Best Ethernet Setup for Streaming Audio

What do you mean by TP-Link.  If it is Wi-Fi then nothing before it matters - it is just Wi-Fi data.  TP-Link in front of the DAC seems to be optimal (eliminates most of injected noise).  D/A conversion always benefits from quiet power supply.  Good line conditioner for DAC (and TP-line receiver) would help.

Kijinki, can you explain the above? How does fiber get interfered by electrical noise?

When transition edges are perfectly square (low transition time) electrical noise added to source or destination makes no difference - level change is recognized at exact moment of transition, but when transitions are slow noise added at the source (varying light intensity) or at the DAC side (varying threshold point) will change moment of level recognition - amplitude variation converts to time variation. Time variation - jitter, converts to added noise.  High end transports have transition times as low as 5ns, typical transports about 25ns and optical many times that.

Where this USB power comes from?  Computer sourced power is not better than wall wort - both likely use crude switching power supplies.  There are very high quality SMPS with lower noise than typical linear supply, but you won't find them in USB supply.  Next best would be linear power supply for USB, but I don't know much about them.  I'm sure they exist and perhaps somebody can suggest a good one?

Fiber optics likely has slow transition edges, making it prone to electrical noise interference (that affect threshold point in time - hence creating jitter). Adding power conditioner would help. 

Thanks Kijinki. Way over my head, but I appreciate the attempt.

Draw on paper two horizontal lines connected by 45 degree transition. Let say, that level recognition point is in the middle of this transition. This will correspond to particular point (draw vertical line). Now, replace this 45 degree straight diagonal line with wobbly line - like sinewave (that’s added noise). Vertical point corresponding to midpoint will change. Vertical line (time) was shifted.
Imagine that you send digital stream producing 1kHz tone. If this digital stream shifts back and forth in time 60 times a second it will result on analog side of D/A conversion with 1kHz tone and two added tones at 940Hz and 1060Hz called sidebands. Amplitude of these sidebands will be proportional to amount of the shift. These sidebands are extremely small, but are still audible since they are not harmonically related to root frequency (in this case 1kHz). With many root frequencies (music) there will be many sidebands, resulting in added noise. This noise is present only when signal is present, so it is hard to measure it. In addition, this shift of the streamed data doesn’t have to be caused by one singular frequency, like 60Hz (correlated) but it might shift at multiple frequencies (uncorrelated).

Perfect transitions (like 0ns) will still have amplitude variations of the midpoint, but it won't change the time moment of level recognition (exact moment of transition).  That way slower transitions are more likely to add noise induced jitter.