https://spectrum.ieee.org/not-your-fathers-analog-computer

 

https://spectrum.ieee.org/not-your-fathers-analog-computer

Seem unlikely. Very unlikely to be practical in this age.

“Because the core of our computer is analog, it can interface directly with sensors and actuators if need be”

Statements like this say these guys are clueless about actual sensor interfacing.

 

I don't think so.

Analog systems are inherently plagued with internal noise, external interference, non-linearity, drift due to environmental conditions, etc.

The focus today is in the senor design, MEMS, nanotechnology.
You don't need fancy amplifiers. You simply digitize the information at the source, along with any environmental parameters, temperature, pressure, humidity, etc. Any correction for non-linearity can be done digitally.

A true 24-bit ADC (i.e. ENOB = 24) is 1 part in 16 million, that is 60 ppb.
Or that is equivalent to measuring a 16V signal to 1μV resolution.

 

the premise is that multiple analog circuitry can do parallel differential equations faster than our digitl system can, the example however lofty, was modeling discrete molecular movement. All in one mS with enough circuits on one chip. It would interface with digital system to output and perhaps further process this approximation data.

 

Seem unlikely. Very unlikely to be practical in this age.

“Because the core of our computer is analog, it can interface directly with sensors and actuators if need be”

Statements like this say these guys are clueless about actual sensor interfacing.

yeah, I’m skeptical also, but if we could do you think there are applications where it could actually work?

 

What I am looking for is computational circuits with neuron equivalents. The odd thing is they will use digital simulations to create the networks , then figure out how to make the networks with analog components. Just MHO.

 

yeah, I’m skeptical also, but if we could do you think there are applications where it could actually work?

There is always a corner-case where something like this might work but as a practical device, no.

 

What I am looking for is computational circuits with neuron equivalents. The odd thing is they will use digital simulations to create the networks , then figure out how to make the networks with analog components. Just MHO.

The brain can compute but it's not an analog or digital computer.

 

There is always a corner-case where something like this might work but as a practical device, no.

I must be thinking of corner cases, like spice modeling and a few other similar cases, where the digital algorithms are horrible because of dithering and or very slow due to complexity.

 

Isn’t an analogue SPICE model… the circuit?

 

Isn’t an analogue SPICE model… the circuit?

Yes! And there are others where we try to model analog in digital and results are less than perfect and less than real time. Therefore, if you for instance used an integrator to model another, it’s results would be quicker than trying to do the calculation in digital form. I believe this is the point of the article. Also that manufacturing is better today and therefore our analog circuits… etc than the ones back in the 70’s.

 

@nsaspook, I think you’re right, there would be limited applications for such a system. It’s still interesting to me.

 

Here are some more articles:

https://blog.degruyter.com/algorithms-suck-analog-computers-future/

https://hackaday.com/2021/10/03/forget-digital-computing-you-need-an-analog-computer/

https://www.analogictips.com/analog-computation-part-1-what-and-why/

 

I've used and repaired analog computer systems. They're not coming back as separate computed device. Anything useful will be integrated into the existing digital framework.

 

https://spectrum.ieee.org/not-your-fathers-analog-computer

I'd agree that there's likely some untapped potential here. I recently bought a couple of contemporary books on this.

The system's ability to compute solutions to differential equations in nanoseconds is one of the key capabilities.

The "structure" of an analog computer is flexible, changes as the modeled problem changes (suggesting that something like an analog FPGA concept might have merits). The system can therefore model a wide variety of physical systems by modeling the mathematics of the system by directly representing its structure.

Unless and until someone decides to seriously invest here we may see little growth but we should not misinterpret lack of investor appeal as an underlying lack of genuine utility.

 

I'd agree that there's likely some untapped potential here. I recently bought a couple of contemporary books on this.

The system's ability to compute solutions to differential equations in nanoseconds is one of the key capabilities.

The "structure" of an analog computer is flexible, changes as the modeled problem changes (suggesting that something like an analog FPGA concept might have merits). The system can therefore model a wide variety of physical systems by modeling the mathematics of the system by directly representing its structure.

Unless and until someone decides to seriously invest here we may see little growth but we should not misinterpret lack of investor appeal as an underlying lack of genuine utility.

Analog subsystems are already included on digital controllers that might need the unique capabilities of analog for signal conditioning.

http://ww1.microchip.com/downloads/en/DeviceDoc/PIC32MK_Information_Sheet_DS50002613C.pdf

The reference voltage (VREF) biases the op amps to VDD/2, therefore bidirectional motor phase current can be sensed using unipolar op amps. The source of VREF can be selected either from the development board or from the internally generated reference voltage using DAC2 and resistors R23/R24, as shown in Figure4. By default, the PIM is configured to source the reference voltage, internally generated using DAC2 by populating R23 and keeping R24 depopulated.To source the reference voltage from Motor Control PIM, R23 needs to be depopulated and R24 must be populated with a zero ohm resistor. The internal op amp configuration and passive resistor-capacitive network configures the filter bandwidth, op amp bias and op amp gain, as shown in Figure4.Table2 classifies the passive components according to their functionality and also specifies the design equations for filter bandwidth and op amp gain.

 

Analog subsystems are already included on digital controllers that might need the unique capabilities of analog for signal conditioning.

http://ww1.microchip.com/downloads/en/DeviceDoc/PIC32MK_Information_Sheet_DS50002613C.pdf

I find analog computers interesting for a similar reason I find functional programming languages interesting. A very refreshing way to look at how to solve problems. Sometimes I think we get stuck in some paradigm and always look at every problem in the same way, this is something Edward de Bono spoke about a lot (he coined the term "lateral thinking").

I work in software much more than electronics, and if you present certain problems to a developer they way they approach it is pretty much dictated by the language they work in, it almost defines the way they think. In some way our tools limit our thinking.

 

I find analog computers interesting for a similar reason I find functional programming languages interesting. A very refreshing way to look at how to solve problems. Sometimes I think we get stuck in some paradigm and always look at every problem in the same way, this is something Edward de Bono spoke about a lot (he coined the term "lateral thinking).

I work in software much more than electronics, and if you present certain problems to a developer they way they approach it is pretty much dictated by the language they work in, it almost defines the way they think.

I find it less refreshing and more of a rehash of things tried and found wanting in general practicality.

 

I find it less refreshing and more of a rehash of things tried and found wanting in general practicality.

Well that's an interesting view, functional languages are based on lambda calculus whereas imperative languages are all based on Turing machines. But lambda calculus and Turing machines are provably equivalent ways of describing computations.

One is not a rehash of the other, they are alternative yet equivalent, in the sense they can each model computation but in quite different ways.

Functional languages are much closer to mathematics, have a purity much closer and that's is an appeal - well for me it is!

 

Here, you guys can try to write a solution to this problem in your preferred language:



The solution must be able to handle a huge input integer, say a million digits or something, it should not crash or hang with very very long inputs.