star grounding vs ground plane for DC board

[ftransform]

So I have a dual layer board board consisting of OPAMPS, linear regulators, passives and an ADC only, the micro controller will be connected to this board via a cable.

The board is dealing with DC signals only. I managed to route everything on the top plane and I am wondering what kind of grounding system I should use for lowest noise.

Should I use the entire bottom layer as a big ground plane or should I try to implement star routing centered around the ADC ground? I read some people on audio forums saying that a star ground is better, but other people just recommend using a solid ground plane. I could implement the star routing using thick traces, like 50-100 mils.

[Smokey]

One big thing to keep in mind is you want to keep high current stuff from one section of your board from flowing through your sensitive circuits.  For instance, you don't want the return from your power supply flowing through the same ground traces that feed your ADC and analog section.  Having a star ground system segregates the current paths of each section while keeping them approximately the same potential since they all meet a one point.  There are a lot of ways to do this, with some more practical than others.

The extent to which you need to keep ground current separate depends on how much resolution/accuracy your circuit is capable of.  If you are using an 8-bit ADC with no external reference, or your board is going inside a machine next to a giant switching power supply with no shielding, or in some cases where all currents are really low and signals are slow, then you probably would never be able to tell the difference.  Unless you know you are doing a high sensitivity design and have specifically planned ahead for your error budget and routed accordingly, then in my opinion, you can probably get away with a solid ground plane.  We aren't usually talking huge potential differences in a solid plane here unless the currents are really high or the signals are really fast or you are looking for super accuracy.

One easy thing you can do, which I guess is a variant on single point for the whole system star grounding, is have ground islands for your sensitive circuit sections.  Just keep all the analog and ADC components in their own local ground plane/pour just big enough for the section, and then connect that plane to the outer global ground plane that is surrounding it with just a single point connection of one trace.   Make sure the connection trace can handle the full current of your isolated section and is big enough to not have a significant resistance relative to that current.  That way anything high current in the global plane will presumably go around your island instead of through it, and all your ADC stuff will be at about the same potential.  That works pretty well and still allows you to make mostly solid planes.

This is one of those topics that has a lot of details.  Check out this app note.  It's dense, but full of good stuff.
http://www.analog.com/library/analogDialogue/archives/46-06/staying_well_grounded.pdf

[Gall]

That's simple. Short answer: generally both are incorrect, and the choice depends upon your circuit.

Do not think about "ground". Think about signal return path. Every signal uses two wires.

On DC, think resistors. Every wire is a resistor. If the return current can choose the path, it chooses the shortest one. You cannot keep the resistance "small enough" but you can use separate wires for separate signals to minimize crosstalk. Or you even may use something like Kelvin connection between parts on the board to minimize resistance effect. In most cases this results in well-known "star grounding" but there are exceptions (more than one star and careful elimination of "ground loops"). A notable example is routing of a stereo amplifier board with RCA inputs; the input ALWAYS forms a ground loop, so at least three "grounds" are used to eliminate the problem, each having its own star and no direct connection with others.

On high frequency AC, think inductors. Every signal wire pair forms an inductor. If the return current can choose the path, it chooses the one that follows the forward path (forms a pair). To allow for path choosing, you will want to use a solid "ground plane". Or jist route everything in wire pairs.

Inbetween think both. Remember that a DC circuit may catch AC interference too. Wire pair routing theoretically works for every circuit but is virtually impossible in practical cases (mostly due to the loops caused by power supply connection), so a reasonable compromise should be found.

How sensitive is your circuit?

[ftransform]

That's simple. Short answer: generally both are incorrect, and the choice depends upon your circuit.

Do not think about "ground". Think about signal return path. Every signal uses two wires.

On DC, think resistors. Every wire is a resistor. If the return current can choose the path, it chooses the shortest one. You cannot keep the resistance "small enough" but you can use separate wires for separate signals to minimize crosstalk. Or you even may use something like Kelvin connection between parts on the board to minimize resistance effect. In most cases this results in well-known "star grounding" but there are exceptions (more than one star and careful elimination of "ground loops"). A notable example is routing of a stereo amplifier board with RCA inputs; the input ALWAYS forms a ground loop, so at least three "grounds" are used to eliminate the problem, each having its own star and no direct connection with others.

On high frequency AC, think inductors. Every signal wire pair forms an inductor. If the return current can choose the path, it chooses the one that follows the forward path (forms a pair). To allow for path choosing, you will want to use a solid "ground plane". Or jist route everything in wire pairs.

Inbetween think both. Remember that a DC circuit may catch AC interference too. Wire pair routing theoretically works for every circuit but is virtually impossible in practical cases (mostly due to the loops caused by power supply connection), so a reasonable compromise should be found.

How sensitive is your circuit?

I want it to be as good as possible as it is for a 2 layer board !
My board is a home made ohm meter. which uses a current source and a unity gain differential amplifier connected to a 16 bit ADC with a kelvin connection to the DUT.
The voltages being read are from 0.2 to 4 volts.

Maybe I should have a ground plane for all my OPAMPS and reference and then do "star grounding" on my voltage regulators?
I tried to fit the regulators as close to the opamps as possible (they are low noise linear regulators)

Or should every OPAMP and its assosiated decoupling capacitors/filter capacitors have a ground leading the the ADC ground?

I guess consider it a thought experiment more then a practical project. 

How is a ground handled in a 8.5 digit multimeter?  I just wanna push it to the limit.


And just to be clear, I read through that article but I felt that it did not go into specific about the actual analog grounding system, just about separating digital and analog grounds.
I guess I want more nitty gritty on just the analog section. Should I do a current measurements, see which parts are drawing the most current and then single those out (I.E. the differential amplifier is drawing alot so it should be separated from the inverting amplifiers ground, but since the voltage reference and low pass filter are drawing hardly any current they can be on the same plane? (THIS IS JUST EXAMPLE I HAVE NO IDEA HOW MUCH THESE PARTS DRAW)

[Gall]

Here's the idea of how this could be done.

I intentionally use "worst-case" discrete opamps everywhere. In practice you'll probably want to use readily-made instrumentation amplifiers or even differential input ADCs. I also "forgot" to draw digital power circuit, obvious decoupling capacitors and local feedback of IC4A. Also extra capacitors in feedback circuit may be needed to ensure stability.

Here I use two star connections - C1 and C2. Each capacitor should be as small as possible so that it could be center of the star. Everything else is just Kelvin-connected. And if I draw two wires like a pair, they should be routed like a pair.

[ftransform]

Hmm, what I did so far is I made separate planes for the linear supply regulators, the current source, differential amplifier/ADC and a the wire shield+current return from the DUT.

 I joined these planes together at the ADC ground. Does this arrangement make sense?
To me it makes logical sense that it is most imporant to isolate the differential amplifer and gain stage from everything else.
I have trouble understanding the diagram/logic above.

my ground plane looks somewhat like this:


Is this reasonable?
Keep in mind the ADC communicate with the digital board through an opto isolator.

[Gall]

Most likely that's not very good. A solid plane has multiple paths for the signal so it's hard to put the exact current path under control.

The idea of proper precision circuit routing is choosing a reasonable reference point for each signal. The reference point is not always ground.

The best you can do in your case is to make true differential inputs everywhere and make pseudo-diffrential connection between ADC and differential amplifier. A typical precision amplifier has an "output reference" pin. This pin should not be grounded directly. Instead it should have a separate trace to ADC's AGND so that this trace fotms a "differential" pair with ADC input wire. The best you can do is to trace ADC IN and diff amp OUT_REF in exactly the same shape on different sides of the board.

It does absolutely not matter how GND is routed provided that OUT_REF is routed correctly.

[ignator]

My board is a home made ohm meter. which uses a current source and a unity gain differential amplifier connected to a 16 bit ADC with a kelvin connection to the DUT.
The voltages being read are from 0.2 to 4 volts.

Maybe I should have a ground plane for all my OPAMPS and reference and then do "star grounding" on my voltage regulators?
I tried to fit the regulators as close to the opamps as possible (they are low noise linear regulators)

Or should every OPAMP and its assosiated decoupling capacitors/filter capacitors have a ground leading the the ADC ground?

I guess consider it a thought experiment more then a practical project. 

How is a ground handled in a 8.5 digit multimeter?  I just wanna push it to the limit.


And just to be clear, I read through that article but I felt that it did not go into specific about the actual analog grounding system, just about separating digital and analog grounds.
I guess I want more nitty gritty on just the analog section. Should I do a current measurements, see which parts are drawing the most current and then single those out (I.E. the differential amplifier is drawing alot so it should be separated from the inverting amplifiers ground, but since the voltage reference and low pass filter are drawing hardly any current they can be on the same plane? (THIS IS JUST EXAMPLE I HAVE NO IDEA HOW MUCH THESE PARTS DRAW)

What dynamic resistance measurement response are you expecting?  i.e. steady state resistance, or fast changing samples.  Is the circuit your measuring from powered and generating noise, that will be coupled to your measurement board?  If your low pass filtering using capacitors to ground, capacitors are marvelous at coupling noise to ground, but if the ground is noisy, they do just the opposite, couple noise back to the signal.  You need to keep noise from ever getting on the ground in the first place. 
16bit ADC, this is on the board, is a serial bus used to get the digital conversion off board?  ADCs make noise, and transfer this to the power connections.  What is your LSB bit voltage?  How does the ADC work successive approximation? It's internal state machine will generate noise.  And I've seen Analog device parts dynamically yank the buffered input all over the place, as well noise on the power connections.
Is this bi-polar, or are you working single ended with ground?
Think of how you can keep the digital noises off the analog signal.  That can mean the connections to the digital ADC has a series inductor, capacitor to digital ground, and on both the Vcc/Vss pins.  If this is sent out on a serial bus, the driver has to charge the transmission line capacitance, this causes little noise pulses to be emitted by this part to the power rails.
You may have a perfect system, but the ADC may bite you in the end with it's idiosyncrasies.  Can you use opto isolator on the remote processor connections?

[ftransform]

My board is a home made ohm meter. which uses a current source and a unity gain differential amplifier connected to a 16 bit ADC with a kelvin connection to the DUT.
The voltages being read are from 0.2 to 4 volts.

Maybe I should have a ground plane for all my OPAMPS and reference and then do "star grounding" on my voltage regulators?
I tried to fit the regulators as close to the opamps as possible (they are low noise linear regulators)

Or should every OPAMP and its assosiated decoupling capacitors/filter capacitors have a ground leading the the ADC ground?

I guess consider it a thought experiment more then a practical project. 

How is a ground handled in a 8.5 digit multimeter?  I just wanna push it to the limit.


And just to be clear, I read through that article but I felt that it did not go into specific about the actual analog grounding system, just about separating digital and analog grounds.
I guess I want more nitty gritty on just the analog section. Should I do a current measurements, see which parts are drawing the most current and then single those out (I.E. the differential amplifier is drawing alot so it should be separated from the inverting amplifiers ground, but since the voltage reference and low pass filter are drawing hardly any current they can be on the same plane? (THIS IS JUST EXAMPLE I HAVE NO IDEA HOW MUCH THESE PARTS DRAW)

What dynamic resistance measurement response are you expecting?  i.e. steady state resistance, or fast changing samples.  Is the circuit your measuring from powered and generating noise, that will be coupled to your measurement board?  If your low pass filtering using capacitors to ground, capacitors are marvelous at coupling noise to ground, but if the ground is noisy, they do just the opposite, couple noise back to the signal.  You need to keep noise from ever getting on the ground in the first place. 
16bit ADC, this is on the board, is a serial bus used to get the digital conversion off board?  ADCs make noise, and transfer this to the power connections.  What is your LSB bit voltage?  How does the ADC work successive approximation? It's internal state machine will generate noise.  And I've seen Analog device parts dynamically yank the buffered input all over the place, as well noise on the power connections.
Is this bi-polar, or are you working single ended with ground?
Think of how you can keep the digital noises off the analog signal.  That can mean the connections to the digital ADC has a series inductor, capacitor to digital ground, and on both the Vcc/Vss pins.  If this is sent out on a serial bus, the driver has to charge the transmission line capacitance, this causes little noise pulses to be emitted by this part to the power rails.
You may have a perfect system, but the ADC may bite you in the end with it's idiosyncrasies.  Can you use opto isolator on the remote processor connections?

I am measuring steady state resistance (it is not something like trying to graph  the impulse using a weigh scale). Since this board uses an opto-isolator to communicate with the processor it will have its own transformer. The full scale of the ADC will be 10V, making the LSB 0.00015258789 volts.

The ADC I am using is powered by a 5V (not bipolar supply like some other ADC's that I have), and it has its own low noise linear regulator right up against it. This linear regulator does not draw off the 15V rail but down converts from the less regulated power. The ADC will communicate off the board using an optical isolator. The OpAMPS are powered by a BIPOLAR supply.

I still need to wrap my head around what gall is saying because my understanding of circuit analysis is poor.... my amplifier does have a reference pin but on my PCB I put the amplifier, so I will give that a isolated ground trace directly to the ADC's ground.

I don't understand what you mean by fully differential inputs everywhere. Are you saying that I should just give the differential amplifier an isolated ground trace to the ADC's ground (its like 3 cm away) and get rid of the ground splitting that I have in my little picture and use one big ground plane? Or should I keep that and just add the isolated trace?

Won't the 5V regulator providing the power to the ADC mess with the rest of the circuitry if there is no barrier put between it and the "more analog stuff", considering that the ADC draws power in bursts when it does its calculation?
Also won't the shield cable return mess up the ground for the other components if some nasty stuff is picked up?
Why does my ground splitting idea not work?

[Gall]

The principle is simple: the amplifier has output reference pin, the ADC has input reference pin. Connect them together with a separate trace, and they'll stay always at the same potential. If you worry about AC crosstalk, make this trace of the same length and shape as the corresponding signal trace. Use the same approach to connect ADC VREF to the precision reference source and inside the current source circuit.

Everything else is differential, so no grounding issues at all regardless of the routing. Power ripple does not matter too since the common mode rejection ratio is probably very high.

Just do not think "ground". Think "reference points". There are not many reference points in your circuit. They do not have to be the same as ground.

[Neilm]

The principle is simple: the amplifier has output reference pin, the ADC has input reference pin. Connect them together with a separate trace, and they'll stay always at the same potential. If you worry about AC crosstalk, make this trace of the same length and shape as the corresponding signal trace. Use the same approach to connect ADC VREF to the precision reference source and inside the current source circuit.

And if the return is via the power pins? Signals flow in a circuit. I've seen a design that had done something similar carefully done with the references. The output of the circuit was incredibly sensitive to the slightest bit of EM noise and the readings went all over the show. The solution was to stitch the "reference" to the ground layer. 

I remember reading that 95% of the return current will flow under the signal trace as this is the path of least resistance.

Neil

[Gall]

And if the return is via the power pins

Just do not design like that if you can. The best solution in this case is to "break" power and/or ground wires using inductors or resistors to prevent the signal going through them. Let the signal always go via the dedicated wire.

A notable exception is the case if the signal is very powerful, i.e. in a precision high-power amplifier. In this case a separate "sense" connection is required to compensate the parasitic resistance, inductance and crosstalk of the power wires.

I remember reading that 95% of the return current will flow under the signal trace as this is the path of least resistance.

Of least INDUCTANCE.
On DC, the return current chooses the "shortest" path (straight path in a solid layer or thickest wire).
On HF AC, the return current follows the same path as the forward current.
On LF AC, it is inbetween and thus almost unpredictable.
A straightforward solution for lower frequencies is to etch a return trace exactly under the signal trace. This works well for both AC and DC. But one still has to avoid using the same return trace for two different signals (especially if one of them is strong).