Snapback, potentiometer degradation and capacitor mod installations

Kadano's guide on snapback, potentiometer degradation and capacitor mod installations

This document both explains snapback and potentiometer issues for customers who want to order a controller or want to have me service their controller and serves as a reference for people looking to adjust their capacitance (scroll down to "finding the optimal capacitance" section).
Apart from that, it's also hopefully a useful resource for people who want to understand controllers better.

Controllers 'naturally' not having snapback, inaccuracies in C-stick inputs, pivot issues and drifting issues are all caused by the same thing: stick potentiometers being worn down. Why that is the case is most easily shown visually, through oscilloscope readings of potentiometers' electrical signal curves in response to stick movement.

In the picture on the right, the yellow line is the potentiometer signal voltage of the horizontal control stick axis. The lowest point corresponds to straight left whereas the highest point corresponds to to straight right.

The two blue lines are the dead zone thresholds (cyan: slightest right, blue: slightest left), the pink line is the dash & smash turn threshold to the right.

Both of these readings show the potentiometer output of the same thumb input: a quick flick to the left, as you'd do it for a reverse airborne neutral-B.

The bottom reading is that of a new potentiometer, whereas the top one is that of a worn one.

You can see that with the new potentiometer, there is snapback – after having flicked left, the potentiometer does read the physical oscillation beyond the dead zone threshold to the right.

The difference in whether the snapback peak is read or not is not the only difference, though. In the worn potentiometer, the voltage curve is irregular, which leads to issues like inconsistent ledgedashes and sometimes getting the wrong aerial on the C-stick.

In the very worst cases, overly worn potentiometers can even drift after directional inputs and take very long to return to neutral. See the reading on the right for an example.

Controllers with this kind of potentiometer condition have issues like dash to short hop nairs becoming fairs instead. That doesn't mean that the controller is beyond saving though, in fact if it has its potentiometers replaced, it will work just like new in terms of stick accuracy.

Original stick potentiometers are rated for 5 million cycles, which might sound a lot, but at 10 hours of dash-dance heavy Melee gameplay a week, that only amounts to 6 months of usage.

So, if you play 10 hours a week on a single controller on average, you have to expect needing your potentiometers to be replaced every 6 months or so.

(Typically, only after about 10-12 months of intense usage the potentiometer response becomes unusable as here, so these 6 months are definitely not a hard limit, but rather an estimate from when on stick performance will be noticeably impacted.)

For pivots, it takes only a small amount of potentiometer weardown in order for pivots to become less consistent.

The comparison on the right is of the exact same potentiometers as the first one, with a new one on the bottom and a moderately worn one on the top, but instead of looking at simple flick inputs, these are pivot inputs – dashing to the left, then smash-turning to the right.

For maximum pivot consistency, the time span spent above the dash threshold (pink line) needs to be very close to 16.67 ms (one frame). In the case of the worn potentiometer, due to the voltage not returning to neutral quickly enough, it's held above the dash threshold for 28 ms (one square is 20 ms). At 28 ms smash turn range time (STRT), the chances of getting polled twice within STRT are (28-16.67)/16.67 = 68%. That means that on a controller with this potentiometer condition, you can only hope for pivot success rates of 32%. The remaining 68% will instead dash to the direction you intended to pivot to.

In the bottom reading with the new potentiometer, STRT was actually just 12 ms, which is also not ideal as this reading can be polled just before reaching and right after exiting the smash turn range, causing the character to not turn around at all.

However, as a new potentiometer reads our inputs properly, this simply means that I didn't flick to the right for long enough. By always flicking for the correct amount of time (16.67 ms), you'd get very consistent pivots.

With the worn potentiometer on the top, there is no way of input that allows for consistent pivot, as the potentiometer inaccuracy leads to every input being delayed, and inconsistently so (every sample is a bit different, even if the manual input was exactly the same).

By adding a capacitor of the ideal size, we can eliminate the snapback that is still present (as you can see by the dip below the blue threshold on the bottom after the smash turn input).

This way, pivots are consistent, yet there is also no snapback.

So, for pivot and flick input consistency, you want to have a controller with new potentiometers and active capacitors. The reason why adjusting the capacitance will be necessary eventually is that potentiometer inaccuracies don't appear all of a sudden, but gradually. As the potentiometer degrades (through usage), less voltage flows through it, increasing the effective delay introduced by the capacitor.

For example, here is a flick input to the left on a new potentiometer with a 334 capacitor (330 nF, medium size):

After 6-10 months of usage, keeping the same 334 capacitor active will make the worn potentiometer respond differently:

You can see that the total amount of time spent below the dead zone threshold to the left is increased by about 100%, which can already be problematic for inputs like dash jump nair (becoming fair instead).

Apart from that, the combination of some potentiometer degradation and capacitance higher than appropriate for the given potentiometer degradation also impacts pivots (not pictured here).

If we completely remove the capacitor, we still have some snapback, though, so while that alleviates the slight drifting issue, it's not an optimal solution.

Accordingly, what we need to do is replace the capacitor with one of less capacitance.

Here, a 154 capacitor (150 nF) was optimal for reliably eliminating snapback yet not introducing drifting delay.

So, for getting the longest span of ideal behavior out of a potentiometer conveniently, we need to have a capacitor mod installation that allows the user to step capacitance up or down, without needing to solder in order to adjust.

That's why when I first documented the modular capacitor mod in 2015, I did so with female jumper wires, that allow quickly sliding capacitors in and out of the sockets, rather than having to desolder and resolder them. The shortcode for this installation on my website is S2b.

Even now, it's still a viable choice, and due to how quickly these wires can be installed, it's the best choice for customers who just want something affordable that gets the job done.

Even more basic installations, such as directly soldered capacitors or jumper wires without providing the full range are not advised as they cannot be tuned by the user.

I always include a full set of spare capacitors to choose from, so that the entire range of 100 nF to 680 nF is available.

Spare capacitors fit perfectly into the backshell below the D-pad, so that they are always available when opening the controller.

Apart from the full set of capacitors, I also include a free tripoint screwdriver with every controller order that a capacitor mod is bought for, since you need to be able to undo the six tripoint screws in order to adjust capacitance.

Adjusting capacitance without an oscilloscope can be a slightly tedious process, as explained below (collapsible):

Finding the optimal capacitance (S2b)

This process is done with Fox or Falco in Melee. I recommend using console only for this, as emulation can drop inputs on some setups, caused by high CPU load or USB driver issues.

1. Make sure that the controller is inserted to the console and properly calibrated (hold X+Y+Start for 3 seconds while not touching or moving the control stick if unsure).

2. Insert a second controller to select another player character, preventing CPUs from interfering. (On 20XX version 4, you can also set the game mode to time, allowing you to advance to the stage select screen with a single character by holding Start for 2 seconds.)

3. Select any stage and hold shine (keep B pressed).

4. To test for horizontal snapback: while holding shine, flick the stick left 40 times (let go of the stick completely and abruptly from full extension) and count how often the character turns to the right. Write down the number and repeat for flicking to the right.

5. To test for vertical snapback: while holding shine, flick downward 40 times. If the character ever jumps out of shine, there is vertical snapback beyond the jump threshold. (Holding shine is not strictly necessary here, however it allows you to not have to focus on the screen constantly – if Fox is not in shine any longer despite you having kept B pressed constantly, he must have jumped at some point.)

6. If for either axis you got 1 or more samples of snapback (wrong direction or jumping), or if you don't have snapback, but do experience stick input issues, take out the six screws that fix the back shell to the front shell and lift off the back shell.

7a. If out of 40 samples, 0 had snapback, yet you do experience input sluggishness (fair instead of dair, side-B instead of neutral-B after flick inputs, dashing problems etc.) you want to reduce capacitance. As a first step, slide out the capacitor of the respective axis (vertical: short wire loop, horizontal: long wire loop) and put it aside into a small box where you'll find it again.

7b. If some of the 40 samples had snapback, you need to step up capacitance. I recommend only going up one step at a time. (See chart below this section.)

8. Redo the shine test for the axes that you adjusted. If you get an instance of snapback before reaching 40 tries, you can stop already, as you've already confirmed that the controller still has snapback and some (extra) capacitance is necessary.

9a. If you reach 40 tries on all directions without a single instance of snapback, your potentiometer is worn to the degree that it doesn't need capacitance any longer.
In this case, put the back shell back on and test the controller for fast directional inputs (like quick reverse short-hop neutral-B, dash jump instant nair). If the controller still has the issue of neutral inputs becoming sideways inputs, your potentiometer is worn out completely and needs to be replaced.
'Easy fixes' such as unclipping and reclipping the potentiometer to the stickbox only alleviate the issue for a short time and are not a reliable solution. I recommend sending the controller to a professional like me for servicing (soldering in new potentiometers).

If instead some snapback was still present on the 40 samples without capacitance, you'll need some capacitance still.
The first recommended capacitance to try depends on the amount of snapback you had. If only 1-8 of the 40 samples had snapback, start with the smallest possible capacitance.
If 9 or more of the samples had snapback, start with a capacitance in the range of 150-200 nF.

With the new capacitance, redo the shine test and if there is still snapback, increase by one step. Keep retesting and increasing by one step until you have no snapback within 40 samples any longer.

9b. If you reached 40 tries on all tested directions after having increased capacitance by one step, you have reached the ideal capacitance for your current potentiometer performance. There is nothing more left to do except putting the controller back together.
If you still have snapback sometimes, you keep increasing by one step at a time and repeating the test.

Of course, you can instead also simply guess which capacitance you should try next, going from 680 nF (684 label) directly down to 220 if you noticed pivots getting less consistent. In most cases, that's completely fine, and you can always readjust if you notice a slight sluggishness in stick response or occasional snapback.

However, if you want to find out the very smallest capacitance that's sufficient to eliminate snapback, minimizing delay and drift, I recommend sticking to the process above.

For customers who prefer paying a bit more for having a faster way of adjusting capacitance, I came up with the idea of installing dipswitch modules in 2018 that toggle capacitors individually. Instead of having to slide capacitors in and out, you flip switches to off or on.

By now, I offer this installation in 3 ways:

S2c 30€ 6x: 4x toggle for the horizontal axis, 2x for the vertical axis.
S2c 40€ 10x: 6x toggle for the horizontal axis, 4x for the vertical axis.
S2d 50€ 10x: 6x toggle for the horizontal axis, 4x for the vertical axis. The capacitor set is intercepted by a transistor-resistor-capacitor circuit that makes it no longer necessary to recalibrate by holding X+Y+Start after plugging in, unlike with S2b and S2c.
(Pictured on the right, with 4 and 10 being enabled for 150 nF on the horizontal axis and 330 nF on the vertical axis)

Finding the optimal capacitance on an S2d installation is largely the same as on an S2b installation, except for the switching between different capacitances being much faster.

Through 62 different capacitance combinations for the horizontal axis, S2d allows for capacitances from 0 to 1130 nF, with steps of 33 nF and smaller, allowing for very precise adjustment.

Of these 62 combinations, only about 8 are typically needed, with the remaining as sub-steps you can adjust to after you've found the rough category. Finally, the capacitances of 700-1130 nF are only needed in few cases, like using Nunchuk or other heavier stick knobs or having a weaker stickbox spring. Typically, they are overly large.

In the sheet below, the 8 main steps are highlighted green. I recommend first choosing one of these, for example, 250 nF by toggling switches 3 and 4 to on, with all others in off position. If you find that this capacitance causes 0 out of 40 shine test samples to have snapback, yet you've found the next-lowest green step (150 nF) to still have occasional snapback, you can try the sub-steps 180, 197 and 230 in this order, stopping as soon as 0 out of 40 samples have snapback.

Now with that being said, keep in mind that these sub-steps are an extra feature for the perfectionist who is obsessed with getting, say, pivots to be as consistent as possible, even if it's just the slightest difference. For the vast majority of users, sticking to the green steps will be enough.

Still, it's good to have the option for maximum precision available – if nothing else, it prevents you from thinking "maybe a capacitance in-between 470 and 680 nF would be just ideal for the controller right now" when having opted for S2b.

Capacitor steps S2b vs S2c-6x vs S2c-10x / S2d

Finding the optimal capacitance (S2d)

This process is done with Fox or Falco in Melee. I recommend using console only for this, as emulation can drop inputs on some setups, caused by high CPU load or USB driver issues.

1. Make sure that the controller is inserted to the console and properly calibrated (hold X+Y+Start for 3 seconds while not touching or moving the control stick if unsure).

3. Select any stage and hold shine (keep B pressed).

7a. If out of 40 samples, 0 had snapback, yet you do experience input sluggishness (fair instead of dair, side-B instead of neutral-B after flick inputs, dashing problems etc.) you want to reduce capacitance. As a first step, write down the currently enabled switches (in 'up' position, for example 4 and 10 in the photo a bit higher up), then disable all the switches for the respective axis (1-6 for horizontal, 7-10 for vertical).

7b. If some of the 40 samples had snapback, you need to step up capacitance. I recommend only going up one step between recommended steps (green highlight) at a time. (See chart above this section.)

8. Redo the shine test for the axes that you adjusted. If you get an instance of snapback before reaching 40 tries, you can stop already, as you've already confirmed that the controller still has snapback and some (extra) capacitance is necessary.

9a. If you reach 40 tries on all directions without a single instance of snapback, your potentiometer is worn to the degree that it doesn't need capacitance any longer.
In this case, put the back shell back on and test the controller for fast directional inputs (like quick reverse short-hop neutral-B, dash jump instant nair). If the controller still has the issue of neutral inputs becoming sideways inputs, your potentiometer is worn out completely and needs to be replaced.
'Easy fixes' such as unclipping and reclipping the potentiometer to the stickbox only alleviate the issue for a short time and are not a reliable solution. I recommend sending the controller to a professional like me for servicing (soldering in new potentiometers).

With the new capacitance, redo the shine test and if there is still snapback, increase by one step. Keep retesting and increasing by one step until you have no snapback within 40 samples any longer.

9b. If you find that you still have snapback in 1 of 40 samples or more, you keep increasing by one (green highlighted) step at a time and repeating the test.

If you reached 40 tries on all tested directions after having increased capacitance by one step, you have reached the ideal broad capacitance step for your current potentiometer performance. If you want to further optimize the capacitance step (completely optional and not usually needed), continue with the step right below the last green step that was not sufficient yet.

For example, if you last found that you still have 2 of 40 samples to have snapback with 430 nF (switches 3+5), but with 527 nF (switches 2+4+5), 0 out of 40 samples have snapback, the next capacitance to test is 463 nF (switches 1+3+5), followed by 503 nF (switches 1+6). (Steps of 33 nF difference or less should be skipped in the higher ranges, as they don't make much of a difference.)

PODE levels and their effects on stick inputs

To describe a potentiometer's condition, it can be helpful to have categories of specific performance states.
I use percentages of remaining lifespan to quantify a potentiometer's performance, so a potentiometer value of 100 means that 100% of the original lifespan still remain.
These ranges are only valid for Noble potentiometers (all black); Matsushita / Panasonic potentiometers (white wiper) do not typically develop the PODE stepping effects, but rather suddenly start drifting and being unusable, needing to be replaced.

90-100: Technical perfection, but low consistency in performing instant smash inputs (smash turn 1-frame dashback without UCF, mashing out of tumble). Will have snapback unless fitted with a capacitor mod. Very consistent pivots, unless an overly high capacitance is fitted.

Typical capacitance necessary to eliminate snapback: 470 to 680 nF, assuming that the stickbox is well-greased with a rather stiff (unworn) spring.

80: Small signs of PODE can be detected with an oscilloscope. These slight irregularities show up as small steps, their effect on gameplay is negligible to non-existent. The main difference is that the remaining snapback is decreased, so the capacitance should be decreased by a bit.

Typical capacitance: 330-470 nF.

60-70: Here, the flick input signal curve has even more steps rather than smooth lines. To the right on the top is a thumb-rest flick input, where the PODE voltage stepping is visible, but has no noticeable functional effect yet, other than further reducing electrical snapback and a very slight delay on inputs (around 2-5 ms).

Typical capacitance: 150-330 nF.

Below is a pre-accelerated thumb flick input, where the faster initial stick speed leads to the voltage stepping being practically instant from deadzone to >0.8 inputs. Potentiometers in these levels thus are consistent with instant smash-level inputs only when doing pre-accelerated inputs. For vanilla Melee, this trait is very desirable for some techniques such as 1-frame dash back. People might call them "dashback controllers" or similar terms.

Important at this stage is setting the capacitance just right. All readings in the 70 range are of the same potentiometers, but the third and fourth ones from the top also have capacitance activated. The third one has the 480 nF that it was likely to be fitted with when the controller was worked on (when the potentiometers were still in the 90-100 range), and here there is a bit of drifting on the return to deadzone:

This can lead to very fast inputs that intended reverse aerial neutral-B resulting in side-B instead, if B is pressed within the ~15 ms of the stick being back in neutral position, but the PODE + capacitance combination still lagging behind electrically.

If we step down to 250 nF, this drifting is eliminated (pre-accelerated thumb input shown):

Keep in mind that PODE is always somewhat irregular, to varying degrees, so not every input sample will look the same. Sometimes, the signal might drift outside of deadzone for just a little bit, and at other times it will stay within deadzone just fine.
By tendency, the more flick inputs you do in a row, the more accurately the potentiometer will read the snapback curve, in other words the PODE level temporarily decreases slightly.

The range of 70 and below is where we not only start to see a noticeable effect of PODE on in-game inputs like dashback, mashing out of tumble and ledgedash, but also have activated capacitance be helpful for that.

Here you can see a thumb-rest flick input done as fast as I can, with no capacitance active:
Not only is there stil snapback, but the vertical voltage step also only happens after a substantial amount of time spent outside of the deadzone.

To better show this, below is the same reading with vertical lines for where the signal leaves the deadzone and where it crosses the smash-level input threshold:

The time that passed between the deadzone and the smash-level thresholds here was 9.0 ms, which means that your effective success rate at things like mash out of tumble / vanilla Melee smash-turn dashback with this input type is (16.67 - 9) / 16.67 = 46%.

When we activate 250 nF of capacitance and do the same type of fast thumb-rest flick inputs, the capacitor holds the slow movement from the center close for longer, causing the drop to smash threshold to happen near-instantly:

At only 1.4 ms TTRT (tilt-turn range time), our success rate is 91.6%.

Again, due to the inconsistent nature of PODE, the initial gradual movement can sometimes continue for a few ms beyond deadzone before the vertical drop, even with 250 nF and very fast thumb-rest inputs

Example reading for that:

Here, we have a TTRT of 6.6 ms, for a success rate of 60%.

So, on medium PODE controllers, we can decrease TTRT by activating an amount of capacitance that 'eats up' some of the initial start of horizontal inputs, yet triggers suddenly with stronger inputs. Keep in mind that this is only possible with a noticeable degree of PODE – on potentiometers in the 90-100 range, there is no such desirable effect on TTRT.

If you want to aim for this effect, it's important to not overdo it. While going with an overly large amount of capacitance is sure to amplify your PODE towards a sudden vertical drop, the drop will be less vertical at the end, and more curved towards the terminal voltage. The threshold above which capacitance stops to be more beneficial for TTRT than lower capacitance is around 600 nF.

Here is an example for the same 70 range potentiometer of all the screenshots before, but with 1050 nF of capacitance active:

TTRT is 3.0 ms, for a smash turn success rate of 82%. The drifting range at the end of the flick input is now increased to a whopping 21 ms, which makes for a very large risk of getting accidental side-B rather than neutral-B.

For this specific potentiometer condition, 200-250 nF is the sweetspot. Any smaller and there will sometimes be snapback, plus the thumb-rest TTRT is undesirably long.
Any higher than 250 nF and there is noticeable drifting on the return path, plus TTRT actually increases again (unwanted).

50: Potentiometers in this range have barely any snapback and only require about 100 nF capacitance to eliminate their snapback. Pivots are usually still consistent, but not quite as easy as with brand new potentiometers. The return path is still decently fast, but can be slightly sluggish (light drifting before returning to deadzone, as seen in the lowest example).

All three example readings are of the same potentiometer, taken within just minutes of each other. This also illustrates how inconsistent PODE at these levels tends to be.

40: There is no longer any snapback, even without any capacitance active. There can be a bit of drifting on the way back, in the range of 10-20 ms.

Typically, this PODE level is reached after the manufacturer-rated lifetime of 5 million cycles or about 6 months of intense usage (10 hours competitive gameplay per week). This is about the earliest that it can make sense to send a controller in for having its potentiometers replaced.

(In earlier stages, having all potentiometers replaced for new ones can still be a good idea as it effectively resets the accuracy lifespan of the controller. I recommend to have this done when you send in a controller for servicing anyway – such as refiling the notches after wearing them down.)

20-30: Drifting is longer, at 20-30 ms, and happens consistently. TTRT is usually very short, for consistent smash turns. Pivots are less consistent, and inputs intended for reverse aerial neutral-B sometimes cause accidental side-Bs due to the sluggish return (only when the B input follows the directional input very quickly).

If the vertical potentiometer is affected with PODE in this range, fast vertical inputs like short hops with Fox / Sheik and Peach instant FC nairs are impacted, becoming full hops and dairs, respectively.

On the right, you can see a very short flick input intended for short hop, but due to the strong PODE, the signal stays above the jump threshold for 47 ms.
Short hops with Fox and Sheik are only 100% consistent if the jump threshold is only surpassed for up to 33.3 ms – this potentiometer response only allows for 18% short hop success rate.

10: Drifting is very long, at 30-40 ms, and pivots are almost impossible. For ledgedashes, having the horizontal potentiometer only be affected to up to this degree can be helpful with the down-forward input method, however at this point the impact on other inputs like reverse neutral-B, pivots and even dash-dancing is rather strong and perhaps not worth that trade-off.

0: The potentiometer is worn out completely and will sometimes drift for ridiculously long intervals, of 40-1000 ms or even longer.

While worn potentiometers will sometimes work properly for a while and can be 'reset' (clipped off from the stickbox and then clipped back on) into not drifting for typically 1-10 days, dealing with their occasional to frequent severe issues is frustrating and at this point, they should be replaced as soon as possible.

Potentiometers usually reach this stage after 12-24 months of intense usage.

Some potentiometers develop drifting to this degree even without other symptoms of PODE being present. On the right are several flick inputs next to each other, zoomed out timescale, and you can see how they never return to neutral properly.

Panasonic / Mitsushita potentiometers (white / light gray wiper) usually transition very quickly from 90-100 performance straight to 0 performance and start drifting.

Strangely, they usually keep having snapback even while drifting. I personally only use them on request and default to Noble potentiometers. For people who want to keep their pivots pristine for as long as possible, the white wiper potentiometers might be the superior choice. I cannot confirm that yet as I have too little long-term experience with them.