 | | | | | | |
| |
|
In lugnet.robotics.rcx, Steve Hassenplug wrote:
> In lugnet.robotics.rcx, Mark Bellis wrote:
> > You can use four touch sensors per input port on the RCX.
>
> While in theory, this sounds very cool, in pratice, I doubt it will work quite
> that well.
>
> The biggest problem will be that the standard touch sensors may not be 100% on
> when you press it. This means the sensor itself can provide some resistance,
> depending on how hard you press it.
>
> You can test this by hooking up a touch sensor (configured to display raw
> values) and softly press it. You'll notice, the harder you press it (to a
> point) the more the number changes. (I forget if they go up or down, using the
> standard software/firmware). Then, as you release the button (VERY slowly) you
> see the numbers change, again.
>
> So, the sensors won't give you a solid on/off reading. I believe the reading
> can easily vary 5 or 10 percent.
>
> That means, if you have 4 touch sensors, there are 16 possible combinations, so
> each can vary by at most 6.25% (100%/16readings=6.25%) or there is 6% between 1
> and 2 or 2 and 3
>
> So, given that each sensor could vary as little as 5%, then at BEST you have
> 1.25% difference. At worst, the reading differ by -20% (readings overlap 20%)
>
> It doesn't exactly come out like that, but let's just say it won't work very
> well.
Good analysis.
Series can work quite well as long as you know that the touch sensors are
touched mutually exclusively, but you are right parallel doesn't work well. I
remember trying to come up with good resistance values to do parallel, and
didn't have much luck.
You might be able to do it with two touch sensors, right? Two sensors can have
four possible combinations, so you get 25% per combination. This is higher than
the 20% worst case, so it might work.
With three touch sensors, you get 8 combinations, at 12.5% per combination. I
don't think that would work worst case.
Kevin
| | | | | | | | | | | | | | | |
| |
|
I have done work in this area.
I chose to solve the problem to permit 4 touch sensors to be pressed in any
arbitrary combination with high probability of getting the answer right :)
A simple R, 2R, 4R, 8R scheme will do.
I chose R = 40.2K. This is high enough that the contact resistance in the
sensors is negligable.
I used 40.2K, 80.6K, 160K and 320K. These resistors all have to be 1%.
First of all, note that 40k is eight 320Ks in parallel. 80K is four, and 160K is
two. Since the RCX's pullup resistor is 10K, you can easily compute the
resistance of the "load" (resistance of none - all of the switch es closed) by
processing the reading (0 - 1023 in passive raw mode).
The load resistance can then be converted to a "number of 320K resistors in
parallel" which must yield a number on the range 0 - 15. This number when
expressed as a binary value 0000 - 1111 then actually respresents which switches
are open and which closed.
The key routine is the conversion from raw reading (0 - 1023) to number of
320Ks. My approach was to round up by "half a unit value", ie; the effect of a
single 320K resistor before doing the final integer division which produces the
0 - 15 "answer".
I would suggest anyone trying to repeat this should take a close look at ohms
law - its pretty simple. To optimise the final quantization, a quick look at a
famous work done centuries ago - Newton's Kissing Sphere's would help. (This
theory is often used in defining signal encoding separation in modems.) The
advantage it has in this application is in only needs to be applied in one
dimension, as opposed to the modem applications which may require 4 or more
dimensions.
We actually sold these simple multiplexors back when but since we had largish
quantities of the four resistor values required, a laborious process of creating
calibrated sets was put in place to get the value matching to better than 0.3%
to increase the liklehood that the processing would guarantee the correct
answer.
JB
| | | | | | |
