What did you do last summer? It was kinda a long time ago, right? Well, I did some stuff that I wrote up and never got around to publishing. For me, last summer was the summer of aspirational Volt-Nuttery, which involved skimming a lot of EEVBlog Forum posts and NIST papers. The truth eventually sank in, though: Josephson Junction voltage standards are super expensive. Calibrations are expensive. Anything you buy cheap from eBay is not going to have a recent cal. Basically, no free lunch.
Regardless, I decided to dip my toe in the waters by getting a 6.5 digit meter for my home lab. Although a 6.5 is pretty pedestrian these days in real labs, the HP 3456A is a big step up for me, whose ‘nice’ multimeter is an Extech EX505 with 0.5% basic accuracy. Looking on eBay, I thought about getting a newer, standard HP/Agilent 34401A, but the prices put me off (and the vacuum fluorescent display eventually wears out). The 3457A, an older model 7.5 digit meter, was trending at $300, whereas the 3456A is south of $150 if you’re lucky. I took a cursory look at Keithleys but they are not really my style.
I ended up making an offer on one of those “we plugged it in and it turns on but didn’t do any further testing cough cough” listings. It was shipped in a huge box, wrapped up in bubble wrap and ensconced in a nest of packing foam. The first look was not promising — Self-Test produced the dreaded “-3” error:
The “-3” error means that the Outguard can’t talk to the Inguard. [Brief overview of the 3456A architecture: The Outguard handles the front panel and GPIB. The Inguard handles the A/D. The Inguard floats with respect to the Outguard, which is referenced to instrument ground. The Inguard and Outguard communicate over a transformer-coupled serial scheme.] The service manual contains quite detailed step-by-step troubleshooting instructions, helpfully. So by unplugging the Inguard comms and sticking in a loopback connection, the problem could be isolated to the Outguard comms section on board A3. With a bit of poking around, I could see that the transistor array U21 on the receiver was not interpreting the signals correctly. For example, the recovered clock looked quite thin, and lacking in sufficient TTL-level amplitude.
Now U21, a CA3046 transistor array, is, shockingly, no longer commercially available. There is old stock rattling around out there, but I didn’t want to bother tracking one down. What exactly does this part do in this circuit, and can I come up with an alternative repair? It appears that this chip is set up to interpret the serial input data; clock and data are extracted, which go to a shift register. For the clock recovery circuit, two transistors are set up to activate when the input voltage exceeds one V_BE from zero, positive (Q1) or negative (Q2). If either of these transistors are on, the following transistor (Q3) will not be turned on, letting the overall output float high.
The other connection at U21 (not shown, sorry) is a little mystifying at first glance: base and emitter hanging off the data line, +5V to the collector. As connected, the transistor doesn’t do anything, so what’s the point? My guess is this connection serves as a diode clamp, since the BC junction will act as a clamp to +5V. Additionally, what’s not shown on the CA3046 schematic are the parasitic diodes that are a fact of life in IC design. The receiver circuit must handle an input voltage that goes lower than -0.6V, which would be a problem for the subsequent TTL shift register, so it stands to reason that any negative excursions should be clamped to ground.
What does this mean? This means that we can replace a CA3046 with some discrete transistors and diodes. Nothing special, just 2N3904 NPNs and 1N914 small-signal diodes. The pinout on the 2N3904s is not a perfect match for two of the transistors, but it’s not a big deal. The CA3046 contains a matched pair, but the design doesn’t appear to utilize it in any way.
I would love to tell you that this fix was an immediate success, but the truth is harsher: it didn’t work. Scratching my head, the signal appeared to be right, but the slopes were a little weak. It took me a while to realize that my loopback jumper, constructed from two pairs of Pomona minigrabber/banana cables, represented a considerable inductance, slowing down the slew rate. Lacking the appropriate jumper, I cut to the chase and hooked up the Inguard. The “-3” was vanquished!
Well, the “-3” turned into a “-4”, actually. So I did some poking around the analog section. A fuse was loose and the big relay was marked “BAD”, but nothing seemed terribly amiss. It was evidence that someone had already been inside, though. Flipping the box upside down, back and forth, eventually I noticed that the Inguard logic board was skewed. Whoever had been inside the box last hadn’t plugged it in all the way! (It’s possible it came loose in shipping, due to those blasted plastic Nylatch things falling apart.)
I also replaced the two caps on the Inguard power supply that support the unregulated +33V supply. At the time, I thought this made a difference, but now I don’t think it was necessary.
So we come to the present day. This box mostly works. If you leave it alone for a while, or look at it wrong, it locks up and the main relay does a twitch of death. So the next suspect is this DIP socket. There was actually a Service Note issued by HP, saying to get rid of the socket and solder the AM8048 directly to the board, so I feel somewhat confident in my suspicions. Removing that 40-pin thing is going to be a major pain though, so I’ll save that work for a rainy day.
Since I’m a glutton for punishment, I ended up buying a second 3456A which I fixed recently, so stay tuned for part 2.