They do. They have to adjust the offset pots for all the discrete opamps too. IIRC, cdsgames said it takes about an hour to manually adjust each Katana dac. Lot of work and has to add to what it costs to make one. But then, Benchmark adjusts harmonic compensation for each dac too.
Thd Compensation
Hi,
I've done the thd compensation on my other board( with dip switch), without i2c pin lifting ( like i2c multimaster mode) data pushed from my esp32 board to the es9038q2m ) , same results on the thd plot (-3dbfs) in green without compensation , in white with compensation ( f800 value in register).
As you can see, upper harmonics are little bit higher but still very low.
Hi,
I've done the thd compensation on my other board( with dip switch), without i2c pin lifting ( like i2c multimaster mode) data pushed from my esp32 board to the es9038q2m ) , same results on the thd plot (-3dbfs) in green without compensation , in white with compensation ( f800 value in register).
As you can see, upper harmonics are little bit higher but still very low.
Attachments
One thing that concerns me a bit about sticking with voltage mode is that all these differential output dacs have some common mode distortion that can only be removed with a carefully balanced differential output stage.
One type of distortion from not doing it the proper way with I/V and differential output is a lot of IMD down at -25dB FS. However, for someone who wants wants a quick and easy improvement, if maybe not the best, the HD compensation register trick with voltage output mode could seem very attractive.
However, I'm not sure how exactly how the trick affects that particular IMD, I only know of it if using I/V without proper differential CM distortion removal.
As usual for me, I kind of suspect ESS would not foolishly recommend doing it the I/V and differential way if there were a simple voltage mode way that could work as well. Of course, it could be nice to believe they are stupid and we are clever, but I'm not so sure if that's how it really turns out.
Still, kudos to occip for this tip (who sent me the original PM on this). I know people will want to try it. I hope somebody will go to the trouble to compare it with I/V and differential and tell us about the results.
One type of distortion from not doing it the proper way with I/V and differential output is a lot of IMD down at -25dB FS. However, for someone who wants wants a quick and easy improvement, if maybe not the best, the HD compensation register trick with voltage output mode could seem very attractive.
However, I'm not sure how exactly how the trick affects that particular IMD, I only know of it if using I/V without proper differential CM distortion removal.
As usual for me, I kind of suspect ESS would not foolishly recommend doing it the I/V and differential way if there were a simple voltage mode way that could work as well. Of course, it could be nice to believe they are stupid and we are clever, but I'm not so sure if that's how it really turns out.
Still, kudos to occip for this tip (who sent me the original PM on this). I know people will want to try it. I hope somebody will go to the trouble to compare it with I/V and differential and tell us about the results.
Last edited:
Can anyone explain how by changing the computations inside the DAC, one can compensate for the THD+N of the external circuitry? There is no feedback and how does the DAC know what the source of the THD+N is coming from and thus compensate for it?
Is it performing an inverse distortion mechanism itself?
On the surface it makes no sense to someone who knows not how these darn chips work.
Is it performing an inverse distortion mechanism itself?
On the surface it makes no sense to someone who knows not how these darn chips work.
Actually, ESS says you can use harmonic distortion compensation to help linearize the external circuitry by introducing an inverse distortion.
Think if it like this, distortion may arise because the transfer function of an opamp is not a perfectly straight line. What if the transfer function is shaped pretty close to a 2nd order polynomial curve, or a 3rd order polynomial? Well, it turns out a second order polynomial curve can't make more than 2nd order harmonic distortion. Similar for 3rd order, it can't make higher than 3rd harmonic distortion. The curve is like a bend in a line that should be straight. What if we first run the signal through a line bent exactly the opposite way so that when the signal has then passed through the opamp too, it has been bent twice in ways that exactly cancel out. It is as though it has only passed through two perfectly straight lines. However, since we cannot shape the bends perfectly as opposites and because the lines may be bent a bit differently for different frequencies or for some other reasons we are unable to perfectly cancel the bending out. That's kind of the idea in a nutshell.
In this particular case it is not really the opamp voltage amplifier that is causing the non-linearity. It is the dac when the outputs are not shorted to an offset-pseudo-ground Reference Voltage (Vref) and the output signal taken to be the current that flows into that short. Nonetheless, we are introducing a non-linearity somewhere upstream in the dac to cancel out the bending that takes place in the dac output stage when the output is connected to a high-z load and signal is taken to be the voltage appearing at that node.
As to how the dac knows how much to bend the line, it is a fixed open loop setting. There is one for 2nd harmonic bending and one setting for 3rd harmonic. Someone has to pick a setting which thereafter remains constant until it is manually changed or the dac reboots.
Think if it like this, distortion may arise because the transfer function of an opamp is not a perfectly straight line. What if the transfer function is shaped pretty close to a 2nd order polynomial curve, or a 3rd order polynomial? Well, it turns out a second order polynomial curve can't make more than 2nd order harmonic distortion. Similar for 3rd order, it can't make higher than 3rd harmonic distortion. The curve is like a bend in a line that should be straight. What if we first run the signal through a line bent exactly the opposite way so that when the signal has then passed through the opamp too, it has been bent twice in ways that exactly cancel out. It is as though it has only passed through two perfectly straight lines. However, since we cannot shape the bends perfectly as opposites and because the lines may be bent a bit differently for different frequencies or for some other reasons we are unable to perfectly cancel the bending out. That's kind of the idea in a nutshell.
In this particular case it is not really the opamp voltage amplifier that is causing the non-linearity. It is the dac when the outputs are not shorted to an offset-pseudo-ground Reference Voltage (Vref) and the output signal taken to be the current that flows into that short. Nonetheless, we are introducing a non-linearity somewhere upstream in the dac to cancel out the bending that takes place in the dac output stage when the output is connected to a high-z load and signal is taken to be the voltage appearing at that node.
As to how the dac knows how much to bend the line, it is a fixed open loop setting. There is one for 2nd harmonic bending and one setting for 3rd harmonic. Someone has to pick a setting which thereafter remains constant until it is manually changed or the dac reboots.
Last edited:
Well, I was hoping to have something to say here maybe even today about the new updated firmware version of Allo Katana dac, another take on what is possible with ES9038Q2M. Unfortunately, it was mysteriously held up in customs for more than 24 hours, then just as mysteriously released. The Fedex tracking said:
"Clearance delay - Import
Itemized breakdown of product composition required."
Never saw that before. Anyway, looks like it maybe just started moving again. Hopefully, I could have some preliminary information to share about the sound tomorrow if it keeps moving along from here on out.
"Clearance delay - Import
Itemized breakdown of product composition required."
Never saw that before. Anyway, looks like it maybe just started moving again. Hopefully, I could have some preliminary information to share about the sound tomorrow if it keeps moving along from here on out.
Last edited:
That explains it, it is the same as calibrating the gamma setting on a display monitor. However in that case a series of measurements are taken and then then an inverse curve is sent to the monitor thereby the resultant output is nearly linear or as close as possible.
I would suspect this is spelled out in the NDA. In the case of Benchmark, with a stable build I could see a practical way of implementation. In a DIY thing, I'm sure it is a lot more luck involved.
Oh, I don't know. Those things can seem pretty sparse at times.
To change the subject a bit, I wonder if anyone is thinking about using the trick occip figured out to hopefully make the stock voltage output stage sound better? To get the best results that way still requires substantially improving AVCC power quality, but most people could probably feel comfortable doing that. Where I don't know if people will want to go would be the digital and software business of writing to the dac registers.
If one were to go there, it would certainly open up a number of other possibilities for improving sound quality.
As it happened, I think occip maybe found a Q2M data sheet out in the open where a dac manufacturer posted it for customer support purposes. I can't and won't say how to find it, but I can say that ESS seems pretty good about giving out data sheets if one asks for an NDA form to fill out. However, it might depend a lot on the distributor representing ESS in region where you are located. I found a distributor who would talk to me but a couple of others didn't seem to want to bother. I suppose contacting ESS directly to ask for help might work if a distributor did not respond to a reasonable number of repeated polite requests for assistance.
Anyway, anyone thinking of having a go at it, turning down the voltage mode output stage distortion? Any questions at all about doing it?
Last edited:
http://www.kelm.ftn.uns.ac.rs/literatura/mpi/pdf/Op Amp Applications Handbook.pdf
Page 230- suggestion for an improved 3.3 v reference.
Page 234, Finally a document that gives a good rundown of the post DAC circuit. The THD compensation is possibly a value that integrates the non perfect nature of both the DAC and outside circuitry and is possibly a trimming value.
Page 230- suggestion for an improved 3.3 v reference.
Page 234, Finally a document that gives a good rundown of the post DAC circuit. The THD compensation is possibly a value that integrates the non perfect nature of both the DAC and outside circuitry and is possibly a trimming value.
Thought I would let people know the new Katana arrived today. It is hooked it up to three power supplies as recommended and playing music continuously to burn in as suggested by cdsgames (although I don't know if I expect much change, but no reason not to try it).
Unfortunately, the old one was returned and at this point its hard to remember exactly how it sounded compared to this one.
However, initial impression is the new one sounds very good and I think noticeably better. I think it's going to be tougher to rank 2nd and 3rd place this time around and I'm not prepared to take a shot at it yet. Besides, I have re-calibrate levels between the dacs and sync up the music playing first so the comparison can be apples to apples.
Choice of filters makes a lot of difference, of course. When the sound seems smoother and less distorted, it also seems less punchy and dynamic. Changing the filter can turn that around so the sound is a little grainier and slightly more distorted, but very nicely punchy and dynamic. This is where I am thinking a carefully designed external interpolation filter with more taps could maybe provide less compromise to achieve smooth and non-grainy, yet at the same time still punchy and dynamic. By, the way these thoughts on filtering apply to all the filters in these systems, those in the dac chips and those in any SRC chips, and not just for Katana, also for my modded dac and for AK4137 SRC. All the filters have a sound, and none of the ones that are configurable ever seem to have the perfect setting. Only DAC-3 with no choice of filters doesn't seem to have that compromise where one part of the sound quality has to traded off for another part of the sound quality. Since the trade offs are apparent when switching interpolation/reconstruction filters, that presumably is where one would look for a solution. Probably part of why DAC-3 uses a fast ASIC chip for filtering rather than using the filters in the dac chip. They seem to talk like it is mostly for avoiding intersample overs, but maybe reasonable to suspect there is probably a little more to the story.
Anyway to get back to Katana, I think cdsgames suggested the apodizing filter sounds best for Katana on small(ish) speakers. For Katana it does seem to give the best balance in overall sound quality, at least as I hear it on my system here, all IMHO of course.
Unfortunately, the old one was returned and at this point its hard to remember exactly how it sounded compared to this one.
However, initial impression is the new one sounds very good and I think noticeably better. I think it's going to be tougher to rank 2nd and 3rd place this time around and I'm not prepared to take a shot at it yet. Besides, I have re-calibrate levels between the dacs and sync up the music playing first so the comparison can be apples to apples.
Choice of filters makes a lot of difference, of course. When the sound seems smoother and less distorted, it also seems less punchy and dynamic. Changing the filter can turn that around so the sound is a little grainier and slightly more distorted, but very nicely punchy and dynamic. This is where I am thinking a carefully designed external interpolation filter with more taps could maybe provide less compromise to achieve smooth and non-grainy, yet at the same time still punchy and dynamic. By, the way these thoughts on filtering apply to all the filters in these systems, those in the dac chips and those in any SRC chips, and not just for Katana, also for my modded dac and for AK4137 SRC. All the filters have a sound, and none of the ones that are configurable ever seem to have the perfect setting. Only DAC-3 with no choice of filters doesn't seem to have that compromise where one part of the sound quality has to traded off for another part of the sound quality. Since the trade offs are apparent when switching interpolation/reconstruction filters, that presumably is where one would look for a solution. Probably part of why DAC-3 uses a fast ASIC chip for filtering rather than using the filters in the dac chip. They seem to talk like it is mostly for avoiding intersample overs, but maybe reasonable to suspect there is probably a little more to the story.
Anyway to get back to Katana, I think cdsgames suggested the apodizing filter sounds best for Katana on small(ish) speakers. For Katana it does seem to give the best balance in overall sound quality, at least as I hear it on my system here, all IMHO of course.
Last edited:
I am looking forward for your Katana review. Try also linear slow. It reveals mycrodynamic and stage is more 3d. Apodizing bring artiffical dynamic. Sound is more *simple*. Maybe better first impressionat on small mid range speakers. Thank you. I develop board in analog domain...and compare it to 5k dacs and more like you Mark . They can reveal what es9038q2m can or not. But not headphones. My ears are not satisfied with headphones so room setup is my review place.
True , but slow linear are usually used for hi res files. Also you need an excellent speaker to have an improvement with slow linear.
For most bookshelves we recommend apodizing.
With regard to filter choices, there are couple of different approaches we use here. They are (1) which one seems to sound 'best,' and (2) which one sounds closest to DAC-3. Because there are trade offs in SQ with filter choice, SQ is not necessarily made only better with a different filter. It can be better SQ in some respects and worse in others. Regarding how one should be chosen, my preference is to have a selection bias towards sounding as good as possible while still sounding as close as possible to DAC-3. I am assuming of course that DAC-3 is more likely to sound like what is really on the CD, which may not be where another dac sounds best. So, when I went through all the Katana filters today I was noticing how they compare to other dac and SRC filters and how they compare to DAC-3. What I found was that neither Katana or my modded dac could be made to sound exactly like DAC-3 filtering, so presumably both are a least a little inaccurate in some (different) ways. Overall for Katana, apodizing is one of the better sounding Katana filters and probably the one that comes closest to DAC-3. It is just that DAC-3 does that sound with more perfection than Katana does. That's okay though, I want to listen to the sound of Katana with it configured to be as accurate as possible. My choice has nothing to do with speaker size, or distance from speakers, etc.
Looking forward to it. If possible try Katana on bigger speakers as well. I find thats when Katana really shines .
The whole SQ is not dependet just from a dac board. The whole setup must be in a certain level that can reveal quality of each component. At yours system one filter is best at my and 2 others another filter. Nothing wrong. I just pointed out 2 best ones... My refference that i know that is one of the best dacs is Total dac...in terms of neutrality. I don*t have it so i don*t promote my eqiupment. If maybe Katana will be in my near I will test it for sure.
Last edited:
Hey, stand back folks. I'm in the imaging business. Everything you are listening to is going through a filter and thus transfer function. In imaging The real thing is when you see it with your own eyes directly with perfect vision and nothing that can produce a transfer function in between. No matter how good a camera sensor is, or the lens is, there is a transfer function. The DAC3 has its own transfer function like it or not. In fact Benchamark inserts one in there to create the desired output. To date it may appear to have the least distortion but when something comes up that produces a transfer function that is different, does that mean to say it is necessarily inferior to the DAC3? Chasing the DAC3 towards better sound might not be the correct thing to do if making an excellent DAC is the goal, but trying to get the same sound as a DAC3 is a different goal. Indeed what is necessary is to compare some other good dacs against the whole roster. Now suppose after a while a DAC4 comes out that sounds like a DAC that "somebody" created today that did not sound like a DAC3 and was dismissed because it did not sound like a DAC3. That is the real risk and the folly of audiophilia at times.
What I have found is that there is no doubt that when the ESS Dacs start to be implented properly, there is indeed a whole lot more information coming out in the higher frequencies as well as LOW. Now when you think about it what CDSgames is saying can make sense. If you increase the balance is the spectrum towards the higher frequencies, you're going to tilt the perception and there is overload in your senses. Small bookshelves might be able to deliver the envelope up there but it needs to be balanced back out with better capabilities down there as well. So maybe the LF information that the small bookshelf is being asked to reproduce is causing distortion and hence the fatiguing sound.
One way of determining what is happen is to put both the large speakers and small bookshelves side by side and feed each one with say 4th order HP filter at 100 HZ and compare the sound of both. That might indicate what is going on about speaker selection.
If this indeed is the situation then maybe a good DAC for jill with full range speakers will not be the same for jack who can only use bookshelves in his small apartment?
What I have found is that there is no doubt that when the ESS Dacs start to be implented properly, there is indeed a whole lot more information coming out in the higher frequencies as well as LOW. Now when you think about it what CDSgames is saying can make sense. If you increase the balance is the spectrum towards the higher frequencies, you're going to tilt the perception and there is overload in your senses. Small bookshelves might be able to deliver the envelope up there but it needs to be balanced back out with better capabilities down there as well. So maybe the LF information that the small bookshelf is being asked to reproduce is causing distortion and hence the fatiguing sound.
One way of determining what is happen is to put both the large speakers and small bookshelves side by side and feed each one with say 4th order HP filter at 100 HZ and compare the sound of both. That might indicate what is going on about speaker selection.
If this indeed is the situation then maybe a good DAC for jill with full range speakers will not be the same for jack who can only use bookshelves in his small apartment?
Last edited:
Mike, we know. I am already taking all that (and more) into account. In fact, we want and need an interpolation filter. The more ideal it is, the more accurate reproduction can be. The more taps we can compute in a digital filter, the closer we can come to ideal filter characteristics in terms of amplitude and phase. The number of taps and the number of bits in a dac chip is more limited that it is in an ASIC. That is why DAC-3 can be made more accurate and correct than the standard ESS dac filters.
Last edited:
@andora76, Regarding Total Dac, I understand it is a very musical dac and a pleasure to listen to. It is also and R2R ladder type dac. I can't find any comprehensive measurements of one though. My best guess would be that one of their dac products might sound great, yet might not be the best choice for a mastering room.
Really good R2R dacs are known for sounding good and not having some of the undesirable quirks of delta-sigma designs such as artifacts resulting from state space variable transitions, but they also are extremely hard to make accurate down to 21-bits where some of the very best delta-sigma dacs are now. While Total Dac may sound wonderful, I sure would like to see some measurements to help us get a better idea about the absolute accuracy part of its performance.
For now, I think it will probably be best to stick with DAC-3 as my reference test instrument for listening comparisons. Having said that, I would still very much like to hear Total Dac sometime if I ever get the chance.
Really good R2R dacs are known for sounding good and not having some of the undesirable quirks of delta-sigma designs such as artifacts resulting from state space variable transitions, but they also are extremely hard to make accurate down to 21-bits where some of the very best delta-sigma dacs are now. While Total Dac may sound wonderful, I sure would like to see some measurements to help us get a better idea about the absolute accuracy part of its performance.
For now, I think it will probably be best to stick with DAC-3 as my reference test instrument for listening comparisons. Having said that, I would still very much like to hear Total Dac sometime if I ever get the chance.
Last edited: