SPIDF DAC with DIR9001 and PCM5102A - Questions and Suggestions

[df audio]

Hi,

I'm looking at designing a DAC using the SPIDF interface, with the following signal path. Power will likely be from a AC adapter, with regulators on the PCB to feed the individual chips. Everything is running in "hardware" mode.

Coaxial/TOSLINK → DIR9001 → I2S → PCM5102A

Datasheets: DIR9001 PCM5102A

I originally looked at a USB-I2S converter like the PCM27xx series or CP2114, but they're sadly all 16-bit devices. I've already got a PupDAC, and am trying to make something that can give 24-bit output fed by the SPIDF jacks on my mobo. Most likely this will be a datasheet based design (I'm still a beginner) and I plan to open-source everything if I ever complete this project.

Now, I've got some questions for the pros in the community

1. Do you recommend that I use coaxial SPIDF or TOSLINK inputs? My motherboard has both, and while I've heard of arguments for both sides (no ground loops for TOSLINK, less jitter on coaxial, etc.) I'm wondering if you guys have any specific advice for my implementation. I could add a switch for both (like in Pavouk's design), but would prefer not to. From the above design, it also looks that I need a transformer to match the coaxial input with the DIR9001 (which would complicate things a bit).
2. How do you recommend I maximize analog performance? I chose the PCM5102A as I has simple line outs directly on the chip, preventing the need for opamps and the like. I'm heard some tips like the fact that analog and digital sections should be seperate, and that the analog circuitry should be as short as possible. Is there anything else I've missed?
3. The two chips I plan to use don't just support I2S, but can also use "left-justified" signals. Which one is better?
4. I've got a lot of clock options here on the DIR9001 - what's the best way to proceed?

I want to state clearly that this project is meant for fun and learning, and that there are many pre-made/DIY DAC options out there that'll save me the time, money, and energy of creating my own design. But what's the fun in that?

[abraxalito]

A trafo on the input helps to reduce common-mode noise, especially if you choose one with low interwinding capacitance.

When you say 'maximize analog performance' do you mean measured or listened performance?

Can't think of any reason one digital interface format would be superior to any other. On DIR9001 clock options, choose the one which gives lowest jitter - there's a graph in the DS from what I recall showing different multipliers.

[df audio]

I'm an objective-subjectivist in audio. The problem is that I (obviously) don't have an AP or Prism to measure everything in detail, but I'll probably be using RMAA and some decent ADCs to keep things in check. Basically, I just don't want my measured specs to fall to far behind from the chip's specs (like many other products I've seen). For subjectivity, I often just rely on A/B testing and try to look for the subtle differences in different pieces of equipment.

[abraxalito]

A/B testing is for listening to differences - but then how would you know if any difference you heard was an improvement?

[df audio]

It's possible, but not ideal, to use A/B testing to listen for things such as clarity and noise. By comparing my design against a solid reference, I might be able to spot out potential flaws - or I might not be able to. Numbers are the most reliable, but of course not everyone has several thousand to blow on an expensive analyser...