We Used an Oscilloscope and Magic to Rescue 60 Year Old Data!

Added: 2026-05-18

[00:00:00]Welcome back to the LGB21. In the previous two episodes, like those episodes were absolute bangers. There is no way that I'm ever going to be able to top the excitement of uh what we were doing in those episodes, which was figuring out what was on the memory of this machine. This machine uses a rotating disc memory that is nonvolatile. And all of the data from when this thing was in use in the 1960s at the Los Alamos Scientific Laboratories was still there intact on the disc. We saw references to the UHTREX, which is the ultra high temperature reactor experiment, as well as the sodium molten plutonium loop experiment. And as we dig through it, we're finding more and more exciting stuff, like we found a reference to TNT.

[00:00:44]Like, why are they dealing with TNT?

[00:00:47]That's pretty wild. Uh, but ultimately, there was four tracks that we were not able to uh pull the data off of on this thing because it had bad heads. Uh, now we only had two bad heads and that's because two tracks are interled under one head and so we moved those over to the spare tracks and the disc is now fully working but the data on those four tracks was unreoverable.

[00:01:09]Maybe we're going to get into that uh a little later, but there is a priority list of things that have to happen here.

[00:01:15]And uh the first is that there's only like two episodes left before VCF Southwest that's coming up on May 29th to May 31st. And this LGBT 21 is going to be at the show hopefully fully up and operational for all 3 days. And so if you want to see it and come and use it and play with it, come down to VCF Southwest. Hopefully it'll be absolutely epic. Uh like I said, there's a priority order here because uh I think the CPU is probably going to be totally fine for all three of those days. But that that flexo rider, man, this thing is already given us plenty of grief. The power roller is blasted. It's like pretty much unsavable. Uh full of cracks and it just not in good shape. It's going to have to come out and get uh recoded, redone, refurbished. I don't know what the correct reword for that is, but uh that can't happen before VCF Southwest.

[00:02:11]There's just not enough time. So, we just hit it with some rubbernew to kind of get it going well enough to hopefully get us to the end of the event. And then after VCF Southwest, I'll pull the whole thing apart and we'll send that off to get redone and it'll be like brand new when it comes back. Uh, also there's a fullon cam shaft in here with four loes and that was fully seized up. Nothing was working like it was supposed to and we broke all that free and that allowed us to uh start inputting and outputting and doing all the stuff that you're supposed to be able to do with this thing. But the problem we were seeing last time is that every time I would press zero or W, it wouldn't shift the value into the accumulator and the CPU would get stuck in this kind of weird loop with the stop light off. It was actually running. Every other key on the keyboard works. Everything works just fine. It's just zero and W. And there's nothing unique about those two keys, at least from an electrical standpoint.

[00:03:08]Every single key on the keyboard activates an encoder bar that makes or breaks contacts on the side of the machine. And this creates the sixbit value that gets sent parallel over to the CPU. Now, when we're inputting into the accumulator, only four of those bits are read, and those four bits are not unique in zero or W. So, whatever it is has nothing to do with the encoding that's going on. But it also has nothing to do with the timing because every other key kicks off the timing in the exact same way. It has to be something mechanical. The when I press the key, it does actually type on the paper, but maybe the type bar is not engaging with the encoder bars correctly or maybe it's not kicking something off. I don't know.

[00:03:52]Something is wonky. So, I think our best bet is to uh pull all of the covers off of the Flexo Rider, pull the encoder down, and just clean and lube everything. Again, just a quick reminder, this Flexo Rider weighs in at a ludicrous 85 lb. That's about the same as 260 cinnamon rolls or 600 donuts. So, if you're planning on going on a pastry binge, remember, lift with your legs and not your back. And and actually we like we don't really have these here in the States, but every time I go back to Japan, I got to load up on balm rolls.

[00:04:27]They're like bite-sized versions of what they call baluhen, which that's really just a lone word from the German bomb kuhen. I'm sure I'm mispronouncing that, but these are also sometimes referred to as tree cakes because they have rings that resemble a tree. And traditionally, it's baked on a spit roast by brushing on layers of batter. And during World War I, Carl Joseph Wilhelm Yu-Him found himself in Japan for reasons. But eventually he started baking baluhan and even opened a bakery. Then there was a big honken earthquake and another world war. But after that he and his wife opened up a chain of bakeries in Japan under the family name uh which not only made Bam Kuhan one of the most popular pastries in Japan, that bakery chain is still around today. Oh, right. We're working on a flexoor writer. Uh, I gave the contacts a good polishing. Uh, though I'm confident the contacts themselves are not the problem here. I also gave the power roller another run with rubber renew. And while that soaks in, I want to wrap the cables that we made in the previous episodes properly. This is just spiral wrap that I picked up at the local hardware store. And I'll secure it at the end with electrical tape, but the rest is free to flex and move about. It was tedious work, but it was one of those things that I definitely needed to do. All right, I haven't changed much, but uh let's give it a try and see what happens. We're in manual input. If I go ahead and hit a zero, uh the stop light stayed on. Oh, there.

[00:05:56]Well, there it went.

[00:06:02]Oh, it's definitely a mechanical issue.

[00:06:10]What is that? Something is sticky. Oh man. Okay, it's uh right it's something something's going on right here right on the side. This is the uh the six bit encoder. And if I hit something like the number two or the number three, you can see that it encodes. Now watch when I hit zero. If I hit zero here, they go forward and stay stuck. And there. There we go. We saw them. We saw them kick back just then.

[00:06:36]They're slow. Sometimes they are, sometimes aren't. There's W. And you can see that they're stuck forward.

[00:06:43]And I don't know what's causing those to get stuck like that. Well, I we're zeroing in on the problem at the very least. At this point, I'm like 99% certain that the problem is the power roller here. It's all cracked and gross and smooth anyways. And I think the only way to fix that is to get this thing uh refurbished, which we're going to do, just not before VCF Southwest. So, I'm going to give it another uh sanding here. try to put a little key in it so that it'll grip the feet a little better and then I'm going to go over it with some more rubber renew here and let this soak in. So, let's move on to the next thing that I wanted to try to get to, which was this rotating disc memory.

[00:07:18]Again, in the previous episode, we discovered that we were able to pull out 60 of these 64 tracks. We were able to dump those. They all read perfectly, and we've been going through the data and figuring out all sorts of amazing stuff.

[00:07:31]But those four tracks that we were unable to recover came back as all zeros. Now, it's a little tricky because there's only 32 heads on the disc. So, that means that two tracks are interled underneath one head. So, if a single head goes bad, you lose two tracks. And that's exactly what we saw. We lost 0001 and 2 Q and 2 W. Now, the head for 00 and 01 measures out a little weird. I get like 38 ohms across uh half of the head and it should be 17 ohms which means that it's just high enough resistance that the amplifiers are probably not picking up anything. And also there's something tricky going on with 2Q2W because even though it read as like 8 megga ohms to 10 megga ohms essentially open that was measuring from one side of the head to the center tab.

[00:08:20]If we measure across the whole head it tells a different story. The head for 2Q and uh 2W is right here on the very end.

[00:08:28]And if I put one probe on the center tap and measure from one side of the head to the other side of the head, I my my multimeter isn't even picking it up. But if I skip the center tap and measure across the entire head, uh you can see right there 36 ohms. Actually, the coils of the head are good. And all that's happened is our center pin connection has failed. Now, I am not going to dive in here and try to resolder that center pin connection.

[00:08:59]That's just a recipe for destroying everything else in here. But I could hook an oscilloscope across the entire head and potentially get the flux transitions off of the disc. Now, from a usability standpoint, this is a totally unnecessary step because we were able to move the wire that goes to the bad head here and move that over to one of the spare tracks. We did the same for track 00 and 01 as well. And that means that this disc is 100% usable by the CPU. This is only to try and recover the data that was already on those tracks. And so if we pull out the scope and take a nice long scope shot of the entire revolution of the disc, capturing all of the flux transitions and compare that against a clock signal, we can then go through and decode the individual binary bits on those four tracks and safely recover them. All right, I've got it spun up and I've got uh a whole bunch of scope probes hooked up to it. And you can see that we have uh the the CP signal. This should just be the bit clock. Um, but the real trick is going to come with the rest of my scope probes here. So, let's turn on number two. This should be the sector address. Uh, and I'm going to have to readjust everything pretty dramatically here. And if I take a single shot of this, uh, yeah, there we go. That I mean, that looks like kind of what we're expecting to see of the sector address.

[00:10:24]Now, here's where it gets really exciting. Number three, are we going to see raw flux transitions? And we are not. How about number four? Are we going to see raw flux transitions on that? Uh, no. We are definitely not seeing raw flux transitions. That is a bit of a shame. Oh, wait. Maybe.

[00:10:44]Maybe. Uh-oh. All right. All right. I just had my scales all messed up. Um, let's move that one down. Let's bring the scale on this one. Yes, that's a raw flux. Let's take a Let's take a scope shot. Look at that. You can see the actual raw flux transitions. Oh, that's so exciting. All right, now I got to uh I got to zoom this way out far enough to get um a couple of rotations of the disc and then take a a scope shot of that and then we'll be able to see uh all of the raw flux transitions per sector address along with the bit clock.

[00:11:25]Let me go into a little more detail about what just happened because the trickery that we just pulled off is a little mental. The head is essentially a single coil of wire with three taps on it. One on each end and one in the center. This tap in the center is where our fault is. It's not making a good connection to the upper side of the disc. But crucially, the windings of the head itself are totally fine. Now, I'm not going to take the disc apart and get in there and try to reattach this center tap. Way too dangerous. Puts the entire disc at risk of being destroyed. So, we have to do something else a little clever here. And what we can do is we can actually isolate the head completely by snipping the diodes, making that head essentially floating, and then connecting our scope probe up to it. And now whenever that disc is sliding by that head, if there is a change in the raw flux on that disc, it's going to induce a little bit of current into that coil of wire and our scope will pick that induced current up. And we see that as these like little upwards or downwards blips. And that's because the raw flux is changing from one orientation to the next or vice versa.

[00:12:39]And so if we can figure out how to time that correctly by putting another scope probe on the bit clock and the sector address, this tells us where the start of a sector is, where the end of a sector is, and then we can look at the raw flux transitions and figure out where the ones and zeros are. Now, this this is the really tricky part because there's a whole lot of bits. So there's no way that we're going to be able to do this visually. We need to be able to take that data and and put it into some kind of file that we can then analyze with a modern computer. And uh that Ksite scope, massive shout out to Ksite for donating that epic scope. It does like 5 giga samples a second, which means that I can zoom it out far enough that we get about two or three revolutions of the disc on one screen.

[00:13:28]And then I can do a single shot of that.

[00:13:30]And there's enough resolution in the the quality of the shot that it takes for me to be able to zoom all the way in and see the individual flux transitions. So now I just need to save that to a commaepparated value file and store that to a USB. Now I've taken that commaepparated value file, combined it, and we can see it here on the laptop using a program called Pulse View. Kudos to Roy for helping me get Pulseview set up here and getting the CSV file combined so that we can actually load it into this application and view it. I've got the entire scope shot here in the program and I can zoom in and zoom out and move around and look at the individual traces as if it were on the actual oscilloscope. So, how do we figure out what the actual data is by just looking at this scope shot here?

[00:14:17]Well, if we go over to our manual here, there's an outline of what the sector address S1 should look like. So, now if we look right to left in our scope shot over here, we can find there's our one start bit. And then we have one, two, three zero bits. And now we start the strange sector address that has one of the bits always set to one. Then we get all of our zero bits. And now we have the real sector address, which is going to be 1 0 0 0 1 1. And then we have two zeros at the very end that line up with the least significant bit and the spacer. So now we know exactly which sector we're on. We know where the bits are lined up with. And so we can look at the flux transitions. And we can see that there's a spike downwards here. So that's probably a transition from a one down to a zero. So now it's going to be zero for all of these bits until we see the next flux transition up. So 0 0 0 0 0 all the way until we get to this point. We see another transition that's going to pop us up to a one. And that's immediately followed by another transition which is going to pop us back down to zero. And then uh no more transitions for the rest of the words.

[00:15:29]So we've already figured out one of the missing 256 words that we weren't able to recover. Now, Ree has been doing some amazing work over on the Discord. He started working on writing a Python script to extract out the data, but using this, he's managed to pull out this data right here. This looks like legitimate real deal LGP21 data to me.

[00:15:54]So, what's in this data? Well, we like we don't know. We don't know yet. We haven't had time to analyze it. I am the living embodiment of just in time manufacturing. I'm very often still editing a video on Saturday night for release the next morning on Sunday. So, the information that you guys are seeing in the video is often the newest, most up-to-date information. And uh well, you know, we're like I'm still filming right now and like we haven't had time to analyze it. So hopefully next week we'll have a better idea of how this data fits into the larger picture as a whole and uh we can figure out maybe if there's some interesting strings hiding out in here. Now, the only program that we've run on the LGB21 so far is the memory dump program, which is an awesome program because it exercises a huge amount of the machine, but it's not a very good demo program for an event. So, the mad lads over on the Discord have been building up a collection of software that we can load into the LGB2 and run at VCF Southwest. That should be pretty exciting to watch. And I want to load one of those programs up right now and see how it works on the machine. But the only way to load new software into the LGP21 is via paper tape. So, I've got my laptop set up with their files on it, and I just need to punch them out on this paper tape punch right here, and then we can load it in. So, I'm just going to go ahead and hit enter.

[00:17:34]All right, I think I've got the uh bootstrap loaded in correctly. It was eight words. Uh I've I've had trouble typing that in in the past. Uh but I think this is correct. So I need to come into one operation, hit an execute, and then our program counter updates.

[00:17:50]And then I need to go to normal and hit start. Yeah. And it's starting to read in the tape.

[00:18:00]And um that's not great. It stopped halfway through.

[00:18:08]That's a new one for me. Okay, that last program was a Serpinsky triangle. That's one of the demo programs that we wanted to run. Uh so maybe we've got something wrong with the program or with the tape.

[00:18:17]So I've got a different tape loaded up.

[00:18:19]This is Fibonacci. So the bootstrap should still be in memory. I just got to get to it. So, we got to do a 0000 uh U3s 000.

[00:18:30]Uh, and then I need to do a fill clear.

[00:18:33]Uh, and then I can fill that all in with zeros on the bottom there. Um, and then we go to one op execute. We see our program counter update. We go down to normal and we hit start and it starts reading in the tape. We can see it uh shifting bits in and shoveling them into various parts of memory.

[00:19:02]It's so cool. Almost finished with the tape. And the typewriter has come alive.

[00:19:08]We did one one two three five 8 K 15 22 237 It's doing Fibonacci in hexadeimal.

[00:19:49]So there we have it, the LGBT B21, printed out the Fibonacci sequence up to some incredibly large number. It's all in hexodimal and I like I can't make any sense of that. But I ran the program multiple times and it came back identical every single time which is an incredibly good sign. There are more programs that I would like to run on this. We got to figure out what's up with the Serpinsky triangle. That's going to be kind of one of the main demo programs I want to run at the event. Uh I also would like to run one-dimensional Conway's Game of Life as well as a Mandler set on this thing. So that way we have a couple of different demos that I can uh punch out interesting things and hand that out to people at the event at at VCF Southwest. Uh now talking about punching out, the typewriter is a bit of a hot mess, but it can actually do typeouts totally fine. I probably need to change the ribbon. I do have a spare ribbon, uh new old stock ribbon over there, so I'm going to take that with me. And then when we get back from VCF Southwest, uh we'll hunker down and figure out exactly what's going on. I think it's probably all in that power roller. Once we get the typewriter fully up and going, we're going to uh then figure out the reader and the punch on the typewriter and then move over to the uh tally punch on this. And then that's like that's it. The LGB21 will be 100% fully working at that point, which is really exciting. So uh next week, that's the last episode before the uh VCF Southwest event. We're going to be right back here right on this machine running those additional demo programs that I want to run and uh probably doing a little bit of cleanup and small cosmetic things to get it ready for the show. So, I want to thank you all so much for hanging out with me today, and I hope to see you next