I had a lot of fun a long time ago by getting i2s right from SBC. I've also used Ian Canada's boards and the isolator, reclocker, and dac made by allo.
I think extracting the i2s from the signal transmitted from the actual usb is a better way than affecting the radiation noise of sbc with a board with a masterclock right above it. It's even more so these days when you can use the high-speed usb isolator easily.
I think extracting the i2s from the signal transmitted from the actual usb is a better way than affecting the radiation noise of sbc with a board with a masterclock right above it. It's even more so these days when you can use the high-speed usb isolator easily.
I don't think it's that simple. The metal shield makes high-frequency noise easily coupled due to the capacitance with the board of the Raspberry Pi, and the metal shield becomes a loop antenna that again radiate the noise.
In any case, even with those various efforts, the noise remains, and I feel like I'm not somehow satisfied, I'm going to do something more, it's not going to end, even though I put in a very good clock and a very good power.
All those tricks were successful every time, including removing all the switching regulators on the Raspberry Pi and trying to make a supercap power source with 3.3v, 1.8v, and 1.2v individually. On the contrary, it means that no matter what you do, you will have another homework to solve.
I think I've come back a very long way from the moment I pick and use i2s right from SBC.
In any case, even with those various efforts, the noise remains, and I feel like I'm not somehow satisfied, I'm going to do something more, it's not going to end, even though I put in a very good clock and a very good power.
All those tricks were successful every time, including removing all the switching regulators on the Raspberry Pi and trying to make a supercap power source with 3.3v, 1.8v, and 1.2v individually. On the contrary, it means that no matter what you do, you will have another homework to solve.
I think I've come back a very long way from the moment I pick and use i2s right from SBC.
Of course this should work, but it's not clear to me why you are using an oscillator + LDO instead of a simple crystal + 2 capacitors?
This is not a DAC oscillator, jitter and other requirements are much less significant here.
Of course, you can use crystals. However, in the actual design, there are many points to consider such as the drive level, load capacitance, and radiation noise of the crystal, whereas for me, LDO + TCXO is much more concise and has good performance and is attractive. Since it is DIY, the price difference in parts is not a big consideration. Probably more than 90% of it reflects my unnecessary greed, but...
how did you manage to remove switching regulators from the pie? as far i know this is not possible with the 3B+ or 4B
any chance to get the gerber files too? or will there be another revision? so we just need to order it
I have one also been working with AD for six months had the original protos here and then me and CG worked out a bunch of stuff.
First there are tons of HS errors. Especially above 96Khz audio. I have both the TEK USB Compliance and 3 different protocol analyzers. The larger the packet is the more errors because of PLL drift.
I used a TXC crystal that would be used in re-transmission of HS USB frames and that seem to work really well. This crystal was tested and had very low phase noise when I used it on another DAC project.
The host side is a little tricky. First proto I used a low noise 3V3 but the power up was a bit slow as there is a timing issue there. So I filtered the heck out of the host side and used the internal there but on the Device side I have a ultra low 3V3 regulator (47nVrms @ 1Hz FS, 108nVrm @ 1Hz HS) and have a really low noise 5V regulator for the VBUS.
Care to route the USB is required with shorter traces better.
I see between 17dB SNR improvement with this over a direct connection to my MacBook Pro i9x8.
Thanks,
Gordon
First there are tons of HS errors. Especially above 96Khz audio. I have both the TEK USB Compliance and 3 different protocol analyzers. The larger the packet is the more errors because of PLL drift.
I used a TXC crystal that would be used in re-transmission of HS USB frames and that seem to work really well. This crystal was tested and had very low phase noise when I used it on another DAC project.
The host side is a little tricky. First proto I used a low noise 3V3 but the power up was a bit slow as there is a timing issue there. So I filtered the heck out of the host side and used the internal there but on the Device side I have a ultra low 3V3 regulator (47nVrms @ 1Hz FS, 108nVrm @ 1Hz HS) and have a really low noise 5V regulator for the VBUS.
Care to route the USB is required with shorter traces better.
I see between 17dB SNR improvement with this over a direct connection to my MacBook Pro i9x8.
Thanks,
Gordon
ah thought so, what you use now and how it compares? just a usbisolator and usb dac on your normal pc?
I'm using a tinker board s for now. I'm using the internal clocks floating them off the PCB on very thin silver wire(as Andrea Mori suggested). Of course, I'm not saying this is the best way to do it, as it's easier to judge the effectiveness of efforts like isolators when there is other power noise present, assuming that the USB signal data is at least flawless. In terms of low noise, I'd rather have a modern MAC or NUC.
Just a note on computers and boards, PI's tinkers, modules ect.
Look these things suck when it comes to noise. I designed 5 PC's in my lifetime and I can tell you stories. We did 980K PC's for the government one time (long story no need to go through it here) and we had them made a DELL because the work was larger than anything in Cincinnati at the time. They had 2 accountants chained to my desk during the design phase to regulate costs. As long as it passed CSA and FCC it was the cheapest thing you could make to get the job done. Changing the clock would probably have minimal effects on everything.
I did a couple of linear supplies for RPI4 modules and then 4166 out without all the extra PCIe stuff (i.e. USB SS, SSD ect) with dual supplies one for the PI one for VBUS and that worked pretty good, because I could house the Compute Module in a can on the board. Biggest problem was entering all the freaken pinouts on that sucker.
Anyway, isolating the computational part of this mess is a good thing. But there are diminishing returns on modifying the CPU.
Gordon
Look these things suck when it comes to noise. I designed 5 PC's in my lifetime and I can tell you stories. We did 980K PC's for the government one time (long story no need to go through it here) and we had them made a DELL because the work was larger than anything in Cincinnati at the time. They had 2 accountants chained to my desk during the design phase to regulate costs. As long as it passed CSA and FCC it was the cheapest thing you could make to get the job done. Changing the clock would probably have minimal effects on everything.
I did a couple of linear supplies for RPI4 modules and then 4166 out without all the extra PCIe stuff (i.e. USB SS, SSD ect) with dual supplies one for the PI one for VBUS and that worked pretty good, because I could house the Compute Module in a can on the board. Biggest problem was entering all the freaken pinouts on that sucker.
Anyway, isolating the computational part of this mess is a good thing. But there are diminishing returns on modifying the CPU.
Gordon
Ah, i asked because i also plan to modify a rpi, tho im not sure yet if its worth it, i appreciate any expierences in this regard
exchanging clocks, replacing voltage regulators (or adding additional capacitance to the voltage regulators outputs that cant be replaced) and adding clean input voltage and adding shielding are some things on list
Modifying these modules or boards is a waste of time.
First you have to have a board or module with a real Ethernet port and if you can order it without WIFI then you are better off. RPI3 and below use USB to Ethernet controllers and are not good. WIFI sucks all around.
Second a good clean supply for these is worth more than modifying.
Third adding the 4166 to this with split isolated supplies is really worth it.
The Compute Module RPI4 is the best starting point because it has real Ethernet, 2 root USB HS and you can wire all that up and buy a muMetal can to go over it. Get a good Halo 10/100/1000 Ethernet magnetics on the mother board with the 4166 on one of the 2 USB ports and then have a split supply to power the RPI4 and VBUS separately. That would out do most of the $3K streamers out there today.
Thanks,
G.
First you have to have a board or module with a real Ethernet port and if you can order it without WIFI then you are better off. RPI3 and below use USB to Ethernet controllers and are not good. WIFI sucks all around.
Second a good clean supply for these is worth more than modifying.
Third adding the 4166 to this with split isolated supplies is really worth it.
The Compute Module RPI4 is the best starting point because it has real Ethernet, 2 root USB HS and you can wire all that up and buy a muMetal can to go over it. Get a good Halo 10/100/1000 Ethernet magnetics on the mother board with the 4166 on one of the 2 USB ports and then have a split supply to power the RPI4 and VBUS separately. That would out do most of the $3K streamers out there today.
Thanks,
G.
i actually have a cm4 + waveshare adapter board to normal rpi4 layout here
i choose it because of your points about ethernet/usb/wifi and aditionally emmctho im really wondering how bad the switching regulators are, these are unexchangeable and im worried that there is a "wall" i hit because it
i actually also have already txco clocks here to swap them on the rpi4 + 80 panasonic caps for filtering the power rails of the switching regulator,
planned is also to add a 3,3V and 1,1V ldo to exchange the waveshare board regulators, add some shielding (is mu metal overall better shielding which can be grounded or do i need aluminium aditionally?), a isolator hat + ian canada ddc hat
maybe i should just try it and see what it gives me, i kinda gave up on the idea because i heared how my desktoppc gives better sq than my tweaked moode rpi4 setup but im currently reevaluating...
thats also why i wanna order this usb isolator, to maybe compare it to the topping hs01 i already have, so if the op is ready, gerberfiles would be really nice so others can just order the pcb already(or mostly) populated with parts
I'm surprised that nobody has commented on this. T'would seem like a big deal, especially in a world where some people rank DACs and other products based on tenths of a dB of SINAD.
getting rid of the dirty pc power is definitely a good (first) step in improving usb
usb isolators do it too
you can also just disconnect vbus and add a 100ohm resistor between grounds, this is basicly what the ifi idefender does and go in with any power supply you want
I agree with everything else, but the above statement is not always true just because in some cases the Vbus wire is still needed even using an external power supply (5VDC).
https://www.denafripsth.com/support/ (USB 2.0 Cable chapter)
https://www.diyaudio.com/community/threads/powering-the-usb-input-chip.394186/post-7226884