2
\$\begingroup\$

At this moment I'm working on a project that includes driving multiple LED interfaces. I've got 16 LED interfaces in total, each interface has two 5V inputs. One input for the high current LED drivers and one input for the digital logic. I kept the logic and load supplies separated due to the fluctuating current consumption from the LED drivers (TLC5940). The interfaces have a common ground, however I designed them so the LED driver grounds and logic ground are only joined together at a single point at the power connector. All the other LED driver and logic ground areas are separated from each other. The LED drivers from each LED interface are drawing about 1A each.

Now I'm not certain about how I should supply all those interfaces. Currently I'm thinking about this. I have four interfaces that are grouped together (so I have 4 groups) that I call 'units'. All the high current LED drivers from each unit are fed by a single regulated 5V/7A supply. All the logic sides of the LED units are fed by a single 5V/10A supply. See the image for a reference. Can anyone tell me if this setup (including the separation of the high current LED drivers and the logic components) is a good idea? Am I avoiding high current ground loops with this setup? If someone has any tips or tricks I love to hear them! Thanks for all your time!

In the image the left side of a unit are the high current LED drivers, ground separated from the logic components at the right side. Both grounds meet at the common power supply ground pin from the interface. The most right PSU is feeding all the logic components. Both the LED drivers and logic components are fed by a 5V PSU.

EDIT: So as @Dave mentioned in the comments the design in the first image isn't a good solution at all. My first design looked something like in the image below. If I'm correct I won't have any ground loops between the interfaces/main processor unit as the grounds are joined together at the power supply (I can call this star grounding, right?). With this setup I'm only concerned about the (possible) voltage drops on the 5V rail. Because the load can vary quite a lot, e.g. if all the LEDs switch on the current demand is increased with roughly 16A. Of course I decoupled the TLC5940's from each interface with some large capacitors. Besides that I forgot to mention that the logic is fed by a 3V3 (LDO) regulator which is connected to the 5V rail of the led interface. However I'm still concerned with the possible voltage drops on the 5V rail when the LED interfaces are driving all the LEDs. In practice would a setup like this with a large single switching power supply work? Or am I asking for trouble with this kind of setup? Thanks for the input!

enter image description here

\$\endgroup\$
  • \$\begingroup\$ I see 5 potential sources for ground loops. More PSUs means more ground loop opportunities. Are you talking about using multiple PSU's to distribute the ground potential so that you can be lazy and not star ground for high current ground returns? Star grounding doesn't stop ground loops. Ground loops aren't inherent in high power supplies (just in poorly isolated supplies, there's honestly no reason to not have full galvanic isolation these days). Or are you referring to poor layout designs where the power and ground form an induction loop? Which is also solved by correct layout design. \$\endgroup\$ – Dave Feb 16 '16 at 3:29
  • \$\begingroup\$ @Dave, I've added some information to my original post. \$\endgroup\$ – WonderTiger Feb 16 '16 at 7:46
1
\$\begingroup\$

Okay I now have enough information to answer. To put it simply, the fewer power supplies you have the better. It is better to have one power supply that can deal with the highest current needs than it is to have two power supplies, one for logic and another for the LEDs. The only time this isn't true is when the two power supplies are internally connected. I.E.- you have a power supply that can provide both 15A @ 5V and 800mA @ 3.3V. Fewer parts is always better plus routing AC cables to all the supplies would be horrifying from a safety perspective.

Yes, that's a star ground. Yes, that's exactly how you do it, though to be honest, star grounding doesn't help the digital circuits, ground planes do better to absorb switching noise form the transistors.

Other thoughts: If you only need to have all LEDs on for a few seconds or less, you could get away with a lower current power supply and buffer the needed current with capacitors. An example: You have 3 10W LEDs (for simplicity). If they are ever one at the same time, they are only ever on for 2 seconds together before one of them switches off. You can't have the 5V line droop below 4.7V before somethings blows up (for simplicity, not actuality). You could potentially buffer the current draw by adding a 1334F capacitor (not realistic). You get that from E = 0.5*C*(V^2) where V is the change in the voltage you can accept, c is capacitance and e is the energy, which we got from 3*10W for 2 seconds (3*10*2=60J). That's just a thought exercise, I don't know your timing. But, if the concept works out to where you can buffer the current draw with a capacitor, you size the power supply based on how fast you need to recharge that capacitor and not on how much instantaneous current draw you have.

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

Henry Ott has convinced me that one solid ground plane over the entire PCB is better than splitting the ground plane -- unless you have an analog-to-digital converter with more than 14 bits of resolution, or you're doing something exotic like deliberately crafting a RF antenna or other RF components from PCB material. "Partitioning and Layout of a Mixed-Signal PCB"; "Grounding of Mixed Signal PCBs"; "General tips for avoiding noise"; " Trace crossing splitted power plane "; etc.

You may also be interested in the "Supplying Power to a Mixed-Signal PCB" FAQ, which mentions

a common misconception that all analog and digital signal currents flow back to the power supply.

This is exactly what you want to avoid.

In a properly designed and laid out board, all signal and transient currents are confined to the PCB. The power supply supplies only dc current to the boards. All signal and transient currents are then supplied from decoupling capacitors on the PCB and flow in small loops confined to the board. To do otherwise creates major EMC and Signal Integrity problems.

Marc Defossez, on p. 23 of his "D-PHY Solutions Application Note", has a nice spectrum of the typical frequencies seen on PCBs, and the appropriate decoupling strategy for each frequency band. (Thank you for the reference, Dave).

Sounds like a fun project. Good luck!

\$\endgroup\$
  • \$\begingroup\$ No ferrite connector between grounds option was discussed, why? Also no reference to what frequencies are best shunted by what forms of capacitance (and thus when to use things like star versus ground plane). xilinx.com/support/documentation/application_notes/… (page 23) \$\endgroup\$ – Dave Feb 17 '16 at 17:31
  • \$\begingroup\$ Henry Ott has convinced me that a solid metal connection between grounds is better than a capacitor or ferrite connector between grounds. That's why I didn't discuss that (inferior?) option. That Xilinx app note looks useful. If I'm reading p. 21 correctly, it seems to be saying to always use power and ground planes, never use star connections for power or ground. It has good tips for laying out a board with a FPGA connected to a low-power 800 Mb/s differential digital bus. I'm not sure how much of that is useful for TLC5940 high-power LED drivers that run at most 30 MHz and likely lower. \$\endgroup\$ – davidcary Feb 18 '16 at 4:47
  • \$\begingroup\$ This note is also for digital logic circuits where the transistors will actually use a ground plane and where the transmission line theory requires a power plane. Star ground is absolutely a better idea in certain applications (like battery connector on one end and having four high power circuits extending outward on the board). I only reference that document as it's the only one I've ever found that references decoupling strategies for most of the used spectrum. \$\endgroup\$ – Dave Feb 18 '16 at 6:07
  • \$\begingroup\$ @Dave: Thank you for the reference. You seem to be fishing for information on "When should I use a ground plane, and when is a star ground better?". I think you should post that as a fresh new top-level question, where it will attract better answers than it would buried in small-print comments. \$\endgroup\$ – davidcary Feb 20 '16 at 21:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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