The prototype Whisper system had been using a unique flat speaker that
offered a small form factor, a lightweight design, and a broad spectrum
(for its size). However, during experiments comparing the HiBall tracker
with Whisper, Nick Vallidis discovered that the speaker was causing problems.
This prompted the change to a more conventional speaker.
The Flat Speaker
Figure 1 below provides information about the flat speaker. Note the language
about "four small speakers for high-frequency sounds." Looking at the
diagram of Figure 2, the four separate areas of the speaker (one in each corner)
can be seen. It is believed that these four "separate" speakers are
actually seen as physically different emitters by the Whisper system. Thus,
for higher frequencies, four different points in space are "battling"
to represent a single point. (Note that future work could investigate using this
feature to our advantage. But for now, it's a problem).
Figure 1: Information about flat speaker
from Digi-Key catalog.
Figure 2: Diagram of flat speaker.
Figure 3 below shows a manifestation of the problem. This plot resulted
from rotating the flat speaker slowly away from the microphone. The vertical
axis represents chips (i.e. range). The horizontal axis can roughly be thought
of as milliseconds. To exaggerate the problem, the speaker was held such that
two opposite corners were perpendicular with the axis of rotation. Thus, as the
speaker is rotated, one of the "mini-emitters" gets closer to the
microphone as a second one gets farther away. The remaining two emitters are
roughly centered on the rotational axis. Note that the speaker was handheld
and hence the range drifts slightly as the speaker is rotated.
Figure 3: Flat speaker range plot with errors.
Left of plot shows speaker orientation.
The shaded area highlights the problem. As the difference between the
near and far emitters increases, the range fluctuation grows. It appears
that the near emitter (which would presumably be louder wrt the microphone)
eventually "wins," causing the sharp decrease of over one chip
slightly before the 6000-millisecond mark. Figure 4 below shows a sample
correlation result for the experiment. Note the two peaks. This is
undesirable to say the least.
Figure 4: Correlation with two peaks.
Go to page: