Biomedical Engineering Reference
In-Depth Information
Mode 1:
Compress the raw neural recordings (action potentials and LFPs) and transmit the full
information to a remote processor where it can be recovered with a given precision for
visualization and/or BMI model building. The bottleneck in mode 1 is the compression
algorithm (i.e., how many channels with which fidelity can be transmitted through the
high-speed wireless channel, with a given power budget).
Mode 2
: Extract features from raw neural recordings (preprocessing) and transmit only the action
potentials or spike counts to a remote station for model building. This is the simplest of
the modes, but the issue is to do spike detection and sorting with sufficient precision be-
fore binning.
Mode 3
:
Locally perform the model building and just transmit the commands to the robotic device
through a low-bandwidth wireless channel, which requires mode 2. In this case, the wire-
less channel needs only low bandwidth (send three numbers every 100 msec).
Mode 4
:
Mix and match among all of the above.
These different possibilities affect the computation and wireless resources, and a flexible
hardware solution would be very beneficial.
Generation 2 moves the amplification and preprocessing of neural data into a fully implanted
device that transmits the data wirelessly to a base station, where data can be visualized. BMI models
are implemented outside the implant and whose outputs control a robotic device through a low-
bandwidth wireless channel. Generation 2 systems are at a research stage. The interfaces typically
consist of a microelectrode array that can record (motor) or control neural networks chemically
(drug delivery) and via stimulation (retinal, cochlear). Because the device is implanted beneath the
scalp, it needs to be self-contained; therefore, the electrodes are directly integrated into the electron-
ics, the hardware substrate, and the packaging, and remotely powered, monitored, and controlled via
a bidirectional telemetry link [
22
]. Reports in the literature have provided a wide variety of layouts
and physical forms with the entire implant surface area of approximately 100 mm
2
and heights
above the skull of no more than several millimeters.
This chapter describes recent progress in each of these areas. several possible generation 2
architectures are possible and will be briefly reviewed.
7.1 SENSINg NEURoNal aCTIVITy: ThE ElECTRodES
The ability to quantify neuronal representations has been fueled by advancements in neural record-
ing technology and surgical techniques. The trend since the last century has been to develop acqui-
sition systems with the ability to spatially resolve the activity of large ensembles of single neurons.
This goal has been met by advancing electrode design, amplifier design, recording D/A, and surgical
