[sdiy] analog or digital ground?

[Magnus Danielson]

From: Seb Francis <seb at is-uk.com>
Subject: Re: [sdiy] analog or digital ground?
Date: Mon, 11 Nov 2002 17:53:30 +0000

> Hi Magnus,

Hi Seb,

> Magnus Danielson wrote:
> 
> >
> > Well, the GND is not of much use for the analog side really, but for the
> > digital side it is mandatory. Assuming it's CMOS inside, then nothing much is
> > actually happening in there, but to be sure, you want to do the propper thing
> > with decoupling caps, having minimal traces between the digital power and
> > ground so that you acheive minimal series-induction.
> >
> 
> Interesting you mention decoupling because this adds a further twist .. there
> is no "digital" power on this MUX.  It has only +V/-V which I guess is very
> much connected to the analog side of things.

Well, actually, it's both. If you look at App-note 684 (then one you've found)
you see that V+/V- is actually mainly used in the internal mixed signal stuff.

> I'm thinking it is a bad thing to decouple to AGND, and connect the GND pin
> to DGND. On the other hand decoupling to DGND will for sure introduce noise
> into the analog +/-15V rails.

Well, the GND seems more like a sense-signal in relation to the Control
signals. Look at what the control signal actuall meets, and diff-amp input
which has a voltage level reference buildin. This is the only point where the
GND pin is actually used, and then it doesn't see much dynamics either.

If you think of it, you have a MOSFET input (think - high impedence), with a
propper diffamp stage (I like). This then drives a pair of CMOS invertes
(which is driven V+ rail to V- rail) which in turn drives the actual CMOS
switch.

> So perhaps AGND is the right choice for the GND pin (yes the MUX is CMOS BTW,
> so digital currents are small).

Well, yes... but I start to think is doesn't matter as much as one may expect.

> > Actually, I recall some very interesting appnotes from MAXIM that gave insight
> > to how their analog switches worked, down to the transistors, so that should
> > help you out.
> 
> I've found this appnote:
> http://www.maxim-ic.com/appnotes.cfm/appnote_number/684

Yeap, this was the one I was thinking about.

> Figure 3 shows that the GND is used only as a reference for the logic level
> comparator section.  So from this point of view it seems pretty safe to
> connect GND to digital ground.

It is. I'd expect the current pulled pretty minor. However, as I've pointed
out, it seems like more of a minor thing now.

> The appnote also says that high-frequency noise on the power rails can couple
> to the analog signals, therefore power supply decoupling is necessary
> (presumably to analog ground).

Well, yes. V+ and V- is the DIGITAL power! This is what it actually says.
You should have 10 uF // 100 nF according to their recommendation. Now, GND
is also digital, and so is actually V+ and V- if you think about it.

> But the question remains whether is it ok to use DGND for the logic GND
> level, but decouple the power rails to AGND. Presumably the digital section
> of the MUX can't cause too much current transients in the power rails or this
> would couple into the analog signals even with external decoupling - so from
> this point of view it seems ok not to decouple to the same ground that the
> digital reference uses.

Well, if my reasoning does not break, all the MAX306 powerlines are "digital",
but in reality we can consider them analog, we want a quite powerfeed if we can
get it. Now, if we run it on analog power (which means V+, GND and V- is tied
to the analog power side), we still want to provide an AC path to the digital
ground. I propose that this is done by tossing a pair of 10 nF caps from the
digital ground, to the V+ and V- but not the MAX306 GND, since this actually
have not much with the actual digital signal. I assume that the digital source
(the PIC in your case) is properly decoupled.

> It's so hard knowing what to do on the PCB .. there's no point even trying to
> measure the effect of using different grounds when the whole circuit is built
> at the moment on a solderless breadboard.  Maybe I should just toss a coin ..
> heads for AGND, tails for DGND :-)

Well... you CAN do that. It's not the GND which is most sensitive one.

Only if you are _REALLY_ picky about exact timing, I would care, since you can
have AC differances entering the GND and modulate the trigger point for the
digital signal input so that you get phase modulations/jitter. I don't think
you have such a problem here, not for a looooong shot.

Hope you got things sorted out, or at least confused at a much higher level ;O)

Cheers,
Magnus