1
\$\begingroup\$

I'm attempting to prototype this curve tracer (http://www.vintage-radio.info/download.php?id=224) mainly following the original design, but altering it to suit my own requirements (serial connection to PC instead of oscilloscope, etc). I've been stung by grounding issues with measurement circuits before, so I thought I'd share this with the community first to avoid any major gaffes.

Here is a high-level component diagram showing my plan so far (click to expand).

enter image description here

Each component box represents a potential circuit board. I'm planning to use a combination of stripboard and copper plane (Manhattan) for the boards. The diagram is a bit of a mess, but essentially there is an effort to separate analog from digital, with separate supplies and a star ground. I've tried to follow best practice regarding the mixed signal board containing the ADC, but I'm not sure I'm there yet and I have a specific question (see below). The main issue I'm faced with is that some components like the step generator use both analog and digital supplies. The step generator is composed of a pulse generator and a counter (both fed from the 5V digital supply) feeding an array of transistors having the analog supply across the CE junction.

Here's an snip of the step generator for reference:

enter image description here

So, my questions are:

1) Regarding components using both digital and analog supplies (like the step generator), what is the best practice approach to grounding? It seems difficult in this case to avoid having the 5V TTL signal at the transistor base polluting the analog supply.

2) Regarding the mixed-signal ADC board. Having done some research it seems the advice is to power the ADC using the analog supply. Since the ADC requires a 5V supply, does this mean I'll need yet another regulator to create 5V from the +12V analog rail?

\$\endgroup\$
  • \$\begingroup\$ 1) I don't see a transistor \$\endgroup\$ – Andy aka May 12 '16 at 13:05
  • \$\begingroup\$ I've just added a snip of the step generator section for reference. You can see how the digital IC is driving the transistor which has the analog supply across its CE junction. However, the question is more general than this, I'm wondering if there are general approaches to be followed in these cases. \$\endgroup\$ – Buck8pe May 12 '16 at 13:15
2
\$\begingroup\$

Have single ground. Divede the board to domains (digital, analog, maybe something else) and place components accordingly. Those who have several supplies will then be on the boundaries between domains. This way no currents will cross those boundaries, so you will not have to separate grounds. Check what supplies will require separate plane (those with many comnections) others may be routed with wide traces on top or signal layers.

Add chassis net on board outline, and bypass ground with capacitors to it. Two capacitors (0.1uF and 10uF) near each screw. So with metal enclosure all those points will be very silent.

My impression is that in terms of supplies your board is quite common.

| improve this answer | |
\$\endgroup\$
  • \$\begingroup\$ I wish I had the time and money to do this properly, but having multi-layer boards printed in small volumes is expensive where I am. It's also quite difficult to get large copper clad boards, so you're suggestion to place everything on the one board is a difficult proposition. Is there any way to extend your ideas into a multi-board setup? \$\endgroup\$ – Buck8pe May 12 '16 at 15:39
  • \$\begingroup\$ Mmm... Yes, but it will be more difficult and require really accurate design. Besides, i really don't believe you will save money this way. If you can't design a four-layer PCB, try using prototyping board. There are boards with pads for smt on one side and ground plane on the other. And tons of wires. \$\endgroup\$ – Gregory Kornblum May 12 '16 at 15:43
  • \$\begingroup\$ If you still will go to separate devices, the secret is to make one (abd just one!) return path for each current. Then if in any place you see you must create another path for existing current- use isolation circuit to brake the loop you would otherwise create. \$\endgroup\$ – Gregory Kornblum May 12 '16 at 15:47
  • \$\begingroup\$ Well, I'm planning to use stripboard (which is similar to prototyping board) and Manhattan style copper board. I'm guessing if copper board is good enough for RF, it's good enough for what I need? \$\endgroup\$ – Buck8pe May 12 '16 at 15:47
  • \$\begingroup\$ That's an interesting comment. Any chance you could update your answer to include a quick sketch of what you mean? This sounds like the answer I'm looking for. \$\endgroup\$ – Buck8pe May 12 '16 at 15:49
1
\$\begingroup\$

OK, this is about ground loop breaking with optocoupler. I hope the sketch is clear enough. The recommendation is very general on one hand and doesn't cover all systems on the other.

sketch of ground loop

So in this image you can see two cases of same system: three devices, first has power supply and provides it to the other two. The ground is passed in the cable with power and signal. Also, the rightmost device sends a signal to the leftmost device, and the ground is there too. The reason for that is that for signal integrity and sane EMI behaviour you need return path to the signal. It's separate discussion on why you need it, so for now just trust me.

What happens is that there are two return paths for the current now (green and orange) and it's bad for many reasons. The simplest is that the bottom wire may not be rated to the current of power supply and just fuse. The more complex, but also more dangerous reason is EMI- system with two return paths will emit EMI and will suffer from other EMI (for example, i once saw system shutting down when cell phone reng).

So there is a way to brake this ground loop: optical (or other) isolation. This way the supply currents will only go one way. Look at case B: the signal is transferred through a barrier, so information passes, while current doesn't. Note that the isolated part requires power of it's own, and it still come with the current, but in that case it's just one return path.

| improve this answer | |
\$\endgroup\$
  • \$\begingroup\$ I see what you mean. You have a component with a single power rail and two ground returns. That's asking for trouble for sure and your suggested solution is a good one. However, my case is slightly different. I have two power rails and a component that has the two intersecting. \$\endgroup\$ – Buck8pe May 16 '16 at 12:00
  • \$\begingroup\$ I'm worried that noise in the (digital) signal on the transistor base will couple to my nicely isolated analog supply. I'm guessing it's inevitable to a certain extent and maybe I'm worrying too much. If, after building, the signal line from the step generator has a lot of noise I'll stick a bead on there and that should solve most of it. \$\endgroup\$ – Buck8pe May 16 '16 at 12:08
  • \$\begingroup\$ Two power rails are not very different from one. Just be sure return paths are fine. The only problem i can think about with two rails is common mode currents- like one behaves as return path of the other for high frequency noises. This may become an issue if you cave capacitors even to GND in various places, which you probably have. Fot that people use common mode chokes, filters, etc. Best is to be sure (by LC filter) that no high frequency noise will even be conducted on power rail. Then the you just have to make sure that you have a star connection (single point somewhere) \$\endgroup\$ – Gregory Kornblum May 16 '16 at 12:41
  • \$\begingroup\$ I am not sure i understand our concern about base coupling to ground. The base-emitter capacity only becomes significant on very high frequencies. But even then you shouldn't worry about the ground, it is always 0V (well, at least if it's a nice thick piece of copper) \$\endgroup\$ – Gregory Kornblum May 16 '16 at 12:43

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.