Not Your Father’s Analog Computer | Hacker News Reader

Not Your Father’s Analog Computer

spectrum.ieee.org

115 pointssohkamyung23 comments

Open articleView on HN

(bit of history on that Tide Prediction Machine)

You can read the paper underlying the mathematics of tidal computation including pictures and description here:

Special Publication No. 98: Manual of Harmonic Analysis and Prediction of Tides (1940)

https://tidesandcurrents.noaa.gov/publications/SpecialPubNo9...

I worked on a Fortran program which I'm pretty sure is the direct descendant of this paper and the analog computer.

It originated in 1967 and was still used in the mid 1990s. And probably still is used.

One of the best analog computers still in use is the F-16's flight control system. Later F-16s use a digital system, but the classic quadruple-redundant analog computer is still in use. Very 1970s electronics.[1]

[1] https://duotechservices.com/pitch-1-board-repair-of-the-anal...

I would have to assume that it's not still in use by the US, lest this be incredibly classified, correct?

F16's fly for many countries.

The Omega Tau podcast discusses analog computers on episode 159. It’s probably one of my favorites.

http://omegataupodcast.net/159-analog-computers/

Prior links of mine on general-purpose analog, analog accelerators, and/or neural analog:

https://news.ycombinator.com/item?id=11731697

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

Also, while it's somewhat slow traffic-wise, there is an analogcomputing sub on reddit.

https://www.reddit.com/r/analogcomputing/

Since they are doing dedicated hardware anyways, I'd like to see it compared to custom 6-bit digital circuit designed for the same calculations; I expect the power consumption would be much closer to the analog computer.

If a generic programmable analog computer -like the one described in the article- can have power consumption similar to a custom-made digital circuit, you are still gaining something.

I said "designed for the same calculations" so it should be as generic as the analog one. The analog computer described in the article is distinct from general purpose compute units, as it performs calculations in a pipeline.

There's always room for Jell-O (as an analog computer) http://hplusmagazine.com/2014/01/10/defend-your-network-with...

An example from the UK

http://blog.liverpoolmuseums.org.uk/2015/08/tide-predicting-...

Kelvin invented the concept: https://en.wikipedia.org/wiki/Tide-predicting_machine

Is there a FPGA equivalent for these analog chips? I'm guessing no but that would be handy to have field programmable analog computers.

Definitely:

https://en.wikipedia.org/wiki/Field-programmable_analog_arra...

There was also a general-purpose, analog computer that people seem to have done similar things with.

http://www.comdyna.com/gp6intro.htm

https://www.scss.tcd.ie/SCSSTreasuresCatalog/hardware/TCD-SC...

I haven't seen any but what you would really need would be a challenge to put into a chip. Essentially you would want a wiring matrix of that you could use to connect groups of integrator / differentiator units to, and then switchable resistors and capacitors to set rates. Not as straight forward as one might like.

An alternative might be a matrix of DSP 'cores', instead of switching an analog signal around you would move around an I/Q stream. That might be an interesting path to wander down for a bit.

A voltage controlled varactor can be made from 2 non-complimentary mosfets with a common body bias. The differential amplifier can be done with as little as 3 additional fets, but probably more on the order or 7-10 for performance reasons. So you're looking at programmable on chip integrators for around 10 mosfets a piece. Of course, you'd use a technology size larger than what is used for digital logic because the analog toolchain would be more sensitive to problems like channel noise, gate tunneling, etc. You would not want to use a differentiator because they tend to amplify noise in an unstable and uncontrollable fashion and are not required for functional completeness (the analog equivalent of Turing completeness).

Edit: IQ analog processing is extremely common in the high performance RF field but tends to be static and not reprogrammable. FPAA's exist on market but are extremely expensive for what you get. The general purpose analog computer was initially developed by Shannon (yes that Shannon) back in the 40s.

I see they re-invented asynchronous logic, such as Chuck Moore's chip today uses, for their digital fabric.

My dad's analogue computer ran the payroll at Raleigh Bicycles.

They had an arrangement with Cadbury, who had the same machine, that if either broke down at wages time, operators could drive the 60miles and run the batch on the other's machine.

I find it interesting my family heritage is

Father - computer mechanic Grandfather - machine maintenance at John Player Cigarettes Great-Grandfather - machine maintenance at a Nottingham Lace factory

Me - computer programmer / analytics

Forgive my ignorance, but wouldn't payroll be particularly unsuitable domain for analog computing? There is very little that can be considered continuous in payroll, and on the other hand the exactness of values is very important?

You can provide sufficient precision simply by scaling the values up until the level of precision you need (e.g. number of decimal points) is represented by big enough steps up/down in whatever medium is used to make the error rate low enough

E.g. consider if you step on pair of analog bathroom scales and you only care about precision to the kg or lbs - you don't then need the scales to be exact. You just need their precision to be sufficiently big that the needle is sufficiently likely to be closest to the right kg to account for what error rate you're prepared to tolerate.

meaning electro mechanical

So the GGP is actually talking about a different kind of "analog" computer compared to what the article is talking about?

Reply on news.ycombinator.com