Description
Development of prosthetic limbs that offer more degrees of freedom in gesture control can benefit from peripheral neural activity recording. A network of miniaturized wireless implants that sit locally near residual peripheral nerves in amputees and record and transmit high resolution neural activity can enhance the functionality of such prosthetics. Such a network can be realized using small ultrasonically operating motes and be interrogated with a single-element external transducer. Multiple access protocols are adopted to permit simultaneous communication with the individual motes.
This overall system is constrained not only by common issues associated with simultaneous multi-transmitter communication, but also by a set of requirements imposed due to the design of the ultrasonic motes, the power/data delivery protocols, the mechanical nature of ultrasonic transducers, as well as computational simplicity on the implant side. Achieving high throughput communication with the implants faces several challenges as a result.
This project aims to address those issues and offers a machine learning (ML) based approach that achieves near an order of magnitude of improvement in the bit-error rate (BER) performance compared to traditional methods. Compared to state-of-the-art, this work provides 4 times higher total channel capacity and the largest number of implants.