Advent of Computing - Episode 172 Analoghybrid

Episode Date: January 11, 2026

In 1945 the first electronic digital computers sparked to life. Number crunching was instantly changed forever! The perfect technology had arrived, and there was never even a competition, right? Well, not so much. The simple fact is that computers sucked for decades. Digital machines have all kinds of inherent pitfalls. There was another entire lineage of computers that existed in the shadow of digital machines: the analogs. Eventually the two technologies would merge in an attempt to create the truly perfect machine: one with the flexibility and accuracy of a digital computer, and the speed and interactivity of an analog computer. The result were hybrids!  

Have you ever noticed how annoying fractions can be? Well, at least if you're using a calculator or a computer. Some will work fine, right? One half turns into 0.5. There's no mess, no fuss there. 1 over 4 equates to 0.25, just fine. But then you get to the weird ones, like 1 over 3. We can understand that value very easily.

It's a third of something, which is a little less than half of something, a little more than a quarter. For a computer, though, turns out that one-third isn't all that simple. The machine may spit out something like 0.33333, or even 0.332. That on its own is kind of annoying, right? But it gets worse when you use that value for more operations. That weird decimal will pollute everything else you do. You can even run into these strange errors,

like, for instance, adding two-third and one-third and getting a number that isn't exactly one. Why are computers so dumb when it comes to thirds? To put it simply, computers have to approximate numbers. Or, said with more sophistication, computers simulate mathematics. computers don't really like working with some numbers because they're already doing tricks to handle mathematics at all. It can be easy to stretch those tricks a little bit thin. As we enter an edge case like this, we have to get a little more specific than usual.

This is a time where we must invoke the full name of the beloved computer. This oddity occurs specifically with electronic digital computers. Those have become the default type of machine, so we don't usually need those qualifiers. But today we do. It turns out the digital part can lead to some strange issues. Is there another option? Is there, perhaps, another type of machine that could handle divide by thirds? It turns out there is, but the non-digital path comes with its own, actually very unique hazards.

Welcome back to advent of computing. I'm your host, Sean Hass, and this is episode 172, Analog Hybrid. I also need to apologize for the lateness of this episode. I was out on a business trip, and I assumed I'd have the time to deal with prepping the episode, but between meetings and being at a trade show for a week, I did not have the energy. As is tradition, my poor planning upsets my schedule every time, but we're back. And today we're going to be discussing something that's particularly fascinating, at least I think. We're going to be looking at hybrid computers, specifically focusing around electronic associates

incorporated analog hybrid machines. I say focus here because EAI is going to serve as the window we use to look into this not quite digital world. That doesn't exactly mean this episode is about EAI itself. So how do we get here, you may ask? Well, it all comes as the result of some ulterior motives on my part. I've been wanting to learn more about analog electronic computers specifically. The big reasons for this are They're just so alien to me

I had a bit of extra time off work around the holiday So I decided to buckle down and You know maybe learn something for once in my life I have a copy of Corn and Corn's electronic analog computers That's been sitting on my shelf for a while So I've been reading up While going through Corn and Corn I've been trying to find a specific machine to focus

on for an episode. That, however, has turned out to be pretty difficult. Electronic analog machines weren't mass-produced at the same scale as later digital machines. And while there were some huge one-off installations, they aren't as well documented as digital computers. Even once we get into mass production, there's still kind of a weird gap in documentation. at least I find it a little strange compared to the great preservation we have around a lot of digital computers. It was on this search for a target

that I ran across a video of all things. In the 1960s, EAI produced a presentation called Understanding Analog Computers. It specifically explains the principle of operation behind their analog hybrid series of machines. These are analog computers then incorporate some digital features. And, oh, what can I say?

I was hooked. I've only really ever heard the term hybrid computer. In my head, that kind of merger isn't possible. Analog computers work on totally different principles than digital computers. Never the twain shall meet, right? So today, we'll be looking at these hybrid machines. How exactly do we get to these blended machines? Are they hideous abominations?

Or are they a lost secret technology? What I really want to get at is how this blending was developed and what it was specifically used for. Because when you get down to it, the detail is really what explains a lot of this type of strange technology. There's also an interesting side question here. Did you catch the year of that video? It's from the 60s. That's well into the digital era. If a digital computer is actually better than an analog or hybrid system,

then why would anyone build an analog computer at all? How could an analog survive post-Iniac? War veterans form electronics company, Princeton, New Jersey, a group of Army veterans who formerly were radar, radio, electronic, and communication experts of the Signal Corps have formed a new company, Electronics Associates Incorporated. It was announced today. Leo R. Jensen, former commanding officer of the new equipment introductory detachment, Fort Monmouth, New Jersey, has been named president of the new organization. During the war, the veterans were engaged in the supervision of design, manufacture, installation, testing, and instruction of personnel on all types

of electronic equipment. A laboratory will be built near here with designing, drafting, and office space. New problems, modifications of Army or Navy equipment, and design of new devices will be carried out. End quote. That notice ran in the Dayton Herald in 1945. Actually, it ran less than a month after the end of the Second World. war. This was the start of EAI. This company would go on to be a titan in the analog and hybrid space, but this is how they initially

started. EAI dealt more in mundane electronics. We're talking test equipment kind of stuff, volt meters and power supplies. There are also early ads for some EAI radios. So this isn't exactly the

new kind of technology they'll set the world ablaze. Now, as mundane as this sounds, there is an immediate connection to the digital realm. The crew that started EAI were all radar technicians. Many digital pioneers would also cut their teeth in radar. I know I've said before that if you can work a radar circuit, then you know all you need to create a computer. You've worked with vacuum tubes and you've worked with memory systems. That's just a short hop away from something like ENIAC.

In fact, some of the people that worked on ENIAC and designed it got their experience in radar. It turns out that radar is also very close to analog computation. In fact, it's probably closer to the analog than digital. Consider, if you will, just what a radar dish does. It produces, receives, and analyzes analog signals. What I find so interesting is that this means radar technology wasn't just an early seed for digital machines. This really speaks to how history is not a straight line. It's more of a meandering path.

Events don't just exist to kick off the next in a series down some line. But I should digress before I get too whimsical. The location of EAI early on is also a bit of a tell. At least one of its founders, Lloyd Christensen, was MIT educated. During the war, Christensen was trained at the Army's electrical engineering program at MIT. Given that EAI was founded in Princeton by a pile of army vets that work in electrical engineering, I would bet that more than just Christensen went to MIT. While EAI didn't start out as a computer outfit, by 1947 it was developing an analog machine.

By 1952, EAI was selling an entire range of analog computers. It's at this point I should probably explain a little bit about what makes electronic analog computers unique. An electronic analog computer represents numbers as voltages. In the digital world, we use set voltage levels to represent binary numbers. It's pretty common these days, at least, to use 5 volts for true and 0 volts for false. You actually technically use a range around 0 and a range around 5, but the point is you have fixed cutoffs for values. Analog, on the other hand, uses a continuous range. To represent the number 10, you could run, say, 10 volts down a wire.

9.5 could be represented as 9.5 volts and so on. In practice, a machine will have some established voltage range, and numbers will all be in reference to that range. So if you have the number 10, then that will be called a machine number. internally, it will be represented as some other voltage. Sometimes these were pretty high voltages, too. A machine number 10 may be 200 volts in some of these things. Circuits are calibrated in such a way that it gets displayed and treated as the final value, that 10. In that way, we have, just like with digital computers, an internal and external representation of data. A digital machine will represent

a decimal number in binary. But it usually doesn't spit binary numbers out at a human. I point this out because, time and time again today, we're going to see how analog electronic computers are remarkably similar to digital computers. These are two forks on a tree, not two unrelated technologies. That said, these are two pretty distant cousins. This becomes especially apparent in looking at how analog machines operate on numbers. One of the major breakthroughs that happens with digital computers is the use of a basic switching element.

Math, logic, and control circuits are all built from the same switches, be they transistors, vacuum tubes, or even something a little more exotic. That allows for very flexible circuit design. When it comes to digital computers, it really is switches all the way down. Analog computers don't work that way at all. Everything is special purpose. Let's take division as an example, since that's the example I understand the best. This uses a voltage divider circuit.

It's actually just a resistor that's tied to ground. You send your input voltage to one side of the resistor, and you can't. get your answer on the other side. The resistance of that resistor is your divisor. What makes this all work is physics. In fact, it's all electronics and magnetism. A series of courses that I only barely passed in college. In fact, my professor threw chalk at me a few times during those classes. Anyway, this all relies on the fact that E&M is very, very, well-understood and very well-defined, mathematically speaking, even if it's not very well-understood by me. We can describe any circuit using a mathematical equation. At least most of them. I'm sure

there are some edge cases where you have to use numeric analysis, but that doesn't matter right now. Let's not think about that. We can describe circuits as mathematical equations. Simple. electronic analog computers work by flipping that around. If you have a little imagination, then you can actually describe an equation using a circuit. On its own, not super useful, right? I think if I had ever turned in calculus homework in the form of a circuit diagram, I would have probably had more than chalk thrown at me. Okay, we have this established way to go back and forth between equiolns.

equations and circuits. It gets useful once you actually pump some power through those circuits. That will give you the solution to the equation. Take the humble voltage divider. Do you know what 10 divided by 2 is? If you somehow forgot, then you can grab a 2-oam resistor, push in 10 volts, and then read 5 volts out the other side. Now, in practice, we have the whole deal with machine values,

values versus internal values. You aren't actually using 2 volts and 2 oms, but you're using some calibrated equivalent that the analog computer deals with for you. As I said, we have equivalent circuits for every mathematical operation, since, you know, we can model circuits with mathematics. Where this starts to matter is when we get to calculus. Circuits with combinations of capacitors, resistors, and amplifiers are used to model derivatives and integrals. With a little bit of doing, you can solve an integral using a capacitor and some vacuum tubes. To me, that's almost magic, right? Calculus is always the end point of this kind of stuff, because, well, once you get to calculus,

you can model just about any physics you want. This is generally applicable as long as you steer clear of really exotic things like nuclear or quantum physics. But who would care about that stuff anyway, right? The shorter summary that's usually used is this. An analog machine can be used to construct simulations. These are direct simulations constructed via analogy. You can just rig up a circuit that simulates the trajectory. of a projectile, accounting for gravity, drag, and starting conditions.

This is subtly different from how electronic digital computers operate. A digital computer simulates mathematics. So any simulation you make is, by necessity, an approximation. It's also an indirect simulation. To be a little bit of a nerd here, mathematics operate over the set of real numbers, at least most common mathematical operations operate over reels. That set is continuous, not discrete. A digital computer operates on discrete values. That means we only have a subset of real numbers that we can work with. As such, we have to approximate. Try dividing one by three and see what you

get. Analog, by contrast, lets you do direct simulation. That is both a blessing and a bit of a curse. The blessing is that you can, in theory, get much more accurate results. There's no approximation involved. The curse is that for accurate results, you have to have components with very high tolerances. But that's all theory. How does this actually get done in practice? Let's look at one of EAI's most popular machines. The Pace 231R. This computer, well, it's a bit of a beast. I know whenever we talk about early digital machines,

you kind of have to say how huge and power-hungry they are. Well, the 231R is up there. This machine comes to market in 1952. It has two big claims to fame. The U.S. Navy used a 231 while designing the same. submarine Nautilus. That's the first atomic submarine. North American aviation used this computer while designing the X-15, a high-speed, high-altitude aircraft that flew to the very edge of space. So this is a very serious machine that was very seriously used. First off, this isn't programmed in the

digital sense. Rather, you build up equations by connecting discrete components. This is a very serious components. This is done at a patch panel. You have access to all these discrete components, and you physically connect them using these plug cables. If you need to divide a number, then you run a wire to a segment on the patch panel that says divider input. It's pretty clear that this is different than programming in any kind of modern sense. By 1952, we don't exactly have Fortran, but we have assembly languages. However, EAI's documentation does call this patch cable work programming. They're even doing so up into the 1960s. So they are viewing this act as the same as, say, writing machine code. In many ways, this would be very similar to early digital computer programming. The patch panel

on the 231R is removable, so a programmer could sit as their desk and wire up a program. This could be done offline, as in without access to the computer itself, without the computer running. This is how programmers worked in the 50s and even in the 60s. Before time sharing became common and before personal computers were a thing, a programmer would work away from the actual machine in offline mode. The experience here would have felt pretty similar. You need to model an equation. You take your tools, either a patch panel or a card punch, and sit down at a desk.

You work out how to tell the computer what to do, then you get your program executed somehow, get its results, and then go back to your desk and figure out what went wrong. I feel like my definition of programming may be getting more broad over time. However, this does really feel like programming when you're looking at the this zoomed-out perspective. The only real difference is the target. In the 50s, there was so much diversity amongst computers that, well, maybe it's not that weird to consider wiring a 231R programming. It's in the details that were snapped back to reality. The simple fact is that analog computers had some fundamental eccentricities. One key spot that's visible is,

the matter of accuracy. On a digital computer, you ensure accuracy by writing more complex and longer running software. Basically, you can make your approximations better by running the computer a little harder. If you only need two decimal places, then you can simulate in maybe a minute. If you need accuracy to three places, you might need to run the machine for 10 minutes. Analog machines are much more physical. Accuracy is dictated by things like signal interference, tolerances, or even temperature of components. This is from a brochure for the 231R. Quote, to provide protection against crosstalk, the Pace 231R, features elaborate signal shielding from patch cord to components.

All patch cords and bottle plugs are of coaxies. type. A unique brass grid work forms individual shielded cells for the 3,450 contacts in the patch bay. To assure maximum accuracy at low voltage, each of these contacts is plated with 24-carat gold. Each of the precision computing networks is individually shielded in a metal can, end quote. Ah, to spend your days in an individual can with your contacts, related in gold. To ensure precision, an analog machine has to be made with really exacting precision. That means precise components, but it also means you have to prevent noise. That's something we don't have to worry about on a digital computer, or at least we don't have to worry about in the same way.

A wire in the digital realm can be pretty noisy and still work since we rely on fixed voltage levels. Anything around 5 volts will look like a 1. But in analog land, your data is just a voltage. Noise can change voltage on a line, which changes your data. In exchange for direct simulation, you have to deal with, well, more direct interference. A further complicating factor is that the 231R, as with many EAI machines, was modular. This is the time to freak out, right? We have a modular computer in the 1950s.

It's just that it happens to be analog instead of digital. You could buy a base system that would include the patch panel, control desk, some dials, power supplies, and a good amount of components. Customers could then increase capacity by adding more cabinets of components. The modularity doesn't stop the cabinet level. Actual components are kept in these circuit packs. These are little replaceable packages that contain a few simple components. The 231R and its expansion cabinets are built up from these component sleds.

You have modules for amplification, combination, summation, and integration. This is actually pretty advanced stuff. Research machines like the TX0 used modular components as early as 19, The PDP1 was seen as pretty wild in 1958 when it came to market built from modular bricks. Over in analog land, we had a similar design approach going on since at least the early 50s. But this all comes at a cost. Programs are entered via a central patch panel, but components are spread all around the machine. That's why noise is such a concern.

This thing is built kind of like a telephone exchange. You have trunks and wires and buses connecting all over the machine and up into that plugboard for the operator. Each of those wires acts like an antenna, ready to receive signal from nearby current. It's easy for noise to build up, and these analog systems are super fragile when it comes to noise. Hence, why shielding is so important.

A stray signal could and would, ruin your data. We're at a point that the design here is pretty refined, though. Little details like gold plating, coaxial cables, and metal cans speak to decades of iteration and refinement. So it's wrong to think of analog computers as just less sophisticated machines. This is a separate lineage with its own history. There's another way that the 2 through 1R assures accuracy. And this one, oh boy, I did not expect it all. Let me just read you a passage from the Maintenance Manual. Quote,

The precision resistors are contained in shielded network assemblies in a constant temperature oven, directly behind the patch bay. The oven temperature is thermostatically controlled. This oven also contains the polystyrene dielectric integrating capacitors. Both the resistors and capacitors are stabilized, and adjusted to an accuracy of plus or minus 0.01% to quote further. As soon as the computer is connected to the power source, the oven heaters and blowers start operation. The oven should never be allowed to cool off.

Thermostats are set at the time of manufacture for an oven temperature of 100 degrees Fahrenheit, end quote. The oven should never be allowed to cool. You can find photos of this thing in the maintenance manual. It's actually a big metal box with a fan. It's a literal convection oven with heating elements and everything. It's used to house these more sensitive components. The modules actually plug in to the oven.

So what's going on here? Why is there an oven in a computer? Well, logically speaking, an analog computer is the opposite of a digital computer. If a digital computer needs cooling, well, an analog machine must need to be heated. But, okay, what's actually going on? It comes down to accuracy. Specific electrical components, like the resistors and capacitors used here, had guaranteed accuracies and behavior at specific temperature ranges.

So the 231R makes sure those components are kept at that optimal temperature. The logical solution is to have a conveyor. oven inside the computer. Analog computers are related, but have all these fundamental differences. So then, what in the world does it mean to be a hybrid computer? And how do we make that jump? Hybrid computers don't really emerge until the 1960s. By that point, we're well into the digital period.

So why then are analog machines still? kicking around? Well, we actually get some first-hand answers. Many texts on analog computers have a section that justifies their existence. And that, I like. I actually saw something similar when I went back to very early books on digital computers. There was always at least a page dedicated to explaining why computers are in fact important and why you should please learn about them. Maybe that would be nice, wouldn't it? With analog machines, we can see a slightly different tone. Corn and Corn has an entire section of its first chapter dedicated to, quote,

an appraisal of DC analog computers, which compares these machines to digital computers. Notably, I've been working out of the second edition of this text, which is published in 1956. By that time, digital computers could beat an analog computer on accuracy. To make an analog machine more accurate, you have to use better and better components and go to increasingly more extreme measures. That isn't cost-effective. So digital will inevitably win out as the machines get better and better. But analog computers are still winning in two regards. The first is cost.

Analog machines were just cheaper. If you're willing to accept a 0.1% or even a 1% of, an error in your calculations, than an analog machine may well be a good deal. Then we get to speed. In some cases, and for some specific applications, an analog computer is plain faster. The reason for this is, well, a simple fact of physics. Electrons move at the speed of light. Well, for the most part.

As soon as a voltage is applied to a circuit, the machine starts humming. Crucially, it doesn't tick. There's no steps like on a digital machine. Voltage propagates through the circuit and you get an output. In some cases, this is nearly instant. Division, for instance, would happen immediately. On digital computers, especially early machines, division was one of the trickier math operations to pull off.

But an analog machine just needs a resistor to do the job. You can also parallelize operations on an analog machine. These computers have isolated components for different operations. The 231R, for example, has eight integrators. That means you can run eight integrals in parallel at the same time. Digital computers, because they operate in discrete steps, can't paralyze anything. True parallel processing only becomes possible as digital computers become more advanced, and are able to cram multiple processors into one machine.

For certain tasks, an analog computer, if well-wired, is just plain faster than a digital one. And it's not even a case of a marginal difference. We're talking orders of magnitude. Keep in mind that speed is traded for accuracy. Oddly enough, that ends up being the same proposition that digital computers face. Due to digital approximation, it's common to run into issues where you can get mostly accurate answers quickly or you can get more accurate answers slowly. So there's this complex break-even kind of analysis that you have to do here when choosing digital versus analog.

Now, one of the corns specifically argues that we shouldn't focus too much on the whole speed versus accuracy versus cost thing. Check out this extended quote. In the writer's opinion, DC analog computers should not be regarded as low-cost substitutes for digital computers. Analog computers can only rarely compete with digital machines in the mass production

of numerical data required as end products of scientific or engineering analyses. In such applications, The high computing speeds of analog equipment are often outweighed by the more advanced automatic programming of digital computers, even if only slide rule accuracy is required. The really important contributions of DC analog computers to modern research and development techniques go beyond mere numerical computation. In many applications, the analog approach functions as a direct aid to a research

workers or engineers thinking process. An analog computer setup serves as a model which helps to bridge the gap between mathematical symbolism and physical reality. There's a lot there. Corn and corn start by explaining that analog computers should be viewed as different tools than digital computers. If that's true, there's no reason to compare the two. In fact, to do so only courts misunderstanding.

One doesn't compare a screwdriver's sharpness to a saw, even if you need both to build a chair. Instead, analog machines should be viewed as a helpful aid for knowledge workers, almost like an augmentative technology. For 1956, this is surprisingly progressive thinking. There's a feel here that's very, very similar to Vannever Bush and Doug Engelbart. Why do I say this is so progressive? This is very explicitly saying analog computers should be used to augment what someone can do. You may not be able to hold an equation in your head very well, but if you throw it onto a plugboard and start playing around with values,

you can understand the equation better. and there's a big implication here. That is that a person is using one of these analog machines on their own, that they're playing around with the machine in real time, that they have access to something like a personal analog computer. For 1956, that is very progressive, and it's treated as an implication here. Corn and Corn never say, oh, and these machines are so cheap and wonderful, they're so small, we can have one on every desk. It will be a revolution.

Instead, they just say that an engineer can mess around with one to better understand physics. That requires prolonged physical access to a computer. It requires someone to have personal access, an interactive use of a computer. For the time, that's wild. Will this be possible in the 50s with a digital computer? No. Of course not. The closest we get is, of course, my beloved LGP30. All kidding aside, though, this is shocking to see in print this early.

And it's shocking to see it in a book about an analog computer system. So mark this as a good reason to stick with analog machines. the fundamental differences in how an analog computer functions leads to fundamental differences in how it can be used. And this isn't just a one-off idea that shows up in corn and corn. I've seen the implication in other places. That EAI video that I opened the episode talking about mentions this exact use case. It says that analog machines are better at allowing an engineer to get a,

feel for a problem. Analog machines simulate directly without the need for approximation or even a program or mathematician to help out. It's a lot more touchy-feely technology. Again, to put it another way, it's personal. A lot of the language around these analog machines are trying to say that they're personal computers. They just don't have that word yet. However, there are some things that analog computers just cannot do. Logic is one of those. A purely analog machine has no way to check values. You can't

tell it to do one thing if the result of an operation is positive, do another if it returns a negative. Things start to change in the latter half of the 50s. James Smalls, writing in ICCI ANLs, points out this change came at the same time as a revitalization of an older technology. And, at least in my head, the two sort of connect. Analog machines can broadly be broken into two categories, single shot and repeating. Single shot machines are what we've been discussing so far. You make a circuit, put in a voltage, get a number out. One shot. Repeating operation machines or rep op are only slightly more complex. In these machines, operations are continually repeated. This is usually paired with an oscilloscope to give you live feedback. This is much better

for feeling out a problem, since you can adjust dials and values and watch the results change in real time. Again, this is an interactive personal computer, but I got to move on from that point. I bring up this shift because, to me, this represents the full flowering of analog machines. These newer repeating operation machines are the ultimate in hands-on computing. They take advantage of the newer technology of the 50s to create these very interactive devices. An engineer can sit down at the desk, twiddle dials, and watch as a graph changes on a phosphor screen. The other interesting point is, with the flourishing, of these repeated operation machines, you're now adding steps to an analog computer.

They may just be cyclic, but you are adding this concept of a ticking clock. Now on the other side of the shift are hybrid machines. If repeating operations are a late flourishing of the analog arts, then hybrid computers are closer to the hollowing out of the technology. Okay, that's a little dramatic. Perhaps it's fair to say that hybrid computers are an evolutionary step towards something different. According to hybrid computation by Becky and Carplus, the first hybrid computers are cobbled together at Conver Aerospace and Ramo-Wildridge Corporation between 1955 and 56.

Note how late that date is. By 55, we have commercially produced digital computers. We aren't at the cusp of new technology anymore. We're into the adoption phase. It may be somewhat early adoption, but we are into adoption nonetheless. I mean, the first computer user group is founded in 1955. Hybrid computers are, very broadly speaking,

systems that combine analog and digital components. I say systems because, hybrid computers are pretty expansive, both as a term and as a device. The first hybrids are actually two computers with some glue in between them. Convair's system used an IBM 704 and our beloved EAI Pace 231R. The two machines were joined with an interface. That interface was a bit of a monster in its own right. Convair used what was called an ad aversher. This device, was essentially a sophisticated analog to digital and digital to analog converter. It could take in multiple digital values at once and spit out multiple analog voltages.

It'd also do the same dance in reverse. The adiverter on its own took up nine cabinets of tubes and resistors, but it was needed to make a digital machine speak analog. Once all set up, Convair could run. write programs for their 704 that could call out to their pace for some calculations. Why was this needed? Initially, it was part of the Atlas ICBM project. The plan was to connect real-life circuits that were being designed for Atlas's guidance system

to the computer. That would allow for very high fidelity simulation and testing. In theory, this hybrid computer could take the guidance system through an entire simulated launch and flight without ever leaving a lab. We don't have very good information on these Atlas projects for kind of obvious reasons. It's not 100% clear why an analog computer was needed in the loop, but we can make some safe assumptions. The Atlas guidance systems weren't digital.

They were analog. We don't have mini-trize computers until the late 1960s, so guidance circuits were all analog in this period. I'd also wager this had something to do with performance. Conver wanted to run these simulations in real time. That's a performance goal. To simulate, say, a real-time flight path, you need a bit of oomph. You may even need to do a multiplication or two

and would likely need to solve something like an integral. An analog computer can do that in real-time handily. but in 1955, a digital machine would probably be a little bit slow on the draw. Despite building up the system, Convair never ended up using their hybrid computer for Atlas. According to Becky and Carplus, the project simply changed. The need for real-time simulation with physical test articles proved less important than initially believed. But we can still see some of the reasons. for hybrid computation here.

Convair was trying to combine two systems to get the best of both worlds. Performance is just one side of the equation. The IBM 704 could carry out multiplication in 240 microseconds. An analog machine can do the same instantly. I don't have numbers offhand because, well, how long does it take current to travel through a resistor? integrals are another example here. An analog computer carries out an integral at breakneck speed. It does so without approximation.

A digital machine has to take an approximation and then take a series of steps to compute an integral. Actually, I keep saying that, so let me slow down here for a second. Do you know what an integral actually is? An integral can be thought of as the area under a curve. Think of an arch, and then imagine coloring in everything below that arch. The integral of the arch will be the area you color in. Crucially, this isn't a discrete value.

A curve's height above some line is always changing. It's continuous. In other words, it's analog. To calculate an integral on a digital computer, you slice that curve up into sections. and then you calculate the area under each of those sections and add everything up to get a total value. This will always be an approximation because you're splitting a continuous curve into a finite number of chunks. Imagine turning that arch that we talked about earlier into a bar graph. It gets a little pixelated.

Technically, the only way to get a 100% accurate value would be to take an infinite number of values. chunks to use an infinite number of pixels. But for lame technical reasons, digital computers can't do infinite things. So you have to live with an approximation. You can choose how accurate that approximation is by taking more chunks, by slicing the arch into more segments. But there is a limit. Each chunk will take a certain amount of time to calculate. So the more accurate your calculation is, the more calculations it takes, the longer it takes to run. You have to trade accuracy for performance. Put a fancier way, the time required for calculation on a digital computer scales by the

complexity and accuracy of that calculation. An analog computer uses a capacitor to do this. You have a little box with a capacitor and some plugs. In the case of the 231R, that box slots into an oven so it stays within its temperature rating. The calculation is almost instant, and there is no approximation involved. I'll tell you that as a programmer, I know which solution I like. I'd prefer the little box that says integrator on it that just gives me an integral. That's so much easier than writing software and thinking about trade-offs.

Another huge draw is the fact that you can hook up other analog circuits into an analog machine. I know that can sound weird, but check this out. Conver wanted to wire up an Atlas guidance system into their computer and run simulations using that real hardware. I don't have a lot of detail on that specific setup, but I do have information from a similar project. In this same period, GE was experimenting with a hybrid system called High Call. Their experience is summarized in a 1962 paper from the Joint Computer Conference. That paper includes about a dozen different case studies that GE had been working on throughout the years. One is testing of missile guidance systems.

GE constructed a full loop. On one side, they had a digital computer that simulated the flight path of a missile. This digital side handled things like drag and thrust. It could be fed commands to adjust the missile's simulated trajectory. The analog side converted and scaled data from the digital simulation, finally outputting voltages to some wires. Those wires were then plugged in to a physical guidance circuit from a missile, into the same plugs that would be used for sensors.

The guidance circuit did its thing and then used another set of wires to output changes to thrust and control surfaces. But instead of those signals going to something like synchros or motors, they were directed back into the analog computer. From there, values were adjusted, some math was done, and then the numbers were digitized and fed back into the digital simulation program. That right there is a powerful setup. GE was able to test a real-world component.

They could use the actual circuits that would be put in a missile and trick that circuit into thinking it was in flight. That's basically what I'd call an integration test, just using a hybrid computer instead of fancy software. Hopefully you're getting the picture. Analog computers have a lot of pitfalls. Digital computers also have a lot of pitfalls. By combining both, you get the best of both.

worlds. The downside is the added complexity. These early hybrid machines are beasts in every sense of the word. We're still, they're all custom. Convair has its own rigged-up system, so do GE and Ramel Woldridge and a pile of other labs. Becky and Carplus spill a lot of ink in definitions. I don't say that as a jab at all. I do the same with bits and bytes. Their text identifies a spectrum of computers. What we've discussed so far, digital and analog machines connected by some linkage are termed as true hybrid computers. That falls in the middle of a spectrum spanning purely digital on one side and purely analog on the other. Once you stray anywhere away from the middle of that spectrum, things start to blur. What

constitutes a balanced blend of analog and digital. Well, that's a bit hard to say. In the 1960s, we start to see production hybrid machines. But where exactly do they fall on that spectrum? This is where we get back to EAI. In 1965, EAI announces a new machine, the 680. From an article in the Ashbury Park Press, quote, Although hybrid computers have long been considered a necessity by aerospace engineers in the design of aircraft, missile, and space vehicles, their complexity and cost, until now, have put them out of reach of engineers and researchers in non-airospace fields.

Lloyd Christensen, EAI president said, with the introduction of the EAI 680, the benefit of hybrid computation, are economically available to the biomedical university and industrial user, end quote. Economic here is still pretty expensive. A base configuration 680 cost $30,000 in 1960s money. That cost went up as you add,

components. But I think it's interesting that the 680, like many digital machines, was meant to be expandable. One booklet from EAI even called it the, quote, machine that grows with you. Machines like the 680 are exactly why so much ink was spilled debating the definition of a hybrid computer. The 680 itself exists in that large blur around the middle of the analog digital spectrum. It's an analog computer with some digital components built in, and interface hardware to connect to an external digital computer. At its heart, the 680 is your classic electronic analog machine. You get a patch panel, analog components, the works.

That's still the core offer. What makes it distinct are all these digital features. That actually made the 680 considerably more complex. As I said earlier, purely analog machines are built kind of like a telephone exchange. You have trunks that connect everything up. Those trunks lead from patch boards to actual components. All that wiring is specifically meant for analog signals. signals. When you add digital to the mix, you have a second set of connectors and trunks,

and you have to have a way to cross over between the digital and analog components in the computer. For that complexity, you actually get a nice complement of new features. The 680 comes tricked out with andgates, timers, counters, and even six digital registers. That's enough to string together a simple computer. There are also some mixed-use devices. This is how things switch over between the analog and digital sides. You get components like comparators, which can take an analog value as input and give a digital output.

To go the other way, you get digital-to-analog switches and relays. These components will pass through an analog voltage if a digital input is set to a specific value. So we have an analog machine with a little bit of digital smarts. on its own, that sounds vaguely useful, right? Well, not so fast. This actually opens up a world of possibilities. This is for two simple reasons, if and steps.

You can use digital components in the 680 to construct conditionals. Using a combination of comparators and digital-to-analog switches, you can build a circuit that carries out a certain calculation if a criteria is met. That's useful for problems that involve partial differential equations. One of the most famous problems it requires a partial differential is Schrodinger's equation. So the addition of a few digital components means an analog machine can simulate quantum physics. That on its own is pretty neat.

You can use counters and registers to sequence execution. do this, next do this, then do this. That means you could, say, wire up a Monte Carlo simulation. Take a step, roll some dice, use that outcome to take the next step. This application is especially neat because the Monte Carlo method was the first digital-only algorithm to exist. It turns out you only need a little bit of logic to make that possible. This makes the 680, on its own, an incredibly powerful machine.

You can then hook it up to your existing digital computer to get an even more powerful setup. You don't even have to have an ad averter. But that's just a more convenient example of what folk like Convair were already doing. Where is the truly wild stuff? Well, believe it or not, EAI would offer their own true hybrid computers. Near the middle of the 60s, EAI starts development of their own series of digital machines. The relevant one here today is the 640.

Now, I've been trying to figure out if EAI's machines are totally homegrown designs or if they're a rebadge of some other computer. I think they were new machines. I say this because BitSaver actually has an internal EAI document that discusses the development and design decisions behind the 8400, EAI's first digital computer. But I don't know. When I see photos of the 640, their second digital machine,

it just kind of looks like one of IBM's small computers. Then again, it was the period for that kind of look, I guess. So I might just be reading into the situation too much. The 640 is a 16-bit parallel digital computer. In other words, it's pretty standard for the period. It has registers, direct memory access, interrupts a whole nine yards. One thing makes it special. To quote from the hardware manual,

the EAI 640 digital computer system provides the user with an outstanding general purpose computing capability for handling a wide range of scientific applications. It includes flexible input-output and interrupt features that make it particularly useful in combined hybrid and special systems for simulation, hybrid computation, online monitor-slash-control, and other uses, end quote. The 640 comes with a little bit of extra hardware for interfacing with analog devices, like an analog computer. So, cool.

EAI has this lineup of analog computer. that are easily interfaced with digital machines, and digital computers that have features that make them easy to interface with analog computers. That's kind of the selling point, right? Both these machines are hybrid-ready. You may be able to guess that there's more going on here. Perhaps the model numbers give it away. In 1967, EAI announces the 690.

It's a 640 and a 680 bundled together. In other words, you can buy a truly hybrid computer direct from EAI. This is a pretty cool maturation of the hybrid concept. We've gone from one-off research machines to products that are mass produced. The downside here, I think, is the digital piece. To facilitate selling a 100% EAI hybrid machine, EAI had to make and sell a digital computer. In this period, that means they also have to make and distribute software for that computer,

since, you know, there isn't exactly a library of software for EAI that's ready and waiting. With the introduction of the digital part of this computer, we actually land back in a classic issue. That is, the software problem. For folk to want to use your hardware, you should really have some software. This isn't so much an issue with analog machines normally because, well, they don't really have software. But that digital piece really leads to some tricky issues, doesn't it? EAI solution to this problem is actually elegant in its simplicity. The 640 had an in-house Fortran 4 compiler.

Thanks to how the 640-690 was positioned, that actually solves almost all the software issue. Remember how we said that analog machines shouldn't be compared directly to digital machines because they fill different roles? That's also the case for hybrid machines. Hybrid computers are very specialized tools. They can do things that digital machines at the time simply could not. One aspect of that is instant feedback. My Touchstone text, the wonderful Becky and Carplus, really hammers this home when discussing software.

When you get down to it, the 690 is basically a hybrid mini-computer. It's intended to be used by one operator, both on the analog and digital side. And it's intended to be used interactively. To be even more specific, it's intended to be used. used by researchers and engineers. That's the exact same audience that Fortran was created and cultivated for. And crucially, those kinds of nerds end up writing their own software as long as they have the right tools.

So EAI just had to have a Fortran compiler and a handful of tools to help the whole analog side of the house. That connection part, the whole hybrid thing, is pretty obviously what makes EAI's offering unique. Technically speaking, EAI ships a standard version of Fortran 4 that's accompanied by a custom hybrid

linkage library. The smart thing about this approach is a new user can run existing Fortran code on a 690. In theory, that means a researcher could migrate to using hybrid features. You come from an IBM

shop. You throw your research simulation onto the new EAI, machine, and then you can slowly improve things by replacing your clunky integration code with a call out to this hybrid library. In practice, the library gives you calls to communicate with the analog side of the machine. This means you aren't so much just saying, divide this number for me. Instead, you call out to set up a division on the analog side, you do some synchronization, and then you fetch the results.

The hybrid 4-Tran code I've seen is kind of clunky but workable. I don't suspect it would be all that gross for a period 4-Tran head, especially when it comes to the more complex operations like integration. A neat addition here is EAI's debugging code. You could compile hybrid-Fortran programs so it would call the physical analog machine. That's the normal workflow. But what if you just want to test things out? EAI provided a simulation library for that.

Basically, it would simulate responses from the analog link. That would let you debug your program in place without firing up anything analog. That's the more conventional part of the offering, if, well, we can call anything hybrid conventional. What's even more off the beaten path is a little known thing called the Hytran operating interpreter. How's that for a name? Hy-Tran is a bit of a beast. Hybrid Fortran is one thing. That's basically just an extension on top of a well-known language.

Hy-Tran is a digital programming language designed for dealing with analog computers. And I don't think I'm quite ready to talk about that, actually. All right, that brings us to, well, I guess a stopping point for now. We've covered a lot of ground when it comes to analog and hybrid computing hardware. But there's a lot more here that I didn't expect. I was actually planning to wrap this up, but the programming aspect runs a lot deeper than I could have ever guessed. Electronic analog computers aren't just some older technology

that was completely outmoded by digital machines. Rather, electronic analog machines are their own lineage. There's a pretty long period where analog machines offer distinct advantages over their digital cousins. Those advantages led to the creation of hybrid computers. And once digital came into the analog picture, well, programming came along for the ride. In the next episode, I'm going to hit the paint hard on hybrid programming languages. Specifically, I want to learn more about Hytrane and this competing language called Apache. There's simply too much here for me to rush things.

So until then, thanks for listening to Advent of Computing. You can find links to Patreon, to merch to everything at Advent of Computing.com. And as always, have a great rest of your day.