DIY ES9018 Hi-end USB DAC

Qusp Have you read the handbook of black magic? and the one on advanced black magic? Both by Johnson & Graham. Essential reads.

not yet, its on the list. marce is a great source of references for this line of thinking, so yes although i've read some Howard Johnson online, havent forked out the coin for those 2, i'm still brushing up on the math needed for some of this stuff, so its a bit slow going.

i've only recently pulled my finger out to start designing my own stuff from scratch after being in sponge mode for the last few years, but i'm pretty happy with how its all coming together. having a background in graphic design/pre-press has helped a lot at the software level. looking at working the beaglebone black into the mix too.

anyway we should probably give these guys their thread back, do shoot me a PM sometime. I wouldnt mind some referrals for local boardhouses for multilayer if you know any that arent too pricey, i'm just grabbing some 4 layer boards at seeed and gold phoenix before going too hardcore.
 
but without coplanar waveguides, being even more alien here :), the thickness (and FS) is a factor for determining the width of the trace needed and waveguides are going to require impedance controlled layout are they not? cheaper to just get a multilayer board in many cases these days. yeah i'm looking 6 layer, not 8, still too pricey unless you have the ability to run it with other company material like you perhaps. depends, maybe by the time i'm done it'll be cheap enough. most of the logic with mine will be off board though with flex PCB interconnects so its not such a big deal.

I wish the ESS came in a BGA package option

Oh definately on the microstrip stuff. I still prefer CPW, as they are much more forgiving of imperfections in the routing and mostly single mode, also board thickness is much less important. However for my board im intending to have a cml transceiver close enough that transmission line effects don't matter and then run differential into my source that will be based on a zynq fpga.

As for BGA, it is not really necessary, or even helpful to go that far. I mean even Hittite uses QFN for most of their stuff. Just makes routing hard.

I am actually most annoyed about the clock being a cmos input, very hard to convert a pure sine wave into a high quality square wave with no internal terminators. I am going to use one of the new LT chips to do this in my design, running the whole design on a 100MHz OCXO. I need a grand total of 16 channels in my system and so i need 8 of the ESS dac chips. Will be designing boards that use two chips each and build 4 of them.

The system should be capable of 64 or more channels though.
 
Last edited:
Deano, i am not saying that they are hit and miss, however i am saying that many people here have little or no understanding of RF electronics, and it shows in how they implement what really is an RF problem. There is a reason why no modern tech uses CMOS voltage levels on anything critical. There is also a reason why high speed circuits are made on 8 layer boards most of the time. The entirety of these problems have a lot to do with parasitics and resonances. I am not saying I am an experienced RF engineer, far from it however there is a lot of disinformation around about things that have been know in hard fact and the rf literature for decades, if people would just think to look for it.

The thickness being a problem is not strictly true. One could design a double sided board that has the correct impedance using co-planar waveguides. Does this board, no. Would I recommend doing an ESS dac with two layers, no. Tis is not however due to there being a problem with impedance matching.

Does a good board for this need a Teflon board, absolutely not. However even 4 layers cannot really do an ideal design to complement this dac. After a LOT of thought over a few years, i have decided that my board will be 2+4+2 layers, and that is as few as I am willing to go. I would actually prefer to have one with 3+6+3 layers, however the cost there is high at about $120 per pcb and i do not quite feel that I can justify that.

Oh definately on the microstrip stuff. I still prefer CPW, as they are much more forgiving of imperfections in the routing and mostly single mode, also board thickness is much less important. However for my board im intending to have a cml transceiver close enough that transmission line effects don't matter and then run differential into my source that will be based on a zynq fpga.

As for BGA, it is not really necessary, or even helpful to go that far. I mean even Hittite uses QFN for most of their stuff. Just makes routing hard.

I am actually most annoyed about the clock being a cmos input, very hard to convert a pure sine wave into a high quality square wave with no internal terminators. I am going to use one of the new LT chips to do this in my design, running the whole design on a 100MHz OCXO. I need a grand total of 16 channels in my system and so i need 8 of the ESS dac chips. Will be designing boards that use two chips each and build 4 of them.

The system should be capable of 64 or more channels though.

Amazing, we are heading in pretty much the same direction.
USB to FPGA ?
 
No, the zynq fpga has an embeded arm cortex A9 built in, this thing will play music direct from my network. No PC required, also allows me to clock the CPU so that it's main clock is derived from the audio clock. This whole system is going to be synchronous baby. I will be able to act as a USB audio sink for a PC though, and I will be designing it so I can send HOA directly into it as a 16 channel pipe and decode it on the fpga.

The FPGA fabric can also be used to implement DSP on the audio. Though i will not need much.
 
No, the zynq fpga has an embeded arm cortex A9 built in, this thing will play music direct from my network. No PC required, also allows me to clock the CPU so that it's main clock is derived from the audio clock. This whole system is going to be synchronous baby. I will be able to act as a USB audio sink for a PC though, and I will be designing it so I can send HOA directly into it as a 16 channel pipe and decode it on the fpga.

The FPGA fabric can also be used to implement DSP on the audio. Though i will not need much.

So now i show my ignorance -
CML = Current mode logic ? (is LVDS comparable).
HOA = ?

So are you implementing FIR crossovers onthe board ?
 
Last edited:
CML = Current mode logic.
HOA = Higher Order Ambisonics.

thought i suppose you could implement crossovers on the board too, that is not how I will be using it. Good old analog crossovers for me.

Is CML comparable to LVDS ?
Ah yes, Higher Order Ambisonics.
I had one of these systems as a kit from Clive Sinclair in the seventies :)
Bolted it onto a 60's radiogram :D
 
Similar but different. Higher speed. Much higher speed.

I doubt you had a HOA system in the 70s but you probably did have an ambisonics system.

HOA includes the higher order spherical harmonics and so has much greater positional accuracy. The goal is to build a system that can do 3rd order horixzontal plus 2nd order periphony.
 
Similar but different. Higher speed. Much higher speed.

I doubt you had a HOA system in the 70s but you probably did have an ambisonics system.

HOA includes the higher order spherical harmonics and so has much greater positional accuracy. The goal is to build a system that can do 3rd order horixzontal plus 2nd order periphony.

Yeah, just kidding.
But it did work, as long as you repaired the class D amps every few days:mad:

I like the ethernet approach, looked at the XMOS stuff but no enough bandwith with there implememtation.

Spend the last 3 years working on how best to get multi channel I2S from a PC.

Started with ASUS Xonar card and LVDS.

Tried PCIe expansion.

Just having a look at USB3.
 
131072 taps per channel, or 131072 taps total, either way seems totally doable on an external non-PC board. You seem to forget that we have Multiple overlap ffts and 100Mhz hardware at out disposal, also modern fpgas are not small. These things do Realtime ffts on adds that work at 10GSPS for the military.
 
131072 taps per channel, or 131072 taps total, either way seems totally doable on an external non-PC board. You seem to forget that we have Multiple overlap ffts and 100Mhz hardware at out disposal, also modern fpgas are not small. These things do Realtime ffts on adds that work at 10GSPS for the military.

131072 taps per channel.

I appreciate they have dedicated fft hardware but find it difficult to compare the real world horsepower,verses say, Intel i7.