To further clarify, the DSD BCLK has been doubled to produce the yellow scope trace. Rising edges of the doubled clock that correspond to rising edges of the DSD BCLK are used to clock in new DSD data bits. Rising edges of the doubled clock that correspond to falling edges of the DSD BCLK are used to perform the RTZ function. So long as the timing is consistent for each clocking function there should be no problems. In case I didn't get that quite right, maybe Marcel can chime in with a comment.
I feel a bit confused.. the 'data' traces are changing right after the clock pulse rising edge, and an RTZ like operation is visible on the Blu (Cyan) data trace, after the second clock pulse.. It all make sense for me if the data traces are the Q outputs of the 74LV574 registers..
Hans,
The experiment I am setting up for is to try generating RTZ input to the shift registers using a crystal clock instead of an XOR gate. Maybe it will make some difference or maybe not. One thing at a time is all.
Also, this experiment is something we already discussed a little bit at some point earlier in the thread.
Because the crystal clock might provide more stable timing for RTZ, it might help provide more stable reproduction of recording amplitude information (since the area under the curve of the RTZ pulses may be more consistent from pulse to pulse). Something more or less like that.
Mark
The experiment I am setting up for is to try generating RTZ input to the shift registers using a crystal clock instead of an XOR gate. Maybe it will make some difference or maybe not. One thing at a time is all.
Also, this experiment is something we already discussed a little bit at some point earlier in the thread.
Because the crystal clock might provide more stable timing for RTZ, it might help provide more stable reproduction of recording amplitude information (since the area under the curve of the RTZ pulses may be more consistent from pulse to pulse). Something more or less like that.
Mark
Last edited:
I proposed a similar approach where the DSD-to-RTZ conversion is done in the USB-board MCU. In theory this should be less noisy since no digital input circuitry is needed as USB board is connected directly to shiftregisters. I even made some tests but in practice performance did not improve which indicates that the analog output circuitry is the main noise source. But the main reason I put this to back burner is that I could not come up with a lightweight DSD-to-RTZ conversion that avoids having to do the conversion bit-by-bit. Even with DSD128 that becomes too CPU-intensive for these MCUs. Once I have my new STM32H7 board honed I may give it another try since it has 2.5x more processing power.
Here is what the present setup looks like:
Sound is getting increasingly precise with the crystal RTZ, but I also hear more clearly an ugly HF distortion and 'S' sounds that are too 'Sh' sounding. Also, at the moment playback is limited to no greater than DSD128 due to use of the reclocked BCLK (instead of using MCLK) to clock the dac.
Maybe will try some things to see about the bright, HF distortion (or correlated noise, whatever it is). Looking at the schematic it appears the are some 22uf, X5R caps to ground at the power pins of the shift registers (on Vref!). May try replacing those with some SMD film caps or something to see if the power supply is a source of some remaining issues.
Also at some point want to get DSD256 working. For now the now-more-clearly-audible distortion is more of a bother.
Sound is getting increasingly precise with the crystal RTZ, but I also hear more clearly an ugly HF distortion and 'S' sounds that are too 'Sh' sounding. Also, at the moment playback is limited to no greater than DSD128 due to use of the reclocked BCLK (instead of using MCLK) to clock the dac.
Maybe will try some things to see about the bright, HF distortion (or correlated noise, whatever it is). Looking at the schematic it appears the are some 22uf, X5R caps to ground at the power pins of the shift registers (on Vref!). May try replacing those with some SMD film caps or something to see if the power supply is a source of some remaining issues.
Also at some point want to get DSD256 working. For now the now-more-clearly-audible distortion is more of a bother.
Mark,
Not at all meant to criticize the hard work you are doing, I'm really impressed how deep you are going, but it's almost as if you are testing a car and at the end the only part that passes the test is the steering wheel, in this case the shift registers, but without digital RTZ circuitry and without the analog reconstruction filter
Hans
Not at all meant to criticize the hard work you are doing, I'm really impressed how deep you are going, but it's almost as if you are testing a car and at the end the only part that passes the test is the steering wheel, in this case the shift registers, but without digital RTZ circuitry and without the analog reconstruction filter
Hans
Another experiment:
4-each, 22uf, X5R caps replaced with Rubycon 22uf, 16v, film P/N - 16MU226MD35750
Not cheap, but sound is very different. Initial impression is full, warm midrange, big soundstage, very nice listening. Not even burned in yet. Will let it run awhile and see if anything changes. (Just noticed one of the caps doesn't look fully wetted with solder on the ground plane side. Will check more closely later and touch up if needed.)
4-each, 22uf, X5R caps replaced with Rubycon 22uf, 16v, film P/N - 16MU226MD35750
Not cheap, but sound is very different. Initial impression is full, warm midrange, big soundstage, very nice listening. Not even burned in yet. Will let it run awhile and see if anything changes. (Just noticed one of the caps doesn't look fully wetted with solder on the ground plane side. Will check more closely later and touch up if needed.)
Wow, €10,- a piece,
Other than mentioning polymer, Rubicon doesn’t tell what polymer the dielectricum is made of.
The same change could be done to the X5R and X7R caps on the analog reconstruction filter, although less expensive 10uF alu smd’s and polypropylene or NP0 100nF could be used.
An idea ?
Hans
Other than mentioning polymer, Rubicon doesn’t tell what polymer the dielectricum is made of.
The same change could be done to the X5R and X7R caps on the analog reconstruction filter, although less expensive 10uF alu smd’s and polypropylene or NP0 100nF could be used.
An idea ?
Hans
Not sure those are going to stay. Give it a day or two. Main thing to note is that Vref tends to be audibly sensitive to changes like that. Its possible at some point an external Vref supply will be tried. Still thinking about it.
My experience is opposite though I don't use X5R. Here is a recent example of what happens with film (or C0G) as bypass caps in a very similar circuit:
https://www.diyaudio.com/community/threads/the-best-dac-is-no-dac.273474/post-7423065
https://www.diyaudio.com/community/threads/the-best-dac-is-no-dac.273474/post-7423065