MMIC filter

Active RFI testing does make it obvious.

Hunting for interference inside a machine is less well-structured, but if you do it often enough you can eventually get to learn what to look for.

Only in the sense of wandering around like a lost sheep. Exploration is goal-directed and an infinite solution space has an infinite number of goals.

Breaking out of conventional wisdom isn't exploring the whole solution space - it's looking over the fence into adjacent areas. There's absolutely nothing wrong with that but it doesn't get you all that far.

Continuing to spend money on a approach that doesn't look too promising isn't a great idea.

I had great fun with that once - the Metals Research/ Cambridge Instruments GaAs single crystal puller relied on an LVDT and spring to very precisely track the weight of the crystal being grown.

The dual emitter transistor used in the oscillator that energised the LVDT went obsolete and I ended up inventing a current mirror version version of the Baxandall class-D oscillator to generate a really clean and stable sine wave to drive the LVDT. Putting in an LT1028 to amplify the output of the LVDT meant that I could use less excitation current in the LVDT, and it stopped drifting on start-up.

Getting rid of the uA741 that amplified the DC output from the demodulator, and replacing with an op amp that did have a 1/f noise spec was the only change the operators noticed - the RF heaters that kept the GaAs molten started running at a steady 30% rather than banging full on for 30 seconds and going right off for the next minute or so.

It took more than a day to pull a crystal, and the operators found the process much more restful after the up-grade.

But they were quick - the fastest stuff you could buy at the time. They never did get the production yield high enough to make money out of them. The signal voltage levels were ECL compatible. Like ECL they were intended to use 0V as their positive rail, but there was a -3.4V rail that sank most of the operating current and a -5.2V rail that provided bias voltage. Like ECL, it made sense to provide a -2V rail to which to return the terminating resistors.

I was stuck with a boss who wanted 10psec granularity. The GaAs parts did let me deliver. I was amazed that price and the single-sourcing didn't put him off, but he was very much into selling his machine on the performance it could achieve, and less interested in making sure that it could deliver that performance day-in, day-out.

On a machine with half a zillion panels, not so easy. Yes, you can prove it but that involves lots of angle grinder use, noise, dust, people not liking you because of the noise and dust, machine shop time, cuts, bruises, metal splinters in the skin, et cetera. That got old over time.

Sure but that gets old as well. Essentially you are always looking and finding other peoples mistakes and then have to find a very polite way to tell the crew. In Asia that can be especially delicate. "I found it is a good system. We could improve it a little by ...".

IOW EMC work isn't very innovative. It's reactive.

That's one way of looking at it. If you begin a design with EMC in mind, you can anticipate a lot of problems. If you have a nasty fast waveform to ship around, making a it half of a balanced drive and shipping it around with with its complement on shielded twisted pair (or paired off in a ribbon cable) can make life a lot easier.

The junior engineers get the message fast if they see you making the design messier and more extravagant than they would have done.

The Cambridge Instruments story was about an electro-beam microfabricator where the patterns written would shift at random by about half a micron for a minute or two, then go back to the right place.

The chief engineer (who was an old technology buff) got shipped over to America to solve the problem.

He had to ride a lift to get up machine, and noticed that it was a very ancient hydraulic lift, which he knew meant that it had a big lump of (magnetic) wrought iron as the floor of the bit that moved.

Problem solved. When the lift was up the magnetic field at the electron beam microfabricator changed enough to move the electron beam by half a micron.

The lift had to stay put while a pattern was being written, which was manageable. It took a trans-atlantic air-fare to solve the problem but with a million dollar machine that was okay.

Orf worse, "we've already wasted $200K on a stupid idea, so we don't want a better, cheaper design now."

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Problem is, consultants are perceived as expensive (they really aren't), so they get called in after all that. When stuff has hit the fan, when a hardcore fail at the EMC lab has happened. Then you are correcting mistake after mistake after mistake.

If they had pulled us in at the beginning it wouldn't have happened.

In John Larkin's previous company building there was a water-driven elevator. That was cool.

I had some of those. 20+ hours of travel each way, 2h work, problem fixed.

The best was when a boss became irate that I didn't show up at a meeting. "I didn't know" ... "But I sent you a fax" ... "When?" ... "At least 15h ago" ... "That's when our flight was over New Foundland" ... "Oh!" ... "By the way, we just fixed the problem this meeting was probably about so maybe we don't need the meeting anymore" ... "WHAT?"

Nobody wastes $200k on a stupid idea, but once you have spent $200k working out that a tolerably sensible idea wasn't actually good enough, the enthusiasm for doing something different isn't great.

There's also the case where over-enthusiastic management can mess up the development of a perfectly sensible idea, by cutting corners in the development of more complicated electronics than the management is used to dealing with. The Cambridge Instruments multisampling electron beam tester eventually worked, but the development took a year or so longer than it should done because its first project manager didn't see the point of design reviews and skipped them as a waste of time. Finding and fixing a bug in a schematic is lot faster and cheaper than finding it in an assembled board.

Exactly. Being goal-directed means that you have already decided the direction to go in.

Maybe the best design is in the other direction.

Our policy is to always stay confused for a while early in a design, and not latch on the the first idea (probably in a textbook) that might work.

The techniques of exploring the greater solution space can be taught, but not simply explained. Like I can't tell you how to play tennis in an email.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Goal-directed means that you have decided where you want to end up, not how you are going to get there.

A goal is a destination, not a route-map

Being confident about where you are need to end up but relaxed about how you are going to get there isn't being confused - just being open-minded.

You aren't great at explanation at the best of times. Your capacity to explain what you have done isn't great, and your incapacity to lead the reader through the solutions spaces that you claim to have explored does suggest that you do more flailing about than exploring.

I'm not an enthusiast for the rip-it-up-and-start-over style of circuit design, but it is a necessary part of the designer's tool-kit. You do have to do a certain amount of work to determine whether a particular approach is likely to pay off, and the indications that warn you that a particular approach isn't promising can point you at an approach which is more likely to succeed.

That's already restrictive. If you allolw yourself to think freely about X, you might accidentally invent Y.

And if you allow your competitors' features, or your marketing or someone to define a product, you will likely miss designing something even better.

Confused is even better. Most engineers and even more managers are uncomfortable with uncertainty and want to lock down a design as soon as they can so they can fire up some project management software and make budgets and schedules with colorful presentation graphics.

It takes a lot of confidence to deliberately stay confused.

That line belongs in my collection of great quotes.

As I said, it can't be simply explained or taught in textbook form. It has too be done.

I think academic EE education is fundamentally hostile to idea generation.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

FLIR came by one day and made a movie that included the elevator. I'll see if I can find it.

They gave us a small thermal imager that wasn't very good.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

This is a well-known problem. Project management course tend to see virtue planning the course of a development project before you know you know exactly what you want to develop - it gives management something to do before the engineers have made up their minds. It's also a waste of time, and runs the risk of forcing the engineers to make premature decisions.

It would. You do seem to have a talent for remaining confused. Stupid people are permanently confused, and your prescription would encourage engineers to drink a lot, all the time.

It probably can, by somebody who finds the right way of looking at the problem. Tom Peters thought that he had

Academics get status by publishing interesting new ideas in the peer-reviewed literature. They do have to generate them for themselves - plagiarism is a crime. They do get snooty about people who claim to have invented new ideas when the idea have been published elsewhere, even if the claimant is too unsophisticated to have found them for themselves.

You would have treated that as hostility if it had happened to you.

If they had known enough to realise that they needed to pull us in, they probably wouldn't have needed to pull us in.

The reactionless drive thread may be a more extreme example of the problem.

It is theoretically possible that they may have lucked onto to something real, but the likeliest explanation of what's going on is that they started off not knowing enough about how their measurements could go wrong, and have now modified them to generate even bigger errors than they started off with.