Original PhiDAC reloaded...
I couldn't resist revisiting the original PhiDAC design and seeing how TI's ultra-cheap low-noise CMOS input opamps (OPA1678) might work in it, swapping out the CFB AD8017s.
The answer after a couple of days' listening is 'rather well'
There are now 3 DACs in parallel rather than just one as OPA1678's voltage noise is a bit higher. The LC filter now works at a much higher impedance to get lower noise - originally it was terminated with 33ohm, now that's risen to 160ohm. Still a lot lower than the optimum resistance for the opamps but subjectively the results are pleasing enough.
With OPA1678 we're no longer limited to 12V supply so I'm running a bit higher and that gives the potential for handling inter-sample overs. I've not examined how well those are handled yet...
I couldn't resist revisiting the original PhiDAC design and seeing how TI's ultra-cheap low-noise CMOS input opamps (OPA1678) might work in it, swapping out the CFB AD8017s.
The answer after a couple of days' listening is 'rather well'
There are now 3 DACs in parallel rather than just one as OPA1678's voltage noise is a bit higher. The LC filter now works at a much higher impedance to get lower noise - originally it was terminated with 33ohm, now that's risen to 160ohm. Still a lot lower than the optimum resistance for the opamps but subjectively the results are pleasing enough.With OPA1678 we're no longer limited to 12V supply so I'm running a bit higher and that gives the potential for handling inter-sample overs. I've not examined how well those are handled yet...
Attachments
Having mentioned that a 160ohm impedance filter was much lower than the optimum for the opamp (which would be around 1k) I was curious to hear if there was any improvement in going higher. So I threw 3 1mH SLF7045s in series for the filter, reduced the caps proportionately and that gave about 285ohm. My wife said that the violin now sounded clearer (we have been listening on YT to a release of Mendelssohn's piano quartets).
I ran the noise gain on LTSpice and this shows about 3dB reduction in noise at 5kHz from the higher filter termination resistance (285 vs 160). But the strange thing is - this noise is already way below the theoretical dither noise of a 16bit recording, according to my estimates at least 20dB below.
So what about higher impedance still? Watch this space.
I ran the noise gain on LTSpice and this shows about 3dB reduction in noise at 5kHz from the higher filter termination resistance (285 vs 160). But the strange thing is - this noise is already way below the theoretical dither noise of a 16bit recording, according to my estimates at least 20dB below.
So what about higher impedance still? Watch this space.
Your Opinion Is Requested
I invite all of you who have constructed one of Abraxalito's NOS DACs with high-order analog output filters to share your assessment of the subjective sound with us on the following thread. We are attempting to combine the best of NOS sound with the best of OS sound:
https://www.diyaudio.com/forums/digital-line-level/371931-makes-nos-sound-38.html#post6671617
We are especially interested to read your assessment of your ling/Phi DAC's sound compared to how it sounds with it's analog filter bypassed. In addition, versus the sound of typical OS digital interpolation-filter DACs. Please share you valuable assessments.
Thank you,
Ken Newton
I invite all of you who have constructed one of Abraxalito's NOS DACs with high-order analog output filters to share your assessment of the subjective sound with us on the following thread. We are attempting to combine the best of NOS sound with the best of OS sound:
https://www.diyaudio.com/forums/digital-line-level/371931-makes-nos-sound-38.html#post6671617
We are especially interested to read your assessment of your ling/Phi DAC's sound compared to how it sounds with it's analog filter bypassed. In addition, versus the sound of typical OS digital interpolation-filter DACs. Please share you valuable assessments.
Thank you,
Ken Newton
Hi Ken - I haven't so far heard of anyone completely bypassing the LC filter on PhiDAC hex or Phi Deca DAC. I also haven't tried it because it was an experiment I tried on the original Phi DAC - that's the single board one with AD8017s as the active elements. When I did this experiment, the sound lost all its lovely '3D' quality and became rather flat but I wonder if the stability of the AD8017 chip was affected by deleting the filter. So I'd not offer this in any way as a conclusive result of what happens without the filter. I recall Kevin (@stellarelephant) listened without the filter on the original lingDAC as removing it was quite simple.
I'm working on a son-of-PhiDAC at present so I'll do the filter deleting experiment on the new design sometime soon - its using OPA1678 which has a CMOS input stage so I'd expect it to be far more robust against high levels of ultrasonics.
I'm working on a son-of-PhiDAC at present so I'll do the filter deleting experiment on the new design sometime soon - its using OPA1678 which has a CMOS input stage so I'd expect it to be far more robust against high levels of ultrasonics.
Hi sorry
it was full, i didnt check but i delete some messages and made more room.Thanks
Thanks, Richard.
So that I'm clear about the DAC's configuration during your original NOS Phi DAC experiments. Without the LC reconstruction filter, essentially being stripped-down NOS, the sound was subjectively lacking in comparison to the sound of NOS with a steep LC reconstruction filter at the output? Do you recall the order of that LC filter?
Last edited:
I can say some fresh notes about NOS mode diskrete R2R dac. With passive LC filter at the dac net output sounds much better. I tried 5 order Shunt C element topology.
.
But I spot that a buffer with germanium transistors, Ken Newton topology was following the passive filter, made very very good impact on sound.
(The power amplifier was Quad II with high input impedance and short non high capacitive cables, so the buffer was lowering ouput but again only dac with passive filter was not so high output impedance). On the masurements I didnt find any significant changes with/without buffer. But w/wo filter was some reduction of HF noise. (Measured with computer and REW soft)
.
I spot one thing about phase shift:
In the simulations and theoreticaly the phase shift of passive LC filter should be much larger than one that I measure in real circuit?
Somebody have an ergument for this maybe?
.
Thanks
.
But I spot that a buffer with germanium transistors, Ken Newton topology was following the passive filter, made very very good impact on sound.
(The power amplifier was Quad II with high input impedance and short non high capacitive cables, so the buffer was lowering ouput but again only dac with passive filter was not so high output impedance). On the masurements I didnt find any significant changes with/without buffer. But w/wo filter was some reduction of HF noise. (Measured with computer and REW soft)
.
I spot one thing about phase shift:
In the simulations and theoreticaly the phase shift of passive LC filter should be much larger than one that I measure in real circuit?
Somebody have an ergument for this maybe?
.
Thanks
The experiment with PhiDAC was bypassing a 3rd order LC (CLC) so not a particularly steep one.
In relation to filters and OS vs NOS I have one other observation which I'm not sure I've set out before. You remarked on my slightly preferring 2XOS to NOS when listening on a PhiDAC hex (I think it was that model). That result was a reversal of a much earlier experiment I did with an unfiltered TDA1387 design when I preferred NOS to 2XOS. At that earlier time I noticed the 2XOS sound was more 'greyed out', had more of a metallic flavour than NOS.
I take the reversal of my preference as being the presence of the filter. TDA1387 being CMOS is rather glitchy (in contrast to TDA1543 which is bipolar) so I figure when run faster there is simply more RF noise on its outputs which affects downstream circuitry. I'm not talking here about ultrasonic images I am talking significantly higher frequencies into the MHz range. It doesn't need a very complex filter to deal with this RF - I played with a string of ferrite beads in series with the DAC output and this helped a lot in reducing the 'greyness' of the raw DAC output. That was after my 2XOS vs NOS experiment.
I take it that in all cases where ther was digital upsampling, it was performed via Foobar(SoX)?
Behind the time
Some 2 years ago Matt had kindly sent a PhiDAC kit to me. Last year I've soldered it - was my first kit from Richard and my first ever SMT solderthing. Then it had been laying around for some time and, after having received a recommended cheap USB board from Aliexpress I did some very experimental cabling and hoorraay - there was and is music. Didn't really expect this.
Contrary to that in case of sound, I expected very good sound and yes, after a few minutes for forming, sound was really great. Amazing as Richards mone amps, I've built in the meantime.
So this nice little DAC in my eyes deserves a proper chassis.
Thanks Richard, thanks Matt!
Cheers
Ernst
Some 2 years ago Matt had kindly sent a PhiDAC kit to me. Last year I've soldered it - was my first kit from Richard and my first ever SMT solderthing. Then it had been laying around for some time and, after having received a recommended cheap USB board from Aliexpress I did some very experimental cabling and hoorraay - there was and is music. Didn't really expect this.
Contrary to that in case of sound, I expected very good sound and yes, after a few minutes for forming, sound was really great. Amazing as Richards mone amps, I've built in the meantime.
So this nice little DAC in my eyes deserves a proper chassis.
Thanks Richard, thanks Matt!
Cheers
Ernst
Attachments
Last edited:
PhiDAC reloaded 3rd prototype schematic is attached. On this one I've raised the CLC impedance even further, to 910ohm. Also I managed to eliminate the I/V stage completely as on simulation the MFB filter has a constant enough Zin vs frequency to work as filter termination. So we're now down to just one dual opamp.
The inductors for this prototype will be P14s - that also allows me to experiment with higher still filter impedances as I can create Ls in the 10's of mH fairly easily with these cores whereas off-the-shelf inductor series (MSS1210 from Coilcraft for example) peg out at 10mH.
PCB layout hopefully will proceed this coming week
Yesterday I got into simulation of noise in LTSpice, I'd not done that before, only simulations of the noise gain of opamps. The difference is that noise gain simulation only takes account of opamp noise, not resistors too.
The results when resistors are included correlated with the listening results - higher impedance filters result in lower noise. But only in the higher frequencies (say above 1kHz or so) - running higher impedances gives less of a 'lift' to the HF noise. There is also a 'lift' to LF noise but I take it that's from the opamp.
This result has given me impetus to try a yet higher impedance. Also to get the best SNR it seems like experiments with higher output voltages than 2VRMS are worthwhile. There is however a downside to higher impedance - higher voltage swings at the output of the cascode MOSFET are bound to raise distortion as its output capacitance is non-linear. Running balanced can mitigate the largest distortion component (2nd harmonic) but I wonder how audible that is in practice. Which is why there's never any substitute for listening...
The results when resistors are included correlated with the listening results - higher impedance filters result in lower noise. But only in the higher frequencies (say above 1kHz or so) - running higher impedances gives less of a 'lift' to the HF noise. There is also a 'lift' to LF noise but I take it that's from the opamp.
This result has given me impetus to try a yet higher impedance. Also to get the best SNR it seems like experiments with higher output voltages than 2VRMS are worthwhile. There is however a downside to higher impedance - higher voltage swings at the output of the cascode MOSFET are bound to raise distortion as its output capacitance is non-linear. Running balanced can mitigate the largest distortion component (2nd harmonic) but I wonder how audible that is in practice. Which is why there's never any substitute for listening...
Ernst - about AD8017. I haven't found any direct replacement for that, AD815 is a bit better (at least no puny 12V rail voltage limit) but it has a much larger package. AD811 is the best sounding CFB amp I've tried but its not available as dual and runs very hot - so hot that the SOIC8 package has a limited upper rail voltage due to thermal constraints.
My affair with CFBs is pretty much over now because in practice they are a bit noisy - I'm interested in VFBs once again and OPA1678 seems to deliver a very good package of features at a very nice price.
My affair with CFBs is pretty much over now because in practice they are a bit noisy - I'm interested in VFBs once again and OPA1678 seems to deliver a very good package of features at a very nice price.