There is a lot of old bad information about doing single-point grounding.
This is partly from historical relevance, and partly from lack of understanding (AC is more complicated), and poor communication (poor explanations, poor availability of them, etc.).
Back in the vacuum tube days, you might only be concerned with 10s of kHz of bandwidth in an audio circuit; signals very much "stay confined to wires", and routing really doesn't matter. In an RF circuit, low frequency ground-loop error is largely irrelevant, and grounding issues tend not to show up -- or rather, they are experienced as shielding issues, inadvertent coupling between sections; and subsequently, oscillation, or images or spurs in the receiver response.
This seems like ancient history, but information filters down slowly through the decades, and many things that get learned, get repeated -- without knowing what assumptions underlie that advice, or what exceptions they may entail.
So it is, we have many otherwise-experienced practitioners to this day building nightmare circuits, and with the ever rising bandwidth of semiconductors, it gets harder and harder to ignore these effects. (The most egregious example of this, I think, is probably in the audiophile community -- one built simultaneously upon tradition and novelty, mining the literature, both old and new, looking for every unturned rock; iterating their designs, incorporating components both old and new, trying to improve the experience. The central conceit, is that we can create objectively near-perfect amplifiers; the search is therefore subjective, and what's being evaluated is not the electrical system thus designed, but ones' experience of it, not just the sound but the experience, indeed even the story of how a given item came to be.)
(And to be clear, while this may sound rather critical, harsh, about one group in particular; I temper that by noting, it's simply the human experience. We all do it, to some extent or another, and even the most focused among us, must fall back on some unchecked assumptions, intuitive or emotional feel, or societal "givens", to handle the sheer number of things we interact with throughout the day. You can find such elements -- subjective experience, superstition even -- in any community, in any topic no matter how technical. Indeed, recognizing this behavior I would say is a key critical-thinking skill.)
So where are we really coming from?
To be clear: star grounding works perfectly fine at DC, where the voltage drops along wires are evident (proportional to current flow and resistance), and keeping them common to a point limits the voltage drops along any given branch to only the current that flows in that branch. Which will be small for most loads, thus the star point can be used as an effective reference.
At AC, the voltage drops along those wire links can be almost arbitrarily high. The impedance of a length of wire rises with frequency, and if you go high enough in frequency (100s MHz, GHz), even a short length (10s cm) can have an impedance of 100s ohms, perhaps even kohms at the peaks.
100MHz may seem high, but it's a very ordinary harmonic for digital logic these days, and SMPS regularly teeter into that range (or work actively within it, as is the case for GaN power supplies with single-ns risetimes!). Transistors can easily oscillate in that range, or ICs with insufficient supply bypass.
The effect is like trying to choreograph an energetic dance on top of a floor of springs and jello.
Clearly, this will not work out in general. We need a more solid ground reference. A reference plane.
Foundationally -- as a basic premise -- we want a ground plane under all traces, so that subsequent assumptions remain valid.
At high frequencies, a signal current (flowing in a trace), induces an image current in the conductors around it (namely, ground plane). This follows the path* of least impedance (the path under the trace), rather than the path* of least resistance (which is taken at DC).
Note that "DC" is a context-dependent concept; for typical PCBs, current flow distributes across the plane for some kHz or below, and so for purposes of current distribution, we might consider frequencies below this cutoff as "DC". Conversely, the inductive path dominates above a MHz or so.
*Incidentally, "the path of least resistance" is a historical misnomer at best; it should really say "distributed according to resistance of the respective path"; or alternately we understand that "path" is a superposition of all possible paths current can take, and thus a distribution is formed -- not that *the singular path* with the lowest resistance becomes the exclusive path of current flow, that would be... weird.
With ground plane under (or around) all traces, we obtain a transmission line structure, thus we have some well-defined propagation velocity and characteristic impedance for every path on the board. This can be used to our advantage, maintaining signal quality, or even constructing filter circuits (typically used in the GHz).
Since image currents flow under the respective signal paths, if we keep signal paths separate, we avoid intersecting current loops, and thus avoid introducing interference between loops.
We don't have to completely discard the concept of a "star ground". But we must modify it to be compliant with the above requirements.
We can have a star topology where there's a central hub, across which all signals are routed (above GND plane), and the spokes are individual sub-circuits (say, a given IC or group of ICs, and related components), laid out on extensions of that ground plane. DO NOT route signals directly between spokes: in general, the space around the ground plane is full of nasty EMFs (external noise sources, common-mode interference from other current loops within the circuit), and routing a signal through that space induces those noises directly in series with it.
Actually going to the trouble of designing a hub-and-spokes PCB, isn't very convenient, or helpful or even meaningful really, and we can morph the topology into a rounded blob (or perhaps more convenient for manufacturing, a rectangle -- which is still "rounder" than a star shape, you'll agree) without significant impact on the EMC (electromagnetic compliance) and signal quality of the design. Indeed it might even be better, as there aren't spokes acting as antennas; the impedance of a blobby board in space is much lower than would-be spokes (at their resonant frequency(ies)), so it's less likely to couple into ambient fields.
For purposes of something like a DAQ system, the analog section can simply be laid out in one region, and supply filtering applied at the regional boundary if necessary; the DAC/ADC placed on the boundary; and digital signals entering/exiting from the other side. Do not route traces across the boundary, unless filtered, to slow enough risetime that the resulting displacement current is small with respect to nearby current loops (thus, minimizing interference even at high frequencies). We thus construct a conceptual boundary, not a real physical one; and ground should be poured solid across the board, including under the boundary.
(At least for signal quality purposes, we could gather traces together which cross this boundary, and cut slots where traces are not, with little impact on behavior or performance. We do, however, introduce more resonant modes by cutting slots -- again, a spikier, concave-ier shape has more antenna quality than a bulky circuit. There may be sneak paths that couple between ambient fields and internal currents or signals -- supply ripple, or logic signals, or RF, might radiate or receive interference in this way, and it's generally better to avoid those opportunities, by not slotting the plane.)
Tim
Home
Help
Search
Login
Register
