Is there any difference between S/PDIF vs USB in terms of audio quality? If yes, then why?

Question

Why are there discussions on the internet S/PDIF vs USB regarding sound quality if both transmit digital signal? Wouldn't be there literally no difference?

Let's take this example: https://www.reddit.com/r/audiophile/comments/8t17rj/dac_question_usb_or_spdif/

someone got upvoted for answering:

Optical gets you complete electrical isolation, however it is typically higher in jitter in comparison to SPDIF or USB in most designs.

Since when "electrical isolation" matters when transmitting digital signal? Cannot digital signal be verified and re-sent on the fly if any packet fail validation, like in the TCP protocol? If you have big enough buffer, you wouldn't even be aware that this happened. Why and how is there any technical difference between SPDIF and USB in terms of sound quality, even if not really perceivable by humans?

Answer 1

Just because someone got upvoted on reddit, that doesn't mean it isn't total and complete nonsense.

OK lets get one thing clear from the start. S/PDIF and USB are simply data transports. In reality they have absolutely nothing at all to do with audio, they are just ways of getting data from one place to another. That data might be related to audio, and in the case of S/PDIF it always is, but it doesn't have to be.

The point at which that data is turned into "audio" is the "DAC" which is the DIGIAL to ANALOGUE CONVERTER.

A DAC needs two things to work (aside from power). It needs a data stream and a sample clock. The clock is used to pull samples from the data-stream through the device.

The DAC doesn't really care where the samples come from as long as it gets them from somewhere.

The clock signal is either generated by hardware or derived from an incoming datastream. In pretty much every case, the re-derived clock signal will be stabilised prior to being sent to the DAC. Particularly in the case of USB, data will be packetised and therefore the clock will be re-derived from timing data and signals contained within the data packets.

Reference to "Electrical Isolation" in any sound-quality related argument in respect of digital audio is complete and utter nonsense.

If audio is sent over TCP/IP, then yes there will be packet-retry as part of the TCP protocol, but this is completely independent of the audio data. I am not aware of any 'retry' function in the USB data transfer protocol. This would be driver dependent.

Answer 2

Electrical isolation matters when hooking up audio devices that have any kind of unbalanced analog connection anywhere: ground loop hum does not really care whether you used a digital or analog connection for completing a ground loop.

It's one reason the standard 31250bps Midi connections were so popular: you could still hook up all your audio as usual without creating additional headaches.

S/PDIF (whether across coax or TOSLINK) is a synchronous direct interface with the data rate being identical to the transmission rate (there is framing involved but that just pads the data to the line rate without leaving reserves for retransmission, and processing does not wait for the end of frame to arrive). In contrast, USB is an arbitrated packaged bus interface with various possibilities for payload and with bandwidth reserves. Audio data is either sent in a streamed manner with fixed alloted bandwidths (in which case there is no possibility for retransmission) or packeted in which case there is no guarantee that the data will arrive timely but you can arrange for retransmission.

Of course, the above quote "optical ... is typically higher in jitter" is nonsensical since "optical" just is a hardware detail and S/PDIF via TOSLINK is not going to have more jitter than S/PDIF via Coax. If we are talking about Ethernet via fiber, Ethernet packaging (whether TCP or UDP or something else) will cause transmission jitter but this should not matter after reassembly at the receiving end.

The high-end audio magic handwaving theory regarding jitter is that the processing of received data puts a load on your power supply, and the more synchronous this load is with the music (and consequently the less jitter you have), the less unrelated noise makes it to the output. Synchronizing the D/A converter rates to the data rates can also be a source of artifacts but the religious mantras surrounding "jitter" don't quite match the problems of PLL designs.

Basically the theories revolve about "what would deliver the theoretic best results assuming all other devices are really really bad" kind of theories.

I would not worry too much about it. Optical isolation will not have significant different properties than well-designed coax interfaces or balanced connections, but "well-designed" is not a given. So it has the advantage of taking one variable off the list: you just cannot botch up an optical connection to the degree of creating a ground loop.