Well, no chance to test again with your loved setup...
while any converter using any FIR gets IMHO 👎 as ringings to avoid (extra sound)
while any converter using any FIR gets IMHO 👎 as ringings to avoid (extra sound)
@olo111 and or @PJotr25,
Noticed this DSD converter can sound a bit bright/hard/distorted at times. Turning down the PCM volume level that feeds the converter seems to fix the problem.
In that regard found the following online:
"... when we're encoding DSD music, the "rule" is that the output level should be around 0dBDSD = PCM -6dBFS. But it is allowable that the signal be a little "hotter" into +3.1dBDSD = PCM -2.9dBFS for "short" periods of time at least."
"...all test DSD signals must be set to peak at +3.1dBDSD (-2.9dBFS PCM)."
Quotes are from: http://archimago.blogspot.com/2022/06/notes-on-dac-dsd-1-bit-pdm-measurements.html
Could it be the above relationship accounts for the volume level issue I am hearing?
Noticed this DSD converter can sound a bit bright/hard/distorted at times. Turning down the PCM volume level that feeds the converter seems to fix the problem.
In that regard found the following online:
"... when we're encoding DSD music, the "rule" is that the output level should be around 0dBDSD = PCM -6dBFS. But it is allowable that the signal be a little "hotter" into +3.1dBDSD = PCM -2.9dBFS for "short" periods of time at least."
"...all test DSD signals must be set to peak at +3.1dBDSD (-2.9dBFS PCM)."
Quotes are from: http://archimago.blogspot.com/2022/06/notes-on-dac-dsd-1-bit-pdm-measurements.html
Could it be the above relationship accounts for the volume level issue I am hearing?
Yes, it is possible. According to our measurements, the DSD converter introduces about 2.9dB of attenuation. Sometimes it may not be enough...
That blog Mark founds includes a link to part 1 of the Scarlet Book, which specifies just about everything except the audio format. Has anyone been able to locate the other parts?
Edit: I found it, there is a button in the upper left corner to select the other parts.
Edit: I found it, there is a button in the upper left corner to select the other parts.
Last edited:
The number of ones in any section of 28 consecutive bits has to be at least 4 and at most 24 to be standard compliant.
That is, when you interpret a one as +1 and a zero as -1, the output signal of an FIR filter with uniform weights of 1/28 over 28 taps (moving-average filter) is to remain in the range from -5/7 up to and including +5/7 at all times. As such a filter has only a limited suppression for ultrasonic quantization noise, doesn't that mean that the desired audio has to stay well below +/- 5/7?
That is, when you interpret a one as +1 and a zero as -1, the output signal of an FIR filter with uniform weights of 1/28 over 28 taps (moving-average filter) is to remain in the range from -5/7 up to and including +5/7 at all times. As such a filter has only a limited suppression for ultrasonic quantization noise, doesn't that mean that the desired audio has to stay well below +/- 5/7?
Last edited:
In addition to Scarlett book and 0dBFS, what about a possible issue due to intersample overs: https://benchmarkmedia.com/blogs/application_notes/intersample-overs-in-cd-recordings
There is another thread for it. May be very good for upsampling PCM. However it doesn't appear to convert PCM to DSD.
Looking at section E.1 of the super audio CD standard, it seems that positive full-scale PCM is supposed to correspond to on average 75 % ones and 25 % zeros in the DSD signal, and negative full scale to on average 25 % ones and 75 % zeros.
The moving average filter of post #686 then produces +1/2 with some quantization noise on top or -1/2 with some quantization noise on top. The range from 1/2 to 5/7 is then available for quantization noise and intersample overshoots.
Tentative impression here from listening to some music with female vocals (including harmonies), is that setting Foobar2000 output level to -6dBFS clears up audible distortion and makes female voices sound more full and natural (the distortion seems more evident with higher frequency sounds). Test song true peak level (reconstructed according to ITU standards) is -.5dBFS.
Also, increasing PCM attenuation also has the effect of audibly reducing SNR/DNR. Sound starts to suffer if excess PCM attenuation.
Will do some more listening later with other tracks including at 16 and 24 bit resolutions. Will also get some more listeners involved to try get a better idea of attenuation level tradeoffs, at least as perceived on the system here.
Also, increasing PCM attenuation also has the effect of audibly reducing SNR/DNR. Sound starts to suffer if excess PCM attenuation.
Will do some more listening later with other tracks including at 16 and 24 bit resolutions. Will also get some more listeners involved to try get a better idea of attenuation level tradeoffs, at least as perceived on the system here.
Another set of data points:
Alan Parsons - I Robot (CD version) True Peak = +.7dB
Alan Parsons - I Robot (24/192) True Peak = -5.4dB
CD version sounds a little cleaner with Foobar PCM output set to -7dBFS.
Alan Parsons - I Robot (CD version) True Peak = +.7dB
Alan Parsons - I Robot (24/192) True Peak = -5.4dB
CD version sounds a little cleaner with Foobar PCM output set to -7dBFS.
I think we have reached a consensus here. There is a tradeoff between reproduction of the stereo illusion of space (soundstage) and minimization of distortion.
Probably an additional 6dB of attenuation gives the best tradeoff. However, in some cases 7dB of attenuation sounds better from a distortion perspective if peak level of the PCM recording is +0.7dB. Thus, it seems like the existing 2.9dB of attenuation is not leaving additional headroom for intersample overs?
Probably an additional 6dB of attenuation gives the best tradeoff. However, in some cases 7dB of attenuation sounds better from a distortion perspective if peak level of the PCM recording is +0.7dB. Thus, it seems like the existing 2.9dB of attenuation is not leaving additional headroom for intersample overs?
Last edited:
A typical SACD uses a sixth-order 64OSR, resulting in a maximum stable amplitude of about -5.7dBFS. Therefore, I believe they have set DSD_0dBFS=PCM_-6dBFS to ensure that DSD remains stable. DSD_3.1dBFS is not mandatory and implies a compromise up to that level. This number is unattainable and unstable with a normal sixth-order DSM. In the quoted article #683, there is an update on June 21, 2022, discussing the characteristics when switching back to PCM.
Looking at DSD64, the THD+N is -116dB. This figure is inferior to the conversion capability of, for example, AK4499EQ, so it cannot be used for measuring AK4499EQ, at least. DSD128 and DSD256 yield satisfactory numbers, but their quantization noise characteristics differ from those of the typical DSM. Usually, DSD256 is simply DSD128 with the frequency doubled.
I'm not familiar with the details, but I believe they intentionally shifted the optimal operating point of NTF to increase the maximum stable amplitude(MSA). As a side effect, THD+N worsened for DSD64, and quantization noise increased for DSD128 and DSD256. I think SACD production facilities make such compromises for signals exceeding DSD_0dBFS. It is not advisable to use them for measurements as it can lead to unnecessary misunderstandings.
Looking at DSD64, the THD+N is -116dB. This figure is inferior to the conversion capability of, for example, AK4499EQ, so it cannot be used for measuring AK4499EQ, at least. DSD128 and DSD256 yield satisfactory numbers, but their quantization noise characteristics differ from those of the typical DSM. Usually, DSD256 is simply DSD128 with the frequency doubled.
I'm not familiar with the details, but I believe they intentionally shifted the optimal operating point of NTF to increase the maximum stable amplitude(MSA). As a side effect, THD+N worsened for DSD64, and quantization noise increased for DSD128 and DSD256. I think SACD production facilities make such compromises for signals exceeding DSD_0dBFS. It is not advisable to use them for measurements as it can lead to unnecessary misunderstandings.
I was also a bit surprised about the rather high maximum level of 5/7 in the standard. You can reduce the out-of-band noise gain to increase the maximum stable amplitude of a single-bit sigma-delta modulator, but that results in less quantization noise suppression in the audio band (per the Gerzon-Craven noise shaping theorem). Of course that 5/7 is including the quantization noise that passes a 28-tap moving-average filter, so the actual desired signal level must be a bit lower.
As I did some DSD SDM implementations, the rule of thump was -6dB of PCM, but some PCM to DSD modulator SW requires some more headroom as only view -0.x dB's only.
The reason for this is that the HW implementations may not floating point related so internal blocks may overdraft.
In addition any PCM inter sample peaks to consider as over drafting even on PCM.
The reason for this is that the HW implementations may not floating point related so internal blocks may overdraft.
In addition any PCM inter sample peaks to consider as over drafting even on PCM.
I have previously successfully implemented a streamer using the PCM2DSD and RTZ DAC with Squeezebox Receiver as the source of the I2S.
I am now doing a similar thing based on Logitech Transporter and using LVDS to transfer the I2S.
I used MCLK in my implementation with the Squeezebox but I am wondering if the PCM2DSD implementation requires it?
Can someone please confirm whether the PCM2DSD mandates the presence of the master clock?
At the moment without MCLK I am getting nothing out of the device but that could also be a fault in my programming of it or a bad solder joint so it would just be great to have positive confirmation before I go digging onto the Transporter mainboard again to pick up MCLK.
Thanks!
Tony
I am now doing a similar thing based on Logitech Transporter and using LVDS to transfer the I2S.
I used MCLK in my implementation with the Squeezebox but I am wondering if the PCM2DSD implementation requires it?
Can someone please confirm whether the PCM2DSD mandates the presence of the master clock?
At the moment without MCLK I am getting nothing out of the device but that could also be a fault in my programming of it or a bad solder joint so it would just be great to have positive confirmation before I go digging onto the Transporter mainboard again to pick up MCLK.
Thanks!
Tony
