Stop band attenuation is how much the filter cuts out ultrasonic noise. I think the max attenuation (least ultrasonic noise) is a fine setting myself, but of course everyone else is free to play around and see what they like.
Is this Studio or Origin?
If you are using an iFi DAC, this DoP option will not appear.
8th order SDM means 8th order (steepest) sigma delta modulator. So it’s not a filter, apodizing or otherwise.
If you want an apodizing filter (apodizing is Greek for “removing the foot” - these filters are used to reduce ringing, especially pre-ringing), you want to change bandwidth, phase, and stop band attenuation. Bandwidth should be reduced, because your filter won’t be steep, so it needs to start cutting early (lower frequency) to do much of anything. Stop band attenuation should be very much reduced, because the less steeply the filter cuts, the less Gibbs effect, colloquially called “ringing.” Finally, in order to eliminate pre-ringing and move the pre-ringing energy to after the signal so it’s more like natural reverberation, you probably want to switch to minimum phase.
There are different FIR filter types used in SDM…
Here’s a graphic from the link I provided above:
Again we get into the Nyquist Frequency of the SDM filter cut-off… at 11.2Mhz (DSD256) the Fc is 5.6MHz… ultrasonic noise is managed by the steepness of the filter… the Nyquist Frequency of the SDM filter cut-off for 5.6MHz (DSD128) is 2.8MHz… Any further artifacting is so below the threshold of hearing as to be inconsequential and of little consequence to a well designed DAC or amplifier.
The graphs for the FIR filters you showed are used in upsampling prior to sigma delta modulation.
Sigma delta modulation then takes the upsampled signal and modulates it so that the signal portion is relatively lower in frequency and the noise is moved to higher frequencies. The analog reconstruction filter can then be made from relatively inexpensive parts and still do an excellent job removing the noise, leaving the analog musical signal.
These results will vary by DAC. Look at the graphs here. With an iFi DAC (not the NEO), you can see that in terms of pushing noise to higher ultrasonic frequencies (which is what we want the sigma-delta modulator to do so that the noise can be effectively filtered by the analog reconstruction filter in the DAC) noise level at 100kHz with DSD128 and the A8 modulator was about -112dB. At 100kHz with DSD256 and the B8 modulator it was about -156dB.
This was measured with an iFi DAC by the developer of Audirvana’s sigma-delta modulators.
We’ve been through this before… Just read my responses on that post you linked…
From the Wikipedia article: “Delta-sigma modulation”
It is worth noting that if no decimation ever took place, the digital representation from a 1-bit delta-sigma modulator is simply a PDM signal, which can easily be converted to analog using a low-pass filter, as simple as a resistor and capacitor.
However, in general, a delta-sigma DAC converts a discrete time series signal of digital samples at a high-bitdepth into a low-bitdepth (often 1-bit) signal, usually at a much higher sampling rate. That delta-modulated signal can then be accurately converted into analog (since lower bitdepth DACs are easier to be highly-linear), which then goes through inexpensive low-pass filtering in the analog domain to remove the high-frequency quantization noise inherent to the delta-sigma modulation process.
Delta-sigma modulation - Wikipedia
This is the same for r8Brain or SoX implementation of SDM where the multi-bit up-sample frequency is subsequently decimated and modulated as a 1-bit PDM signal encapsulated in the DoP or raw 1-bit DSD signal presented to the output of the computer/OS platform.
Dynamic-range is relevant when there is perceptible and contextual musical elements that may be revealed in the recording…
I only upsample frequencies 16/44,1 to 32/88,2 and 24/48 to 32/96.
I think the result is fantastic when listening to old good recordings (Bach Toccata, Andre Isoir ; Agnes Obel ; Radiohead…). Over x2, there are often artefacts on less well-recorded albums (PJ Harvey, U2, The Cure…)
When I upsample DSD, PCM>24/96, I can’t listen any improvement, but sound is less natural and large with some albums.
What I posted has to do with this process. As I mentioned, the higher DSD rate (mostly) and the ‘B’ modulator do a better job at pushing the noise into higher frequencies where the final analog filter can more easily remove it.
Is the difference audible? Don’t know. I just figure starting with substantially lower noise to be filtered out can’t be a bad thing.
Changing 16 and 24 to 32 doesn’t really help anything, but it doesn’t hurt either. The change is done by “zero padding,” which simply writes zeroes into the new 8 or 16 bits. It’s like writing 1 as 1.00000000 - the quantity doesn’t change.
But converting 44.1 to 88.2 or 96 will change things. The rest of the upsampling will then be done by your DAC internally.
Up-sampling will not change the inherent signal that is encoded in the mastering process, so if you are detecting these things, it is because of the filter settings you have employed or you are now hearing the quality of the mastering process and fundamental recordings or DAC related distortions… The 16/44.14kHz encoded signal will have a brick-wall filter at approximately 22kHz, so no more high-frequency energy is created or passed on than was initially recorded in the A/D encoding.
Converting to a higher bit depth will reveal low-level energy that was recorded in the 16-bit A/D encoding… This low-level energy may or may not exist in every master recording…
So, you see… there are different perspectives on the results… However, you will find agreement from most, that having Audirvana/r8Brain or SoX do the conversion, is the right thing to do, in order to remove the processing overhead from the DAC platform, so to give it the best chance of providing the highest-quality of reproduction it is capable of, with any given digital-audio file.
I listened for a few hours Flac converted to DSD256.
It was really good, but again I thought something was missing, I don’t know how to define it.
Then I tried to use Flac for X2. It turned out sensational.
Of course this is my experience, but that’s what I thought was great with the iFi Neo DAC.
In the case of modulating to DSD, I believe 5.6MHz (DSD128) is the best target…
How much system RAM does your Mac system have? DSD256 requires a lot of system resources…
Obviously, all that matters is how it sounds to you, on your playback system… I have gone through the gamut of sample-rates on playback and by far, PCMxxx modulated to DSD128 provides the most satisfactory quality of playback sound on my system.
My Mac mini has 8Gb and is running the latest MacOS operating system.
It worked perfectly Flac for DSD256, but in my opinion it wasn’t so natural.
Now with Flac X2 it was sensational and that’s how I’m going to leave it.
What is your impression when up-sampling/modulating PCM 16/44.1kHz → 5.6MHz (DSD128)?
I personally see an 8GB system RAM to be a limiting factor in processing reserves for combined operations of r8Brain SDM processing to create the 1-bit 11.2MHz DSD256 signal data-stream and the subsequent DoP packetizing of that 11.2MHz DSD256 file data in Audirvana… It would be nice to get some technical insights that are measurable or from the Audirvana development team regarding computer performance related elements that may play into what you are experiencing with up-sampled PCMxxx → DSD256
DSD256 alone requires a lot of system resources… You may think it worked perfectly, but more-likely-than-not, your system is being taxed by the computational overhead that requires a lot of RAM… I can’t prove this to you, however, if modulating 16/44.1kHz → DSD128 sounds better, this will be evidence that DSD256 is taxing your MacMini/Audirvana system.
I don’t remember if you were running the MacMini headless or with a dummy video accelerator or not, What CPU Processor speed is running in the Mac? Is your LOCAL Library on an external storage device?
Sensational is good.