Or, Duck Dodgers in the 51st Megahertz and beyond!

I'm interested in beginning to work with circuitry again, and would love some advice. I've dealt with analog, discrete things in the past. And on the formal side, I've studied a fair bit (for a college dropout) of EE/CE/CS. But I've no idea how to put the two together.

I know things change once you start dealing with higher frequency circuits. But how does one reasonably prototype boards? I've always been very incremental in how I do things - start with a minimum implementation, test it, and then embellish. Sending away for manufacture and paying tooling fees doesn't seem well suited to that. Do I just need to design a complete board on paper, and hope i got it all right?

More specifically: Where can I not reliably test without a PCB? Where might the practical threshold be where you *need* a dedicated ground (or power) plane? 30MHz? 60MHz? It depends? (I know I can print two layer boards at home, thus the concern for needing more layers.) Equally, where do I need to switch to surface mount over discrete packages?

Many thanks,
Douglas

Hi.

I used to build prototype boards on a 0.1" grid eurocard board using point to point wiring, and with some care have done successful graphics controller, video, and processor systems at up to about 60MHz. It takes a lot of careful decoupling (Hugo always says you can tell a board I've designed, there are so many decoupling capacitors it keeps working for five minutes after you turn off the power ), but it can be done.

I've also implemented a multichannel SVGA controller on a double-sided PCB with no ground plane, which worked fairly well once I added extra thickness to the power traces with soldered-on wire.

At the RF end of the scale, it's perfectly possible to go to many gigahertz on double-sided PCBs, assuming you both know more or less what you're doing and use the right PCB material. IE, the standard FR4 board (fiberglass) isn't really ideal above about 2-2.5GHz, and is not actually recommended at that frequency although it will work if thin enough. RT Duroid (PTFE substrate) is essential for anything much higher, but is a lot pricier. If you really want to go nuts, use ceramic (alumina) PCBs with gold traces!

For pretty much anything above a few MHz, a proper 4-layer board will improve almost everything except cost It certainly makes the layout easier. More layers may become necessary if you go to particularly high packing densities, or use certain packages, such as BGA and uBGA.

The cost of PCB manufacturer has come down a hell of a lot over the last few years, to the point that it's almost not worth prototyping by any other method unless you really only want a one off, or are on a very very limited budget. There are companies that will do very low cost PCBs for small production runs or prototypes, such as www.eurocircuits.be or www.pcbpool.com . They have some limitations, but overall offer a very useful service.

As far as going surace mount, for all but fairly simple chippery you may have no choice. Most components that do something interesting are available in SMT packaging only, since they are all aimed at volume manufacturing where surface mount technology saves a lot of money. It makes playing with them at home a bit fiddly though. Also, bear in mind that a lot of chips listed on a manufacturers website may be available in theory but not practice, unless you want a couple of hundred thousand or so.

Don't bother designing anything other than an overview of a circuit on paper, since you won't be able to do much of anything useful with it if you want a PCB. Get a decent PCB CAD package and use that, it will give much more useful results and you will slowly build up a library of custom parts that will make subsequent projects easier. I would recommend Seetrax Ranger 2 for Windows (it was used to design the Empeg players), which you can get from www.seetrax.com/r2win.htm . The demo version will give you an idea of what it does, while the full version is only 60 quid. I've been using various Ranger packages for 16 years, and on the whole they're pretty good. The other main low-cost PCB package is Ultiboard, which I haven't used, but friends of mine tell me it's quite good as well.

Hope this helps.

pca

>Hope this helps.

Indeed it does. The only point I could hope for clarification on is the discrepancy between "point to point wiring successful up to 60MHz" and "above a few MHz, 4-layer improves almost everything". I understand it's a fuzzy line, but "with some care" is too fuzzy.

As for "on paper" - I shudder at the literal thought. It appears to be a shorthand for anything between "in my head" and "physical".

But you've been very helpful, at a level I've found very few resources for. (I sympathize with Goldilocks!) I appreciate it. And find a reasonably optimistic attitude from it.

Regards,
Douglas

By "with some care", I meant of course "if you know what you're doing"

To clarify: It is perfectly possible to break the rules (actually, they're more guidelines) and use unsuitable construction such as wire-wrap or point to point techniques for quite high-speed digital circuits, but you have to be aware of what is likely to break and how to fix it. Common issues are signal crosstalk, power spikes, ground bounce, signal reflections (which takes us into transmission line theory, and you don't want to go there), and voltage drops due to inductance and capacitance issues. All of these problems can be overcome, assuming you can correctly identify them and know the magic passes needed to rectify them

Doing everything on a proper PCB makes life much easier. Double-sided boards are the easiest and cheapest, but still have limitations, some of them the same as above, but will usually allow more consistent results and to higher frequencies than wire-wrapping. In addition, with a PCB you can use surface mount chips easily, whereas wire-wrapping an 80-pin quad flat pack chip is fiddly. Not impossible (you glue it down upside down and use lots of very fine wire), but fiddly.

Going to a 4 layer board allows you to run the power and ground planes completely separately from the signal layers, which makes layout easier and improves the chances of the circuit working correctly, especially at high frequencies. With a solid ground plane you can dramatically lower the track inductance, increase the current available to the circuitry thus reducing the amount of decoupling capacitance required, and shield the circuit both from transmitting and receiving RFI much more effectively. Mixing multiple different power planes in one layer, or separating analogue and digital ground planes is merely a matter of experience.

If your circuit is more complex, you can of course go to more layers. The most I have ever required is 10 (the empeg is 8), and the majority of the extra layers are signal ones rather than power layers. I know of PCB manufacturers that can go to around 24 layers, and you can even utilise buried vias (ie connect one layer to another without the connection going all. the way through the PCB as they normally do), and even buried components, although this last is incredibly expensive and very rare.

One last tip: always make your power tracks as wide as you can manage. It helps.

pca

I'll have to dig out the board in question. It was a prototype for the graphics enhancer product I designed for the Acorn Archimedes about 12-13 years ago, and it's buried in a box of 40 or 50 other prototypes from those days out in the garage. I'll have a look later.

pca