UC San Diego

The connection between our brain and our muscles via the peripheral nerves enables us to communicate and act upon the world. Individuals suffering from disease and injury that affect these connections have limited treatment options and often rely upon prostheses. A fundamental challenge for such prostheses is the development of a control system that is capable of interpreting desired intentions accurately. Existing control schemes utilize activation by other functional systems using clever mechanical linkages or electromyography signal based control. Increasingly, neural interfaces are being considered for these applications due to their potential ability to restore behavioral function without co-opting existing motor functions. Implanted electrode based neural interfaces, such as electrocorticography (ECoG) hold promise for providing high signal to noise ratio measurements that are stable over long time periods while electroencephalography (EEG) provides convenient, non-invasive scalp surface measurements. Although these techniques provide powerful approaches to measuring neural signals, their application currently yields limited performance in neural interfacing applications.

Addressing the complexity of interpreting neural signals, we explore different neural decoding algorithms and evaluate their performance and limitations. Taking advantage of the time-varying properties of neural oscillations, we propose algorithms for sequence learning and sequence transfer learning. We also explore the different user interaction scenarios and the associated performance limitations. Furthermore, in advancing the adoption and understanding of non-invasive neural interfaces, we outline a portable system capable of measuring complex montages of multi-modal bio-sensing data. The perspective gained from these analyses provide better insights to practical neural interface design.