Most of this has been said here already, but since we have nothing better to do while waiting for the first Olympus batch to ship, here you go:
PCM/DSD
When analog data is converted to digital (zeros and ones), it uses a digital format like PCM or DSD. Think of these as the methods used to digitally represent music. These are the formats for encoding the audio signal itself. That digitized signal can be stored in memory (hard drives, USB drives, SSD drives, RAM, etc.) and can be read by digital processors (i.e., computers). But if you want that signal to get out of the computer, you need a protocol to transmit it from point A to point B.
I2S
This is where I2S comes into play. I2S is the standard protocol for transmitting digital data (think PCM, DSD, etc.) between integrated circuits (chips). DAC chips take I2S and convert it to an analog signal. The DAC chip in the Taiko XDMI analog daughter card, for example, takes I2S. So does the Lampizator DAC. So, yes, I2S is absolutely necessary for both of these cases.
I2S is a synchronous protocol and uses separate signal lines for data, clock, and word select. That works really well for short-distance transfer (i.e., a fraction of an inch) between ICs. But when you need to transfer this over long distances, there are all kinds of issues. More on that later.
S/PDIF
To be able to transfer digital data over longer distances, such as between two components like a CD transport and a DAC, Sony and Philips created their own digital interface in 1983 called S/PDIF (Sony/Philips Digital Interface).
S/PDIF is an asynchronous protocol that includes the clock data inside a continuous stream of data. S/PDIF uses coaxial (RCA) or optical (TOSLINK) cables to transfer data. It is relatively low voltage (0.5 to 1 volt peak-to-peak) and can reliably transfer data up to 10 meters.
AES/EBU
The 10m distance wasn’t enough for the pro-audio industry. The Audio Engineering Society/European Broadcasting Union created a new standard a couple of years later, in 1985. They called it AES/EBU and it was specified to transfer digital data up to 300 meters over a relatively higher voltage (2 to 7 volts peak-to-peak for balanced connections). Similar to S/PDIF, AES/EBU is an asynchronous protocol with clock and data mixed together.
USB
USB is a general-purpose data transfer interface used for various types of data, including digital audio, video, and other types of data. On paper, USB has some advantages over all other protocols mentioned above - it has built-in error checking/correction, it supports higher resolution and more audio channels, it can carry power (i.e., to power up your DAC), it has bi-directional communication which allows some flexibility, and it is available on almost every computer. With all its advantages, wide compatibility, and low cost, it easily became the most common interface to carry digital audio.
Other
There are many other digital audio transmission protocols designed for various reasons, such as HDMI, Bluetooth, Ethernet-based protocols such as Dante and AVB, MADI, AES67, and many more. Each of these satisfies different design goals and comes with its own pros and cons. But they are not super popular in high-end audio.
So, how does digital data transfer look end to end? You have a streamer that plays a digital file, say PCM for example. That PCM data gets transferred via one of the interfaces mentioned (S/PDIF, AES/EBU, or USB) between your server/streamer and DAC. Then the DAC converts it to I2S, which is then converted to an analog signal. And yes, those conversions penalize the sound quality.
I won’t go into the limitations and problems of each of these protocols, but it’s worth mentioning that S/PDIF and AES/EBU were created in the early '80s and USB took over because it did not have the limitations of the other ancient protocols.
But what about I2S? Why don’t we use I2S? We can, but I2S is designed for very short distances between ICs on a PCB board. You can transfer plain I2S between your digital source (CD player, streamer, server, etc.) and DAC, but the sound quality deteriorates. There are a lot of issues associated with that - signal degradation over longer distances being one of them (skew, jitter, EMI, crosstalk, etc.).
There are some practical ways to implement I2S over longer distance, though. One of them is converting the I2S to LVDS (Low-Voltage Differential Signaling). But as it was mentioned above (re: PS Audio’s I2S implementation not sounding that good) we are using I2S to avoid any conversions but then we end up converting I2S to LVDS to avoid long distance problems - what’s the point, then… So, really I2S is not suitable here.
I am going to sneak in a couple of comments from Emile here as well:
I don't think it's a particularly good approach to generate an I2S signal, consisting of data, word clock and a bit clock line, and transfer all of that over long sensitive copper interconnects with a multitude of impedance mismatches along the way and a general strong sensitivity to transmission line quality.
It's worth noting that some types of jitter are perceived as improving sound quality. Which certainly adds a twist to any technical superiority discussions. This can enable technically inferior solutions to be touted as sounding better.
For the record, a few years ago I was super convinced that I2S was the best sounding protocol between a server and a DAC. I’ve tried every possible solution I could find on both ends - the digital source and DAC. It sounded better than USB at the time, but a properly implemented USB solution (on both - the server and the DAC) has surpassed even the best of the I2S I’ve had.
Here is me quoting myself on my very first post on WBF 5 years ago (I was still posting as
@vassils back then):
The minimalist in me, makes me think that USB is not the best interface for transferring digital from your computer to your DAC. There are too many conversions, clocking, etc. It can sound really good if done with care, but I like simple. In a way AES/EBU is more simple, but it also has its limitations. Many DACs would end up converting the SPDIF/AES/EBU signal to I2S at the end, so I think a good clean I2S signal out of the computer might be a better option for most DACs. If clean, clocked well, and transmitted cleanly (this is where LVDS helps), I2S signal could be superior to USB and SPDIF. YMMV and it is DAC dependant of course... I really wish server manufacturers and DAC manufacturers get together and create a unified standard and stop using USB in high end audio. Take the signal out of a PCIe slot and convert it to the unified standard that could be reclocked before the DAC or fed directly into the DAC without reclocking. I2S seems to be doing many things right, but it was designed 33 years ago and could use some refresh. I really like the MSB idea - proprietary I2S over fiber. And I really wish to see something like that going out of every music server into the DAC.
I've learned a lot since then...
