After briefly researching the idea, he discovered the numerous
papers describing the underwater parametric array and the
earlier attempts of its application as an audible sound
source.  From these papers, he saw that there were two key
concepts which were overlooked in the previous attempts,
mitigating their success:

Preprocessing

Earlier attempts used simple AM modulation to generate the
ultrasound signal, which does create audible byproducts, but
also substantial distortion.  The nonlinear transformation
from ultrasound to audible sound is much more complex than AM
demodulation.  Therefore, in order to reduce distortion, this
specific transformation needed to be mathematically modeled,
inverted, and then applied as a preprocessing algorithm.   The
lowest-order preprocessing method, used in the earliest MIT
prototypes, was derived from a simple model [5] proposed in
1965.
 
Transducer Design

The transducers used in previous attempts were common
piezoelectric transducers used for ultrasonic ranging.  These
transducers are highly resonant, and do not have sufficient
bandwidth to reliably reproduce the preprocessed ultrasonic
signal.  Thus, even with a preprocessing algorithm,
substantial distortion would continue to result until we
developed transducers capable of reliably reproducting the
broadband preprocessed signal.

As a side project during his Master's work, he continued his
development of these ideas, studying nonlinear wave
interactions and ultrasonic transducer design, eventually
deciding to pursue the area as the focus of a doctoral
dissertation.  Of all the universities that he applied to, he
decided that the free-wheeling nature of the MIT Media Lab was
the ideal environment for developing the idea.

The first full size prototype was demonstrated in April 1998
to our Media Lab Sponsors, and performed beyond all
expectations. The first demonstration was a John Coltrane
solo, whose saxophone was heard loud and clear, projected like
a spotlight all around a movie theater, and flying right over
the audience. Power consumption was nominal (<30W),
construction was straightforward, and distortion had been
reduced by several orders of magnitude compared to all earlier
attempts.

                              -97-