Aardvark Daily

Please visit the sponsor!

Is analog computing about to make a come-back?

Although most of the time, when asked to think of the first computer I ever built, my mind casts back to the collection of logic gates, memory chips and the feeble microprocessor that I threw together in 1977, there was a system I built some ten years before that.

What's more, the computer I built in 1967 was surprisingly fast and used much less power than that 1977 work of rainbow-cable macrame.

Unfortunately, my '60s computer was limited to addition and subtraction - although it could have been expanded to perform basic multiplication and division.

The cool thing about this computer was that it had just four components.

In essence, it was a battery, two potentiometers (variable resistors) and a moving-coil meter.

By placing the two potentiometers in series, the knobs could be turned to represent any range of values between the upper and lower limits -- those values being summed and the resulting current driving the carefully calibrated meter.

This was a very simple analog computer of the kind I'm sure many kids built and quickly grew tired of.

The big problem was that it didn't deal in exact amounts -- only approximations.

If you set one knob to '2' and the other to '3' then the meter would read '5' -- but there was always a margin of error produced by our limited ability to set the knobs and read the meter. Parallax was a major issue with this simple computer.

This kind of error was also present in the world's most popular analog computer -- the slide rule. The accuracy of figures calculated using a slide-rule was very much dependent on the user's ability to accurately "read between the lines" so to speak.

Of course this is where digital computers shine. If you enter a value of 2 then it's exactly 2 and not 2.00000001 or 1.9999999. Similarly, if the computer says the answer is 5 then you know that's what it is. Digital computers are deadly accurate (at least when dealing with integers) whereas analog computers would always produce a certain amount of error.

For trivial calculations then this analog error may not be much of a problem but once you start trying to computerise long chains of complex equations, the analog error can rapidly accumulate and produce massive output errors -- which is why we now tend to rely almost solely on digital computers.

However, I see there may be a resurgence in the popularity of the analog computer simply because it's ideally suited to some tasks that don't require perfect accuracy and which would otherwise require an awful lot of digital computing resource.

Think about some of the applications that digital computers are currently put to and you can see the potential for their analog brethren to make a comeback. Anything which involves "fuzzy" logic might be a candidate. It's somewhat ironic that in order to get a digital computer to come up with imprecise results, we have to fudge things -- whereas an analog computer introduces fuzziness as a standard feature.

The term being used for this kind of imprecise computing is "probabilistic" and is indicative that the answers will probably be close to the true answer but need not be perfectly accurate.

One example of a good candidate for probabilistic computing in this ArsTechnica report is the issue of spam filtering. In theory, an analog (probabilistic) processor could be used to perform key parts of the spam filtering algorithms far more efficiently than its digital peers.

I guess that in computer technology, just like fashion, what goes around comes around.

Even solid-state magnetic memory seems to be set for a come-back.

So what other applications can readers think of where the probabilistic results produced by an analog computer may be a viable alternative to digital MIPS?

Please visit the sponsor!

Have your say on this...

PERMALINK to this column

Oh, and don't forget today's sci/tech news headlines

Rank This Aardvark Page

Change Font