33
Next Steps
hut edited this page 2 months ago
These are potential directions that the project could go in the future:
Applications
- Turning it into a musical instrument
- Connecting it to an extra thumb (see paper) (By going to https://plasticity-lab.com/publications and clicking on the link with the same title as the paper, one can access the full paper for free)
Hardware
- Switching to another instrumentational amplifier that works at 3.7V and switching to a lithium battery that supplies 3.7V, eliminating the need for a power converter
- The MPC6N11-100 instamp seems to work fine for this.
- Try I2C for communication between the individual PCBs. Add a 4-channel ADC with I2C output to each electrode module. Connect each board via daisy chain
- Switching to operational amplifiers instead of instrumentational amplifiers (to save cost, and perhaps space on the PCB)
- Adding multiplexers to allow for more than 8 signals
- Fixing interference issues on signal 2
- Use linear regulator chip instead of opamp for generating reference voltage
- Try decent electrodes
- 3M Red Dot 2248-50: See blog post "3M Red Dot electrodes"
- 3M Red Dot 2660
- Metal parts soldered directly onto the PCB
- Use surface-mounted (SMD) pin headers (alternative) and pin sockets instead of through-hole ones to eliminate points of contact on the underside of the PCB
- A way to avoid through-hole components for attaching the Arduino is to add contact pads on the power module to which the arduino can be directly soldered on to.
- Surface-mounted AAA battery clips are available
- Redesign it so that a PCB manufacturer can assemble it completely, except for screws, wires, and elastic bands.
- Rectification
Software
- Better neurofeedback
- Generating sound from signals
- Upgrading to KiCad 6
- Move neural net evaluation into the Arduino with TensorFlow Lite and imitate bluetooth keyboard/mouse so you can use PsyLink on any bluetooth-capable device without software (once it's calibrated)
- Software rectification of signals
- Support for pre-trained neural network weights as a starting point for users to train a custom neural network for their personal EMG signals and labels/key presses, to improve prediction quality and training speed
Documentation
- Publish datasets of recordings -> Sample Signals blog post
- Publish trained NN model
- Record a video of setting up a Linux environment to run PsyLink
- Create data sheets for recent prototypes
- Create tutorial for running PsyLink UI
- Add bill of materials to prototypes -> BOM blog post
- Add blog post on P8/P9's electrode module configurations
Community
- Add read-only mirrors on GitHub
- Automate synchronization
Marketing
- Design and print stickers
- Find sponsors