Hi, I am using an ADC with 8 single ended inputs, and for a referensvoltage
I have a REF02. In the datasheet for the ADC it is quite specified how the
GND-pins should be connected (to AGND), but how about the decoupling
capacitanses on the inputs. First I have 2 decoupling capacitanses on the
VCC-input and then I have 2 on the Vref-pin. But are they suppose to be
connected to AGND or GND. My intention is to have 2 separate ground-planes
at the same potential (0V) and then just connect them in one single point,
also here decoupled. So I have the REF02-circuit connected to AGND and the
ADC, but how should I do about the capacitanses in the inputs?? should
they be connected to GND or AGND??

Thank you!!

Frida wrote:

Further reading in an application note reveals that the ADC
is supposed be mounter on the AGND with its AGND and DGND
pins on the AGND. Go ahead, there are quite extensive
application notes at the various manufacturers websites.

Rene
--
Ing.Buero R.Tschaggelar - http://www.ibrtses.com
& commercial newsgroups - http://www.talkto.net

Hi,

I am sorry if I wasn't clear. You are absolutely right, the datasheets
specify clearly how the pins on the ADC should be connected. But it
doesn't say anything about the decoupling capacitances. So my questiod was
really about them and not really about the ADC itself. Should the
decoupling capacitances be connected to AGND or GND?? Or maybe is it
implicitly saying that the decoupling capacitances should be connected to
the same ground as the ADC, but that is what I can't find out.

Thank you!

"Frida" <friost@foi.se> a écrit dans le message de news:
359858ca088f8d139dc3b6df38590f25@loc...cequipment.com...
Use a unique, large, full ground plane, and connect all pins that should be
connected to all xGND to this ground plane. This is by far the safer
alternative for 99,999% of the designs. Dual-ground plane architectures are
VERY difficult to design, and are the root cause of serious headackes when
doing EMC testing... Except of isolated / high voltage designs of course.

Anybody disagree ? (this is a subject discussed quite often so I expect
thousands of answers ranging from "stupid answer" to "of course"....)

Friendly,
Robert

Frida wrote:
To GND. The underlying principle is that there should be no
currents flowing in the AGND circuit. Currents have spikes, and
cause IR drops, and generally create havoc for analog inputs.
Decoupling capacitors shunt current in order to maintain voltages.

--
"If you want to post a followup via groups.google.com, don't use
the broken "Reply" link at the bottom of the article. Click on
"show options" at the top of the article, then click on the
"Reply" at the bottom of the article headers." - Keith Thompson

Robert Lacoste wrote:

The answers will be far more likely to be correct if the question
is asked in sci.electronics.design where it is on-topic. While it
is true that some EEs are also interested in embedded systems
programming, many are not and thus don't read comp.arch.embedded.

Hello Robert, Hello Frida,

Amen. Meaning I agree and thus I guess I'll fall into the category "of
course". In my 20 years on the beat I have never seen a split ground
system work well. A few of them kind of worked but then they didn't pass
EMI until we performed a common ground relayout.

One way to think about it is that a typical system has more than one
connection to the world. So while there may be just that single ground
connection under the ADC there are signals coming in from the analog
side to be converted and then other signals going to displays, keyboards
etc. on the digital side. All these will also carry ground leads. This
causes lots of not well controlled loops with each of them having to be
broken by ferrites or other means, something that rarely works well.
This is just one of the reasons why a split ground architecture can lead
to sleepless nights.

On my designs I always start out with a nice full ground plane and make
sure that any currents on that plane won't cause more transients at the
converter than can be tolerated. It has never been a problem, as long as
the placement of components is suitable.

Cordialement & med bästa hälsningar, Joerg

http://www.analogconsultants.com

Frida wrote:
It's there - between the lines: if the only ground connected
to the chip is AGND, the by-pass capacitors have to be connected
to it.

A different story is that the incoming power supply
should be decoupled by a choke to avoid polluting
the analog ground with the crud on the supply line
via the caps.

--

Tauno Voipio
tauno voipio (at) iki fi

"Joerg" <notthisjoergsch@removethispacbell.net> wrote

Depends on what is meant by 'split ground'.

I regularly use a star ground that is then grounded to the 'system' ground
at the center of the star. This creates a 'ground plane', though a
pretty cruddy one as the lines going to the star are not gridded.
Often more than one star.

And I have had boards that were split horizontally - left half of
the board analog and right half digital. 5v and Dground layers for the
digital side and +/- V and Aground layers on the analog side.
One connection between the two grounds made at the power-supply
reference point.

For process control instrumentation it is not uncommon to have one
ground per process input/output. The back panel of process control
cabinets often have a drilled copper bar for _the_ signal ground
(used to be labeled 'Mecca Ground'). Often sensors are grounded at
the source.

I haven't had any problems with the exception of clients who insist
on 2-layer boards for high speed designs. The problem has been
EMI susceptibility.

YMWV.

--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer: Electronics; Informatics; Photonics.
To reply, remove spaces: n o lindan at ix . netcom . com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/

Hi Nicholas,

EMI is usually the first problem I find with approaches that didn't use
a single gound plane. It was more prevalent than performance issues
(although there were tons of these, too). A split or otherwise separated
plane makes for a nice big dipole antenna that radiates and receives.
Any RF currents flowing across the node where they are connected are at
the same time imposed onto that unwanted dipole antenna. Inevitably any
system will have some physical connections to the rest to the world.
Then these dipoles become huge loop antennas and it all gets even worse.
Just my experience, and I have seen it over and over again.

Until now I or rather my clients haven't gotten any mileage out of
separate ground approaches, except the miles for the repeat trip to the
EMC lab ... ;-)

Then again, many EMC labs are located in rather pristine areas of the
country so everything has an upside.

Regards, Joerg

http://www.analogconsultants.com

"Joerg" <notthisjoergsch@removethispacbell.net> a écrit dans le message de
news: dleKd.7658$8Z1.1585@newssvr14.news.prodigy.com...
Fully in agreement with Joerg. In fact in you use a "split" ground plane
with a unique connection under an ADC chip, then in order to avoid any
EMI-receiving/emitting coil you have only one solution : All signal lines
going from one side to the other must be routed EXACTLY ABOVE the connexion
point between the two grounds ! If not you have a marvelous current loop...

Cheers,
Robert

Hello Robert,

Yes, and when the product is now behind schedule the only way to fix
these loops is with the 'Aspirin method'. Lots of #43 material toroids
and other cores. The worst case I ever saw required about two pounds (!)
of ferrite to make it perform and pass EMC. Later after a redesign with
a common ground structure that dropped to zero pounds, with even better
EMC margins. The only sad person was the sales guy of the ferrite supplier.

Regards, Joerg

http://www.analogconsultants.com

Frida wrote:
techonline has a good analog lecture on A/D grounding you may want to
play
http://www.techonline.com/community/...g/course/13479

On Thu, 27 Jan 2005 08:50:38 -0500, "Frida" <friost@foi.se> wrote:


The VCC supply is for the digital circuitry of the ADC chip. It will
therefore be noisy, so the decoupling capacitor should be connected to
digital groud.

The suggestion of having only one xGND plane will obviously work, and
is a good solutions for many applicaitons if you have a very solid
ground plane and are not critically concerened with ADC accuracy.
However, if you need to measure very small signals or get ultimate
accuracy, then you need to have separate AGND and GND planes that are
connected at only at a single point.

regards,
Johnny.

Johnny wrote:
I think the folks here (as well as the a/d manufacturers) are
suggesting a single ground plane is the best solution (lowest noise)
not a comprimise solution for the majority of cases when using
commerical A/D's. I think the reason is that an A/D is already an
integrated mixed signal device and you have to deal with the
complications and limitations of the grounding system thats brings.
Connecting its GND to a seperate GND plane(assumed noiser) can
internally couple noise into the analog portion of the a/d via stray
capacitance. For the rest of digital circuitry on the board the idea is
to physically seperate them from the analog devices on the board so
that the returns of the noisy digital chips don't have to cross over
analog returns to get back to the power supply return.

Obviously there are exceptions to this rule especially for strange
shaped PCB's or motherboard/daughterboard designs, or discete optically
isolated A/D's but for large single pcb systems using integrated A/D's
or microcontrollers, its a pretty reliable option.