Hi Ian,
Not sure if you are aware of our thread to mod cheap imported ES9038Q2M dacs? We have found that the Q2M seems to sound best with an AK4137 upsampling and converting all input to 11.2MHz DSD. Sounds way better than PCM. Don't know why, but think one of the Benchmark Media blog posts on their website said they were feeding some version of DAC-2 or DAC-3 with something like "DSD wide." They say it provided the most favorable oversampling ratio for minimum THD. We also found over here that reducing Q2M DPLL bandwidth for DSD to the minimum stable setting gives a noticeable but small additional improvement in sound quality. Upsampled DSD for, say, CD playback, sounds so much better I would strongly suggest giving it a try and see if you don't agree. We are doing it in hardware immediately before the Q2M to help preserve any improvement in jitter due to upsampling. If done too early in software, any jitter reduction can get lost by the time the signals get to the dac. Between that, a Crystek 100MHz clock on its own regulator, LME49720 AVCC buffers following an LTC6655 3.3v reference, and a carefully made LME49720 output stage, we find Q2M sound quality can be vastly improved.
If you would like to take a look we are over here: https://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-281.html#post5559608
Not sure if you are aware of our thread to mod cheap imported ES9038Q2M dacs? We have found that the Q2M seems to sound best with an AK4137 upsampling and converting all input to 11.2MHz DSD. Sounds way better than PCM. Don't know why, but think one of the Benchmark Media blog posts on their website said they were feeding some version of DAC-2 or DAC-3 with something like "DSD wide." They say it provided the most favorable oversampling ratio for minimum THD. We also found over here that reducing Q2M DPLL bandwidth for DSD to the minimum stable setting gives a noticeable but small additional improvement in sound quality. Upsampled DSD for, say, CD playback, sounds so much better I would strongly suggest giving it a try and see if you don't agree. We are doing it in hardware immediately before the Q2M to help preserve any improvement in jitter due to upsampling. If done too early in software, any jitter reduction can get lost by the time the signals get to the dac. Between that, a Crystek 100MHz clock on its own regulator, LME49720 AVCC buffers following an LTC6655 3.3v reference, and a carefully made LME49720 output stage, we find Q2M sound quality can be vastly improved.
If you would like to take a look we are over here: https://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-281.html#post5559608
Last edited:
ES9038Q2M dual mono DAC HAT
I did some ESS DAC HAT comparison test a couple of months ago.
https://www.diyaudio.com/forums/pc-...i2s-dac-hats-raspberry-pi-31.html#post5440848
Based on the actual listening test. I found the ES9038Q2M dynamic performance and quality was not as good as ES9028Q2M. The output impedance of ES9038Q2M is 774 ohm higher than ES9028Q2M's 403ohm. I suspect that is the main reason.
So I designed this ES9038Q2M dual mono DAC HAT. In this configuration, the internal impedance of ES9038Q2M will become 387 ohm, lower than ES9028Q2M.
I'll do a comparison test between signal and dual mono configuration ES9038Q2M DAC HAT very soon to see if there is any improvement.
ES9038Q2MpiDualHAT_2 by Ian, on Flickr
Good weekend.
Ian
I did some ESS DAC HAT comparison test a couple of months ago.
https://www.diyaudio.com/forums/pc-...i2s-dac-hats-raspberry-pi-31.html#post5440848
Based on the actual listening test. I found the ES9038Q2M dynamic performance and quality was not as good as ES9028Q2M. The output impedance of ES9038Q2M is 774 ohm higher than ES9028Q2M's 403ohm. I suspect that is the main reason.
So I designed this ES9038Q2M dual mono DAC HAT. In this configuration, the internal impedance of ES9038Q2M will become 387 ohm, lower than ES9028Q2M.
I'll do a comparison test between signal and dual mono configuration ES9038Q2M DAC HAT very soon to see if there is any improvement.
ES9038Q2MpiDualHAT_2 by Ian, on Flickr
Good weekend.
Ian
Hi Markw4,
I have to agree with you on the PCM to DSD conversion. I find it sounds much better as well. It think it would be possible to add the AK4137 on a HAT and upsample everything to native DSD256 instead of going the DoP route. Also, since RPi does not support native DSD through I2S it would be a good solution. Only thing is that the upsampling to DSD is done better with software like Daphile, Roon and HQplayer, although the AK is quite good solution too.
Do
Exactly. I have been experimenting with these dacs for several months. Use a Benchmark DAC-3 as a reference dac for SQ comparisons. Also, I have been a reviewer for Allo Katana so far. My modded dac has always sounded as good or better than Katana. In the last listening tests I was able to do with Katana 1.1, the modded dac definitely came closer to the sound of DAC-3 which so far as I know is the only Sabre dac on the Stereophile recommended equipment list. Also, state of the art in terms of measurements. Anyway, I wouldn't be surprised if my modded dac sound quality is one of the reasons they are still working on Katana. They know what I am doing and I sent their engineers an email to tell them any of their competitors could easily do what I am doing. In addition, since we don't need to use master mode to get good sound, it means the modded dac can be used as a general purpose dac, not just a RPi hat.
That's the approach taken by Ted Smith when he designed the PS Audio DirectStream DAC ($6,000!). He converts everything coming in (from all inputs) to DSD. The firmware automatically detects the incoming format and upsamples to by some huge amount to DSD. You can find YouTube videos of Ted describing why he does this and the nature of his single bit conversion stage with custom Hammond transformers. He also talks about the importance of minimizing jitter in single bit converters. All interesting stuff.
Cheers
Whit
Hello Ian,
Really enjoying reading about your experiences. A note on DAC output impedance - for a current output DAC like the ESS parts, the output impedance is supposed to be high. In fact the impedance of an ideal current source is infinitely high.
A possibility is that one of the DACs is producing more harmonic distortion than the other, giving the impression of greater dynamics. All sorts of other possible reasons too.
I have the same intuition. Don't know why internal resistance matters so long as I/V stage is configured appropriately to avoid opamp cross over distortion , if using opamps that is. Also, some opamps can be affected by local RF such as from DECT wireless phones. Lots of possibilities. Other things external to the dac chips could also be involved. Lot's of possible variables.
Also, don't know why dual mono dacs would be preferred. I always go back to DAC-3 which performs quite well. It uses one ES9028PRO and some LME49860 opamps. There is an SRC4392 also.
What is different about it is whatever is going on in the Spartan 6 chip, and the unique clocking scheme using 27MHz and 30MHz clocks that one probably has to buy a whole reel of to get one, perhaps Taitien brand. It does use some specialized filtering to improve performance. The built-in interpolation filters in most dac chips are a weak point. The modulator is most of what you need, and one may as well find it's sweet spot and run all audio through it with those settings.
Last edited:
I have the Terraberry DAC2, which has the AK4137 chip. That too sounds significantly better in PCM to DSD upsampling mode.
Hi Markw4,
Thank you so much for the suggestions.
Yes, I've already noticed that ES9038Q2M sounds very good when it's playing native DSD music. AK4137 is an interesting SRC chip which can up-sampling PCM into DSD256 on the fly. There are some AK4137 boards on eBay. I'll buy one to give it a try once I get time. If I'm happy with the result, I'd be more interested in figuring out how to run it at synchronized mode rather then ASRC.
As you mentioned the 100MHz Crystek clock, I think you are still using ESS DAC at async mode. I would highly recommend you trying the true sync mode, it really makes big difference.
Regarding to the sound quality of power supply, I found passive power supply such as ultra capacitor or LifePO4 battery are superior to any LDO or voltage reference. So I open a new R&D thread which is here:
Develop ultra capacitor power supply and LiFePO4 battery power supply
I believe your LME49720 AVCC buffers following an LTC6655 3.3v reference is also very good. But maybe you are interested in getting them compared to see what's the real difference. Both ultra capacitor and LifePO4 battery cell are very easy to be tested in a system.
BTW, I still have 6 or 7 of the ES9038Q2M prototype PCBs, and maybe also a couple of standard OP amplifier I/V stage PCBs available for sharing. Please let me know if you or any other community members need them for your/their project. PM me the shipping address, I can send them to you for free. I'd be very glad if they can be any help. For the well known reason, no schematic will be provided. But I don't think there is any problem to figure out.
Regards,
Ian
Attachments
Actually, I was thinking before about how to run an SRC4392 instead of an AK4317 in synchronized mode, but the same scheme should probably work the same for either chip. The way I was thinking of doing it would not require me to change the existing 100MHz clock, since I can divide it down and use it to clock the SRC. Although it would not be using a standard sample rate it should still work fine. I wrote a short explanation of the idea in the other thread in this post: ES9038Q2M Board - Page 230 - diyAudio
By that means, they could run in master mode (synchronous), or use ASRC, whichever is configured by register programming. Should be doable so long as there is still a separate clock that can be enabled for AK4137. Might use a NB3L553 to buffer the ES9038Q2 master clock output so I could disable that signal and enable another clock for the SRC, if wanting to run in ASRC mode. So, maybe nice to be able to switch it either way in software is all I was thinking. Certainly would make for quick comparisons or perhaps some ability to use synchronous or ASRC depending on what works best for a very high jitter incoming source.
Yes, I have been interested. The thing that puzzles me is that Allo Katana runs in synchronous mode and so far my modded dac using ASRC has sounded as good or better. I do have to tweak the Q2M DPLL bandwidth register to get the best sound quality though. Also, the Benchmark DAC-3 here always sounds better than my modded dac and Katana, and DAC-3 runs with ASRC enabled. It may be doing upsampling of some kind though. However it works it seems to show there is some way to make ASRC give excellent sound quality. Maybe the actual problem is figuring out how they do it.
Right, I would like to try ultra caps. I did try an LiFePO4 AA cell battery for AVCC once and it didn't sound realistic to me. It was pumping a little like a dynamics transient processor used in recording studios sometimes, but it didn't sound like I think the record was supposed to sound. Could be I needed to use a bigger battery so as to improve regulation some.
I also have a couple of low ESR .33F ultra caps to play around with, but they are probably not big enough to store enough charge like a battery. I would also worry about interconnection wire resistance. If I use LME49720 buffers the sense lead (inverting input) can be very close to the dac AVCC power pin so that might make lower impedance than wires running around. Don't know though, haven't tried it yet. I will try to get to it sometime. Also I will read your R&D thread. Thanks for the link.
Don't know about your kind offer yet regarding your spare parts/boards, but I think it might be interesting to try your IV stage board, for one possibility. See if it would work well with what we are doing. Please let me think about that a bit more get back by PM later.
Regards,
Mark
Last edited:
* Katana 1.0 had quite some (minor) (sound impacting) flaws onboard - and it's been recalled - that DAC IMO shouldn't be taken as a reference to anything for now -
Allo did quite some work to improve the situation - we'll see soon how things
are going on, on that one
And if you reference to e.g. Katana - you need to specify how you run it.
There are numerous options (HW and SW) , which all sound differently.
* A Benchmark DAC-3 is a different animal with a different DAC
* Sabre is not doing native DSD processing - just keep that in mind!
* Sync mode has always shown quite a step up on these Sabre DACs,
if you run async you could still be talking about a different subject.
The worse the DAC or it's implementation the more impact had these ASRCs
in the past. The better the DACs the more it became obvious that you better pass these devices by. Things might have changed.
Bottom line. All this crap is much too complex to draw any generic conclusions.
Allo did quite some work to improve the situation - we'll see soon how things
are going on, on that one
And if you reference to e.g. Katana - you need to specify how you run it.
There are numerous options (HW and SW) , which all sound differently.
* A Benchmark DAC-3 is a different animal with a different DAC
* Sabre is not doing native DSD processing - just keep that in mind!
* Sync mode has always shown quite a step up on these Sabre DACs,
if you run async you could still be talking about a different subject.
The worse the DAC or it's implementation the more impact had these ASRCs
in the past. The better the DACs the more it became obvious that you better pass these devices by. Things might have changed.
Bottom line. All this crap is much too complex to draw any generic conclusions.
@soundcheck
That sounds very reasonable. I'm fully agreed with you. It seems you did a lot of research into ESS DACs. Thank you so much for sharing.
To get best sound quality out of a given ESS DAC, I think the following three things are very important:
1. Power supply;
2. master clock;
3. I/V stage
Talking about the DAC chip itself, I normally tend to design this section as clean and pure as possible to keep the style of that particular DAC no change.
I'm not against having a DSP or SRC before DAC, but they should to be designed as optional and independent module to make it easy to switch back to native stream. Also, have to run them in sync mode and make sure do not degrade any to the clock quality.
Regards,
Ian
That sounds very reasonable. I'm fully agreed with you. It seems you did a lot of research into ESS DACs. Thank you so much for sharing.
To get best sound quality out of a given ESS DAC, I think the following three things are very important:
1. Power supply;
2. master clock;
3. I/V stage
Talking about the DAC chip itself, I normally tend to design this section as clean and pure as possible to keep the style of that particular DAC no change.
I'm not against having a DSP or SRC before DAC, but they should to be designed as optional and independent module to make it easy to switch back to native stream. Also, have to run them in sync mode and make sure do not degrade any to the clock quality.
Regards,
Ian
On my side I have not done DSD tests on ES90XX DACs but only on AK4490 (Has DSD Direct) and Signalyst DSCv2 (DSD only DAC) which are totally different in their architectures.
My last experience with ES9018 was BII with NTD1 I/V stage, had it for years and really liked it.
Your products are always top quality so it will be interesting to know how it stacks against those, at least for me personally.
All the best!
Do
My last experience with ES9018 was BII with NTD1 I/V stage, had it for years and really liked it.
Your products are always top quality so it will be interesting to know how it stacks against those, at least for me personally.
All the best!
Do
Yes, know what you mean. Things have improved somewhat after ES9018. But the dacs can still produce excess harmonic distortion and are very resolving of it, so it is easy to hear what it is. The best that can be done short of external interpolation filtering, would be to the point where the dac sounds okay across the room with the volume turned up a little, but careful listening close up still reveals the same problems, but perhaps at a lower level. Only with DAC-3 have I heard a Sabre design that almost completely removes that last bit of harmonic distortion to leave a warm, neutral, but very detailed sound. Across the room with the volume turned up a little, the best Sabre dacs can sound close to each other, but not with careful, near field listening. The last fix that separates DAC-3 from the others doesn't seem to have to do with clocking or ASRC, it seems to replace the interpolation filters inside the Sabre chip with an external filter with much lower distortion. It sounds good, as I think cdsgames agreed. On the other hand, the best I have been able to do so far with ES9038Q2M is somewhere between Katana 1.1 (with 3 clean power supplies, and the filter cdsgames recommended which sounded best to me also) and DAC-3. At that level my Q2M dac is pretty good, but not good enough for the most discerning listeners. Maybe good enough for a back up dac, or something like that. I will say it takes a lot of work and care to get it that good, even though it would be nice if the sound quality from Q2M could be a little better than I know how to get it (at least so far
).
Last edited:
But the new version of Katana is not available yet - so you should be getting the improved version soon. Why comment when Allo has said they are working on an improved design?
Nope. Not at all.
* I just clocked my ES9023 HAT synchronously with the Allo Kali. And removed the single onboard clock. That pretty well showed me what's happening if you swap sync/async.
* When testing DSD features/functionalities for the PicorePlayer project I had a closer a look at DSD and how different DACs are handling it. That showed that Sabre converts DSD back to some kind of PCM.
If then somebody says that DSD sounds much better. The first question I'd ask myself
would be how a Sabre would sound with upsampled PCM!?!?
Not as good as with DSD. AK4137 can do it either way.
By the way, there was no preconceived notion upsampled DSD would sound best, actually the opposite. However, listening tests proved otherwise with DSD definitely sounding better.
In addition, the test you described with ES9023 does not show how well it can work with ES9028 or 38 if done properly. It can sound just as good with ASRC on. Eventually, you will probably get a chance to see for yourself.
Last edited: