The key to this is to look at what the GTO filter does. I could speculate about how it works based on iFi’s marketing but to be honest it may be better to engage them directly to figure out what is happening. Someone from the company did use to post here. Having owned a DAC which was audibly superior with DSD then I’d just configure Roon to deliver DSD as it was better than the DAC converting PCM to DSD.
They are 167 MB each. The DoP FlAC file is 166 MB. Here are the links.
Original DSF: iCloud
DoP FLAC: iCloud
VinylStudio DSF: iCloud
Spoon confirmed that dBPoweramp cannot do the re-conversion.
Did that. Thanks. 
My thoughts exactly. However, I didn’t know if something like that was possible either by design or as a bug. Also, @Marian suggested I consider perceptual bias, and so I will.
I first noticed the issue in Colibri, and I used Roon to verify. I can try the other changes that you suggested to Roon’s audio settings, but at first blush it looks like they’re not involved in the cause. They may be an alternate solution to the problem, but right now the easiest way to remedy the issue for me is simply to select bit-perfect mode at the DAC.
Thanks, Henry. There’s a white paper on their site about the GTO filter. It goes into very technical detail and I wouldn’t classify it as marketing. If you haven’t seen it, here’s the link:
Jussi Laako said that the GTO filter is like “one of the shortest minimum-phase filters, like minringFIR-mp for example is most similar”.
Original DSF and VinyStudio DSF are bit-identical, as expected. VinylStudio file has no metadata in it, so it’s a bit smaller than the original.
1 Like
Looks like marketing to me. While other manufacturers (e.g. Chord) boast unusually large number of taps, iFi seems to go the other way and use a 32-tap filter. What kind of in-band ripple and out-of-band attenuation can you get with that?
Thanks for confirming that the audio in both files is bit identical.
32 taps, millions of taps, anywhere in between. It’s way over my head so I will leave that to mathematicians and engineers to debate. 
I think the debate is between audiophiles, not mathematicians or engineers.
1 Like
This is off topic:
To understand these two (opposite) schools of thoughts, you may look into literature of linear phase vs minimum phase. The former camp is represented by Chord. The latter camp is represented by MQA and Ayre (although not a MQA supporter). Interestingly enough, the default choice in Roon DSP engine is minimum phase.
My standpoint in this (after studying relevant arguments and have users did blind tests) is that this is subjective preference dependent. Some users prefer one to the other, regardless of what different theories say what should sound right.
Off topic, but interesting 
The way I see it, the most important aspects of a [FIR] filter are passband ripple and stopband attenuation. The more taps you have, the smaller the ripple and the higher the attenuation, regardless of the kind of filter. Now, if you get high stopband attenuation, you necessarily get ringing, again regardless of the kind of filter. (If the iFi filter has reduced ringing, it most probably has slow roll-off, i.e. is “leaky”). Linear phase filters have symmetrical ringing around the impulse, so you get both pre- and post-ringing. Minimum phase filters, being causal, have all the ringing after the impulse, i.e. you only get post-ringing. I personally prefer linear phase, fast roll-off (“brick-wall”) filters for a few reasons:
- Ringing is not objectionable if passband is at least 20kHz, since our hearing acuity vanishes to zero around that mark.
- They are the best approximation of the theoretical reconstruction filters, as prescribed by the sampling theorem.
- As a consequence of the second reason, they are the best at preserving the original samples - although that’s not a requirement by any stretch.
That being said, I don’t think I could ever tell one from another when done properly.
Thanks for that. With regards to the marketing I was talking more about when the filter was engaged because my belief was that it shouldn’t be if a DSD file is encapsulated but I suspect a DOP FLAC file is not necessarily a scenario that was envisaged.
1 Like
Would it better to convert my files to DSD 256 or DSD 128 using dBPoweramp or is it better to let Roon do it? (Space not an issue for larger files.)
I don’t see a good reason to convert outside of playback, especially if you want to add some DSP to the signal (e.g. headroom, volume, PEQ or room EQ). No DSP can be applied directly to DSD.
1 Like