Earlier Research Processor Implementations at UCSF, 1975-1981
UCSF's General Purpose
Non-Real-Time Speech Processor/Stimulator System ~1975-1979
-
UCSF developed an 8-channel non-real-time speech
processing system -- with the assistance of Marshal Fong and Linsay
Vurek. This system was designed to be a highly versatile speech
and stimulus processor. Implementing High-rate, interleaved
pulsatile processors was one of the principal design objectives for
this system when designed and built in 1975-79. As it turns-out,
only a small part of this system was actually used for patient testing
-- for basic psychophysical experiments and synthesized-speech
experiments. This versatile system required substantial software
and hardware for that era (documented in Chapters 4 and 5 of
White's Ph.D. thesis, 1978). This first UCSF
computer-speech-processor/stimulator system was designed before UCSF
had had any significant experience with human implant psychophysics.
Very few implant laboratories had had any experience at all.
UCSF's design specifications were partly based on the very
preliminary work of other implant laboratories. For the
most part the design specifications were appropriately conservative --
given the lack of knowledge about cochlear implants at the time.
However, for one important specification M White was not
conservative enough: the design did not "allow for" enough
digital-to-analog amplitude resolution (only 8-bits) for each of the
electrode channel outputs. Ideally, M White should have added at
least 4 additional bits of "reserve resolution." Unfortunately,
changing the D/A resolution for the 8 electrode stimulators was not as
simple as it would be today. There were 3 primary reasons
why UCSF did not go ahead and upgrade this system:
-
UCSF's small engineering group was already heavily
committed: designing and producing new patient electrodes, a
percutaneous connector system, a matching 4-channel trans-cutaneous
transmitter-receiver, and matching compact portable speech processor -- to be
used by the few research patients after they had completed 3-6 months of tests using the per-cutaneous
system. In addition to this, a "clinical version of
this implant" was being developed and produced "under-the-gun" by
the same
engineering group for distribution to a substantial number of
hospitals around the U.S. and in Britain!
See this page for more
information about the allocation of engineering
resources.
-
Although any real-time system created in that era would, by
necessity, significantly constrained speech processing options, UCSF
decided to construct two Real-Time Processing Systems.
[Amusing time-frame reminder: M White's 1978 Berkeley Ph.D. Thesis
was produced using a typewriter, not a word-processor!]
The real-time processor option offered significant advantages:
-
The available staff for testing speech-processors were
largely audiologists, with very limited experience in using
computer systems -- let alone a "non-real-time,
digital-speech-and-delivery system using
a parallel processor." In contrast, the testing/audiology
staff had already become, or were becoming, capable of using
the much simpler bench-top
real-time processor for testing patients.
-
Real-time systems offered significant benefits for
audiologist-patient communications: (1) improved testing
efficiency and (2) improved patient-audiologist interpersonal
communications
-
Real-time processors did not require often-lengthy computer
processing prior to each day of patient testing. Such
processing could constrain patient testing schedules and
require additional human resources.
-
The first real-time system was designed and
constructed relatively quickly. This was important because
patients were already being implanted by the UCSF surgeons at
the time.
-
Initially, it was believed that the 2nd real-time system
could also be constructed relatively quickly -- particularly
since the 2nd real-time system's architecture and parametric
specifications were already specified; and because the
"analog-front-end" of the second system was practically
identical to the first system (e.g., highly-flexible modular,
reconfigurable, and adjustable preprocessing, band-pass
filter-bank, compressors, safety-clippers; only
variable-cutoff-frequency-envelope-detectors had to be added to form the
"analog-front-end" for UCSF's real-time, high-pulse-rate,
interleaved-pulses processor. More information on the
UCSF analog real-time system is available immediately below.
-
Real-time systems offered a strong potential advantage for
patient learning, even though such learning would have to
occur relatively quickly, i.e., in a laboratory environment where
researchers,
audiologists, the patient, and friends interacted during and in
short-intervals between testing sessions.
UCSF's Real-Time, Analog, 4-Channel Processor
-
In 1978, when it became apparent that patients would be implanted and
"ready for testing" at a much earlier time than had been planned for, or expected, UCSF
researchers decided to build an analog real-time
processor. An analog system was chosen because it could be
built quickly and because there was some evidence that multichannel
analog systems could offer some benefit to patients. In
1979-1981 UCSF designed and constructed an analog real-time laboratory (table-top)
processor.
-
The first real-time, laboratory processor was a 4-channel
analog system. It was rack-mounted and small enough to place on
a table. Each processing module's (e.g. microphone preamp,
pre-emphasis filter, band-pass-filters, compressors, safety-clippers,
etc) input and output was accessible from the front panel. All
of the processing modules were connected to each other on the front
panel -- using coax-cables with bnc connectors -- and could be easily
interconnected in different configurations for testing. The
modules were very flexible, with adjustable processing parameters for
most of the modules. The table-top laboratory system electrode drivers
were isolated from each other, and could be switched to current-controlled or
voltage-controlled mode. The voltage-controlled mode allowed us to estimate the
consequences of using the simpler voltage-controlled drivers in the take-home
and subsequent transcutaneous rf systems. The take-home system only had
isolated voltage sources.
-
It was relatively simple to design and build. In some
cases, we purchased commercially available signal processing modules.
For example, the four compressors were sophisticated commercial
units that were extremely versatile (571 compressor chips and a "Gain Brain" for the front-end).
For the "designed in-house"
components, Lindsay Vurek did most of the circuit-level design.
He also constructed all "in-house" modules, and then "put
everything together" into one table-top box. He also designed and
constructed the take-home analog system.
* In this investigators opinion, it is quite possible that the
rivalry between the UCSF and the House Institute surgical staff was at
least partially responsible for the many extra tasks that the engineering
group was assigned.
** It should be noted that almost all early multichannel analog systems
used unisolated, voltage-controlled stimulators. To create a truly
multi-channel system required that each channel's stimulator be
electrically isolated from the other channels. It should be noted
that the electrode contact impedances were far larger than the tissue and
perilymph impedances in multichannel implants (see //TBD). As a consequence,
defining such systems using unisolated channels as "multichannel" is
questionable at best. In contrast, Don Eddington's real-time multichannel
analog system did use channel-isolated, current-controlled stimulators --
as did UCSF's 8-channel non-real-time speech processing laboratory system.
UCSF's portable, take-home device had isolated channels, but used voltage drivers. Whereas,
UCSF's table-top real-time systems had isolated current sources (as well as isolated voltage sources as an option, for comparison purposes).
See pp.
166 of Merzenich, 1983 for a diagram of one commonly-used configuration of UCSF's
table-top real-time 4-channel analog processor. An
aside: I remember hearing that Dr. Hall
and Dr. James Flanagan from Bell Labs designed and built one or more
optically-isolated current sources in the 1960s for cochlear implant
research that they had conducted with Blair Simmons. When I was doing the same
thing 10 years later for animal experiments, I certainly felt in good company!