Multichannel,
Temporally-Interlaced, Pulsatile Speech Processors
1987 - 1997
Descriptions and Quotes from the Literature:
There was some good news sometime in 1987 or 1988. A
significant section
of an RTI progress report detailing the proposed design of RTI's
new portable real-time processor (the portable processor was designed by Charlie
Finley) indicates significant progress in understanding channel
interactions had been made by RTI. Although there was still more
to learn. This section was also included in the April, 1988 RTI
contract proposal described in the next paragraph.
Between the end of summer 1987 and April 1988, RTI and Mark White
agreed on a set of projects for a sub-contract for NCSU (M White, PI).
The sub-contract was part of an RTI proposal to the Neural Prosthesis
Contract Office of NIH. The
first version of the proposal was submitted in April, 1988.
During the initial part of this planning process, RTI's first priority
was for NCSU to develop a robust F0 extractor for their cochlear implant
processor (using artificial neural network technology). As before, Mark
White indicated that this would not be very interesting to him. He
reminded RTI that there was significant evidence that the CNS could
extract f0 cues "on its own" -- if channel pulse-rate were high
enough (and, of course, assuming that the envelope-detectors' cutoff
frequencies were set higher). Furthermore, he also reminded them that
raising the pulse-rate (far above F0) could communicate to the CNS other
important information (e.g., F1) -- that was not currently made
available to the CNS by RTI's low-pulse-rate IP processor (i.e., pulse
rate = F0 during voiced-speech, 200-300 PPS, at most, during unvoiced
speech).
The different view-points of the two groups were again
discussed: RTI was concerned that pulse-rates above 200-300 pps were
above the maximum discharge-rate of auditory nerve fibers. At the time,
RTI felt that it made no sense to drive the nerve at rates above their
maximum discharge rate. M White indicated that fibers may not be driven
to such high discharge rates (i.e. many, or all, fibers might be driven
stochastically -- i.e., at lower discharge probabilities). Also, RTI
had become concerned about inter-channel interactions that could occur
during such high-rate non-simultaneous stimulation. M White indicated
that such interactions were far less substantial than those interactions
caused by simultaneous stimuli; and that the "time-constant"
and level of non-simultaneous interactions varied widely across patients
(with some patients having unmeasurable levels, even at the shortest
interpulse intervals), electrode types, and inter-electrode distances.
All of the information in the previous 2 paragraphs had been published
one or more years before RTI first tested a patient at UCSF: please
carefully examine the "basic science" sections of this web site for the
research that caused M White to repeatedly suggest and explain the
evidence for the proposed processor improvements and related
experiments.
During the planning and writing of a very large basic
science program project proposal (1987-1989) with other scientists, RTI
learned a lot about the nerve's stochastic behaviour and how such
behaviour might impact stimulus coding and processing for cochlear implant
systems. Once again,
science played an extremely important role. Via the "Road-Map" you can access a larger overview of all
of UCSF's and other's work on understanding the auditory nerve's stochastic behaviour.
Finally! During this planning process RTI finally understood the
different key research areas, most of the data, and the rationales for
testing at much higher pulse rates than 300 PPS. This broke the
logjam. During a fateful week during this planning period, Charlie and
Blake responded with more understanding than in previous years about
testing at much higher pulse rates than 300 PPS. Blake was particularly
emotional about it. The
following excerpt and email
from Charlie Finley describes
just one
of the key breakthroughs in RTI's understanding that had "held it
back."
In RTI's April 1988 contract proposal listed the new processors they
were planning to test. The 1st new processors that they listed were
"promising variations of the existing IP strategy" (see
excerpt 1 and
excerpt 2 from that proposal). The resulting sub-contract for M
White continued to support the "development of new signal processing
strategies for cochlear implants" but, needless to say, did not mention
the "development
of a robust F0 extractor!" See
pages B-13 and B-14 in Appendix B of
that RTI grant proposal.
The RTI contract proposal to NIH was finally funded after
re-submission. In 1989, RTI tests a relatively low-rate "maximum-rate" IP processor
on a subject with a per-cutaneous connector. This processor is the
"maximum-rate" processor described in the patent application: excerpt
pp. 11-12 (from RTI
QPR N01-DC-9-2401QPR01) of processor description that can be
compared with that in the patent application above. The testing results
were somewhat encouraging: excerpt
p. 23 and "follow-up studies... to evaluate additional variations
of the 'maximum rate' IP processor" were scheduled for the next quarter
excerpt
p. 33 from RTI
QPR N01-DC-9-2401QPR01.
The next quarterly progress report from RTI indicates the
"maximum-rate" IP processor's (now named the "super-sampler") parameters
(i.e., in particular, pulse-rate) were finally adjusted over a wide
range and then renamed: excerpt
p. 3 (see excerpt p. 33 from previous QPR) and excerpt
p. 4 from RTI
QPR N01-DC-9-2401QPR02. Better speech perception was obtained by
substantially increasing the pulse-rate of the "max-rate" IP system
(i.e., the "super-sampler," and renamed again to the "CIS processor")
and making the corresponding change in the low-pass-filter's integration
window (excerpt
p. 7). It was postulated that the increased pulse-rate allowed
more fine-grain temporal information to be communicated: excerpt
p. 8. The equivalence of the "super-sampler" and the "max-rate"
variation of the IP processor class as described in the patent
application is made particularly evident in: excerpt
1 p. 6 and excerpt
2 p. 6. Compare this description of the "super-sampler" to the
description of the "max-rate" version of the IP processor in the patent
application.
In the third RTI
quarterly progress report (1989-90), the following excerpt
again establishes that the "maximum rate IP" processor, the
"super-sampler" processor, and the "CIS" processor are identical. The
only difference in RTI's mind, apparently being that they finally tested
the processor at relatively high pulse-rates and got better results and
renamed the system.
It's possible that there were secondary factors that were involved in
RTI's delay in testing at higher rates: e.g., Their ability to drive a
sufficient number of electrode-channels at high enough rates may have
been limited (1) by their laboratory speech/stimulus-delivery system,
and (2) by those patients without a "transparent,
trans-cutaneous-connection to their intrascalar electrodes." And (3) in
some cases, patient bandwidths were severely limited by several factors
(e.g., see this
section of an RTI progress Report). At least in regard
to item (1) above, given RTI's significant motivation and technological
ability, it is likely that such 'technical problems' would have been
quickly circumvented -- and indeed they were!
A note about the difficulty in
obtaining accurate, historical documentation: None of RTI's QPRs for
Contract # N01-NS-9-2401, covering the period 1988-1992, were
available on NIH's web site -- until recently: click-here
to see a "snap-shot" of the previous web site's contents. All
other RTI QPRs were available (QPRs before and after
this 1988-1992 period) through the web site. That may be one of the
reasons that many investigators have been unaware of the history of IP
processor development. After I made a request, a new staff member at
NIH made all of RTI's QPRs available at
this website. I thank him for that.