Movement direction decoding of local field potentials using time-evolving spatial patterns

A main disadvantage of using intra-cortical recordings for Brain Computer Interface (BCI) is their inherent non-stationarity and instability. Thus developing direction decoders for Local Field Potentials (LFP) that are robust over time becomes a difficult task. In this paper, we show the superior performance of qualitative information over the absolute power of the recorded signals by introducing a novel method, that uses time-evolving spatial patterns. This method over-performs the baseline method by 30% on an average over a two week testing period and provides a bit-rate of 0.98 per trial. Further, these spatial-patterns provide robustness against learning when new field-forces are introduced.