Stick to the schematic
Tales From The Cube: How do you get a 1.6-GHz oscillation from a 1.2-GHz part? When you assigned an intern to build your circuit.
By Barry Harvey, Intersil Inc -- EDN, October 30, 2008
I recently needed to re-create a settling-time-measurement jig I built 25 years ago. It transmits and receives 10V swings and is flat enough and magnifies enough to see 0.01%/division within 35 nsec on an oscilloscope. You can’t buy such a step generator, and you can’t use an oscilloscope unassisted to see 0.01% absolute-settling features.
I instructed our intern to build it. I made sure that he stuffed only the first stage, which is a wideband differential amplifier employing a 35-year-old CA3127 transistor array from RCA Semiconductor, which General Electric Semiconductor and then Harris Semiconductor absorbed. Now, Intersil has absorbed it, so we have the part in inventory.
He proudly showed it working in the lab. Square waves at the input came out square. He commented, though, that the differential outputs seemed to have a large offset. I checked the operating currents, and they were correct. I then applied a triangle wave to the input. Holy schmaltz! The output was curved! The other of the differential outputs was simply railed—not good. I inadvertently rubbed the board with a finger and saw the entire waveform warp back to triangle shape and shift its dc level.
OK, we had an oscillation there, one that was too fast for our 400-MHz oscilloscope to see. I discovered that I could get any manner of triangle distortion or offset depending on where I touched. A finger presents an impedance of approximately 100 pF in series with approximately 100Ω to ground, and this condition loads and modifies oscillations. A couple of things happen during oscillation. The oscillation grows in amplitude until it overloads devices, causing dc rectification and operating-point shifts. Thus, the offset varies with where you touch your finger. A consequence of this variation is that low-frequency signals variably encounter the dc shift, so gain and offset variations and even nonlinearities on the signal can occur.
The other result of oscillation is radiation, meaning that you can find the oscillation using an antenna and a very-high-frequency-sensitive lab instrument, such as a spectrum analyzer. So, I made a small antenna from a coaxial cable cut and a small loop connecting from the center conductor to the shield. This small transformer secondary coupled energy efficiently enough for a spectrum analyzer to show.
So, holding my loop near the part, I saw a 1.6-GHz spectral spike that disappeared when I powered the system off. I then found that I could put my finger anywhere, even in ground regions, and influence the spectrum. Unfortunately, I couldn’t find a place that killed the oscillation. I used a metal probe, fingering down to keep the probe length and inductance low.
|
The grounds were hot everywhere! I had our intern drill holes in a grid every ¾ in. between top foil and bottom grounds, shorting them all over.
After three sessions, I made no progress, so I looked at the data sheet of the CA3127. The peak gain-bandwidth product was 1.2 GHz. How do you get a 1.6-GHz oscillation from a 1.2-GHz part?
At this point, I didn’t trust the part. I asked what part it was, and the intern assured me that it was a 3127—the HFA3127! Looking at that part’s data sheet, I noticed an improved 8-GHz array at the gain-bandwidth-product curve, and, although an 8-GHz part can oscillate at 1.6 GHz, our board layout couldn’t handle such fast transistors.
He replaced the array with a CA3127, and all was fine—no offset, no distortions. I didn’t recommend the HFA devices, so I blamed the intern.
You can share your Tales from the Cube and receive $200. Contact edn.editor@reedbusiness.com.
-
It's really difficult to remember what we didn't know once upon a time. The interesting thing here is that the intern in question had used a related part ( the Gilbert cell version of the high speed transistor
array family ) in his senior design project, which won an award for best project in his class at Stanford. So it may have been ill advised, but not completely illogical for him to use a similar part in his next
project. In his class project, he was instructed to include base resistors to quell the oscillations from these high speed devices. We sometimes have to run into these sorts of problems a few times before we understand and remember them, which is why interns have supervisors. So I hope the intern in question, and his peers (who have circulated this all the way back to their senior project adviser), don't take this too hard. One has to learn to beware of both high
frequency devices and distracted supervisors.
Dave Ritter - 2009-8-5 13:19:00 PDT -
Yes, just kidding. I would not treat a young person badly, I hope. In fact, I am known around the office as SOB (sweet ol'''''''' Barry).
Barry Harvey - 2009-26-3 08:29:00 PDT -
Test.
Joe Sixpack - 2008-15-11 12:04:00 PST -
You blamed the intern?!!! I hope you're just kidding.
There is no way the poor schmuck could be expected to know the difference between CA3127 and HFA3127 unless you specifically warned against it at the outset. You've been around at least 25 years (so have I). Back then, there was usually only one of each thing so such mistakes were not even possible (I exaggerate just a bit). If that was his only fault, it hardly seems worth looking to assign blame to him in such a public manner - this article.
The experience surely taught him at least two good lessons. 1) Don't substitute parts unless you know enough to ascertain the usefulness of the new part. 2) Watch your back because you never know who will be gunning for you.
Oh, if you were just kidding, then never mind.
Kirt Blattenberger (RFCafe.com) - 2008-14-11 16:13:00 PST -
I've often wondered how practical those HFA parts are, given the package parasitics. It doesn't take much inductance to make a few-GHz resonator with a pF or so.
Years ago I used to work very hard to get performance out of fast devices that were only available as TO72-ish packages (or in some cases raw die, which I had no way of using). Besides their inferior low-frequency parameters, by the time I got things to work there were so many damping Rs and so forth that I eventually concluded that I'd have been better off with slightly slower parts. And these were tight-as-possible layouts---close-packing of circles, air connections, etc. These were one-off hand-built machines, not ready for manufacture.
With SMD now things have gotten quite a bit more practical.
Brad Wood - 2008-5-11 12:51:00 PST
