General purpose analog computer | Hacker News Reader

General purpose analog computer

en.wikipedia.org

41 pointsch19 comments

Introduced by Claude Shannon, the father of modern information theory, as we know.

Ahhh Shannon. The guy who invented the entire field of digital freaking electronics for his master's thesis [1], and then moved on to even greater things. This was it seems five years after his Master's thesis. He was 25 when he proposed this.

The bane of every grad student. "Shannon invented digital electronics for master's thesis. Come up with anything interesting yet?" Add to this, he also appears to have been located at Medford. The shame. What have y'all been upto lately, Boston/Cambridge area grad students? : (

I wonder if our obsession with profits and quick returns and 'hacker culture' is distracting people like him from making similarly great contributions?

[1] http://www.cs.virginia.edu/~robins/Shannon_MS_Thesis.pdf

As Peter Thiel said, "We wanted flying cars, instead we got 140 characters."

I always find it a bit surprising that Shannon is remembered far more for Communication Theory than his Master's thesis (which I think was very significant).

It seems to me that everything about universities and funding these days is very conservative, making it quite difficult to do the kind of research that can lead to great contributions.

I have been wondering recently if analog computing would make more sense for certain search algorithms, like protein folding simulations. Basically, let a set of potentials be placed in the op amp setting representing position, and let them miller integrate the surronding charges and watch it evolve. Basically, physical modelling directly. Im not sure of the state of analog device precision, drift, and temp stability these days but it might be worth a try.

Take a look at these, esp math coprocessor:

https://www.schneier.com/blog/archives/2015/07/friday_squid_...

The trick is using analog circuits for just what it's good for. The digital does the rest. Need to have them integrated well. Good news is analog works well with inexpensive process nodes like 350nm and digital stuff on them is inexpensive. Tekmos even can integrate 90nm digital with 350nm analog plus MPW's to keep things cheaper.

If you can keep the chains short enough and the noise down then analog can solve some things that you'd have a hard time solving digitally.

Even mechanical computation has some unique problem solving capabilities.

It is my dream one day - to sell my startup, and get some form of mechanical computer, and have it in my house as an art display.

I would like to integrate it somehow to my existing systems - so that, no matter how little it can contribute, it can click and clack away.

I shall name him Switch

Wouldn't Clutch be a better name? I mean if you litterally mean mechanical.

Switches were mechanical long before they were electrical. And I don't mean 'electro-mechanical', no just mechanical. Think of where the name 'railroad switch' comes from. That lever by the side of the road is the switch (the thing itself is called a 'point').

Mechanical vs electronic is pretty much orthogonal to analog vs digital. I would consider slide rule for example analog mechanical, while abacus digital mechanical.

Analog computer I saw in Berlin a few weeks ago, very cool stuff - https://twitter.com/maljx/status/650690241574080512

If this interests you, you may enjoy modular synthesizers, eg: http://www.makenoisemusic.com/maths.shtml

I recommend the docu-movie "I dream of wires". I think it's on Netflix.

I worked with a medium-sized analog computer in the mid 1970s. It may have had around 20 op-amps, covered a desktop, and used an HP XY-plotter for output. I could set up different models of differential equations. I remember simple harmonic oscillators and predator-prey models [0]. I was studying physics rather than computer science and didn't then know about Turing computability. I found the GE BASIC timesharing system easier to program and use than the analog computer, but the graphical output of the analog system was much nicer.

[0] https://en.wikipedia.org/wiki/Lotka%E2%80%93Volterra_equatio...

There's been lots of practical work on this including a math coprocessor with 100x claimed acceleration. I posted a list of links below for general purpose analog:

https://www.schneier.com/blog/archives/2015/07/friday_squid_...

The field has more potential than most think. Needs more research. One angle I thought about was applying it to neural networks somehow. Turns out there's a Deep Learning company doing that right now. ;) Many possibilities left to explore.

I thought I was the only one think along these lines. This field has incredible potential...I honestly think there is an Intel-level (or greater) company waiting to be built out of this line of research.

Every once in a while, I hear about FPAA's (field programmable _analog_ arrays), but the concept never seems to take off despite serious efforts, even by Motorola in the 90's.

Isn't the GPAC the same thing?

I suppose the reason these aren't popular is because the use-cases are rather narrow (or at least usually implementable with an ordinary digital approach)?

Well the GPAC is a model of computing, much like a Turing machine, so they are not the same thing, though certainly this model influences the work of FPAA's.

By not take-off, I'm guessing you just mean there's not many people buying these:

http://www.anadigm.com/fpaa.asp

Reply on news.ycombinator.com