Trill Flex Base noise

giuliomoro

When you run it, check the console (at 115200bps) and paste here what gets printed

Desan

The script crashes my board. I can't get a COM connection anymore. I recovered the board by putting in into bootloader mode, reuploaded it, and then loose it again.

giuliomoro

I'm wondering if it runs out of RAM (although that should be highlighted at build time). Try changing constexpr size_t kNumReads = 1000; to 10. If it works, then try 100 and if it works keep increasing by 100 until it stops working and report the results with the largest number you can get it to run with.

giuliomoro

Also, if you have a Bela board, you can run the same code there and that should work without problems.

Desan

Hi,
I tried it with 10, still looses connection (or never regains connection after upload). Arduino IDE only says 9% programm memory and 18% dynamic memory is being used after compilation. At kNumReads = 1000 its 9%/63%. So not the RAM.

I don't have a Bela board. I only have the Uno R3 to test additionally (which also shows the the problems with the rev.C1's baseline; not a Nano 33BLE problem generally as I thought originally)

max

Desan just letting you know three people are currently working on getting to the bottom of the issue you described.

Desan

Wow, thanks, I really appreciate it! I hope we can get this worked out! 🙂

Was anyone able to replicate the issue in anyway? Let me know if you need more info.

Desan

Don't know if it helps but I wanted to clarify again that not all C1`s behave the same.

This is the C1 we talked about, in Diff mode the first pad just doesn't zero:

This C1 works, its has some reads for the first bin in the beginning, but then equilibrates:

The A2 show zeros from the plugin/reset right away, absolutely no issues with that one.

I checked the C1s a bit closer. If you put tape over the microcontroller one can read the label, and I noticed that there are differences in what I assume is the manufactorer or plant (TWN vs PAI).

working C1:

BLissue C1:

all A2:

Might be nothing but wanted to share.

Desan

After a very quick testing of all my C1 TRILLs without flexPCB:

Out of 8 TWN trills: all work
Out of 8 PAI trills : 4 work, 3 show first pad BLissues, 1 sometimes works

This was a very quick screen, would have to look deeper into it.

Edit:
All 4 BLissue PAI-C1s have the same issue with the flexPCB inserted.
1 out of all the 8 TWN-C1s seems to have an an issue with the flexPCB inserted.

giuliomoro

@Desan the tests done so far seem to show that the C1 revision seems to be slightly noisier than other revisions, even though it has the same grounding scheme as the earlier A2. We swapped a chip from C1 and A2 and the noise didn't follow the chip but stayed with the PCB. That's puzzling because the A2 and C1 versions are very similar. The C2 we tested are also fine.

Anyhow, long story short: the workaround to increased noise is to increase the noise threshold . This is set to 40 by default, but you can set it all the way up to 255. trillSensor.setNoiseThreshold(100) (with appropriate sleeps on either side) should do the job.

If the boards prove to be unworkable even with the increased noise threshold, let us know and we will happily send you replacements.

Desan

Thanks a lot for the deep dive! I really appreciate it!
Very quick test seems to get rid of the first pin miss-reads, but I have to look deeper into it and then expecially with my custom flexPCBs. Increasing the NoiseThreshold is not ideal for my aplication, as I am working with very slight touches, but I will have to test it out.

Also I realized today, after I tested all of the ones I have yesterday, that I actually have some C2s as well (4), and all of those work perfectly fine (from the TWN group from above)!

We can talk more per email, if you like, but are you working on another rev or is the C2 gonna stay for now? Just because if everything works out for me here I will need quite a lot more and then it would be nice to have the same revision.

Again, thanks a lot to all of you ( @giuliomoro , @max, and your third one^^), awesome support, has to be said!

giuliomoro

Desan Increasing the NoiseThreshold is not ideal for my aplication, as I am working with very slight touches, but I will have to test it out.

For reference, on Gliss we use a noise threshold of 245 with the prescaler at 3 and it still feels pretty responsive to light touches. There may be more targeted filtering options (e.g.: 2-dimensional median filtering) that can be applied but require more RAM than it's available on the chip itself so it needs to be done on the host processor.

Desan , but are you working on another rev or is the C2 gonna stay for now?

C1 and C2 use chips with the same die but slightly different form factors. The reason we developed them both was to be able to use whichever chip is available at any given time. Following this in-depth dive, I think we'll work on making a C1.1 (name TBC) where we fix any issue with the PCB and achieve performance on par with the A2 and C2, so we are back to having a working design for each chip.

max

To expand on the answer by Giulio, the C2 design seems to work fine, we still need to find the actual source of the issue some TRILL FLEX BASE C1 boards show. So yes: C2 will stay as it is, no reason to change it at this time.

« Previous Page