Well that entire power supply was very ayashī !! Congrats!!! Man, debugging things that are connected in a big giant loop is tricky. It looks like it's faulty at every single point, and it's nearly impossible to distinguish the cause from the effect. Plus you had the crowbar masking the effect! The way I usually debug these is by breaking the loop somewhere and feeding the input at that point with something reasonable, then following it around like in a normal non loopy circuit. But in power supplies that can also result in an entertaining ElectroBOOM if you are not careful. Great job getting it going!
@@CuriousMarc it's really interesting how over engineered they are willing to go when they know the customer base is going to be crazy small and the margins are going to be high. Even as popular as the pdp11 was my guess is during the period of time before LSI they knew they would only be selling a few thousand of these things. Kind of like high end medical they were willing to go to the nTh degree to just make sure that high price tag was justified. Especially in an age where components level maintenance was commonplace, it would be unimaginable if a company like DEC cut corners and it got out.
@@lindoran DEC always over-engineered everything. But in doing so they did not fail as often. Now 40 years later well who would have thought people would still be using these. Kind of like COBOL. The people that wrote the code 50+ years ago never expected code to still be in use.
Woah I got mentioned in a Usagi Electric video!? Am I dreaming? I’m very glad to see the 11/44 back up and running. It has been so much fun to work on the machine with such an experienced and talented group of people on your discord. I have learned so much from them. I am forever grateful to everyone who has shared their knowledge and experience. Thank you for bringing such an incredible community together!
13:13 This is a constant current source. D34 has 3.9V, the emitter of Q24 has 3.3V. So at resistor R111. So 0.65mA flows into the capacitor C40 via the resistor. This can then be used to calculate the time.
The emitter is always the one with the arrow (pointing out, NPN, pointing in, PNP.) Or, as I learned it, “NPN - Not Pointing iN, PNP - Pointing iN Positive.” 😊
Thank you so much! I really wanted to try to understand what was going on inside this bonkers PSU, so it was actually a lot of fun following the trouble around the schematic.
The pnp transistor is a constant-current source for the time constant capacitor. Makes the capacitor charge up with a linear slope. Simpler 555 circuits just use a pullup resistor that makes the capacitor charge exponentially. That transistor open would cause the 555 to not run right, possibly run too slow and cause over-voltage. Good find.
CONGRATULATIONS on finally identifying that bad transistor and getting all six power supplies up and running!! The PDP-11/44 was seriously an incredible machine. It would support multiple users, heavy calculations, and keep on running, day and night, almost forever. It would run RT-11, RSX-11-M, and even RSTS/E, as well as UNIX!! One last hint: Put the bus termination card immediately after the memory and console card, don't count on grant continuity knuckle-busters and possibly dodgy backplanes/sockets. Then try again. Good luck!!
Thank you so much! I'm quite excited to get a multi-user OS up and running on the machine. I've never used UNIX before, so it could be quite fun to work on getting a time-sharing UNIX setup up and running on it! Good call on the moving the termination card up and not ruling on the knuckle busters. The PDP-11 is an interesting machine for sure, and I reckon it'll keep me on my toes for many years to come!
If you look back to the place where the PSU was turned on with the multi meter attached to the 12V rail you can see 20V moments before dropping to 2.3V.
I saw it too, he was behind when that happened. But the oscilloscope recorded it i guess. See Adrian Black?, the UT-210E works fine. Fix your mode selector switch!, why is yours loose?
I once had a Fujitsu Eagle land on me. I was carrying it with a friend on an icy sidewalk and slipped under it. I thought that I was going to have to go to the ER. That is absolutely a proper hard drive.
It is kind of fun and cool that the purple color you see on some levers and handles inside the PDP11 is the same color used on all hot-swappable parts in a brand new HPE server. The color got carried over to Compaq when they bought DEC. And then later to HP/HPE when they bought Compaq. 😁
What amused me most is the fact that four pretty large boards comprise the innermost part of a single processor core which nowadays takes like couple of square millimeters of silicon on a single chip. And this revolution happened during my lifetime. I'm more or less this the same age as this machine. Also congratulations on bringing it back to life!
Q24 is used as a constant current source. This causes the voltage over C40 to rise with a linear curve and also the timing will not be affected by the exact voltage of the +15 rail.
Hi, I’ve been watching your channel for a while now and get a charge out of your enthusiasm about this stuff, keep it up. Also I worked at Signetics from 1978 to early 1984. I worked building a fab in Sunnyvale CA then moved to Albuquerque to build another fab. In that fab we had a PDP11-44 used for production and product inventory control. I didn’t have any interface with that but when you show closeups of the boards I see a lot of Signetics components that we built during that time. Also many years ago I had a big Healey wish I hadn’t sold it☹️Cheers
Absolutely wonderful!! When you get to the 4116 dram boards. Keep in mind they love to self destruct - they are very sensitive to bring up and take down voltage rail management. Turns out if +12 is applied before -5 on bring up it can damage the ics. And then you have to reverse the process when you bring it down. A replicated app note in MOS's data book sort of downplays this. I think I remember from watching the power supply repair that the power supply handles this correctly. The other thing that can be done is replacing all the 4116s with 5v only equivalent you can manually bodge 4164 and 41256 to fit in the same footprint (and they are generally cheaper and more reliable). If you start to see failure in the 4116s they tend to destroy other chips connected to the 12v rail. While I hate to not restore things to factory it's a really good idea to replace the 4116s if they start failing because it's typically a sign that they are all going to fail soon.
Apparently not having -5V applied first is bad for anything uses it. That's a big reason why I avoid 2708 EPROMs whenever I can. (that and not having a programmer that supports them)
Thank you so much! I've had a few 4116s go down on the Centurion memory boards before, but the modified 4164 replacement chips work an absolute treat. They're an excellent alternative and pretty common to boot. Once I figure out the halting problem, we'll start digging into memory tests and seeing if we can suss out if there are any bad chips floating around!
@@UsagiElectric I looked about a week ago and unicorn electric had tons of 41256 for very cheap. Free shipping to the Continental US. I just bought ram off them to fix that specy I got last year. That's were I learned about the 4116 bring up issue -- now I'm looking at making a drop in dc-dc board for the issue 3/3b spectrums
Congrats on solving the Power supply issue. This system is well build, it protects way more expensive cpu/memory boards from faulty power supply. One word of cation regarding Console connection. Based on number of wires at RS-232 cable it most likely uses hardware flow control. If so, merely swapping RX and TX will not result fully working cable but one must adjust other connections accordingly. Also the standardization was a bit weak so there were some 'odd' pinouts of DB-25 at various machines. Wrong cable *should not* break anything, but it's better to be safe than sorry. I do remember spending quite long hours testing and making special cables for various CNC machine to PC data transfer like 30 years ago..
the pin of the transistor (q27 pnp) that was making the lower one (q23 npn) into conducting to ground is the collector. The emitter is always the one with the arrow.
A PNP just spontaneously died? Doesn't look like it handles much current. I thought transistors rarely fail. That's so bizarre!! Congrats on getting to a boot prompt!
Oh, they most surely do. I think that attitude is a hangover from when they moved in over valves, when they randomly failed alot less, at least in the average techies experience. They still suffer wear, even modern sot-32 parts will either drift out of spec or just decide they no longer want to be a transistor any more. I'm not sure why, but PNP parts seem more susceptible than their NPN siblings. That's just gut data from repairing a few old audio things. Fortunately it doesn't seem to be a time bomb scenario, if one goes the rest are close etc. They just sometimes... Don't work any more. Or worse, don't work quite right, which can make troubleshooting a bunch of fun.
I really enjoy watching these videos today. Having worked with all these giant and loud computers, I am glad to NOT have any in my house today. My modern, totally silent machine will do me just fine. Ahhh.... silence is golden! But it was an exiting time back then! Of course, when the place I worked at got a "micro computer", all management did with it is figure out how to "micro manage" the staff!
Q27 is a PNP transistor. It's flipped upside down with respect to your regular NPN transistor, so the emitter is at the positive voltage rail. The arrow always points in the direction where the current flows as the base-emitter pair is a diode.
26:30 Thank you so much for sharing this journey with us. These machines were long before my time and this is a fascinating look into computer history.
Great job! It would be nice to see the before/after on the scope for the 555 so that we can see what effect the bad PNP was having on the output signal from the 555. Keep on truckin'!
Amazing stuff Dave. I might have similar issues with my 11/24 which also has a beast of a PSU, so seeing someone work on such a complex unit and get results gives me hope. That's a distant project though. I too grinned like a loon when the DMM showed the mythical 12V had returned 😁
Thanks for that debugging tour! You are inspiring me to get off of the dime. I have an 11/23 with the WOMBAT disk controller, an RX floppy, RL02 and a SuperEagle that can run RSX-11/M, RT-11 and Fuzzball. My long-term goal is to have the only GPS-disciplined Fuzzball NTP server on original hardware up on the Internet. (I also have a KS-10 but no drives. Trying to get access to plans for one of the two rumored MASSBUS-to- interfaces I've heard have been built. I am not sure one loves computers until you've had to move a Fujitsu Eagle or SuperEagle between houses at least twice. I mean, theyre a whopping 474MB. Too bad they're >100 pounds. My movers hate me.
Honestly man I love your stuff I have an Eagle 2 that often leaves me hopeless Im the only one known in my town to own one and its very rough not running so I am often feeling the same with it lol but seeing you bring these ancient beauties to shine again makes my 1982 eagle 2 project not seem so hopeless keep doing what you do man keeping the history alive !
Question: What is an Eagle 2? I had 4 of the same type of Eagle drives he has. But gave them to someone that could actually make use of them. Never heard of an Eagle 2.
@@timradde4328 If only i could explain it well enough a spin off computer from another company called AVL , in the early 80s only lasted til about 87 LGR has a good video on the history but yeah a CP/M system a copy of an IBM lol unique and short lasted company as many systems like it A fairly common system you would think but a surprisingly rare old computer
@@timradde4328 a unique computer I couldn't tell you enough about it enough lol I'm still learning myself ,although LGR has a good video on the companies history a CP/M computer and IBM clone practically lol
@@BluesToOL77 Oh, it's a computer. Cool. Still never heard of one. How did you come by it? I am a fan of CP/M but never really used it back then. I do have a Kaypro 2x that I find fascinating. Don't know what LGR refers to. Is that a channel here on YT?
It is really easy to get terminal buffer over-runs on the PDP-11 console. I would set up the Rainbow for 1200 bps, 8N1 with jump scrolling. If you’re doing XON/XOFF in-band flow control, it won’t be seen soon enough to prevent loss of data at the terminal. You won’t see it on the short lines spit out by the monitor program, but you will lose characters on program listings and the like. The Rainbow and DEC terminals of that era store the screen contents as a series of linked lists in order to conserve memory and provide a larger scrolling window in the terminal itself. Every now and then the terminal will run out of memory to allocate to constructing a new display line. When the unallocated memory runs out, the terminal will stop listening to the serial input and do garbage collection to consolidate enough fragmented memory to construct a new display line. DEC terminals apparently don’t do DMA of data coming in from the UART into a ring buffer so that terminal CPU can occasionally quit paying attention to the UART in order to do house keeping, HP terminals have a better architecture and do not lose data. Most HP terminals can keep up with a constant 9600 bps data stream without ever having to do XON/XOFF or RTS/CTS hardware flow control.
You and your accomplishments are epic. A couple things really jumped out at me. @20:06 - RS-423? I've only dealt with RS-232 and RS-422, and read up on RS-485. @21:13 1.5 stop bits??? How do you have half a stop bit (or half of any bit)?!? And of course, the epic victory @22:48. And that ADM 3A failed? 😞 A future repair video, I suspect? So happy you got this PDP working as far as this, and looking forward to the next step.
Half stop bits (actually 1.5) is an old teletype thing that we've been cursed with ever since it was invented. And I am also interested in what happened to the ADM-3A. It was said as a sort of throw-away line, but it should take something very interesting to cause a working ADM-3A to fail!
The existence of half a stop bit makes sense once you realize that they're using bit as a unit of time. At 9600bps 1 bit is 104.16us, 1.5 bits is 156.25us.
Wow, that was a deep dive into that 12V issue! Enjoyed seeing how that extra bit of diagnostic info (scope on startup) helped you narrow down the problem. Thanks for taking us along on this journey!
11:47 Q24 is actually a constant current source. This is the way it works: * The Zener makes a constant voltage of 15 V-3,9 V= 11,1 V on the base. * The voltage on the emitter is 0.7 V higher, that is 11,8 V * Therefore resistor R111 sees a constant voltage of 15-11,8= 3,2 V. * Therefore there is a constant current through R111 of 3,2/5,11 0,62 mA * This current is also the collector current because the base current is very small. * To my understanding, this constant current charges capacitor C40.
Firstly, thanx for being awesome! Here's A tip I learned while I was in field support for XENIX/UNIX systems and later an engineer at the world's largest maker of serial ports (I might be old...) for testing whether you NEED to swap pins on an RS232 style connector will seem obvious once you think about it. Power everything up. Stick a voltmeter between ground and measure to both TX and RX. You should see about the same voltage on each pin. Why? Because that side is driving the other pin to an idle voltage. If the pins are swapped (you have effectively two null modems) you'll see both the output from the terminal and the board each driving the TX lines (whichever line they're actually on) and each listening on the RX line. We used to carry little devices with LEDs on the important pins and if your cabling was basically correct, TX and RX will be driven to the same color. The fancy ones had jumper cables and dip switches so you could open and rejumper for quick testing without hand-making the kind of cable you showed. There was just never any consistency in doc saying "This pin IS TX vs. "This pin should be CONNECTED TO TX", so it was just defense to trust nothing. This is easy with the scopes and meters you have. Carrying on past your (deserved!) victory, if either side couldn't independently drive ONE of the pins, suspect the 1488 line drivers. If you could see a flicker when you pressed a key (you have access to scopes and analyzers that we didn't in the field) but the other side couldn't "hear" it, even if you were sure the cabling was right, suspect the 1489 line receiver in the receiving direction. The line drivers and receivers were essentially fuses for all the electrical garbage picked up over the mile of romex and chicken wire that people would inevitably use to connect serial devices. They were socketed for a reason. ;-) These same basic electrical principles govern RS-232 from the manly days of +/-15V on through today where we just bareback glorified TTL (which isn't even TTL any more; it's often 3.3V and sometimes even 1.8V on one board I have!) when you're attaching a GPS to your ESP32 or whatever. If the pin voltages don't approximately match, you have the TX and RX swapped the wrong number of times. I'd recommend having one of those testers in place any time you're connecting a serial device up the first time. Good luck and thank you!
I really love this series and the detail of your work. Really brilliant. Most of us will never get to touch this equipment but you bring your equipment into our home.
Mention of the 2114 memory chips gave me a warm fuzzy feeling. A pair of those formed 2/3 of the main memory of my homebrew SC/MP and later 6800 based machine. (The other 1/3 was four 2112 256x4 chips. Memory was expensive in those days!)
Omedetou! Good to see this old beastie finally showing signs of life. This is shaping up to be another great addition to your collection of minis. Can't wait to see what's on that Eagle drive.
I think you were clairvoyant in that you knew you could fix it 😉 You have the collector and emitter backwards. The line with the arrow is always the emitter regardless of NPN or PNP.
Next time you're looking for the RS-232 boot output, throw your scope on the data out line (TxD) and see what's up. Might need to tie down some of the handshake lines first, but you should be able to reverse engineer baud rate and at least see if there's activity. Also, 9 bit memory is usually used for byte+parity, so if you want to monitor for errors due to cosmic rays, power supply glitches, or flaky TTL chips or RAM cells, you write a byte and write the parity in the 9th bit, then when reading, if the parity doesn't match the data you know something got glitched. Happy to see you bring this guy back to life!
My father always talks about learning to program on a PDP-11 - using punch cards and punch tape! It's interesting to see one of these machines finally.
Something I would greatly appreciate is a summary of the faults/replaced components on a particular part of the system at the conclusion of a series of episodes focused on it. In this case, the power supply. I know this ep covered a blown fuse and bad transistor, but I have since forgotten the previous power issues in the episodes leading in lol
I built an octopus curve tester to be used on in circuit transistors when I was a Diplay Tech in the Navy. It would check for open or shorted transistor. The computer was all transistor. 32 K of 8bit memory.
I'm really happy you got this beast of a machine running, so many things to get right... Congrats! A couple of very minor corrections: First on the PNP transistor the leg with the arrow is always the emitter not the collector, regardless if it is NPN or PNP... you had them labelled backwards in your diagram, the data sheet had it labelled correctly, regardless, you still found the faulty transistor!. A NULL modem connection doesn't just swap TX/RX but the handshake signals as well... sometimes if the CTS doesn't go low (TTL logic) then the UART won't transmit if flow control is enabled... Don't ask me how I know this, we've all been there LOL. Still I'm amazed you got data coming out to your terminal, they must not be set to honor the flow control signals... or you swapped more than RX/TX, the signals as well, and just mentioned on the video that you swapped TX/RX and left the other details out... only us old guys that have fought with RS-232 in the day know the real pain of stubborn hardware... LOL.
Having feedback circuits, basically a loop of dependencies is a hard thing to diagnose. Sometimes the quickest way is too just desolder all transistors and just test them with a transistor tester. With a proper desoldering tool that can be done really fast. If you do one by one, just might be lucky and find the broken one long before the end. Same for other silicon.
@UsagiElectric Your hope to "see you all next time" continues to be frustrated. We shall need to wait until our cameras are working full-duplex, & you have a monitor with enough pixels to display at least a *thumbnail* of each viewer... Plus a processor with more memory (not to mention speed) than that Bendix with its rotating drum...😉😉 But congrats with the PDP 11/44!
You've successfully brought me back to my high school years, when I occasionally got to see a field technician fixing the school's HP 2000 mini. I never had enough free periods to watch him do a repair start to finish, so watching videos like these is fascinating to me on multiple levels. So, here's a question: The year is 1982, and Company A's PDP 11/44 has gone down. Their DEC field technician is paged, shows up shortly thereafter and diagnoses a fault in the power supply. Does he a) swap in another power supply, bring the mini back up, and bring the failed power supply back to HQ for repair, or b) pull out a soldering iron, open a veritable tackle box of components and get on the phone with an engineer at HQ who helps him through fixing the fault? My question isn't intended to be rhetorical. In my experience watching minicomputer field technicians from 1980 to 1993, repairs could take anywhere from a couple of hours to several days depending on what was wrong and, probably, what sort of service contract was in place, and I rarely understood even half of what those technicians were doing. So I really want to know.
Thank you! I think it depends on the machine, but for the 11/44 the troubleshooting manual actually states that the PSU plug-in units should just be replaced. So the tech would diagnose the PSU enough to figure out which plug-in unit had the fault, then they would just replace the bad unit and get the machine back up.
@mark4l, @usagielectric: Component level diagnosis was rarely performed in the field. If you had a service contract, the goal was to get you up and running within N hours, even if it meant someone was booking a ticket on a plane to carry a replacement board in their briefcase. With power supplies, in particular, they're almos always swapped because A) they can be hard to do component level diagnostics because the PS can be the cause (as shown here) or the victim (some component on the backplane can bring the signals down) B) A bad PS can take out other systems. Imagine the carnage of that 35V on the 5V rail. (Or imagine a 15V rail that's sagging to 10 intermittently. Is it wimpy on the output or is something else dragging it down? Did you bring a big fan-cooled resistive dummy load?) and C) safety concerns. Because these aren't exactly 10W wall wart phone chargers, they're big beefy supplies with grownup-pants power involved and customers don't LIKE popped breakers, the smell of smoke, or (gasp) fire, they were best to just swap out. Now back at the regional service center, there was a guy that had a big stack of these on one side of his desk that were broken that he slowly converted to a small stack that were working and then handed back to the field techs for their trucks/kits for deployment in rapid swaps. Those guys had the special soldering iron tips that could heat all the pins on a DIP package at once to get it out in one shot, scopes, analyzers, and all those other toys you really didn't want to drag onto site. As I mentioned above in a comment, I was in the business in this approximate era. Our regional service center was on the opposite side of a wall from us.
You want to get a faster acting continuity tested and perhaps even a faster autoranging multimeter. You will also save a LOT of time if you invest in or make up a box of LEDs that will monitor RS-232 signals on the D25 connectors. If you plug it onto a cable you will immediately be able to see that the correct handshake signals are lit and the polarity. Swapping to the other side you can detect that the other pins are providing signal voltage and there are no clashes. This will test all the signals at a glance without having to guess or measure individually or try to remember which way the device was or should be wired.
If you have a TU58 that would be a good place to start especially if you can get hold of a diagnostics tape. The TU58 is Bootable through the console device and only needs minimal hardware.
Congrats! I marvel at your relentlessness and thoroughness in debugging hardware. I don't have the patience for that, which is why I switched from computer engineering to computer science in college. 🙂 It does seem like you have a huge community to bounce ideas off of, though; I'm sure that helps.
Congratulations! Getting a power supply is working can be H A R D ! It's insane how they have managed to get all that functionality into a single chip in a mobile phone charger, even with the little bit of cheating they can get away with (it's supplying few enough watts not to matter that the transformer isn't passing any energy through it for at least half the time, and more like three quarters or five eighths on 230V mains. That's not wasting energy, because there is no current in the primary side most of the time, but it won't scale well). Any money you like says the fault is a bad connection somewhere. A switch that needs a drop of contact cleaner, maybe, or a dry solder joint. The way the CPU includes a little microprocessor to talk to the terminal Anyway, looking forward to the next episode ..... If you want a clickbaity title, you can always say it's a video of you solving the halting problem ;)
When the PSU is switched on at 4:35, the output voltage run-away already van be seen on the voltage meter. If you only had seen this at that moment, of would have saved you so much time!
(@7:58) Before looking at the signal path going through Q23, I’d first check R112; if it’s open, R93 will keep the base of Q10 logic low. Also, Q10 could have a base-emitter short, or R93 could be shorted, or very low resistance, making it difficult for the 555 to drive the base to logic high. This could be checked down stream of R112 if you cut the connection between the junction of R112 and D29, and the junction of D23, D26, and Q23. Finally, one of D28 or D29 could be open. 😊
4116 memory is quite fragile, if one of the voltages are missing they tend to fail. I hope all those chips work! 500ns is very slow, I think the 4116s I have are all 150ns. I'm amazed that a 555 is used in a power supply like that for a "real" application, I thought the 555-based nixie tube boost converter I built using a 555 was pretty sketchy.
Man...I saw that sea of 4116s and got really scared that the missing 12v rail killed a few at some point in the past....but hopefully a PDP-11 is a bit cleverer than a ZX Spectrum ;)
I tried to give you a null modem adapter (the very on in the article listed, but only 9 pin) at VCF Southwest, but such is life. So very happy the old lady came up.
Great video and success! Your videos are well done and narrated. Congratulations on the fix. Question, after you get all these running, what are you going to do with them?
