Part of the wiring harness before removing it from the chassis.

There’s two main power harness that run to most of the systems in the machine, which are thankfully still in good condition. As parts of them are hardwired directly into the transformer, it would have been a big pain to replace parts of these. They were pulled out to inspect, but looks like they’ll just get reinstalled as-is since there’s no damage.

The I/O pod removed from the back of the machine.

The I/O harness runs from the front of the card cage to a removable “pod” containing all of the ports on the back of the system. The mice that were living in this machine ate large sections of insulation off the wires of this harness (to use as nesting material I assume?), so a lot of it is going to have to be re-wired. I spent the afternoon today beeping out each connector and drawing a schematic of the harness in EAGLE, which should help keep track of everything as I rebuild the harness.

Part of the I/O harness showing damaged wires from mice.

Next step will be to figure out which specific wires need replacing, and how much wire I need to get. Then it’ll be getting a whole bunch of stranded wire and crimp pins for the connectors and replacing wires one by one!

Completed EAGLE schematic of the I/O harness.

Starting out with everything except the transformer disconnected and the system connected through a current limiting bulb and variac, I brought the voltage up slowly and… nothing. No current draw at all, no activity. Pretty quickly realized that I didn’t have the power distribution board connected yet, so the power wasn’t getting from the input to the transformer. Connected that and brought the voltage back up, but still nothing. Figured at this point I needed to trace the wiring harnesses out more to figure out exactly where the power was going.

I started tracing and found that the primary of the transformer actually looped through the linear board of the PSU, through the thermal cutout I didn’t understand the purpose of earlier. It turns out the power actually goes through the cutout on the linear board, then to the switcher board, which starts itself up first then closes a relay to give power to the linear side. Without the entire PSU in place, nothing was going to get power at all. I figured out which terminals the whole chain of boards would eventually loop together, and joined them with a wire for the time being. Turning the voltage on the variac back up lead to a loud humming and 60VAC out of one of the transformer taps, finally!

Had to jumper the connector to get the transformer powered without the rest of the PSU. Just like with an ATX power supply, except 120V!

I measured all of the transformer taps and documented them all for future reference, then connected the linear board up. I knew from reverse-engineering the board that it didn’t rely on the +5.2V from the switcher board at all, so the switcher having control over the linear board’s power must be to sequence power for the rest of the system. Since there was no actual load connected, not having sequencing wasn’t an issue and running the linear board on its own was safe.

With a meter on the +24VDC rail, I slowly brought up the variac again and watched the voltage on the rail climbing until it hit +24VDC, then it stopped, right where it should! Looking at the rail status LEDs, all but the +5.2VDC rail were up. I briefly started troubleshooting this before someone else at the lab pointed out that the +5.2VDC rail came from the switcher board, which was not yet connected. Oops.

First powerup of this PSU in probably 25 years, all rails except +5.2V (whch wasn’t connected) working!

I ran over the board with a thermal camera, and everything looked relatively okay, so I added a fan next to the board (as it would have in normal operation) and let it run for half an hour or so, with no issues.

This picture has nothing to do with anything really, but it looked like the capacitors were looking over the edge of the board and were quite surprised.

The switcher board seemed like the next part, I pulled the two largest capacitors off first and made sure they seemed OK, both had <25uA leakage which was fine. I hooked the board up to the linear board, slowly brought up power, and the rail didn’t come up. I vaguely remembered seeing this in the past, where the switching controller wouldn’t start if the line voltage rise time was too slow. Since it took the full 120V without excessive current draw or heating, I switched the line voltage off and back on straight to 120V, and the +5.2V rail came up! I let it run for a bit after confirming the voltage was good, with no problems at all.

The two boards of the PSU connected up for testing with all rails up and running.

Since the +5.2V rail has the highest power output (at least, from what I can tell by the big bolt terminals for connections and the fact that it has an entire board just for it), I wanted to run it with a test load on it for awhile. I figured it would supply at least 10-20A, and I didn’t have any resistors handy which would handle that for more than a few seconds. I remembered we had a spool of fairly heavy magnet wire left over from another lab member’s project though, and plenty of buckets. As I’d learned from a Mike’s Electric Stuff video awhile back, I measured out enough magnet wire to get about 0.35 ohms (~15 amps) out of the PSU, pushed it all under the surface of a bucket of water, and connected it up. It powered up fine and ran the load for half an hour or so with no issues and nothing getting even remotely hot outside of a couple diodes which was expected.

Using a bucket of water and magnet wire as a load for the +5.2V rail.

I tried cutting the length of wire down to increase the load, but it looks like the +5.2V rail shuts down above 15A output. Given how much total this machine draws and how large the +5.2V board is, I have trouble believing it can only supply ~75W, but with no documentation available for it I don’t really know. It’s possible it only supplies ~75W, or potentially some current limiting component has drifted over the years and it’s shutting down early. The fact that the heatsinks only rose a couple degrees after half an hour under the full load it would supply makes me think maybe something is wrong, but we’ll see. The tests I’ve done at least show that the output on all rails is stable under load though and won’t go over-voltage, so I’m comfortable proceeding at this point.

Checking over both power supply boards with a thermal camera.

I do want to finish reversing the linear board and then the switcher board at some point, but at this point I think I’ll move on to other parts of the system and come back to this later. The switcher board has a lot of interesting-looking stuff on it, so reversing it should be pretty enlightening.

The PSU schematic so far. +24VDC rail is completely reversed out, and starting to work on the +/-12V supply.

Next up is going to be taking lots of pictures of where every connector on every harness is, stripping it all out of the system, and starting to trace all the wires. Mice have eaten a lot of the insulation off all the data cabling and some of the power cabling, so tracing where all the wires go will be the first step in replacing them.

PSU reassembled and installed back in the case.

I had a box sitting around the HackLab which I’d built it to control various parts of our laser cutter many years ago (but which I’d since replaced with a fancier PLC-based solution), so modifying that into a current limiting box seemed to make sense. I rewired the switches so I have one regular bank of switched outlets and one bank of 2 or 3 outlets in series. This lets me plug in the device I’m testing into one, a light bulb into another, and (optionally, with a switch to select either way) an ammeter into the third. Should make testing the power supply a bit easier!

The front of the current limiting box.

The back of the current limiting box.

Limiting current through the light bulb in the lower left using another light bulb, and measuring how much current is being used. Still have to make a proper cable to go to the meter, but this works for now.

Working on reversing the PSU at HackLab.TO.

Reverse engineering older power stuff like this is not that difficult overall, just tedious at times. It’s a double-sided board, no ground planes, and mostly fairly thick traces, so following traces around isn’t that hard. All the components are through-hole (or bolt-on, in the case of the big power parts), so reading values off (most of) the passives is easy too.

I started by taking some good shots of the board, turning them to B&W and increasing the contrast, then printing out one of each side of the board. I found a fairly random spot to start (the first transformer tap input) and followed where the trace went, drew that into the schematic, and marked it off on the paper copy to show it’s been transcribed. Then repeated that for the next few hours. When traces go under other parts, a combination of bright light through the board and my multimeter on continuity mode helps find where they go, and I took the big caps off the board to help with that too, since they just unbolt.

Tracking sheet I use to mark off each trace that’s put into the schematic, after the first few hours of reversing.

The hardest part so far is that most of the components have Xerox-specific part numbers on them, so I can’t just look up datasheets to find pinouts. TO-3 transistors have a fairly standard pinout, so I’m assuming they use that one until I get the schematic a bit more complete. I’m still not sure if they’re PNP or NPN, but as more of the schematic gets completed it should become more obvious which they are, as should the pinout.

Thankfully the crowbar chips have the original part numbers (MC3423) which lead me to a datasheet. This finally explained what all the SCRs on each board are for, which is to crowbar the supply in case of overvoltage, not power sequencing on startup like we’d guessed. If the voltage on a rail goes too high, the crowbar chip triggers the SCR to short out the voltage rail and blow the fuse.

An example schematic of how the MC3423 can be used, which was helpful in figuring out what the designers were doing.

At this point the connectors, a small part of the rectification/regulation circuitry, and the crowbar circuitry for the +24V rail is reversed out. Since this board has 4 rails and 4 of the crowbar controller chips, the next 3 crowbars should be easy to reverse, just copy/pasting in EAGLE and beeping out each trace to confirm that it matches the same design.

The schematic of the PSU’s linear board at the end of the first few hours of reversing.

Each rail has an LED and a labelled meter probe test point.

It seems like they didn’t pay much attention to safety on connectors back then, as there are a number of connectors with 120V next to low voltage rails! I’ve documented all the pinouts that I’ve identified so far in the documentation section of this site.

Shining a bright light through the board helps the traces stand out, and makes it easier to see traces on the opposite side of the board.

I’m actually considering putting schematics together for at least sections of the linear board, since it would help me understand how it works as well as assist in any required troubleshooting. It’s complex enough that it would be at least tens of hours to do just the linear board though, so I may just do small sections. This is one of those times where I need to decide whether I just want to understand enough to troubleshoot anything that comes up and get the system working, or whether I want to understand how all the bits work in detail. I tend to lean towards the latter, since reverse engineering boards and understanding the circuits within is something I really like doing.

The PSU installed in the system. The red connector is all the transformer taps, and the screw terminals are the high-current +5V.

Ever since I’d first opened the system’s case I’d wondered why there was both a power supply box as well as a giant transformer in the unit. It turns out that the transformer is part of the power supply, it was just too big to fit in the box! There’s a wiring harness to take 120V into the transformer and send all the various secondary windings/taps into the power supply.

The transformer used to power the linear board of the PSU via many windings/taps.

I had expected the PSU to be a switchmode power supply, but it looks like that’s only half true. There are a total of 5 rails (or possibly 6, it lists +5.2V and +5V which may be the same rail). It looks like 4 of the 5 (the lower current rails, all except +5) are done via linear regulation, and +5 has its own entire board that implements a primitive-ish switchmode power supply. A lot of the components have Xerox-specific part numbers, so figuring out how things work is a bit difficult in places.

One of the two PSU boards, seems to be linear supplies for all the lower-current rails.

One of two PSU boards, seems to be the +5V SMPS.

Tonight I started pulling the biggest caps off the board (which is easy, since they connect with bolts!) and putting DC across them to check leakage and see if any need reforming. Got two done tonight, and both look OK after a couple hours of voltage across them, 4 more to go.

Testing capacitors to see if any need reforming.

Lots of cool and unusual things on these boards, at least for someone now used to 2015 electronics! A bunch of us at the lab spent a lot of time looking over the boards in detail marveling at all the high-power parts that bolt into the board and unusual parts like 4-terminal wirewound resistors and now-uncommon IC/transistor packages.

An interesting package on the PSU’s linear board. Old op-amp maybe?

An SCR bolted to the +5V output. Not sure what’s going on with this, maybe for power sequencing?

A 4-terminal wirewound resistor used on the output of the +5V supply for current feedback.

The list of caps that need testing to confirm <= 50uA leakage and/or reforming is:

• Linear board C1 – Not yet tested
• Linear board C2 – OK
• Linear board C5 – Not yet tested
• Linear board C6 – OK
• 5V board C5 – Not yet tested
• 5V board C6 – Not yet tested

The maintenance panel on the front of the Dandelion

The Maintenance Panel (MP) board normally connects to the I/O Processor (IOP) board as well as the power distribution board, but I wanted to test it individually so connected it to various bench power supplies and an oscilloscope instead. I found a schematic for the MP in the Bitsavers archive, which made this testing go much faster than it would have otherwise.

The MP board normally takes +12VAC from the power distribution board, which is on at all times even when the power switch is off. This power is used to run the RTC, and the 60Hz line frequency is divided down/conditioned and used to clock the RTC. In addition, when the system is turned on the MP then draws power for the LEDs and LED drivers from the system itself.

The maintenance panel PCB

I first brought up the board on +12VDC through a current-limited power supply to confirm everything seemed OK, which it was. After checking the board over with a thermal camera and making sure the output was +5VDC, I switched the input to a +12VAC adapter I had around. The 1PPS clock output from the board was working fine, which was a good first sign!

Checking the MP board for any hot spots once it was powered up

The LEDs are powered from the system itself, so I added a +5VDC power supply to run that. Bringing the blanking line low made the LEDs light up, and manually toggling the count line confirmed the display incremented properly. I found a 10Hz line on the board in the RTC clocking circuitry, and connecting it to the count input, the LED display started counting up as intended.

The MP also provides two pushbuttons (Boot Reset and Alternate Boot), which worked fine, as well as power failure detection circuitry. This circuitry sends a notification to the system that the power has failed and therefore the clock is invalid, and the system then asserts a line saying the clock has been set, which clears the power failure notification line. This all worked fine as well, setting and clearing as expected.

So the first bit of electronics in the system works perfectly! Next I’ll probably pull the power supply out of the system and start looking it over.

Rusty bottom fan cover

I tried to just vacuum it off, but there was so much paint/rust flaking off all the time that it was no use. I ended up using a wire brush on an angle grinder to remove the paint/rust over large areas and get down to bare metal, then a wire brush on a drill to get into the corners and places I couldn’t get with the bigger brush.  Once that was done, I gave it a fresh coat of paint. Once it dries it should be ready to install again, all shiny and rust-free. I expect I’ll probably have to repeat this process for a number of other metal parts in the system as I proceed.

Bottom fan cover stripped down to bare metal

Bottom fan cover with a fresh coat of paint

HackLab.TO moved into a new larger space in late 2014, meaning I finally had a place to keep the machine while working on it, as well as a space full of all the tools I’d need to restore it. Today I had some time and a ride to the lab (thanks Daria for the ride and Jeff for helping us get it up the stairs!), so the Star came to the lab to start getting restored.

Dandelion in its new home