the second 80 percent

Submitted by Ed_B on Tue, 10/13/2009 - 12:51

Printer-friendly versionPrinter-friendly version

Larry Wall once said about the progress of a large project that "we're well into the second 80 percent". Designing a printed circuit take a lot of patience. It's worth saying that almost none of the troublesome design constraints on this circuit are electronic. Most of the work has been in making things fit into a particular form.  Finding parts that would fit was a big challenge. Ths board was to have no surface mount parts, in order to make hand soldering easy. The object of the board-making excercise is, obviously, to turn an electrical schematic into a physical thing. This is the final schematic for the board. It's the same revision as the final board design. Do a save-as on the image to get a better look at the diagram.

A small enough through-hole replacement for the SMT Radio-frequency noise supressor filter on the other board wasn't available (that I could find) from Mouser, DigiKey, or Newark, or any of their Manufacturers. So afer a day or so of digging, I decided that a ferrite bead in the positive lead of the power bus would have to do (MuRata BL01RN1A1D2B from Jameco). It's not so different from what was there before, at least in spirit. And they're cheap. It's called "inductor" in the circuit.

Interestingly, finding three matched red, green, and blue LED's in a T-1 package was tough. It's best to stick with a small number of well-known vendors and that makes specing tough sometimes. Not Jameco, or any of the other big suppliers we normally use could fill the to the rescue. But still, it was weird that it was hard to find three matched led's.

The polyswitch self-resetting fuse on the first version of the ArtBus board got used more than I ever thought, but what am I thinking. Of course it did. People are klutzes. That polyswitch (on the first board) was too big for the board, but I couldn't find one smaller. I wanted SMT, but could only find through-hole. Now I'm wanting through-hole, but small. In a 150 mA device there aren't any small ones. But I did find a rather chunky SMT one that should be okay for hand-soldering. It took about a day of off-and-on to get a good part number. It's F1, FUSE, in the circuit, and the sole pair of of surface mount pads (red rectangles) on the board. I hope it's a good electrical fit in the circuit (Bourns MSMF014-2 from Mouser).

When I got my first batch of ArtBus boards stuffed, I thought I needed to call the bluff of my design. The chip firmware has pretty good error checking, so I did my experiment. I set up a running system with my laptop doing a Processing sketch talking to the board. Then I took my keys out of my pocket (work and home keys on one chain) and dropped them on the board while it was running. I did it twice. No USB system failure at all on the laptop, and both times the board restored operation automatically and properly, and everything kept on ticking. I was happy. The reverse protection diode acts with the polyswitch to make the board immune to reversed power polarity at the inputs. This saved a large installation piece that was being wired up a little too close to their deadline.

The final layout looks like this. Red is the top layer copper, green is the bottom layer copper, blue is the silkscreen drawing layer, tan is plated through drill holes that span both sides of the board, grey is the board outline, and yellow is the registration mark. Except for the registration mark, each layer is a separate CNC file that is extracted from the digram in the last step of the process. That step is called (I think) "posting", in the same sense that it is in other CAD/CNC situations.

So getting the parts to fit the board was the main issue. In the 3-d rendering of the board in my last posting, everything looked fine, but there were points of interference between components in several places, so a lot of pushing and pulling had to be done to get the fits right. Printing the foil pattern on paper and poking the parts through the holes is a good way to check the look and feel, and verify the mechanical fit of the parts. It's usually about three hours between major revisions of the PCB. What doesn't show is that there is a lot of ripping up stuff that isnt part of the changes in order to clear a workspace to make the changes. Often it goes back together the way it was, plus the changes. Other times several things will change subtly between revisions.

The differences between the two paper models here include changing the polyswitch footprint and moving it slightly, and moving several things around it to make room. Also the big thing is rotating the crystal. The body of the crystal has a "skirt" of sorts that extends well beyond the device pins. I failed to account for this and didn't realize my mistake until the paper-stuff test. I've been bitten by this problem on another board, and felt sure I did it right this time, but alas.

At that point in the design, I had the choice of tearing up all the connections to the chip and moving it 4 or 5 millimeters to the center -- which is basically starting the whole thing over -- or turning the crystal and its pair of capcitors 90 degrees. Starting over was out of the question. Turning the crystal was an unpleasent and a bitter last resort because of the amount of board real estate it chewed up. On the up side, it provided the opportunity to rerun the traces between the chip and the output pins. They didn't look nice before, and people will judge you on that.


There were a couple of other revisions between here and the first beta version, but this took Saturday night from about 9:pm until about 3:am. Sunday and Monday were spent posting out the CNC files for the board house.

And so on.