8
\$\begingroup\$

I am still new to PCB design but I often run into this.

Say I have a USB connector on the bottom of a board, a SoC on the top, an LDO somewhere in the middle, and miscellaneous stuff on the sides. It's often not clear the best place to tap into VCC for everything while I'm still moving things around and I end up connecting it in multiple places back to the source, creating a 'ring of power.'

Loosely related, I've found when running long runs of LED strips that effectively form a loop, it works better to connect both ends to +ve and ground. Otherwise the furthest end of the LED strips have more resistance and they get dimmer towards the end. Does the same thinking apply?

My approach thus far has been "it can't hurt," but can it? Should there just always be only one path?

\$\endgroup\$

5 Answers 5

Reset to default
10
\$\begingroup\$

Having low impedance on the power rail is always a good thing. So multiple power paths aren't bad in general. However a star-pattern or power plane can be better if emi is a concern. Any ripple caused by sudden power draw can propagate to other parts better if you only have a single "circle" line. What you may also watch out for is proper blocking capacitors so you don't get any ripple into your components.

\$\endgroup\$
1
  • 1
    \$\begingroup\$ okay good to know. Every IC has accompanying 0.1uf, and for bigger things also 10uf positioning right next to the IC. I know that much from inspecting other designs and reading data sheets, but I'm still ignorant to nuance with these things. \$\endgroup\$
    – Kez
    Commented Oct 26, 2023 at 8:31
10
\$\begingroup\$

A related peripheral issue worth being aware of:

If you have analog measurement points on a board, whether internally sourced or from external sources, then at the signal levels usually dealt with by uC ADCs substantial error can occur if earth reference voltages are not meticulously managed.

This often requires

  • Microcontroller grounds

  • Signal source grounds to be star connected to a common point

  • Taking exceptional measures to ensure that unrelated signals do not flow in the ground paths.

Example:

On a PCB with onboard aDC (maybe in a microcontroller or separate) Measure the voltage between eg power supply input ground point and various ground points on the board. An oscilloscope allows you to see AC and DC.

As the circuit operates note how the voltages do (or don't) vary.
If you are driving eg a relay you may see some mV or maybe 10's of mV difference. A 10 bit ADC with a say 2.048V reference has 2 mV per bit resolution. Use it on a circuit board without due care and you will easily reduce it to 9 or 8 bits effective. Maybe even worse. Ask me (and many others) how we know :-).

Comment on these claims welcome.

\$\endgroup\$
16
  • \$\begingroup\$ thanks for that tip. I am indeed playing with some external ADCs, but they appear to work just fine bread boarded or on a home milled PCB laid out without much thought. I am however using really fat traces so I guess that helps. \$\endgroup\$
    – Kez
    Commented Oct 26, 2023 at 8:25
  • \$\begingroup\$ @Kez Such things work well sometimes always and sometimes while Murphy is looking the other way. Measure the voltage between eg power supply input ground point and various ground points on the board. An oscilloscope allows you to see AC and DC. As circuit operates note how voltages do (or don't) vary. If driving eg a relay you may see mV yo maybe 1's of mV difference. A 10 bit ADC with a say 2.048V reference has 2 mV per bit resolution. Use it on a circuit board without dur care and you will easily reduce it to 9 or 8 bits effective. Maybe even worse. Ask me (and many others) how we know :-). \$\endgroup\$
    – Russell McMahon ♦
    Commented Oct 26, 2023 at 9:40
  • 1
    \$\begingroup\$ I don't really understand why everyone speaks of "star" layouts - in year 2023 nobody is using 2 layer boards any longer. At a bare minimum 4 layers with at least 1 GND layer in the stack-up. Which in turn means you'll have a GND copper pour all over the board. And the way to deal with sensitive parts (like RF or voltage regulator power ground) is to have a separate copper pour area for those things and connect it at a single point to the main GND near the GND connector of the board. I'm not seeing any stars :) What am I missing? \$\endgroup\$
    – Lundin
    Commented Oct 26, 2023 at 11:45
  • 1
    \$\begingroup\$ @Lundin TBD in my case. I'm about to implement a solar powered latrine lighting system for use in regfugee camps. It needs to work enough magic to look after itself in all sun conditions, plus monitor usage and state of usability, plus provide (probably) mesh radio connectivity to other latrines (who would have thought ? :-) ), plus cellphone network interface for management and more - and cost under about $US14. Every $ corner that can be cut will be - while still working brilliantly (pun noticed) and super reliably. If N layers cost little more compared to other things they will happen. \$\endgroup\$
    – Russell McMahon ♦
    Commented Oct 27, 2023 at 11:44
  • 1
    \$\begingroup\$ ... design very careful due diligence is required. VERY> Components are usually genuine. Their are out of spec risks/failed test parts/ ... . It's doable and most and maybe all mayve be genuine. I can (probably) get about 1 km range regulatory acceptable mesh modules for around $US1. LORA transcievers for under $US2 and GPRS 2G (cellphone) modules for under $US5. (These can be gateways to a mesh system so cost is spread). 2G is being turned off in NZ next year :-( - likely available in eg Africa for some while yet. 4G modules are dearer but still affordable. || \$\endgroup\$
    – Russell McMahon ♦
    Commented Oct 28, 2023 at 0:36
3
\$\begingroup\$

A closed rail loop is a nightmare for signal processing since any magnetic flux changes through it (and normal household environments have magnetic fields of the 50Hz or 60Hz variety, causing hum in all audio equipment) will induce a circular voltage leading to a circular current offsetting the magnetic flux change. That makes the circular rail useless as a reference.

Now naturally audio and measurement circuits are worst affected by this: the solution is to keep ground reference tracks parallel to the corresponding signal tracks without introducing loops, and to use solid ground pours for voltage supply and bypass capacitor purposes.

A ring architecture is not really good as a ground pour; a criss-crossed area works reasonably with reduced copper requirements.

So the answer is: for many purposes a ground loop can and does hurt. An LED strip is not really concerned with signal processing; whether it makes a difference to your holiday season decorations when the central tree is hit by lightning is debatable.

\$\endgroup\$
13
  • \$\begingroup\$ But unless you can get rid of all ground loops (and you often can't), it's better to have more of them. The worst case is a single ground loop. "Let ground abound" is usually good advice. \$\endgroup\$
    – John Doty
    Commented Oct 27, 2023 at 11:12
  • \$\begingroup\$ "Necessary" ground loops come about because of multiple channels with grounded inputs joining, and amps having grounded outputs. But most amplifiers these days use opamps, and a summing/inverting opamp with a "grounded" input can get its reference ground from input and and output grounds in the exact same proportion as on the positive input, but with resistances in the order of 10ohms. That puts enough resistance into the ground loops to let basically all of the ground loop voltages end up at the resistors forming the virtual ground point, and no net hum survives in the differentials. \$\endgroup\$
    – user107063
    Commented Oct 27, 2023 at 11:47
  • \$\begingroup\$ Single-ended interfaces between subsystems are very common in general. And remember, crosstalk is reciprocal: it's often not enough to make your low-level external interfaces differential. Can you make all your logic differential? Can you do differential interfaces between every on-board chip? \$\endgroup\$
    – John Doty
    Commented Oct 27, 2023 at 13:30
  • \$\begingroup\$ @JohnDoty You don't need "differential interfaces", you just need to have the ground reference paths accompanying the signal paths. That way magnetic fields don't have much area to go between them. Only when you need to splice a ground reference do ground bus implementations (the mentioned resistor networks) come into play for breaking up a ground loop. I have an overland line passing in 200m distance and work with amps; I have a bit of experience with managing magnetic field impacts and other noise sources... \$\endgroup\$
    – user107063
    Commented Oct 27, 2023 at 13:37
  • \$\begingroup\$ That's a start. But suppose you have two subsystems, on common power, with a signal connection between them. You put in a ground conductor by the signal. Good, but now you have a ground loop that will carry the current due to any external AC flux. That's often not so good. But if you have a lot of low impedance ground loops, they share the current, so it's less of a problem. A single ground loop is the worst case. \$\endgroup\$
    – John Doty
    Commented Oct 27, 2023 at 17:11
1
\$\begingroup\$

I would say avoid sharing impedance between loads. Star connection is best for that and make sure you have some capacitance there.

Point is when you have multiple loads sharing a trace from the source you will have common impedance and resistance. Meaning that if load one suddenly increases current consumption load 2 sees a voltage drop. This would be bad in sensitive circuits.

BAD

Loads behind shared trace

STAR

enter image description here

Star is a better way to insulate subsystems from each other. Capacitance in the junction helps to cancel the effects of shared impedance to the left of the junction.

If you plan on using a solid supply plane you should have at least 6 layers and use only two for signal routing. See this https://www.youtube.com/watch?v=ySuUZEjARPY&t=4826s&ab_channel=Altium

\$\endgroup\$
1
\$\begingroup\$

Every current that flows from your power supply on the power conductor returns to your power supply by some other path. It's these loops that are EMI trouble, not loops in the power conductor itself. The more area such a loop surrounds, the more potential trouble.

So, ideally, the power distribution geometry should reflect the current flow in the signal network, keeping the loop area small. Bypass capacitors around current concentrations help by allowing the high frequency components of the current to return through ground as well as power distribution conductors.

A star network for power distribution is usually only a good idea if the signal flow matches it. But sometimes it's a good idea to isolate patches of circuitry, routing all power, signal, and ground connections through a narrow "bridge". A disciplined form of star connection.

\$\endgroup\$

Not the answer you're looking for? Browse other questions tagged or ask your own question.