Where do analog supplies go on mixed boards?

I have yet to see any split ground structures work.

Yep, usually :-)

Had to do that in nearly all cases where clients were stuck with noise because of such splits.

I've done many 'half-split' grounds and they worked great. Note they were not split grounds in the usual sense; they were split with voids for current return path shaping to keep high current return paths away from high speed reference planes.

Cheers

PeteS

You bring the supply across the bridge.

One of the places where I regularly consult puts 'moats' in the ground (and power) planes around noise sources such as switching power supplies, with one controlled bridge to bring all leads in and out.

They do quite well, with switching power supplies on boards that are quite noise sensitive. They mix video, precision motion control, and nasty switchers all in one happy pile, with quiet video and few motion control glitches (other than software problems).

Were these entirely split planes, with one-wire connections, or were they moated, with all signal lines that must cross going over the bridge?

I like the term 'moated' - it's exactly what I do to isolate noisy areas (SMPS for instance) from other areas, or protect quiet areas from the general noise.

I've done it for quite a while on many boards and it works when done right; that's why companies pay us money to show them how ;)

Cheers

PeteS

We usually have the same. Sensitive ultrasound, nasty switchers and motion control (for mechanical transducers).

On most complicated boards there just have to be numerous lines across. That's one reason why it often doesn't work. Then multiple ADCs etc.

That's why I ask -- at the client that I'm talking about they make the moat just as wide as it needs to be to accommodate all those signals. Think of all that circuitry in a bag, with the wires sticking out. Now put a rubber band around the end of the bag -- the edge of the bag is the moat, and the mouth is the collection of wires.

It can work but often the bundling of traces increases the crosstalk among them. Typically I ask client whether it is ok if I butcher one of the boards. Then I short all the moats and re-route some traces by running enamel wire and glueing it to the board surfaces. Afterwards there is either no change but in most cases a lot of noise is gone. Often there is also a miraculous vanishing of digital glitches, like a DSP that suddenly doesn't hang anymore. "And here we all thought it was a software problem...".

I'll occasionally make a thin C-shaped cut in a ground plane and tuck low-level stuff into the sheltered inside part of the C. This keeps big ground-plane currents from making small potential errors in the delicate stuff.

I've also "isolated" a square of copper on the plane and then reconnected it to the main ground with 4 or 8 skinny traces, so that the resistance of the traces is large compared with the resistance of the square, sort of like a thermal. That makes it hard for anything to jam a significant potential gradient into the square.

In both cases, the special region is not seriously AC/HF/EMI isolated.

This sort of thing is only necessary if you're mixing amps and microvolts on the same board, like a big heater PWM driver close to the thermocouple stuff. For something like a uP with relays and such, just pave over everything.

You can also make a hefty slit in ground and power planes, bridged by a number of skinnier traces, to mostly block thermal gradients.

But actually split ground planes? Nope.

John

I've just done such a board. It's on Vectorboard right now so we can experiment but it is a version with a full ground plane. Several watts of switcher for the TEC of a laser, delicate regulating loops, plus a microvolt RF amp. All on that same board. Works great, except for the DFB module which is too freaking slow to tune. Even diode current tuning is like molasses. Which brings up a question since you have worked with VCSELs: Are there any in the 1540nm range yet? Or at least above 1300nm?

I'll second that "nope".

It can be made not to work, sure. But it can be useful for those who know how to do it. I have been doing it for > 15 years now without failure. Peter put it quite well how and when it is useful to be done, I can confirm it from my experience:

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

------------------------------------------------------

Joerg wrote:

In article , Joerg wrote: [....]

I have seen many. In most cases, the trick is to make the "split plane" its own layer over the general ground plain and then connect it where needed. This works well for switching supplies and the like because it forces the AC currents to run within the power supply circuit and not get over into the op-amps.

In article , Tim Wescott wrote: [....]

I find it works better if you use an extra layer to make the ground under the DC-DC converter. In some cases this same layer can be used as the place to put the signal reference ground of the analog circuit. The logic just gets the normal Vcc and ground planes.

There are rumors of 1310 vcsels, but I haven't got my hands on one. I am using some nice, fairly clean uncooled F-P lasers at 1310, made by Appointech, that will launch a milliwatt or so into singlemode fiber, and somewhat more into multimode. I think they have 1550's, too.

I wonder if you could heat/cool one of these yourself, potentially better/faster than the packaged units do... they probably weren't intended for what you're trying to do.

Sounds like an interesting thermal design: solder the TO-18 can to the end of a brass (?) rod whose other end is buried in a peltier-cooled block. Add a heater of some sort right near the laser. Do the math and keep the thermal taus low. Close some loops. This sounds primitive, and I'm sure there is a more elegant way to do this, but it's bedtime for johnzo.

And oh, I have a friend down your way, a crazy Russian optical genius, who is making interesting electrically-tunable optical filters at these sorts of wavelengths. One of them could well be tightly coupled into a laser and essentially become part of the cavity. And if you need any consulting in general, he's da man.

John

Could you send me his contact info? jsc analogconsultants com Then I could pass that on to the client.

If we could create reasonably priced external cavity modules that are tunable this might work. For the fine tuning range we'd need to be able to traverse a 100pm or ideally 200pm range in about 10msec or less. Of course, if we could get up to 2nm that way it would be even better because then we could dump the TEC loop.

I have also seen them "work" but more in terms of not causing harm. However, too many times they did cause grief, mostly when lots of other stuff to the external world had to be connected to the digital and the analog sides. That forms a lop antenna which picks up pager system, police radio repeaters, embassy or coast guard transmitters, AM stations etc. All those signals then typically maxed at the split.