Issue Concerning Rebound
Although the texture analyzer does measure the switch rebound, we discovered that there is an issue with our testing method that throws off the measurement slightly.
Note that in our charts, at the end of the key travel there’s a sharp upward spike in the force measurement. This spike is where the probe found the end of the key travel and encountered resistance. (Because of this spike, we can look at the raw data and determine the end point of each switch’s key travel.)
Bear in mind that switches are made of plastic, which is of course a material that compresses, and further, the probe can apply high amounts of force and can flex even a sturdy metal backplate. Therefore, the spiked portion of the line measures the force the probe applied and the distance the probe traveled, not necessarily the force and distance of the switch, because the probe will continue to depress (and compress) the switch far beyond what human fingers can.
When the probe retracts, the force drops down from that spike. By looking at where the spike ends, we can also find where the rebound of each switch begins, and the texture analyzer measures the force and distance of the rebound.
For individual switch measurements, this is not an issue. In the graphs, you can plainly see where the travel ends (at the beginning of the spike) and where the rebound begins (at the end of the spike). Looking at the raw data, we can further pinpoint the end of the travel and the beginning of the rebound.
However, when we combine all the switch data together to look for a median, we encountered two problems, because we were comparing raw data line-by-line.
Because the ends of the travel are all slightly different switch-to-switch, and the amount of force/distance during the spike was slightly different switch-to-switch, the beginning of the rebound switch-to-switch didn’t line up well at all.
This first problem is easily solved by identifying the ends of the travel for each switch and the beginning of the rebounds of each switch and aligning them. However, this method leaves some small gaps in the data that we had to account for.
The second problem is that, upon close examination of our raw data, we discovered that the distance measurements, switch-to-switch, no longer aligned precisely once the switches begin their rebounds.
For example, below is a spreadsheet showing measurements of multiple switches on the keypress - note that the distance is identical across all cells, and thus the distances between each measurement are uniform:
| Row | [Name of keyboard and key X] | [Name of keyboard and key X] | [Name of keyboard and key Y] | [Name of keyboard and key Y] | [Name of keyboard and key Z] | [Name of keyboard and key Z] |
|---|---|---|---|---|---|---|
| Row 0 - Cell 0 | Force (g) | Distance (mm) | Force (g) | Distance (mm) | Force (g) | Distance (mm) |
| 1 | 53.9 | 2.838 | 54.9 | 2.838 | 55.4 | 2.838 |
| 2 | 54.4 | 2.848 | 55.6 | 2.848 | 54.8 | 2.848 |
| 3 | 54.9 | 2.858 | 55 | 2.858 | 54.4 | 2.858 |
| 4 | 54.6 | 2.868 | 54.8 | 2.868 | 54.5 | 2.868 |
| 5 | 54.2 | 2.878 | 55.1 | 2.878 | 55.1 | 2.878 |
| 6 | 54.9 | 2.888 | 56.1 | 2.888 | 54.6 | 2.888 |
| 7 | 55 | 2.898 | 55.7 | 2.898 | 54.1 | 2.898 |
But during the rebound, those same switches give us data like this (below). Note that some of the distances are different across the cells, and also the distances between each measurement are not uniform:
| Row | [Name of keyboard and key X] | [Name of keyboard and key X] | [Name of keyboard and key Y] | [Name of keyboard and key Y] | [Name of keyboard and key Z] | [Name of keyboard and key Z] |
|---|---|---|---|---|---|---|
| Row 0 - Cell 0 | Force (g) | Distance (mm) | Force (g) | Distance (mm) | Force (g) | Distance (mm) |
| 1 | 52.5 | 3.958 | 53 | 3.871 | 46.9 | 3.727 |
| 2 | 49.7 | 3.928 | 51.3 | 3.827 | 50.8 | 3.677 |
| 3 | 48 | 3.892 | 49.4 | 3.777 | 54.4 | 3.627 |
| 4 | 46.5 | 3.848 | 51 | 3.727 | 54.2 | 3.577 |
| 5 | 44.2 | 3.798 | 55 | 3.677 | 52.9 | 3.527 |
| 6 | 45.4 | 3.748 | 56.1 | 3.627 | 54 | 3.477 |
| 7 | 49.1 | 3.698 | 56.5 | 3.577 | 54 | 3.427 |
Were the deltas between each distance measurement still uniform, our solution to the first problem (manually aligning the beginning of each switch’s rebound) would solve this second problem. But because the deltas are not uniform, it cannot.
Therefore, in looking at the median performance of a given set of switches, we have had to sacrifice a small amount of precision (artificially filling in gaps in measurements) for the sake of accuracy (better showing the median of the ends points and of the rebound beginnings). Further, because of our current testing procedure, the data showing the median of the full rebound is also accurate switch-to-switch but not precise as a true median.
Again, none of the above affects the measurements and graphs of the individual switches, including the rebound. However, because we use the median of all the switches as a baseline from which to evaluate tolerances and spot outliers, this means that we will refrain from making any evaluative determinations about switch rebound performance.
Further, these differences, we should point out, are fairly minute - often a matter of less than a tenth of a millimeter - but even so, we want to be careful about making judgments without precise, objective measurements.
Still, our rebound data is instructive in some ways. For example, it shows the force curve of the rebound of each switch, and that’s informative. And if there are any aberrations in the force curve of the rebound of any particular switch, such as a particularly ugly dip or spike, that’s also useful data.
MORE: Best Deals
MORE: How We Test Mechanical Keyboards
MORE: All Keyboard Content
