I never tried to use it, only heard about it. Sorry.
Jitter from clocks themselves is hard to measure properly, it takes some very expensive test equipment to measure accurately down to 1Hz offset. But, it is important for good audio. I would suggest only considering well documented clocks and or those with reputations for sounding good. Skip the rest for a dac.
Also, I would agree with you that buying cheap clocks on Taobao seems risky. You might ask our other Taobao-using, and dac-building forum member, Abraxalito, about it.
Jitter from clocks themselves is hard to measure properly, it takes some very expensive test equipment to measure accurately down to 1Hz offset. But, it is important for good audio. I would suggest only considering well documented clocks and or those with reputations for sounding good. Skip the rest for a dac.
Also, I would agree with you that buying cheap clocks on Taobao seems risky. You might ask our other Taobao-using, and dac-building forum member, Abraxalito, about it.
There is some discussion about measurement here: The Well Tempered Master Clock - Building a low phase noise/jitter crystal oscillator There is a new GB round here for the 5.x and 6.x crystals that should close soon.
Trying to find very low phase noise, 80 to 100MHz clocks is very difficult.
I believe even the Crystek clocks are all over the place when it comes to LF
phase noise.
If you can at all do it, better off running 9038 at 45 / 49MHz and use NDK
NZ2520SDA clocks. The 'A' ones are, apparently, more consistent WRT phase
noise and also slightly lower.
T
Next dac, I'll try something different. But, I still want it to be software configurable to be able play the highest sample rates.
Regarding NZ2520SDA, that's what Allo Katana uses, IIRC. My modded dac can sound just like it, better, or worse. Depends how I have the clock delay set up, and also AK4137. In terms of jitter affecting SQ, I think the modded dac is coming in about the same as Katana. That's what it sounds like. And, I still think I might be able to get existing jitter down a little lower if I were to keep working on it.
Regarding NZ2520SDA, that's what Allo Katana uses, IIRC. My modded dac can sound just like it, better, or worse. Depends how I have the clock delay set up, and also AK4137. In terms of jitter affecting SQ, I think the modded dac is coming in about the same as Katana. That's what it sounds like. And, I still think I might be able to get existing jitter down a little lower if I were to keep working on it.
Last edited:
Anyone here experience with this board?;
Lusya ES9038Q2M decoder board I2S input ES9038 asynchrone USB module kan worden gebruikt met Italiaanse interface in Lusya ES9038Q2M decoder board I2S input ES9038 asynchrone USB module kan worden gebruikt met Italiaanse interface van Versterker op AliExpress.com | Alibaba Groep
seems really interresting for the small footprint and upgraded op-amps...
Lusya ES9038Q2M decoder board I2S input ES9038 asynchrone USB module kan worden gebruikt met Italiaanse interface in Lusya ES9038Q2M decoder board I2S input ES9038 asynchrone USB module kan worden gebruikt met Italiaanse interface van Versterker op AliExpress.com | Alibaba Groep
seems really interresting for the small footprint and upgraded op-amps...
Yes. I got one with the upgraded opamps to check out. Think the amps are probably fakes. Sound quality is like junk, to me. Others may feel differently, of course. There are buttons for volume, but no way to change the filters.
Last edited:
Related to our sound quality concerns in this thread, I posted a review of Ling dac, a $5 dac kit, that may be of interest to some. It can sound pretty impressive with a couple of mods or upgrades: https://www.diyaudio.com/forums/dig...-rbcd-multibit-dac-design-30.html#post5712455
Hi Mark
Interesting read, thanks... and astonishing result for a few bucks... quite different from what we want to achieve here, but for sure intriguing and I guess good value for money.
A while back you said that to evaluate audio projects it would be advisable to have a good headphone amp. Hmm, makes indeed sense, and as a "group" it could make sense to have similar tools to share findings...
If I remember correctly a few months back you recommended this one:
1PCS LME49720+LME49600 headphone amplifier KIT | eBay
My understanding is it comes already fully assembled, bare the power supply and jacks.
Is that correct... and would that still be your recommendation?
If so, where do you put the volume potentiometer and what parts do you use to finalise the kit?
I browsed again through this thread and found a small post with some pix and hinting to additional caps recommendations / mods, but couldn't find an exhaustive list of all the mods to be carried out. Some looked quite complicated on the pix (shield, pin lifting??), others quite feasible (caps addition?)... would like more info...
Being completely honest, at this stage it is just for my curiosity, so please no need to waist your precious time and to put something together if it doesn't readily exist - I merely want to evaluate cost and feasability, in fact I don't need this part at all but perhaps for future project it could make sense to have the same tools as you... and to get started
Have a nice day
Claude
Interesting read, thanks... and astonishing result for a few bucks... quite different from what we want to achieve here, but for sure intriguing and I guess good value for money.
A while back you said that to evaluate audio projects it would be advisable to have a good headphone amp. Hmm, makes indeed sense, and as a "group" it could make sense to have similar tools to share findings...
If I remember correctly a few months back you recommended this one:
1PCS LME49720+LME49600 headphone amplifier KIT | eBay
My understanding is it comes already fully assembled, bare the power supply and jacks.
Is that correct... and would that still be your recommendation?
If so, where do you put the volume potentiometer and what parts do you use to finalise the kit?
I browsed again through this thread and found a small post with some pix and hinting to additional caps recommendations / mods, but couldn't find an exhaustive list of all the mods to be carried out. Some looked quite complicated on the pix (shield, pin lifting??), others quite feasible (caps addition?)... would like more info...
Being completely honest, at this stage it is just for my curiosity, so please no need to waist your precious time and to put something together if it doesn't readily exist - I merely want to evaluate cost and feasability, in fact I don't need this part at all but perhaps for future project it could make sense to have the same tools as you... and to get started
Have a nice day
Claude
@ClaudeG,
Those come as kits. I did recommend modding one to make it sound better than the stock design which isn't very good. Main problem is with the PCB layout, and the lack of any ground plane. I posted at least one pic of how mine ended up, and until now nobody has expressed any interest in modding one so I left it at that. What I did post should be at: ES9038Q2M Board - Page 63 - diyAudio
Those come as kits. I did recommend modding one to make it sound better than the stock design which isn't very good. Main problem is with the PCB layout, and the lack of any ground plane. I posted at least one pic of how mine ended up, and until now nobody has expressed any interest in modding one so I left it at that. What I did post should be at: ES9038Q2M Board - Page 63 - diyAudio
Just wonder how this concerns the DAC you modify? IMHO multibit NOS are much more forgiving distortions. And even then the sound is rather “interesting” than “true”. A DS one with such a level of distortions will sound like ....
But, yes the design of lingDac is pretty interesting indeed, and most importantly it is not an eBay mod.
Last edited:
That was more my point, and also that it could be an interesting modding project if people think this has been interesting. In addition, it could be an opportunity for the newbies to learn some more electronics.
In another vein, you may be interested in some posts of mine in Blowtorch today. Starting with: https://www.diyaudio.com/forums/the...wtorch-preamplifier-iii-1451.html#post5713303 and continues on the next page where there is some guessing about why reverb tails are low. As I am thinking about that, it occurs to me we may be on the cusp of understanding how a dac can measure flat FR but have 'weak' bass.
Hi Mark,
Here's a thought, if the clock frequancy is increaced, the jitter delta will be correspondingly smaller, the problem then will be dividing the clock without increacing the delta - ripple counter into a flipflop maybe? (sorry, showing my age, FPGA)
The next bit was - in the UK we have the MSF signal, it's a 60Khz time code signal locked to three atomic clocks (formally known as the Rugby clock) Time from NPL (MSF - Wikipedia). It has been used it to create accurate reference frequancys with a stability of parts per billion, maybe something similar could be implimented here; GPS signals?
Here's a thought, if the clock frequancy is increaced, the jitter delta will be correspondingly smaller, the problem then will be dividing the clock without increacing the delta - ripple counter into a flipflop maybe? (sorry, showing my age, FPGA)
The next bit was - in the UK we have the MSF signal, it's a 60Khz time code signal locked to three atomic clocks (formally known as the Rugby clock) Time from NPL (MSF - Wikipedia). It has been used it to create accurate reference frequancys with a stability of parts per billion, maybe something similar could be implimented here; GPS signals?
Hi Itsmee,
I think you need to be a little more clear about what you mean for me to understand what you have in mind. For example, when you suggest increased clock frequency, by what means is it to be increased? A higher frequency clock module, or maybe keep the original clock and use a frequency multiplier to double it?
Also, I'm not quite sure how you define 'jitter delta.' Some difference in jitter, but where does the difference come from? From a higher frequency clock vs the original clock? And how is 'jitter delta' to be measured?
So, you can see I am a bit confused, but if you want to brainstorm over your idea, I'm willing to try.
Regarding the MSF signal, atomic clocks have near-absolute clock accuracy over longer periods of time which is what they are good for. Maybe for keeping track of all the seconds that have transpired over several years very accurately. But, atomic clocks tend to be jittery over short periods of time like the period of a dac clock. That's because atomic decay is random and that makes it noisy. Over long periods of time the noise can be averaged out and that makes them super accurate over longer time periods.
For dacs, nothing beats a really good crystal clock, because crystals have very high Q which makes their resonance spectrum very clean and stable in the short term. The only thing absolute frequency would contribute would be super accurate reproduction of pitch. Maybe reproduced accurate to .0001 semi-tones. Unfortunately, the ADC that was used to digitize a recording probably had a clock that was a little more off in absolute terms, and sometimes recordings are at odd, non-standard pitches due to tape recorder vari-speed processing. Sometimes a song sounds better sped up a little, so they speed up the tape during mastering, but note pitch goes up with the speed too.
I think you need to be a little more clear about what you mean for me to understand what you have in mind. For example, when you suggest increased clock frequency, by what means is it to be increased? A higher frequency clock module, or maybe keep the original clock and use a frequency multiplier to double it?
Also, I'm not quite sure how you define 'jitter delta.' Some difference in jitter, but where does the difference come from? From a higher frequency clock vs the original clock? And how is 'jitter delta' to be measured?
So, you can see I am a bit confused, but if you want to brainstorm over your idea, I'm willing to try.
Regarding the MSF signal, atomic clocks have near-absolute clock accuracy over longer periods of time which is what they are good for. Maybe for keeping track of all the seconds that have transpired over several years very accurately. But, atomic clocks tend to be jittery over short periods of time like the period of a dac clock. That's because atomic decay is random and that makes it noisy. Over long periods of time the noise can be averaged out and that makes them super accurate over longer time periods.
For dacs, nothing beats a really good crystal clock, because crystals have very high Q which makes their resonance spectrum very clean and stable in the short term. The only thing absolute frequency would contribute would be super accurate reproduction of pitch. Maybe reproduced accurate to .0001 semi-tones. Unfortunately, the ADC that was used to digitize a recording probably had a clock that was a little more off in absolute terms, and sometimes recordings are at odd, non-standard pitches due to tape recorder vari-speed processing. Sometimes a song sounds better sped up a little, so they speed up the tape during mastering, but note pitch goes up with the speed too.
Last edited:
Introduction to FPGA FIR Filter Design
Today I am writing to provide a general update on my progress with looking into FPGA programming. First though, I picked up one of these: Arty S7 [Reference.Digilentinc] ...which may help make life easier. They are very affordable, and there is an example that one can load and run to use 24-bit I2S I/O to make a simple digital volume control.
However, just getting the software for that up and running on linux has taken me a couple of days, not in small part because I haven't used linux in awile, and there is a lot of typing out commands rather than the simple GUI interface of Windows.
So, that's the hardware side of things.
In terms of the nitty-gritty of low level FIR filter implementation at the logic gate level, there is a great resource available from Xilinx which they call something like 'Xilinx DSP University.'
Once the DSP University material is well understood, then there is the question of learning about the resources in a modern FPGA chip, and then how to program in VHDL to best use the FPGA hardware to implement a Filter. Fortunately, as is the case with Arduino, there are lots of examples and tutorials on the web.
Once HDL code to make it all happen is put together, then it must be compiled into low level hardware language, routing to FPGA pins for I/O specified, etc. That part is done in Xilinx ISE for Spartan 6, or in Vivado for Spartan 7. It is another programming environment that has to be learned.
One more thing, to know how many taps one needs to allow for in the FPGA filter design, one needs to figure out the parameters of the passband, transition band, stopband, required stop band attenuation, and allowable passband ripple. The are various software tools for that, including a very popular one in Matlab. They can determine how many coefficients will be needed and their values in order to meet the filter performance specification.
So, learning about and then stitching together all the above seems to approximately summarize what has to be done to get a basic FIR interpolation filter running over I2S, which is to say, at least a basic test filter to try with a physical dac.
Today I am writing to provide a general update on my progress with looking into FPGA programming. First though, I picked up one of these: Arty S7 [Reference.Digilentinc] ...which may help make life easier. They are very affordable, and there is an example that one can load and run to use 24-bit I2S I/O to make a simple digital volume control.
However, just getting the software for that up and running on linux has taken me a couple of days, not in small part because I haven't used linux in awile, and there is a lot of typing out commands rather than the simple GUI interface of Windows.
So, that's the hardware side of things.
In terms of the nitty-gritty of low level FIR filter implementation at the logic gate level, there is a great resource available from Xilinx which they call something like 'Xilinx DSP University.'
Once the DSP University material is well understood, then there is the question of learning about the resources in a modern FPGA chip, and then how to program in VHDL to best use the FPGA hardware to implement a Filter. Fortunately, as is the case with Arduino, there are lots of examples and tutorials on the web.
Once HDL code to make it all happen is put together, then it must be compiled into low level hardware language, routing to FPGA pins for I/O specified, etc. That part is done in Xilinx ISE for Spartan 6, or in Vivado for Spartan 7. It is another programming environment that has to be learned.
One more thing, to know how many taps one needs to allow for in the FPGA filter design, one needs to figure out the parameters of the passband, transition band, stopband, required stop band attenuation, and allowable passband ripple. The are various software tools for that, including a very popular one in Matlab. They can determine how many coefficients will be needed and their values in order to meet the filter performance specification.
So, learning about and then stitching together all the above seems to approximately summarize what has to be done to get a basic FIR interpolation filter running over I2S, which is to say, at least a basic test filter to try with a physical dac.
Last edited:
Whaou!!!!
Tenous you are, Sir, and courageous!
"Apart" from learning "all what is needed just to make it run", indeed a main obstacle will be to dertermine all the filter parameters. After all, the manufacturers spend for sure a great deal of resources on that since decades (Wadia...), it is popular to have very calculation intensive filters (Chord) enabled by today's chips, but then the question is: what is really good and really better than what is implemented first hand?
Good look Sir
Claude
Tenous you are, Sir, and courageous!
"Apart" from learning "all what is needed just to make it run", indeed a main obstacle will be to dertermine all the filter parameters. After all, the manufacturers spend for sure a great deal of resources on that since decades (Wadia...), it is popular to have very calculation intensive filters (Chord) enabled by today's chips, but then the question is: what is really good and really better than what is implemented first hand?
Good look Sir
Claude
The filter parameters are easy. Maybe I can explain later.
In relation to very long filters vs less long ones, there are always engineering trade offs in design. A first cut at a filter would need to be followed by listening tests to see how it sounds. Could be the tap coefficients need changing, or could be more taps are needed. If more taps, then a longer filter would have to be designed. But starting with an existing design, that should be fairly easy, at least until the FPGA starts to run low on ALU resources. At that point with some more design work, perhaps some additional taps could be implemented in the FPGA 'fabric.'
In any case, I don't think we are going for the chord sound. For one thing, it costs too much to do. I'm pretty sure they are doing 64-bit processing in multiple FPGA chips, and 64-bit FPGAs are not cheap. One chip can cost as much or more than a very high performance server class Intel CPU. That's part of the reason Chord Dave costs $12,000.+
The reason custom filters can be better than what is in the DAC is because there can be more taps, and because the filter can be designed to take advantage of pre-upsampling to a non-standard sample rate. That's basically what Benchmark DAC-3 does.
In relation to very long filters vs less long ones, there are always engineering trade offs in design. A first cut at a filter would need to be followed by listening tests to see how it sounds. Could be the tap coefficients need changing, or could be more taps are needed. If more taps, then a longer filter would have to be designed. But starting with an existing design, that should be fairly easy, at least until the FPGA starts to run low on ALU resources. At that point with some more design work, perhaps some additional taps could be implemented in the FPGA 'fabric.'
In any case, I don't think we are going for the chord sound. For one thing, it costs too much to do. I'm pretty sure they are doing 64-bit processing in multiple FPGA chips, and 64-bit FPGAs are not cheap. One chip can cost as much or more than a very high performance server class Intel CPU. That's part of the reason Chord Dave costs $12,000.+
The reason custom filters can be better than what is in the DAC is because there can be more taps, and because the filter can be designed to take advantage of pre-upsampling to a non-standard sample rate. That's basically what Benchmark DAC-3 does.
Last edited:
I wish I had the skills to assist you on such an endeavor, I feel useless
I'm sure with enough time and persistence you could do it, although you don't have the time to do it now.
To anyone reading who thinks it might be too hard to consider undertaking study of DSP and FPGAs, at least it doesn't require calculus like Laplace transforms do. Besides, sometimes things that initially appear hard and impenetrable can seem obvious once all the details of how they work have been learned.
Last edited: