Zen Mod recommends film motor run capacitors for decoupling power amp PSU's, if ZM recommends something, it's worth looking at, and they're much cheaper than 'branded' film caps.
Last edited:
Motor capacitors tend to come in higher voltages and therefore may be physical larger size per uf. The Wima film caps I used are 63v types which is the smallest physical size I could find. I also chose stacked film over other film cap construction types. It is better for low ESR and low ESL. Don't know as much about the motor run caps the other guys may have in mind. That being said, I do have large electric car motor inverter 'DC Link' film caps here to try sometime. They are reportedly favored by some audiophiles in Japan, but they cost about $1/uf so not so cheap as the Wima caps.
It may also be that there are voltage regulator types that don't need film caps nearly as much. Jung Super Regulators come to mind. I am planning to try that sometime too, but too many things to do to get them all done quickly. For now, the film caps we have work quite well for their intended purpose.
It may also be that there are voltage regulator types that don't need film caps nearly as much. Jung Super Regulators come to mind. I am planning to try that sometime too, but too many things to do to get them all done quickly. For now, the film caps we have work quite well for their intended purpose.
I figured out that you need to disconnect the ''VCC10'' power supply internally to use external IS2 with a Topping D50, otherwise it will only play at fixed low volume with the DAC turned off. hopefully that info will help others in similar situation, because the fact that this actually fixed the problem is a miracle... what could possibly need 10V that would stop the DAC chip accepting external IS2? nothing else was effected by disconnecting this power source.
Anyway does anyone know what the resistors on DACs IS2 input are for? they are 22 ohm, ive seen different values used in another dac with sabre chip, not sure if the value is dependant on the DAC or USB interface. I see this board doesnt have any
Anyway does anyone know what the resistors on DACs IS2 input are for? they are 22 ohm, ive seen different values used in another dac with sabre chip, not sure if the value is dependant on the DAC or USB interface. I see this board doesnt have any
Not sure what Topping VCC10 stuff refers to exactly.
Regarding 22 ohm resistors on I2S inputs, they are probably there to help damp out any ringing caused by long wires, etc.
the topping takes a 5V input and converts it to various voltages internally for different parts of the DAC. There are multiple 3.3V regulators labeled 'DAC' and 'USB' on silkscreen, self explanatory, but there was one step up converter marked 'VCC10' and whatever it was powering was stopping the DAC chip accepting an external IS2 signal and disconnecting it didnt effect the functionality of the DAC in anyway. just found it strange.
Do you use these resistors with your board? the IS2 lines are only a couple cm long so they might not be necessary here, if there is any point to removing/bypassing them is another question
The Topping thing does sound strange. You could probably try to ring out where the output of VCC10 goes to. If you have a DVM with a continuity beeper mode, then one probe can be held or otherwise attached at the VCC10 output and the other probe touched around at various points in the circuitry. It can go pretty quick and maybe give you some way to track down what it might have to do with I2S not working.
Regarding the 22 ohm resistors, there is no exact best value. If they are too big they may prevent the dac from working, but in the sizes usually found they were selected by the designer as a good fit for expected I2S conditions.
The Chinese dac boards that be typically mod don't have resistors there and when I have looked with a scope I don't see ringing aberrations in the waveforms that look like they need fixing. Various distributed and lumped capacitance and inductance on a PCB might cause more ringing than with the wiring we usually are using with our modded boards.
Anyway, probably best to leave resistors like that as you find them unless you have some reason to believe there is a problem. Most likely there wouldn't be as they are pretty low value.
Last edited:
Just an update. I'm still going through and taking note of the various setting combinations between ES9038Q2M and AK4137 when Q2M is I2S/DSD clocking source (master mode). Initial impression is that jitter is not as low this way as the lowest I have ever had it. Although it sounds very good as is, it would still be nice to know how to make it the best I can figure out. The reason I think jitter is not lower than it is (all based on subjective listening impressions only) is because in master mode the highest frequency clock being sent to AK4137 by Q2M is BCLK, not MCLK. That means AK4137 most likely is running an internal PLL to synthesize its own MCLK, and PLLs like that typically can't compete with a good crystal oscillator in terms of jitter. Anyway, a couple of approaches come to mind. One is to install a 25MHz or 20MHz clock on AK4137. That would cause it to produce non-standard output sample rates, but that's okay. Since its clock would then be an integer subharmonic of the 100MHz Q2M clock, it is much more likely they could be injection pulled or injection locked together. If it worked, that would probably be lowest cost way to get them in exact timing with each other and hopefully at audibly lower jitter than AK4137 produces now. However, clocks we use are not specified for that type of pulling, so although it might be made to work, it might not work with different clocks than the exact ones I used, and might require some fiddling to make it work on other boards. Another approach might be to use a low-jitter clock divider chip and use the 100HMz Q2M clock as the divider input to then generate a lower frequency exactly synced clock for AK4137. Cost for experimenting with it and implementing it that way might be higher, but chances of it working reliably and doing so sooner rather than later are probably better than playing around with injection pulling schemes. So, I am taking a look at some prospective clock divider chips and evaluation boards I might be able to try. In case I don't like any of the options, I do happen to have 25MHz clock module here I could swap onto an AK4137 board to see how that might work out.
Its also possible I will leave clocking as it is for now, and turn attention back to thinking about what might be done with interpolation filtering using Spartan 6.
Its also possible I will leave clocking as it is for now, and turn attention back to thinking about what might be done with interpolation filtering using Spartan 6.
Last edited:
Enough sorting through master mode options that I have it figured out pretty well. Switched back to normal mode and like the sound quality that way better. Therefore, doesn't seem like there is any reason to rush into modding AK4137, or even to play around more with master mode operation for the time being. Getting AK4137 crystal clock more in sync with ES9038Q2M clock seems like what is really needed to get any better sound quality any with clocking changes.
Maybe I will swap that 25MHz clock I have in stock onto an AK4137 board and see how that works. Probably worth giving a try.
Maybe I will swap that 25MHz clock I have in stock onto an AK4137 board and see how that works. Probably worth giving a try.
25MHz clock has been installed on one of the AK4137 boards. Just have to make sure it oscillates, replace a couple of 0.01uf SMD caps I removed to improve access, and then I can swap this AK4137 board into the test setup. Should be interesting to see what happens.
Also, thinking about changing how power is fed into the ES9038Q2M board clock. I bypassed the filter inductor since the clock has its own dedicated voltage regulator, but for pulling experiments it could be having the inductor in the circuit might help. When I was experimenting before pushing on boards to bring then into low-jitter sync, I sort of got the impression that power supply and or close-by ground coupling might be more effective than trying to pull them by way of the output terminals. Just have to try some things to see what happens. Maybe get lucky, maybe not. If not, I'll keep thinking about it some more.
Also, thinking about changing how power is fed into the ES9038Q2M board clock. I bypassed the filter inductor since the clock has its own dedicated voltage regulator, but for pulling experiments it could be having the inductor in the circuit might help. When I was experimenting before pushing on boards to bring then into low-jitter sync, I sort of got the impression that power supply and or close-by ground coupling might be more effective than trying to pull them by way of the output terminals. Just have to try some things to see what happens. Maybe get lucky, maybe not. If not, I'll keep thinking about it some more.
Well, the ($2) 25MHz test clock works about the same as the old 22.5MHz clock (that came with AK4137). Not easy to pull its frequency where I want it and keep it there with adequate stability.
However, it turns out I found an ultra-low jitter 25MHz voltage controlled crystal oscillator. It doesn't have a very big tuning range, and it costs more than a Crystek clock, but it might work to allow ultra low jitter upsampled DSD in the modded ES9038Q2M dac. If needed, might even be able to servo it using Arduino or RPi. The DAC chip has a DPLL number that can be accessed via I2C bus. The number shows where the DPLL is tracking at the moment. If it looked like it was drifting out of adjustment range, maybe a little PWM dac, or probably better, an external 12-bit instrumentation dac controlled by Arduino could keep the VCO in optimal tune. The assumes of course that changes are slow and and of small magnitude, but it might be an option if it happened to come to be needed.
First though, before going as far as ordering a test clock I am going to get a low jitter clock divider chip here and use it to do some tests and help take some measurements to better understand what clocking of AK4137 works best for low DPLL bandwidth operation of ES9038Q2M. Hopefully, some things can be better understood that aren't described in AK4137 and or ES9038Q2M data sheets.
Probably take about a week before I can get the divider here. Then I will need find a power supply setup for it, and probably mod an AK4137 board with a u.fl connector to patch it in. Might want to further mod a dac board too to add another u.fl connector to it. The u.fl connectors would be used to patch clock connections to the divider chip.
However, it turns out I found an ultra-low jitter 25MHz voltage controlled crystal oscillator. It doesn't have a very big tuning range, and it costs more than a Crystek clock, but it might work to allow ultra low jitter upsampled DSD in the modded ES9038Q2M dac. If needed, might even be able to servo it using Arduino or RPi. The DAC chip has a DPLL number that can be accessed via I2C bus. The number shows where the DPLL is tracking at the moment. If it looked like it was drifting out of adjustment range, maybe a little PWM dac, or probably better, an external 12-bit instrumentation dac controlled by Arduino could keep the VCO in optimal tune. The assumes of course that changes are slow and and of small magnitude, but it might be an option if it happened to come to be needed.
First though, before going as far as ordering a test clock I am going to get a low jitter clock divider chip here and use it to do some tests and help take some measurements to better understand what clocking of AK4137 works best for low DPLL bandwidth operation of ES9038Q2M. Hopefully, some things can be better understood that aren't described in AK4137 and or ES9038Q2M data sheets.
Probably take about a week before I can get the divider here. Then I will need find a power supply setup for it, and probably mod an AK4137 board with a u.fl connector to patch it in. Might want to further mod a dac board too to add another u.fl connector to it. The u.fl connectors would be used to patch clock connections to the divider chip.
Last edited:
thanks for the info
will leave the resistors be.VCC10 leads to the 16 pin ribbon cable which connects the display and controls to the DAC PCB, cant inspect at the other end of the cable due to how they designed the display and controls (permanently stuck inside the DAC enclosure), was probably something to with the microcontroller.
It would be odd if no loss of functionality with the now missing power supply rail.
there was some loss that i didnt notice at first, the display usually displays bit depth (16, 24,1 for DSD) in small text at bottom, but now its stuck on 1 bit always
also hardly a loss but when there is no USB cable attached it defaults to displaying 96kHz, previously it would display the last used samplerate
Last edited:
Now in the process of getting ready to do some new dac tests. The test 25MHz clock has been removed from AK4237 board and a u.fl connector installed instead. A u.fl connector was also installed on dac board at Q2M GPIO #1 pin. The u.fl connectors will be used to send a copy dac board MCLK to LMK01000 clock divider board where it will be reduced in frequency, possibly phase adjusted, and then sent to AK4137 board to act as master clock for DSD upsampling. The idea is that the resulting AK4137 reference clock signal should be an exact integer subharmonic of Q2M MCLK. It should also be reasonably low jitter since it will be derived from the dac board Crystek clock (MCLK) with only a very small amount of additional jitter from the low jitter divider chip. Shouldn't imagine AK4137 output reference clocking could get much more optimal for producing a DSD BCLK signal allowing dac DPLL bandwidth to be very much minimized while maintaining good stability (same should also be the case for PCM I2S BCLK and LRCK, assuming we set AK4137 to output that mode). That's the idea anyway, and what the test hopefully will show can be accomplished by the above described means.
Hi Mark,Hi Paul,
Guess I should have edited the schematic before now. It does look like it might be a little easier for builders to read.
Good to hear of your plans to get started.
Please let us know if any problems or questions. Don't forget the output stage sound quality may be improved with film caps added to the +-15v rails. I haven't tried to figure out the minimum amount that works best (which may vary with type of +-15v supply used), but tested with about 100uf per +-15v rail.
Also, we always like to look at pictures of builds. Be great if you get a chance to take some along the way.
Mark
Got all my parts on hand so I'm going to try to build the output stage this week. One question about the added capacitance on the power rails, can electrolytics be used bypassed with film. I've got some films on hand, but only in lower values ~ 1 uF, got a lot of small higher value electros on hand. I was planning to put them on the output board along with the voltage divider for the AVCC Vref. That way I would avoid overly hacking up the DAC board.
Paul
Hi Mark,
Do you think these film caps will be ok for the +-15v rails?
MKS4D053306G00KSSD WIMA | Mouser France
It is 100uf per rail, so I will need 6 of them, is that right?
Thank you for your help.
Do you think these film caps will be ok for the +-15v rails?
MKS4D053306G00KSSD WIMA | Mouser France
It is 100uf per rail, so I will need 6 of them, is that right?
Thank you for your help.
I saw this one too, 100uf for 13€. do you think it will perform well too?
R60EW61005000K KEMET | Mouser France
Hi Guy,
Looks like lots of questions about film caps. A little history from here: At first I had 1,000uf electrolytics on the output stage board and some small film caps of around 1uf. That's when I started experimenting to see if I could affect sound quality by changing caps. I added 10,000uf electrolytics to the output stage and no improvement.
That's when I soldered together all the Wima stacked film caps I had as an experiment. They are of the Wima type terry22 linked to, but the values for each rail add up using the following values: 10+10+22+33+33 (all numbers in uf)
The total comes out to 108uf per +-15v rail. That made really good difference and that is also what I tried using to make Allo Katana dac sound better.
Some things I don't know are: 1) What is the minimum value in film caps per rail needed to get most of the improvement in sound quality? (I only know that very small values, such as 1uf, don't help much), and 2) Can other types of lower cost film caps work as well as the Wima caps?
Regarding the second question, the low cost Kemet film caps are probably not at good. For one thing they say they caps are either stacked or wound, but they don't say which for each value. Stacked is presumably better for lower ESR and ESL, especially up to 100kHz or so where our dac filters are still working to remove HF noise. In addition to questions about construction, the Wima caps have a very good reputation as being good for sound quality. The foregoing observations still don't answer the question as to whether or not the Kemet caps are good enough to get, say, 90% of the benefit of Wima caps in our application. I don't know. Someone would have to try both ways to find out in order for us to know for sure. I can add some of the Kemet 100uf caps to my next Mouser order, but don't know when I can get to it. Seems like it might be a long shot for Kemet to work as well, though. For those that want or need to order all parts at once, I would say I can only vouch for the effectiveness of the Wima caps I tried. In addition, 100uf isn't a whole lot of filter capacitance for an output stage, probably about the minimum. Obviously, there are also lots of electrolytics in use here too, but they don't seem to do anything that the film caps are doing. Therefore, I don't think 100uf is probably an excessive amount of film caps to use.
Lastly, regarding the question of: whether or not 33+33+33 (all in uf) of Wima film caps per rail is good enough, rather than using the particular combination of caps I did? My guess would be that might be a reasonable way to help reduce cost and get nearly all of the benefit. In fact, I now have another bundle of 5x 33uf Wima caps (165uf total per rail) that I can probably test with before very long. Even without testing them though, I would say that buying a total of 6 each of the Wima 33uf (3 in parallel per rail) is something I could recommend. If we later find that some smaller values in parallel help a little more, then more could always be added in parallel later (but unlikely to be necessary is my guess).
Sorry for all the verbiage, but I want to be clear on what I know and don't know. It would be better to be able to give absolute numbers, but that probably depends on the exact power supply the caps are used with too. If using something like Jung Super Regulator or maybe Salas then extra film caps might make much less difference (but might still help some), or they could even cause stability problems for Salas (you would have to ask him for advice about that). For most regulators, including LDOs, the film caps should probably be used.
Regards,
Mark
EDIT: Please see next post below for some additional thoughts.
Looks like lots of questions about film caps. A little history from here: At first I had 1,000uf electrolytics on the output stage board and some small film caps of around 1uf. That's when I started experimenting to see if I could affect sound quality by changing caps. I added 10,000uf electrolytics to the output stage and no improvement.
That's when I soldered together all the Wima stacked film caps I had as an experiment. They are of the Wima type terry22 linked to, but the values for each rail add up using the following values: 10+10+22+33+33 (all numbers in uf)
The total comes out to 108uf per +-15v rail. That made really good difference and that is also what I tried using to make Allo Katana dac sound better.
Some things I don't know are: 1) What is the minimum value in film caps per rail needed to get most of the improvement in sound quality? (I only know that very small values, such as 1uf, don't help much), and 2) Can other types of lower cost film caps work as well as the Wima caps?
Regarding the second question, the low cost Kemet film caps are probably not at good. For one thing they say they caps are either stacked or wound, but they don't say which for each value. Stacked is presumably better for lower ESR and ESL, especially up to 100kHz or so where our dac filters are still working to remove HF noise. In addition to questions about construction, the Wima caps have a very good reputation as being good for sound quality. The foregoing observations still don't answer the question as to whether or not the Kemet caps are good enough to get, say, 90% of the benefit of Wima caps in our application. I don't know. Someone would have to try both ways to find out in order for us to know for sure. I can add some of the Kemet 100uf caps to my next Mouser order, but don't know when I can get to it. Seems like it might be a long shot for Kemet to work as well, though. For those that want or need to order all parts at once, I would say I can only vouch for the effectiveness of the Wima caps I tried. In addition, 100uf isn't a whole lot of filter capacitance for an output stage, probably about the minimum. Obviously, there are also lots of electrolytics in use here too, but they don't seem to do anything that the film caps are doing. Therefore, I don't think 100uf is probably an excessive amount of film caps to use.
Lastly, regarding the question of: whether or not 33+33+33 (all in uf) of Wima film caps per rail is good enough, rather than using the particular combination of caps I did? My guess would be that might be a reasonable way to help reduce cost and get nearly all of the benefit. In fact, I now have another bundle of 5x 33uf Wima caps (165uf total per rail) that I can probably test with before very long. Even without testing them though, I would say that buying a total of 6 each of the Wima 33uf (3 in parallel per rail) is something I could recommend. If we later find that some smaller values in parallel help a little more, then more could always be added in parallel later (but unlikely to be necessary is my guess).
Sorry for all the verbiage, but I want to be clear on what I know and don't know. It would be better to be able to give absolute numbers, but that probably depends on the exact power supply the caps are used with too. If using something like Jung Super Regulator or maybe Salas then extra film caps might make much less difference (but might still help some), or they could even cause stability problems for Salas (you would have to ask him for advice about that). For most regulators, including LDOs, the film caps should probably be used.
Regards,
Mark
EDIT: Please see next post below for some additional thoughts.
Last edited:
Just looked over the Kemet film cap data sheet some more. Noticed that dissipation factor is not rated for 10kHz. Also, in the maximum current vs frequency graphs, 100uf and 47uf caps are not rated for use over 10kHz. Those are worrisome signs. We should probably forget those, and stick with Wima instead.
EDIT: Looking at similar graphs for the Wima caps, looks like 22uf are the biggest ones fully rated up 100kHz. They don't show a graph for 33uf, but for 47uf, the graph ends at about 50kHz. That suggests the combination of Wima caps that I used might have benefits over using only 33uf values in parallel. Thus, probably best to use the exact values I did if that is an option for you.
EDIT: Looking at similar graphs for the Wima caps, looks like 22uf are the biggest ones fully rated up 100kHz. They don't show a graph for 33uf, but for 47uf, the graph ends at about 50kHz. That suggests the combination of Wima caps that I used might have benefits over using only 33uf values in parallel. Thus, probably best to use the exact values I did if that is an option for you.
Last edited:
It seems a total nonsense to spend approx 40 box ($) for wimas in the power rail. Just save money for audio quality opa-s ! Mark did you try audio grade electrolitic caps (Nichicon FG, KZ; Elna Silmic, Cerafine) there? What usually people do just bypass 220 uF Elnas with smaller (100 nF MKP) wimas.
Last edited: