data wasn't correctly read, but the question is....does the device know? disclaimer: I am not too technically informed about the exact workings of the process, so please excuse me if I'm talking gibberish

USB, at least, uses CRC to detect bit errors. This should be reliable enough to reduce the undetected error rate to effectively zero, so I would say yes, the device does know and can request a resend of bad data. However, the audio devices we're generally talking about may not allow time for that to happen before the data must be sent to the DAC for playback, so you might end up playing bad samples anyway. In the case of SPDIF, only a simple parity check is done which should detect all single-bit errors, but multibit errors (per frame) may not be detected. SPDIF interfaces generally dump the data directly to a DAC as it comes, so there is almost never time to resend the bad frames, and I don't think the interface spec provides a mechanism for the device to request a resend anyway.

Personally I think that with current devices, the SPDIF/AES protocol has the potential for better quality as the clock recovery PLL can be externally tuned and modified much more easily than with USB. Assuming the source has a decent clock (which is also more likely and easier to achieve than USB), a good PLL should be able to recover a good clock with SPDIF, while USB timing is a lot more difficult to recover as the signal is a lot more complicated and 'bursty'. In any of these options the actual data itself should be trivial to recover, it's getting a decent clock out of it that's the problem.

 

[tot]

Quote:

Originally Posted by error401 /img/forum/go_quote.gif USB, at least, uses CRC to detect bit errors. This should be reliable enough to reduce the undetected error rate to effectively zero, so I would say yes, the device does know and can request a resend of bad data.

In the most common streaming audio USB mode there is no resend, it is just one way street like SPDIF.

I would imagine that any properly cabled system should have no issues getting bits correctly to the receiver, sound quality is mostly down to jitter.

For those interested how much abuse a toslink can take see http://www.seanadams.com/jitter_fiber_test/

 

[holland]

Quote:

Originally Posted by MaloS /img/forum/go_quote.gif Well I am going with assumption that when we sent a 1, incorrect would be classified as 0. There is a region that can be classified as neither, and the results coming out of that can get very interesting - since D/A conversion is not exactly working by clicking 'if 1 - play back this frequency', and on top of that, the chip will SOMEHOW respond to this non-#, which will produce results other than what is considered correct response to the data by specifications. On top of this, 1 is classified as for example 3 to 3.5 V. For example if we feed this into a gate that has a buffer function - depending on the gate we could still get a variation in resulting output. There is a possibility that if we feed 3.2V, we will get 3.4V, when in reality we are hoping to get only 3.5V when feeding the range of 1 into a buffer. Now why does my question of preferable medium come in - ? Well, suppose you have a cable, with certain impedance. It will invariably take some power from the signal transfer, and if our output Voltage from the PC was a 1, but right on the edge - it can fall into the non-defined voltage when received by the chip - and results at that point can be questionable. That is not all - even more interesting scenario is known as response time of a gate. If we have an inverter (feed a 1 get back a 0 and vice versa), there is a certain amount of time it will take the gate to respond, since when switching from 1 to 0 it has to go through the voltage where the digit is unidentified by our layout - and all the stages of the DA conversion have to respond to that (which also has effects on fidelity). With not so quality medium this can be especially adverse if the change we are undergoing is not so great (from margin of 0 to margin of 1 will not have very fast response as opposed to going from the edges - 0V to 3.5V). If the medium, such as USB, is somehow lossy (meaning losing voltage) - there is going to be fidelity loss in the data WITHOUT actual wrong data being sent/received just because things will slow down. Now all of these things occur at very high speeds, so it obviously should not be noticeable, just like mp3 compression, which at its most basic methods removes frequencies that are not audible to human ear...eh? eh? Get where I am going with this? There is absolutely no need for interference to create problems, but I am trying to figure out which one of the 3 mediums can produce the least probability of it...might end up being the best idea just to get a quality cable for each and test it all back to back... As you can see - digital actually has a whole lot of characteristics you seem to expect from analog Computer does not know what 1 or 0 is, its just a bunch of designations that are consistently refined so we, humans, can perceive as digital data. Oh, and another addition. This is what Emperical Audio is basically working with - since such differences can cause problems when working with data real time - one of their products has a neat effect to accept the signal from USB or coax or optical - and store it on new memory, filtering it and setting the voltages as close to extremes as possible - and then feeding the refreshed, neat information to the DAC.

Much of what you describe can be dismissed. The reasons are such. The transient issues with edges and signals are typical of timing violations and each signal is required to maintain a hold time before each clock edge to guarantee quiescence and the clocks themselves are required to have hold times as well, to discard jitter in the clock signal. If it is out of spec, it's a poor design, and a bad component to even consider.

Marginal voltages from the transmitter and receiver are marginally spec compliant, and as such are marginal components themselves.

That really just leaves the question of which is the more robust interface.

 

[MaloS]

Do I now have to explain the sampling theory and how all of these small things that do not matter for computer operation will have an influence on SQ? Marginal differences, but they are there.

Quote:

Also, just so you know, there is either a 1 or a 0, either a pulse is read or it isn't. So if a pulse is not recognized, it is considered a 0. There is no "neither". It is one or the other.

Certain points within the noise region can and will be read as points within noise region, without a clear cut position one way or the other. There are always many-step processes to attempt to clarify that, but in reality we are just minimizing this hole.

I repeat, the differences in the end are minute, but they are there. Otherwise there would be NO DIFFERENCE between my Stello DAC and an EMM Labs CDP.

 

[Ahriman4891]

Quote:

Originally Posted by MaloS /img/forum/go_quote.gif Otherwise there would be NO DIFFERENCE between my Stello DAC and an EMM Labs CDP.

Are you implying that the difference in the digital path of these devices produces more effect on analog output than the difference in analog paths? (Schematic, caps, opamps, interference--which matters way more for analog, etc. etc.?) I'm willing to bet that the difference is due to the CDP being way more expensive and less compromised than your Stello.

BTW, Stello is nothing special if you look at the innards -- no shielding for the PSU, digital and analog paths probably share the same ground plane (wild guess). It's basically a 0404USB in a shiny box with a bit better parts, I wouldn't expect it to rip apart a $10K CDP. EDIT: it actually has full Class-A output, so I removed one of my "complaints"--not that all Class A circuits are created equally.

P.S.
Quote:

Do I now have to explain the sampling theory and how all of these small things that do not matter for computer operation will have an influence on SQ?

Sure, go for it

 

[holland]

Quote:

Originally Posted by MaloS /img/forum/go_quote.gif Do I now have to explain the sampling theory and how all of these small things that do not matter for computer operation will have an influence on SQ?

I have to disagree with you. Bitwise errors are much more critical to normal operation than it is to SQ. Errors on a bus are not taken lightly as they can easily cause system wide faults.

Quote:

Originally Posted by MaloS /img/forum/go_quote.gif Certain points within the noise region can and will be read as points within noise region, without a clear cut position one way or the other.

No. It's either on or off in binary logic. The noise region is not high enough to trigger a high voltage signal, and it will not have the necessary hold time for a pulse. Because of that, it will be treated as low. In other words, it has to reach the minimum for the high signal and maintain it for xxx ps before the clock edge for the flop to latch. It's rooted in physics, transistor design, in order to breach the physical junction and diodes and pass voltage through.

I am not familiar with your components that you mentioned, but there is much more to component level variances than a common bitstream or a single transistor. So, I think your view on the matter is overly simplistic, to try and emphasize a point.

Dirty fluctuating DC can play a factor, the analog side would play a very significant part in the sound. As long as the inputs up to the DAC are identical (I don't know as you brought up the components) then the differences are elsewhere, most likely analog, power, shielding, etc.

 

[thomaspf]

Quote:

Originally Posted by Riboge It appears to me from searching around in our forums and elsewhere that the only DACs that can process 24/96 via USB are the new Benchmark DAC1 and the Empirical Audio USB to SPDIF/12s converters.

There are quite few USB adapters out there that support 24/96 data. The Audiotrak Optoplay is an entry level adapter that supports 24/96 just fine. I also have an Edirol UA-1EX that does support 24/96.There are many others...

Quote:

Originally Posted by tot In the most common streaming audio USB mode there is no resend, it is just one way street like SPDIF.

That is true for all the audio modes defined in the USB audio spec. The samples always travel in an isochronous stream with no retransmissions.

One feature of the USB audio spec that is rarely implemented is the ability to have the clock that is driving the playback residing in the USB device. Unlike S/PDIF connections with toslink or coax a USB cable is a bus that works in both directions. With async USB audio the DAC can pull the data at the rate required by it's local clock.This way you can build a very stable low jitter local clock and can avoid asynchronous resamplers or more complex schemes trying to achieve low jitter from an S/PDIF connection.

Cheers

Thomas