I have long had a conviction that I want to check: "Signal grounds should always be build in a strict tree structure, because that makes ground loops impossible." Am I correct?

 

Are you referring to Star point ground?
There is also a method called equi-potential bonding

 

You avoid ground loops by not connecting two grounds together. This is typically a problem found with house wiring when a signal source and an input are connected together and plugged in to different outlets.

A tree does not avoid other problems. You do not want any high current trace shared as the ground for anything else, hence the star configuration.

 

I would consider the opposite?
Earth bonding is where different or separate earth planes are connected/bonded together as shown in the Siemens Pub.
Also equi-potential bonding can now eliminate the need to just earth one end of a shielded cable, As seen in CH 6, both ends of a shielded cable can be connected using this method, which now eliminates EMI radiation.

 

I have long had a conviction that I want to check: "Signal grounds should always be build in a strict tree structure, because that makes ground loops impossible." Am I correct?

Not entirely correct. One cannot make such as blanket statement without looking at the specific application, e.g. magnitude and frequency of currents, mixed signal application, physical design and layout of components, etc. One can still have ground loops in a solid piece of copper plate.

 

Ok, this was just a problem of terminology for me then. There's no "star" in my background (math). What you're calling a star looks like a tree to me -- a tree of height 1.

Thanks, this explains some confusing conversations I've had. Interdisciplinary engineering catches me out on vocab all the time. Don't get me started on i and j. (Current? Imaginary numbers? Unit vectors?)

 

I would consider the opposite?
Earth bonding is where different or separate earth planes are connected/bonded together as shown in the Siemens Pub.
Also equi-potential bonding can now eliminate the need to just earth one end of a shielded cable, As seen in CH 6, both ends of a shielded cable can be connected using this method, which now eliminates EMI radiation.

Ok, this was a critical factor I was missing as well. It's "looping" of the shielding connections that I was actually concerned about in this harness design I'm looking at.

 

Not entirely correct. One cannot make such as blanket statement without looking at the specific application, e.g. magnitude and frequency of currents, mixed signal application, physical design and layout of components, etc. One can still have ground loops in a solid piece of copper plate.

I'm aware of the edge cases. I should always say "generally not" and never say "never". Especially when engineers (read "pedants") may be present. And I'm not criticizing. I usually am that pedant.

Strictly, I should have said "Generally, you should arrange your grounds in a tree structure because that eliminates the production of ground loops purely as a result of wire connection topology." And I appreciate the drive to be more precise (and thus accurate).

 

Welcome to AAC!

I have long had a conviction that I want to check: "Signal grounds should always be build in a strict tree structure, because that makes ground loops impossible." Am I correct?

I've never heard of a tree ground structure.

A star ground is used for high current situations where voltage drop in ground traces/wires could affect circuit operation. For signals, you design in enough voltage margin so that variations in ground voltage won't affect correct operation.

 

Ok, this was just a problem of terminology for me then. There's no "star" in my background (math). What you're calling a star looks like a tree to me -- a tree of height 1.

In practice, Star GND generally refers to a common earth termination point in a control panel, for example, where all grounded circuits end up terminating, this together with the service GND.

 

A star topology, but also being careful that the 0V/Ground/Earth/Sheild/Screen is not connected twice to any separate subsystem. That is, if you are distributing subsystems and using shielded cables to connect them together, the shield should be bonded to the 0V point on the output side and not connected at the input.

It should run all the way to the connector and then stop. As you can see, if you do connect these shields you are creating loops that could be at different potentials, leaving them disconnected doesn‘t affect their efficacy as shielding, but it stops them from being another path to ground that is not at 0V relative to the device they are being connected to.

 

I have long had a conviction that I want to check: "Signal grounds should always be build in a strict tree structure, because that makes ground loops impossible." Am I correct?

If you are referring to ground layout for PCB's, it is called a "star" ground layout. It is a ground trace topology where all ground traces originate from one central point. Star grounding really isn't necessary with the proper use of "ground planes", but there are exceptions where separate ground planes are used but are interconnected at a central point. If you call this a "tree structure", no one will know what your referring to, as "star" is commonly used in the industry.

 

Just to reiterate what everyone else has said, we don't use the word "tree" grounding. "Star" is the word that is commonly used.

 

A star topology, but also being careful that the 0V/Ground/Earth/Sheild/Screen is not connected twice to any separate subsystem. That is, if you are distributing subsystems and using shielded cables to connect them together, the shield should be bonded to the 0V point on the output side and not connected at the input.

Initially the method was to only connect shielded cables at one end of the shield, one problem was that the open end shield acted as an antenna !
With the later advent of instituting equi-potential bonding between separate areas of a system, where GND loops were eliminated, it was recommended and made possible to earth GND both ends of the shield of a shielded cable.

 

Initially the method was to only connect shielded cables at one end of the shield, one problem was that the open end shield acted as an antenna !
With the later advent of instituting equi-potential bonding between separate areas of a system, where GND loops were eliminated, it was recommended and made possible to earth GND both ends of the shield of a shielded cable.

I built many recording studios using the method I described. So long as the shield is well bonded to the output side device 0V point, and that device is well bonded to a good earth ground, there is no issue with RF induction. This also relies on good, low impedance shielding on the cables.

But, you are correct that if you can ensure and equipotential state for all devices, the open shield is redundant. Since we often had no control over the internal arrangement of the device shielding and connections, we couldn't be sure the ground connection of the jacks would be equipotential.

 

I built many recording studios using the method I described. So long as the shield is well bonded to the output side device 0V point, and that device is well bonded to a good earth ground, there is no issue with RF induction. This also relies on good, low impedance shielding on the cables.
But, you are correct that if you can ensure and equipotential state for all devices, the open shield is redundant. Since we often had no control over the internal arrangement of the device shielding and connections, we couldn't be sure the ground connection of the jacks would be equipotential.

Most of my experience has been with industrial electronics, mainly machine control, CNC etc.

 

I have long had a conviction that I want to check: "Signal grounds should always be build in a strict tree structure, because that makes ground loops impossible." Am I correct?

It all depends on what is important in the particular application. Ground loops are just one consideration. You also have crosstalk due to reference voltage rejection (also known as ground bounce) and controlling EMI by minimizing image current loops. To a significant degree, these are competing effects. You can combat reference rejection using a star configuration, but that is all but sure to really aggravate the EMI issues in high speed designs, which are best addressed by having a highly distributed grounding scheme (ground plane being ideal).

Fortunately, these are generally at opposite ends of the frequency spectrum, so you can use ground planes that split and laid out in a star configuration for power but then tie those planes together, and also to the power planes, with stitching capacitors to create a distributed ground at high frequencies. Putting stitching capacitors close to any place where signal traces cross splits or change layers will then keep the image currents close to the signal currents, thus reducing the effective loop antennas for either EMI emission or susceptibility.