And you will get a lot of distortions and parasitic tones. especially after attack.
Many peoples tried to do this, but it is not so simple - you have to use or AFIR technology, or RTZ technology, or both together.
After that you will find that AK4497, ES9038 and BD34301/352 are much better
Alex.
The bottom line is the 300 average of the -1dB stimulus, the green is the active -1, the red the one from memory relative/referenced to the -1dB (Full Scale="0dB"), but with a stimulus of -60dB.
That -60dB is just to display no problems caused by uneven rise and fall times, but I'm pretty sure you know those issues;-)
Signal is generated by converting a 1KHz (998?) 24 bit pcm signal to dsd512 with HQPlayer. Screen dump is from 2019, so I don't remember all details, but once I'm up and running again, I'll be back to this.
Loved fiddling with these things.
Btw kudos for your efforts developing and sharing those same type of DAC designs (tube one and the latest). They're fun to do, right?!
Alex, it's doable when fast logic is used. CML works great but is very expensive, FIR is in principle easy and cheap. Many of them actually sound very good, but they probably will never beat the better chips without considerable effort, but this is DIY audio and I actually got to appreciate both the chips as those self made designs even more. One of the best performing designs like Mola Mola and a design here on DIYaudio a few years back, are DSD (or PDM, DSM or whatever you fancy;-) in a FIR arrangement.
@marcel: the -54dB might have been caused by having to change the level of the original PCM signal in HQPlayer with the volume control. At that time I was experimenting with the maximum signal combined with least distortion, so fiddling with the volume to get the max headroom might have been the cause of the shift of that 0dB, or -60/-54 dB level, as -6dB PCM level is usually used for a 0dB DSD level.
You mean "1-bit output from the fast logic + LPF"?
No AFIR, no RTZ?
Did not tried with CML, but with 74LVC there were a lot of overtones at 1-bit+LPF configuration.
Changing to differential 16-bit AFIR helps.
Alex.
Knowing Bruno's love for class D amps he should use PWM, right?! ;-)
It's still discrete with 32 taps FIR though and it's hard to prove it has flaws even with an AP measuring it. Great feat. Not yet heard it though.
Great review, which also describes a bit the subjective differences between some of these DAC architectures. I agree with those.
https://parttimeaudiophile.com/2020/09/25/mola-mola-tambaqui-dac-review/?amp
You could argue that because of his early love of DSD (his ADC design used it) then he'd go that way - maybe along the lines of PS Audio's top DAC(s). But I'm not completely sure he's really such a big fan of DSD nowadays. Bruno's amps use non-quantized PWM and are self-oscillating, without a master clock. The DAC can't do without one though. As I understand it, the flaw is that the latency is indeterminate due to the ultra-low jitter phase locking method he chose which precludes it being used in active systems where all channels need a consistent and equal time delay. I heard at least one customer sent it back for this reason.
Yes there is an output transversal filter after the PWM generation.
Yes there is an output transversal filter after the PWM generation.
@Alex, yes no FIR, just a low pass passive filter, symmetrical output straight into the measurement device.
The trick was to use the HQPlayer AMSDM7 modulator.
You'd have to ask Jussi why that is, but previous experience shows he won't let you share why exactly that also works best with most of the other discrete FIR DACS.
The trick was to use the HQPlayer AMSDM7 modulator.
You'd have to ask Jussi why that is, but previous experience shows he won't let you share why exactly that also works best with most of the other discrete FIR DACS.
Bruno's PWM Noise Shaper patent: https://patents.google.com/patent/WO2004049561A1
Some related developments:
https://patents.google.com/patent/US6972704
https://ieeexplore.ieee.org/document/5537341
https://www.teledynedalsa.com/en/pr...gn/design-products/pwm-sigma-delta-modulator/
Some related developments:
https://patents.google.com/patent/US6972704
https://ieeexplore.ieee.org/document/5537341
https://www.teledynedalsa.com/en/pr...gn/design-products/pwm-sigma-delta-modulator/
'Pure DSD' produced how? From a modern sigma-delta ADC?
In most modern cases there isn't fundamentally 1-bit A/D conversion, and back when maybe there was (like when SACD first came out?) it was suboptimal because it was DSD64 or DSD128.
DSD doesn't start sounding really good until at least DSD256. Also turns out the sweet spot for highest SQ for many DSD dacs is at DSD256.
In most modern cases there isn't fundamentally 1-bit A/D conversion, and back when maybe there was (like when SACD first came out?) it was suboptimal because it was DSD64 or DSD128.
DSD doesn't start sounding really good until at least DSD256. Also turns out the sweet spot for highest SQ for many DSD dacs is at DSD256.
From .dff/.dsf file with music, also generated signals directly in DSD.
Doesn't matter - no connection with this discussion. I talked about high THD and (which is more important) - about the "echo".
@abraxalito,
Thanks, great info.
I can imagine this being a problem when the NIC or USB is used, since in that case the receiving end dictates the exact speed of playback and it inherently can't get synced accross the other converters. Shouldn't be a problem with spdif though?
He designed that Grimm converter and afaik Channel Classics used it for their recordings, but despite the high quality of that unit, there was something along the lines of it being a somewhat "lossy" format, because several back and forth conversions between DSD and PCM would add too much hf noise to the signal. Of course there's truth in that, especially at 64fs.
Thanks, great info.
I can imagine this being a problem when the NIC or USB is used, since in that case the receiving end dictates the exact speed of playback and it inherently can't get synced accross the other converters. Shouldn't be a problem with spdif though?
He designed that Grimm converter and afaik Channel Classics used it for their recordings, but despite the high quality of that unit, there was something along the lines of it being a somewhat "lossy" format, because several back and forth conversions between DSD and PCM would add too much hf noise to the signal. Of course there's truth in that, especially at 64fs.
IIUC Bruno's adaptive ASRC/buffer/phase-locker with variable latency was to remove jitter from SPDIF. Again IIRC, he wanted the PPLL (polyphase locked loop) servo corner frequency to be very low. To do that he had to buffer incoming SPIF and estimate its average sample rate over a long-ish integration time. If the SPDIF jitter were low enough then the buffer delay could be adaptively reduced. Something like that.
@altor , it is just hard (impossible?) to find good measurement stimulus in "pure" DSD.
Besides, the distortions you talk about, ISI, noise floor etc are more prevalent in higher rates, so it'seasier to test at these rates. A discrete dac at dsd64 can be without ISI, but at higher rates become obvious. It comes crawling out of the woodwork as the difference between rise- and fall times of the logic in relation to the width of the pulse gets smaller. So both matter.
Typical cmos logic has a rise/fall time of a few nS, Analog CML logic has a lower voltage swing, but does that in 17 pS.
It was just my way of having fun, an object for study, I don't see this as a viable way of producing a commercial (or diy) DAC.
Anyway, I didn't buy HQPlayer for this, I got it to play all the music I have and not be limited to any source format. There is also a direct DSD mode, which also bypasses the volume control, if that's what you fancy;-)
Edit: I'd like to add to this that the previous efforts of likewise designs added to my sheer admiration to these dacs. For me it didn't start with the Philips Bitstream, or Sony's but when I build the DSC design by Jussi. Later came the raving experiences about Marcel's Tube dac, which I still have laying around but haven't build for myself and a good friend yet, (sorry and the other solid state design here on DIYaudio, which name slipped my mind right now.
It just seems more logical to me to have adjustable, advanced PC software and a dumb converter than mediocre software in a dumb converter, which is the other description of a standard DAC chip.
Besides, the distortions you talk about, ISI, noise floor etc are more prevalent in higher rates, so it'seasier to test at these rates. A discrete dac at dsd64 can be without ISI, but at higher rates become obvious. It comes crawling out of the woodwork as the difference between rise- and fall times of the logic in relation to the width of the pulse gets smaller. So both matter.
Typical cmos logic has a rise/fall time of a few nS, Analog CML logic has a lower voltage swing, but does that in 17 pS.
It was just my way of having fun, an object for study, I don't see this as a viable way of producing a commercial (or diy) DAC.
Anyway, I didn't buy HQPlayer for this, I got it to play all the music I have and not be limited to any source format. There is also a direct DSD mode, which also bypasses the volume control, if that's what you fancy;-)
Edit: I'd like to add to this that the previous efforts of likewise designs added to my sheer admiration to these dacs. For me it didn't start with the Philips Bitstream, or Sony's but when I build the DSC design by Jussi. Later came the raving experiences about Marcel's Tube dac, which I still have laying around but haven't build for myself and a good friend yet, (sorry
and the other solid state design here on DIYaudio, which name slipped my mind right now.It just seems more logical to me to have adjustable, advanced PC software and a dumb converter than mediocre software in a dumb converter, which is the other description of a standard DAC chip.
Last edited:
At the moment we have here for listening tests a prototype discrete resistor DSD-only FIR dac with RTZ up to DSD256. It uses SOA ultra low phase noise clocks at 22/24MHz, and includes a galvanically isolated FIFO buffer/reclocker. it sounds fabulous. Clearly better than AK4499 or AK4499EX. Soundstage and detail of 24/192 PCM converted to DSD256 sounds best. DSD64 (SACD rips) soundstage is more narrow, and sounds are a little less finely-detailed/less-natural sounding.
Last edited: