Here a question that problably will have equal answers as there are engineers but I'll give it a shoot: Where is the best place to splitt digital and analog ground? For example, on a PCB there are both digital curcuits like FPGA or DSP and sensitiv analog ICs like sensors and op-amps. Beetween these digital and analog world there are ADCs and DACs. The most intuitive would be to split the ground at the different ADCs, but should this be done at a very narrow point and should the ground planes be seperated with a ferrite bead or just "0Ohm"..?
There are several ground methods wich people like to put forward as the 'perfect' solution (ie solid aluminium ground chasis, star gnd,...) each one is good in its own way, but basically you need to consider the noisy currents from the digital supply grounds and make sure they dont cross through any part of the quiet ground plane so that they can introduce a voltage wich is added onto your quiet signal.
However if you have multiple signals with multiple DACs, if they are all coming down the same shielded signal cable and on the same board then you just have one quiet gnd plane and one point at wich digital and quiet gnd are conected.
I gues if you put conections with 0ohm you can always swap it for a ferrite bead, or even a 10ohm resistor. It realy all depends where the worst noise in the system is as to wether things like this make it better or worse.
Oh and watch those supply decoupling caps too, dont forget they have most of the noisy currents through them. often the SPI interface introduces a more intrusive noise when it is sending bits, so it is worth making sure the currents in this path are contained.
If you have seperate sources and seperate boards it gets more complicated as the current paths become more dificult to identify. It may be best to consider each one as seperate and not need to connect the quiet grounds together, however if their grounds are connected together at the far end (or at the case) a ground loop could arise wich acts like a loop antena.
It is probably not practical to completly isolate the quiet grounds but a common mode torroid filter on the signal+gnd lead might help to get rid of induced circulating currents in the ground. A ferrite bead between the grounds might help to reduce this too, it might be best to connect the digital and quiet grounds together at just one place, but again wether this makes it better or worse depends on other things as the DACS might complain if there is a large signal between thier analogue and digital gnd.
So suck it and see, or copy something similar that works well, or look at something that doesnt work very well and dont do it that way.
As others have indicated, controlling noise in circuits is a very large subject of which grounding is just one part. If you have a component or circuit that is particularly sensitive, running a separate ground and power traces from the power supply capacitors is probably the best strategy. With some of the new very low power microcontrollers, it is now possible to design boards that generate very little digital noise to begin with.
Don't do it. Use a single solid ground plane and handle any low-level problems locally.
Splitting planes implies an impedance between them, hence high relative AC potentials between "grounds". That will generally cause a lot more trouble than it cures.
Actually you can do it - and sometimes you must do it if you want your design to work. The best suggestion was already given by Graham,
works for me as well.
If you are doing just analog and digital, well, placement often (if not always) can be sufficient. However, if you have full scale input signal in the mV or tens of mV range, plus high speed sampling (1, 10 or more MSPS) at 14+ bits and, say, some smps on the same board which has to be small..... well, you have to split the ground plane - and some others as well. And again, there is no better advice which can be given than the one Graham already gave - if it is not detailed enough for you, you need to gain more experience/knowledge rather than an advice.
I'll second John's comment. In 20+ years I have yet to see a split ground situation that performed well enough. Except where safety concerns or regs mandate a complete isolation, there we just had to make it work.
It hasn't for my clients.
On the ultrasound systems we designed the signals range from almost down in the thermal noise to tens of mV. Works fine with a solid common ground plane but didn't with splits under the ADCs.
OTOH I won't complain about designs where people have split the ground plane. After all, that brings in re-design business for me :-)
how many bits conversion? Do you have an SMPS on the same board (tiny enough - like 100 x 50 mm ).
Well like I said it works for me. Check my hi-spec gamma spectrometry module for an example.
Well then give it a shot at my competitors. A number of them would be glad to match my above mentioned module - they all have yet to do so four or five years after I introduced it. I'll be "gespannt" . :-)
I think it's a case of horses for courses. I did actually once try a non-split ground for a DSP processor in some audio gear I was designing for the heck of it and the digital section introduced 'birdies' into the audio at low level. Back to plan A !
Not always. Several of our chips have separate AGND and GND pins, but the intention is *not* to go to different grounds. It's to isolate the on-chip digital noise from the PLL. It's the package inductance along with huge switching currents that's the problem, not card plane noise.
12 bits at over 100MSPS. Plus a time-gain control amp in front to cover the required range.
Yes. But the boards are typically larger, about pizza size. Except one that was a litte under 100mm x 150mm. Size didn't really make a difference. However, this stuff has to perform under heavy duty RF loads such as a nearby diathermia machine. The worst case would be a defibrillator shock while the patient is still connected. That's not supposed to happen but cardiologists can make mistakes after a week of hard overtime and lack of sleep.
I've never worked in that field. I like the bus designator "Nukebus" on your web site.
I wouldn't know who your competitors are but you can give them my web site address ;-)
Very low frequencies is where star-grounds and stuff like that can indeed work. Ultrasound guys consider audio to be DC ;-)
Problems can still creep up when you install such split ground gear in the vicinity of a strong RF emitter. Radar bases, coast guard stations, AM talk radio towers, 13.56MHz "blasters", you name it, I have seen them pretty much all. Mostly I have to come out there to find out what's causing EMI and then sit down with the client to gently break the news that they need a few re-layouts and chassis mods. Usually the engineers take that in strides but often the enclosure designers don't. Frequently they are artists rather than engineers and they can become quite livid when I suggest to move away from an injection molding scheme.
Just recently, I put a power supply (linear and switching regs), a temperature controller (thermocouple-input, 150 watt PWM heater output), a PPM-noise-level, 3-amp NMR gradient coil driver, and a uP on a single small board, with a single ground plane. Works fine and passed FCC and CE first time.
And another board has a VME interface, switching regs, uP, and twelve
12-bit ADCs clocked at 40 MHz, with sub-LSB RMS noise levels. Single plane.
I also did the NIF timing system: about 2000 client points triggered to within a few picoseconds accuracy, spread over the world's biggest laser, the size of a football stadium. No split planes there, either.
Crap. Graham can't even give good advice on how to wipe ;-)
!
Yep. You don't want a common wirebond to cross-couple noise.
Very standard practice in the microchip world.
...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems | manus |
| Phoenix, Arizona Voice:(480)460-2350 | |
| E-mail Address at Website Fax:(480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |
I love to cook with wine. Sometimes I even put it in the food.
The answer is pretty straight forward once you figure out if you dealing with high level digital current, low level digital currents, single or multiple boards, see
Plus professors often teach that stuff. That, and those reference designs, have kept my business humming. When I started out I really didn't want to go into EMI but split grounds have kind of sucked me into that funnel.
It can also mean that the chip designers didn't have much of a clue about the system side of things ;-)
The worst case was a slew of field failures. Some pulses had hit into the substrate diodes and much of the stuff across the AGND-DGND devide fried. Nobody knew where the pulses came from but after a re-layout with a common plane there were no more field failures.
In EMI we have an old rule: The EMI performance of a given design is inversely related to the number of toroids it requires to make it behave.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.