limits of analog electronic pid?

[coppercone2]

For low frequency PID control, say of a large mechanical system, how does it compare to a digital system from a performance perspective?

Looking at a simple PID circuit with the best capacitors and resistors, it still looks like you would get at least a 50ppm/c setpoint drift with temperature.

But anyway, are there places where they are still used?

[T3sl4co1l]

How does a PID loop's quiescent output depend on RC values?

Assuming P is very small.

Explicit analog PID controllers are never used, but a variant is common, like the pole-zero or lead-lag compensators.  The difference is having the high frequency zero without limiting DC gain.

Tim

[coppercone2]

How does a PID loop's quiescent output depend on RC values?

Assuming P is very small.

Explicit analog PID controllers are never used, but a variant is common, like the pole-zero or lead-lag compensators.  The difference is having the high frequency zero without limiting DC gain.

Tim

I guess you are right that the actual setpoint does not appear to be effected since its really outside of the loop on the first op-amp *you can use thermally matched resistors*, but the equation terms drift so the response drifts, so the best case tuned oscillation will vary I think.

I think this means the average 'power' will be the same but the weeble wobble will vary as the circuit changes

How come they are never used? D-term noise?

[T3sl4co1l]

I guess you are right that the actual setpoint does not appear to be effected since its really outside of the loop on the first op-amp *you can use thermally matched resistors*, but the equation terms drift so the response drifts, so the best case tuned oscillation will vary I think.

I think this means the average 'power' will be the same but the weeble wobble will vary as the circuit changes

What?

How come they are never used? D-term noise?

Because a type 2 compensator is an op-amp, two resistors and a capacitor.

Tim

[radioactive]

You might find this paper interesting.  They point out the advantages of an analog PID as well as performance compared to digital for their application.  I've always implemented PID digitally, but I can see where there are probably many cases where a programmable-analog approach like this would be highly preferable.

https://aip.scitation.org/doi/pdf/10.1063/1.5010181

[Mechatrommer]

How does a PID loop's quiescent output depend on RC values?
Assuming P is very small.
Explicit analog PID controllers are never used, but a variant is common, like the pole-zero or lead-lag compensators.  The difference is having the high frequency zero without limiting DC gain.
Tim

i think what he meant is this... http://www.ecircuitcenter.com/Circuits/op_pid/op_pid.htm

How come they are never used? D-term noise?

i guess because tuning is easier made in mcu than on analog circuit. mcu's are cheap nowadays, you just snap in a single pic10f206 or attiny10 and voila! there goes your pid controller.

[KE5FX]

Explicit analog PID controllers are never used

I wouldn't say they're "never used."  What about boxes like this one?

[T3sl4co1l]

i think what he meant is this... http://www.ecircuitcenter.com/Circuits/op_pid/op_pid.htm

Yes, that's what I had in mind.  Thank you for adding a visual reference!

i guess because tuning is easier made in mcu than on analog circuit. mcu's are cheap nowadays, you just snap in a single pic10f206 or attiny10 and voila! there goes your pid controller.

PID is popular because it's easily done digitally, even if it has drawbacks like noise sensitivity or poor DC stability (when P is used).  Noise can be dealt with using a small FIR filter, say a 1-2-2-1.

Explicit analog PID controllers are never used

I wouldn't say they're "never used."  What about boxes like this one?

It's not PCB mount.

Tim

[Mechatrommer]

Explicit analog PID controllers are never used

I wouldn't say they're "never used."  What about boxes like this one?

It's not PCB mount.

yes it is, take a look inside anyway that one is general purpose controller.

[IanB]

For low frequency PID control, say of a large mechanical system, how does it compare to a digital system from a performance perspective?

Looking at a simple PID circuit with the best capacitors and resistors, it still looks like you would get at least a 50ppm/c setpoint drift with temperature.

But anyway, are there places where they are still used?

Did you know that in the old days analog PID controllers were not even electronic, but were pneumatic and operated on instrument air?

https://visaya.solutions/article/pneumatic-pid-controller/

I've always thought it would be amusing to model one of these electronically. What is the smallest number of op amps you would need to do so? I have a sense that a single op amp would be sufficient, but I have never really tried to work through it. (The single lever balance system corresponds to an op amp, and the various bellows correspond to direct/lead/lag feedback loops.)

[Mechatrommer]

Did you know that in the old days analog PID controllers were not even electronic, but were pneumatic and operated on instrument air?

not even pneumatic, the first recorded close loop control system is just bare metal some sort of inverted pendulum regulating boiler pressure. mechanical air-gas mixture regulator is still in existence today the last time i saw it in low cost electric generator's gas engine.

[jbb]

Obviously it will depend on a lot of factors, but I think there are a lot of places for analog PID.  Or at least analog PI.

It's very common to see PI or PI + lead-lag in power supply control circuits. Some of them add a feedforward capacitor to the feedback path to provide something like derivative action (specifically, they add a zero to H(s)).

I think that analog PID (or similar all-analog methods) can do well in the following situations:

• Fixed P, I, D gains
• Time constants of a few seconds or less (guesstimate)
• No processor in the box already.
• Especially important for high bandwidths (>1MHz is possible, and rather awful to do digitally)
• Don't need super-fine gain settings
• Can live with some unit-to-unit variation

The following would push me towards digital:

• Variable gains
• High precision gain settings
• Long time constants greater than a few seconds.  (Time constants are achieved with R*C, and there's a limit to how much C can be reasonably deployed.  This means that large Rs may be required, at which point bias and leakage currents will come into play and cause DC or low frequency drift.)
  
• Large control systems with many parameters
• Self tuning
• Advanced control features like dead-time compensation, state estimators or parameter estimators

Hope that's helpful.

[coppercone2]

ah the high value problem.

I wonder if there is a suitable precision capacitance multiplier or gyrator circuit that can be used to deal with the C term. It might end up having better specifications then a gigantic resistor.

Also, what do people mean when a PID controller has a 'complicated' response? Do they mean one that uses higher order calculus operations? Or has some kind of peicewise response (2 PID sections or PID that overlap?)

If you wanna add a jerk, crack, snap, pop, do you just put another parallel analog circuit into the summer (like the math equation?)

[Conrad Hoffman]

AFAIK, analog PID just as shown in the schematic above has been used all over, certainly "never used" isn't right, though today it's probably less common. PID is great until you have a system with a dead-band, say loose mechanical joints, hysterisis or a few other problems. Then it can be a nightmare to get right and (IMHO) digital solutions may do better because you can add "complications" to deal with special conditions.

[rstofer]

ah the high value problem.

I wonder if there is a suitable precision capacitance multiplier or gyrator circuit that can be used to deal with the C term. It might end up having better specifications then a gigantic resistor.

Also, what do people mean when a PID controller has a 'complicated' response? Do they mean one that uses higher order calculus operations? Or has some kind of peicewise response (2 PID sections or PID that overlap?)

If you wanna add a jerk, crack, snap, pop, do you just put another parallel analog circuit into the summer (like the math equation?)

In terms of analog computing, you rewrite the differential equation with the highest derivative, by itself, on the left hand side of  the equation and divide off any coefficient.

Now, assume you actually have that derivative in your hands (you don't, but you soon will!).  You add a series of integrators to work your way back up to the variable itself and you deal with the coefficients as they come up with adders, multipliers, coefficient pots, whatever.  Eventually, you will form the entire right hand side of the equation.

And then there is the magic of the '=' sign...  In the beginning we assumed we had the highest order derivative and now we just created it on the right hand side.  Just add a jumper wire (the '=' sign) to close the loop.

Initial conditions can be applied at any stage but usually it will be the variable itself that has an initial condition, not the 3rd derivative.

This solution of ordinary differential equations was invented by Lord Kelvin.

[rstofer]

I wonder how many PID loops there are in a nuclear power plant?  Initially, these were all analog and I doubt that they have been changed.  Newer plants are probably digital.

How about the carburetor on a car (when they had them).  It had P and D terms but no I term if we assume that the accelerator pump was somehow responsive to a change in throttle position (which it was).

Pneumatic controls for HVAC systems are still pretty common.  The loops aren't usually very tight and intalling pneumatic tubing is fairly simple.  Integration is a piece of large diameter pipe.

This is a really good presentation on HVAC control systems:
http://www.caee.utexas.edu/prof/Novoselac/classes/CE397b/Notes/389H_NO_24_Control.ppt

[LaserSteve]

I often work on galvanometer scanners used for positioning laser beams.  Some of them scan at 4-6 KHz small angle with harmonic components in the drive amplifier in the 10s of KHz.  Until the last year, the cost of DSP based control systems was too expensive for most users, so everything is  analog.  We use a P-I-D-D loop with separate high frequency and low frequency damping. We have  two feedback paths, one from the coil current series resistor  and one from the capacitive or optical position sensors that get weighed and summed.
Usually we have one to three notch filters in the feedback path as well to handle shaft torsional and bending resonances that are often in the tens of kilohertz.

As our scan velocity is a complex function based on angle and time, the loop equations are very complex.  Still we have fast discrete jumps over a sixty degree mechanical angle if we want,  with settling times in the 10s to hundreds of microseconds depending on jump size and mirror mass for the fastest of devices.

We love non technical  beginners in the laser show and marking industry who are suddenly confronted with 12 ten turn pots for adjusting  each axis.  The control interactions are complex, but can be conquered by starting with the basics and using a scope and test patterns.    Often users get frustrated and go to EE school or get a degree in Physics to try to understand what they are working on, and no I'm not kidding, I've nominated more then a few laser show folks for undergraduate or graduate school.

 As you can actually  see subtle changes in the PID settings in your projected  images or marking, they can be quite fun to work on once you get the hang of the adjustments.   Some wags have referred to galvos as the fastest production analog PID loops on the planet, but I'm more then aware of other applications, often automotive or aerospace, that are considerably faster.

A good galvo with a low mass load  is between 10 and 100 times faster in terms of angular velocity  then the fastest hard drive positioner arm...


Steve

[coppercone2]

do the digital systems you use now outperform the analog loop you described or are they simply cheaper and easier?

if they outperform, can you see analog engineering surpassing the currently accepted digital solution?

And since its outdated, can you post some schematics?

[Mechatrommer]

if they outperform, can you see analog engineering surpassing the currently accepted digital solution?

iirc you usually discuss on generalized topic. you cant make a conclusion to the whole or applicable to the whole. each system will be suitable for certain condition. you cant snap HDD arm at 6KHz something might have been broken, or moment of inertia is too great control servo will unable to catch up. engineering is about learning many techniques and applying one particular (or few combinations) technique to a particular problem.

[Kleinstein]

The linear part of a PID regulator can be done quite well in the analog domain, at least if there are no very long time constants. There usually is no need to have the time constants very accurate, so something like capacitor drift is usually not a problem (unless one uses the wrong type of MLCC).

For the long time constants one can avoid the very large resistors / capacitors a little longer by having a divider after the integrator.  Dividing the integrator output by 100 corresponds to a 100 times longer time constant. This limits the possible vales of the integral contribution, but this is often more of a good thing than a problem. One problem with the analog PID is integrator windup, thus a rather slow response when coming out of the limits of linear operation (e.g. after an actuator reached saturation).

Today digital PID essentially always includes some kind of good anti-windup, as in a DSP / µC implementation this is easy and really helps at the bounds. The more normal implementation may not even need another parameter, except the actuator limits. Another advantage of digital regulators is that one can include nonlinear transfer-functions of the sensor or actuator. Quite a few commercial digital PID controllers include some kind of tuning tools / auto-tune.

[LaserSteve]

Here you go, with a few things removed.  This is a simple one without notch filters  or digital inputs.  Feedback is from a pair of opposed photodiodes excited with a current source driven LED. A a sensor  linearity enhancing  circuit has been removed as has the active current source.

Yes, with care and careful compensation, an LM3886 makes a great DC coupled power opamp as long as your gain is more then 20...

 You can learn here as well:

http://www.ecircuitcenter.com/Circuits/op_pid/op_pid.htm

All this is covered in a basic EE controls class the first week of class, no magic...

Steve

[rstofer]

do the digital systems you use now outperform the analog loop you described or are they simply cheaper and easier?

if they outperform, can you see analog engineering surpassing the currently accepted digital solution?

And since its outdated, can you post some schematics?

There is new interest in analog computing.  There are videos and articles all over Google including:



Apparently, analog computing does a better job of pattern recognition among other things.  Neural networks are the current research topic.

... the power budget of the human brain is around 15W, and its computation capabilities range in the 10¹⁷ FLOPS

Does anybody think the 10¹⁷ FLOPS is a staggering number?

from: https://users.ece.cmu.edu/~pgrover/teaching/files/NeuromorphicComputing.pdf

So, it we want to do artificial intelligence and if neural networks are the path, analog is the way to get there.  There is a lot of work being done with advanced analog computing.  It's all over Google!

[chris_leyson]

@LaserSteve. The Cambridge Technology analog galvo controller is a classic bit of design. I had to design a new galvo driver as part of a product redesign and the spec was you have to use a 5V power supply !. It was more or less a copy of the classic CT galvo driver but with a fully differential opamp, LT6362, driving a pair of OPA567's for the output stage. It worked OK for small galvos like CT 6100H and a Chinese PT40 galvo but wouldn't drive anything bigger like a CT 6810. I think I got it down to 4 adjustable pots and one of those was for the notch center frequency, I was about to change the design to use digital pots with an aim to make it auto tuning and then the company went belly up.
It was good fun building an analog galvo controller and I even managed to build a reasonable Spice model for the galvo, didn't get around to modeling bearing stiction but the simulated and actual results were very close. I even started work on a digital control loop but didn't really pursue the idea because you still need a lot of analog electronics for the interface, a digital controller would have saved maybe 4 op-amps. It would have been a nice project if I had the chance to see it through to the end though. Galvo controllers are just one application where a digital control loop would probably out perform an analog loop just because it's so easy to tweak the control algoritms.

[coppercone2]

That circuit is very interesting

It probably belongs in the art of electronics if something similar is not there already. I like the power limiter going into the JFET, where the JFET is like a AGC type deal.

[coppercone2]

on a note irrelevant to PID, how do you come up with 20k/0.1uF to earth ground on the right side?

was that somehow designed in or did you use some kind of measurement to come up with that? Those semi-floated connections freak me out (not safety just design)