Maybe I'm old but that's common knowledge, the reason more and more computations are moving to the digital world is weirdly the simplicity of implementation and not accuracy, speed, price, reliability or anything else super sophisticated.
It's much easier, faster, maintainable and usually cheaper to add a cheap controller and peripherals than to find an electrical engineer to do the math and design a circuit.
Actually it's usually better in the digital domain for all of those reasons. Most systems these days have minimal transducer processing stuff in the analogue domain and get it into the digital one as quickly as possible.
Analogue systems have a lot of non-ideal characteristics such as drift, thermal sensitivity, parasitic capacitance and inductance, impedance matching problems, insane cumulative errors and manual tuning requirements that don't exist in the digital domain. They are also more expensive to modify, and the price goes up significantly with complexity and speed. Not to mention small and fast is really difficult in the analogue domain thanks to everything being a transmission line.
I fly a plane with an old analog computer autopilot/flight director. It works perfectly well. Its failure modes are well understood and easy to diagnose. The boards and components can be easily troubleshot and repaired by competent electricians.
Modern autopilots are all sensors, DACs, ADCs, and various digital computing resources. Cheaper, faster, more featured, and mystifying in their failure modes.
Although I am a big digital person, even for photography, audio and other seemingly analog things - I admire and appreciate the one major analog computer in my life.
Alas, after 7 years, it finally needs a repair. Sigh.
(It is broken because:) When engaged, the altitude hold function commands a 10 degree pitch up attitude in all cases. This is a reasonably extreme pitch. Even the built in “take off/go around” mode usually commands only about 7-8 degrees nose up.
The altitude is “held” using a “static” pressure sensor and pitch is adjusted in a control and feedback loop to attempt to keep that signal constant. When it works, it is very smooth.
The system is not as a whole redundant, but there are many failsafes. You can put it into a mode where the system does not fly the plane but tells you what it thinks you should do. This is called a flight director. You fly it almost like a video game, get the pointers aligned.
Each component has some redundancy. It has a self test, an annunciator panel saying what mode it is in, a mode control panel, several controls on the yoke including a kill switch, several circuit breakers (one, the electric trim, which has a big red flange on it - see the recent 737 crash artickes), etc.
The workflow of using the autopilot is always command a change, check that the change is reflected on the annunciators and that the plane is adjusting attitude and such appropriately, then resume the normal scan. If at any time anything looks off my instant reaction is always to hit the big red button and take the plane’s control back. When you do this you need to be wary for an out of trim condition and other sudden forces on the yoke and rudder.
TFA explains the nuances of the trade-off. For certain domains (simple calculations at extremely high frequency with tolerance for approximation), analog is a good choice.
I still have a few operational transconductance amplifiers lying around which I tried to use for my guitar effects projects(none of them ever reached completion of course).
Not obvious to work with, but a great solution if you want to multiply two signals.
Other ways to do this:
- Vactrols (a LED/LDR combination)
- Blackmer VCA Chips (e.g THAT2181 or SSM2164)
- Analog Multiplier ICs (these are actually four-quadrant multipliers), (e.g. AD 633)
- PWM switching
- ...
>"Q: What are some of the issues associated with analog computation?
A: They are similar to the ones that are of concern whenever analog functions (op amps and more) are used for signal processing or signal conditioning. These include linearity, offset, temperature-related drift, aging, and bandwidth."
I understand bandwidth but could someone elaborate in lay terms what these other "concerns" are in analog circuits? Specifically what are linearity and drift?
Linearity and drift already covered by the other the other folks...
Offset: When something is supposed to read a specific value (usually voltage), what is it actually reading? Most op-amps (configured as buffers) won't actually read 0V if 0V is applied. Usually this offset is super small (10s of mV, or even way less) in modern op-amps. Sometimes it's offset current. Many op-amps specify both current and voltage offset.
Temperature-related drift: This is component value change specifically related to temperature. Multi-layer ceramic capacitors are especially susceptible to this. A 'good' MLCC might 'only' drift +/-20% across temperature
Aging: As a component ages, how does its value change? Electrolytic capacitors have this problem. If you've ever heard of people talking about replacing capacitors on old TVs or what-not, it's almost always those caps. The paste in the electrolytic caps dries out and affects the value of the capacitor. Once this reaches a certain point, the circuit won't work anymore.
Note I provided specified examples for each of those problems, but they exist across pretty much every type of component to various degrees. For example, linearity can be a big issue on ceramic capacitors. Their values can change as much as 80% (yes, 80%) with applied voltage which can cause some really weird non-linear behavior to some circuits. Most modern silicon probably doesn't have many problems with aging unless it's driven to the ragged edge of its performance or there's an actual defect, since typically chips are qualified to run for a minimum of 100,000 hours. That being said, eventually it WILL fail if it's running constantly. Electromigration affects all silicon. By comparison, usually electrolytic capacitors you'll be lucky to get 10,000 hours out of them. Less the hotter they get and the more the voltage applied is.
Linearity: you expect a circuit to e.g. always multiply an input with 2x. Turns out it doesn't, it does so close to e.g. 0, but multiplies by less or more at higher values. Somewhat generalized, for other components with other curves of behavior, they also don't stick perfectly to the mathematical model, especially when the environment changes. Which leads us to
Drift: temperature changes, behavior of component changes (e.g. resistors change slightly in resistance, circuits to generate precise voltages produce slightly different voltages, oscillators change in frequency). E.g. precision frequency sources often have a heating element to keep their core oscillator at a stable frequency, audio devices thermally couple two components of which the ratio is important, so the changes cancel out as much as possible, ...
Linearity: if a system is linear you can draw a nice straight line on an X-Y plot of input vs output. Nonlinearity is a measure of how much it deviates from that line.
Drift: component values are not fixed! Most vary slightly with temperature, and capacitors change value over time too. Can be seen in any kind of old equipment which needs to "warm up".
Not really, they're separate parameters. Things like crystals and diodes are subject to drift without being linear at all. And ADC/DAC designers worry about linearity a lot while not being concerned about drift, because that gets outsourced to the reference voltage.
Maybe I'm old but that's common knowledge, the reason more and more computations are moving to the digital world is weirdly the simplicity of implementation and not accuracy, speed, price, reliability or anything else super sophisticated.
It's much easier, faster, maintainable and usually cheaper to add a cheap controller and peripherals than to find an electrical engineer to do the math and design a circuit.
Actually it's usually better in the digital domain for all of those reasons. Most systems these days have minimal transducer processing stuff in the analogue domain and get it into the digital one as quickly as possible.
Analogue systems have a lot of non-ideal characteristics such as drift, thermal sensitivity, parasitic capacitance and inductance, impedance matching problems, insane cumulative errors and manual tuning requirements that don't exist in the digital domain. They are also more expensive to modify, and the price goes up significantly with complexity and speed. Not to mention small and fast is really difficult in the analogue domain thanks to everything being a transmission line.
> insane cumulative errors
Any sufficiently advanced analog computer is indistinguishable from a noise generator.
I fly a plane with an old analog computer autopilot/flight director. It works perfectly well. Its failure modes are well understood and easy to diagnose. The boards and components can be easily troubleshot and repaired by competent electricians.
Modern autopilots are all sensors, DACs, ADCs, and various digital computing resources. Cheaper, faster, more featured, and mystifying in their failure modes.
Although I am a big digital person, even for photography, audio and other seemingly analog things - I admire and appreciate the one major analog computer in my life.
Alas, after 7 years, it finally needs a repair. Sigh.
Does it have some kind of built-in redundancy? And how are you aware that it needs repair?
(It is broken because:) When engaged, the altitude hold function commands a 10 degree pitch up attitude in all cases. This is a reasonably extreme pitch. Even the built in “take off/go around” mode usually commands only about 7-8 degrees nose up.
The altitude is “held” using a “static” pressure sensor and pitch is adjusted in a control and feedback loop to attempt to keep that signal constant. When it works, it is very smooth.
The system is not as a whole redundant, but there are many failsafes. You can put it into a mode where the system does not fly the plane but tells you what it thinks you should do. This is called a flight director. You fly it almost like a video game, get the pointers aligned.
Each component has some redundancy. It has a self test, an annunciator panel saying what mode it is in, a mode control panel, several controls on the yoke including a kill switch, several circuit breakers (one, the electric trim, which has a big red flange on it - see the recent 737 crash artickes), etc.
The workflow of using the autopilot is always command a change, check that the change is reflected on the annunciators and that the plane is adjusting attitude and such appropriately, then resume the normal scan. If at any time anything looks off my instant reaction is always to hit the big red button and take the plane’s control back. When you do this you need to be wary for an out of trim condition and other sudden forces on the yoke and rudder.
TFA explains the nuances of the trade-off. For certain domains (simple calculations at extremely high frequency with tolerance for approximation), analog is a good choice.
Same reason why the majority of CPU's are synchronous instead of asynchronous, cheaper to design by far.
Omega Tau podcast has quite a long episode on analog computers.
Episode link: http://omegataupodcast.net/159-analog-computers/
I still have a few operational transconductance amplifiers lying around which I tried to use for my guitar effects projects(none of them ever reached completion of course).
Not obvious to work with, but a great solution if you want to multiply two signals.
Other ways to do this: - Vactrols (a LED/LDR combination) - Blackmer VCA Chips (e.g THAT2181 or SSM2164) - Analog Multiplier ICs (these are actually four-quadrant multipliers), (e.g. AD 633) - PWM switching - ...
The FAQ states:
>"Q: What are some of the issues associated with analog computation?
A: They are similar to the ones that are of concern whenever analog functions (op amps and more) are used for signal processing or signal conditioning. These include linearity, offset, temperature-related drift, aging, and bandwidth."
I understand bandwidth but could someone elaborate in lay terms what these other "concerns" are in analog circuits? Specifically what are linearity and drift?
Linearity and drift already covered by the other the other folks...
Offset: When something is supposed to read a specific value (usually voltage), what is it actually reading? Most op-amps (configured as buffers) won't actually read 0V if 0V is applied. Usually this offset is super small (10s of mV, or even way less) in modern op-amps. Sometimes it's offset current. Many op-amps specify both current and voltage offset.
Temperature-related drift: This is component value change specifically related to temperature. Multi-layer ceramic capacitors are especially susceptible to this. A 'good' MLCC might 'only' drift +/-20% across temperature
Aging: As a component ages, how does its value change? Electrolytic capacitors have this problem. If you've ever heard of people talking about replacing capacitors on old TVs or what-not, it's almost always those caps. The paste in the electrolytic caps dries out and affects the value of the capacitor. Once this reaches a certain point, the circuit won't work anymore.
Note I provided specified examples for each of those problems, but they exist across pretty much every type of component to various degrees. For example, linearity can be a big issue on ceramic capacitors. Their values can change as much as 80% (yes, 80%) with applied voltage which can cause some really weird non-linear behavior to some circuits. Most modern silicon probably doesn't have many problems with aging unless it's driven to the ragged edge of its performance or there's an actual defect, since typically chips are qualified to run for a minimum of 100,000 hours. That being said, eventually it WILL fail if it's running constantly. Electromigration affects all silicon. By comparison, usually electrolytic capacitors you'll be lucky to get 10,000 hours out of them. Less the hotter they get and the more the voltage applied is.
Linearity: you expect a circuit to e.g. always multiply an input with 2x. Turns out it doesn't, it does so close to e.g. 0, but multiplies by less or more at higher values. Somewhat generalized, for other components with other curves of behavior, they also don't stick perfectly to the mathematical model, especially when the environment changes. Which leads us to
Drift: temperature changes, behavior of component changes (e.g. resistors change slightly in resistance, circuits to generate precise voltages produce slightly different voltages, oscillators change in frequency). E.g. precision frequency sources often have a heating element to keep their core oscillator at a stable frequency, audio devices thermally couple two components of which the ratio is important, so the changes cancel out as much as possible, ...
Linearity: if a system is linear you can draw a nice straight line on an X-Y plot of input vs output. Nonlinearity is a measure of how much it deviates from that line.
Drift: component values are not fixed! Most vary slightly with temperature, and capacitors change value over time too. Can be seen in any kind of old equipment which needs to "warm up".
Thanks for the wonderful and detailed responses. I really appreciate it.
One follow up - is it accurate to say that linearity and drift are always related then?
Not really, they're separate parameters. Things like crystals and diodes are subject to drift without being linear at all. And ADC/DAC designers worry about linearity a lot while not being concerned about drift, because that gets outsourced to the reference voltage.
They will come back in the form of continuous variable quantum computers.
Distinct quanta ... variable?