Biomedical Engineering Reference
In-Depth Information
taneously within the NSP. Once the prediction model completes an epoch or computation cycle, the
program must interrupt the MSP and transfer any required data. This process also involves creating
the correct packets for transmission to the appropriate server (off-board).
The final layer of code resides in the on-board CPLD. This hardware-based VHDL code is
responsible for correctly shuttling data between all of the hardware modules. It achieves this pro-
cessing through a series of interrupts and control lines that are provided by the individual hardware
components. Depending on the functionality of the system, this code is responsible for all the inter-
nal and external communication of the NSP system.
We use a high-speed USB connection to transmit 104 binned neural channels and 3D trajec-
tory data to compute an optimal linear combiner using 10-tap FIRs per channel trained with the
normalized LMS algorithm for a total of 3120 parameters [
55
]. The average amount of time for
a single prediction when computed over 90 predicted outputs is 211 µsec. This test achieves the
same timing results on the computation of predicted output because this was the same DSP used in
our previous systems [
56
,
57
]. On a 600-MHz PIII laptop running Matlab 6, the average predic-
tion takes 23 msec. The factor of improvement or speed gain is about 100× for the DSP over the
laptop running Matlab. The LMS output results collected at the receiving computer were directly
compared to Matlab computed outputs and they agree within seven decimal places. The NSP uses
approximately 90 to 120 mA, which equates to 450-600 mW. Overall, this is much lower than the
4 and 1.75 W previously attained by other acquisition hardware [
56
,
57
]. In addition, the board size
is 2.5 × 1 in. board size and weighs 9 g.
7.5 FloRIda wIRElESS IMPlaNTaBlE RECoRdINg
ElECTRodES
The second-generation systems will have to deal with an immensely more restrictive set of require-
ments because they will be subcutaneous. We envision a modular design with tens to hundreds of
channels. The present requirements that we are working with is an implantable rechargeable device
called the Florida wireless implantable recording electrodes (FWIRE) that will measure 1.5 × 1 ×
0.5 cm, weigh 2 g, and can collect data from 16 microelectrodes for 100 hours, amplify and send
them wirelessly within 500 Kbits/sec to a remote system within 1 m of the subject. A conceptual
building block is presented in Figure
7.15
.
As illustrated in Figure
7.15
, the FWIRE microsystem consists of a flexible substrate that
serves as the platform for the signal processing integrated and fine integrated circuit and wireless
telemetry radio frequency integrated circuit chips, transmit antenna, receive and power coil, and
microwire electrode array. A low-profile rechargeable battery is located below the flexible substrate
and is used to power the implant electronics during recording sessions. The external coil wraps
