Yes.
If you look at the pictures I have posted a few times of my modded board the clock can be seen. You can probably even (barely) read the part number if you look at the picture at full resolution which you can do by clicking on it once to open it, then clicking again on the white X in the lower left corner when the mouse is in that area. I mention that because the full resolution pictures contain a lot of information but few people seem to notice just how much.
Anyway, you can see how much material I trimmed away and that I messed up a trace and fixed it. You have to look closely to see that though.
There is also a more recent picture that shows the new regulator mounted on the back of the board, you can get a hint from the file name if you hover over the picture most recently posted here: http://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-84.html#post5415832 it is the one that has 'osc 3-3v supply' in the name.
Last edited:
I tried upsampling with foobar running with wasapi and in combination with the minimum phase slow rolloff filter and did not like what I heard. The plugin was the standard one included in the latest foobar library. There was something fundamentally wrong with the higher midrange and the higher octave. It did not sound correct. Then I switched back to the standard FLAC at 16 bit 44.1 and it actually sounded more natural. Then last evening I switched to the slow rolloff linear phase filter and I thought it actually improved in that attacks and transients were sharper. The minimum phase filter sounded too sweet making transients sound softer and slower or more dulled. It does make it sound smoother however and depending on your system might make it better. The apodizing filter was a complete no go, same for the others.
Maybe this weekend I will put in a new IV section and see what happens. Mark, I see in your IV it looks like the balanced to single conversion was carried out on the existing OP amp. I am not up to drilling those precise holes for the feed through to the bottom so I am going to take the complete IV off board and retain the 3042 for the time being for supplying the 3.3v. The 330uF boost to the decoupling for AVCC appears to be fine for the time being.
Regarding upsampling, I prepared some high quality CD rip files at the original sample rate and with very high quality SRC at 24/96. You can download them here: DBX
In addition, please check that Windows or whatever OS is really sending out the correct sample rate for the file. If my 24/96 upsampled file sounds wrong when sent to the DAC (without any resampling by the OS) it means the DAC is still not working well enough to accurately play the file.
Regarding differential combining of my I/V stages, I reconfigured the components on the top of the board to make two differential amps using all 0.01% tolerance 10k thin film resistors I had in stock, and plugged an LME49720 DIP version into the socket. (EDIT:No original parts are still in use for that, not even the socket.) Hope that all seems more clear when looking at the pictures again at full resolution (click on the picture to open it then hover around the lower left corner until you see the white X, clicking on that blows it up to full size. You can also right click and download to view at full size).
The holes were not precision drilled at all. They were made with a small HSS engraving ball-cutter and a dremel tool. I just used the ball as a drill bit and hand drilled every one by eye. Easy.
In addition, please check that Windows or whatever OS is really sending out the correct sample rate for the file. If my 24/96 upsampled file sounds wrong when sent to the DAC (without any resampling by the OS) it means the DAC is still not working well enough to accurately play the file.
Regarding differential combining of my I/V stages, I reconfigured the components on the top of the board to make two differential amps using all 0.01% tolerance 10k thin film resistors I had in stock, and plugged an LME49720 DIP version into the socket. (EDIT:No original parts are still in use for that, not even the socket.) Hope that all seems more clear when looking at the pictures again at full resolution (click on the picture to open it then hover around the lower left corner until you see the white X, clicking on that blows it up to full size. You can also right click and download to view at full size).
The holes were not precision drilled at all. They were made with a small HSS engraving ball-cutter and a dremel tool. I just used the ball as a drill bit and hand drilled every one by eye. Easy.
Last edited:
You experience mirrors mine. Upsampling or adding dithered bits to make it 20-24 bits causes the sound to not sound natural no matter what algorithm I use.
Slow roll off (linear phase) sounded the best to me on raw 44/16. Min Phase was smooth like you described. I find a similar effect when I take a sabre DAC deep into current mode. My current setup is - 44/16 -> Linear Phase -> Transformer -> Class A Preamp -> Class D Amp
Cheers
Last edited:
Mark
QQ - Have you managed to get rid of the 2nd and 3rd Harmonic spike from the DAC? If so what specific mod caused it to disappear.
Regards
Looks like I HAVE to implement a new IV section soon. That might take some of the hardness of the sound ( maybe 3rd harmonics) out yet maintain most of what I want.
So also my current set up is
44/16 > Linear Phase > Stock output 9038Q2M> 744/811 Preamp with Jung Super Regs> Leach Double Barrel Amp 300W Dual Mono with Fully Regulated Power Supplies & Leach Low TIM 3 Fully Regulated Power Supplies (Sub) > B&W 802F + VMPS Larger Sub Array.
I haven't done much distortion measuring, still have some more work to do on the notch filter options. I can see that the harmonic distortion is way down but I didn't measure after each mod. I will have to go back to that when I get done with what I am working on now.
Regarding the test files I would agree there was something wrong with the 24/44 version. I put up a couple of other files, and I also listened to them all carefully using the DAC-3 and AHB2 amplifier. They all sound different in some ways. Please listen and let me know your listening impressions.
Link: DBX
Edit: Also, please remember to change the OS default sample rate to the same as whatever file will be playing. Thank you.
Edit 2: I would also say there is a problem if it sounds different with different reconstruction filters. If it does, that means you probably aren't hearing it the way it is supposed to sound, only the way that is preferred with your total reproduction system and listening preferences. Anyone who says DACs have inaudible distortion must not be listening very carefully, it sure seems like (digital reconstruction filters produce mostly linear distortion, group delay being a significant one. There are some specs on that in the datasheet but can't say what they are. Maybe a tiny bit of nonlinear distortion and noise produced too but probably hard to measure.)
Regarding the test files I would agree there was something wrong with the 24/44 version. I put up a couple of other files, and I also listened to them all carefully using the DAC-3 and AHB2 amplifier. They all sound different in some ways. Please listen and let me know your listening impressions.
Link: DBX
Edit: Also, please remember to change the OS default sample rate to the same as whatever file will be playing. Thank you.
Edit 2: I would also say there is a problem if it sounds different with different reconstruction filters. If it does, that means you probably aren't hearing it the way it is supposed to sound, only the way that is preferred with your total reproduction system and listening preferences. Anyone who says DACs have inaudible distortion must not be listening very carefully, it sure seems like (digital reconstruction filters produce mostly linear distortion, group delay being a significant one. There are some specs on that in the datasheet but can't say what they are. Maybe a tiny bit of nonlinear distortion and noise produced too but probably hard to measure.)
Last edited:
For filtering switchers you might want to consider a LC filter with MLCC ceramic caps rather than a single huge cap, reasons being, electrolytic caps are very bad at high frequency and also as you found out most switchers don't like huge bulk capacitances either. (They already have bulk capacitance at their output if they're designed properly with a LC filter)
Right, understood. The caps were put in to help with another issue, it was just a byproduct that the switcher was affected.
Also, agree that LC filters with appropriate HF caps should work better to filter out switcher noise that is conducted RFI/EMI. The radiated stuff is another issue. For low-cost off-the-shelf designs the EMI tends to go everywhere.
Anyway, the linear supply just didn't have those issues. Diode commutation noise was already taken care of. But I know well designed switchers can work very well. Some quasi-resonant designs produce little noise to begin with, for example.
To cut to the quick with the sample files, it turns out the sample rate conversion process also changed the levels a little. The effect of that is that 16-44 version sounds better in some ways, punchier for example. Turning the volume down on it makes it sound weak and muddled in comparison. Frequency response is the same in all of them throughout the audible range. Loudness differences can easily make that sound different.
The thing to listen for to start with are the cymbals and cymbal tails as they fade out. How much like a real cymbal do they sound. Not like bursts of noise or gritty/grainy. So long as no resampling by the OS, the higher bitrate files should sound more smooth and real when listening to that. Once you start to latch onto the cymbal SQ then the HF on everything else starts to sound more real, when the brain starts processing that information. That being said, there is no chance the cymbals can get near to real sounding at all until most or all of the mods have been implemented. Once the distortion is down, it is possible to hear pretty much exactly how well the recording was made. On the album used to make the samples there is some distortion, just a little though. However, its easy to think you have removed all your distortion contribution to the sound and then find out not so when you lower it some more.
I should probably just go back and renormalize all the files. Its a bit more processing but it should help get rid of the loudness effects.
The thing to listen for to start with are the cymbals and cymbal tails as they fade out. How much like a real cymbal do they sound. Not like bursts of noise or gritty/grainy. So long as no resampling by the OS, the higher bitrate files should sound more smooth and real when listening to that. Once you start to latch onto the cymbal SQ then the HF on everything else starts to sound more real, when the brain starts processing that information. That being said, there is no chance the cymbals can get near to real sounding at all until most or all of the mods have been implemented. Once the distortion is down, it is possible to hear pretty much exactly how well the recording was made. On the album used to make the samples there is some distortion, just a little though. However, its easy to think you have removed all your distortion contribution to the sound and then find out not so when you lower it some more.
I should probably just go back and renormalize all the files. Its a bit more processing but it should help get rid of the loudness effects.
Last edited:
Hi Markw4. Thank you for you effort. You did great job.
But this download test is not the right way. Please do that way. Use one original 44.1khz, 24/96 or 192kz file that we can unload it. Than send it trought dac and Adc and we will compare original file vs dac out file. I am interested how was changed vs original.
But this download test is not the right way. Please do that way. Use one original 44.1khz, 24/96 or 192kz file that we can unload it. Than send it trought dac and Adc and we will compare original file vs dac out file. I am interested how was changed vs original.
androa76, I have an idea I want to try first. It will take me awhile to get to it though. Everything takes time, and right now my time I have available for this is focused mainly on getting the DAC into master I2S mode. Progress continues, but there have been some issues to sort and I am working on another new one now.
I'll get back to this later though, when I can.
I'll get back to this later though, when I can.
Just a quick update. Got DAC into I2S master mode (128fs_mode) with all clock signals coming out. Sent those signals into the SRC chip operating in slave mode, and it is producing I2S data going back into the DAC.
Problem is the DAC is staying muted and not clear yet exactly why. I have tried some different settings, and sent off a question through the DAC chip distributor for clarification of some data sheet information (the data sheet is a bit on the minimal side, as data sheets go).
So, kind of waiting for a response and waiting to go at it again maybe tomorrow. Possibly sleeping on it for a night will help, sometimes it does actually.
In the meantime any questions are of course welcome. Happy to help if I can.
Problem is the DAC is staying muted and not clear yet exactly why. I have tried some different settings, and sent off a question through the DAC chip distributor for clarification of some data sheet information (the data sheet is a bit on the minimal side, as data sheets go).
So, kind of waiting for a response and waiting to go at it again maybe tomorrow. Possibly sleeping on it for a night will help, sometimes it does actually.
In the meantime any questions are of course welcome. Happy to help if I can.
Regarding upsampled test files, I am still thinking about that one too. People are very easily fooled or misled by volume level differences at little as .1dB. Sample rate conversion will cause come volume level change.
What I usually do is adjust the volume up and down for each file and try to ignore any perceived changes in frequency response, punch, sound stage, etc., that change with volume. Unfortunately, it takes some practice to learn how to hear the other things that don't perceptually change with volume level.
In addition, a reproduction system has to be good enough to not obscure those small differences. Many or most systems tend to be lacking in that respect, which is part of the reason I encourage adding a low-cost modified headphone amp to the project. But, the DAC will need to modded enough or otherwise good enough to reveal subtleties too.
What I could do with the sample rate test files to help keep people from getting distracted by perceptual effects of level differences could be to level match them to around .01dB.
However, I can't just normalize them digitally to the same level because the sample locations in the analog waveforms have changed due to resampling. Therefore they would have to be level matched after passing through reconstruction filters to find the true levels. I can do that, but it is a slower more laborious process. Maybe I will play around with it while waiting for a reply on the data sheet question.
Lest anyone think the subtleties I refer too above must be too small to bother with, once you listen to a good DAC for awhile and get used to more of the good sounding little details being there, you sure notice it when they are missing. Doesn't work the other way around though maybe because people don't suddenly learn to notice new things instantly unless the new things are big enough or particularly stick out in a way that tends to attract attention.
Besides long term exposure or long term listening, there are other ways to learn to hear the small details faster and better that are taught to recording and mastering engineers, but usually that is done when people are sitting in the same room and listening to the exact same source. I can tell you though that the training generally tends to start with listening very, very carefully to cymbal sounds. Why? I don't know, it just seems to work with humans. If you try different things with people its usually the cymbals they seem to be able to learn fastest with at first.
What I usually do is adjust the volume up and down for each file and try to ignore any perceived changes in frequency response, punch, sound stage, etc., that change with volume. Unfortunately, it takes some practice to learn how to hear the other things that don't perceptually change with volume level.
In addition, a reproduction system has to be good enough to not obscure those small differences. Many or most systems tend to be lacking in that respect, which is part of the reason I encourage adding a low-cost modified headphone amp to the project. But, the DAC will need to modded enough or otherwise good enough to reveal subtleties too.
What I could do with the sample rate test files to help keep people from getting distracted by perceptual effects of level differences could be to level match them to around .01dB.
However, I can't just normalize them digitally to the same level because the sample locations in the analog waveforms have changed due to resampling. Therefore they would have to be level matched after passing through reconstruction filters to find the true levels. I can do that, but it is a slower more laborious process. Maybe I will play around with it while waiting for a reply on the data sheet question.
Lest anyone think the subtleties I refer too above must be too small to bother with, once you listen to a good DAC for awhile and get used to more of the good sounding little details being there, you sure notice it when they are missing. Doesn't work the other way around though maybe because people don't suddenly learn to notice new things instantly unless the new things are big enough or particularly stick out in a way that tends to attract attention.
Besides long term exposure or long term listening, there are other ways to learn to hear the small details faster and better that are taught to recording and mastering engineers, but usually that is done when people are sitting in the same room and listening to the exact same source. I can tell you though that the training generally tends to start with listening very, very carefully to cymbal sounds. Why? I don't know, it just seems to work with humans. If you try different things with people its usually the cymbals they seem to be able to learn fastest with at first.
Last edited:
Markw4
One question I have to ask is this. That suboptimal (bad) recordings exist is a fact. Can you comment on whether the better the DAC, the more this class of recording becomes more obvious? to the point of atrocious and wanting to stop the playback? How does the DAC3 handle bad recordings? Is it more listenable or less? I need to know that because is getting this board closer to the DAC3 going to make more of my library "bad"?
One question I have to ask is this. That suboptimal (bad) recordings exist is a fact. Can you comment on whether the better the DAC, the more this class of recording becomes more obvious? to the point of atrocious and wanting to stop the playback? How does the DAC3 handle bad recordings? Is it more listenable or less? I need to know that because is getting this board closer to the DAC3 going to make more of my library "bad"?