Multi-Touch Proto 2

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

I don't think it would work well, there are enough scanned key (and scanned output LED) devices that have a bad feel to me. Add in the pressure sensitivity and the large capacitance of the antennas slowing down your possible scan rate.

Plus TDM methods wouldn't scale well to 16 I/O for example. Scan times would go up quickly.

Inventor · Posted: Tue Dec 30, 2008 1:16 pm Post subject:

Randy Jones · Joined: Dec 15, 2008 Posts: 12 Location: Seattle

Nicholas, you certainly do have some filter tricks up your sleeve! Your solution is very clever and uses building blocks I hadn't considered. thank you for writing it up so clearly. I will have to spend some more time in order to fully grok the CIC filters one of these days.

I am still not sure what the embedded hardware is going to look like. Depending on that, some of these tricks might come in handy. One issue is that I might not know the phase of the carrier signals: imagine that they are generated by some waveform synthesizer chip, separate from the demultiplexer.

I'm not too worried about optimizing the demultiplexer yet because generating the carriers is proving to be the tough part of the design. I want to make a significantly higher-resolution matrix next, on the order of 16 by 16. (or 4 by 64, etc.) The carriers need to be very accurate in frequency, and each on a separate conductor of course. If they are all in the same octave (ok, octave and a fifth), the design might work with square waves as carriers, which could lead to some cheaper solutions. I'm currently simulating some of these ideas on my prototype.

----

bugfight, time-division multiplexing could work-- it's normally how people do multi-part capacitive sensors. In principle it's fine, but in practice it's usually done way too slowly. Also, it creates timing offsets between the different parts of the surface that would probably have to be corrected for in software or hardware.

A more DSP-based solution lets me leverage hardware that is coming down in price all the time. I may end up using the internal mux in something like a DSPic.

----

mistercooper, the monome is definitely an inspiration. I'm committed to keeping all the software and hardware for this project open.

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

Thanks Randy.

I am not sure carrier phase is important. Actually when I first started the design I offset the phase between the carrier and the pickup but I don't think the simulations in my paper were done that way.

Sine generation is not a problem, for FPGA at least. They can be generated with very nice dynamic range and frequency resolution.

I think I could build a hardware version for $100-200 that would demodulate all of the signals, but I wouldn't know what to do with the data next. I didn't pay as close attention to the centroid finding algorithm or the actual synthesis methods. Would it be possible to have 5 CV/gate outputs and cap the maximum number of centroids to 5? How else could the data be easily loaded into a PC (minimal drivers, etc)?

I think putting all of the (pulse) carriers in 1 octave makes post processing more difficult.

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

Inventor: in his paper Randy mentioned the minimum update rate he wanted. I don't remember the specifics but 100Hz would not have met the requirements, his runs at 1300Hz.

I meant that detecting pressure in addition to which "switch" is pressed adds another variable to something that is traditionally done pretty poorly.

Why do you think the capacitance is small?

Build a TDM version, I am interested in seeing what you come up with!

Inventor · Posted: Tue Dec 30, 2008 3:14 pm Post subject:

bugfight · Joined: Aug 02, 2007 Posts: 188 Location: Arlington, TX USA

i won't be building this anytime soon, the queue is too fat and bloated already. his fatness is making him lethargic too...

a 1ms response is not too much to ask for though.

i am curious what would happen to a pulse when 2 or more, erm... appendages are applied to the same row...

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

You'd have to ask someone who has built the device on what the necessary update rate is. His reasoning and experience seems sound. I think that "feel" of an interface is often underrated, I can't stand laggy buttons, LCD, etc.

I maintain that the capacitance is large. Exclamation

I believe I have taken a look at your project before but I will when I get home tonight. Smile

Randy Jones · Joined: Dec 15, 2008 Posts: 12 Location: Seattle

Inventor · Posted: Tue Dec 30, 2008 5:35 pm Post subject:

Oh, I see now, you need such a high update rate for expressiveness, including scratching and rubbing... OK. I was just thinking in terms of finger motion but you're doing a lot more than that!

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

1pF is less than I expected.

If 6 kHz is the minimum that is passed, the R in the RC time constant is 26 Mohms, right? (1/(2*pi*6e3*1e-12). I'd expect the R to be a lot smaller, closer to whatever the input resistance is on the soundcard. Then again RC networks aren't my specialty.

Here's an example of why I don't think a capacitive switch network would work very well. Take an LED. I can't imagine it has much capacitance but I don't feel like looking it up. If you send a PWM to an LED with a high enough frequency (I thought it was around 1 kcps but its been awhile, its definitely not into 100s of kcps) the LED will appear to be on all the time even though it may only be on 50% of the time. I don't believe its because the eye doesn't react that fast because if the LED is moved around quickly (and the PWM frequency is high enough) the trails are a continuous blur rather than discrete on/off states (like some of those annoying LED brake lights).

I think if a pulse was run through a capacitive sensor and TDM the same sort of smearing would occur even if the capacitance were small.

How is finger motion different than expression? Confused

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

I did a simulation and my filter doesn't seem sensitive to phase differences between the carrier and pickup input. I doubt your FFT would be either.

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

...And I wouldn't worry too much about CIC filters. They are similar to boxcar integrators...or basically just a FIR with N taps equal to 1. When you do it in CIC form you save some additions and put it in a structure more friendly to sample rate changes. Actually I can probably be quoted as saying the CIC filter is mostly useless in the past couple months since its only useful in a couple applications like huge sample rate changes and signal smoothing/averaging. It just so happens that CIC fits this application extremely well since all of the undesired tones are known, so the zeros can be placed accordingly.

I also don't think tone generation is the biggest problem. You have a few choices. If you really like your FFT demodulator you could use an IFFT to generate the tones. You could also use a DDS/NCO which can be a lookup table with some dither or Taylor series correction or a CORDIC. Either way extremely precise tones can be had.

If I had the time/skills I would be building myself a 16x16 Randy-brand multitouch pad. Twisted Evil

Inventor · Posted: Tue Dec 30, 2008 8:54 pm Post subject:

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

Blink an LED at 100 Hz 50% duty cyclone and wave it around and tell me its a smooth line. Its not (as I remember). The modulation breaks the eye perception rule.

In fact the LED Christmas lights running at 60 Hz irritate my eyes quite a bit, along with monitors.

Inventor · Posted: Tue Dec 30, 2008 9:37 pm Post subject:

blue hell · Posted: Tue Dec 30, 2008 9:44 pm Post subject:

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

But I do see CRT monitors pulsing at 60Hz, I cannot work on them. 75Hz is OK but 85Hz is pretty much necessary for me. Same with Christmas LED.

I understand refresh rates and all that, my point was at a certain fairly low frequency (say 1 kHz but I don't actually remember), devices with even a small capacitance like an LED can behave differently, going from discrete on-off points to a continuous on (when waved around) within a matter of Hz. I don't know if the diode capacitance is modeled as shunt or series though.

Either way I predict problems with a TDM approach. And why go there, Randy's method is sweet.

Randy Jones · Joined: Dec 15, 2008 Posts: 12 Location: Seattle

Inventor · Posted: Tue Dec 30, 2008 10:28 pm Post subject:

urbanscallywag · Joined: Nov 30, 2007 Posts: 317 Location: sometimes

Are you sure about multiplying 2 sinusoid that are 180 degrees out of phase giving no signal? If you were summing them sure. My simulation indicates no problem.

I think you're right about the IFFT, you'd need 1 per carrier unless you filtered each signal out. I was looking at it as an OFDM type problem but the IFFT automatically sums the waveforms making it less useful here.

Randy Jones · Joined: Dec 15, 2008 Posts: 12 Location: Seattle

Randy Jones · Joined: Dec 15, 2008 Posts: 12 Location: Seattle

Shameless bump with recent developments.

Some students working with Dan Overholt at Aalborg University in Denmark are finishing up a controller in a guitar neck. For audio interfaces they are using some dirt-cheap USB 2x6 boards made by some Chinese company, then making them into an aggregate audio device on MacOS. See their videos [vimeo.com], and await more information as I am doing. Smile

I started a company, Madrona Labs. It has a website.

monokinetic · Joined: Aug 01, 2006 Posts: 100 Location: prague

Hi Randy,
just read through your site, wow it looks very interesting. Really glad you pursuing this idea. I'm just about to leave on holiday for a week but after that I plan on getting involved on your forum Smile

Good luck, looking forward to this one!

David