BMI Systems - Brain-Machine Interface Engineering - page 198

C H A P T E R 7
BMI Systems
additional Contributors: Erin Patrick, Toshikazu Nishida, david Cheney,
du Chen, yuan li, John g. harris, Karl gugel, Shalom darmanjian, and
Rizwan Bashirullah
The success of translational neuroprosthetics to aid patients who suffer from neurological disor-
ders hinges on the goal of providing a direct interface for neural rehabilitation, communication,
and control over a period of years. Without the assurance that the neural interface will yield
reliable recording, preprocessing, and neural encoding/decoding in a small portable (or implant-
able) package over the patient's lifetime, it is likely that BMIs will be limited to acute clinical
interventions. However, the reality of the present systems is far from long life or fully implanted
systems.
The present state of BMI technology requires that human or animal subjects be wired to
large equipment racks that amplify and process the multidimensional signals collected directly from
the brain to control in real time the robotic interface as shown in Figure 7.1 . However, the vision
for the future is to develop portable BMIs as shown in Figure 7.1 c, which incorporates all of the
complexity of the laboratory system but in a portable package. Data bandwidths can typically ex-
ceed 1Mb/sec when monitoring neural signal activity from multiple electrodes and as a result, many
of the systems with wireless links are not fully implantable and require large backpacks [ 1-7 ]. To
overcome these challenges, several parallel paths have been proposed [ 7-12 ] to ultimately achieve
a fully implantable wireless neural recording interface. But the truth is that the development of a
clinical neural interface requires beyond state-of-the-art electronics, not only DSP algorithms, and
models as treated in this topic. The purpose of this chapter is to review the basic hardware building
blocks that are required for a portable BMI system, establish the bottlenecks, and present develop-
ments that are currently underway.
As discussed in Chapter 1 , the design space in portable BMIs is controlled by three axes:
computation throughput (for real time), power consumption, and size-weight. Obviously, all three
depend on the number of recording electrodes, so BMI technology should be specified by the size -
weight, power, and computation per channel . Each electrode can sense action potentials and LFPs,
and creates a digital stream of 300 kbits/sec with conventional analog/digital (A/D) converters