FIR is a digital filter so it must be applied before the DA stage. But a FIR needs a multibit file to process. So for R26 to perform better on the SINAD it must turn the DSD back to PCM. Then speculation from the L27 measurements you posted indicates it then runs them through the R2R PCM ladder as indicated by the much lower noise floor
Thanks. Can a FIR work on DSD directly? I thought a FIR needs multibit ie PCM to process?
So it looks like for R26 to play DSD without loads of noise, it turns it back to PCM (back as most DSD was originally recorded and mastered in PCM) and then uses the R2R ladder.
Thanks. Can a FIR work on DSD directly? I thought a FIR needs multibit ie PCM to process?
So it looks like for R26 to play DSD without loads of noise, it turns it back to PCM (back as most DSD was originally recorded and mastered in PCM) and then uses the R2R ladder.
External clocks need to be at stable temperature, so many leave on 24/24. Too bad if there's a power surge.
FIR is a digital filter so it must be applied before the DA stage. But a FIR needs a multibit file to process. So for R26 to perform better on the SINAD it must turn the DSD back to PCM. Then speculation from the L27 measurements you posted indicates it then runs them through the R2R PCM ladder as indicated by the much lower noise floor
This oversampling/interpolation process takes a lot of weight off any low pass filter used (analog fir filter) and spreads the quantization noise over a wider region which improves SINAD.
From my understanding volume adjustment on DSD is not possible, so if one can adjust volume when DSD direct is off R26 must internally convert from DSD to PCM.
From my understanding volume adjustment on DSD is not possible, so if one can adjust volume when DSD direct is off R26 must internally convert from DSD to PCM.
Notice how you can’t use the volume when PCM NOS is turned on either. This also bypasses the interpolation fir filter. It’s not about needing a conversion.
It’s about how digital volume control works, word length. and how many times we have to represent that word length every second. 16 bit, 20 bit, 24 bit. That's the length of the "word". The number of times per second we have to represent that word is the sample rate: 44,000, 88,000, 192,000 etc.
When you use digital volume control you are lowering the original number (word length) and losing bits. 1 bit for every 6dB.
Oversampling/upsampling increases the word length from the original size. 16 bit goes to 32bit. Then you have 16 bits of volume or 96db of dynamic range.
So imagine what happens when you turn the oversampling off. With DSD you only have 1bit.
External clocks need to be at stable temperature, so many leave on 24/24. Too bad if there's a power surge.
FIR is a digital filter so it must be applied before the DA stage. But a FIR needs a multibit file to process. So for R26 to perform better on the SINAD it must turn the DSD back to PCM. Then speculation from the L27 measurements you posted indicates it then runs them through the R2R PCM ladder as indicated by the much lower noise floor
This will help to understand what is going on here.
Also a good video to watch that shows this happening is GoldenSound discovering that the Denafrips Ares 2 does not have a true NOS mode. He caught them lying lol.
You can skip to the 23:45 mark
Smart. There's a simple test anyone can do remote in hand. So turn "DSD off" and try R26 volume: if the volume changes then => DSD is being processed as PCM. DSD in and PCM out
@Nada Notice how you can’t use the volume when PCM NOS is turned on either. This also bypasses the interpolation fir filter. It’s not about needing a conversion.
It’s about how digital volume control works, word length. and how many times we have to represent that word length every second. 16 bit, 20 bit, 24 bit. That's the length of the "word". The number of times per second we have to represent that word is the sample rate: 44,000, 88,000, 192,000 etc.
When you use digital volume control you are lowering the original number (word length) and losing bits. 1 bit for every 6dB.
Oversampling/upsampling increases the word length from the original size. 16 bit goes to 32bit. Then you have 16 bits of volume or 96db of dynamic range.
So imagine what happens when you turn the oversampling off. With DSD you only have 1bit.
Look at this photo
DSD isn’t converted to PCM in this case.
Do you think that oversampling DSD a PCM conversion has to take place?
Are you saying a finite impulse response (FIR) filter can be done on DSD one bit files directly?
Smart. There's a simple test anyone can do remote in hand. So turn "DSD off" and try R26 volume: if the volume changes then => DSD is being processed as PCM. DSD in and PCM out
Here is a test R26 owners can perform to verify if they hear the same noise difference as I do.
(I hope Jussi does not mind I share his test file with you)
The test goes as follows:
1. Download the following file (containing an inaudible 1k sine at -120dB): https://www.signalyst.eu/tmp/sine1k_-120dB-2.wav
2. Let Roon or HQPlayer upsample it to DSD and stream it to R26
3. On R26 toggle Direct DSD = ON and OFF in the menu and listen if you hear a difference in the noise level (best using headphones)
Can anyone here reproduce my findings, do you also hear increase in noise when DSD Direct = ON ? ( in other words using the dedicated DSD DAC)
@Nada
DSD direct “off”
DSD signal -Digital Interpolation fir filter (oversampled) -Digital modulator (noise shaping and dither) - 1bit Dac- Analog output filter - Analog signal
Volume works
DSD direct on
DSD signal -1bit DAC-Analog output filter -Analog signal
Volume doesn’t work.
You are oversampling the DSD signal. Gustard is calling this a interpolating Fir
Filter. Because it is oversampled is exactly why the volume works. There is a greater bit rate. Dynamic range is huge.
Volume doesn’t work when this interpolation/oversampling is bypassed because there is no bit depth.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
