Pease Porridge
Once upon a time, back a Philbrick Researchers about 20 years ago, we were asked to quote on a spec generated by Loral, a military customer located in New York City. They wanted a very fast DAC to draw vectors. They wanted a 10-bit multiplying DAC with very low phase error, very fast response to digital code changes, and low glitches. In other words, when they changed the code—such as the worst-case carry for 0111 1111 11 to 1000 0000 00—they wanted the output jump or glitch to be small and brief. That meant the output had to settle very quickly. I figured out that a double DAC could do this (Fig. 1).
One DAC (Q1, Q2) had an offset current thought R1, plus a main signal path through R2, plus a linearity correction path from a Howland current pump through R3. The second DAC (Q3, Q4) just had an offset through R4. The circuit also had a current reflector, so that when Vin was zero, the current though R1 would cancel out the current through R4.
When the input signal was not zero, the main signal path came directly through R2 quite fast. But the emitter of Q1 isn’t a constant voltage. If you rely on the current through R2 alone, it wouldn’t be linear. So I built a dummy bit cell and a Howland current pump. And by using A1, the current through the 20-kΩ resistor from its output to it noninverting input would cause the emitter current of Q111 to be a very linear function of Vin.
Then the current through R3 would cause the total current fed to Q1 to be very linear. Of course, the bit-switch was repeated about nine more times at various levels of precision and current scaling.
We built up a rather large and floppy breadboard of this. I was able to fit it inside a test oven that was about 18 in. X 12 in. X 12 in. inside. However, I soon discovered that no precision tests could be done because of thermal wobble and gradients in the oven. I needed some thermal baffling. I went out to a grocery store and bought some Turkey bags-the kind you can roast a Turkey in. These would not melt even at 125°C. So I put the circuit in the Turkey bag, and I got the tests done. Because of the Turkey-bag, we called it a Turkey-DAC. We bid on the project. We won the business.
However, in those days, we had three departments at Philbrick: Engineering, Manufacturing, and Give-It-Away. For some reason, the salesman, to make sure we got the business, quoted a really low price. But, to get this complicated circuit into the smaller package, it took several hybrid parts, all of which were shoehorned into a small potting shell that was about 4 in. X 2 in. X 0.5 in.
The Current Reflector, for example, was a hybrid subassembly. This wasn’t simple of cheap to manufacture, and it required lots of trims. It had offset trims, gain trims, and linearity trims. We had to build up the basic circuit, check it out, and put in the first round of initial trims or coarse trims. Then we filled up the potting shell about half way and cured the epoxy. After that, we did the next round of medium trims, which was followed by a little stabilization bake for the fine trims. More testing was done, and then we topped up the shell with epoxy. Finally, once a few more room-temp tests and high-temp tests were done, we could ship. We were selling this whole kluge, with all hermetic mil-spec parts, for a lousy couple hundred bucks. We were losing money on each sale, and we were NOT making it up in volume. In fact, the test and trim labor were larger than usual, and we could not get out very many DACs in a week—we were falling behind schedule on shipments despite our best efforts. This project was turning into a real Turkey. A Turkey-DAC.
Our management figured out that we had to find a way to stop losing money on this contract. We held some discussions with the customer’s engineers. We asked a lot of questions about how it was being used. I asked at length, “What kind of storage registers are you using to feed the bit lines to this DAC?” They replied that they were feeding in the bits via a couple of 5495 series-to-parallel shift registers. I could not believe what they were saying!! I asked them, “Are you saying that when the code is changed from 1000 0000 00, to 0111 1111 11m the code that the DAC actually sees goes through all 10 stages in the shift register, such as:
1000 0000 00 to
1100 0000 00 to
1110 0000 00 to
1111 0000 00 to
1111 1000 00 to
1111 1100 00 to
1111 1110 00 to
1111 1111 00 to
1111 1111 10 to
1111 1111 11 over a period of a full microsecond?”
The guy said yes. I proceeded to point out that that wasn’t the right way to get a low glitch. They were paying for the low-glitch feature but not using it. They were paying a LOT for a feature that they could not see nor use. I asked them how they managed to avoid seeing a big glitch during this awful transition? They said, “Oh, we blank the CRT.” I said, “We’ll see you in a couple weeks, because we can make a DAC that’s much easier to build, and easier to ship on schedule. It won’t have as low a glitch, but it will work just as well. And if you are blanking the CRT during a glitch, you’ll never see the difference.”
I cobbled together a quick breadboard with a bunch of 30-Ω JFETs to do the switching. It worked just as I expected, and we carried it down to the customer in its big metal box for a demonstration.
The Loral engineers looked at it, plugged it into their system, and agreed that it seemed to work well at room temperature. But they were curious—how well would it work over its rated temperature? They had an oven, and were interested to see how our circuit would play. We knew it would work fine, but our breadboard was much too big to fit inside their small oven. So we taped our breadboard to the face of their oven—in place of the oven’s door. In fact, the box extended well beyond the oven. After taping it really well, they ran the tests, and the circuit worked perfectly at all temperatures. They were impressed.
When we got back to Dedham, we drew up the schematic for the “NUT-DAC”: the “New Un-Turkey DAC.” I have reconstructed from memory what the circuit was approximately like (Fig. 2). MUCH simpler to build, MUCH simpler to trim.
And Frank Dota, the manager of our Drafting Department, used a special pencil to draw, on the schematic, a sketch of a Turkey skedaddling out of an open cage. A fitting label for an “Un-Turkey DAC.” And, because Frank used the special pencil, the picture of the Turkey didn’t show upon the print when we made a copy of the schematic, and we could look at the original, and he could not see what we saw—he could not see the Turkey skedaddling out of the cage.
This DAC was so much easier to build, and trim, that we actually got out shipments back on schedule, and shortly began to turn a little profit. Bob Goodell, our General Manager, called our whole team into his office, and presented me with a 44-lb. Frozen Turkey as a bonus for bailing us out of the Turkey-DAC business. I took this big Frozen Turkey home. We had a big freezer, so we had no trouble finding a home for it. But we could not fit it into our oven.
So my wife took it down to our neighborhood butcher, and she had him saw it in half, and theat way we got to cook and eat our Turkey 22 pounds at a time.
That wasn’t quite the end of the Turkey DAC, because we thought it would be a lark to present a technical paper on this DAC. I submitted the paper to NEREM (The IEEE’s North-East Regional Electronics Meeting), and then presented it in the spring of 1973. I showed the complete circuit, and all the trims, and all the technical specs, with a straight face. Really good specs! Afterwards, I overheard a number of people saying that they could not believe that a DAC like that could be produced. Well, they weren’t entirely wrong…
And, that’s STILL not the end of the Turkey-DAC. An aviator friend of mine pointed out that the Lockheed S-3A is still flying for the Navy, and quite possibly, probably, some of those old Turkey-DACs are up there flying right now.
All for now./Comments invited!
RAP/Robert A. Pease/Engineer
Address:
Mail Stop D2597A
National Semiconductor
P.O. Box 58090
Santa Clara, CA 95052-8090
Bob’s Mailbox
Dear Mr. Pease:
I also still find analog computers very useful in designing simulators and training devices. One analog computer I designed was a propeller-torque computer. It calculates torque as a function of RPM and pitch, and sends the resulting torque set point to a dynamometer controller in a complete (less propeller) ship’s engine room located in a school building. Another analog computer calculates generator speed as a function of time, electrical load, and turbine throttle valve position in an electronic governor trainer.
Both of these could have been done digitally, of course, but the analog design was simpler (cheaper) to design, build, document, and maintain. No software, no a-d to d-a converter. A classic trade-off example: Choose any three- better, faster, cheaper. I don’t think digital can come close to the cost-effectiveness of analog for any application requiring analog in and out, and requiring a fairly small number of add/subtract, multiply/divide, and log/antilog terms.
GLEN WALPERT, PE
Yeadon, Pa.
I tend to agree that analog circuits can still have advantages over complex digital of Fuzzy systems. –RAP