Martti,
Noise floor in the audio range for DSD128, DSD256 and DSD512 with a DSD coded silent track including large HF noise, I don't think you will find any sidebands.
And finally also a much lower DSD512 noise floor when just playing 010101....., a noise floor that's being the same for all three DSD rates that I tested.
Hans
P.S. My FFT can handle up to 1Mio points, in case of a 100Mhz FFT, bin width will be ca 100Hz and not 6K as mentioned in #1280, but this doesn't help either to make the BCLK jitter visible.
Noise floor in the audio range for DSD128, DSD256 and DSD512 with a DSD coded silent track including large HF noise, I don't think you will find any sidebands.
And finally also a much lower DSD512 noise floor when just playing 010101....., a noise floor that's being the same for all three DSD rates that I tested.
Hans
P.S. My FFT can handle up to 1Mio points, in case of a 100Mhz FFT, bin width will be ca 100Hz and not 6K as mentioned in #1280, but this doesn't help either to make the BCLK jitter visible.
Attachments
Last edited:
There is a lot of stuff that could be measured, but isn't. Why focus on this over all else?
https://www.thedrum.com/opinion/201...-the-focusing-illusion-key-better-advertising
https://www.thedrum.com/opinion/201...-the-focusing-illusion-key-better-advertising
Hi Hans,
Okay. But it is practical to take steps to reduce jitter and see if the sound is improved.
Besides, we also find that cleaning up power supplies and grounds helps. Have we focused on measuring power supply rail noise at device pins? No. Why not?
Then there is the question of, what will be done with anything learned? Use it to design a dac-4? Something else?
Mark
Okay. But it is practical to take steps to reduce jitter and see if the sound is improved.
Besides, we also find that cleaning up power supplies and grounds helps. Have we focused on measuring power supply rail noise at device pins? No. Why not?
Then there is the question of, what will be done with anything learned? Use it to design a dac-4? Something else?
Mark
Martti got a much lower noise floor after changing only the way the bit clock was generated, I got a lower noise floor after splitting the clock from the data part of my flatcable. That is, there is some circumstantial evidence.
I don't know who this mysterious "we" is but I have presented measurements of power supply rail noise:
Vref and clock doubler ps
https://www.diyaudio.com/community/attachments/rtz_bg_noise-jpg.1199181/
+15V pins
https://www.diyaudio.com/community/attachments/lt3045-15v_noise-jpg.1199182/
I doubt your power supplies are much cleaner than those. Especially due to radiated mains noise.
Measurements have shown that cleaner DCLK results in lower noise floor. So obviously measuring noise floor is important.
BTW some members seem to be using PCM2DSD board which has no reclocking of output signals. It is quite probable that it will result in elevated noise floor unless reclocking is added.
I generally like to get rid of artefacts that can easily be avoided.
Regarding the out-of-band noise of a single-bit sigma-delta modulator, it is not really simple additive noise. It's not even quantization noise that has its first and second moments made independent of the signal by triangular dithering. In fact there are often idle tones in it that are frequency modulated by the music. If one can avoid or minimize mixing that rubbish down to the audio band, I'm in favour of it.
Just tried the experiment we talked about offline to slow done common mode attenuation circuitry so that it mostly only is correcting for DC offset.
What I hear in SE mode is that the sound has opened up, exposing more low level musical details, and eliminating most of the remaining blur. Problem initially was that it also exposed a lot of low level grungy stuff that sounds bad. Learned not too long ago that most of the grunge can be filtered out with a long cable made out of the right kind of 3-conductor wire, with the 3rd conductor only grounded at the source end. Test cable is around 14' - 16', one cable for each channel between the dac output and the line amp input. No further information on the cable is available. That said, different types of cable may be found to sound different so might be worth some trial and error to see.
The changes were made to be reversible with 1M resistors on pin headers in series the the 2k opamp input resistors, and a pin header in parallel with the C0G feedback cap. Here is a pic of the modified filter board:
Also in case of any interest, a pic of the soldering iron to show its size:
What I hear in SE mode is that the sound has opened up, exposing more low level musical details, and eliminating most of the remaining blur. Problem initially was that it also exposed a lot of low level grungy stuff that sounds bad. Learned not too long ago that most of the grunge can be filtered out with a long cable made out of the right kind of 3-conductor wire, with the 3rd conductor only grounded at the source end. Test cable is around 14' - 16', one cable for each channel between the dac output and the line amp input. No further information on the cable is available. That said, different types of cable may be found to sound different so might be worth some trial and error to see.
The changes were made to be reversible with 1M resistors on pin headers in series the the 2k opamp input resistors, and a pin header in parallel with the C0G feedback cap. Here is a pic of the modified filter board:
Also in case of any interest, a pic of the soldering iron to show its size:
Last edited:
On that point, have been thinking about the blurred output in balanced mode versus more low level detail in SE mode. If there is a difference in sound (as there seems to be), then perhaps it arises at least in part because the inverting and non-inverting FIRDAC components are not closely matched enough for very small details to always render in equal but opposite phase. In that case what is equal and opposite is retained, and anything else is discarded as a common mode noise. Well, it appears that what is equal and opposite at low levels is mostly some blur. That's conjecture only, of course. YMMV.
Last edited:
It sounds as if you're getting close to what I was hearing from the RTZ when I listened back along using HQPlayer 4 to upsample to DSD256 (with the 'AMSDM7 512+fs' modulator) and the JLSounds USB board on the RTZ input (via an interface board with very short tracks). Initially that was with my Noir HPA but it got even better when I used my 300B SE-OTL HPA.
Mark, I`d almost ask a rhetorical question if you think those twisted power supply cables wouldn`t need to be of similar quality as your signal cables.
My opinion is that twisted cables should only be used when AC is running through them, on top of that they are rather thick for the currents running through them.
It could be that the thickness you ultimately like the signal cables to be, are skewed because the supply cables are a bit too thick, so you might be over compensating, if you get my drift.
My opinion is that twisted cables should only be used when AC is running through them, on top of that they are rather thick for the currents running through them.
It could be that the thickness you ultimately like the signal cables to be, are skewed because the supply cables are a bit too thick, so you might be over compensating, if you get my drift.
Probably not at the point where that's the biggest problem. Pretty sure I could make better wiring to the power supplies using, oh, say, maybe star-quad geometry, with some possible modifications. Besides what makes you think there isn't some AC running through the cables?
For now, if we don't remove certain common mode noise in order to preserve more low level musical detail then we may need more aggressive analog output filtering to help compensate? Something like that.
For now, if we don't remove certain common mode noise in order to preserve more low level musical detail then we may need more aggressive analog output filtering to help compensate? Something like that.
The reason why the DAC is differential is precisely that the signals are not exactly equal but opposite. If they were, it would suffice to just look at one of them and ignore the other - the other would not add any information.
That is, I've used the usual layout tricks to get the flip-flops to match, but the shift register output resistance is different for a high than for a low level. The effect of this is compensated to some extent thanks to the differential structure. Also there is common-mode noise coming from the reference (very little, but still) that cancels out and of course various sources of even-order distortion.
The reason for using a fast common-mode loop is to get most of these advantages even when you use the output single-endedly. If you were listening to the differential output signal, I could maybe come up with some far-fetched hypothesis why a slow common-mode loop might be slightly better than a fast one, but it doesn't make any sense to me when you listen to the single-ended output.