[From Peter Cariani [960514.2000)
from Ellery Lanier [960514 11:00 AM mst}
An appeal for courtesy. The posts of Peter Cariani are fascinating but it
would be real nice if he would include a brief description of the basics of
what he is writing about for us non-initiates. CSGNET is an
interdisciplinary group and I am sure that we all want to know what the
other fields are doing.
If the meeting at Flagstaff will be just a lot of technical jargon about
pitch detection it will be a major disappointment. The broad foundations of
Control Theory need more discussion.
I think you can rest assured that the meeting in Flagstaff won't be just
alot of technical jargon about pitch detection (I won't be there to divert
attention away from the broad foundations... 
My apologies for the lack of an introduction on neural coding and pitch,
but Bill and I got sucked into this thread in the heat of the argument.
Initially, people were discussing "consciousness", and I proposed that there
could be a theory relating the organization of neural signalling processes
to various "states of consciousness" (e.g. awake conscious awareness, sleep,
anesthesia) and to various perceptual experiences. I think the discussion of
the relation between our internal experience and what is going on neurally is
rather spent out. I think I brought up the perception of pitch because I
wanted to ground the discussion more in concrete observations that have been
made about neural activity. I am currently studying the "neural coding
problem" as it relates to the perception of pitch, i.e. what patterns of
neural activity carry the information that the brain uses to perceive
pitch? What is the neural "code" that subserves pitch perception?
Its hard these days to find textbooks that discuss "information" or
"neural codes" or the brain as an adaptive signalling system, but the issue
of the nature of the neural signals bears on the fundamental functional
organization of the brain.
So, a very simple and very rough primer on neural codes. Neurons send signals to each
other using trains of all-or-nothing spikes (pulses, action potentials),
and whatever information is sent must be carried in some aspect of
these spike trains. Rate-codes assume that the average firing rate (# spikes/second)
of a neuron is the signal it sends to other neurons (1 signal/neuron at
any given time, a number, a scalar quantity). Temporal codes rely on
time patterns of spikes, so it is possible to send multiple independent
signals in a single spike train (multiplexing, multidimensional signalling).
The debate between rate-codes vs. temporal codes is related to fundamental
issues in control theory because it bears on the question of how complex
control systems might be organized. In the rate-coded perspective,
one thinks of the brain as a huge telegraph network in which every
telegraph station (e.g. a neuron) gets inputs from many, many others,
adds them together in some way, and produces one signal that it sends
out on wires to other stations downstream. Individual, specific wire connections
are absolutely critical here because they determine what is connected to what
and how various signals are weighted, and there is no way to distinguish what
a signal means except by which input line it came in on.
Imagine what happens if suddenly we have the ability to encode information
in different frequency bands (like radio) and each node in our network
can propagate all frequencies in its output.
A given node can pay attention to a particular frequency
channel, and alter the signal in that channel while passing through all
of the other frequencies. Signals can then be distinguished on the basis
of their temporal patterns (frequency channel 1 is smell, channel 2 is
pitch, channel 3 is color, etc.) rather than where they came from, and
this means that one no longer needs highly specific sets
of connections between individual nodes to keep all of the signals
separate and distiguishable. One also no longer needs to organize the
system in an elaborate hierarchy in order to get a message
from point A to point B; More decentralized, and "anarchistic"
modes of organization are then possible. One can think about different sets of distinct
temporal spike patterns in place of the frequency channels, and one has
some sets of signals that are independent and others that are not.
The issues take you into whether every signal that is controlled must be
a scalar, how many signals are combined together (is it a sequence of
scalar controllers? are they in parallel? how does association work?),
whether the brain is a hierarchy or a heterarchy, etc.
Like color, pitch is a complex percept. Monochromatic illumination produces
particular color percepts (all other things being equal),
and pure tones produce particular pitches. However, spectrally-complex light
(having multiple frequencies/wavelengths) can produce
color percepts that are not directly related to any of the
individual component wavelengths that are present.
Light with very different spectra can give rise to the same
color percept (metamery). Similarly, periodic (harmonic) sounds
(complex tones) can be built up from adding sinusoids of
different frequencies, amplitudes, and phases. These
"complex tones" that are made up of multiple frequencies
(partials, stimulus components, harmonics, pure tones) can
give rise to ("low") pitches that are not associated with any one of the stimulus
components. Sounds with very different spectra can produce the same low pitch
(metamery). For a harmonic complex tones, we almost invariably hear a low
pitch at the fundamental frequency F0, which is the greatest common denominator
of the component frequencies (harmonics). Each harmonic frequency is an integer multiple
of the fundamental frequency (F0). The space of the pitch percept is a
spiral, with "pitch height" the monotonically increasing quality
(e.g. pitches of a series of pure tones increasing in frequency) and "pitch
chroma" or the "musical pitch" or the musical note (e.g. C, D, E, F, G, A, B)
having a cyclic structure. Pitch height spans the range from less than 50 hz to
over 15 kHz, while the cyclic, pitch organization (the octave, the fifth,
"tonality", "harmony") is restricted to low- and medium-frequencies (below 5 kHz).
The patterns of pitches that are heard for complex tones can be quite complex,
and there have been extensive debates over the last 150 years as to
the nature of the auditory representations and information-processing
mechanisms that are responsible for them. These aren't esoteric issues,
because most pitches that we hear (voice pitch in speech, musical pitch)
are complex-tone pitches and there is often not much spectral energy
at the fundamental frequency (but we hear loud and clear the
"missing fundamental"). There is also a very elaborate set of perceptions
we have around particular frequency ratios and combinations of frequencies that
bear fundamentally on how we construct auditory objects (what do we group together
and why, what do we separate and why, auditory Gestalts, "scene analysis"). These
properties of the auditory system and the structure of our perceptions may be
due to the actions of elaborate central processors that have learned all of the
requisite associations needed to recognize harmonic structure. Alternatively,
they may be direct consequences of the nature of the underlying neural codes
that are used by the auditory system. I've been experimentally studying the
neural coding of sounds at the level of the auditory nerve and cochlear nucleus
for the last six years, and it appears to me, on the basis of huge amounts of
spike train data, that interspike interval codes (times between spikes in a
spike train) account for almost of the major pitch phenomena that I know of.
No other neural code at the level of the auditory nerve
corresponds to patterns of pitch perception anywhere near as well as this one,
with as much breadth and economy of explanation. Stimulus-specific timing patterns
can be seen in every other sense modality if you look carefully in the literature,
but these phenomena are rarely ever discussed in the textbooks. So caveat emptor!
I have a 1000 word summary of temporal coding evidence; if people are interested
I could post it. The upshot of all of this perceptual coding stuff is that the
nature of the perceptual signals may tell us about the nature of the neural
control mechanisms that use them, and this may lead to new kinds of mechanisms
and organizations that we hadn't thought of before.
I hope this is somewhat helpful -- it's been written off-the-cuff very hastily
so it's not as well organized and straightforward as I would have liked, but maybe
it will give you and others some way of thinking about this seemingly inscrutible
argument that some of us are having.
Peter