Routing data and reference for multiple 20-bit DACs

[ezalys]

I've been reading various app notes as well as Ott trying to figure out how to lay out multiple converters on a board. Most explain how to route the data lines but not much about routing the power supply or reference. Ott mentions layout A for multiple converters. This makes sense for the data lines but this would seem to mix up the reference and data lines. Layout B doesn't partition the ground plane and just passes the reference around in the analog half of the board. This seems to make more sense to me but Ott mentions that some partitioning might be helpful when using converters with bit depths beyond 16-bits. He also mentions stripline, which appeals to the microwave designer in me, but this will be the first run of a board and it would be helpful to be able to cut and bodge traces. Any suggestions? 20-bits takes some serious discipline I reckon so any help is appreciated.

[David Hess]

In the ideal situation, if the reference inputs are differential, then the second input can be used to sense the remote ground for the reference source.  Otherwise remote sense can be provided by an operational amplifier or instrumentation amplifier for each DAC.

Another way is to distribute the reference as a current to each DAC with a precision resistor at each DAC to convert it back to a voltage, but I have not tried this in precision applications; it works well at lower precision when common mode noise getting into the reference is a problem.  Generating multiple precision current outputs is an exercise for the reader but ultimately no more complex than using an operational amplifier or instrumentation amplifier to sense remote ground.

Any solution involving resistor matching or operational amplifier offset voltage is likely to require calibrate of each DAC individually and careful error and noise analysis.

[ezalys]

This makes sense but I’m still confounded as far as how the routing should go. I don’t know how to keep the return currents from the SPI lines out of the reference. Either you get return currents spreading through the whole plane because it’s uncut or you cut it and have to run the reference near the digital lines and you’re back at square one — seemingly. Should I just bite the bullet and use strip lines and via fence the crap out of everything?

[Marco]

One option is just to have regulators and voltage references for each ADC.

Does it really make much sense to split the planes if you don't?

[ezalys]

Using LTZ1000s makes that somewhat unrealistic lol.

[Marco]

Why can't you split the planes, but have a digital connection to the split plane on top and an analogue connection on the bottom?

[RandallMcRee]

Both method A and B suffer from problem of reference voltage to one DAC interfering with a different DAC. As you know from Ott shared current is bad and that is why star ground is *good*.

So, to remedy that every DAC should have its own REFP and REFN connection directly back to the REF (put the REF in the middle maybe).

[ezalys]

Yes sorry, the lines are meant to represent busses and not individual wires. The DAC symbols are meant to include a pair of reference buffers and the pink connections are kelvin connections to the reference voltage.

But still — I’m really looking for answers that draw contrast between A and B or offer a third option.

I should also say I’m far more interested in stability and precision than accuracy

[dietert1]

Some time ago i built a 20 Bit DAC with a MAXIM part and if i remember right in the proposed application note both reference and DAC were used with four wire connection, with some zero drift amplifiers in between. I think these buffers would be multiple in a multiple DAC board and then it may be more obvious how to design it.

Regards, Dieter

[jbb]

Spitballin’ here...

This looks like the sort of equipment where cost is driven by the specification. If you’re not making huge quantities, you should also factor in design / verification time.

With that in mind, how about this?
- Single reference
- Split ground planes for reference and DAQ
- Bring analog ground planes together at single star point
- Use star distribution for reference directly above star ground
- Use digital isolators on SPI busses

A big question: what kind of equipment will you connect to the DACs? How will this equipment be grounded? Could you get ground loops?

Maybe you could have a single reference which is distributed with differential methods and an INstrumentation Amplifier (INA) adjacent to each DAC (set gain = 1) to translate the reference to local ground? The integrated chips should have quite good resistor matching - they are often factory trimmed. You’d need to calculate offset, drift and noise contributions to see if it’s sufficient.

[T3sl4co1l]

FYI, guessing this is an extension of this thread, for different channels/devices?
https://www.eevblog.com/forum/projects/design-review-help-placing-smps-regs-ref-for-16-bit-dac/

For my part, without knowing anything more about the design, I'd give the same advice: REF in the middle, ground plane all around.  As mentioned earlier, if REF can be differential, do it; there's nothing wrong with wiring a ground trace on top of a ground plane, just use a net bridge (or 0-ohm jumper, or..) to enforce it in the design.

I did a design with a 24-bit ADC a couple years ago, which seemed to perform at spec.  I didn't have the tools or time to fully evaluate it, mind, just that it seemed to give the expected noise levels and gain.  (Which also had a switching converter nearby -- for the analog supplies -- adequately islanded and filtered.  I did propose ground slots as an option, which the client declined, so solid ground plane it was.)

Tim