In general, downsampling works best (i.e. best quality per CPU usage) when downsampling by an integer multiple. “Asynchronous” downsampling is generally going to sound worse if implemented within the same performance envelope.
There are lots of technical shades of detail in sample rate conversation, and certainly some advanced techniques exist that accomplish non-integer rate conversions efficiently enough and without a significant quality compromise, but Roon’s down-samplers don’t work that way. So, when downsampling for compatibility, we divide by 2,4,8,16,… and don’t attempt the asynchronous conversion from 352.8 to 192k.
DSD formats are a multiple of 44100hz (DSD64 is 44100x64hz), so they get downsampled to 352.8, 176.4, 88.2, or 44.1.
The SonicOrbiter SE’s s/pdif port has a unique (and unfortunate, IMO) quirk: It only supports the following rates:
44.1, 48, 96, 192
This is extremely unusual in the world of audio products–it’s customary to support rates in pairs–meaning that a device that supports 96k should support 88.2k as well, and 192k support should imply 176.4k support. As I recall, this limitation comes from the iMX6 itself, since the S/PDIF interface used on the SonicOrbiter SE is built into the SoC.
The result you’re seeing is a result of combining of our “power of 2” downsampling strategy and this device’s format support quirk.
Squeezelite is willing to take data from us at 176.4k (which is why it shows up that way in the signal chain view). Since we know that the s/pdif port doesn’t support that rate, Squeezelite (or ALSA) must be performing a sample rate conversion. I’m not sure what the ultimate output rate of that conversion would be, or the quality level of that resampling process. Personally, I trust ours more (since I know the implementation details!).
I like the fact that Sonore uses the same processor as the Dialog.
The iMX6 is very popular for products in this space because unlike many comparable ARM SoC’s it’s available in small volumes with modest entry costs at a tolerable price point. Looking a few feet to my right at the pile of gear we use for testing, there are at least half a dozen products that use that chip–it’s the current “default” choice, sort of like how most sub-$5k USB DACs use the same XMOS, FTDI, ESS Sabre, and Burr Brown components.
The “best” SoC’s (Snapdragon, Exynos, Tegra, …) go to the cell phone and tablet manufacturers who measure sales volume in the millions, not high-end audio manufacturers who measure sales volume in the thousands.
The RPi2’s Broadcom SoC is definitely in a class below the iMX6 (but that’s why CuBox starts at $89 and an RPi2 is $35…).