The big question is whether there are spurious transitions from noise or EMI. Generally, the pull-up or pull-down is strong enough so that, except while switching, the raw signal will stay reliably high or reliably low. If this is true, then you can act as soon as you see the first change in the raw signal. Then you have a time constant before you will consider a change in the opposite direction to be valid. That way, no matter how quickly the user taps a button, you will a) act as soon as possible, and b) not miss a tap.
lambdaone · 5h ago
I do something very similar, but use two time constants. One to determine how many consecutive samples are needed to change from the established state, and a second one, as suggested here, to ignore subsequent input for a bit after a registered state change to allow for bouncing before going back to the receptive state. This is a bit more robust, as practical systems are always at the mercy of occasional glitches in electromagnetically noisy environments.
With well-chosen constants, I've found similar behaviour to the above; low latency and high performance at once.
progbits · 6h ago
This is mechanical bouncing.
lambdaone · 4h ago
Real environments have both mechanical bouncing and unavoidable EMI spikes, even when you have done your best to eliminate EMI; particularly when you're dealing with switches, which are typically on the end of long wires which are effectively antennae. Source: actual experience with crappy real-world systems.
lambdaone · 4h ago
A side-note: I have a friend who is an electronics genius who managed to construct a production system that combined a multi-kilowatt electric arc lamp and a CCD array. Microvolts of CCD signal at megahertz of bandwidth mixed with hundreds of watts of RFI all in the same chassis. Oh, and of course lots of low-jitter digital timing and sequencing electronics and wideband ADCs.
It all worked perfectly. He used every trick in the book, and some new ones he invented from scratch, to implement it - I've never seen anything like it, before or since.
progbits · 3h ago
Sounds fascinating, can you share what it was for?
Regarding your parent message, I agree, but the EMI stuff will depend on lots of other factors and so you will have to do your own modeling / measurements. While the mechanical bouncing should be pretty uniform so this is a nice data source for picking a well behaved switch and designing debouncing network.
Of course I'm just realizing this is N=1 for the switch so hard to say what manufacturing tolerances each one has.
londons_explore · 6h ago
Mechanical bouncing should be pretty predictable - ie. A particular bit of metal briefly resonating at a frequency based on the speed of sound in the material.
However, these traces look more chaotic.
Anyone know why?
progbits · 4h ago
A bad switch will simply put contacts together at the same rate as the human pressing it, so it's very random and variable as you press with different strength and speed.
A good switch will decouple your finger from the contact by some nonlinear/bistable linkage, so once you accumulate a fixed amount of travel distance it will snap the contacts with the same speed/force. But of course even that isn't perfectly decoupled nor perfectly deterministic.
lambdaone · 4h ago
The system is hit so violently that it goes into a non-linear regime, repeatedly hitting the end-stops of its motion. Consider how adding even a small amount of nonlinearity to a simple harmonic oscillator can create chaos, and I think you should be able to see how this generates chaos.
tonyarkles · 2h ago
Plus, at the interface, there’s some small amount of oxide that has grown on the surfaces. When those pieces of metal are being banded together and separated there’s tiny electric arcs that are working at getting rid of that.
formerly_proven · 6h ago
I’ve never seen a collection like this, it’s quite amazing.
It’s worth pointing out that due to the x1 probe there’s an input capacitance of about 50-60 pF, which would result in a 200 kHz low-pass for opening and dampen high speed oscillations in general. You can see this in the graphs with the rising transitions being slightly rounded. MCUs have much lower capacitance in their pins, so reality would tend to look even messier than this, especially zoomed in.
lambdaone · 4h ago
A bit of analog low-pass filtering and electrical hysteresis definitely helps reject EMI - but you shouldn't use these alone to do debouncing as if you do, you introduce substantial latency for both up and down transitions. The (very simple) software algorithm described above gives you both low latency and complete resistance to bouncing. Combining the two is a complete solution.
nickdothutton · 6h ago
Knew this would be good, was not disappointed.Glad I don't work in hardware.
joezydeco · 3h ago
Jack Gansssle wrote the definitive guide on debouncing for embedded systems.
This is a very trivial problem, it doesn't even feel right calling it a 'problem'. It's probably the second lesson for the greenest Arduino newbie, right after blinking an LED. Don't let it discourage you from embedded hardware.
lambdaone · 5h ago
Doing it right (via the algorithm above) is a trivial few lines of code. The thinking behind the state machine that gives both high reliability and low latency isn't trivial. It is, however, obvious once you know the trick of maintaining an extra auxiliary variable in addition to counter/timer variables, and understand that you don't need to wait for the bouncing to finish before registering a change. Think sort algorithms; everyone comes up with bubblesort first, but you can do much, much better with only tiny changes after a bit of thought.
tonyarkles · 2h ago
And yet… I’ve seen it done wrong so many times, resulting in either spurious inputs or lost inputs, depending on how it’s been implemented.
I agree with you, don’t let it discourage you from embedded hardware. If you want to be successful in the embedded space, though, this is a great easy example to show that you need to take into account a bunch of physical effects when you’re doing hardware. Capacitors change their value with temperature and applied voltage (!), resistors have capacitance and inductance (which may or may not matter), clock crystals change frequency with temperature, switches bounce, ADCs have sampling capacitors that need to charge, chips power themselves up from active-high TTL lines, etc.
I’ve been doing embedded for 20 years now and one of the biggest things I’ve encountered with new learners is that the Arduino platform is amazing for people to get into the field but they’ll hit a point where something mostly works and they don’t have the rest of the background necessary to debug or understand why it doesn’t work as well as they thought it would. Or works great and then “randomly” blows fuses/burns out of there isn’t a fuse.
formerly_proven · 4h ago
On the one hand yes on the other hand I’d boldly claim incorrect debouncing is the most common (and usually exceedingly obvious) defect in embedded systems.
https://www.ganssle.com/debouncing.htm
With well-chosen constants, I've found similar behaviour to the above; low latency and high performance at once.
It all worked perfectly. He used every trick in the book, and some new ones he invented from scratch, to implement it - I've never seen anything like it, before or since.
Regarding your parent message, I agree, but the EMI stuff will depend on lots of other factors and so you will have to do your own modeling / measurements. While the mechanical bouncing should be pretty uniform so this is a nice data source for picking a well behaved switch and designing debouncing network.
Of course I'm just realizing this is N=1 for the switch so hard to say what manufacturing tolerances each one has.
However, these traces look more chaotic.
Anyone know why?
A good switch will decouple your finger from the contact by some nonlinear/bistable linkage, so once you accumulate a fixed amount of travel distance it will snap the contacts with the same speed/force. But of course even that isn't perfectly decoupled nor perfectly deterministic.
It’s worth pointing out that due to the x1 probe there’s an input capacitance of about 50-60 pF, which would result in a 200 kHz low-pass for opening and dampen high speed oscillations in general. You can see this in the graphs with the rising transitions being slightly rounded. MCUs have much lower capacitance in their pins, so reality would tend to look even messier than this, especially zoomed in.
https://www.ganssle.com/debouncing.htm
I agree with you, don’t let it discourage you from embedded hardware. If you want to be successful in the embedded space, though, this is a great easy example to show that you need to take into account a bunch of physical effects when you’re doing hardware. Capacitors change their value with temperature and applied voltage (!), resistors have capacitance and inductance (which may or may not matter), clock crystals change frequency with temperature, switches bounce, ADCs have sampling capacitors that need to charge, chips power themselves up from active-high TTL lines, etc.
I’ve been doing embedded for 20 years now and one of the biggest things I’ve encountered with new learners is that the Arduino platform is amazing for people to get into the field but they’ll hit a point where something mostly works and they don’t have the rest of the background necessary to debug or understand why it doesn’t work as well as they thought it would. Or works great and then “randomly” blows fuses/burns out of there isn’t a fuse.