Biomedical Engineering Reference
In-Depth Information
TaBlE 7.4:
Pico Remote power consumption
SySTEM
SPIKE SoRTINg (mw )
BIN CoUNTINg (mw )
Pico Remote
100
87
Pico Remote with TDT
with headstage
143
130
Once the tests are completed and the design is stable, the system will be repackaged in a small
rat backpack not larger than 1.5 × 3 in. in size and weighing a few ounces.
7.4 PoRTaBlE dSP dESIgNS: ThE NEURal SIgNal
PRoCESSoR
Because the modeling is in the critical path of BMIs, an effort to create low-power flexible DSP
boards was pursued in the scope of generation 1 implementations. The system described here serves
as the digital portion of the overall BMI structure and is responsible for mapping the neural firings
into action in the external world. To do this mapping, the system first acquires digitized neural data
from the analog module. After acquisition and spike detection, the system then computes a linear or
nonlinear model to map the neural data to a trajectory. Once a trajectory is determined, the system
finally transmits commands wirelessly to an off-board robotic arm as described in mode 3. Mode
2 probably can be implemented without the use of a DSP. Mode 1 also requires a DSP for data
compression. Recent results with a specially designed self-organizing map [
54
] or a vector quantizer
tuned to action potentials [
54
] show that compression ratios of at least 100 are possible preserving
spike shape.
Because the system needs to be carried by a subject (human, primate, or rat) without wires, there
are portability and wireless constraints that will affect the design choices. First, the size and weight
must be small enough to accommodate the various subject species. This involves the actual size of the
components and the printed circuit board layout. Second, power consumption must be low to allow
the use of lighter batteries and yield longer experimental trials. With power and size constraints placed
on the system, the choice of processor and wireless transceiver becomes a concern. Because most of
the prediction models require the use of floating point arithmetic, the processor needs to be floating
point with enough speed to attain real-time model computations yet low power enough to appeal to
the system constraints. In addition, the wireless transceiver must not only be small with a sufficient
transmission bandwidth for current and near-future needs, it must also be low power. In the following
