Show us your signal path

I monitor how the underspecced (as per Roon) NAS handles the load. Pretty good so far.

Thanks. You are both right… to a degree. To test this, I added a 44.1kHz WAV convolution filter. With no sample rate conversion, I got this:

image

Roon has no problem doing convolution at 44.1kHz with 235 taps. But when I added sample rate conversion to 176.4kHz, I got this:

Now, Roon is up-sampling before convolution, and the number of taps increases to 2k. Roon must have converted the filter from 44.1kHz to 176.4kHz. This is inefficient compared to filtering at 44.1kHz as before, since you have more taps and more samples to process - not to mention the need to up-sample the filter. I then set the sample rate conversion to 705.6kHz:

Again, all up-sampling happens before convolution. This is even more inefficient, with 6k taps and 16x more samples to deal with. But now comes the strangest part: when I convert to DSD, I get this:

Now, up-conversion happens in 2 steps: 8x before convolution (with 3k taps) and 8x after convolution. How weird is this? For one, Roon is totally fine doing an up-conversion after convolution, so there is no real reason not to do it this way in the previous two cases. Then, although Roon has no problems convoluting at 705.6kHz, it chooses to do it at 352.8kHz. This does save quite a bit of cycles, as you can see in the processing speed: conversion to DSD is more efficient, even if there’s more processing involved. It’s clear that most savings would be obtained convoluting at 44.1kHz, so doing it at 352.8kHz is dubious. I thought maybe Roon doesn’t want to do a 64x up-conversion in one step, so I removed the filter, kept the DSD conversion and got this:

The 64x sample rate conversion is done in a single step, so no limitations there.

Bottom line: I think there’s a bug in Roon’s choices when combining convolution and up-conversion. The results are most probably equivalent from an SQ point of view. However, purists may find this upsetting, and the added inefficiencies could impact the ability to do DSP on low-powered machines.

I am speculating here when I say that convolution and other DSP actions work only in the PCM format

You’re right, you can’t perform DSP on a two-level signal. I just don’t understand Roon’s inefficient mix of convolution and up-sampling.

Maybe this is a good question for Roon developers around here. Would be interesting to know the pros and cons of either option.

Not quite correct when using DSD source files, check out Roon - DSP - Sample rate conversion settings:


Also, Brian explained it here.

1 Like

I don’t think “64-bit DSD” is a thing, but it’s just semantics.

2 Likes

Lyngdorf TDAI 1120

2 Likes

Time to dust off this old nugget… Been toying a bit more with ARC over the last few days – does not sound half bad: :smiley:

2 Likes

The living room however is a tad more complex these days:

3 Likes

lovely

Screenshot 2023-06-27 at 21.46.13

2 Likes

Hehe… Two can play that game: :smiley:

2 Likes

Headphone mode. :slight_smile:

3 Likes

Here is a copy of my Signal Path from my iPadPro controller…This is shot with my Convolutions turned on! They are built by HAF’s Thierry…since I do not use convolutions in my Office system shown below…,.

But can someone show me where I would TURN OFF Volume Leveling? Seems that it completely destroys dynamics!

Actually not possible, but I’ll show you anyway…


Marin, Thanks for the help! Don’t get me wrong I use Headroom Adjustment to compensate for my Convolutions…I was surprised that I had to go as high as -8 db to the Red Light to stop the red blinking light…But you are saying Volume leveling can not effect music dynamics? I can’t seem wrap my brain around that…

It‘s a static adjustment and doesn’t dynamically vary during a song/album, so how should dynamics be affected, technically?

1 Like

You are thinking of dynamic range compression, but this is not the same thing.

Volume leveling analyzes the files and makes it so that their their louder parts are approx. the same. It’s to avoid readjusting the volume all the time because different albums are mastered with different peak loudness.

  • With the “Track” setting it does that for each track, e.g. for playlists, but the loudness differences within a single track remain unchanged.
  • With the “Album” setting it takes the whole album into account; the loudness differences within a single album remain unchanged.
  • (The “Auto” setting uses “Album” when playing albums and “Track” when playing playlists).

The DSP calculations are the same as it uses for headroon adjustment, it’s just that it chooses a target loudness based on the analysis of tracks/albums.

Note that volume leveling nearly always reduces the volume, so it can sound more lifeless because the human perception always hears louder as better. So you have to turn up the volume on the amp by the same amount to make an SQ comparison (and there will be no difference once properly adjusted)

There is a whole standard around how to do this:

5 Likes

Thanks for the reply…it makes more sense they way you explain it…Maybe I was thinking about DRC…Headroom adjustment does also lower the volume some, but like you say…There is more volume to be had…and I do like volume when it’s an Album or a Track I really like…!

Thanks again Suedkiez! My mistake…

2 Likes