Current experimental protocols limit awake and behaving primate
electrophysiological studies to timescales of a few hours with
behavior constrained by chairing and head posting. Tracking neural
signals over days involves patching together data sets separated by
many hours, leaving unobserved periods. The HermesB system provides
the ability to overcome these limitations by recording broadband
neural signals, allowing for extraction of both LFP and spike data,
from a chronically implanted 96-electrode array (Cyberkinetics, Inc.)
in an unrestrained freely behaving rhesus macaque. The HermesB system
is battery powered, self-contained, and programmable; it allows for
two broadband neural channels to be selected and recorded from
simultaneously, while measuring head acceleration from a 3-axis
accelerometer. Channel selection is programmable, allowing autonomous
sweeps of multiple channels. We validated recording quality by
comparing data recorded from HermesB to data recorded from a
commercial system; spike sorted data from the same channels recorded
on these two systems show nearly identical unit isolations. We
conducted analyses on continuous 54 hour data sets to demonstrate
neural correlates for physically active and inactive time periods, as
defined by the response of the accelerometer. Neural data was
collected from area PMd. During active periods, 5-25 Hz LFP power was
significantly reduced. Using a single threshold fit to LFP power, 93%
of blocks tested were correctly classified as active or inactive. A
prosthesis could therefore infer when its user is asleep, which would
prevent undesired movements and save power. Also in this volume (see
Linderman et al.), we show spike waveforms changing over the
timescales of our recordings and in response to large head
accelerations. These initial results strongly suggest that continuous
recording of neural waveforms is central to neuron identity
verification and can be used to improve prosthetic performance.