Using feedback in long term trajectory decoding from Local Field Potentials

In this paper, we study a feedback mechanism in the design of a Local Field Potential (LFP) based Brain Computer Interface (BCI) that decodes arm movements. A major setback of using Local Field Potentials based BCI is their non-stationarity. In addition, many proposed BCI devices are usually trained and simulated in an open-loop environment, neglecting the effect of user adaptation in the loop. To tackle these problems, a Hammerstein-Wiener based decoder is proposed to model the nonlinearities of the system in a modular fashion. Furthermore, partial feedback is incorporated to achieve high decoding performance, particularly when decoding data from trials conducted as many as 14 days following initial BCI training.