How do I make a seperate analog and digital ground?

I have often told people, (and been paid for doing so), that split ground systems fall into 2 categories: those in which the split does nothing; and those where the split does harm.

By the time you arrange that return currents do not have to flow across your split, you have done most of the work needed to avoid shared impedance coupling. The rest of the job is keeping field coupling below tolerable levels. Once that is done, the split will have no effect in ordinary circuits at ordinary frequencies. The exception is for DC where small potentials are significant.

If you do not see why this is true, I suggest you procure and study "Noise Reduction Techniques in Electronic Systems" by Henry W. Ott. Or hire someone who has and can explain it better than I can here.

The description I like on how to use split grounds (on a PC board) is thus:

Split your ground planes, except for one "bridge" segment. RELIGIOUSLY pay attention to ground return paths for ALL signals, keeping them over the correct plane. ANY signals that gave to cross from one domain to the other goes over the bridge.

Then when you're done make the ground plane one continuous pour.

I'm trying to design a circuit which has both analog and digital power supplies but I'm not sure how to make seperate analog and digital grounds. I assume just making seperate ground planes isn't enough because in the CAE schematic that would merely look like two seperate ground symbols connected to the same net. I think I've seen some schematics where an inductor is placed between the analog and digital ground nets. How does this work?

TIA

Hello Henk,

In my experience it often doesn't.

While it can be done in pretty much any CAD system I found that split grounds are a real hassle when circuits become large and complex structures. Noise coupling becomes nearly unpredictable, EMI certification turns into a nightmare. You connect the grounds at one spot and get noise. You connect them at another and get another kind of noise. When you connect them at multiple places pandemonium starts, hair turns gray, the aspirin consumption goes up. I could go on. Bottomline I have never designed a split ground system in my 20+ years of circuit design and probably won't in the next 20 years. By then I am hoping to retire.

The lone exemption is isolation for safety or code reasons. But then the grounds are usually never connected anywhere and the barrier needs to withstand a few thousand volts. I did design a few of those.

Regards, Joerg

The reason for splitting ground planes is to have the current from the digital devices return to the digital supply and have the current from the analog devices return to the analog supply.

In most cases I draw a circuit diagram of the ground net. Then I determine through which legs large currents are flowing. These legs should not carry any current to sensitive devices. Picture each leg as a tiny resistor, so each leg has a voltage drop across it.

Placing an inductor between ground planes doesn't sound like a good idea because the signals it rejects are superimposed on the power supply rails. Also, ICs which are connected to both ground now have 2 ground levels. If these are too far apart, you may run into trouble. In general, ground should be rock solid.

Hi Larry,

That makes two of us ;-)

Regarding harm: The most horrifying scenario I have seen was where the single point connection jumper between AGND and DGND began to glow. The lab smelled like a camp fire. Then, kapoof, the switcher decided it had enough of this.

Regards, Joerg

Right. Except in the most exotic situations, splitting planes is insane.

John

In article , Larry Brasfield wrote: [...]

William Shatner has acted and been paid for doing so. :>

(Yes I know that was a cheap shot but I just couldn't resist)

I have a copy of said fine work.

Quoting near the bottom of page 58: "The seperate ground system (parallel connection) shown in Fig. 3-7 is the most desirable at low frequencies"

This means star ground or split ground plain. He later suggests that "low frequency" mean less than 10MHz. Less than 10MHz includes most audio and switching supply work.

The book does not actually talk about PCB layout. Neither does Ralph Morrison's "Grounding and Shielding techniques in instrumentation", which is a little light on the math but a bit better on practical issues than the Ott book.

In article , Joerg wrote: [...]

I'll see your wimpy little jumper wire fire and raise you about 6 feet of ribbon cable + about 150K of electronics. It seems that somewhere there is a designer who thinks that fuses should go in the (-) power line on DC equipment but who also thinks that the (-) power should be connected to signal ground.

Probably not; an old tube shortwave receiver had a million times that GBW on a single metal chassis. Exotic would be, maybe, nanovolts at DC. Well, maybe not, since I have done nanovolts at DC (along with a power supply, a uP, and a 10 amp PWM) on a single ground plane.

John

Yes. Mr. Shatner deserves better.

He says "Normally, at frequencies below one megahertz a single point ground is preferable; above 10 MHz, a multipoint ground system is best."

If you look at his diagrams of grounding topologies, you will see no signal connections between the separately grounded circuits. If you look a few pages back (start of section "SIGNAL GROUNDS"), you will see the comment "In the following discussion of grounding techniques, two key points should be kept in mind: 1. All conductors have a finite impedance, generally consisting of both resistance and inductance. At 11 KHz, a straight length of 22-gauge wire one inch above a ground plane has more inductive reactance than resistance."

Now, think for a moment about what happens if those separate circuits pass single-ended signals between separately grounded subsets. The return currents have to travel through the star or the equivalent split plane join point. Those currents give rise to voltage drops along the ground conductors, and for frequencies over 10-100 KHz, depending on what level of interference is significant, B-field coupling becomes significant, unless the return currents have been made to closely follow the signal currents.

People who are enamored with split ground systems generally do not consider B-field coupling, and rarely understand where return currents actually flow when a shared ground plane is used. Typically, they will invoke return current paths distant from the nominal signal path, failing to appreciate how, at higher frequencies, the return currents concentrate beneath the signal traces, unless blocked by splits and such.

I believe Mr. Ott's advice, with respect to what frequencies the split ground schemes can be used, ceases to apply when signals spanning a wide dynamic range are involved.

Morrison's work is mostly good for low frequency circuits. Many of the techniques he suggests begin to be counter- productive above the ~10 KHz transition frequency where conductor reactance exceeds resistance.

One of my burned-in memories involves a system that was put together by a "holy ground point" advocate. It had a magnetically deflected CRT display system, enough digital stuff to store and display captured ECG signals, and a high bandwidth (200 Hz!) ECG acquisition module. Routed to the holy point were hundreds of black wires. It took some time to figure out that the reason it blew its fuses whenever turned on was because of extraneous coupling that arose solely because of that misguided grounding scheme. The deflection power amplifier would give its all replicating a fast signal that seemed to come from nowhere, at least until the actual whole circuit was considered rather than what the holy point advocate had idealized. For some strange reason, those signals would just not content themselves to stay *in* the wires.

The schematic capture software are not handling this well. Either you choose different symbols for analog and digital GND and short them at one point leading to a single identifyable error but is technically autoroutable, or you assign them the same symbol, and do the routing yourself and don't get any errors at all. The digital GND is a copper pour on all layers, while the analog GND is star-connected. Now there usually are signals crossing from one to the other domain. There you have to consider what current is flowing in repect to what and where the return current flows.

Rene

It might not even include a 50 Hz lamp dimmer! The harmonics of the 50 Hz typically extend beyond 10 MHz.

No, it does not include switch mode supplies. I recently had a 600kHz switcher for a LED from 1.2V to 3.5V or so at 20mA. The switching spike repeating every 1.5us or so had harmonics extending far beyond 600MHz as it was just 1.5ns wide. These are exactly those that make the ADCs measure whatever that is not there. And if your design doesn't cope with these 600MHz up, the spike remains there. Filtering the 600kHz amounts to nothing, there is nothing there, just the rep rate.

Rene

Any variation on the input of a SAR ADC will cause it to misread. Delta mod, ADCs are much less sensitive to it. All the noise components from the switcher matter at the ADC.

If your design couples 100mV of 600KHz into the ADC the 600MHz junk won't matter much. You are right about the filter having to cope with the high frequencies but it also has to cope with the low frequencies. The source is the right place to try to stop such things.

No boiler plate rule can substitute for good design. The design of the grounding and power routing is an important part of the design.

In article , John Woodgate wrote: [...]

Move to North America and you won't have to deal with 50Hz lamp dimmers :>

But seriously: I'd expect there to be a whole bunch more 100KHz than

Actually on a mixed PCB, the digital stuff usually forces you to have more layers than the analog needs. This allows you to set aside one layer under the analog stuff as the ground for it. This leaves the digital ground in place to conduct the currents past and an independant ground over it to act as the reference for the analog signals. This works well at frequencies below 1MHz.

In very dense boards where the number of layers is very high, you can set aside three extra layers of power/ground. This lets you have low impedance power lines and grounds within the analog section. There can be as much trouble from AC coming in on the power as on the ground. The decoupling caps run the AC current to the ground near the op-amp. This is exactly where you don't want an AC current injected. Small value decoupling resistors on the power lines coming into the section and low ESR caps to ground are helpful here.

In article , John Larkin wrote: [...]

Would you call an amplifier with a gain of about 1 million at 4KHz "most exotic"

I read in sci.electronics.design that Ken Smith wrote (in ) about 'How do I make a seperate analog and digital ground?', on Tue, 11 Jan 2005:

The spectrum does fall with increasing frequ3ncy up to about 1 MHz, and then stays flat to around 10 MHz, then drops again quite steeply. At least, on those I've measured.