This is amazing. I use the MakeNoise Maths module on my synth to experiment with similar ideas, as I think the next leaps in scaling quantum computing are going to come from a generation of kids who developed a musical level intuition for wave dynamics playing with their parents' analog synths.
Digital has been cool and interesting, but when young hackers have the tools and incentives to become de-facto physicists, they're going to tear holes in the fabric of reality. This is such a cool product.
One thing to note: the waves in quantum physics have complex amplitudes[0] (e.g. a real number + an imaginary number), rather than the waves in most synthesizers and analog computers which only have a real amplitude. This means that they can have very different behavior e.g. wave interference is different for real-valued classical waves and complex-valued quantum waves. Experience with any wave dynamics is certainly still valuable though.
OK, this is increasingly off-topic, but I’m just getting into Eurorack and I’ve been become curious about analog computers… not because I know anything about them, but because they share a user-interface style.
Can you speak to similarities between them? You seem to know a bit about both topics.
Thanks! With the caveat that I profoundly lack depth on both, and am a bit artsy craftsy in my intuitions, I did buy the Maths module precisely because it was an analog computer.
The intent was to create some generative music, and seeing if I could use it to create 1/f noise, and exploring self-similarity using a musical ear instead of in code. Still working on it.
When you look at what you can do with magnets hooked to a synth with a ferrofluid, e.g. https://youtu.be/Q3oItpVa9fs , the sound and your ear helps to develop an sense for what kinds of waves can be used to yield physical effects. If you can functionally suspend and maipulate a ferrofluid with waves, there are likely analogies to laser pulses and electrons for quantum computing. Waves gonna wave, etc.
The other piece that grabbed me recently was "analog fractals," (https://hackaday.com/tag/analog-fractals/ + that rabbit hole) where if you can get those artifacts instantaneously as the effect of physical feedback instead of rendering them computationally, there are likely faster functional approaches to a lot of other problems, only a subset of which we use ASICs and FPGAs to implement today.
It's speculative based on laughably incomplete understandings on my part, but that's hacking.
The last time this came up, I spent some time investigating the use of analog computers for Monte Carlo simulations and it is absolutely fascinating how differential equations or stochastic differential equations are being implemented with analog components. There are already some chips available that integrate these components similar to FPGAs and some toy examples show promising results, but it looks like we are not completely there yet.
Nevertheless, if these chips get further developed, they could have a huge impact on neural networks (at least in terms of power consumption) and maybe Monte Carlo.
That would be really interesting - how do you generate the randomness?
Every so often when Matlab or Simulink won't integrate something simple, probably because of something I did, I wish I had an analog computer to compare to.
Randomness is impossible to eradicate from the analogue world, so you just take an existing noise source like a semiconductor junction and connect it to an amplifier.
Some care is needed to get something that works across a wide temperature range, and avoids being easy to overwhelm with outside noise sources, but it's a problem with lots of existing solutions.
I mean in a practical way for a monte carlo simulation - usually in analog setup you set parameters with discretes, so you are going to need to do things like feeding a transconductance amplifier with your noise source. You're also going to worry about shaping and otherwise parameterizing your distribution - do you need uniformly distributed noise? gaussian noise? what are your limits?
I can imagine a lot of practical pitfalls and awkward half-solutions to trying to do an analog monte carlo; I was wondering how the OP went about it.
There is actually a fascinating thesis by Yipeng Huang ("Hybrid Analog-Digital Co-Processing for Scientific Computation") that discusses toy models (i.e. Black-Scholes and variants).
As usual, you have a Wiener process and thus need Gaussian noise. Yipeng Huang found that some noise stemming from a resistor ladder of the chip provided Gaussian white noise and he could control the mean by feeding it with a DAC and he also had some way to control the variance by changing some multipliers (but I can't tell you exactly how that worked). Nevertheless, this was the analog part and he faced issues with DC drift.
Alternatively, he looked into generating the noise digitally with a microcontroller.
It's cute, but, having used analog computers, it lacks some important things. Analog computers are usually reset to initial conditions, then started. As time goes on, the outputs change. This thing has integrators, so it will generate time-varying outputs. You need some way to watch those time-varying outputs, such as a plotter or an oscilloscope. This needs a graphical display. Something like one of those US$20 digital oscilloscopes available on Alibaba.
While this is a cool idea, it's not terribly useful for anything more complex than a toy example. The fact that I need to physically wire the components and manually adjust coefficients every time I create a new program means that I can only really leave one complex program configured at a time. How do I switch between programs? Do I buy new THATs then?
It would be much more practical if the analog components could be configured digitally through a script, so that multiple programs could be constructed and run on the same compute platform. A hybrid platform of sorts.
> The fact that I need to physically wire the components and manually adjust coefficients every time I create a new program means that I can only really leave one complex program configured at a time.
This is basically the reality of every musician making music with guitar pedals and modular synthesizers. And the answer is, yes, you physically make the thing you want. Once you're done with it, you tear it down to make the next thing. That's part of the magic.
It's like building something out of LEGOs. If you want to re-use the pieces, you need to take apart your creation.
Why can't it work like an FPGA? I set the coefficients in software, it programs the analog computer through a serial interface, and then I don't need to tweak knobs every time I want to change something.
Perhaps a more interesting question is to ponder why an artist might not want that flexibility. How does it affect their creative process when in order to make a new sound the old one must be destroyed? Are there positives to that workflow?
It can do. Digital switching of analogue signal paths can absolutely be done. MAX335/336/337-type chips are one way of doing it, and they've been around for decades.
From a mainstream perspective, there are mixed mode ICs that contain a bunch of analog components (op amps, comparators, etc.) and allow you to configure them digitally. E.g. the whole PSoC series from Cypress (formerly, now Cypress is part of Siemens).
Slightly less mainstream, there is the concept of FPAA (field programmable analog array), which is the equivalent of an FPGA but for analog instead of digital.
i’m not sure how to find the photos but on some old analog computers, the entire patch board is swappable. to run a program, you would choose from your physical library of patch-boards.
kind of like moving an AVR chip from one breadboard to another
FPGAs are certainly far more useful due to the fact that you can design and simulate them digitally and then mass produce them from that design. You can work on as many FPGA designs as you want in parallel without having to physically set up and tear down each one in turn.
This looks really cool, I found the line "inherently safer than digital computing in the face of cyber threats" pretty funny. There's no analog internet so what could the cyber threats even be?
The quote is saying that digital computing faces cyber threats, which makes analog computing inherently safer. Not that analog computing faces cyber threats.
I feel like that's a silly claim though. If you have no networking, wireless, interface for loading code, or really any code at all, then of course it's more secure.
You're not doing the same tasks as a digital computer though.
Exactly. And the claim stands, that if you can effectively compute («model dynamic systems») through an analog computer, you expand your toolbox with items that are beyond a framework involving security issues. (Yes, like paper, but paper, while it helps the computer, does not compute.)
Another rephrasal of the concept they are proposing: "Don't be the man with a hammer" [not all problems are to be treated as nails - including the health and safety related concerns nails demand].
I've joked for a number of years that I expect to see boutique professional offices (lawyers, doctors, dentists, etc) springing up that offer 'paper records only' as a feature.
I love devices like this, buying a Behringer Neutron completely opened my eyes to the possibilities of patching on a bigger level than just music.
I do have to wonder who would buy this though. Analog computers like this are virtually useless unless you have some particularly applied use-case for them. I'd be interested to hear some user testimony if anyone owns one of these!
> Analog computers like this are virtually useless unless you have some particularly applied use-case for them.
Did you miss the educational part?
Electronics is physical math. Amplifiers are a perfect example, take an input and multiply it. Resistors can be used to subtract or divide. Combining these functions is how an op-amp works. Combining more components now allows you to do integrals, derivation, and other functions.
From there you take these basic building blocks and apply them to real world problems like audio amplification, processing and filtering. Of course a lot of this is replaced by digital stuff but digital has one huge disadvantage: obsolescence. I can repair a 40 year old analog amplifier. I cant repair a 10 year old stereo with a dead ASIC or DSP/SoC.
> Analog computers like this are virtually useless unless you have some particularly applied use-case for them. I'd be interested to hear some user testimony if anyone owns one of these!
You've already highlighted a clear use-case. The Neutron is a semi-modular analogue synth. The whole modular-synth scene is based on this kind of analogue computing, when I look over at my eurorack modules, I have:
* Function generators
* Gates
* AND/ OR / NOR / XOR / NAND / XNOR
* Summing
* Multipliers
* Comparitors (generate a signal based on conditions)
* And more!
That pretty much covers everything in 'The Analog Thing' computer.
Gates are built using transistors that are designed and biased so that they "threshold" themselves. You can just-about rig a gate so that it's output will be uncertain; but in general, gates are not amplifier circuits with some kind of threshold on the output. They are intrinsicaly electronic switches.
Yes. Logic gates are essentially the bridge between the analog and digital world. Your digital computer is ultimately based on analog components that use thresholds and high-low voltages.
The site emphasizes the educational aspect. I think they have a point. Nothing teaches mathematical relationships like hands-on instant feedback from a physical system. Patching your program and turning knobs in a literal sense jives well with the naive mental physics model some of us rely heavily on for reasoning and learning.
I think they missed the mark only having a single lcd panel for output. I think a grid of leds, maybe 10x5, with each column having an input jack and a dial to set the range would allow much better visualization. That is a lot of extra parts though.
The LCD voltmeter is used for precisely setting coefficients, to view the output you need some kind of scope. This is typical of most analog computers.
Because it is an analog computer it is a kind of breadboard though, isn't it? ;-)
No, having all the Op-Amps wired up for me, powered for me, "nulled" for me is a huge improvement over the mess I would have to make on a breadboard to replicate even a part of it.
I only wish it had more than one output (meter) component so I could observe steps along the way.
This looks awesome, but as someone who knows nothing about analog computing, what’s a good educational resource to go along with this so that I can use it to actually learn?
Analog computing is about making analogs (today we would say usually say simulations) usually of physical systems, usually with operational amplifiers.
Learn to model physical systems with differential equations first (and in 2021 you will probably also learn to solve/integrate them numerically on a computer in the process) then go to op amps.
> As digital computing approaches the limit of Moore's Law, analog computing offers a strategy to diversity today's digital monoculture.
What sort of industrial roles/domains is analog computing particularly fit for. And whats the cancer that we see them adopted either replacing or alongside digital computers in those roles.
This is so fascinating. I'll be keeping an eye on this. I've been doing binary computing all my life, so it's kind of mind bending to think of computing using analog circuitry without immediately trying to turn it back into binary again.
> With "The Analog Thing", you can model dynamic systems including: ° market economies; ° the spread and control of diseases; ° population dynamics; ° chemical reactions; ° mechanical systems; ° a variety of mathematical attractors
This is way better than my Radio Shack 200-in-1 for understanding the logic part better than jumping directly to resistors/capacitors/diodes/etc. What a time to be alive.
"Why Analog Computation?", oldie but goodie unclassified NSA document.
> An introduction to analog computation containing a brief description of the analog computer and problems in which it can be advantageously applied. Both analog computers and systems combining analog and digital techniques are discussed in order to show why the Agency's interest in this computation area has increased.
I looked a while back, and it seems that most books on the topic fall into one or the other of two categories:
1. Really old, and really expensive
or
2. Somewhat modern, and also pretty expensive
FWIW, here's an Amazon List on this topic that I put together a while back. Looks like there are one or two that can be had for fairly cheap, if used copies are acceptable.
There's a fairly simple reason for the superiority of digital computing, which is that error accumulation in analog is impossible to avoid, so the accuracy of results is limited by the quality of the components, which makes them large/expensive and even then maybe only 1% accurate. Programming them is less flexible, too. The applications are pretty niche.
The analog advantage is that you can directly implement differential equations, so they can be more power-efficient and faster than the equivalent digital simulation.
Thanks for this. I am most interested in mathematical comparisons. That is to say, ways in which optimal analog computation (idealizing away from noise propagation, among other things) might be sometimes better than its digital counterpart. Your last point about diff equations is directly relevant to this.
Digital has been cool and interesting, but when young hackers have the tools and incentives to become de-facto physicists, they're going to tear holes in the fabric of reality. This is such a cool product.