They are not hot, I can hold my finger on them indefinitely. Not hot enough to hurt them. Temperature-wise, they are about like my first modded dac. You can see the AVCC opamp by itself, and the two I/V opamps close together in the pic below. The AVCC opamp ran cool, ,the I/V and differential summing opamp on the topside of the board all ran much warmer. Not so at all for 2nd modded dac.
Anyway, I'm thinking of trying an experiment to separate the output stage board from the dac board and maybe do a little bit of RF filtering. Would be interesting to see what happens.
Attachments
Not sure if you are using the Rpi wifi. FWIW I keep it turned off. The Pi is on and ethernet cable to a wireless router which provides the wireless gateway to control Volumio.
I don't know anything about either the Kanata or the Dual ESS dac. I have a very different set up. I do have very clean upper register SQ generated from the FIFOpi. I will say I did not like the upper frequency until I removed the 3 caps on the bottom of the XO adapter and installed an FCA cap. But I think a minor change in sound compared to what you describe.
I like the Benchmark DAC-3 faraday cage around the power supply.
I did take some effort to provide shields using brass plate or double sided PCB stock. The DAC, its regulators and the tube I/V output are all in their own little copper box. The PI/FIFO etc are all on the other side of the copper plane. The ufl cables poke through a grommet to provide PCM signal in and RCA lets analogue out. The power transformers and filtering stages are all in their little metal rooms. Perhaps the effort was worthwhile.
The benchmark pic gives me an idea to experiment with ferrite beads on the signal wire. Might attenuate some of the cell phone action. I have never been a fan of ferrite clamps on power cables as they kill dynamics, but perhaps a small one on the signal cable would kill some noise and leave the signal in tact.
I don't know anything about either the Kanata or the Dual ESS dac. I have a very different set up. I do have very clean upper register SQ generated from the FIFOpi. I will say I did not like the upper frequency until I removed the 3 caps on the bottom of the XO adapter and installed an FCA cap. But I think a minor change in sound compared to what you describe.
I like the Benchmark DAC-3 faraday cage around the power supply.
I did take some effort to provide shields using brass plate or double sided PCB stock. The DAC, its regulators and the tube I/V output are all in their own little copper box. The PI/FIFO etc are all on the other side of the copper plane. The ufl cables poke through a grommet to provide PCM signal in and RCA lets analogue out. The power transformers and filtering stages are all in their little metal rooms. Perhaps the effort was worthwhile.
The benchmark pic gives me an idea to experiment with ferrite beads on the signal wire. Might attenuate some of the cell phone action. I have never been a fan of ferrite clamps on power cables as they kill dynamics, but perhaps a small one on the signal cable would kill some noise and leave the signal in tact.
There are a number of different types of ferrite material that attenuate in different frequency ranges. Most likely some types will not cause audible problems.
Just put the output stage on an extender cable. Now its starting to sound more like clock problems will be the next thing to look at.
So, you think power quality is the clock issue? Maybe too much lead inductance with the plug-in clock boards?
Also, notice that Vref on the output stage is not set to AVCC/2. Rather it is AVCC/3. Wonder if that might be to reduce the big DC offset currents though the very low resistances around the opamps. Looks like the circuit was optimized more for low noise than for low distortion. May have to come back to that later.
Attachments
Last edited:
I think all improvement to clock power yields results. Ian suggests battery power bypassing the onboard LDO. I suspect it is very good... even better if you add a supercap. I use very good linear plus supercap dedicated rail to the OX and delete the ceramic caps on the adapter. Result is very clean upper registers and everything natural.
sq225917, sorry to hear you're not getting the SQ you expected from Ian's GB gear. From what you've posted, it seems like you're doing the right stuff... LiFePO4 supply with FiFoPi LDO's bypassed, sync mode (regular sync or true sync?),
AND your overall setup sounds good.
Just to get a baseline, I spent some time comparing some options today. One was to verify what @wlowes said, upgrading the clocks. This was with my standard RPi 2B powered by a modified K&K Audio low voltage supply and Ian's LiFePO4 supply powering the rest of the DAC stack.
To check the upgrade I got from the Crystek CCHD-957s, I put the 'gimme' clocks back in. This was on the trusty ES9028Q2M prototype I got from Ian about a year ago that I use as my basic test bed, with the unobtanium Onetics transformers on the outputs, which are the best I've tried so far. Yes the 'gimme' clocks were a significant downgrade and the highs were much trashier. So I second Walter's comments, this is an area to explore.
Next to get a baseline on some opamp options, I first put back the 'gimme' opamps into the IVStd I/V stage. This was on the ES9038Q2M Dual Mono board I have here with Crysteks in the FiFoPi. Similar reaction to the clocks... yuck!
Then I ran through the OPA1642, OPA1612, Sparkos discretes and LKS discretes in fairly short succession:
- I REALLY like the mids of the OPA1642, but the bass is less defined than the other options and the highs a bit hashy, though not really badly. Still I could live with them, especially for those mids.
- The LKS discretes sounded different. I need to spend more time with them to characterize their sound. BUT at this point I would not put them in my favorites category. That may change with more burn-in or other changes to the IV stage and DAC. Bummed about that.
- The OPA1612s were the best of the monolythics so far. Detailed, extended, articulate. Highs were a little more pronounced than the Sparkos.
- AND the Sparkos are my favorite. The 'class-A'-ness comes through (just under 30mA for a dual versus just under 10mA for an OPA1612) in the separation between instruments and the microdynamics. Highs were very detailed and articulate, but less pronounced compared to the OPA1612.
@Supersurfer, I'm puzzled by what you experienced with the Sparkos. They have been trouble-free for me. I checked the temps today and got about 120F, which compares well to the 105F or so I got on the OPA1612 (and I'd expect that difference based on the specified quiescent currents).
Also, @markw4, I'll be curious to hear what you find if/when you dig into what you suspect is RF issues. I'm not seeing, hearing, or experiencing them here. I HAVE increased the I/V stages feedback filter capacitor from 270pF stock to 1500pF currently, but I didn't feel any significant difference from the opamp heating with the stock Cs there (though I never checked with a thermometer). Possibly something going on with your opamp power supply? What are you using there? AND what temps have you measured on various OPA1612 setups?
There's been some comparisons of Ian's GB gear suggested to the Benchmark DAC3 & @markw4's developmental ES9038Q2M DAC. I can't comment on either of those having never heard either. BUT I can compare to my Allo Katana, which has a pretty high-test power supply setup and the few mods detailed in the 'Getting the best out of Allo.com's new Katana...' thread here: Getting the best out of Allo.com's new Katana DAC... . I would not say it is the best sounding Katana out there, but I suspect it is in the top few (@John Luckins likely has us ALL beat on this!... see post 183 in the 'Getting the best out of Allo.com's new Katana... thread).
In my setups and to my ears and preferences, my top Katana is performing at a very good level. My 2 Twisted Pear Buffalo-IIIPro setups need some additional tweaking and refinements to match the Katanas. Same with my 2 Soekris DAM setups, though I suspect I won't get them up to the same level no matter what I do. I'm sad about that... there still is something about an R2R DAC's sound, though the Katana with the 'Sound Quality' board edges in that direction.
At this point, my better Ian GB setups are in the same strata as the Katana, but overall a bit below it. Those best setups are:
- RPi -> FiFoPi (Crystek CCHD-957s & LDO's bypassed) -> Ian DAC (both of the GB ones and the ES9028Q2M prototype).
- RPi powered with my modified K&K Audio low voltage supply, the Ian GB stack by his LiFePO4 supply.
- IVStd with either the OPA1612 or Sparkos discrete OR a transformer output stage with either the LL1544a (good) or Onetics (better). Also note that the prototype IV861 units I have from Ian show strong promise so far.
- I/V & transformer output stage high frequency filtering cut-off lowered to roughly the same range as the Katana single-ended output filters (see my post a few pages back about too little high-frequency filtering causing some ringing in my ears). I'm also experimenting with the Rasmussen alternative pre-IV filtering as used in the Katana 'Sound Quality' Opamp board with promising results.
Areas I feel are worth exploring to bring out higher-performance:
- Enabling the harmonic distortion compensation feature on the DAC chips as suggested by @Markw4
- The power supply tweaks used on the Allo Katana... 20uF & 1uF SMD film caps on the AVCC lines along with supercaps (after the LP5907 LDOs they use for those rails).
- @markw4's polyprop cap bank on the I/V stage +-supplies.
- Even more extreme power tweaks such as Ultracap supplies and the local Ultracaps @wlowes posted about recently.
- Bypass caps on the clocks
- Other upgrade clocks (@blueneogeo, where are you getting the Accusilicon clocks?)
- Separate power rails for the clocks (As @wlowes did, 3 total for the FiFoPi) and for the 3 rails on Ian's DAC boards (as @jacklee did). I'm setting up a 2nd LiFePO4 board to try this and also have 3 UCPi I can set to 3.3V to power the DAC board rails.
What other thoughts?
Greg in Mississippi
AND your overall setup sounds good.
Just to get a baseline, I spent some time comparing some options today. One was to verify what @wlowes said, upgrading the clocks. This was with my standard RPi 2B powered by a modified K&K Audio low voltage supply and Ian's LiFePO4 supply powering the rest of the DAC stack.
To check the upgrade I got from the Crystek CCHD-957s, I put the 'gimme' clocks back in. This was on the trusty ES9028Q2M prototype I got from Ian about a year ago that I use as my basic test bed, with the unobtanium Onetics transformers on the outputs, which are the best I've tried so far. Yes the 'gimme' clocks were a significant downgrade and the highs were much trashier. So I second Walter's comments, this is an area to explore.
Next to get a baseline on some opamp options, I first put back the 'gimme' opamps into the IVStd I/V stage. This was on the ES9038Q2M Dual Mono board I have here with Crysteks in the FiFoPi. Similar reaction to the clocks... yuck!
Then I ran through the OPA1642, OPA1612, Sparkos discretes and LKS discretes in fairly short succession:
- I REALLY like the mids of the OPA1642, but the bass is less defined than the other options and the highs a bit hashy, though not really badly. Still I could live with them, especially for those mids.
- The LKS discretes sounded different. I need to spend more time with them to characterize their sound. BUT at this point I would not put them in my favorites category. That may change with more burn-in or other changes to the IV stage and DAC. Bummed about that.
- The OPA1612s were the best of the monolythics so far. Detailed, extended, articulate. Highs were a little more pronounced than the Sparkos.
- AND the Sparkos are my favorite. The 'class-A'-ness comes through (just under 30mA for a dual versus just under 10mA for an OPA1612) in the separation between instruments and the microdynamics. Highs were very detailed and articulate, but less pronounced compared to the OPA1612.
@Supersurfer, I'm puzzled by what you experienced with the Sparkos. They have been trouble-free for me. I checked the temps today and got about 120F, which compares well to the 105F or so I got on the OPA1612 (and I'd expect that difference based on the specified quiescent currents).
Also, @markw4, I'll be curious to hear what you find if/when you dig into what you suspect is RF issues. I'm not seeing, hearing, or experiencing them here. I HAVE increased the I/V stages feedback filter capacitor from 270pF stock to 1500pF currently, but I didn't feel any significant difference from the opamp heating with the stock Cs there (though I never checked with a thermometer). Possibly something going on with your opamp power supply? What are you using there? AND what temps have you measured on various OPA1612 setups?
There's been some comparisons of Ian's GB gear suggested to the Benchmark DAC3 & @markw4's developmental ES9038Q2M DAC. I can't comment on either of those having never heard either. BUT I can compare to my Allo Katana, which has a pretty high-test power supply setup and the few mods detailed in the 'Getting the best out of Allo.com's new Katana...' thread here: Getting the best out of Allo.com's new Katana DAC... . I would not say it is the best sounding Katana out there, but I suspect it is in the top few (@John Luckins likely has us ALL beat on this!... see post 183 in the 'Getting the best out of Allo.com's new Katana... thread).
In my setups and to my ears and preferences, my top Katana is performing at a very good level. My 2 Twisted Pear Buffalo-IIIPro setups need some additional tweaking and refinements to match the Katanas. Same with my 2 Soekris DAM setups, though I suspect I won't get them up to the same level no matter what I do. I'm sad about that... there still is something about an R2R DAC's sound, though the Katana with the 'Sound Quality' board edges in that direction.
At this point, my better Ian GB setups are in the same strata as the Katana, but overall a bit below it. Those best setups are:
- RPi -> FiFoPi (Crystek CCHD-957s & LDO's bypassed) -> Ian DAC (both of the GB ones and the ES9028Q2M prototype).
- RPi powered with my modified K&K Audio low voltage supply, the Ian GB stack by his LiFePO4 supply.
- IVStd with either the OPA1612 or Sparkos discrete OR a transformer output stage with either the LL1544a (good) or Onetics (better). Also note that the prototype IV861 units I have from Ian show strong promise so far.
- I/V & transformer output stage high frequency filtering cut-off lowered to roughly the same range as the Katana single-ended output filters (see my post a few pages back about too little high-frequency filtering causing some ringing in my ears). I'm also experimenting with the Rasmussen alternative pre-IV filtering as used in the Katana 'Sound Quality' Opamp board with promising results.
Areas I feel are worth exploring to bring out higher-performance:
- Enabling the harmonic distortion compensation feature on the DAC chips as suggested by @Markw4
- The power supply tweaks used on the Allo Katana... 20uF & 1uF SMD film caps on the AVCC lines along with supercaps (after the LP5907 LDOs they use for those rails).
- @markw4's polyprop cap bank on the I/V stage +-supplies.
- Even more extreme power tweaks such as Ultracap supplies and the local Ultracaps @wlowes posted about recently.
- Bypass caps on the clocks
- Other upgrade clocks (@blueneogeo, where are you getting the Accusilicon clocks?)
- Separate power rails for the clocks (As @wlowes did, 3 total for the FiFoPi) and for the 3 rails on Ian's DAC boards (as @jacklee did). I'm setting up a 2nd LiFePO4 board to try this and also have 3 UCPi I can set to 3.3V to power the DAC board rails.
What other thoughts?
Greg in Mississippi
Last edited:
Hi,
could you fill me in on reg sync & true sync.
By reg sync did you mean async?
Thanks
Just a quick note on the NS2 clock mod - installing two NS2 clocks with the optocoupler described earlier in the thread unfortunately do not provide sufficient isolation between the clocks.
Apparently as the two clocks' output are tie together the signal from the active clock will still be affected by the inactive clock via the coaxial cable connecting to the NS2 board. This sometimes lead to one clock not locking so could not possibly be good for sound either.
I'm investigating using a RF micro relay to physically isolate the clocks when inactive. Will report back when I have a chance to implement that.
Apparently as the two clocks' output are tie together the signal from the active clock will still be affected by the inactive clock via the coaxial cable connecting to the NS2 board. This sometimes lead to one clock not locking so could not possibly be good for sound either.
I'm investigating using a RF micro relay to physically isolate the clocks when inactive. Will report back when I have a chance to implement that.
@soundcheck,
I prefer the minimum-phase slow and have since the early Katana. Same on Ian's ES9038Q2M DM board.
It was @Jonathan P who liked the Apodizing.
Also @nicoch58, who linked to this article on filters:
H I F I D U I N O: WM8741 Digital Filters
I forgot to say thanks for that, was a good re-read. I saw it back when it was published, but know so little about digital at the time that it didn't really take much away. I wish he was still posting... a LOT of what I learned about digital came from his site!
After @nicoch58's suggestion and re-reading that article, I tried apodizing again. Still a no for me.
@ichiban, with the early ESS DAC chips (ES9008 & ES9018), you could set the DAC to run in sync mode by feeding a clock that was from the same family as the source material and the DPLL would not do the typical asynchronous upsampling... I heard the term 'free-wheeling' used, as I remember.
You can still do that with the current generation ESS DAC chips (ES9028PRO, ES9038PRO, ES9028Q2M, ES9038Q2M). That is what you get with one of Ian's DACs when you connect the MCLK output from the FiFoPi and turn off the local clock using the small switch on the DAC board. Once you have done this, you can also use Ian's ESS Controller to set the PCM and DSD DPLL Bandwidth to the lowest level 1.
BUT these new chips also offer a more advanced version of sync mode, true sync mode. This is where you actually disable the DPLL entirely. Ian's controller also allows you to do this. BUT there are limitations:
- PCM only, no DSD
- MCLK should be 128FSR for best results. For 44.1 source, that is a clock of 5.6488! While you can use the more common 22.5792 or 44.1584 clocks and it will work, you are not getting the full benefit of full sync mode at that point.
Still, even with the wrong clock frequencies, I personally prefer full sync to sync and sync to the typical ASRC mode.
Greg in Mississippi
I prefer the minimum-phase slow and have since the early Katana. Same on Ian's ES9038Q2M DM board.
It was @Jonathan P who liked the Apodizing.
Also @nicoch58, who linked to this article on filters:
H I F I D U I N O: WM8741 Digital Filters
I forgot to say thanks for that, was a good re-read. I saw it back when it was published, but know so little about digital at the time that it didn't really take much away. I wish he was still posting... a LOT of what I learned about digital came from his site!
After @nicoch58's suggestion and re-reading that article, I tried apodizing again. Still a no for me.
@ichiban, with the early ESS DAC chips (ES9008 & ES9018), you could set the DAC to run in sync mode by feeding a clock that was from the same family as the source material and the DPLL would not do the typical asynchronous upsampling... I heard the term 'free-wheeling' used, as I remember.
You can still do that with the current generation ESS DAC chips (ES9028PRO, ES9038PRO, ES9028Q2M, ES9038Q2M). That is what you get with one of Ian's DACs when you connect the MCLK output from the FiFoPi and turn off the local clock using the small switch on the DAC board. Once you have done this, you can also use Ian's ESS Controller to set the PCM and DSD DPLL Bandwidth to the lowest level 1.
BUT these new chips also offer a more advanced version of sync mode, true sync mode. This is where you actually disable the DPLL entirely. Ian's controller also allows you to do this. BUT there are limitations:
- PCM only, no DSD
- MCLK should be 128FSR for best results. For 44.1 source, that is a clock of 5.6488! While you can use the more common 22.5792 or 44.1584 clocks and it will work, you are not getting the full benefit of full sync mode at that point.
Still, even with the wrong clock frequencies, I personally prefer full sync to sync and sync to the typical ASRC mode.
Greg in Mississippi
It has puzzled me also. The sparkos were burning in during the night with a new smps I got. The previous evening all sounded ok but the next morning there was a lot of noise coming from the speakers when I turned on the amps. I measured the output of the sparkos and one of them was putting out around 3,5v dc. I took them out and put the stock opamps back, the noise was still there but when I changed to another smps it was quiet again.
The only thing I can think of is that the noise from the smps has caused the damage. I am now running the set for a week with another ps and no problems.
Next week I will receive a new sparkos and will try again. I hope the sound is as good as you say!
Regards,
Regarding ES9038Q2M digital interpolation filter choices, what I found along the way in my dac modding journey was that which filter sounds best depends on what is still wrong with the dac. Once every single problem affecting SQ that can be fixed is fixed, then Apodizing (the ESS default) should probably sound best. To put it a bit differently, to my current way of thinking for ES9038Q2M, if Apodizing doesn't sound best (for PCM), it means keep looking for things that can be fixed with the dac and the reproduction system. Of course, my current way of thinking is subject to updating over time in light of new information. Also of course, there is the issue of personal opinion as to what sounds 'best.' That can be complicated, so I will take the easy way out and stop here.
My initial impression is that the extender cable may have reduced some output stage distortion. Picture in earlier post: https://www.diyaudio.com/forums/pc-...essions-tweaks-mods-hints-33.html#post5788176
...If anyone else decides to try it, please post your listening impressions.
However, cymbals still sound too crunchy. Probably time to try some clock swapping next.
Also, the output stage schematic shows C9, C10 as optional for additional filtering. May try installing those soon.
I remain concerned about the very low value resistors used around the opamps (for low noise) in terms of possibly affecting opamp distortion performance. Replacing all the resistors and filter caps would be more work than I want to try at this point, but at a minimum adding C9, C10 (NPO or C0G) seems like a good idea.
Also, probably worth trying to send the dac I2S from a good USB to I2S card rather than from RPI. That would probably clarify how much hash can be attributed to RPi.
...If anyone else decides to try it, please post your listening impressions.
However, cymbals still sound too crunchy. Probably time to try some clock swapping next.
Also, the output stage schematic shows C9, C10 as optional for additional filtering. May try installing those soon.
I remain concerned about the very low value resistors used around the opamps (for low noise) in terms of possibly affecting opamp distortion performance. Replacing all the resistors and filter caps would be more work than I want to try at this point, but at a minimum adding C9, C10 (NPO or C0G) seems like a good idea.
Also, probably worth trying to send the dac I2S from a good USB to I2S card rather than from RPI. That would probably clarify how much hash can be attributed to RPi.
Today, I tried Volumio on my setup which is Rpi3B+, FifoPi, ES9038q2m with default ps boards and standard I/V convertor. I couldn't check sacd on Volumio since it doesn't support it yet, but dsd files plays very good. No noise for dsd files like in sacd Archphile distribution. Clocks are NDK sda, 44 and 49 MHz. For the moment dac chips receive and decode DoP with output dac model R-PI.
Maybe I did something wrong during the installation of sacd option in archphile, although I followed the manual.
Maybe I did something wrong during the installation of sacd option in archphile, although I followed the manual.
Forgot to mention that folding over and taping the ribbon cable as shown in the picture put the dac +- outputs for each channel next to each other. The capacitance between them may have attenuated some HF junk, which is why I taped the ribbon cable in the position I did. Probably not as good as twisting the +- outputs for each channel together and maybe running the pair next to a ground plane, but my impression was that it helped some. I did not stop to investigate more carefully at this point, however.
I recently had a 8*TDA1387 HAT from Audiophonics on my desk. I was building a HAT tower (the SW part) for a friend of mine.
Just for fun I've been using the FifoPi with the 1387.
I never expected what was happening.
Suddenly instruments sounded like real instruments on my Adam speakers.
I'll soon get my own device delivered. If further tests turn out to confirm my earlier experience, I have to question all these Sabre resp. Delta/Sigma DACs in general.
Yep. Many of us have been in that area with 1543/1541 or R2R projects.
Klaus, I'm keen to hear more about that.
Cheers, Ernst
Just tried the Crystek 957 clocks instead of NDK. Sounds closer to right to me, although still some remaining sound quality problems.
Cleaning up remaining distortion should help a lot. Now it sounds more like IMD rather than jitter, as the bigger problem. The sound seems a bit muffled and jumbled together in the midrange which sounds like a lot of HD/IMD contributing to that. But adding artificial brightness and pseudo-clarity from NDK jitter is not the fix, it just adds another problem.
Harmonic distortion compensation should probably be the next thing to look at. If an FFT shows much in the way of higher than 3rd harmonic, then some hardware troubleshooting would be necessary.
To summarize configuration so far:
*DVCC on ADM7150
*VCCA on ADM7150
*AVCC on NewClassD UWB2 (with 5.7v regulator powering dac board)
*Output stage +-15v using parallel film caps from each rail to ground (10+10+33+33+33, all values uf, Wima mks4)
*Output stage on ribbon extender to help attenuate RF leakage into output stage
*Crystek 957 clocks, rather than NDK.
Note: Still on my to-do list:
*Test ADM7150 with a 33-ohm 1/2-watt resistor to ground to see if that makes it more usable for AVCC (or otherwise improves SQ elsewhere).
*Try adding C9, C10 to output stage board (NPO or C0G).
Also, checked I2S signals on clean 40-pin GPIO bus. Don't know why termination resistors would be needed. Pic of BCLK below.
Cleaning up remaining distortion should help a lot. Now it sounds more like IMD rather than jitter, as the bigger problem. The sound seems a bit muffled and jumbled together in the midrange which sounds like a lot of HD/IMD contributing to that. But adding artificial brightness and pseudo-clarity from NDK jitter is not the fix, it just adds another problem.
Harmonic distortion compensation should probably be the next thing to look at. If an FFT shows much in the way of higher than 3rd harmonic, then some hardware troubleshooting would be necessary.
To summarize configuration so far:
*DVCC on ADM7150
*VCCA on ADM7150
*AVCC on NewClassD UWB2 (with 5.7v regulator powering dac board)
*Output stage +-15v using parallel film caps from each rail to ground (10+10+33+33+33, all values uf, Wima mks4)
*Output stage on ribbon extender to help attenuate RF leakage into output stage
*Crystek 957 clocks, rather than NDK.
Note: Still on my to-do list:
*Test ADM7150 with a 33-ohm 1/2-watt resistor to ground to see if that makes it more usable for AVCC (or otherwise improves SQ elsewhere).
*Try adding C9, C10 to output stage board (NPO or C0G).
Also, checked I2S signals on clean 40-pin GPIO bus. Don't know why termination resistors would be needed. Pic of BCLK below.
Attachments
Last edited: