Under what circumstances would it be necessary to implement more than one ground in a single circuit? For example, in addition to the usual 0V ground, you would have other grounds, possibly on other boards within the same design, that have non-zero voltages on them WRT to the main 0V ground?
Ground is whatever you want it to be. We have a setup somewhere around here that has a rack-full of electronics sitting at +100kV. The stuff inside the rack isn't even aware that its ground potential isn't the same as the local earth.
Of course, you have to make sure that any interface between that and local earth will withstand the potential difference, so getting signals and power across is a bit special.
To keep life simple, you'll want *one* ground inside a piece of electronics. Anything connected to the outside will somehow have to be translated to that ground. This may be very easy and simple, say, like a digital line receiver, of intermediate complexity, like differential line receivers, isolation amplifiers, transformers or opto-couplers, or downright cumbersome, like motor-generator sets with long insulating rods between them.
You should realize that for certain applications, notably RF, high-current or precision circuitry, there is no such thing as simple 'ground'. Everything moves, even if it's solidly connected together.
We do multi-channel analog input and output boards with isolated channels. Each channel has its own circuit common, which sure looks like "ground" on a schematic sheet.
formatting link
Some people like to have separeate analog and digital grounds, connected at a single point. That's usually a bad idea.
One of our customers made us float our interface circuit to their system. On our board, the ground of that circuit is isolated by optocouplers and dc/dc converters, and they furnish *their* ground to us via a connector pin. It's silly, but they insist.
--
John Larkin Highland Technology, Inc
picosecond timing precision measurement
There's a lockin I made (way back when), the reference oscillator output is on an isolated BNC and get's a "single point" ground back at the common power supply. If not isolated, I get some currents flowing through "case" ground and ~10-100's of uV of "signal" coupling into things. (This was before I discovered SED, so maybe there a better way.)
To be honest I still see some coupling (capacitative?) at the highest frequency from the oscillator into the front panel "case" ground.
There are many cases where multiple grounds are used. Basically, where one wants to keep currents local a separate ground is used.
Some places I use local ground islands:
- Switching regulators: the grounds of the input capacitors, regulator (switch/power ground) and output capacitors are connected together on an island and this is then stitched to the overall ground under the regulator, if it has a power pad. The capacitors are oriented so that their ground connections are close to each other.
- Microphone input circuits: A local ground is used for the bias circuits, any low level signals, and the analog ground pins of the ADC. Again, this is stitched to the overall ground in one place, often under the ADC, though sometimes near the connector ground pin.
- Analog grounds: Any low level signal grounds where these are sharing a board where there are also high current signals. Again, this island is in addition to the overall "digital" ground.
The bottom line is that any additional grounds are in addition to the overall ground plane. The ground plane never gets cut and care is taken such that vias don't form cuts in this plane. It's nice if all digital signals can be routed on a layer very close to this plane, as well.
For things like switchers or microvolt amps, we sometimes make a C-shaped cut in the ground plane for that circuit. That works well for
4-layer boards where we don't have a spare plane layer for private grounds.
I guess you isolate a part of the ground plane, a little square maybe, and couple it to the main ground with ferrite beads or something. I haven't done that.
--
John Larkin Highland Technology, Inc
picosecond timing precision measurement
The RF guys do that but they don't often have high current devices on their boards (or at least at that end). I don't like cuts of any sort on the overall plane.
No ferrite bead. It gets staked directly to the ground plane in one place. For switchers, the input and output caps and the power ground pins connect to a pour on the IC side of the board, so the only vias connect this plane to the overall plane. This pour is also the heat sink (though the plane helps too) so is as large as practical.
If I deviate off on a tangent at this stage I'd never get the info I need, so I may return to that aspect later but at the moment I'm getting some excellent responses - and the more the better!
Any signal ground carries some return current. Ditto, any power supply common conductor. It is useful to move one's ground connections around, in order to reduce the conductor impedance (inductive or resistive) contribution to sensitive signals, and sometimes that means clean-ground/dirty-ground decisions.
LVDT signalling is beneficial because it makes ground-connection flaws irrelevant. Balanced wiring in microphones, PCIe, USB, Firewire, and all the way back in history to Futurebus and RS-485 serial, are intended to get around ground flaws (and they all work).
"Ground" is a common connection - usually to 0V, though as Jeroen Belleman points out, it doesn't have to be.
As such, there is only one "ground" in any one circuit. I designing how thi s "ground node" is laid out as an extended conductor , you have to keep tra ck of what currents are flowing through it - where they enter the conductor and where they leave it - and the corresponding voltage drops due to the r esistance and the inductance of the path traversed.
"Single point" or "tree" grounding tries to prevent the voltage drops from one part of the circuit from disturbing the operation of other parts of the circuit, and it's common to separate "digital" and "analog" grounds for ex actly this reason.
One of my more productive bug-fixing exercises found that the precision vol tage reference setting the current through the final lens on an electron mi croscope was "grounded" at the wrong point on 0V connection that was carryi ng some 100mA in PROM supply current. There was a 100mV drop across the rel evant bit of track, which subtracted from the reference voltage.
The quick and dirty fix was to solder some heavy copper wire onto the groun d track. The longer-term solution was to separate the relevant ground retur ns.
If there has to be a cut in the ground plane then there should be a rule that no traces that carry any AC current or that are susceptible to induced voltage are permitted to pass over (or even near) the cut. That pretty much means no traces at all are allowed to cross over the cut.
This could be nice, provided the circuits on the "leaves" of the tree do not communicate with each other through signal wiring that is sensitive to induced voltages or carries any AC current, and provided also that there is very little electromagnetic field in the vicinity of the instrument.
In situations where the "leaf" circuits are connected together with signal wires, and the system must pass EMC tests and work outside a shielded chamber, a tree ground will form all sorts of loop antennas unless every signal wire from each "leaf" is routed all the way back to the "trunk" of the ground tree before it goes to any other "leaf", so that there is very little loop area between the signal current and its corresponding ground return path.
IOW a ground plane is much more practical in most cases.
If you think it would be a good idea to have other ground traces on top of the ground plane, especially multiple stacked ground planes connected at one magic "star point", then watch out! It can form a very high Q resonator. e.g. in my youth I was asked to build a prototype board to connect up a RF amplifier chip (in a TO-3 style package but with more pins, I think from Avantek or one of those companies). I bolted the package body flat onto some double-sided copper clad FR4 with the pins sticking through clearance holes, and to "solidly connect" the copper "ground" planes on both sides of the board, I wrapped copper tape over all edges of the board and soldered it on both sides. Of course where the RF pins of the amplifier passed through the FR4, they were passing through the high impedance part of a high Q cavity resonator. The thing oscillated like a banshee, at about a gigahertz. I spent some time wondering why it didn't work at low frequencies, and why the scope trace was slightly fuzzy, until someone lent me the spectrum analyser to see the oscillation. I guess I was about 15 at the time so that is my excuse. If you really must have multiple planes, at least stitch them all over with vias, close enough together that the impedance between the planes will be low all the way up to the highest frequency where any device has gain. Maybe putting resistors in series with some extra vias would help to deaden the resonances too. The situation could be similar when you have a power plane and a ground plane, only the DC voltages are different. Some fraction of the bypass caps could have resistors added in series to deaden that also.
w this "ground node" is laid out as an extended conductor , you have to kee p track of what currents are flowing through it - where they enter the cond uctor and where they leave it - and the corresponding voltage drops due to the resistance and the inductance of the path traversed.
rom one part of the circuit from disturbing the operation of other parts of the circuit, and it's common to separate "digital" and "analog" grounds fo r exactly this reason.
Sure. You tend to track a signal and its associated ground as paired tracks , side-by-side or as stripline or microstrip if on different layers.
You need to work out the characterisitc impedance of the transmission line involved (which tends not to be wildly different from 50R - twisted pair te nds to come out from 90R to 150R)
I had a learning experience in the early 80's, when I built an instrument with 100mv FS chart recorders and a 10 amp motor drive circuit. Every time I turned the motor on, the chart recorder went FS. Deeply imprinted lesson. Another lesson from High School machine shop class, a student using a file on a lathe, the jaw on the chuck hit the file and pushed it, under the skin of his forearm.
Mikek,
PS. We had a great machine shop in my HS, about 10 big lathes, 4 milling machines, several drill presses, Tool Crib, Foundry, Shaper, torches, welders. My son's school didn't even have a wood shop, let alone a machine shop. Times change, I guess. :-(
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.