Open-source 40um/1.6thou capacitive encoder

[nichlock]

I built a capacitive linear encoder and made the whole package (hardware, firmware, testing equipment, etc...) available open-source at: https://github.com/nichlock/open-linear-scale

This build has an electrical accuracy of about 40µm/1.6thou. Right now I can accept that... but I'm working towards a goal of ±10µm on the electrical side, which is what the pro-grade ones seem to claim.

I went deep into the design and algorithms on the readme, so I won't put the long-form here. It's the same operating principle as the digital calipers you could get on amazon (and the older pro-grade ones, but these days they use inductive).

I'm make improvements to the design for a rev B. The mechanical testing fixtures are probably contributing to my error too, so I need to learn more about mechanical work.

The project isn't by any means a professional job right now. I'd love to hear what people think, and as I evolve the goals of the project, I want to do more proper analysis of the accuracy and noise figures.

There are design details in the repository, including PDF schematics. Please let me know what you all think! How could I improve on this? Do you think you could use this project at its current state? Could you use it with some changes or improvements? What would make it more useful to you?

Any suggestions on improving the design are welcome! The one thing I'm pretty dead-set on right now is that the next revision will have two measurement scales for absolute position sensing.

[Renate]

If everyone's gone to inductive, why not this too?

What's the thermal coefficient of the PCBs? What does the error work out to be?

[nichlock]

I might shoot for inductive in a later version. It has a benefit in that it is resilient to machining oil and other moisture. The accuracy is comparable. Honestly, the real reason is because I was more interested in the capacitive scales inside the digital calipers I had at home when I started this project. Inductive still has lots of valid patents out, whereas capacitive patents have expired from what I can tell. So I have been weary about that too.

Thermal expansion is definitely something that would demand some sort of compensation. Fortunately it's mostly linear. I'm only doing small-scale calibration tests right now, but if this were used in a machine it might need a 200mm travel. In that case, uncompensated thermal expansion would lead to a linear error of roughly 30µm across the full scale over 10°C.

[Renate]

What's the scanning rate on the EM's? Is it 1 of 8 or 4 of 8?

[nichlock]

The pattern is 4-of-8 if you mean the spatial period? Not all four necessarily activate on every scan step.

The emitters get a repeating pattern where 4 get a positive pulse, 4 negative. The transitions are:

0: PPPPNNNN
1: NPPPPNNN
1: NNPPPPNN
...up to 7

As written, that would require multiple positive pulses in a row. Something needs to happen in between those. So the real scan pattern is:

1: PPPPNNNN
2. 0NNN0PPP
3: NPPPPNNN
...up to 15

Where 0=no transition. It is still 4-of-8 periodic. The 3-emitter signals are also still usable to get a position.

[negativ3]

Looks really nice. What's the PCB material?

[nichlock]

My builds are made with JLCPCB's FR4-TG135.