Figure 4
From: Leveraging neural dynamics to extend functional lifetime of brain-machine interfaces
Contribution of rotational modes to decoding. (a) Contribution of rotational modes (versus purely exponentially decaying modes) to the decoded speed (Methods) in Monkey J. Error bars denote standard deviation. While both decoders demonstrate a decrease in contribution from rotational modes as electrodes are lost (slopes significantly different than 0, p < 0.01), the trend in the HNDF is substantially shallower than in the NDF (difference in slopes, p < 0.01). (b) Same as (a) but for Monkey L. (c) Contribution of rotational modes in Monkey J to the NDF, decomposed by frequency of the mode. Contribution from higher frequencies decreases as electrodes are lost (gray arrow) while contribution from lower frequencies and DC modes increases (white arrow). (d) Same as (c) but for Monkey L. (e) Same as (c) but for the HNDF decoder. The contributions from rotational modes is more consistent when dynamics are remembered, even as electrodes are lost. (f) Same as (e) but for Monkey L.
