Battle of S/PDIF vs USB: which is better?

[Blumlein 88]

Blumlein 88, do you understand what JA actually measured? I read his review in Stereophile several times, but I must be stupid. I guess that he used a PC or Mac to generate a J-test signal. He connected the V-Link to a USB output of the computer. Then he says that he measured the eye pattern by connecting a 15' plastic cable from the TOSLINK output of the V-Link... to what? I don't see a TOSLINK input on the AP 2722. This eye pattern is so good that it defies belief. The sample rate of the excitation appears to be 96kHz. If TOSLINK works this well at 96kHz, I guess it would work well at 10x this sample rate -- maybe even 100x -- yet conventional wisdom is that optical cables don't support sample rates in excess of 96kHz. I suspect that he connected the optical cable to a TOSLINK-S/PDIF converter and then connected its S/PDIF output to the AP 2722. If so, I wonder whether the converter filtered out the jitter.

I am assuming from his description he connected the Vlink (which is a USB to SPDIF converter) to his Macbook via USB. Then fed via Toslink from the Vlink to the AP to record and measure.

Some Toslink gear does work at 192 khz though most gear stopped at 96 khz. I don't know if early Toslink was limited by cable or the optical circuits. I would think the latter.

ADAT uses the same optical circuits and opti-cable, but is a different format than Toslink. It can carry 8 channels at 48/24 or 2 channels at 192/24. I don't know if that is a physical limit of the medium or just an agreed upon standard from years ago. The type encoding of ADAT uses twice the baud rate of Toslink so it would be similar to 384/24 over Toslink. That is said to be compatible with the earliest Toshiba based transceivers of the optical signal. My guess is the cable is not the bottleneck. Probably the circuits.

Early suggestions were to limit optical cable Toslink to 10 meters. But lots of low cost gear now promises and delivers good performance at lengths of 40-140 meters. So probably for shorter distances modern cable/circuitry could work at much higher rates those simply aren't part of the standard. I know I have a piece of cheap 35 foot cable that works with no glitches or hitches with satellite TV boxes and a 12 year old AVR. At different times I have used it for ADAT transfer with some inexpensive gear.

 

[3beezer]

My model has it so I am sure his too. It is very hard to see though as they are tiny connectors that are normally covered. I have circled them in this picture of it:

View attachment 8849

Obviously, I have no direct experience using the AP (although I actually wrote some of the software many years ago), so I appreciate your clarification. However, I am still wondering whether the measurement that JA made is valid. Surely the AP locks to the signal presented at any of the digital inputs so that one can make measurements such as frequency response entirely in the digital domain. If so, then any jitter measurement is actually measuring the jitter of the signal after the clock recovery circuit in the AP. Furthermore, the AP must assume that any signal presented on a digital input is binary, so it is also requantizing the amplitude. To measure an eye pattern, one would have to connect the digital signal to an analog input so that one can see the waveform at points between "high" and "low". It would be complicated to make such a measurement with an optical signal, but the measurement would be very interesting. In fact, that measurement is exactly what we all want to see to understand better the extent to which optical cables introduce jitter. Surely engineers somewhere have made such measurements (at Toshiba?), but I have never been able to find them on the web.

 

[3beezer]

The point here wasn't which sounds better. But rather, given a choice, which one is better from technical point of view. To the extent you have a choice and one is cleaner than the other, it makes sense to use that.

I can certainly appreciate the view of some here that any discussion of S/PDIF vs. TOSLINK is moot if the answer is USB. If, as Amir has stated, the point is to identify the interface that is best from a theoretical point of view, then I am still inclined to agree with him (and most of you, it seems) that USB wins -- a view that I held without qualification until recently. USB has the clear advantage of putting the master clock in the right place. However, in engineering, the choice is rarely clear cut, so it is worth taking time to fully appreciate the advantages and disadvantages of every possible solution. S/PDIF has some objective advantages that are easy to appreciate. I was astonished to discover that it has a subjective advantage too with certain DACs (Sal1950: in repeatable tests under blind conditions with a group of listeners of varying listening experience). That discovery forced me to reconsider my fervent belief in the supremacy of USB. Look at it this way: IF it is the case that engineers are now able to recover a clock from an S/PDIF signal with sufficiently low jitter, then isn't the primary advantage of USB moot? If so, then surely it follows that we should be considering other factors as well. I am not sufficiently dogmatic to assert that the primary advantage of USB is moot, but my discovery of evidence that it might be has me wondering.

 

[Cosmik]

I don't know if people are aware of how trivial all this stuff is. If S/PDIF is not as theoretically clean as USB, we are really only saying that a bidirectional link is better than one where the DAC has to adapt its clock rate to the source. The electrical stuff (differential versus optical, etc.) is very simple in hardware terms.

Want to isolate it? Buy a couple of chips from a catalogue for $2 and accommodate them on your board. Want to implement your own 'TOSLINK asynchronous USB'? A few hours of effort by a handy engineer could create such a link.

Want your own bidirectional protocol? Fiddling about with firmware code is all that is needed.

Converting a S/PDIF type DAC to asynchronous USB would probably be trivial if you had access to the designs and source code - but would be a waste of time because you could buy something equivalent for less effort.

It's pretty simple stuff - just shuffling bits from one place to another.

 

[Sal1950]

I was astonished to discover that it has a subjective advantage too with certain DACs (Sal1950: in repeatable tests under blind conditions with a group of listeners of varying listening experience).

If that was the case, either the involved DACs were poorly designed or broken.

 

[Old Listener]

Almost every computer made in the last 10 years has USB port(s). Few have coax SPDIF connectors and not many have Toslink connectors. Laptops outsell desktop PCs and they have no slots for a card with a SPDIF outut., Even $ 100 USB DACs now use async mode. A decent USB to SPDIF convertor is likely to cost more than that. Unless you have an old SPDIF DAC you want to continue using, USB makes far more sense for audio output from a PC.

Smart TVs often have a Toslink connector but no USB output that works with a DAC (just mass storage.) Still relevant in that application.

 

[March Audio]

I can certainly appreciate the view of some here that any discussion of S/PDIF vs. TOSLINK is moot if the answer is USB. If, as Amir has stated, the point is to identify the interface that is best from a theoretical point of view, then I am still inclined to agree with him (and most of you, it seems) that USB wins -- a view that I held without qualification until recently. USB has the clear advantage of putting the master clock in the right place. However, in engineering, the choice is rarely clear cut, so it is worth taking time to fully appreciate the advantages and disadvantages of every possible solution. S/PDIF has some objective advantages that are easy to appreciate. I was astonished to discover that it has a subjective advantage too with certain DACs (Sal1950: in repeatable tests under blind conditions with a group of listeners of varying listening experience). That discovery forced me to reconsider my fervent belief in the supremacy of USB. Look at it this way: IF it is the case that engineers are now able to recover a clock from an S/PDIF signal with sufficiently low jitter, then isn't the primary advantage of USB moot? If so, then surely it follows that we should be considering other factors as well. I am not sufficiently dogmatic to assert that the primary advantage of USB is moot, but my discovery of evidence that it might be has me wondering.

Any subjective or objective advantages of spdif are entirely down to individual DAC design. Both spdif and USB can produce good results but it is down to implementation. However USB does have a head start, its asynchronous principle with word clock in the dac is the right thing to do. All things being equal, the only variable you need to be concerned about is galvanic isolation. It is however true that optical solves the isolation issue and that most electrical spdif has transformers, but the word clock still has to be extracted. Yes this can now be done practically jitterless.

So its implementation. The point is be concerned about the dac performance in this area and not the transfer medium. Well designed dacs these days should operate well with both, however if measured Im fairly sure usb will perform better in most cases.

If you have a dac that performs better with spdif then there is something wrong with its usb implementation.

As far as Im concerned spdif is an anachronism. I moved to computer playback many years ago and usb dacs are ubiquitous.

 

[watchnerd]

Unless you have an old SPDIF DAC you want to continue using, USB makes far more sense for audio output from a PC.

Why physically connect a PC at all?

That's what networking is for.

 

[Sal1950]

Why physically connect a PC at all?

That's what networking is for.

All that extra complication crap when the PC is like 1 meter away for the DAC?

 

[watchnerd]

All that extra complication crap when the PC is like 1 meter away for the DAC?

Yours is?

Mine isn't -- my server / NAS is in a completely different room, other side of the house. The easiest way to avoid the whole 'noisy PC' problem is to have it elsewhere.

 

[Old Listener]

All that extra complication crap when the PC is like 1 meter away for the DAC?

If you only have audio in one room and one computer in the same room, I agree.

 

[Old Listener]

Yours is?

Mine isn't -- my server / NAS is in a completely different room, other side of the house. The easiest way to avoid the whole 'noisy PC' problem is to have it elsewhere.

Audiophiles don't talk much about functionality but it does matter. I wanted to play music from 3 computers to audio systems in three rooms. I've used different architectures in the 11+ years that I've had computer based audio. My current approach involves USB output from my personal PC in our home office, WiFi communication over a lan to a dedicated NUC in another room and a WiFi connection to a smart TV which is a DLNA player device. I have a WiFi network and a NAS so I use them.

Unless you want to use a PC for gaming, it is quite possible to get a quiet computer without heroic or expensive measures.

 

[Jakob1863]

Trying to assess the effectiveness of a digital link by measuring the output of a DAC is a bit of a flawed experiment. It may appear to be the 'all in one' measurement, but it is the result of multiple mechanisms, including (as mentioned before) shifting clock frequencies, random grounding schemes, stray RF, etc.

Studying the characteristics of the link itself and the mechanisms by which various DACs function would be far more useful. Then you could actually predict what was going to happen worst case rather than just sucking it and seeing.

@Cosmik,

"assess the effectiveness of a digital link" is a bit too ambiguous imo; usually the methodology depends on the research question/hypothesis the researcher tries to assess, so one can´t dismiss the "measuring the dac output" .

Unfortunately rarely the specific research question is exactly specified so often the reader has to work it out backwards and sometimes it seems to be something quite different than the author itself intended to do. (or thinks to have done, judging from the conclusions given)

 

[Jakob1863]

@svart-hvitt,

Jakob,

Archimago posts the result data and charts of his measurements of toslink, coax spdif and usb in three different articles. A comparison between them should be possible by jumping from one article to the next.

Toslink has obviously poorer channel separation!?

But jitter is quite low, and below hearing limit?

I´m really sorry, but i still don´t understand your question regarding my post about measurements done 25 years ago and archimagos measurements. Unfortunately i don´t have spare time to search archimago´s blog to find all the articles about TOSLINK related measurements.

But as stated before, what one is able to find at the analog outputs of a DAC depends on the variables mentioned and at the end on the jitter transfer function of the DAC (and any isolation given by design).

If something measured is considered to be "inaudible" depends on the model used to assess the audibility and might be wrong but there is evidence needed from controlled listening tests.

 

[Jakob1863]

I don't know if people are aware of how trivial all this stuff is. If S/PDIF is not as theoretically clean as USB, we are really only saying that a bidirectional link is better than one where the DAC has to adapt its clock rate to the source. The electrical stuff (differential versus optical, etc.) is very simple in hardware terms.

Want to isolate it? Buy a couple of chips from a catalogue for $2 and accommodate them on your board. Want to implement your own 'TOSLINK asynchronous USB'? A few hours of effort by a handy engineer could create such a link.

Want your own bidirectional protocol? Fiddling about with firmware code is all that is needed.

Converting a S/PDIF type DAC to asynchronous USB would probably be trivial if you had access to the designs and source code - but would be a waste of time because you could buy something equivalent for less effort.

It's pretty simple stuff - just shuffling bits from one place to another.

Yeah, if someone has discovered the problems it is much easier to find solutions, the former is often the more important and exhausting part.
Have you seen the second part of the Burr Brown development story?
http://www.eetimes.com/story/OEG20020220S0064

I´ve linked to that articles quite often after people in german forums didn´t want to believe me that the usb specification couldn´t ensure that no jitter effects feed through.

 

[Cosmik]

Yeah, if someone has discovered the problems it is much easier to find solutions, the former is often the more important and exhausting part.
Have you seen the second part of the Burr Brown development story?
http://www.eetimes.com/story/OEG20020220S0064

I´ve linked to that articles quite often after people in german forums didn´t want to believe me that the usb specification couldn´t ensure that no jitter effects feed through.

I find it hard to put myself in the place of a person who has never written any real time software or looked at pulses with an oscilloscope or whatever i.e. typical audiophiles and reviewers. As I said before, shuffling bits around in asynchronous links is supremely trivial. Implementing PLLs and adaptive systems less so - but entirely graspable intuitively. It is possible to grasp that a unidirectional link can never be entirely free of topology-induced jitter, while an asynchronous one is. And also that jitter in the non-asynchronous system can be attenuated and/or 're-distributed' at the expense of some latency.

I worry that people are imagining that digital links are rocket science, or have an element of 'art' in them, and can only be evaluated by listening to string quartets and jazz from DAC outputs. The reality is much, much more prosaic and predictable than that. As someone said earlier, it isn't an audio interface. It's a digital interface.

 

[Fitzcaraldo215]

I find it hard to put myself in the place of a person who has never written any real time software or looked at pulses with an oscilloscope or whatever i.e. typical audiophiles and reviewers. As I said before, shuffling bits around in asynchronous links is supremely trivial. Implementing PLLs and adaptive systems less so - but entirely graspable intuitively. It is possible to grasp that a unidirectional link can never be entirely free of topology-induced jitter, while an asynchronous one is. And also that jitter in the non-asynchronous system can be attenuated and/or 're-distributed' at the expense of some latency.

I worry that people are imagining that digital links are rocket science, or have an element of 'art' in them, and can only be evaluated by listening to string quartets and jazz from DAC outputs. The reality is much, much more prosaic and predictable than that. As someone said earlier, it isn't an audio interface. It's a digital interface.

Amen. However, that would not seem to preclude some implementations by some manufacturers of being downright schiity.

My impression from looking at measurements over the years seems to indicate that the now widespread asynchronous USB method together with proper galvanic isolation is hard to equal in measured performance with proper design. Earlier, proprietary asynchronous schemes via HDMI, like Sony HATS, also achieved some success, but are now defunct.

Amir's published measurements of his ancient Levinson DAC also showed that excellent performance was achievable by spdif, but that seemed to require careful and expensive design choices. There may be other similarly excellent choices today via spdif. However, it seems extremely rare that unidirectional approaches via HDMI, spdif, AES/EBU or Toslink can measure as well as asynch USB. They are also much more constrained in bandwidth than USB, hence limiting supported hi rez sampling rates and Mch to a greater or lesser degree.

Once, the AT&T or the proprietary Theta single mode unidirectional glass fiber were deemed superior to spdif coax by audiophiles and reviewers, but I never saw measurements for them. They seem to have disappeared.

A fringe element today believes that the unstandardized I2S interface or the use of an expensive external super clock into a similarly expensive DAC is the holy grail. But, they are still a unidirectional answer to a problem better solved by two-way asynchronous under control of the DAC clock.

Others believe dedicated CATx cable is the ultimate answer, although equipment compatible with that is extremely rare. And, of course, the measured advantages over asynch USB are not to be found, although cable length constraints on that are generous to an extreme. But, that might become an issue for active speaker hookup unless wireless Ethernet can deliver adequately. Personally, I would not trust wireless, not yet, anyway.

Audibility-wise and in spite of measured differences but within their individual sampling rate and channel count constraints, there may not be too much difference today with decent design and implementation, even with the once widely disdained Toslink.

But, I am a happy camper with asynch USB into a galvanically isolated DAC even over a 5 meter, non-exotic cable run. DSD256 or PCM 352k are no problem for me with 5.1 or 7.1 channels over a single wire. I have not personally heard better sound anywhere.

 

[amirm]

I remembered that the Musical Fidelity V-DAC II has coax and optical input so here are its results of USB compared to coax:

We see identical performance even though this is an older DAC. So the notion that S/PDIF is better is yet again disproven.

 

[amirm]

Yet another DAC I found that has both S/PDIF and USB: The S.M.S.L. Mini DAC Sanskrit.

While the noise floor is generally lower in S/PDIF, it is peppered with jitter components that are much more problematic as far as audibility. And at any rate, they peak above the noise floor around our main signal and by more than 20 db relative to USB.

So another loss for S/PDIF.

 

[Blumlein 88]

I remembered that the Musical Fidelity V-DAC II has coax and optical input so here are its results of USB compared to coax:

View attachment 8859

We see identical performance even though this is an older DAC. So the notion that S/PDIF is better is yet again disproven.

Or alternatively, the idea USB results in better measured performance is disproven for this device.

What it really means is jitter over decent SPDIF and USB is low enough it isn't a limiting factor in performance among most DACs. Except for notable failures (like the USB of Schiit Modi 2 or Sanskrit SMSL), jitter is effectively a non-issue.

I have a recording interface that can have a free running internal clock fed via ASIO, use an external clock over Toslink or use an external clock with the ADAT format, use external clock over coax SPDIF or external clock over BNC coax connection. When fed to my Forte ADC you get identical results with those. Even as close as .4 hz from the main tone the results are the same. The only differences are noise levels below 10 hz between the connections.

Now it is possible the Forte ADC has higher jitter than anything else and is dominating the result. It however has pretty low jitter or it would manifest itself in sum and difference tones and other noise. All of which are very, very low with the Forte.

So checking the Jtest for the various inputs is worthwhile to be thorough. If you don't get sub-par results the type of digital connection doesn't matter audibly.