Grounding Scheme within a DAC

[5th element]

OK I've finished by PCB layout for the DAC im gonna build. This is largely based on MWP's DAC but substituting a differnt DAC chip.

So the basic layout is PCM2902 USB interface => DIR1703 SPDIF receiver => PCM1794/8 DAC chip.

Well as im a complete newbie to digital I basically just copied the schematic for the 2902 and the 1703 and then implemented the DAC chip as per TI's data sheets.

Anyway my main concern is grounding.

I have created three seperate ground planes around each section, a USB ground for the 2902 chip, a digital ground for the 1703 and digi side of the 1794/8, and an analogue ground for the analogue side of the 1794/8 and I/V stage.

So on the PCB I have three sets of copper islands around each different section all isolated from each other. My question is how do I link all these together? Do I run a separate feed to the star ground or do I link them together someway on the PCB?

Cheers Matt

[tiroth]

Hmm...well, we're assuming I guess that the USB ground is total drek. (I think this is safe) If you have a TTL S/PDIF output from the USB receiver then I would be tempted to transformer couple to the main digital ground. I have never heard of people using transformers over such a short distance (inches) so you may want to wait for the pros to weigh in on this.

If your I/V stage output is balanced out then the analog ground should connect to the digital ground through a nice fat bridge under the DAC IC, preferably along the division between analog and digital circuitry in the die.

[skippy]

"tiroth" gives good advice. Join the grounds ender the DAC with a fat piece of copper - 10mm is good.

Joining the ground for the 2902 can be done with a wide strip under the digital signals that go between the sections.

Think about ground return currents. Every digital signal has a return current and it's best to make this go the shortest distance.

The best fun I ever had with a design was a 3.4GHz radio, 14V and 120V switched-mode PSUs, audio CODECs and lots of DSP all one one 4-layer board. IT HAD ONE SOLID GROUND PLANE ALL OVER and performed excellently.

[Guido Tent]

Originally posted by skippy

The best fun I ever had with a design was a 3.4GHz radio, 14V and 120V switched-mode PSUs, audio CODECs and lots of DSP all one one 4-layer board. IT HAD ONE SOLID GROUND PLANE ALL OVER and performed excellently.

Hooray !

You see guys, it IS possible (as long as you control the currents)

well done

Guido

http://www.tentlabs.com/Info/Article...decoupling.pdf

[5th element]

Well the PCB is already designed and its a bit impossible to put copper islands under or ontop of the DAC area because im using SSOP to DIP converters and ive got traces on both copper top and bottom nagating any islands around those area. There are large amounts of copper around everything anyway so I'll just try it and see how it turns out.

I was simply asking if there are any do's and dont's with digital stuff like making the analogue ground digital ground or something or keeping then completely separate appart from the star ground.

[tiroth]

ive got traces on both copper top and bottom nagating any islands around those area.

Sounds like trouble brewing. You want your ground plane to be as continuous as possible, if there are large slots or voids you could have trouble.

I've built digital circuits without ground plane though so it will probably work, there is just no guarantee.

[Guido Tent]

Originally posted by tiroth


I've built digital circuits without ground plane though so it will probably work, there is just no guarantee.

Hi

Data-wise they will work most of the time, if not always

Jitter and SNR wise the layout, circuit and grounding scheme are of utmost importance.

cheers

[5th element]

All I can say is we shall see!

[5th element]

OK i have taken a pic of the copper top and bottom so you can see exactly how I have done this. If its complete trash then say so, I have made an active xover board like this and it works superbly. But if you think it will be OK then thats good too. This is already the second attempt at the baord design (the first one was for SSOP, until I realised that might be a bit ambitious so I made is DIP) so I dont really want to go about designing another one.

[guido]

Mmm,

Does not look too well to me. Seems the current loops will be quite big, since the returnpath is not directly under a trace but has to go all the way around. There should be ground under a trace to provide the returnpath. In that way you know where that current will go and you need to make shure there are no other currents near. Now all will go through same ground.

Two stories here (not much experience, so you might ignore this post ):

One guy made a testpcb one day using dig circuit on veroboard.
Did not use groundplane but one big track for ground and connected all to it. Then he had 0.5 volt difference on the groundplane, say over 10 centimeters (from memory, long ago).
This difference looked ugly (wasn't DC, but noise, spikes, crap).
So he was soldering big massive wires on the groundplane, to reduce resistance, result: no change.

Story two: made a pcb myself for a dsp at school. ADC on the edge, than fast memory and then DSP. ISA bus on the underside.
No groundplane, except around the ADC. Two layer only and lots of traces needed. So ground was also big trace around the edge of the pcb. So we got the pcb, stuffed it and plugged it in. Turned the knob, downloaded testprogram and ...... it worked (led started blinking). Hurray!!! So we started using it. All fine, except now and then the DSP did not see the ADC interrupt anymore and needed a reset. Happend randomly. So we got a new DSP and plugged that in (lots of $$$ for the school). Same result, bit less but still it stopped working sometimes.

In the end a solution was a wire close to the ADC-DSP traces (say ~i2s connection) connected to ADC and DSP ground.
Than it worked and we never had to reset again (though the groundloop was ). Resolution was not that good, we did not get the 16bit of the adc. Too much noise, but than again it was not sheelded (yet) and inside a PC. Think we got 14bit.

So i would get a groundplane. If you have too many traces to connect, use wires over the ground layer instead of cutting through it.

Hope this helps,

Regards,

Guido B (so not T )