Hi Hans, I've hidden the answer in the opening post of https://www.diyaudio.com/forums/dig...n-zero-shift-register-firdac.html#post6847540
Marcel was so generous to lend me his solid state dac, giving me the possibility to look at it from different angles and more importantly, to audition the Dac.
The review I made is rather lengthy, that why I attached it as PDF.
Only the resume is repeated here.
Hans
Resume
The review I made is rather lengthy, that why I attached it as PDF.
Only the resume is repeated here.
Hans
Resume
- My current Dac is a Bel Canto 3.5VB equipped with the PCM1792, but in it’s final shape, Marcel’s Dac improved on it by producing more clearness at the high end of the audio spectrum, without any sharpness.
I can highly recommend this excellent product.
- This Dac needs a very good low noise 15Volt supply, like with a LT3045 LDO regulator to prevent 50Hz AM modulation.
- The LT3042, providing the critical 5V supply, needs an alu or tantalum electrolytic cap on its set pin, a ceramic cap is not suited for the job.
- In its original version, the reconstruction filter is quite noisy, worsening the noise produced by the digital section by several dB’s.
- Not to change too much components, modifications where made that lowered the noise but must not be seen as the one and only solution, because the op-amps with this mod have to process signals that are 4 times as large. Because of that small IMD distortion emerged with 19Khz+20Khz signals both at -6dB. That’s 10dB more signal as used in Fig 12 in the review.
Marcel’s original noisy filter did not show any IMD at all levels.
However, the filter used for Marcel’s RTZ Firdac has a much better noise figure because of A) the used OPA2210 for the first stage, having almost 4 times less voltage noise, and because B) of the much lower 3K01 Firdac resistor values instead of the 32K4 in this design. I already used 6K98 without any noticeable negative effects.
So, this shiftregister design seems the best option for the lowest additional noise and the lowest possible IMD.
- Result of the reconstruction’s filter noise reduction:
Table 2, Effect on overall noise by reducing the reconstr. Filter’s noise
- Be very careful when placing large caps in the Dac’s output, because that may have a direct impact on perceived sound reproduction. Most likely you won’t need them because of the very low output offset voltage of below 2mV. If so you can also simplify the output buffer by directly connecting output to minus input.
- I have a preference for separated digital and analog ground, which is more or less the case with the Shiftregister RTZ Firdac. However given the quality of the reproduced sound, there was no reason to believe that with this design separating the two would have brought any benefit and the achieved S/N of around 108dBA seems more than adequate.
Actually, I haven't a clue how large or small the intermodulation distortion with 19 kHz and 20 kHz signals of the filter of the RTZ shift register FIRDAC is. I haven't got the equipment to measure it at home, only to do harmonic distortion measurements (and those not nearly as accurately as Hans).
The pole positions of the RTZ shift register FIRDAC filter are different from those of the 74AHC02/74AHC08 logic gate DAC. The logic gate DAC's filter is optimized for a good phase response, because the resulting droop of the magnitude response is corrected for in the digital part, at least up to well above 20 kHz. The shift register DAC is a DSD DAC with no such correction, so I chose a Butterworth reconstruction filter for it. Up to 20 kHz, the differences are only small, of the order of 0.2 dB.
It was a big surprise that Hans measured a far lower noise floor than I did. I knew the filter noise contributed substantially to the total noise, but now that the total noise turns out to be lower than expected, it actually dominates the total noise entirely. I will check what went wrong with my measurements. One hypothesis is limited alias suppression combined with the out-of-band noise of the DAC, but whether that is the true explanation remains to be seen. Once that is debugged, I intend to do some filter redimensioning experiments of my own.
Anyway, I'm happy I leant the 74AHC02/74AHC08 logic gate DAC to Hans, as the results of his measurements and experiments are quite interesting.
The pole positions of the RTZ shift register FIRDAC filter are different from those of the 74AHC02/74AHC08 logic gate DAC. The logic gate DAC's filter is optimized for a good phase response, because the resulting droop of the magnitude response is corrected for in the digital part, at least up to well above 20 kHz. The shift register DAC is a DSD DAC with no such correction, so I chose a Butterworth reconstruction filter for it. Up to 20 kHz, the differences are only small, of the order of 0.2 dB.
It was a big surprise that Hans measured a far lower noise floor than I did. I knew the filter noise contributed substantially to the total noise, but now that the total noise turns out to be lower than expected, it actually dominates the total noise entirely. I will check what went wrong with my measurements. One hypothesis is limited alias suppression combined with the out-of-band noise of the DAC, but whether that is the true explanation remains to be seen. Once that is debugged, I intend to do some filter redimensioning experiments of my own.
Anyway, I'm happy I leant the 74AHC02/74AHC08 logic gate DAC to Hans, as the results of his measurements and experiments are quite interesting.
Last edited:
Hi Marcel,
IMD at 0dBFS with the modified filter is shown in the attachment.
There´s little to worry about with second order at 3.7ppm and third order at 11ppm.
And to show things in perspective, I´ve also added the IMD from my Bel Canto 3.5VB, as measured by Stereophile and being at similar levels.
But with the original noisy filter, IMD was not measurable at any level, so the right filter could be less noisy and without IMD.
Hans
IMD at 0dBFS with the modified filter is shown in the attachment.
There´s little to worry about with second order at 3.7ppm and third order at 11ppm.
And to show things in perspective, I´ve also added the IMD from my Bel Canto 3.5VB, as measured by Stereophile and being at similar levels.
But with the original noisy filter, IMD was not measurable at any level, so the right filter could be less noisy and without IMD.
Hans
Attachments
Hi Hans ... very interesting results you have obtained with the RTZ FIRDAC - thanks for sharing your changes, and in such detail. It makes it easier to follow what causes which changes and also the intricacies involved in audio design 
Impressive also that these designs of Marcel can reach this level of performance
Kudos to you & Marcel for making this possible ...
Just as a side remark I have recently been experimenting with a No-DAC approach and without yet optimizing anything in terms of filter capacitors or output resistors I get from - 100 to - 105 dB 2H from a 74AUP1G74. And, remarkable I think, the output spectrum is "incredibly" clean with very few harmonics and an attenuating noise level dropping up till ~60 kHz where the noise level starts to rise again (HQPlayer ASDM5ECv2 modulator - 1 dB output, AD7760 EVM board). The overall noise level, however, to my memory is a bit higher than what you measure on the RTZ FIRDAC's - around - 130 dB.
One question in relation to your review document, if I may: When reading through the review document it seems to me that the changes you made were only to the power supply and the reconstruction filter, is this correct? If this is so then these figures were obtained with the RTZ FIRDAC converter section "as is" designed by Marcel?
Cheers, Jesper
Impressive also that these designs of Marcel can reach this level of performance
Kudos to you & Marcel for making this possible ...Just as a side remark I have recently been experimenting with a No-DAC approach and without yet optimizing anything in terms of filter capacitors or output resistors I get from - 100 to - 105 dB 2H from a 74AUP1G74. And, remarkable I think, the output spectrum is "incredibly" clean with very few harmonics and an attenuating noise level dropping up till ~60 kHz where the noise level starts to rise again (HQPlayer ASDM5ECv2 modulator - 1 dB output, AD7760 EVM board). The overall noise level, however, to my memory is a bit higher than what you measure on the RTZ FIRDAC's - around - 130 dB.
One question in relation to your review document, if I may: When reading through the review document it seems to me that the changes you made were only to the power supply and the reconstruction filter, is this correct? If this is so then these figures were obtained with the RTZ FIRDAC converter section "as is" designed by Marcel?
Cheers, Jesper
Hi Jesper,
Yes, all I did was trying to reduce the reconstruction´s filter noise, reduce 50Hz AM with a stabilized supply, change the ceramic cap on the LT3042´s set pin into a electrolytic cap and modify the output buffer into a straightforward buffer without caps.
But, this was all with the PCM Dac described in this thread, which includes indeed a RTZ Firdac with 74AHC02 and 74AHC08 gates, but has not to be confused with Marcel´s Shift Register RTZ Firdac using a 74LV574A, to be used for DSD only.
And all Kudos are for Marcel, I didn´t change anything to the perfect basic concept.
Hans
Yes, all I did was trying to reduce the reconstruction´s filter noise, reduce 50Hz AM with a stabilized supply, change the ceramic cap on the LT3042´s set pin into a electrolytic cap and modify the output buffer into a straightforward buffer without caps.
But, this was all with the PCM Dac described in this thread, which includes indeed a RTZ Firdac with 74AHC02 and 74AHC08 gates, but has not to be confused with Marcel´s Shift Register RTZ Firdac using a 74LV574A, to be used for DSD only.
And all Kudos are for Marcel, I didn´t change anything to the perfect basic concept.
Hans
Hi again, Hans ... & thanks for your feedback.
I think you read between the lines here because I had mistaken this DAC for "being an earlier version" of the RTZ DAC for DSD that is currently being talked about in another thread. Looking deeper into the schematic of this DAC I can see that this is not the case. My mistake - sorry!
Anyway, impressive results Marcel & you have reached (and I reckon the main objective is that it sounds as wished/hoped for ) ...
Have a fine evening,
Jesper
I think you read between the lines here because I had mistaken this DAC for "being an earlier version" of the RTZ DAC for DSD that is currently being talked about in another thread. Looking deeper into the schematic of this DAC I can see that this is not the case. My mistake - sorry!
Anyway, impressive results Marcel & you have reached (and I reckon the main objective is that it sounds as wished/hoped for
) ...Have a fine evening,
Jesper
It can't be aliasing, actually. I measured about the same audio noise floor in all three sigma-delta modulator modes, while the out-of-band quantization noise is very different. I should have measured far worse audio noise in PWM8 mode than in chaotic mode if my measurements had been affected by aliasing of out-of-band noise inside the FR2-LE field memory recorder that I used for measuring.
hi MarcelvdG,where did this module like firdsd6 and fir384interpol2 come from ?
They were generated by the Xilinx FIR compiler version 5.0. See https://linearaudio.net/design-data-package-marcel-van-de-gevels-valve-dac-vol-13 and specifically https://linearaudio.net/sites/linearaudio.net/files/ThevalveDAC_PCB_FPGA_files for web posting.zip for details. firdsd6 is a six-bit version of firdsd.